Electrostatically Induced Recruitment of Membrane Peptides into Clusters Requires Ligand Binding at Both Interfaces

Protein recruitment to specific membrane locations may be governed or facilitated by electrostatic attraction, which originates from a multivalent ligand. Here we explored the energetics of a model system in which this simple electrostatic recruitment mechanism failed. That is, basic poly-L-lysine binding to one leaflet of a planar lipid bilayer did not recruit the triply-charged peptide (O-Pyromellitylgramicidin). Clustering was only observed in cases where PLL was bound to both channel ends. Clustering was indicated (i) by the decreased diffusional PLL mobility DPLL and (ii) by an increased lifetime τPLL of the clustered channels. In contrast, if PLL was bound to only one leaflet, neither DPLL nor τP changed. Simple calculations suggest that electrostatic repulsion of the unbound ends prevented neighboring OPg dimers from approaching each other. We believe that a similar mechanism may also operate in cell signaling and that it may e.g. contribute to the controversial results obtained for the ligand driven dimerization of G protein-coupled receptors.


Introduction
The association of proteins with the surfaces of plasma membranes or intracellular membranes is tightly regulated. Membrane affinity may be solely provided by electrostatic attraction of amino acid residues, which concentrate in the tertiary structure to form a binding surface [1]. Alternatively, phospholipid binding domains may be engaged, including e.g. pleckstrin homology domains and Fab1 domains [2]. When these domains interact with membranes, it involves stereospecific recognition of membrane targets like diacylglycerol and phosphoinositides. Coincidentally, protein attraction to the membrane may be aided by electrostatic or hydrophobic protein-lipid or proteinprotein interactions.
Simultaneous involvement of several detection mechanisms is believed to be responsible for the restricted, rather than uniform distribution of recruited proteins across intracellular membranes. For example, detection of both phosphoinositides and small monomeric GTPases directs the four-phosphate adaptor protein-1 to the trans-Golgi network [3]. While the benefit of cluster formation for signaling purposes is immediately evident, the affinity requirements for protein and lipid recruitment into these clusters are less clear [2].
For example, it takes both lipidation and electrostatic lipid protein interactions to target the polybasic myristoylated alaninerich PKC substrate peptide (MARCKS) [4] at the membrane. Once anchored to the membrane, MARCKS laterally recruits negatively charged phospholipids. A generalization of this phenomenon suggests that any basic peptide may recruit multivalent membrane anions into clusters [5]. However, when exchanging PiP 2 for the triply charged gramicidin peptide opyromellitylgramicidin (OPg), no clusters were observed, i.e. unilateral binding of the basic poly-L-lysine (PLL) fails to recruit OPg [6].
OPg forms ion-conductive dimers due to C-termini interactions at the membrane midplane where M and M 2 denote the monomer and dimer, respectively.
The ratio K of the respective association and dissociation constants, k R and k D , is equal to the ratio of the equilibrium surface concentrations A and AA (both in units of mol cm 22 ) of M and M 2 , respectively: PLL's failure to induce the formation of conducting OPg dimers cannot be explained by the lack of a lipid anchor. PLL and OPg possess different mobilities and they rapidly exchange binding partners only in case of unilateral PLL binding. However, the lipid anchor is still absent when bilateral PLL binding is allowed. And yet in this case, measurements of current flow through OPg channels revealed cluster formation [7].
The goal of the present paper is to clarify the mechanism and energetics of this simple OPg -PLL model system. Thus, we aspire (i) to distinguish whether the transmembrane cluster emerges by registration of half-clusters in the individual leaflets or by recruitment of entities that register across both leaflets immediately upon cluster formation and (ii) to understand why PLL binding to both channel entry and channel exit is required for cluster formation (Fig. 1).

Planar membranes
Vertical planar bilayer lipid membranes were formed by painting diphytanoyl phosphatidylcholine (DPhPC) solution in decane (20 mg/ml) over an aperture (500 mm in diameter) in a diaphragm separating two aqueous solutions. Horizontal planar bilayers were folded from DPhPC monolayers to cover the aperture (100 mm in diameter). The apertures were pretreated with 2% DPhPC in n-decane or with 0.5% hexadecane in hexane, respectively. The volumes of the lower and upper chambers were 3 and 0.5 ml, respectively. We observed formation of planar membranes (i) optically, either through a front window or through a cover glass in the bottom of the lower chamber [8] and (ii) electrically, via the determination of membrane capacitance. Electrical current was measured by means of a picoamperemeter (Keithley Instruments or VA-10 amplifier, npi, Tamm, Germany).
Most of the experiments were performed in a solution containing 25 mM KCl, 10 mM HEPES, and 0.1 mM EDTA buffered at pH 7. The ethanolic stock solution (0.015 mg/ml OPg) was mixed with the DPhPC/decane membrane-forming solution or OPg (a generous gift from N. S. Melik-Nubarov, Moscow State University, Department of Chemistry) was added from a stock solution of 2 mg/ml to the DPhPC/hexane monolayer which was formed on top of the aqueous solution.
Cy3 conjugated gramicidin A (gCy3, a generous gift provided by V. Borisenko and G.A.Woolley, University of Toronto, Canada) was prepared as previously described [9]. Cy3 attachment did not affect channel activity [9]. PLL (PLL, Sigma, Vienna, Austria) was added to one or both compartments of the cell as stated. In some experiments, PLL was labeled with Atto633. PLL, HBr (average molecular weight 24,000, 115 Lysines, Sigma), was dissolved in double-destilled water (pH 9, 2 mM) and mixed with equal amounts of peroxidefree dioxan. 1 mM Atto633-NHS (Atto Tec, Siegen, Germany) were added and incubated in the dark for 2 h under Argon. The Atto633 labeled PLL 115 was subsequently lyophilized, redissolved in bi-destilled water and stored at 4uC. Experiments were carried out at room temperature (21-23uC).

Fluorescence correlation spectroscopy (FCS) measurements
The surface diffusion coefficients of gCy3 and PLL(Atto633) were measured by fluorescence correlation spectroscopy (Con-foCor 3 attached to the laser scanning microscope LSM510, Carl Zeiss, Jena, Germany). The dyes were excited at 561 and 633 nm. We calibrated the confocal volume by measuring the residence time t R of rhodamine 6G in solution. Based on a diffusion coefficient of 426 mm 2 s 21 [10], we obtained confocal plane radii v of 0.22 mm and 0.25 mm for the two different lasers.
Autocorrelation functions G(t) of the fluorescence temporal signal from PLL-Atto633 were fitted to the two-dimensional equation where N is the number of particles. We performed so-called Z-scans to exactly position the horizontal membrane in the focus. We gradually changed the vertical position of the laser focus relative to the phospholipid surface plane [11]. Recordings made right in the focus were used to estimate D PLL . The absolute value of D PLL was determined with an accuracy of about 20% [12]. Only relative changes in D PLL are important for the scope of the current work and these were determined with much higher precision.

Evaluation of changes in the dissociation kinetics of the OPg dimers induced by cluster formation
The association of two OPg monomers from two different leaflets resulted in the formation of a transmembrane pore. These cationconducting dimers of labeled gramicidin derivatives are comparable to the wild type channel, both are stabilized in their head-to-head association by six hydrogen bonds [13]. The lifetime of these OPg dimers depends on the force exerted to decrease the membrane thickness to the size of the dimer [14]. If the number of open channels is small, it can be monitored by single channel recordings. If hundreds or thousands of open channels are reconstituted, the decay time of the transmembrane current subsequent to sudden removal of functional monomers can instead be used [15].
The monomer-dimer equilibrium instantaneously shifts upon photodynamic monomer inactivation [16,17]. The photosensitizer, aluminum trisulfophthalocyanine (AlPcS 3 , Porphyrin Products, Logan, UT), therefore adsorbed to the membrane and generated singlet oxygen 1 O 2 by a flash of light. The 1 O 2 diffusion span within the membrane [18] is sufficient to target tryptophan residues of gramicidin monomers [19]. Since both the duration of the flash and the lifetime of 1 O 2 [20] are at least two orders of magnitude smaller than the characteristic decay time (t P < 0.2 s) of the membrane current through the OPg dimers [7], they can generally be neglected. t P is obtained from a single exponential fit of the equation: to the current trace recorded after a flash of light. I ' denotes the final membrane current. The addition of appropriate concentrations of polymers leads to two-exponential kinetics: It can be shown that in the case of a mixture of two exponentials, single exponential fit should reveal t P : We calculated t P from an exponential fit (Eq. 5, Eq. 6) or we obtained t P by fitting the data with a single-exponential function (Eq. 4).
AlPcS 3 was added to the bathing solution at the trans-side. The flash lamp was attached to the cis compartment. The current I was measured under voltage-clamp conditions by a current amplifier (model 428, Keithley Instruments), digitized by using a LabPC 1200 (National Instruments, Austin, TX) and analyzed using a personal computer with the help of WinWCP Strathclyde Electrophysiology Software designed by J. Dempster (University of Strathclyde, UK). Ag-AgCl electrodes were placed directly into the cell and a voltage of 30 mV was applied to the lipid bilayer. The value of the current was about 1 mA on average which corresponded to 7.5610 6 conducting channels in the bilayer. Planar lipid bilayers were illuminated by single flashes produced by a xenon lamp with flash energy of about 400 mJ/cm 2 and flash duration ,2 ms. A glass filter was placed in front of the flash lamp to cut off light with wavelengths ,500 nm. To avoid electrical artifacts, the electrodes were covered by black plastic tubes.

Results
Free-standing planar bilayers doped with gCy3 were placed into the focus of the laser scanning microscope. Diffusion of the dye into and out of the focus resulted in fluorescence intensity fluctuations. Calculation of the corresponding autocorrelation function ( Fig. 2A) allowed determination of gCy3 residence time t R in the focus. Computation of the membrane diffusion coefficient D M according to Eq. (7): resulted in a value of 8.960.8 mm 2 /s. D M is two to three times larger than that measured by single particle tracking under comparable conditions [21]. However, it was close to the D M of lipids which was determined to be 8.160.4 mm 2 /s [12,22]. The similarity between D M of lipids and peptides with one membrane helix is in line with measurements of fluorescence recovery after photobleaching [23]. Addition of PLL in any concentration to one or both sides of the membrane did not alter D M . This result nicely agrees with the observation made by Ghambhir et al. [24] that recruitment into clusters requires the number z of charges per molecule to be $2. To test this hypothesis we substituted gCy3 for OPg (z = 3). We measured a PLL diffusion coefficient D PLL of 661 mm 2 /s ( Fig. 3B) upon unilateral PLL adsorption to the surface of freestanding planar bilayers made of DPhPC (Fig. 3A). We calculated D PLL similar to D M (Eq. 7). Increasing the PLL concentration from 10 27 to 10 24 M had no effect on D PLL . Even reconstitution of OPg in a concentration of up to , 50 dimers per mm 2 did not alter D PLL (Fig. 3B). Since OPg was not labeled, we determined its surface density as the ratio of the transmembrane conductivity to the single channel conductivity and the membrane area. Eq. (2) revealed that A matched AA for K = 1.2610 14 cm 2 mol 21 [25], i.e. the monomer was present at a surface density , 50 mm 22 . Taking into account that every OPg bears three negative charges, we obtained a density of about 300 charges per mm 2 . According to the Gouy-Chapman theory, this charge density s corresponds to a surface potential y 0 of: Assuming that the permittivity e at the membrane surface is equal to 10, we find that the Debye length l is equal to 0.68 nm, which results y 0 < 20.37 mV. Thus, OPg makes a negligible contribution to membrane surface potential, because membranes formed from neutral lipids have a surface potential of around 26 mV [26,27]. As a consequence, OPg has only a very modest effect on PLL adsorption to the membrane (Fig. 3A).
Despite the small y 0 , D PLL dropped significantly when PLL was added to both sides of the membrane. The effect depended on PLL concentration. Above a threshold bulk concentration of about 10 27 M (per lysine monomer), it became apparent that D PLL reached its minimum at about 10 26 M (Fig. 2B, green curve) and returned to the initial value for bulk concentrations .10 25 M (Fig. 3B). The peak of cluster formation was observed at a polymer density of 5/mm 2 (Fig. 3A). If we take the drop in D PLL as an indicator for transmembrane cluster formation, several OPg dimers must have been embedded between two PLL molecules.
The hydrophobic thickness of lipids and gramicidin differ by l , 0.3 nm [28]. As a result, a line tension around an isolated OPg dimer or around a cluster with several OPg dimers must exist [14]. We extrapolated s , 5 pN from tension values measured as a function of hydrophobic mismatch between lipid clusters [29]. Assuming that the cluster is circular to minimize the energy per boundary length, we calculated the lipid deformation energy DG which is spent when an OPg dimer forms, i.e. the channel opens as: where r OPg is the radius of an OPg dimer. It is calculated assuming an effective cross-sectional area of OPg, A OPg , 5 nm 2 , consisting of 2.5 nm 2 for OPg itself [30] and 0.63 nm 2 for each of the four bound lipids [31]. The result of the oversimplified Eq. (9) is in good agreement with calculations of DG from a so-called phenomenological spring constant H B [32]. For bilayers made from dioleoylphosphatidylcholine in decane, H B is equal to 56 kJ mol 21 nm 22 [33] so that DG = H B 6(26l) 2 < 8.2 kT. For a cluster of n OPg molecules Eq (9) transforms into: where DG PLL is the lipid deformation energy per cluster. According to Eq. 10, the lipid deformation energy DDG per OPg dimer in the cluster is smaller than DG: Eq. 11 indicates a reduction of strain on the six hydrogen bonds between the monomers of an individual OPg dimer. Hence, OPg dimer lifetime increases in a cluster [34][35][36].
To estimate n, we probed OPg dimer dissociation kinetics by photoinactivation of OPg monomers. The transmembrane current (which was initially about 1 mA) decayed with time constant t P , which was roughly equal to OPg dimer lifetime of 0.2s. t P increased about a hundredfold to the new value t PLL < 20s, when PLL bulk concentration reached , 1 mM PLL (per lysine monomer). It returned to control values at both much lower and much higher PLL concentrations (Fig. 4). t PLL and t P relate to DDG: Inserting the values for DDG and DG into Eq. 11, we calculate that n = 4 OPg dimers are bunched into one cluster. In contrast to the case of bilateral PLL presence discussed above, the addition of PLL to only one side of the membrane did not result in the deceleration of the photo-inactivation kinetics, indicating that clusters do not form isolated in one leaflet. Similar results have previously been reported for experiments carried out at higher ionic strength [37]. The observation agrees well with D PLL measurements which also indicated the absence of clusters.
To test the hypothesis that clustering was opposed by the electrostatic repulsion of the unbound channel ends, we increased the ionic strength in the compartment lacking PLL. We observed only a very modest increase of t p from 0.25 s to 0.5 s (Fig. 5), which indicated the persistent lack of OPg clusters. Due to its small contribution to the overall surface potential (,300 charges/mm 2 ), OPg had a small effect on PLL surface concentration. Unilateral and bilateral additions are marked with the indices 1 and 2, respectively. The subscripts a and b indicate the different leaflets (A). Both PLL surface and D PLL were determined from FCS autocorrelation functions. D PLL is equal to the ratio v 2 /4t R , where v and t R are the radius of the confocal plane and the mean residence time of PLL-Atto633 in the focus, respectively. The ratio of unlabeled to labeled PLL increased from zero to ,1000 with increasing PLL concentration. In the absence of OPg, D PLL was insensitive to the amount of PLL added (curve labeled DPhPC). The same observation was made when PLL was allowed to adsorb to one leaflet only. D PLL decreased in a narrow PLL concentration interval (B), only when present at both sides of an OPg-containing membrane. The membranes were folded from DPhPC monolayers (20 mg per ml hexane). OPg was added at a concentration to form , 50 dimers per mm 2 which was checked by conductivity measurements prior to the addition of PLL. The single channel conductance was assumed to be equal to 4 pS. The bathing solution was 25 mM KCl, 10 mM HEPES, 0.1 mM EDTA, pH 7. doi:10.1371/journal.pone.0052839.g003

Discussion
There are two alternative ways that a membrane-spanning cluster may form: (i) The constituents are first separately recruited in each individual leaflet into half-clusters, which in a second step form a transmembrane cluster or (ii) components from different leaflets first align forming a nucleus, which in a second step develops into a cluster by the concerted recruitment of more constituents from both leaflets. PLL's inability to cluster OPg monomers excludes mechanism (i), while recruitment of multiple OPg dimers to bilaterally bound PLL confirmed mechanism (ii) (Fig. 6).
The OPg and PLL concentrations at which cluster concentration reaches a maximum, translate into average distances of about , 447 nm and , 100 nm between neighboring PLL molecules and neighboring OPg molecules, respectively. Even a fully extended PLL molecule with a length of only , 45 nm does not bridge this distance. However, the molecule diffuses so fast (D PLL = 5 mm 2 /s), that within the 0. The electrostatic attraction between OPg and PLL acts to increase this population.
Once an OPg monomer binds to an already existing PLL-OPgdimer complex, the increased surface density augments the likelihood of dimerization (compare Eq. 2). The fact that 2 OPg molecules cannot be more than 45 nm apart translates into a minimal OPg surface density of 1.3610 5 /mm 2 , which shifts the minimal equilibrium dimer to monomer ratio to 0.98:0.02. Because OPg clustering reduces the energy DG which is incurred by bilayer thinning at the peptide OPg interface to DDG, K increases a hundredfold. This conclusion is based on the observation that t PLL , 1006t P . The combined effect of increased OPg surface density and the augmented K value of 1.2610 16 cm 2 / mol are equivalent to a shift in the dimer to monomer equilibrium from 0.5:0.5 (Eq. 2) to 0.998:0.002.
From the difference in DDG and DG, we estimated a cluster size of four OPg dimers. This corresponds very well to the area , 22 nm 2 of a condensed PLL molecule (115 residues) on charged planar bilayers [22]. Our current FCS measurements also agree with the anticipated cluster size. Cluster formation implies that OPg and PLL diffuse as one entity. That is, D M decreased threefold from its initial value of 8.9610 28 cm 2 s 21 to 2.8610 28 cm 2 s 21 . The first number stems from the assumption that D M of gCy3 and of OPg are similar. The second value reflects the situation in which most of the PLL molecules are part of a cluster (minimum in Fig. 3B). Such a change in D M indicates a ninefold increase in molecular weight, i.e. the radius of the diffusing entity increased from that of one OPg dimer (0.89 nm) to that of a tetramer of OPg dimers with 16 tightly bound lipids (2.6 nm). This calculation assumes that the 8 lipids bound to the dimer must be interchangeable, because they do not contribute to D M , while the 32 lipids are locked in the cluster which is sandwiched between two PLL molecules (Fig. 6B). The inability of the clustered lipids to exchange with the surrounding lipids has been previously observed and has been attributed to the line tension around the cluster [22]. Cluster size estimates are further based on the inverse proportionality of D M and the molecular radius [23]. The logarithmic dependence of the membrane diffusion coefficient on cluster size as described by the Saffman-Delbrück formalism [38] should be used for larger clusters (radius .3 nm).
When PLL unilaterally binds, the above analysis does not explain the lack of clusters. Insight is expected from a closer look at the unfavorable total energy balance: The sudden decrease of the OPg monomer fraction by photorelease of singlet oxygen (from 1 mM AlPcS 3 ) results in an exponential decay of the OPg-mediated current I through planar bilayers. The time constant of the decay 1/t P depends on the PLL concentration in the bathing medium. t P is obtained from a single exponential fit of Eq. (4) to the data (dashed gray lines). t P is equal to 0.23 s (concentration per lysine residue = 180 mM PLL) and 0.34 s in the absence of PLL (control), respectively. Double-exponential fitting (dashed gray line) is required (according to: I = I 0 +A 1 exp(2t/t P1 )+A 2 exp(2t/t P2 )) in the presence of 1 mM PLL at both sides of the membrane. The best fit is attained with the parameters: A 1 = 6.7%, t P1 = 0.53 s, A 2 = 6.2%, t P2 = 19.5 s. The initial value of the current I 0 was , 1 mA. Insert: the dependence of the averaged t P (compare Eqn. 6) on bulk PLL concentration. doi:10.1371/journal.pone.0052839.g004  where W d , W att , W rep and W e denote the dimerization energy, the attractive electrostatic energy, the repulsive energy and the entropy-induced amount of energy, respectively.
To calculate the dimerization energy W d , we take into account that the concentrations of OPg monomers and dimers are equal to each other at the highest cluster abundance (Eq. 2). The assembly of four OPg molecules in the small spot of one PLL molecule must consequently be accompanied by the formation of two new OPg dimers, which both contribute to E D . The dissociation energy of a gramicidin dimer E D is about 228.7 RT [25], where R is the gas constant and T the absolute temperature. Dimerization therefore contributes W d = 2 x E D = 257.4 RT to cluster formation. DG PLL is equal to 26 DG (Eq. 10), so that neither DG PLL nor DG appear in Eq. 13.
The attractive electrostatic energy W att is fourfold larger than the electrostatic energy W el between one OPg and PLL [39]. For the assessment of W el we use the textbook equation: where R = 0.9 nm, r = 0.9 nm, and q OPg = 23e are the OPg radius, the average distance between the interacting charges, and the OPg charge, respectively. e is the elementary charge. We estimate e to be , 10 on the membrane surface. Because PLL is treated as a point charge in Eq. (14), we have to derive an effective charge q PLL . q PLL accounts for (i) the size of the polymer and (ii) the steric restraints which arise from its interactions with other (OPg) molecules. For an assessment of q PLL we used data from a previous publication [37] in which t PLL, induced by the 60 residues large PLL (PLL 60 ), was a function of the ionic strength of the bathing solution. Krylov  At the PLL free leaflet, the unbound pyromellityl groups repel each other with W rep . Substituting q PLL for q OPg and assuming r = 0.9 nm allows utilization of Eq. (13) for calculation of the repulsive electrostatic energy W el < 25 RT between any pair of OPg molecules. In a cluster of four OPg dimers, there are six such pairs so that the total repulsive energy W rep amounts to 150 RT.
The entropy-induced amount of energy W e depends on the probability p that 4 OPg monomers (two from each monolayer) and 2 OPg dimers simultaneously hit the 22 nm 2 large spot, which is occupied by one condensed PLL molecule. There is only one such spot per 200,000 nm 2 at the highest cluster abundance. As it is tenfold more abundant, an individual OPg monomer or an individual OPg dimer encounters one PLL molecule with a probability of p m/d < 1.1610 23 . Thus, a rough estimate for p < (p m/d ) 6 It confirms our hypothesis: the repulsion between the charged OPg groups dominates. Cluster formation is only energetically favorable in the absence of W rep . To further validate the result, we decreased W rep by increasing the ionic strength (Fig. 5). In a solution of 1 M KCl, W el is equal to 6.1 RT between any pair of OPg molecules, (Eq. 14), which translates into W rep < 37 RT. Thus, W tot amounts to , 4.2 RT, suggesting that cluster formation remained unfavorable. This conclusion is in perfect agreement with the experiment (Fig. 5).
Eq. (16) also explains the dependence of cluster concentration on PLL and OPg concentrations in case of bilateral PLL addition, at least on a qualitative level. A tenfold increase in PLL concentration shifts the interfacial OPg/PLL ratio to 2:1, causing a tremendous increase in entropy-induced amounts of energy for the simultaneous binding of 4 OPgs to one PLL. A tenfold increase in the OPg concentration shifts the OPg dimer: monomer to 13:1, which vanquishes W d . W att cannot drive cluster formation on its own, because it is smaller than W e .
In summary, the PLL-induced buildup of OPg clusters isolated in one leaflet is opposed by electrostatic repulsion from the opposite leaflet. We believe that the insight gained by studying this model system may be helpful for understanding the much more complex aggregation of receptors in the cellular plasma membrane. For example, it may shed light onto the highly controversial issue [41] of ligand-induced dimerization of certain G-proteincoupled receptors (GPCRs) [42]. The critical extracellular GPCR ligand binding sites and the intracellular docking sites for Gprotein both contain charges. These charges are conserved throughout the G protein-coupled receptor family. For example, negative charges are excluded in peptide-GPCRs, whereas positive charges are excluded from the critical extracellular locus in amine-GPCRs [43]. Certain charged residues of the cytoplasmic loops are likewise crucial for C-protein coupling, as was e.g. shown for the second inner loop of the muscarinic receptor [44]. Our study suggests that charge shielding by both ligands and G-proteins may in part regulate the extent to which some of the GPCRs form dimers.