This Article Abstract Full Text (PDF) All Versions of this Article: biophysj.105.068387v1 89/5/2972    most recent Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Tzlil, S. Articles by Ben-Shaul, A. Search for Related Content PubMed PubMed Citation Articles by Tzlil, S. Articles by Ben-Shaul, A.

Flexible Charged Macromolecules on Mixed Fluid Lipid Membranes: Theory and Monte Carlo Simulations

Department of Physical Chemistry and The Fritz Haber Research Center, Hebrew University of Jerusalem, Jerusalem, Israel

Correspondence: Address reprint requests to A. Ben-Shaul, Tel.: 972-2-658-5271; E-mail: abs{at}fh.huji.ac.il.

	ABSTRACT

TOP ABSTRACT INTRODUCTION THEORY RESULTS AND DISCUSSION CONCLUDING REMARKS ACKNOWLEDGEMENTS REFERENCES

 
Fluid membranes containing charged lipids enhance binding of oppositely charged proteins by mobilizing these lipids into the interaction zone, overcoming the concomitant entropic losses due to lipid segregation and lower conformational freedom upon macromolecule adsorption. We study this energetic-entropic interplay using Monte Carlo simulations and theory. Our model system consists of a flexible cationic polyelectrolyte, interacting, via Debye-Hückel and short-ranged repulsive potentials, with membranes containing neutral lipids, 1% tetravalent, and 10% (or 1%) monovalent anionic lipids. Adsorption onto a fluid membrane is invariably stronger than to an equally charged frozen or uniform membrane. Although monovalent lipids may suffice for binding rigid macromolecules, polyvalent counter-lipids (e.g., phosphatidylinositol 4,5 bisphosphate), whose entropy loss upon localization is negligible, are crucial for binding flexible macromolecules, which lose conformational entropy upon adsorption. Extending Rosenbluth's Monte Carlo scheme we directly simulate polymer adsorption on fluid membranes. Yet, we argue that similar information could be derived from a biased superposition of quenched membrane simulations. Using a simple cell model we account for surface concentration effects, and show that the average adsorption probabilities on annealed and quenched membranes coincide at vanishing surface concentrations. We discuss the relevance of our model to the electrostatic-switch mechanism of, e.g., the myristoylated alanine-rich C kinase substrate protein.

	INTRODUCTION

TOP ABSTRACT INTRODUCTION THEORY RESULTS AND DISCUSSION CONCLUDING REMARKS ACKNOWLEDGEMENTS REFERENCES

 
The lipid bilayer, constituting the central structural element of biological membranes, is a two-dimensional fluid mixture, composed typically of many lipid species. Owing to the lateral mobility of lipids at physiological temperatures, the membrane can respond to interactions with integral and peripheral macromolecules by mobilizing those lipids interacting favorably with the macromolecule into the interaction zone. This process leads to local changes in lipid composition around the guest molecules which, under certain conditions, may evolve into larger-scale reorganization of membrane components, resulting in domain formation. The molecular composition and phase characteristics of the domains, as in lipid rafts, are different from those of the surrounding membrane (1–3).

The ability of integral proteins to induce local and global changes in lipid composition has been extensively documented experimentally (4), and amply analyzed theoretically (5,6). Similarly, experiments reveal that when charged macromolecules, such as certain kinds of proteins or DNA, are adsorbed onto a mixed membrane containing a small amount of oppositely charged lipids, the charged species migrate toward the adsorbed macromolecule (7–9), tending to achieve local electrical neutrality. Although lowering the electrostatic (free) energy of the system, the segregation of charged lipids induced by the peripheral macromolecule may involve a non-negligible entropic penalty. Several recent theoretical studies have carefully analyzed the energetic-entropic balance associated with electrostatic adsorption of rigid macromolecules (e.g., DNA and globular protein) onto fluid membranes (10–16). It was shown, for instance, that the extent of lipid segregation, the corresponding entropy loss, and the interaction free energy depend sensitively on the shape, charge, and concentration of the adsorbing macromolecule (17,18).

This work focuses on the energetic and structural characteristics of the interaction between flexible, electrically charged, macromolecules (polyelectrolytes), and mixed, oppositely charged, fluid membranes. We shall consider three-component membranes composed of one electrically neutral species and two differently charged lipids. Electrostatic adsorption on such membranes may involve significant changes in the spatial configuration of the adsorbing polyelectrolyte and, consequently, a substantial loss of conformational entropy. The two kinds of entropy loss, those associated with lipid segregation and those which lower the macromolecule's conformational freedom, tend to offset the gain in electrostatic interaction energy between the oppositely charged molecules. It should be noted, however, that both degrees of freedom, namely, lipid mobility and macromolecule flexibility, enable the interacting complex to select, with higher probability, those mutual membrane-macromolecule configurations of lowest free energy.

A delicate balance between the energetic and entropic contributions to the adsorption free energy on mixed fluid membranes is exhibited in various biological processes (19,20). One important example is the electrostatic-switch mechanism underlying the operation of the myristoylated alanine-rich C kinase substrate (MARCKS), and several other proteins (21,22). MARCKS is a prominent protein C kinase substrate, implicated in a variety of signaling pathways involving, for instance, the control of lipid second messengers and the regulation of cytoskeletal actin. This natively unfolded (and thus flexible) protein binds electrostatically to anionic lipids in the inner leaflet of the plasma membrane. Of special importance in this binding is the multivalent (here z = –4, but generally varying between –3 and –5) anionic lipid (phosphatidylinositol 4,5 bisphosphate, PIP₂). The average membrane concentration of PIP₂ is typically 1%, yet it tends to localize in viral envelopes and membrane rafts, as well as in the binding zones of various proteins involved in signal transduction pathways. Among these proteins is MARCKS, which binds to the plasma membrane through its relatively small (25-residue) but strongly charged effector domain that comprises 13 basic residues. A 150-residue-long flexible polypeptide chain separates the effector domain from the myristoylated N-terminus, and a comparably long and flexible peptide chain connects the effector domain with the C-terminus (for more details, see e.g., Gambhir et al. (9)). The myristoyl chain inserts into the hydrophilic core of the lipid bilayer, serving to anchor MARCKS to the membrane.

Experiments reveal that the effector domain sequesters approximately three PIP₂ molecules (9,23,24), suggesting that these multivalent (z = –4) lipids provide most of the negative charge required for neutralizing the 13 basic charges of the MARCKS' effector domain (25). Considering the small average PIP₂ concentration in the membrane (1%), versus the 10–30% abundance of monovalent acidic lipids (primarily phosphatidyl-serine, PS), it is clear that the protein must import the multivalent lipids from remote membrane regions. Another significant observation is that upon lowering the net charge of the effector domain from +13 to +7, MARCKS detaches from the membrane, thereby exposing PIP₂ to cleavage by phospholipase C and initiating a series of signal transduction events (26,27). The change in charge is generally achieved through phosphorylation of three serine residues in the basic domain by protein kinase C.

In addition to electrostatic interactions, MARCKS interacts with the membrane hydrophobically as well, through the myristoyl anchor at the N-terminus and via five phenylalanine side chains within the effector domain (28). A subtle interplay between the electrostatic and hydrophobic interactions, as well as the entropies associated with the long flexible chains on both sides of the effector domain, governs the intricate electrostatic switching process involving MARCKS and PIP₂. Motivated by this notion, our goal in this study is to examine the coupling between electrostatic interactions, membrane composition, lipid mobility, and polymer flexibility, and elucidate its role in the adsorption of charged macromolecules onto oppositely charged membranes. Although inspired by the biological relevance of these interactions and processes, our present calculations do not attempt to mimic in detail the behavior of any particular biological system. In fact, our simulations involve a rather short (20-segment) flexible polyelectrolyte chain, interacting with a three-component fluid membrane containing neutral, monovalent, and tetravalent lipids. More specific simulations, modeling MARCKS-membrane interaction, are in progress.

To demonstrate the important role of lipid mobility, our results for the fluid (i.e., annealed) membrane are compared to those obtained for a frozen (i.e., quenched) membrane of the same average lipid composition. The structural and energetic characteristics of a polyelectrolyte interacting with a fluid membrane are qualitatively and quantitatively different from those pertaining to any specific quenched lipid membrane. However, from the purely formal-computational aspect, as argued in the next section, the statistical aspects of polymer adsorption on a fluid membrane can also be derived using a biased superposition of statistical averages corresponding to polymer adsorption on an ensemble of quenched membranes. We shall also show that, in the limit of vanishing polymer concentration (in the bulk solution and hence also on the membrane surface), the average adsorption probability on a Boltzmann-weighted ensemble of quenched membranes equals the adsorption probability on an annealing, fluid membrane. Differences between the two kinds of membranes appear at nonzero concentrations of macromolecules, and will be accounted for using a simple cell model.

In addition to the fluid and quenched lipid membranes, some of our calculations describe polyelectrolyte adsorption on uniformly charged surfaces, such as those of metal oxides or metal electrodes (29,30). We shall see that, in general, uniformly charged surfaces adsorb more weakly than either a quenched or a fluid lipid membrane. Another limiting case of interest is that of a stiff polyelectrolyte interacting with a charged membrane. Contrasting the adsorption behavior of such a molecule with that of a self-avoiding freely jointed chain will emphasize the role of polyelectrolyte flexibility. As we shall see, weak electrostatic interaction may not suffice to overcome the loss of chain flexibility upon adsorption. This will be demonstrated by considering the limiting cases of a weakly charged polyelectrolyte and a membrane containing only monovalent lipids.

In the next section we first describe the basic statistical-thermodynamic background underlying the adsorption of a flexible macromolecule onto the various types of lipid membranes mentioned above. Later in that section we introduce the model system and describe our extended version of the Rosenbluth Monte Carlo (MC) simulation scheme (31,32), which enables efficient simulation of polymer adsorption on annealing, fluid, membranes. We then present and analyze the results of the simulations and close with a brief summary of the main conclusions.

	THEORY

TOP ABSTRACT INTRODUCTION THEORY RESULTS AND DISCUSSION CONCLUDING REMARKS ACKNOWLEDGEMENTS REFERENCES

 
The lipid molecules comprising a fluid (annealed) membrane are mobile and can thus diffuse into and out of the interaction region with the adsorbing macromolecule. The statistical thermodynamics of macromolecule adsorption onto such membranes involves simultaneous averaging over all the polymer and lipid degrees of freedom. In a frozen membrane, on the other hand, the lipids are immobile, and thermodynamic averaging is obtained by tracing over many quenched arrangements of the membrane lipids. Qualitative differences between these two membrane types are reflected in various structural and thermodynamic properties, such as the local lipid composition around the adsorbed macromolecule, and the average adsorption free energies. Formally, however, it can be shown that the various characteristics of macromolecule-lipid interaction on a fluid membrane, can be derived by a biased averaging of the same property over a Boltzmann-weighted ensemble of quenched membranes. (This suggests an indirect way of simulating adsorption onto the fluid membrane using quenched membrane simulations.) We shall see that the differences between average macromolecule adsorption characteristics on fluid and quenched membranes are substantial at nonzero concentrations of macromolecules, but disappear in the limit of vanishing concentration. We open this section with a detailed discussion of the relevant thermodynamic background, and then describe the model system and our method of simulation.

Adsorption thermodynamics
All our simulations involve one adsorbed macromolecule, as formally appropriate to low surface concentrations. However, using a simple cell model, the simulations can also be used to account for the adsorption behavior at higher surface concentrations. In this model, the membrane area A is divided into an array of A/a noninteracting cells, all with the same area, a, and the same lipid composition. The cell area is large enough to comfortably accommodate one adsorbed macromolecule. The model thus, approximately, accounts for excluded area effects but ignores other intermacromolecule interactions. Note that to contain a flexible macromolecule, e.g., an unfolded protein, a membrane cell should typically consist of several hundreds of lipid molecules. For the three-component membranes of interest here, this implies an enormous number of two-dimensional lipid arrangements.

Single molecule partition functions
We treat the quenched membrane as an ensemble of independent cells, each characterized by a specific frozen lipid configuration, m. To compare the quenched membrane with a fluid membrane having the same lipid composition, we equate the fraction, P_q(m), of quenched membranes in configuration m, with the Boltzmann weight, P(m), of m in the bare fluid membrane, i.e.,

(1)

where is the partition function per cell of a bare fluid membrane,

(2)

In these equations and below we use the subscripts q and f for the quenched and fluid membranes, respectively. Also, the potential energy U(m) of the lipids in configuration m, as well as all other energies are hereafter expressed in units of k_BT, the thermal energy, where T is the temperature and k_B is the Boltzmann constant.

The partition function, per cell, of a fluid membrane occupied by an adsorbed macromolecule is given by

(3)

Here U(m,p) = U(m) + U(p|m) is the potential energy corresponding to the membrane-polymer configuration m,p. The term U(p|m) stands for the energy of a polymer in state p, interacting with a membrane in a given configuration m. It includes the self-energy of the polymer (i.e., the sum of its intersegment potentials), and its interaction energy with a membrane in state m. By p = ,r we refer to the polymer chain conformation, , and the position, r, of the polymer relative to the membrane plane (see below). The sum over p thus (tacitly) involves integration over r, implying that is a configurational partition function, bearing the dimensions of volume. As above, U(m) is the interlipid interaction energy.

The sum of Boltzmann factors,

(4)

introduced in the third equality in Eq. 3, is, of course, the partition function of a macromolecule adsorbed onto a membrane of a specific lipid configuration m. From Eq. 3 it thus follows that the partition function of a macromolecule interacting with a fluid membrane can be expressed as a Boltzmann average of the partition functions corresponding to the ensemble of quenched environments. Note that the constant lipid energy, U(m), is not included in our definition of i.e., the energy of the occupied m-cell is measured relative to the ground state energy U(m). On this energy scale we obtain for the empty cell. The quantity introduced in the last equality of Eq. 3 may be interpreted as the average partition function, per cell, in a Boltzmann-weighted ensemble of quenched membranes.

From Eq. 3, it also follows that the average of any property A (e.g., adsorption energy, polymer radius of gyration, etc.) of a polymer adsorbed on a fluid membrane, can, in principle, be evaluated as a biased average of A in the ensemble of quenched membranes. Explicitly, let

(5)

denote the average (over polymer conformations) of A, for a polymer adsorbed on a quenched membrane in configuration m. Using Eqs. 1– 4, we find

(6)

Note that all quantities on the right-hand side of this equation depend only on quenched membrane properties. The second equality thus offers a way of calculating A_f as a weighted average of the polymer conformational averages, A(m), in the ensemble of quenched membranes. In this biased average, here denoted as the weight, (), of the quenched membrane configuration m, is the product of the fraction of such membranes (P(m)) with the statistical weight ( ) of all polymer conformations on this membrane. It will be shown below that this formal relationship between fluid and quenched membrane averaging may be given a physical meaning in the limit of vanishing macromolecule concentration. In this limit the probability of finding a macromolecule adsorbed onto a quenched membrane m is proportional to and hence is the average of A(m,p) in the ensemble of quenched membranes. Note, however, that this average differs from the simple average

Membrane partition functions
To account for the adsorption behavior at nonzero surface concentrations we should consider a many-cell membrane in equilibrium with a solution of macromolecules. Suppose the bulk solution is of volume V, and contains N_b macromolecules of chemical potential µ. For simplicity we assume dilute solution behavior, in which case µ = –ln q_b + ln _b, where _b = N_b/V is the bulk density of macromolecules, and

(7)

is the internal partition function of a macromolecule in the bulk solution. Note that the summation here is over all possible conformations of the macromolecule, ensuring that its center of mass (or one of its segments) is kept fixed in space. Note also that q_b, like all partition functions in our treatment, is a configurational partition function. The momentum factors in the partition function cancel out identically in all relevant expressions (33). Note, however, that µ refers here to the configurational part of the chemical potential, also known as the excess chemical potential (32).

The cells comprising a fluid membrane are identical. Treating the membrane as an open system with respect to macromolecule exchange, the grand-canonical partition function of the membrane is

(8)

where = exp(µ) is the absolute activity, and is the two-state (i.e., empty and occupied) partition function of a membrane cell. Using N_s to denote the number of macromolecules adsorbed on the membrane surface, the fraction, _f = N_s/M = N_sa/A, of the membrane area occupied by macromolecules (or, the surface coverage), is given by We thus obtain a Langmuir-like adsorption equation

(9)

where in the second equality we have used the dilute solution limit of the activity, = _b/q_b = exp(µ). In the third equality we have introduced the dimensionless bulk concentration where v is a volume per macromolecule defined in more detail below. Thus, may be regarded as the volume fraction of polymers in solution.

The grand partition function of the quenched membrane is given by

(10)

where M_m = P(m)M is the number of membrane cells with a two-dimensional lipid distribution m. The probability of finding an m-cell occupied by a macromolecule is so that

(11)

Averaging over all the quenched configurations, m, and using Eqs. 3, 9, and 11, we find

(12)

or, equivalently,

(13)

Actually, with the definitions of F_f and F_m given in Eqs. 9 and 11, the last equality follows directly from Eq. 3.

Both _m/(1 – _m) and exp(–F_m) are convex functions of their arguments. Using Jensen's inequality of convex functions (34), it thus follows from Eqs. 12 and 13 that

(14)

for any probability distribution P(m). In other words, on average, macromolecule adsorption onto an ensemble of quenched membranes (whose lipid configurations appear with probabilities P(m)) is always weaker (lower and smaller –F) than adsorption onto a fluid membrane of the same lipid composition.

From Eq. 12 it follows that the equality _mq = _f is obtained only in the limit of vanishing surface coverage, i.e., when (F_mq F_f requires that all F_m are negligibly small, as can be seen from Eq. 13.) Underlying the limiting behavior _mq = _f is the fact that although the lipid molecules of a quenched membrane are, indeed, immobile, the adsorbed macromolecules can nevertheless explore all membrane states, m. This may be achieved via lateral diffusion on the membrane surface and/or by desorption from one local membrane region and adsorption into another. Note also that in the limit we obtain (see Eq. 11). Thus, in a Boltzmann-weighted ensemble of quenched membranes, the number of macromolecules bound to membranes of lipid configuration m, is proportional to P(m)_m and hence to If measured over an ensemble of quenched membranes, the average of a physical observable A(m,p) would then be given by which, as noted in Eq. 6, is equal to A_f. Thus, in the 0 limit, both the average surface concentration, and the -weighted average of A(m,p) over the quenched membrane ensemble, approach the corresponding quantities in the fluid membrane.

A physical interpretation of the last conclusion can be given in terms of the cell model, as follows. When 0 (and in the absence of kinetic constraints), each of the adsorbed macromolecule can freely and independently visit all membrane environments m, thereby sampling all possible lipid-polymer configurations m,p. Furthermore, since none of the cells is blocked, all m and hence all m,p are sampled according to their Boltzmann weights, just like the states sampled by a macromolecule on a fluid membrane. The difference is, of course, that a macromolecule adsorbed on a fluid membrane need not migrate from one cell to another to sample the entire configuration space. Consequently, = _f is the same for all cells of the fluid membrane, whereas a wide distribution of _m values characterizes the ensemble of quenched membranes. Note however, that in practice, even this formal equivalence between a fluid membrane and an ensemble of quenched membranes may not materialize, even when 0, because of kinetic barriers to macromolecule mobility.

Interestingly, it is possible that for some quenched membrane states m, F_m F_f (and hence, _{m f}). This may appear surprising in view of the fact that the free energy, F_f (not F_f), of a macromolecule interacting with a fluid membrane is invariably lower than the free energy, F_m, of a macromolecule adsorbed onto any quenched membrane, m. (This is because involves summation over both m and p and is therefore larger than any ) Hence, After adsorption, however, the distribution of lipid arrangements in the fluid membrane is no longer the Boltzmann distribution before adsorption, implying a loss of lipid-mixing entropy. Of course, no loss of lipid entropy is involved upon adsorption onto a quenched membrane, which explains why certain quenched states can be more attractive to macromolecule adsorption than the fluid membrane.

Upon increasing the concentration of macromolecules in solution, the more strongly adsorbing m-values of the quenched membrane will be occupied first. Once these favorable local environments (or cells) are populated, further adsorption is necessarily suppressed. This implies _{mq f}, because in the fluid membrane every cell can independently anneal its lipid distribution, thereby enhancing adsorption.

Our conclusions regarding the relationship between macromolecule adsorption on quenched versus fluid membranes agree with previous works pertaining to polymer statistics in random media. Cates and Ball (35) have studied the behavior of a single long polymer chain in a random medium and concluded that, as long as the environment is infinite, the quenched and annealed averaging will yield the same statistical chain properties. Our fluid and frozen membranes are analogous to the annealed and quenched random potentials in the treatment above. Inequalities valid for the multichain adsorption, analogous to Eq. 12, have been obtained by Andelman and Joanny (36,37) for neutral chains adsorbing on annealed and quenched flat surfaces. The main conclusion there is that the density of polymers on an annealed surface (membrane) is always higher than in the frozen case.

The adsorbed state
The MC simulations presented in the following sections enable evaluation of all the partition functions encountered above, as well as a variety of relevant structural properties. First, however, we have to clarify what distinguishes an adsorbed macromolecule from a free macromolecule in solution.

We noted above that the sum over p in Eqs. 2 and 3 involves all possible chain conformations, , as well as all possible positions, r, of the macromolecule, relative to some arbitrary point on the membrane. Identifying the membrane surface with the (x,y) plane, the only relevant coordinate is the distance, z, of the macromolecule from the membrane surface. This distance can be expressed in terms of the normal displacement of any chain segment (or the center of mass) from the membrane plane. In the simulations, we find it convenient to measure this distance in terms of z₁, the normal displacement of the first (more precisely, terminal) chain segment (see Fig. 1). Beyond a certain distance from the membrane surface, comparable to the range of membrane-macromolecule interactions, the macromolecule is not affected by the membrane. This cutoff distance, , may be defined in terms of z₁, such that for z₁ , the macromolecule is considered adsorbed, and otherwise as free in solution. An alternative, yet practically equivalent definition of can be given in terms of the average segment density profile (see below).

View larger version (22K): [in this window] [in a new window]   FIGURE 1  A schematic drawing of the simulation model. A 20-segment long chain of spherical segments, each carrying a single point charge in its center, interacts with a mixed membrane composed of neutral, singly charged and tetravalent anionic lipids, which occupy the sites of a two-dimensional hexagonal lattice. Lipid charges are concentrated in the centers of the corresponding discs. The lipids can diffuse (exchange positions) within the membrane plane. The polymer chain is flexible, but subjected to electrostatic and short-range spatial repulsion between its constituent segments. The diameters, d, of polymer segments and lipid disks are equal.

(15)

with v a, and

(16)

In the last equation we have introduced q_f(z₁), the partition function of a macromolecule whose first segment is fixed at z₁. The factor a in Eq. 15 results from the fact that the partition function per molecule must be proportional to the cell's area. The volume element, v = a, may now be interpreted as the volume of a membrane cell. Thus, represents the (average) partition function per unit volume of an adsorbed macromolecule, or, equivalently, the average partition function per unit length along the membrane normal. Note that for large z₁ (practically for z₁ ) we have U(m,;z₁) = U(m) + U(), and hence

We may now rewrite Eq. 9 in the form

(17)

Similarly, for the quenched membrane

(18)

where q_m(z₁ > ) = q_m () = q_b. In Eq. 18, to enable straightforward comparison with the fluid membrane, we keep using the same concentration units,

For chain molecules composed of L segments, the segment density in the bulk solution is (z = ) = _b = L_b. Near the membrane surface the segment density, (z), is different from _b and is given by

(19)

where (z₁) is the density of macromolecules whose first segment is at z₁, and n(z|z₁)dz is the average number of chain segments between z and z + dz due to chains originating at z₁.

The surface excess of adsorbed macromolecule is, by definition,

(20)

where, in the second equality, is the average (three-dimensional) density of chain segments within the surface layer. Note that the upper limit in the integral defining can be replaced by (or any larger value). The second equality may also be regarded as the definition of the surface layer thickness . Note that in the limit of vanishing surface density, a_s/L = .

A slightly different definition of , useful in our numerical calculations, can be given in terms of the ratio q(z₁)/q() = (z₁)/_b. Namely, we can choose as the smallest value of z₁, beyond which this ratio is practically 1. In practice, the two definitions are indistinguishable, because so that the integral over (z) in Eq. 20 can be replaced by the integral over (z). The equality of the two integrals follows from the fact that for practically all z₁ within , all chain segments will be found inside the surface layer. For chains originating near , say at z₁ = – ( << ), some conformations will cross the z = surface, contributing less than L segments to the surface layer density. By symmetry, however, chains originating at z₁ = + will compensate for the loss of segments from the z₁ = – chains. The near-equivalence of chains originating at z₁ = ± follows from the fact that these chains are hardly affected by the membrane.

Equations 19 and 20 are applicable to the fluid membrane, as well as any quenched membrane state m. For small values of we can use (z₁) = _bq(z₁)/q() with q(z₁)/q() derived from our single-chain simulations. Approximate density profiles for nonzero surface concentrations can be derived by expressing (z) as the product of the probability () to find the cell occupied and the normalized density profile corresponding to one adsorbed molecule. With the aid of Eqs. 17–19 we then find that for z ,

(21)

For z > we must require (z) = _b. Note that this equation applies to the fluid membrane, as well as to any quenched membrane in state m.

The model system
Our model system consists of a single polyelectrolyte interacting with a finite size membrane, large compared to the size of the polymer and the range of intermolecular potentials. With the exception of several limiting test cases, in all simulations we consider polyelectrolyte chains composed of L = 20 spherical segments of diameter d, interacting with a considerably larger two-dimensional membrane cell consisting of M = 2500 lipid headgroups. The lipid membrane is modeled as a perfectly flat and impenetrable two-dimensional hexagonal lattice, with lipid headgroups occupying all its lattice sites. The lattice constant is set equal to d. The membrane may thus be regarded as an hexagonal array of closely packed disks of diameter d, as illustrated schematically in Fig. 1. Using a typical lipid headgroup area of 65 Ų we find d = 8.66 Å. We simulate three-component membranes, composed of electrically neutral (z = 0), monovalent (z = –1), and tetravalent (z = –4) headgroups. These may be regarded as representing, respectively, the phosphatidyl-choline (PC), phosphatidyl-serine (PS), and phosphatidylinositol 4,5 bisphosphate (PIP₂) lipids mentioned in the previous section. The lipid charges are treated as point charges residing at the grid points of the hexagonal lattice, and the electrostatic repulsion between them is modeled in the Debye-Hückel (DH) approximation. Explicitly, the interaction potential between lipids of valences z₁ and z₂ at distance r apart, and in units of k_BT, is

(22)

where l_B = e²/ k_BT is the Bjerrum length, and ^–1 is the Debye screening length; e denotes the elementary charge and is the dielectric constant. In all calculations we use l_B = 7.14 Å, appropriate for water ( = 78) at room temperature, and ^–1 = 10 Å, which corresponds to typical physiological conditions (monovalent ionic strength of 0.1 M). Note that ^–1 is comparable to the other relevant length scale in our system, namely, the distance (d = 8.66 Å) between adjacent lipid charges, as well as between adjacent polymer charges.

In the simulations each polymer bead carries a unit positive charge (z = +1), localized at its center. Although the polymer bond length d is fixed, there are no other restrictions on bond angles, except for those implied by electrostatic and spatial (excluded volume) repulsion between nonbonded segments. For the electrostatic interaction between polymer charges we again use DH potentials. The spatial repulsion is modeled using the shifted and truncated Lennard-Jones potential:

(23)

Note that only the short-range repulsion of the 6:12 Lennard-Jones potential is retained. Setting 2^1/6 = d and ensures the onset of steep repulsion as soon as r falls below d (38).

The electrostatic attraction between the oppositely charged polymer and membrane is also modeled using screened DH potentials. In addition, the membrane surface is treated as an impenetrable wall to the polymer, implying a minimal distance of d/2 between polymer and lipid charges. At this distance the electrostatic attraction between a polymer (z = +1) segment and a monovalent (z = –1) lipid headgroup is 1.07 k_BT. For comparison, the electrostatic repulsion between neighboring monovalent lipids or adjacent polymer beads, taking the distance of closest approach to be r = d, is 0.35 k_BT.

Since the distances between charges in the system are either comparable to or larger than the Debye length, i.e., r d ^–1, screening by counterions is expected to be effective. Under physiological conditions, when ^–1 is small (of the order of few Ångströms), the long-range character of the electrostatic interactions is screened and DH potentials offer a reasonable approximation. These potentials are commonly employed in simulation and theoretical studies of polyelectrolyte-surface interactions (see, e.g., (39,40)).

Henceforth, we shall measure all distances in units of d. Recall also that energies are measured in units of k_BT.

Simulation method
The Rosenbluth MC method (31), or its configurational-bias variant, provides an efficient means for simulating polymer statistics (32). In this approach, chain conformations are generated, segment after segment, with preference for conformations of large statistical weight. Based on these ideas we present below our extension of the Rosenbluth scheme for modeling polyelectrolyte adsorption on fluid, as well as frozen and uniform membranes.

Frozen membrane
Consider first a polymer interacting with a membrane of quenched lipid configuration m. The simulation begins by placing the first chain segment at distance z₁ above the center of the membrane cell, where its interaction energy with membrane lipids is u(z₁,m) (see Fig. 1). We then sample k random directions (and hence positions, r₂) for segment 2 and select one, say with probability where is the interaction energy of segment 2 with segment 1 and the membrane, and is a local partition function. This procedure is continued until all segments of the chain are generated. Repeated applications of this scheme (for the given (m, z₁)) yield an ensemble of conformations { = r₂,...,r_L;z₁,m} with probabilities

(24)

As above, is the total interaction energy of polymer segments with each other and with the membrane. The partition function,

(25)

with w₁ k exp[–u(z₁,m)], is the complete Rosenbluth factor of the polymer-membrane configuration (;z₁,m). Note that W becomes independent of k in the limit k . In our calculations we generally use k = 50. Note also that some of the k vectors pointing from segment l to l + 1 may cross the membrane interface, especially if segment l is near the surface. Their probability, and likewise their contribution to w_l+1 (and hence to W) is zero, reflecting the loss of entropy associated with the presence of the hard membrane wall.

Since every possible conformation is sampled with probability proportional to exp[–U()]/W(), proper Boltzmann averaging requires weighting each by its Rosenbluth factor W(); i.e., the average (over , for the given z₁,m) of any structural or energetic polymer property A is given by

(26)

Note also that the partition function corresponding to all polymer conformations originating at z₁ is

(27)

where it should be stressed that the sum runs over all the generated by the Rosenbluth scheme.

For z₁ > we have q_m(z₁) = q_m() q_b. Averaging A(z₁,m) over all z₁ we obtain the average of A (over all conformations) for molecules adsorbed on a frozen membrane of lipid configuration m,

(28)

Similarly,

(29)

is the partition function of the adsorbed polymer (see Eq. 18).

Fluid membrane
From Eqs. 3 and 6 we know that the thermodynamic and structural properties of a fluid membrane can be modeled based on simulating an ensemble of quenched membranes. However, this procedure is rather indirect and often impractical. Alternatively, adsorption on the fluid membrane could be simulated by combining the Rosenbluth and Metropolis methods. That is, after generating a polymer in conformation p = (;z₁) for a given lipid configuration m, the membrane is allowed to relax to a new configuration m' through a series of Metropolis moves. Another polymer conformation p' can then be generated for m', letting the membrane relax to m'', and so on. The problem here is that the relaxed membrane is no longer the one which served to generate the last polymer conformation. A retracing procedure (32) can be used to improve this scheme, but not fully eliminate its inconsistencies. We have adopted, therefore, an alternative simulation method for the fluid membrane whereby, in the spirit of the Rosenbluth sampling scheme, we generate simultaneously both polymer conformations p and membrane configurations m, as follows.

Any joint polymer-membrane configuration p,m is fully specified by the coordinates of K = L + M^(–1) + M^(–4) particles; that is, L polymer segments, M^(–1) monovalent lipids, and M^(–4) tetravalent lipids (M⁽⁰⁾ = M – M^(–1) – M^(–4) neutral lipids occupy all other membrane sites). We now generate a joint (p,m) configuration by randomly adding either a polymer segment or a charged lipid, until all particles have been placed. More explicitly, suppose the new configuration is already partly grown, consisting of a polymer chain of length l, and a partially charged membrane containing m^(–1) and m^(–4) anionic lipids. One of the remaining (K – l – m^(–1) – m^(–4)) particles is now randomly selected and added to the system. If this is a polymer segment it is added as the (l + 1)^th segment of the chain. As before, this segment is placed in one of k possible positions, with probability exp[–u(l+1;l,m^(–1),m^(–4))]w_l+1, and u(l+1;l,m^(–1),m^(–4)) is the interaction potential of the added particle with all those already placed, and w_l+1 is defined as usual. If the new particle is, say, a monovalent lipid, it is placed with probability in one of n randomly chosen membrane sites, where u(m^(–1) + 1;l,m^(–1),m^(–4)) is the interaction energy of this lipid with the rest of the system, and is the sum of the Boltzmann factors corresponding to the n membrane sites. (In the simulations we usually sample n = 1000 sites, some of which are possibly occupied already and thus do not contribute to w.) This procedure is repeated until all chain segments and all charged lipids are placed, resulting in a statistical distribution of p,m configurations, whose probabilities are

(30)

where

(31)

is the (generalized) Rosenbluth factor of configuration p,m.

As for the quenched membrane, we generally sample many polymer-membrane configurations corresponding to various z₁ values and only then average over this variable. The averaging procedure is analogous, e.g., the average of A for a given z₁ is

(32)

where

(33)

is the partition function introduced in Eq. 16. Similarly,

(34)

	RESULTS AND DISCUSSION

TOP ABSTRACT INTRODUCTION THEORY RESULTS AND DISCUSSION CONCLUDING REMARKS ACKNOWLEDGEMENTS REFERENCES

 
From the simulations we have derived the basic thermodynamic characteristics of macromolecules interacting with fluid, quenched, and uniformly charged membranes. In parallel, for every system considered we have calculated a variety of structural properties, such as the two-dimensional distribution of charged lipids in the membrane plane, or the density profile of chain segments along, as well as perpendicular to, the membrane normal. Two membrane compositions were analyzed in detail:

1. PC:PS:PIP₂ = 98:1:1 membrane, i.e., a membrane containing 98% neutral (z = 0, or PC) lipids, 1% monovalent (z = –1, PS) lipids, and 1% tetravalent (z = –4, PIP₂) lipids.
   
2. PC:PS:PIP₂ = 89:10:1.
   

Note that the average charge, per lipid, corresponding to these membranes (hereafter also referred to as the weakly-charged and the strongly-charged membranes) is and respectively.

For both compositions, simulations were performed for fluid, quenched, and uniformly charged membranes. In the uniformly charged membrane all lipids carry the same partial charge,

We repeat the numerical values of the various parameters in our model: The polyelectrolyte is a freely jointed homopolymer chain composed of L = 20 spherical segments, each carrying a z = +1 charge (see Fig. 1). The membrane cell is an hexagonal array of M = 50 x 50 lipid molecules. The headgroup diameter, is equal to the polymer's bond length. The distance d also marks the onset of steep excluded volume repulsion between nonbonded chain segments (see Eq. 23 where = d/2^1/6 7.72 Å and ). The Bjerrum and Debye lengths are l_B = 7.14 Å and ^–1 = 10 Å, respectively.

For the sake of comparison, we have also performed a limited number of simulations for a stiff (rodlike) polymer, as well as for a weakly charged polymer. Recall that simulations are performed for varying values of the first segment position, z₁, and that corresponds to a free polymer in solution. For the three-dimensional case of a polymer in solution we have also carried out, for comparative reasons, one set of simulations for an electrically neutral polymer.

The number of chain-membrane conformations generated for each z₁ value of a polymer adsorbed on a fluid membrane is 10⁶. The number of chain conformations generated for each z₁ value of a given quenched membrane m is 10³, and the number of membrane configurations is 10⁴. The increments in chain origin positions are z₁ = 1. (Recall that distances are measured in units of d.) The number of possible bond directions when generating polymer conformations is k = 50. The number of possible positions for lipid addition in our simulation scheme of the fluid membrane is n = 1000.

A pictorial illustration of the polymer-membrane configurations generated by our simulations is given in Fig. 2. The figure shows top and side views of two (rather arbitrary) simulation snapshots of a polyelectrolyte interacting with a fluid membrane of composition PC:PS:PIP₂ = 98:1:1. Only part of the membrane is shown, yet it is apparent that the local concentration of charged lipids in the vicinity of the polymer significantly exceeds the membrane average.

View larger version (52K): [in this window] [in a new window]   FIGURE 2  Side and top views of two, rather arbitrary, simulation snapshots (left and right), of a polyelectrolyte interacting with a weakly charged fluid membrane (1% PIP2, and 1% PS). For visual clarity only a section of the membrane is shown, and polymer segments and lipid headgroups are depicted as small spheres, (recall, however, that short range repulsions keep these segments at distance d). PIP2 and PS lipids are represented by blue and purple spheres, respectively. Note the localization of the charged lipids in the vicinity of the polymer.

(35)

The differential adsorption free energy onto the fluid membrane is given by with a similar definition of F_u(z₁). The corresponding free energy change for the quenched membrane is defined here as It should be noted that the net (or integral) adsorption energy of the quenched membrane is not a simple integral of Similarly, the net free energy change of all membranes is not a direct integral of F(z₁). These issues will be clarified after analyzing Figs. 3 and 4 .

View larger version (10K): [in this window] [in a new window]   FIGURE 3  The differential energy of adsorption (a), and free energy of adsorption (b), of a flexible macromolecule adsorbing on a membrane of lipid composition PC:PS:PIP2 = 98:1:1. The value z1 is the distance of the first polymer segment from the membrane plane. The solid, dashed, and dotted curves correspond to the fluid, frozen, and uniformly charged membranes, respectively. The dotted-dashed curve in b is for a weakly charged (z = +1/2) polymer interacting with a fluid membrane. The free energy change corresponding to this polymer is not shown because it very nearly overlaps the dotted curve in a.

View larger version (11K): [in this window] [in a new window]   FIGURE 4  The differential energy of adsorption (a), and free energy of adsorption (b), of a flexible macromolecule interacting with a lipid membrane of composition PC:PS:PIP2 = 89:10:1. The solid, dashed, and dotted curves correspond to the fluid, frozen, and uniformly charged membranes, respectively.

The entropy losses, TS(z₁) = E(z₁) – F(z₁), associated with polyelectrolyte adsorption are quite substantial. In the quenched and uniform membrane cases these entropy losses reflect the lower conformational entropy of the adsorbed molecule, compared to that of a polymer in solution. The entropy loss is even higher, reaching 70% in the case of the fluid membrane, see Figs. 3 and 4. The origin of the enhanced entropy deficit experienced by this membrane is the additional loss of lipid mixing entropy.

From Eq. 13 we know that An analogous equality is also valid for the differential partition functions, q(z₁). That is,

(36)

The last equality here is a reminder that, in the limit of low surface coverage, q(z₁)/q() is equal to the ratio between the density of chain molecules (more precisely, chain termini) at distance z₁ from the membrane, and the corresponding density in the bulk solution. From Eq. 36 we also note that explaining why in Figs. 3 and 4.

In Fig. 5 we show, for our three model membranes, how the partition function (equivalently, the first segment density) ratio defined in Eq. 36 varies with z₁. It should be emphasized that the partition functions corresponding to the quenched and fluid membranes have been obtained using the two different MC simulation schemes described in the previous section. Apart from the small numerical noise, we indeed find that the partition functions corresponding to the fluid and the ensemble of quenched membranes are essentially identical, reassuring that the different simulation methods indeed yield identical results.

View larger version (11K): [in this window] [in a new window]   FIGURE 5  The partition function ratio, for a macromolecule interacting with weakly charged membranes of composition PC:PS:PIP2 = 98:1:1 (a), and strongly charged membranes where PC:PS:PIP2 = 89:10:1 (b). Solid, dashed, and dotted curves correspond to the fluid, quenched, and uniformly charged membranes, respectively.

View larger version (13K): [in this window] [in a new window]   FIGURE 6  Segment density profiles along the membrane normal, (z), relative to the segment density in the bulk solution b = (). The membrane composition is PC:PS:PIP2 = 98:1:1 (a), and PC:PS:PIP2 = 89:10:1 (b). The solid, dashed, and dotted curves correspond to the fluid, quenched, and uniformly charged membranes, respectively. Two curves are shown for each type of membrane; the upper curve corresponds to the low density limit (and hence 0), and the lower one is for

In Fig. 6, we also show representative density profiles corresponding to high surface concentrations of macromolecules (see Eq. 21). Under these conditions we expect different adsorption probabilities on the fluid and quenched membrane. Indeed, for a macromolecule bulk density of Eq. 17 yields _f = 0.5 for the strongly charged fluid membrane, whereas Eq. 18 implies a much smaller surface density for the quenched membrane, Additional values are given in Table 1. Note that the average free energy of adsorption in the ensemble of quenched membranes is zero, indicating that some membrane environments must be repulsive (see below). Also repulsive is the weakly charged uniform membrane, as clearly seen in Fig. 3 b. Indeed, the ratio ), which may be interpreted as the three-dimensional density of macromolecules very near the membrane, is

Given the -values we have calculated, the integral adsorption energies, free energies, and surface concentrations for the fluid, quenched, and uniform membranes. Fig. 7 shows the distributions, P(F) and P(E), of adsorption free energies, F_m, and energies, E_m, for the ensemble of quenched membranes. The adsorption energies are defined here by and their average is The integral adsorption energies for the fluid and uniformly charged membranes are where E(z₁) are the differential adsorption energies shown in Figs. 3 and 4. The adsorption free energies are given by with Their average is In Figs. 3 and 4 we have shown Thus, as noted above, is not simply the integral of The relationship between F(z₁) and F of the fluid membrane is different, namely, with a similar relationship for the uniform membrane.

View larger version (11K): [in this window] [in a new window]   FIGURE 7  Probability distributions of adsorption free energies P(F) (a), and adsorption energies P(E) (b), for a Boltzmann-weighted ensemble of quenched membranes. Solid and dashed curves correspond to membranes with PC:PS:PIP2 = 89:10:1 and PC:PS:PIP2 = 98:1:1, respectively. See Table 1 for more details.