摘要:

Lipidñprotein interactions in the membrane Studies with model peptides Sanjay Mall Ram P. Sharma J. Malcolm East and Anthony G. Lee* Division of Biochemistry and Molecular Biology School of Biological Sciences University of Southampton Southampton UK SO16 7PX Received 27th November 1998 We have used —uorescence quenching of tryptophan-containing trans-membrane peptides by bromine-containing phospholipids to study the speci–city of peptide»lipid interactions. We have synthesized peptides Ac-K where m\7 and n\9 (L16) 2GLmWLnK2A-amide and m\10 and n\12 (L22). Binding constants of L for dioleoylphosphatidylserine 22 [di(C18 1)PS] or dioleoylphosphatidic acid [di(C18 1)PA] relative to dieoleoylphosphatidylcholine [di(C18 1)PC] were close to 1. However for L16 whilst the bulk of the di(C18 1)PA molecules bound with a binding constant relative to di(C18 1)PC close to 1 a small number of di(C18 1)PA molecules bound much more strongly.Assuming just one high affinity binding site on L for anionic lipid the affinity 16 of the site for di(C18 1)PS was calculated to be ca. 8 times that for di(C18 1)PC. The relative binding constant was little aÜected by ionic strength and close contact between the anionic headgroup of di(C18 1)PS and a lysine residue on the peptide was suggested. The relative binding constant for di(C18 1)PS at this high affinity site was less than for di(C18 1)PA. Cholesterol interacts with L with an affinity about 0.7 of that of 22 di(C18 1)PC. The structure of the peptide itself is important. The peptide Ac-KKGYL6WL8YKKA-amide (Y2L14) incorporated into bilayers of dinervonylphosphatidylcholine [di(C24 1)PC] whereas L did not incorporate into this 16 lipid.It is suggested that thinning of a lipid bilayer around a peptide to give optimal hydrophobic matching is less energetically unfavourable when a Tyr residue is located in the lipid/water interfacial region. The membrane-spanning region of an intrinsic membrane protein adopts either an a-helical or a b-sheet structure to ensure formation of the maximum number of intramolecular hydrogen bonds of these structures the a-helix is the most common.1 The transmembrane a-helices of a membrane protein contain largely hydrophobic residues because the solubility of polar residues is low in a non-polar environment such as that of a lipid bilayer ; since the thickness of the hydrophobic core of a bilayer composed of naturally occurring phospholipids is about 30 ” a stretch of about 20 hydrophobic residues is required to span the bilayer.Charged residues are usually found on either side of the hydrophobic stretch of residues the charged residues being located in the interface region of the lipid bilayer interacting either with the polar backbone region of the lipid bilayer or with the charged headgroups or with water. The distribution of hydrophobic residues in the transmembrane region of a membrane protein is non-random.1 The most common amino acid is Leu but aromatic residues particularly Trp and Tyr are often found at the lipid/water interface in both a-helical and b-sheet proteins ;1 it has been suggested that these residues could act as ì—oatsœ with 127 Faraday Discuss.1998 111 127»136 their polar groups facing the water and their non-polar aromatic regions penetrating into the hydrocarbon region serving to –x the helix within the lipid bilayer. The importance of the lipid bilayer for the proper functioning of a membrane protein has been demonstrated for the Ca2`-ATPase of skeletal muscle sarcoplasmic reticulum. A high rate of ATP hydrolysis by the Ca2`-ATPase requires that the fatty acyl chains of the surrounding phospholipids be between C16 and C20 in length ; this presumable ensures that the bilayer thickness matches the hydrophobic thickness of the Ca2`-ATPase.2,3 Outside this chain length range the activity of the ATPase is low and changes on the ATPase can include a decrease in the rate of phosphorylation of the ATPase by ATP and a change in the stoichiometry of Ca2` binding.4,5 High activity also requires that the phospholipids be in the liquid-crystalline bilayer phase; low activities are observed in the gel phase and in the hexagonal H phase.6,7 Finally higher activities II are observed in bilayers of zwitterionic phospholipids such as phosphatidylcholine (PC) or phosphatidylethanolamine (PE) than in bilayers of anionic phospholipids such as phosphatidic acid (PA) or phosphatidylserine (PS).8 The region of the Ca2`-ATPase embedded in the lipid bilayer is a bundle made up of 10 transmembrane a-helices,9,10 and so it must be these a-helices that are sensing the character of the bilayer.To characterize the nature of the interaction between lipid bilayers and a-helices we have turned to simple synthetic peptides of the type Ac-K -G-L -W-Ln-K2-A-amide (Pm`n) consisting 2 m of a long sequence of hydrophobic Leu residues capped at both the N- and C-terminal ends with two polar lysine residues with a centrally located tryptophan residue to act as a —uorescence reporter group.11 It has been shown that peptides of this type (without the tryptophan residue) when mixed with phospholipid form stable a-helices spanning the phospholipid bilayer the ends of the helix being anchored in the phospholipid headgroup region by the charged Lys caps.12,13 A —uorescence quenching method has been used to measure the strength of interaction between the peptides and particular lipids in bilayers in the —uid liquid crystalline phase.14 The peptide is incorporated into bilayers containing the brominated phospholipid dibromostearoylphosphatidylcholine [di(Br C18 0)PC]; di(Br C18 0)PC behaves much like a conventional phospholipid 2 2 with unsaturated fatty acyl chains because the bulky bromine atoms have similar eÜects on lipid packing as a cis double bond.14 In mixtures of brominated and non-brominated phospholipids the degree of quenching of the —uorescence of the tryptophan residue in the peptide is related to the fraction of the surrounding phospholipids which are brominated and thus to the strength of binding of the non-brominated lipid to the peptide.Experimental Dimyristoleoylphosphatidylcholine [di(C14 1)PC] dioleoylphosphatidylcholine [di(C18 1)PC] dierucoylphosphatidylcholine [di(C22 1)PC] dinervonylphosphatidylcholine [di(C24 1)PC] dioleoylphosphatidylserine [di(C18 1)PS] and dioleoylphosphatidic acid [di(C18 1)PA] were obtained from Avanti Polar Lipids.Di(C18 1)PC was brominated to give di(Br C18 0)PC as 2 described by East and Lee,14 and the same procedure was used to brominate di(C18 1)PS to dibromostearoylphosphatidylserine [di(Br C18 0)PS] and di(C18 1)PA to dibromostearoyl- 2 phosphatidic acid [di(Br C18 0)PA]. Cholesterol was brominated to give 5,6-dibromocholestan- 2 3b-ol (dibromocholesterol) as described in Simmonds et al.15 Peptides Ac-KKGL7WL9KKAamide (L16) Ac-KKGL10WL12KKA-amide (L22) Ac-KKGYL6WL8YKKA-amide (Y2L14) and Ac-KKGYL were synthesised using t-Boc chemistry,16 and purity 9WL11YKKA-amide (Y2L20) was con–rmed using electrospray and MALDI-TOF mass spectroscopy.Peptides (20 nmol) were incorporated into phospholipid bilayers by mixing peptide and lipid at a molar ratio of peptide to phospholipid of 1 100 in chloroform»methanol (2 1 v/v). Solvent was removed under vacuum and the mixture was resuspended in buÜer (400 ll) by sonication in a bath sonicator for 5»10 min. Aliquots (100 ll) were diluted into buÜer (2.5 ml; 20 mM Hepes 1 mM EGTA pH 7.2) and —uorescence intensities were recorded at 25 °C using an SLM-Aminco 8000C —uorimeter with excitation and emission wavelengths of 280 and 340 nm respectively. Faraday Discuss. 1998 111 127»136 Theory Tryptophan —uorescence is quenched by bromine-containing molecules by a process of heavy atom quenching which requires contact between tryptophan and bromine.The —uorescence life- 128 time for tryptophan is considerably less than the time for two lipids to exchange position in a bilayer so that quenching can be considered to be a static phenomenon.17 Quenching can then be analyzed in either of two ways. The –rst is a lattice model of quenching.14,18 The degree of quenching in the lattice model is proportional to the probability that a brominated lipid occupies a lattice site close enough to the peptide to cause quenching. For a random distribution of lipids the probability that any lattice site is not occupied by a brominated lipid is 1[xBr where x is Br the mole fraction of brominated lipid in the bilayer.The probability that any particular peptide will give rise to —uorescence is proportional to the probability that none of the n lattice sites close enough to the peptide to cause quenching is occupied by a brominated lipid. Thus (1) F/F0\Fmin](F0[Fmin)(1[xBr)n where F and F are the —uorescence intensities for the peptide in non-brominated and in bro- 0 min minated lipid respectively and F is the —uorescence intensity in the phospholipid mixture when the mole fraction of brominated lipid is xBr . In a hexagonal lattice if quenching can only be caused by immediate neighbours n will equal 6. An alternative description of quenching is the sphere of action model.19 This assumes the existence of a sphere of volume around a —uorophore within which a quencher will cause quenching with a probability of unity.Adapting this model to the two-dimensional case of a biological membrane gives (2) F/F0\Fmin](F0[Fmin)exp([pr2CBr) where pr2 is the area of the quenching circle around each tryptophan and CBr is the concentration of brominated lipid in units of molecules per unit area. Here we have assumed an area of 70 ”2 per lipid molecule. The lattice model can be readily extended to describe quenching of the peptide in a mixture of two lipids of diÜerent affinities for the peptide in the case where the sites around the peptide are equivalent. At each site an equilibrium will exist PL]Q¢PQ]L where PL and PQ are complexes of peptide with non-brominated lipid (L) and brominated lipid (Q) respectively.The equilibrium can be described by an equilibrium constant K given by (3) K\[PQ][L]/[PL][Q] where K is the binding constant of the brominated lipid relative to that of the non-brominated lipid. Fluorescence quenching then –ts to the equation (4) fBr .14 The F/F0\Fmin](F0[Fmin)(1[fBr)n where the fraction of sites at the lipid/peptide interface occupied by brominated lipid is fraction of sites occupied by brominated lipid is related to x by (5) Br fBr\KxBr/(KxBr][1[xBr]) The lattice model is not readily extended to the case where the sites around the peptide are non-equivalent. However the quenching sphere approach can be readily extended to the case where there is a single high affinity binding site for the brominated lipid together with a large number of non-speci–c lipid-binding sites.Binding at the high affinity site can be described by an equation analogous to eqn. (3) giving the probability that the site is not occupied by the brominated lipid as (1[x )/(1[xBr]KxBr). Quenching from the non-speci–c sites can be described by Br eqn. (2) so that F (6) \Fmin](F0[Fmin)A1[ 1 x [xBr exp([pr2CBr) Br]KxBrB F0 Results We synthesized two peptides Ac-KKGL and Ac-KKGL 7WL9KKA-amide (L16) 10WL12KKAcontaining a tryptophan residue in the centre of a hydrophobic domain composed of amide (L22) Faraday Discuss. 1998 111 127»136 129 Leu residues and a corresponding pair of peptides Ac-KKGYL6WL8YKKA-amide (Y2L14) and Ac-KKGYL WYL KKA-amide (Y2L20) in which the Leu residues at each end of the hydropho- 9 11 bic stretch have been replaced by Tyr residues.Peptides were incorporated into lipid bilayers at a molar ratio of peptide lipid of 1 100 by mixing peptide and lipid in organic solvent followed by removal of the solvent and hydration of the mixture. The —uorescence emission spectrum of the tryptophan residue is environmentally sensitive the emission maximum moving to shorter wavelength with decreasing environmental polarity.19 The emission spectra of all the peptides incorporated into bilayers of lipids containing oleoyl chains [di(C18 1)PC di(C18 1)PA or di(C18 1)PS] are centred at about 323 nm indicating a very hydrophobic environment for the tryptophan consistent with the expected localization in the middle of the bilayer.Fluorescence quenching 16 and di(C18 1)PC decreases with increasing content of di(Br C18 0)PC and –ts 2 the data in either di(Br C18 0)PC/di(C18 1)PC and L The —uorescence intensity for the peptide L incorporated into bilayers of di(Br C18 0)PC is 16 2 about 5% of that in di(C18 1)PC demonstrating highly efficient quenching of the tryptophan by the bromine-containing fatty acyl chains (Fig. 1). The —uorescence intensity for L in mixtures of di(Br C18 0)PC 2 2 to eqn. (1) with a value of n the number of ì sites œ around the peptide where binding can result in quenching of 3.4 ; the same value of n –ts the quenching data in mixtures of di(Br C18 0)PS and di(C18 1)PS (Fig. 1). For the longer peptide L22 2 or di(Br C18 0)PS/di(C18 1)PS again –t to eqn.(1) but now with a value of n of 2.1. These 2 results suggest that the peptides L adopt diÜerent structures in the lipid bilayer and 16 22 di(Br C18 0)PC 2 (L) or 2 22 Fig. 1 Fluorescence intensities for L and L22 in mixtures containing brominated phospholipids. L (A) or 16 L 16 (B) were incorporated into mixtures of di(Br C18 0)PS and di(C18 1)PC and di(C18 1)PS (K) at a molar ratio of peptide phospholipid of 1 100. Fluorescence intensities are expressed as a fraction of that recorded for peptide in di(C18 1)PC (A) or di(C18 1)PS (B). The solid and broken lines show best –ts of the data to eqn. (1) and (2) respectively with the parameters given in the text. Faraday Discuss. 1998 111 127»136 130 16 ”; L ref. 20) the hydrophobic length of is greater (36 ”) L so that will be 22 16 indeed whereas the hydrophobic length of L ”) is close to the thickness of a bilayer of (27 di(C18 1)PC (29.8 22 tilted in the bilayer.11 As shown in Fig.1 the quenching data also –t to the circle of quenching model eqn. (2) with values for the quenching radius ” r of 9 and 8 respectively for L and L22 . Relative binding affinities of anionic and zwitterionic phospholipids 2 di(Br C18 0)PC. If diÜerent relative binding constants are obtained in mixtures of PC and PS or PC and Fitting the —uorescence quenching curve for a peptide in a mixture of di(Br C18 0)PC and lipid X to eqn. (4) gives the peptide binding constant for lipid X relative to that for 2 all the binding sites on the peptide are equivalent then –tting the —uorescence quenching curve for the peptide in a mixture of di(C18 1)PC and brominated lipid X should give the same relative lipid binding constant assuming that bromination of the chains does not signi–cantly aÜect interaction with the peptide.The quenching curve for L in mixtures of di(Br C18 0)PS and 22 2 di(C18 1)PC (Fig. 2) in buÜer of low ionic strength –ts to eqn. (4) with a binding constant for PS relative to PC of 1.3^0.4 showing no signi–cant selectivity in binding. Consistent with this interpretation the quenching curves for L in mixtures of di(Br C18 0)PC and di(C18 1)PS –t 22 2 to eqn. (4) with a binding constant for PS relative to PC of 0.8^0.1 (Fig. 2). Similarly relative binding constants for PS and PC are close to 1 in media of high ionic strength and relative binding constants for PA and PC are also close to 1 (Table 1).In contrast to these results with L22 L16 for Fig. 2 Fluorescence intensities for L in mixtures of PC and PS. The experimental points show —uorescence 22 (L) or intensities in mixtures of di(Br C18 0)PS and di(C18 1)PC (K) as di(Br C18 0)PC and di(C18 1)PS 2 a function of the mole fraction of PS. The buÜer was 20 mM Hepes 1 mM EGTA pH 7.2. The solid lines 2 show –ts to eqn. (4) with the relative binding constants given in Table 1. Table 1 Relative binding constants for peptide L22 Lipid binding constant relative to PCa High saltc Low saltb Lipid mixture di(C18 1)PS»di(Br C18 0)PC di(C18 1)PC»di(Br2 2 C18 0)PS di(C18 1)PA»di(Br C18 0)PC di(C18 1)PC»di(Br C18 0)PA 1.3^0.4 0.8^0.1 » 1.6^0.1 0.8^0.4 0.8^0.2 1.6^0.3 0.8^0.2 2 2 a Concentrations of phospholipids expressed as mole fractions.b 20 mM Hepes 1 mM EGTA pH 7.2. c 20 mM Hepes 1 mM EGTA pH 7.2 300 mM KCl. Faraday Discuss. 1998 111 127»136 131 Fig. 3 Fluorescence intensity for L in mixtures of PC and PS. The experimental points show —uorescence 16 (L) or intensities in mixtures of di(Br C18 0)PS and di(C18 1)PC (K) as di(Br C18 0)PC and di(C18 1)PS 2 a function of the mole fraction of PS. The buÜer was 20 mM Hepes 1 mM EGTA pH 7.2. The solid lines 2 show –ts to eqn. (4) with the relative binding constants given in Table 2 and the broken lines show –ts to eqn. (6) with the parameters given in Table 3. PA depending on whether the brominated lipid is the phosphatidylcholine or the anionic phospholipid (Fig.3). The large relative binding constant for the anionic phospholipid derived from experiments in which the anionic phospholipid is brominated together with a relative binding constant close to 1 derived from experiments in which the phosphatidylcholine is brominated suggests that the anionic phospholipids bind to a small number of sites with high affinity on the peptide whereas the zwitterionic phospholipid binds to a large number of non-speci–c sites. This Table 2 Relative binding constants for peptide L16 Lipid binding constant relative to PCa High saltc Low saltb Lipid mixture di(C18 1)PS»di(Br C18 0)PC di(C18 1)PC»di(Br2 2 C18 0)PS di(C18 1)PA»di(Br C18 0)PC di(C18 1)PC»di(Br C18 0)PA 1.4^0.1 1.7^0.1 1.1^0.2 4.2^0.4 0.9^0.3 3.0^0.3 1.1^0.1 4.2^0.1 2 2 a Concentrations of phospholipids expressed as mole fractions.b 20 mM Hepes 1 mM EGTA pH 7.2. c 20 mM Hepes 1 mM EGTA pH 7.2 300 mM KCl. Table 3 Relative binding constants for anionic phospholipids for peptide L16 Binding constant relative to di(C18 1)PCa High saltc Low saltb Lipid mixture di(Br2C18 0)PS»di(C18 1)PC C18 0)PA»di(C18 1)PC 8.6^1.4 2.9^0.4 23^3 21^4 di(Br2 a Concentrations of phospholipids expressed as mole fractions. b 20 mM Hepes 1 mM EGTA pH 7.2. c 20 mM Hepes 1 mM EGTA pH 7.2 300 mM KCl. Faraday Discuss. 1998 111 127»136 132 (L) ( or dibromocholesterol and di(C18 1)PC K) and —uorescence is Fig.4 Fluorescence intensities for peptide L22 in mixtures containing cholesterol. The peptide was mixed with di(Br C18 0)PC and cholesterol plotted as a function of the mole fraction of the brominated component. The solid lines show the best –t to the 2 data with n\2.1 with a relative binding constant K for cholesterol of 0.7 for mixtures of di(Br C18 0)PC 2 and cholesterol and with n\6 and a relative binding constant K for cholesterol of 1.4 for mixtures of di(C18 1)PC and dibromocholesterol. 2L14 in (a) di(C14 1)PC; (b) di(C18 1)PC; (c) di(C22 1)PC; (d) di(C24 1)PC. (B) shows —uoand Y2L20 in lipid bilayers. (A) shows —uorescence emission 2L20 Fig. 5 Fluorescence emission spectra for Y spectra for Y rescence emission spectra for Y 2L14 in solid line di(C14 1)PC; broken line di(C18 1)PC dash-dot line di(C22 1)PC; dotted line di(C24 1)PC.133 Faraday Discuss. 1998 111 127»136 situation is better analyzed in terms of the quenching circle approach. The data shown in Fig. 3 can be –tted to eqn. (6) assuming a single high affinity site on the peptide for anionic phospholipid with the relative binding constants for anionic phospholipids given in Table 3. Interaction with cholesterol We can also use these —uorescence quenching methods to study interactions between peptides and cholesterol in the membrane. Fig. 4 shows that the tryptophan —uorescence emission of the peptide L is quenched by dibromocholesterol in bilayers of di(C18 1)PC so that dibromocholes- 22 terol must be able to interact with the peptide.Conversely cholesterol can displace di(Br C18 0)PC from around the peptide as shown by the increase in —uorescence intensity with 2 increasing cholesterol content again showing that cholesterol can interact with the peptide. The experimental quenching curves for L C18 0)PC»cholesterol mixtures can be –tted 22 in di(Br2 to eqn. (4) with n\2.1 and a relative binding constant [peptide.cholesterol]/ [peptide].[di(C18 1)PC] of 0.7. This same relative binding constant also –ts the data for L in 22 di(C18 1)PC»dibromocholesterol mixtures assuming a value for n for dibromocholesterol of 6.0 suggesting a hexagonal packing arrangement for cholesterol around the peptide. EÜects of Tyr residues on hydrophobic matching incorporated fully into bilayers of phosphadid not incorporate at all into a 16 L16 combined with a shift to longer wavelength.11 Thus whilst is completely excluded In a previous paper we showed that whereas L22 tidylcholines with chain lengths between C14 and C24 L16 bilayer of di(C24 1)PC and only partly incorporated into a bilayer of di(C22 1)PC.11 The level of incorporation of the peptide can be quantitated from the —uoresence spectra since tryptophan —uorescence intensities are very low for unincorporated peptide due to the formation of non- —uorescent aggregates in water.11 Fig.5B shows the —uorescence spectra of Y2L20 in phosphatidylcholines of chain length between C14 and C24. The very similar —uorescence intensities observed for the peptide in all bilayers indicate full insertion into these bilayers ; the higher wavelength of the emission maximum in di(C14 1)PC than in the other lipids suggests that the tryptophan residue is buried less deeply in bilayers of di(C14 1)PC than in thicker lipid bilayers.For Y2L14 although the —uorescence intensity in di(C24 1)PC is less than that in di(C14 1)PC the decrease in —uorescence intensity is only 40% (Fig. 5A) compared to an 85% decrease in intensity for L from a bilayer of di(C24 1)PC Y2L14 is at least partially incorporated. was fully incorporated into bilayers of phosfailed to incorporate normally into a ” ” calculated using a helix translation of 1.5 per residue and a hydrophobic will match the hydrophobic Discussion In a previous paper11 we showed that whereas L22 phatidylcholines irrespective of lipid chain length L16 bilayer of di(C24 1)PC and was only partly incorporated into a bilayer of di(C22 1)PC.If it is assumed that the peptide adopts an ideal a-helical structure then the hydrophobic length of L16 will be about 27 stretch of 18 residues in total. Thus the hydrophobic length of L16 thickness of a bilayer of di(C16 1)PC or di(C18 1)PC. In bilayers of sub-optimal thickness signi –cant changes in the a-helical structures of the peptides are unlikely because of the stability of the a-helix an expectation con–rmed in infrared studies of peptides of this type.21,22 There are then two mechanisms which would allow matching of a thin bilayer to a long peptide the fatty acyl chains could stretch increasing the thickness of the bilayer or the peptide could tilt away from the direction of the bilayer normal reducing its eÜective length across the bilayer.However when the hydrophobic thickness of the bilayer is greater than that of the peptide there is only one way to achieve matching and that is by compression of the fatty acyl chains. The observations with L suggest that any stretching or compression of the fatty acyl chains is rather limited and 16 becomes energetically unfavourable when the hydrophobic thickness of the bilayer exceeds the hydrophobic length of the peptide by more than about 10 ”.11 The peptide L22 however incorporates normally into all bilayers with chain lengths in the range C14 to C24 and presumably any mismatch is accommodated by tilting the peptide. There will of course be an energetic cost Faraday Discuss.1998 111 127»136 134 associated with stretching or compressing the lipid fatty acyl chains which will be re—ected in values of relative lipid binding constants. Thus strongest binding of lipid to L is observed for 16 di(C18 1)PC and for L strongest binding is observed with di(C22 1)PC as expected for optimal 22 matching.11 16 Here we have explored the importance of charge eÜects on the interaction between anionic phospholipids and the positively charged peptides. For the peptide L anionic phospholipids were found to bind more strongly than the zwitterionic phosphatidylcholine (Table 3). The binding constant for PS relative to PC in a medium of low ionic strength was found to be 8.6 (in mole fraction units) corresponding to a diÜerence in unitary binding energies of [5.3 kJ mol~1.At pH 7.2 PS bears a single negative charge.23 The binding constant for PS changes little with ionic strength (Table 3) suggesting that the interaction with the positively charged peptide does not follow simply from a high positive potential in the vicinity of the positively charged Lys residues on the peptide increasing the local concentration of anionic phospholipid. The energy of interaction between two ions U is given by (7) U\z1 z2 e2/4%e0 er r z e and z are the charges on the two ions is the relative permittivity (dielectric constant) 2 r where 1 of the medium and r is the distance between the two ions. Assuming a dielectric constant of 78.5 (water) an energy of interaction of 5.3 kJ mol~1 corresponds to a distance of separation between two monovalent ions of 3.3 ”.This therefore suggests that strong interaction requires the anionic headgroup of PS to be in close contact with one of the Lys residues on the peptide. Once this strong interaction with a single PS molecule has been made other PS molecules will then interact with L relatively non-speci–cally with a binding constant relative to PC close to 1. This picture 16 is consistent with results of a molecular dynamics simulation of the individual a-helices of bacteriorhodopsin in bilayers of di(C14 0)PC carried out by Woolf.24 This showed that a small proportion of the lipid molecules interacted with the a-helices much more strongly than the others and that these strong interactions were dominated by electrostatic terms rather than van der Waals terms.24 The relative binding constants for PS are less than for PA and are more sensitive to ionic strength (Table 3).For phosphatidylserine the presence of the positively charged ammonium group as well as the negatively charged carboxyl group in the headgroup region may reduce interaction with the positively charged peptide. 16 L22 In contrast to L the binding constants for anionic phospholipids to are very similar to those for zwitterionic phospholipids with a relative binding constant close to 1 (Table 1). In principle it is possible that a high affinity binding site for anionic phospholipids exists on L but 22 that binding of a brominated anionic phospholipid to this site does not result in efficient quenching of tryptophan —uorescence.However the considerable —exibility of the fatty acyl chains in the liquid crystalline phase make this rather unlikely. It therefore appears that tilting of L in the 22 bilayer necessary to match the hydrophobic length of L to the hydrophobic thickness of a 22 bilayer of di(C18 1)PC locates the Lys residues on the peptide too far from the lipid headgroup region to allow a strong interaction between the anionic phospholipid and the peptide. In general binding constants for phospholipids to membrane proteins show relatively little selectivity for anionic phospholipids. For example binding constants for PA and PS relative to PC are close to 1 for the Ca2`-ATPase8 and for the (Na`-K`)-ATPase binding constants for PA and PS are about twice those for PC.25 However there is evidence for the presence of a small number of special phospholipids binding to some membrane proteins acting as ìcofactorsœ.An example is provided by cytochrome c oxidase whose crystal structure shows the presence of a lipid molecule bound between the transmembrane a-helices.26 Speci–c high affinity binding sites for anionic phospholipids on a membrane protein could involve close interaction between the anionic headgroup and a positively charged residue on the protein as suggested by the results presented here. The studies with cholesterol shown in Fig. 4 show that cholesterol binds to the peptide with a binding constant only a factor of about 2 less strongly than di(C18 1)PC. This is rather surprising given the relatively rigid structure of the steroid ring of cholesterol and the molecularly rough surface of the peptide.In other studies we have shown that cholesterol binds relatively weakly at the lipid/protein interface of the ATPase;15,27 comparison with the peptide studies reported here 135 Faraday Discuss. 1998 111 127»136 suggests that weak binding of cholesterol to the ATPase involves interactions in the lipid headgroup region rather than interactions between the sterol ring and the hydrophobic transmembrane a-helices. Aromatic residues particularly Trp and Tyr are often found in membrane proteins at the lipid/ water interface and it has been suggested that these residues could act as ì—oatsœ serving to –x the helix within the lipid bilayer.1 This is consistent with the results shown in Fig.5A for the peptide Y does not incorporate into bilayer of di(C24 )PC,11 Y2L14 does (Fig. 5A). 2L14. L16 Whereas Incorporation of a short peptide into a thick bilayer requires dimpling of the lipid bilayer around the protein ; the results shown in Fig. 5 suggest that this distortion of the lipid bilayer is energetically more favourable when Tyr residues are located close to the lipid/water interface. Acknowledgements We thank Dr R. J. Webb for helpful discussions and the BBSRC for –nancial support. Paper 8/09299K References 1 D. C. Rees A. J. Chirino K. H. Kim and H. Komiya in Membrane Protein Structure ed. S. H. White OUP New York 1994 p. 3. 2 A. G. Lee K. A. Dalton R. C. Duggleby J. M. East and A. P. Starling Biosci.Rep. 1995 15 289. 3 A. G. Lee Biochim. Biophys. Acta 1998 1376 381. 4 F. Michelangeli E. A. Grimes J. M. East and A. G. Lee Biochemistry 1991 30 342. 5 A. P. Starling J. M. East and A. G. Lee Biochem. J. 1995 310 875. 6 A. P. Starling J. M. East and A. G. Lee Biochemistry 1995 34 3084. 7 A. P. Starling K. A. Dalton J. M. East S. Oliver and A. G. Lee Biochem. J. 1996 320 309. 8 K. A. Dalton J. M. East S. Mall S. Oliver A. P. Starling and A. G. Lee Biochem. J. 1998 329 637. 9 C. J. Brandl N. M. Green B. Korczak and D. H. MacLennan Cell 1986 44 597. 10 A. G. Lee in Biomembranes V olume 5. T he AT Pases. ed. A. G. Lee JAI Press Greenwich Connecticut 1996 p. 1. 11 R. J. Webb J. M. East R. P. Sharma and A. G. Lee Biochemistry 1998 37 673. 12 J.C. Huschilt B. M. Millman and J. H. Davis Biochim. Biophys. Acta 1989 979 139. 13 F. A. Nezil and M. Bloom Biophys. J. 1992 61 1176. 14 J. M. East and A. G. Lee Biochemistry 1982 21 4144. 15 A. C. Simmonds J. M. East O. T. Jones E. K. Rooney J. McWhirter and A. G. Lee Biochim. Biophys. Acta 1982 693 398. 16 E. Atherton and R. C. Sheppard Solid phase peptide synthesis a practical approach IRL Press Oxford 1989. 17 J. M. East D. Melville and A. G. Lee Biochemistry 1985 24 2615. 18 E. London and G. W. Feigenson Biochemistry 1981 20 1932. 19 J. R. Lakowicz Principles of Fluorescence Spectroscopy Plenum Press New York 1983. 20 B. A. Lewis and D. M. Engelman J. Mol. Biol. 1983 166 211. 21 Y. P. Zhang R. N. A. H. Lewis R. S. Hodges and R. N. McElhaney Biochemistry 1992 31 11579. 22 Y. P. Zhang R. N. A. H. Lewis R. S. Hodges and R. N. McElhaney Biophys. J. 1995 68 847. 23 G. Cevc Biochim. Biophys. Acta 1990 1031 311. 24 T. B. Woolf Biophys. J. 1998 74 115. 25 M. Esmann and D. Marsh Biochemistry 1985 24 3572. 26 S. Iwata C. Ostermeier B. Ludwig and H. Michel Nature 1995 376 660. 27 J. Ding A. P. Starling J. M. East and A. G. Lee Biochemistry 1994 33 4974. Faraday Discuss. 1998 111 127»136 136

ISSN:1359-6640    

DOI:10.1039/a809299k

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

General Discussion Dr Templer opened the discussion of Dr Leeœs paper Are you able to tell us whether or not the ììbilayer œœ is moved sideways by the adhesion of the AFM tip to the top layer of DOPE in the —uid phase? Dr Lee responded The fact that the lipid bilayers (and monolayers) can be scanned repeatedly with the AFM probe without changing the –lm morphology indicates that we are not irreversibly deforming them. However there is a signi–cant increase in the lateral force measured by the cantilever as the probe passes from the solid lipid domains to the DOPE domains which we associate with an increase in friction.1 The increased friction on the DOPE phase may result from an increased probe»surface contact area or viscous dissipation. 1 Y. F. Dufre� ne W. R.Barger J-B. D. Green and G. U. Lee L angmuir 1997 13 4779. Prof. Klein commented MD calculations can perhaps help explain some of your observations. Unpublished calculations by Tarek et al.1 have examined the behaviour of lipid monolayers and bilayers supported on SAMs. These calculations –nd that the supported lipid in the L -phase is rather mobile. If this –nding is correct the AFM tip may indeed displace the liquid-like outer –lm. In the gel-like phase this situation does obtain. As expected the gel-like bilayer has crystalline alkane chains. a 1 M. Tarek and D. J. Tobias Molecular dynamics simulation of the glassy behaviour in globular proteins 1999 to be published. Dr Lee responded Molecular dynamics simulations have provided powerful insight into the nature of the AFM probe»surface interaction in UHV.I think it would be extremely interesting to simulate a probe indenting into a supported lipid bilayer in water. Prof. Barclay commented When cells come into contact there is normally a clear gap between opposing lipid bilayers as proteins interact. I can see the lipid interactions being important in the membrane fusion events but are they important in adhesion events as suggested ? Dr Lee responded You are absolutely correct in noting that the lipids we have used in this study are more biologically relevant to membrane fusion than cell adhesion. However this work demonstrates that the AFM can be used to map the surface forces chemistry and mechanics of multicomponent lipid systems on the nanometer scale and thus has set the stage for studies of more complex systems.For example we are preparing a manuscript that describes the interaction of several GM1s with hydroxy carboxyl and methyl functionalized probes. Prof. Tiddy asked In Fig. 8B the chains are drawn almost straight. What is the consequence of the double bond? How much hydrophobic contact is there between the DOPE chains in the contact region and water? Do you know the state of hydration of the headgroups of chains in contact with the surface and how this diÜers from a normal layer in water? Dr Lee responded One would expect the DOPE double-bond to disorganise the lipid chains. It is clear that the unsaturated DOPE phase is thinner than the saturated phases but we really have very little direct information about the molecular structure of the –lm.The weakly repulsive surface forces on DOPE suggests the –lm naturally has a low density of hydrophilic groups at the solid/lipid interface (or a higher density of hydrophobic groups). As the probe comes into contact 137 Faraday Discuss. 1998 111 137»157 with the –lm one would expect that the lipid headgroups would be displaced from their equilibrium conformation to present hydrophobic groups to the probe/lipid interface. The level of perturbation of the –lm should vary across the probe contact area. However we really do not have any direct evidence about the molecular structure of the probe/surface interface and this is why it would be great to have a MD simulation of this system. Dr L. Fisher asked Could you clarify whether your experiments were performed with the specimen and probe tip fully immersed in aqueous solution ? If in aqueous solution what were the Debye lengths ? Are these signi–cant in the context of your experimental results ? Dr Lee responded The bilayer and alkanethiol monolayer studies were performed with both the probe and surface immersed in water or 0.15 M NaCl.The Debye length in the triply distilled water is large but electrostatic eÜects were ruled out on these nominally uncharged surfaces using the fairly high concentration salt solutions. The study of the mixed lipid monolayer was performed in air. Prof. Roux said The probe is coated by a SAM of C OH. Is there some degree of entangle- 16 ment between the long chains of the probe and the chains of the surface you are trying to probe? Would it make a diÜerence to use a coating with shorter chains ? Dr Lee responded The physical properties of the SAM –lm are dependent on the length of the alkane chain and for saturated alkanethiols the transition from solid to liquid behaviour takes place somewhere between C and C12.C16OH The SAM is a highly ordered solid phase and its 6 elastic modulus is signi–cantly higher than that of the lipid bilayer. We have also measured the surface forces between shorter alkane thiol –lms (C and C11) and found a signi–cant diÜerence in 6 the surface forces. Prof. Evans asked Could you explain what thickness you expect for a monolayer»including estimates for headgroup and chain dimensions»and how this compares with data from X-ray diÜraction measurements? Dr Lee responded The thickness of a lipid monolayer is of course dependent on the deposition conditions and the speci–c lipids used.Theoretical estimates and ellipsometric measurements of the thickness of the lipid –lms used in this study are presented in Table 2 in our paper. In order to relate molecular dimensions to monolayer thickness one must know the orientation of each part of the lipid in the –lm. It has been established that the maximum chain length of saturated lipids in 0.154]0.1265](number of carbons) nm.1 For DSPE and DGDG we can expect the carbon chain length to be 2.4 nm which would make the chain contribution to the –lm thickness 2.1 nm if the molecule is tilted 30° to the surface normal. The chain thickness of unsaturated DOPE is expected to be signi–cantly less than the saturated lipids due to the disorder the cis double bond introduces into the –lm.Experimental measurements of chain and headgroup contribution to –lm thickness come from X-ray or neutron diÜraction studies. Unfortunately most of this work has been done on suspensions of multiwalled vesicles so it does not exactly correlate with our work on surfaces. It should be noted however that there are a growing number of studies of lipid –lms on surfaces using the extremely powerful techniques of small angle X-ray scattering (SAXS) and grazing-incidence X-ray diÜraction (GIXD). Both phosphatidylethanolamine2 and DGDG3 vesicle systems have been studied by X-ray diffraction. The repeat distance of the DSPE bilayers is approximately 6.4 nm which is in reasonable agreement with our ellipsometric DSPE monolayer thicknesses of 3.3^0.1 nm.It is not possible to directly compare the AFM and X-ray results for DOPE because the liquid-like isotherm of this material makes thickness highly dependent on surface pressure. McDaniel suggests that the DGDG headgroup is oriented parallel to the plane of a lipid bilayer forming a tightly packed 0.8 nm layer. If we add the X-ray DGDG headgroup thickness to the calculated chain thickness the total thickness of the monolayer becomes 2.9 nm which again agrees with our ellipsometric measurements. In conclusion I think that the thicknesses we have Faraday Discuss. 1998 111 137»157 138 measured with ellipsometry are in fairly good agreement with X-ray diÜraction supporting our interpretation of the AFM images and force curves.1 C. Tanford T he Hydrophobic EÜect Wiley New York 1980; J. N. Israelaivili D. J. Mitchell and B. W. Ninham J. Chem. Soc. Faraday T rans. 1976 72 1525. 2 J. M. Seddon G. Cevc R. D. Kaye and D. Marsh Biochemistry 1984 23 2634. 3 R. V. McDaniel Biochem. Biophys. Acta 1988 940 158. Dr Chen asked AFM is a powerful tool to study the surface properties of the membrane and lipid peroxidation is a big problem of the biomembrane. Have you done some work related to lipid peroxidation of the membrane with AFM? Or do you have an opinion on it ? Dr Lee responded No. I havenœt done any of this work. I do not have any opinion on this. Prof. Holzwarth asked Do you have any hard evidence about the smoothness of the tip used during your AFM experiments for example an inspection by electron microscopy? Did you try to measure at diÜerent temperatures especially in the temperature range of a phase transition to see diÜerent states of order coexisting as claimed by my group.1 1 R.Groll A. Boé ttcher J. Jaé ger and J. F. Holzwarth Biophys. Chem. 1996 58 53. Dr Lee responded The AFM probes have been examined with FI-SEM and SrTiO calibration 3 gratings.1 SEM con–rmed that the microfabricated probes were pyramidal in shape and had a radius of curvature \100 nm. The SrTiO analysis was not performed on the probes used in this 3 paper but in subsequent studies these calibration gratings have been used to characterize the radius of curvature of each probe used.The conclusion of our recent work is that the force curves presented in our paper are consistent with the measurements made with spherical D50 nm radius probes. We have not made measurements on the LB –lms at diÜerent temperatures. In principle AFM can be performed at the solid/water interface at temperatures between 0 and 100 °C but in practice it is not easy to obtain high-resolution images due to the in—uence of thermal drift. However the experiment you are suggesting is tractable and would identify the diÜerent states of order. 1 S. S. Sheiko M. Moé ller E. M. C. M. Reuvekamp and H. W. Zandbergen Phys. Rev. B 1993 48 5675. Dr Pawlak asked You investigated membrane surfaces using de–ned hydrophilic tip/ hydrophilic surface and hydrophobic tip/hydrophilic surface contacts on —at mica.Are there differences in the surface roughness of these two measurements? In the case of hydrophobic/hydrophilic contacts would it be possible that hydration forces or a diÜerent water structure in close vicinity of the hydrophobic surface generate an apparent increase of surface roughness in topographical images which is not related to the real surface roughness of the support? We have actually measured such an increase (in the range of 1 nm rms) during imaging of very hydrophobic self-assembled layers of C -alkane on metal-oxide using Si3N4 tips. The 16 roughness increase was not observed when comparing clean hydrophilic substrate surfaces and surfaces with a terminating monolayer of lipid.Dr Lee responded In this manuscript we describe force measurements in aqueous solutions between (i) a gold coated CH -OH alkanethiol functionalized probe and alkanethiol functional- 16 ized surfaces (CH -OH and CH -CH3) ; (ii) a gold coated CH -OH alkanethiol functionalized 16 15 16 probe and lipid –lms (DSPE/DOPE and DGDG/DOPE). We also analyze the contact imaging mechanism of the CH -OH tip on the lipid –lms in water and air. 16 I think the issue that you are trying to get at is Can AFM measure surface hydration at the nanometer scale and if so can this information be used to understand the chemical homogeneity of a surface ? Our work on these model systems clearly shows that local hydration mechanical and topographical properties each contribute to the AFM contrast and changes in any one of these factors will produce a change in the observed rms roughness in an AFM image.However I do not advocate the using of surface roughness as a means of characterizing surface homogeneity because the AFM imaging mechanism is dependent on several other factors e.g. probe radius and feedback parameters. Rather friction phase and force modes can be used in parallel with imaging modes to identify nanometer scale surface inhomogeneties. The power in using surface force 139 Faraday Discuss. 1998 111 137»157 mapping is that the force curves can be used to semi-quantitatively interpret the contribution of each imaging factor. Dr Sansom opened the discussion of Prof. Smithœs paper In the graph (Fig. 4) of displacement vs. residue can one see (a) correlations between residues with high displacement and structure (loop vs.helix) of the protein ; and (b) correlations between those residues which give high displacement at D150 K and those which give high displacement at D250 K? Prof. Smith responded Fig. 4 indicates that transitions take place at D150 and D240 K. This result is found to be reproducible with the present simulation model and diÜerent initial conditions. However I feel that the model may be too crude to permit meaningful investigation of the correlations you referring to. We need to repeat the calculations with bacteriorhodopsin in its trimeric form and in an explicit membrane. Prof. Holzwarth asked Did your simulations include the special lipids which are strongly bound to bacteriorhodopsin (BR) and are located between the helical structures ? Those lipids are still mysterious in respect of their function.Did your simulations account for lipids which are arranged outside the helical structure of BR especially how far the in—uence of BR might reach into its surroundings? We tried to measure the in—uence of BR on its lipid environment and found that this protein in—uences as many as –ve to six lipid layers around itself.1 1 A.Boé ttcher N. Dencher R. Groll F. Meyer and J. F. Holzwarth Reactions in Compartmentalized L iquids ed. W. Knoche and R. Schomaé cker Springer-Verlag Berlin 1989 pp. 105»115. Prof. Smith responded Our simulations did not explicitly include any lipids. It is something we would hope to do in the future.The in—uence of a protein on its lipid environment is something that molecular simulation is indeed beginning to explore as evidenced by other papers in the Discussion. With Dan Mihailescu we are also beginning to explore this for a smaller lipid/ membrane system gramicidin S and see a signi–cant change in the order parameters of peptideassociated membrane lipids. Prof. Finney said I would be interested in your comments on Fig. 3. (1) The transition identi- –ed at 150 K looks a little diÜerent from that measured experimentally by Reç at et al.1 Rather than showing an increase in dSu2T/dT between two regions the simulated transition depends on a previous almost horizontal plateau region. Do you have any comments on this ? (2) Have you looked at the processes that give rise to this plateau region ? What eÜects prevent the normallyexpected increase of Su2T with T ? (3) dSu2T/dT before the plateau is greater than dSu2T/dT after 150 K.What does this imply with respect to changes in the —exibility behaviour? 1 V. Reç at H. Patzelt M. Ferrand C. P–ster D. Oesterhelt and G. Zaccai Proc. Natl. Acad. Sci. U.S.A. 1998 95 4970. Prof. Smith responded The present work demonstrates anharmonic transitions in the protein simulation at D150 and D240 K similar to the temperatures found experimentally by Reç at et al.1 Beyond that fact it is unclear to what extent experiment and simulation resemble each other. Direct comparison of the Su2T data indeed show some signi–cant diÜerences. Further information may be obtained from a more detailed simulation model and direct experiment»theory comparison at the level of the dynamic structure factor.To answer your second point a clue to this might be seen in Fig. 4. At the transition some residues increase Su2T and others decrease it. For the latter the transition may have involved moving into a steeper-walled potential. The run eÜect can conceivably average to a plateau. To answer your third point I am not sure that the gradient diÜerence is signi–cant. If it were it would imply a higher ìaverage force constantœ above 150 K such that on average the atoms move in a steeper-walled region of the protein potential surface. 1 V. Reç at H. Patzelt M. Ferrand C. P–ster D. Oesterhelt and G. Zaccai Proc. Natl. Acad. Sci. U.S.A. 1998 95 4970.Prof. Klein asked Neutron scattering data provide a powerful probe of protein dynamics. However it is important to calculate the full scattering function from the MD simulation and to treat the MD data in the same fashion as experimentalists treat their data. In the case of the Faraday Discuss. 1998 111 137»157 140 soluble protein BPTI a large ììdiscrepancyœœ existed between the calculated and measured values of mean square displacements. This ììproblemœœ is at least partly due to intrinsic limitation on the instrumental resolution. With careful treatment of instrumental eÜects the simulation and neutron data agree rather well.1 A similar situation may obtain for bacteriorhodopsin. So I urge you to make more direct comparison with neutron data. 1 M.Tarek and D. J. Tobias Molecular dynamics simulation of the glassy behaviour in globular proteins 1999 to be published. Prof. Smith responded We have performed direct comparison of the full scattering function including instrumental eÜects many times in the past (see ref. 1 and 2 here and references cited therein for examples ranging from soluble proteins to molecular crystals). The present preliminary simulation data on bacteriorhodopsin are sufficiently encouraging to warrant such an investigation but I would prefer to do this with a more complete model. Concerning BPTI our experiments with John Finney and Steve Cusack on this protein concentrated on the inelastic scattering and were not used to derive mean-square displacements (see ref. 3 for a review).Therefore I –nd the existence of new experimental mean-square displacement data on this protein of much interest. 1 N.-D. Morelon G. R. Kneller M. Ferrand A. Grand J. C. Smith and M. Bee J. Chem. Phys 1998 109 2883. 2 A. Lamy J. C. Smith J. Yunoki S. F. Parker and M. Kataoka J. Am. Chem. Soc. 1997 119 9268. 3 J. C. Smith Quart. Rev. Biophys. 1991 24 227. Dr Smart asked What eÜect if any does the value taken for the harmonic restraint constant (used to stabilize the structure) have on the transition temperatures? In addition what about the value taken for the relative permittivity on the transition temperatures? Prof. Smith responded The force constant for the harmonic restraints would be expected to aÜect the dynamics if set high enough. We did not have sufficient computer time to examine this and chose the smallest possible value that would maintain the average structure to within A D2 o é f experiment.In the future one would hope to do away with the restraints and include the environment explicitly. Variation of the relative permittivity would be of interest. We have examined the eÜect of electrostatic model variation on the density of states of a soluble protein1 but it would be of fundamental interest to extend this to membrane proteins and the temperature dependence. 1 J. C. Smith S. Cusack B. Tidor and M. Karplus J. Chem. Phys. 1990 93 2974. Mr Schuler commented MD of protein structures in many cases has been shown to be rather sensitive to equilibration times. Would you agree that there might be a diÜerent (much larger) mean square displacement in this particular system if the equilibration times averaged were extended to say 1 ns instead of 110 ps? Prof.Smith responded The question is somewhat ambiguous. What in the present paper is referred to as ìequilibrationœ is a 5 ps period in which the temperature was rescaled if necessary prior to 110 ps ìproductionœ which was performed at constant energy. If the production periods had been extended then larger mean-squared displacements would be expected. But the –nite energy resolution of the neutron instrument also limits the timescale of the experimentallyaccessible mean-square displacement.1 One also –nds that protein —uctuations sampled over a given production period can depend on the equilibration length.An example of this is given in Mihailescu and Smith.2 1 R. Daniel J. C. Smith M. Ferrand S. Heç ry R. Dunn and J. L. Finney Biophys. J. 1998 75 2504. 2 D. Mihailescu and J. C. Smith J. Phys. Chem. B 1999 in the press. Prof. Tiddy said Your simulation places the protein in an environment which does not include details of the lipid membrane and water. Is it a coincidence that the 240 K transition is close to the water melting temperature? What is the transition behaviour of the membrane lipids ? 141 Faraday Discuss. 1998 111 137»157 Prof. Smith responded The 240 K transition has experimentally been shown to be strongly aÜected by the hydration state of the membrane and is indeed close to the water melting temperature. However we also see a transition at that temperature in a simulation without hydration indicating that the protein also has an intrinsic dynamical anharmonicity that reveals itself at D240 K.I am not aware of data on the transition behaviour of the membrane lipids. Dr Templer asked Would you tell us which parts of BR are rigid and which parts —exible ? Prof. Smith responded The loop regions are more —exible than the transmembrane helices in our simulations. Beyond that there is some diÜraction evidence for —uctuation asymmetry across the membrane and by combining site-directed labelling (H/O) with incoherent neutron scattering Reç at and co-workers1 were able to demonstrate relative rigidity of the retinal binding pocket. Simulation work by several groups is providing additional information in this regard.1 V. Reç at H. Patzelt M. Ferrand C. P–ster D. Oesterheltz and G. Zaccai Proc. Natl. Acad. Sci. U.S.A. 1998 90 9668. Prof. Evans asked Are there calorimetric data for the transitions indicated by neutron scattering and implied by the MD simulations ? Prof. Smith responded I am unaware of such calorimetric data on this system. Our experience with soluble proteins is that the thermodynamic and dynamic transitions do not necessarily coincide and that there can be a timescale-dependence of the mean-square displacement pro–le. Good calorimetric data would certainly be useful and would allow comparisons to be made with glass formation in which the dynamical transition occurs at temperatures above the glass transition temperature as predicted by the mode-coupling theory of non-linear coupling between density —uctuation modes.Dr L. Fisher asked In your 3D –gure which shows mean square displacement as a function of residue number and temperature a number of residues show comparatively large displacement at the highest temperatures examined. The residue numbers are widely separated but it is possible that the actual residues in the folded three-dimensional structure might be close together in space. Can you tell from a comparison with the known crystalline structure of bacteriorhodopsin whether this is in fact the case ? Prof. Smith responded Yes one can tell from the simulation (or the crystal structure) whether residues non-local in the sequence are close together in space. It is also useful to go further than this and to examine correlations between —uctuations of diÜerent atoms.An example of this in which the interatomic distance —uctuation matrix of crystalline lysozyme was examined and corresponding X-ray diÜuse scattering intensities determined using molecular dynamics is given in Heç ry et al.1 For bacteriorhodopsin I think we need a more accurate simulation model before examining these properties. 1 S. Heç ry D. Genest and J. C. Smith J. Mol. Biol. 1998 279 303. Dr P. N. Edwards opened the discussion of Prof. Needhamœs paper The ionisation behaviour of polymeric acids or bases cannot be described by the Henderson»Hasselbach equation. Only titration can yield the proportions ionised at particular pH values since the multiple microscopic pKa values are structure and chain-length dependent and typically in your system would occur over a wide range»probably from D3 to D10.Prof. Needham responded Dr Edwards is correct in his point that the only way to obtain the actual proportions of ionised polymer at a particular pH is by titration. In response to this question the titration experiment for the polymer used in the liposome release experiments has now been performed and the methods and results are brie—y described below. PEAA was dissolved in an equimolar amount (with respect to monomer) in 0.01 M NaOH and 0.137 M NaCl was added to match the salt concentration used in our liposome release experi- Faraday Discuss. 1998 111 137»157 142 Fig. 1 Titration curve for the PEAA polymer. ments.The polymer was then titrated with 0.01 M HCl under a nitrogen atmosphere. The pH response with added HCl is shown in Fig. 1 above. Using the electroneutrality condition de–ned below1,2 we calculated the degree of dissociation of the polymer a. a\[CNa`]CH`[CCl~[COH~]/Cmon are the concentrations of Na` and Cl~ ions due to initial NaOH NaCl and COH~ are the concentrations of OH~ and H` ions and were calculated from is the concentration of monomer units calculated as the titration C Here C and C Na` Cl~ added HCl. and H` the measured pH values. Cmon proceeded. All concentrations were in mol L~1. Using this equation to calculate a the plot shown in Fig. 2 here was constructed showing pH vs. percent protonation ([1[a]]100). Also shown in this plot is the Henderson»Hasselbach equation assuming a single pK of 7.9 i.e.the pH at which a\0.5 as determined from the degree of a dissociation equation. This plot serves to support Dr Edwards assertion ; the protonation behaviour of the polymer does cover a much wider pH range than the Henderson»Hasselbach equation would predict. With regard to the correlation drawn in the paper between membrane cohesion and the degree of protonation of the polymer Fig. 3 is a plot showing the corresponding change in elastic expansion modulus for each SOPC cholesterol membrane composition (relative to pure SOPC) vs. the Fig. 2 Comparison between titration data and Henderson»Hasselbach approximation for the percent protonation of the polymer versus pH. 143 Faraday Discuss.1998 111 137»157 Fig. 3 Increase in the elastic area expansion modulus (over and above that for the pure SOPC bilayer) versus the increase in percent protonation of the polymer (over and above that which occurs at pH 6.7) at the pH where dye is released for bilayers containing increasing amounts of cholesterol (mol% cholesterol is given in parentheses next to each data point). change in percent protonated at the critical release pH for each composition (relative to the percent protonation required to release CF from SOPC vesicles). Thus although the percent protonation actually does cover a diÜerent range than we originally presented in Fig. 4 in the paper the trend between elastic modulus and protonation of the polymer still holds. An increase in the elastic expansion modulus KA is directly proportional to an increase in the percent protonation required to cause release of —uorescent dye from the vesicles.1 U. K. O. Schroé der D. A. Tirrell and K. H. Langley personal communication. 2 M. Mandel Eur. Polym. J. 1970 6 807. Prof. Almgren commented The interaction of PEEA with vesicles covered with PEG-lipids is perhaps not surprising in view of the pronounced attractive interaction between polyacrylic acid at low pH and PEO as well as PEG-surfactants.1 It would be of considerable interest to assess the ability of PEG lipids to hinder the approach of a polymer/polyelectrolyte that interacts only with the bilayer interface ; the described experiment is not informative in this respect. 1 M. Vasilescu D. F. Anghel M.Almgren P. Hansson and S. Saito L angmuir 1997 13 6951 and references cited therein. Prof. Needham responded We are already aware of some work that indicates the formation of inter-macromolecular complexes between the two diÜerent polymers as evidenced by the PEGinduced contraction of poly(methylacrylic acid) cross-linked hydrogels in neutral medium. Thank you for pointing out other work by Vasilescu et al.1 and Petrova et al.2 that speaks of the same association between the two polymers although this work studied the interaction between PEG and PAA not PEAA. Data from Petrova indicates that ììpolycomplexœœ formation occurs between PEG with molecular weights greater than 6000 and a 250 000 molecular weight poly(acrylic acid) (PAA) i.e. at a PEG PAA molecular weight ratio of 1 42.This molecular weight cut-oÜ is greater than the polymer we used which was a 2000 molecular weight PEG grafted to the lipid-bilayer and the molecular weight ratio between the 2000 PEG and 30 000 PEAA is 1 15. There are then some discrepancies between the PEG PAA system and our grafted-PEG PEAA system. However if as other data indicates the addition of ììanchorœœ groups to the PEG (5-nitro-8-quinolinoxyl groups) aids in the formation of complexes with 2000 molecular weight PEG then it is possible that the grafted PEG could interact with the ethyl groups on PEAA in a similar fashion to that with PAA thus aiding in the process of polymer»polymer association and bilayer disruption. Faraday Discuss. 1998 111 137»157 144 In the PEG PAA system the ììpolycomplexesœœ form between the two diÜerent polymer molecules due to hydrogen bonds between the oxygen atoms of PEG and the hydrogen atoms of the carboxyl groups of PAA.The literature indicates that this interaction is prevalent at low pH (data is shown for pHs below 5.0) where the carboxyl groups are highly protonated to allow for the hydrogen bond interaction.2 The pH of our solutions is generally much greater than 5.0 and so the same polymer»polymer interaction is not as favoured for our PEG PEAA system as in the lower pH PEG PAA case. However there still could be an interaction between the protonated groups on the PEAA at our working pHs of 7.5»6.0 and the PEG molecules. I appreciate that it is important to evaluate the eÜects in isolation and –nding a polymer that does not interact with PEG but does interact with the bilayer is an interesting idea.Whether the two events can be uncoupled remains to be seen since a polymer that doesnœt interact with PEG may well have structural and chemical elements that also prevent it from interacting with the bilayer interface. The point of the experiment that we reported in the paper though was to determine if PEG acted as a repulsive barrier to the PEAA polymer in much the same way that it does to more globular macromolecules. It didnœt and the PEAA did cause contents release at the same or even slightly higher pH than for the unmodi–ed bilayer. As we discussed PEG does not represent a repulsive barrier to PEAA and its passage through the PEG layer may also involve polymer»polymer associative interactions (in line with the above references in previous literature).1 M. Vasilescu D. F. Anghel M. Almgren P. Hansson and S. Saito L angmuir 1997 13 6951 and references cited therein. 2 T. Petrova I. Rashkov V. Baranovsky and G. Borisov Eur. Polym. J. 1991 27 189. Prof. Robinson asked Can you comment on the size/shape of your vesicles before and after the addition of PEAA? Have you any information on how the PEAA is distributed in the vesicles ? What is the loading ? i.e. the number of polymers/vesicle ? (Please could you give the molecular weight of polymer in your response.) Your kinetic processes at low pH take place rather rapidly. What is the eÜect (on structure and dynamics) of decreasing the PEAA concentration ? Have you considered doing these experiments in a diÜerent way? For example by adding the polymer to the vesicle at high pH (D7) and then suddenly decreasing the pH to say 6.Could you speculate as whether the results will be similar ? Prof. Needham responded We have not done any size or shape determinations of the vesicles. Size analysis by light scattering is difficult because of the small amount of liposomes that are present in our experimental samples. However Borden et al.1 showed previously that at a lipid polymer weight ratio of 1 1 the size of the measured particles in the sample (which were initially liposomes) changed in diameter from 90 to 16 nm with a drop in pH in the presence of the polymer after a transition at pH 6.5.We have far more polymer in our experiments than this and so we would suspect that the vesicles do undergo a transition to micelles when the polymer interacts with them. Since we have not yet measured the actual amount of polymer bound per vesicle the number of polymers/vesicle at this point can only be approximated by calculating the surface area of a liposome the cross-sectional area of the polymer and –nding the number of polymers that could potentially cover an area of this size. Polymer molecular weight \30 000 g mol~1 SA of one 100 nm liposomes Radius of gyration Rg of polymer\3.37]10~9 \3.14]10~14 m2 Cross-sectional area of polymer \n(R /2)2\8.92]10~10 m2 g Number of polymer liposome \35 201 Decreasing PEAA concentration from 1 to 0.002 mg mL~1 has no eÜect on the critical release pH for pure SOPC membranes.A close examination of the release kinetics has not been done but in all cases release is noted within minutes and so it does not seem that for these concentrations the process has reached a diÜusive limit. 145 Faraday Discuss. 1998 111 137»157 Yes we have considered doing the experiment as you suggest but controlling the pH to as –ne a degree as is required here is not practical. I would speculate that the results would be similar though. Preadsorbing the polymer to the vesicles and then dropping the pH would eliminate any diÜusion in the kinetics and would put the polymer in direct contact with the lipids. This is something we will try in bulk (even though the level of control over pH is not as good as the way we have so far conducted the experiment).The best way of investigating this interaction is with the micropipet manipulation technique which allows us to take a single vesicle with preadsorbed (or electrostatically bound) polymer on the membrane and then change the pH by either transferring the vesicle to a second chamber or using a second pipette to blow the new pH solution over the vesicle. These techniques have been developed and used by us and Prof. Evans in a series of experiments concerning molecular exchange with vesicle bilayers and are described in detail elsewhere.2,3 Preliminary work in collaboration with Evans Thomas and Tirrell that is as yet unpublished the pure lipid bilayer (KL) and an ììeÜective L/KM against the fractional change in area S indicated that this method can provide important and unique information including the area per molecule a of the polymer in the lipid bilayer.Assuming a superposition of compliances for the S overall modulus of the bilayer with polymer (KM) modulusœœ due to polymer partitioning (KS) a plot of K of the vesicle due to polymer partitioning yields a slope of 7.5. K is approximately given by KSB2kB T /aS which gives a as B30 ”2 a very reasonable number for the cross-sectional area of S the polymer in the membrane. 1 K. A. Borden K. M. Eum K. H. Langley J. S. Tan D. A. Tirrell and C. L. Voycheck Macromolecules 1998 21 2649. 2 E. Evans W. Rawicz and A. F. HoÜman Bile Acids in Gastroenterology Basic and Clinical Advances ed. A. F.HoÜman G. Paumgartner and A. Stiehl Kluwer Academic Publishers Dordrecht Boston London 1994 pp. 59»68. 3 D. Needham N. Stoiceva and D. V. Zhelev Biophys. J. 1997 73 2615. Dr Jones asked In the 6-CF release experiments from vesicles (Fig. 6) below the pH where release of 6-CF occurs you show the —uorescence for both pure SOPC and PEGylated SOPC vesicles presumably normalized to the same —uorescence. Did you in fact –nd that the extent of 6-CF encapsulated in the vesicles was reduced in the case of the PEGylated vesicles due to the space occupied by the PEG? Prof. Needham responded The —uorescence is normalised to the —uorescence equivalent to 100% release of the entrapped CF dye upon dequenching whose —uorescence is largely quenched when inside the vesicles.The measured —uorescence therefore represents the relative percents of the total amount encapsulated not necessarily the absolute concentrations of dye either entrapped or released. The —uorescence prior to release was not of direct concern here. It also pertinent to note that a constant loading of CF was difficult to achieve from experiment to experiment which is why we decided to control for it by using the relative percent release measurement. So the small diÜerences in —uorescence intensity for the liposome samples before pH-induced/polymer release of dye seen in Fig. 6 may or may not be signi–cant and (without appropriate calibration) can not be used to estimate whether in fact PEG on the inside of the vesicles does reduce the available space for CF which is at a fairly high concentration.The data actually show that the —uorescence with PEG on the inside of the vesicle is in fact higher than with out it and so this would not a priori –t with an inference that the presence of PEG reduces the available space for this particular molecule. But to reiterate the amount of loading per vesicle sample did vary from preparation to preparation whether PEG-lipids were present or not and so we cannot infer anything from the present data regarding available space inside the vesicles. Prof. Holzwarth asked Cholesterol tends to cluster in lipid bilayers and I wonder how you could have as much as 60% cholesterol in a bilayer without destroying its structure ? You might inspect the literature about the in—uence of cholesterol on bilayers especially the appearance of clusters of cholesterol which was observed at concentrations as low as 7%.1 1 A.Genz J. F. Holzwarth and T. Y. Tsong Biophys. J. 1986 50 1043. Prof. Needham responded Cholesterol concentration is represented as the ììnominalœœ mol% cholesterol in the lipid sample from which the vesicles were made. In an earlier paper1 we dis- Faraday Discuss. 1998 111 137»157 146 cussed this and showed that the saturation of cholesterol composition as measured by the elastic area expansion modulus occurred at D58% (nominal) mol% cholesterol in SOPC vesicles. So 60 mol% cholesterol does represent the maximum possible bilayers simply do not form at concentrations greater than this and excess cholesterol is probably present as crystallites.Thank you for referring us to the Genz et al. paper. We will certainly consider the issues of cholesterol clusters on polymer uptake further. 1 D. Needham and R. S. Nunn Biophys. J. 1990 58 997. Prof. Tiddy commented There is segregation in between PC and cholesterol with the formation of cholesterol-rich patches. Could this occur with your PEG-2000 lipid particularly with the low poly(ethylene oxide) fraction ? Prof. Needham responded The purchased PEG-lipid was chosen such that its hydrocarbon chains (diC18 saturated DSPE-PEG) matched as closely as possible to the bilayer-lipid SOPC (C18 0/C18 1) and so we expect that at these low PEG-lipid mole fractions of 5 mol% the two lipids would be ideally mixed and no PEG-lipid rich domains would be formed.In all our experiments with PEG-lipids using X-ray diÜraction (with Tom McIntosh) and exchange of avidin and lysolipid micelles which are blocked by the presence of PEG-lipids the behavior is well modeled by a continuous well distributed layer of PEG irrespective of density up to the point of saturation. Prof. Evans asked Does uptake of the protonated polymer into cholesterol»lipid bilayers correlate with the excess mole fraction of lipid (i.e. lipid»cholesterol) which would indicate partitioning predominately in the lipid component? Prof. Needham responded This is a very interesting idea and goes back to what Prof. Tiddy mentioned regarding cholesterol-rich domains. To what extent cholesterol-rich patches actually occur in the SOPC»cholesterol system is I believe at the moment unknown.The elastic modulus varies in a fairly continuous way with increasing cholesterol1 with the possible exception of a slight jump at around 40 mol% cholesterol and so mechanically the bilayer behaves as a continuum. We have not actually yet measured the uptake of polymer in a quantitative fashion and so cannot answer this question directly. However if the critical percent protonation required for release of contents can be taken to indicate the amount of polymer bound (which is a tenuous assumption at best) then the question would be is the reciprocal of this critical protonation linearly related to the mol% cholesterol in the membrane? It turns out that it is (analysis of data carried out subsequently to the meeting).So yes it may be that the polymer is –nding the pure lipid component and could be relatively impermeant for cholesterol-rich domains. 1 D. Needham and R. S. Nunn Biophys. J. 1990 58 997. Prof. Neumann asked Is the release of contents from the liposomes dependent on the chain lengths of PEAA at otherwise equal conditions ? Also a comment PEAA is an anionic polyelectrolyte. The protonation of PEAA is multiple binding each EAA residue having its own pK value. However at low degrees of protonation (less than or equal to 0.2 the H` binding is to separated sites independent and of high electrostatic affinity associated with practically one highaffinity pK value. Similarly at high degrees of protonation (greater than or equal to 0.8) the remaining unprotonated sites may be considered as localised and independent islands with one low-affinity pK value.Dr Needham responded All of the experiments that we and others have performed looking at release from vesicles use approximately 30 000 molecular weight PEAA so we do not have any direct data to correlate chain length with release. However the critical pH of the collapse transition of the polymer itself in aqueous solution has been studied as a function of molecular weight. This transition has been noted to decrease with decreasing molecular weight,1 and so we might expect that for smaller polymers the critical pH for release of contents might be reduced to lower pH. In answer to your comment please see my answer to Dr Edwards (earlier) where I present new data.147 Faraday Discuss. 1998 111 137»157 1 U. K. O. Schroeder and D. A. Tirrell Macromolecules 1989 22 765. Dr P. N. Edwards said The amazing sensitivity of your system to changes of pH seems to require a very high order of co-operativity in the disruption of membranes. Perhaps multiple polymer molecules form a channel through the membrane? Prof. Needham responded PEAA certainly can form channels in lipid membranes. Data from Chung et al.1 shows that at relatively low polymer lipid weight ratios near 1 100 PEAA causes the formation of cation selective pores that are often stable for many seconds. However our experiments were carried out at much higher ratios and it is unlikely that the mechanism of release is related to pore formation since as I responded to Dr Robinson earlier at the polymer lipid ratios we were using the polymer induces a vesicle to micelle transition and so probably dissolves all the bilayers making the role of channels a moot point.1 J. C. Chung D. J. Gross J. L. Thomas D. A. Tirrell and L. R. Opsahl-Ong Macromolecules 29 4636. Dr Bezrukov asked Did you try experiments on planar lipid bilayers to understand the mechanism(s) of dye release in more detail ? Prof. Needham responded No we have not looked at black lipid –lms just extruded vesicles and we plan to examine giant lipid vesicles next. Such techniques are not presently in the laboratory tool box although I have worked on black lipid –lms in the past with Dennis Haydon and I can appreciate that this would be an interesting study to perform.Miss Bucak communicated You suggest a possible binding of the protein to PEG as it can make its way to the membrane in the presence of PEG whereas the same size micelles cannot. Do you think it is possible to look at the kinetics of this in the presence and absence of PEG to –nd out if there is an affinity between PEG and the protein ? Prof. Needham communicated in response In the paper we do not talk about protein binding to PEG or the membrane only PEAA interactions. Evidence to suggest that certain proteins like avidin do not traverse the PEG layer (that are the same size as micelles that also do not traverse the layer) have been carried out and published.1,2 What we found and have since modeled for proteins or particles of diÜerent size3 is that the polymer essentially behaves as a hard sphere and that the on rate for the polymer compared to the bare membrane can be orders of magnitude slower.Debbie Leckband seems to have data that suggests that an immobilized layer of protein will be bind to an immobilized layer of PEG in the surface force apparatus if the force applied is large. 1 D. Noppl-Simson and D. Needham Biophys. J. 1996 70 1391. 2 D. Needham N. Stoiceva and D. V. Zhelev Biophys. J. 1997 73 2615. 3 D. Needham T. J. McIntosh and D. V. Zhelev L iposomes. Rational Design ed. A. S. JanoÜ Marcel Dekker New York 1998. Dr Fontana communicated Looking at your data the role of cholesterol is to increase either the cohesiveness or the hydrophobicity of the membrane. Can you suggest a control experiment to separate these two contributions ? Prof.Needham communicated in response This is an interesting question and is one that we have recently been investigating with respect to water permeability. Using the micropipet method on single giant lipid vesicles has shown that water permeability through bilayers progressively decreases as the compressibility of the bilayer decreases showing the role of cohesion as —uctuations in surface density which determine both compressibility and permeability to water. However for the unsaturated lipids whilst the area expansion modulus of bilayers made from lipids that contained multiple unsaturation did not change much as the unsaturation was changed from 1 to 6 double bonds per molecule the water permeability coefficient increased from 20 lm s~1 for 18 0/1 bilayers to 100 lm s~1 for di 18 2 bilayers.This observation indicates that whilst membrane interface cohesion is important in restricting water transport for the more condensed mem- Faraday Discuss. 1998 111 137»157 148 branes such as ones containing cholesterol it appears to be the absolute solubility of water in the membrane that dictates the rates of transport for the softer membranes i.e. this experiment does isolate hydrophobicity at constant interface cohesion since it is known that the solubility of water in alkenes is slightly greater than in alkanes. These compositions have not yet been studied with respect to polymer insertion but it will be of interest to check the polymer (and other molecule) uptake into bilayers made from lipids with increasing unsaturation (decreasing hydrophobicity) to complement studies on bilayers of increasing cohesion.Dr P. N. Edwards opened the discussion of Prof. Neumannœs paper To what extent does the electric –eld existing across plasma membranes of living cells change the expectations relative to the liposomes you describe ? */ind\[(3/2) Ea o cos h o where a is the radius of the is directed ind E Prof. Neumann responded The electric –eld m\[*/ind/d which is induced by interfacial polarization across the membrane of thickness d is dependent on the positional angle h relative to the direction of the external –eld E and is very much larger than E. The induced potential diÜerence drop in the direction of E is given by spherical vesicle or cell.At the pole cap facing the positive condenser electrode */ind from the outside to the vesicle inside whereas at the other pole cap facing the negative electrode */ is in the direction inside to outside. In living cells there is a natural potential diÜerence */nat\/in[/outBfrom [50 to [200 mV where the electric potential / * * / /nat of the outside is taken as the reference /out\0. Note that out exists even in the absence of an external –eld. Generally the total potential diÜerence m\[Emd\*/ind]*/nat\[M(3/2)Ea]*/nat/cos hN o cos h o. Therefore for living cells at and at the other one is in the same direction as */ind facilitates electric pore formation on */nat one pole cap (h\180° cos h\[1) */nat (h\0° cos h\]1) */ is opposite to */ind .Hence nat one side and reduces it at the other side of the cell. Prof. Klein asked Can you comment on the eÜect of the applied –eld on the orientation of the PC headgroups? Molecular dynamics simulations indicate that the P]N dipoles have a broad distribution of angles with respect to the bilayer normal.1 Some lipids even have P]N dipoles perpendicular to the interface. Is it possible for the –eld to increase the population of such lipids which in turn could locally encourage pore formation? 1 K. Tu D. J. Tobias and M. L. Klein Biophys. J. 1995 69 2558. *T ~/T 0 and *A~/A0 . We Prof. Neumann responded The chemical modes *T `/T 0 and *A`/A are slower than the 0 shape deformations indicated by the respective dichroitic modes associate the chemical modes with the orientation of the lipid headgroups toward the direction of the transmembrane –eld allowing the entrance of additional water and ions into the surface of the membrane in the pole can regions ; see Fig.10 in our paper. We have so far not used lipids with dipolar groups perpendicular to the interface. Prof. Bohne asked Is there a dependence of the relaxation times (fast and slow) on the length of time that the –eld is applied ? Prof. Neumann responded The relaxation times of the rapid transients are independent of the pulse length. However at longer pulse durations additional slower deformational modes become apparent. Prof. Robinson asked Is there useful information to be obtained from the amplitudes of the perturbation ? Some of your relaxations are extremely fast»of the order of 1 ls.In the paper you do not give much kinetic information based on the time resolution of these transients ? What are the technical limitations to achieving good resolution ? e.g. –nite rise/decay time of electric –eld pulse. 149 Faraday Discuss. 1998 111 137»157 and q(I)\0.15 ls and *T ~(II)/T Prof. Neumann responded The turbidity relaxations *T /T and *T `/T (Fig. 4 in our paper) 0 0 both indicate that in the presence of the –eld there are at least two kinetic phases phase I and phase II. At E\8 MV m~1 the kinetic analysis yields the amplitudes and the time constants *T ~(I)/T0\0.080 0\0.043 and q(II)\1.24 ls. The time resolution is limited by the spark gap discharge to switch on the electric –eld ; here this machine time is O20 ns.The kinetic analysis shows that both modes re—ect shape deformations of the vesicle due to a –eld-induced increase *S in the membrane surface S at constant vesicle volume. Applying a Mietype numerical code program by Farafornov and co-workers,1 the turbidity data yield the axis ratio p of the ellipsoidally elongated vesicle as a function of time and –eld strength (Fig. 13). On the other hand p is related to the surface fraction f\*S/S with S0\4pa2 being the total surface 0 area of the vesicle of radius a by (Kakorin and Neumann unpublished data) (1) f\p~2@3/2]p1@3 arcsin[(1[p~2)1@2]/[2(1[p~2)1@2][1 The question is now how can the applied electric –eld cause a biphasic increase in the fractional surface area ? in phase I re—ects membrane stretching and smoothing of thermal undulations whereas the slower phase II is caused by electric pore formation.Clearly the electric –eld E interfacial polarization (Maxwell»Wagner). The time constant qpol of this polarization is given by m qpol\aCm (jin ~1]jex ~1/2) where Cm\5]10~3 F m~2 is the speci–c lipid membrane capacity and tively. Here a\50 nm and jin\jex\6.9 mS m~1. Therefore qpol\0.06 ls being about twofold smaller than the experimental value of q(I)\0.15 ls. m The experimental data are quantitatively consistent with the following assignments The rapid across the vesicle membrane reaches its –nal value by the ionic j and jex are the conductivities of the vesicle interior and of the external solution respec- The analysis of the relaxation modes in terms of amplitudes and time constants is rather demanding because the experimental values re—ect the inhomogeneous position-dependent –eld eÜects of the vesicle membrane the pole caps experience a larger induced –eld than the equatorial regions.In addition the –eld-induced membrane tension T T has two components T and which g g l both are dependent on the positional angle h relative to the –eld direction E. T refers to the global vesicle shape deformation and the sin h-average is given by (Kakorin and Neumann unpublished data) (2) STgT\ 3 40 a Me0 ewE2[[64i(p[1)/3a3]N w\80 at T \293 K the relative permittivity of water and i gT\1.7 where e is the vacuum permittivity e 0 is the bending rigidity. For E\8 MV m~1 iB2]10~20 J and p\1.37 we obtain ST ]10~4Nm~1.The local electrostrictive contribution is given by STlT\CmS*/ind 2 T/2 where */ind\[(3/2)Ea o cos h o and S*/ind 2 T\(1.5Ea)2 Pp cos2 h sin h dh/2 0 For E\8 MV m~1 we have S*/ind 2 T\0.12 V2 and STlT\3]10~4 N m~1. The relative surface area increase due to stretching is given by Needham and Hochmuth.2 (3) fT\*ST/S0\(STgT]STlT)/K where KB0.2 N m~1 is the compression modulus. The time constant for stretching is given by and the initial lateral tension T T\2.3]10~3 where g is the viscosity. Here we obtain f q T\2.5]10~10 s. TBga/K q The smoothing of thermal undulations by the electrical Maxwell stress in the membrane with 0\1 mN m~1 yields an additional increase in the projected membrane Faraday Discuss.1998 111 137»157 150 area according to Kloé sgen and Helfrich.3 (4) ]lnM[p/4a2](T fund\*S/S0\[kT /(8pi)] g]Tl]T0)/i]/[p/4a2]T0/i]N where k is the Boltzmann constant and T the absolute temperature. The time constant qund of the undulative bending can be estimated from eqn. (17) in the paper. For E\8 MV m~1 and T \298 K we obtain fundB1.2]10~3 and qundB5]10~8 s. Insertp ing 0\1.35^0.01 from Fig. 13 in eqn. (1) we –nd that fB1.5]10~2. The surface fraction f (I)\fT]fund\0.35]10~2 yields the surface ratio xf(I)\f (I)/f x and f(II)\1[xf(I). From x (I) we readily obtain the ratio well with the experimental ratio x\(*T ~(I)/T x(I)\[p(I)[1)/(p0[1)]\[xf(I)]1@2B0.5. This value compares 0)/*T ~/T0)\(p(I)[1)/(p[1)B0.6. pol\0.06 ls and CHP is C?P f Since (p[1) is proportional to f 1@2 we obtain the ìsurfaceœ relaxation times (at E\8 MV m~1) q (I)\q(I)/2\0.07 ls which is close to the interfacial polarization time q q (II)\q(II)/2\0.62 ls.f At higher –elds the rate constant k of pore formation according to the scheme 1/q\kC?P]kCHPBkC?P . For E\8 MV m~1 we obtain kC?PB1/qp\ f (II)\1.6]106 s~1 comparing well with previous estimates for asolectin vesicles (1/q approximated by 1/q B6]105 s~1).4 fpB(jm/ji)]100% of conductive openings –lled with the intravesicular medium j S m~1 linearly increases from fpB0 at E\1.8 MV m~1 to fp\0.017% r f 1 V. G. Farafornov N. V. Voshchinnikov and V. V. Somiskov Appl. Opt. 1996 35 5412. 2 D. Needham and R. M. Hochmuth Biophys. J.1989 55 1001. 3 B. Kloé sgen and W. Helfrich Eur. Biophys. J. 1993 22 329. 4 S. Kakorin S. P. Stoylov and E. Neumann Biophys. Chem. 1996 58 109. Dr Gon8 i asked Do you have an estimate of the size of the electrically-generated pores in your system? And what would be the density of pore per unit of membrane surface ? What is the in—uence of surface potential (i.e. electrically charged lipids) on the number and electrical properties of the pores? Prof. Neumann responded To answer your –rst two points we have calculated the radius of electropores in lipid membranes in three ways. First from the –eld dependence of the equilibrium constant K\[P]/[C] for the electroporation reaction scheme CHP where C is the closed membrane state and P the electroporated membrane state of the membrane respectively.1 At small –eld strengths EO2 MV m~1 pulse duration tp\10 ls and for the vesicle radius a\160 nm the average pore radius is r6 p\0.35(^0.05) nm of the assumed cylindrical pore of thickness d\5 nm suggesting an average cluster size of SnT\12 (^2) lipids per pore edge.1 The percentage of membrane area of conductivity i\2.2 at E\8.5 MV m~1.The membrane conductivity jm was estimated from the deviation of the linear dependence of ln[K(E2)] on E2. The percentage of the conductive membrane area fp\ 0.017% refers to 142 pores per vesicle or the pore density o\4.35]10~4 nm~2.1 Second the pore radius was obtained from the efflux of electrolyte through the electropores of salt-–lled vesicles. The kinetic analysis of the conductivity increase yields at –eld strength E\1.0 MV m~1 and in the range of pulse durations 5Ot /msO60 the number of water-permeable electropores to be N\35^5 per vesicle of radius a\50 nm with mean pore radius of r6 p\0.9 E ^0.1 nm under Maxwell stress.2,3 Third pore radii are obtained from the kinetics of uptake of dyes and DNA by electroporated cells.For instance the fractional surface area for the dye (SBG M 854) conductive pores is fp\0.035^0.003% and the mean pore radius is r6 p\1.2^0.1 nm. The maximum pore number N is p\1.5^0.1]105 per average electroporated intact FccR~ mouse B cell (cell diameter 25 lm E\2.1 kV cm~1 and tE\200 ls).4,5 The kinetics of the direct transfer of plasmid DNA (YEp 351 5.6 kbp supercoiled MrB3.5]106) by membrane electroporation of yeast cells (Saccharomyces cerevisiae strain AH 215 diameter 5.5 lm) yields the mean radius of the pores in the DNA permeable porous patches r6 p\0.39^0.05 nm at E0\4.0 kV cm~1.The correspond- N ing mean number of pores per cell is p\2.2^0.2]104. The maximum membrane area which is involved in the electrodiÜusive penetration of adsorbed DNA into the outer surface of the electroporated cell membrane patches is only 0.023^0.002% of the total cell surface.4,6 151 Faraday Discuss. 1998 111 137»157 To answer your third point the surface potential acts in two main ways on the membrane electroporation. First diÜerent surface potentials on the two membrane surfaces leads to a transmembrane potential diÜerence which either adds to the externally induced membrane –eld on one pole cap or is opposite to it on the other one.This asymmetry in the total transmembrane –eld leads to an increase in the extent and rate of membrane electroporation on one pole cap and to a corresponding decrease on the other. Second if the repulsive forces between the charged lipid molecules are larger in the outer membrane lea—et than those in the inner one because of asymmetrical ion clouds and unequal Debye lengths at the two interfaces the extent and rate of membrane electroporation are increased to produce the conical electropores. Thereby the distance between the charged headgroups is increased and the free energy due to the repulsive forces is reduced in the outer membrane lea—et. 1 S. Kakorin S. P. Stoylov and E. Neumann Biophys.Chem. 1996 58 109. 2 S. Kakorin and E. Neumann Ber. Bunsen-Ges. Phys. Chem. 1998 102 670. 3 S. Kakorin E. Redeker and E. Neumann Eur. Biophys. J. 1998 27 43. 4 E. Neumann and S. Kakorin Radiol. Oncol. 1998 32 7. 5 E. Neumann K. Toensing S. Kakorin P. Budde and J. Frey Biophys. J. 1998 74 98. 6 E. Neumann S. Kakorin I. Tsoneva B. Nikolova and T. Tomov Biophys. J. 1996 71 868. Prof. Tiddy commented The area per headgroup for PC is ca. 65»70 ”2. How much does this increase in the pore region ? Can you stabilise the pores by the inclusion of larger headgroup lipids ? Prof. Neumann responded The surface area of a primary electropore of radius r6 p\0.35(^0.05) nm induced by 10 ls –eld pulse of E\2 MV m~1 in the membrane of lipid vesicles of radius a\160 nm is Sp\4pr6 p2\0.39 nm2.S compares well with the area B0.5 nm2 of a lipid mol- p ecule in the membrane surface plane.1 Presumably the pore radius increases stepwise with increasing –eld strength and pulse duration because of a discrete number of lipid molecules composing the pore and because of the hydrophobic force oscillating with distance between pore walls.2,3 We have so far no information about the potential stabilisation of pores by the inclusion of larger headgroup lipids. 1 S. Kakorin S. P. Stoylov and E. Neumann Biophys. Chem. 1996 58 109. 2 E. Neumann K. Toensing S. Kakorin P. Budde and J. Frey Biophys. J. 1998 74 98. 3 E. Neumann S. Kakorin I. Tsoneva B. Nikolova and T. Tomov Biophys. J. 1996 71 868. Prof. Morantz asked Can you comment on the charge distribution within the pore? Prof.Neumann responded We expect that in the curved surface of a hydrophilic pore the charge density due to lipid headgroups is smaller than the planar membrane part. Prof. Laggner commented One would expect the pore formation property to depend on the lipid composition»e.g. the presence of PEs»and also on the presence of additional solutes in the vesicles solution e.g. salts or alcohol. How far has this been explored? Prof. Neumann responded Actually the pore formation properties do depend on the lipid composition and on the concentration of salt and other substances in the external and internal vesicle media. We have found that the salt concentration gradient across the membrane has a profound in—uence on the electroporation of membranes containing charged lipids if the membrane surface charge density is pP10~2 C m~2.The diÜerence in the Debye lengths of the inner and outer compartments of the vesicles and cells has a considerable eÜect on the spontaneous curvature. The spontaneous curvature caused by diÜerent chemical composition of the internal and external membrane lea—ets appreciably aÜects membrane electroporation. Analysis of electrooptical data obtained with asolectin vesicles doped with 1,6-diphenyl-1,3,5-hexatrien (DPH) in terms of Gibbs area diÜerence elasticity energy suggests that the fraction of electroporated membrane area increases 2-fold if the concentration of NaCl in vesicles inside increases from cin\ cout\0.2 mM up to cin\0.2 M[cout\0.2 mM where cin and cout are the concentrations of salt in the inner and outer vesicle compartments respectively.Faraday Discuss. 1998 111 137»157 152 Prof. Holzwarth asked How large is the temperature jump resulting from the decay of the electric –eld ? Can you distinguish between the electric –eld eÜect and the temperature eÜect ? Prof. Neumann responded The temperature increase due to Joule heating is up to 3° depending on the applied –eld jump lasting only 10 ls. The additional electrooptical signal is negligibly small compared to the signal change of the direct –eld eÜect. Prof. Svetina commented An estimate presented for the relaxation time for the orientation of non-spherical vesicles in an external –eld is in the order of 100 ls. Couldnœt this time be shorter if the vesicle aligned itself by appropriate shape transformations rather than by a rotation ? Prof.Neumann responded Actually if the electrooptical signal were due the orientation of initially elongated vesicles (e.g. resulting from the extrusion procedure) the time constant should be equal to the rotational relaxation time qrot\130 ls at –eld zero. The measured time constants are in the range 0.1Oq/lsO3. Therefore we deal with initially spherical vesicles which are rapidly elongated by the electrical –eld and rapidly relax to the spherical shape after the electric pulse. However the vesicle elongation either requires an increase in the membrane area or a decrease in the internal vesicle volume. The decrease in the vesicle volume by efflux of intravesicular medium through electropores is very slow; the characteristic time constant is about 10 ms.1,2 The increase in the surface membrane area due to membrane stretching and smoothing of thermal undulations as well as membrane electroporation in the electrical –eld must be much faster than orientation of non-spherical vesicles (\1 ls).However membrane stretching and smoothing of thermal undulations can only rationalise about 23% of the measured increase in the membrane area. The residual increase (of 77%) in the membrane surface area is due to electroporation with a time constant of about 1 ls. 1 S. Kakorin and E. Neumann Ber. Bunsen-Ges. Phys. Chem. 1998 102 670. 2 S. Kakorin E. Redeker and E. Neumann Eur. Biophys. J. 1998 27 43. Prof. Roux opened the discussion of Prof.Leeœs paper How much are the conclusions about the affinity of DOPC vs. DOPS model-dependent? Prof. Lee responded The key observation shown in Fig. 3 in the paper is that quenching is more efficient when the bromine is on phosphatidylserine than when it is on phosphatidylcholine. This lack of symmetry means that there must be at least two types of site on the ATPase. The observation that displacement of brominated phosphatidylcholine by phosphatidylserine –ts to a relative binding constant close to 1 suggests that there are a large number of sites at which phosphatidylserine and phosphatidylcholine bind with equal affinity. The observation of more efficient quenching by brominated phosphatidylserine than phosphatidylcholine then means that there must be a small number of sites at which phosphatidylserine can bind with greater affinity than phosphatidylcholine.The experiments do not show what the number of sites is but assuming there is just one such site we get the relative binding affinities at this site given in Table 3. If we were to assume two sites we could –t the data equally well but with a lower relative binding constant for phosphatidylserine. We picture phosphatidylserines bound to these ì sites œ as phosphatidylserine molecules interacting strongly with the positively charged lysines in the peptides. There is likely therefore to be only one of these sites per peptide on each side of the bilayer and since the peptide contains a spacer Gly residue on one side of the membrane and not the other it may be that there is only one such site and not two.We should be able to distinguish between these possibilities by suitable design of new peptides. Prof. Tiddy asked What are the error bars for data in Fig. 2 and 3? Prof. Lee responded Errors in measuring —uorescence intensity are very small and will be smaller than the data points in the –gures. The greatest problem in these experiments is ensuring total inclusion of the peptide into the lipid bilayer. This seems to be largely a matter of ensuring that the peptide is kept totally dry. 153 Faraday Discuss. 1998 111 137»157 Dr Dijkstra commented An alternative to the electrostatic interpretation of the lipid protein interaction presented in the paper could be a non-speci–c bending energy induced eÜect.Arguments to support this are (1) The data shown in Fig. 3 appear to –t more or less as well to eqn. (4) as to eqn. (6) indicating all ìbindingœ sites would be of equal importance rather than implying a single high affinity site on the peptide. 16 for one lipid over the other. (2) From the results presented in Table 2 it is observed that in general only slight preferences at most are found from the peptide L for the anionic lipids versus the phosphatidylcholine (PC) if the PC is brominated rather than the anionic lipid. This is in line with the lack of a preference from the L22 (3) This is in contrast to the data where the anionic lipid has been brominated (Table 3). This diÜerence would imply the bromination technique does not quite leave the saturated brominated lipid functionally equivalent to the unsaturated non-brominated lipid.Intuitively one would also not expect these species to be the same e.g. one could expect bromination to induce a dipole at that site which a double bond would not. (4) The 5 kJ mol~1 diÜerence in interaction energy observed between PA relative to PC could also be explained by a diÜerence in the hydrophobic length of the one lipid vs. the other. Calculations show the energy diÜerence associated with hydrophobic mismatches of about 1 Aé to be of the same order of magnitude.1 It can be expected that anionic headgroups will experience more headgroup»headgroup repulsion than zwitterionic headgroups and therefore result in large hydration and headgroup areas per molecule and thus shorter hydrophobic lengths.(5) A hydrophobic mismatch eÜect rather than speci–c binding would also explain why no was observed. Here the mismatch diÜerence in the relative affinities for either lipid from the L22 between the peptide and the lipids is so large 1 Aé more or less would relatively speaking make less diÜerence here than close to the ì perfect œ as with L16 . The paper suggests no diÜerence between the lipids affinities is seen due to the tilting the L peptide locating the Lys too far from 22 the lipid headgroup region to allow strong interactions thereby minimising the diÜerence between the anionic and zwitterionic lipids. However tilting will out of geometrical necessity bring all parts of the peptide outside the membrane closer to the membrane surface.This would lead to increased interactions with the lipid headgroups unless tilting goes so far as to place the Lysresidues inside the hydrophobic region. In this case the efficiency of incorporation into thinner bilayers would be expected to decrease which is not the case based on earlier reports from the authors that full incorporation has been observed for L into PC-bilayers with chain lengths 22 between C14 and C24 . (6) A distance of 3.3 Aé is calculated between the anionic headgroup and the Lys-residue using the relative permittivity of water. Given that this distance is the same order of magnitude as the size of a water molecule the validity of the calculation using a bulk relative permittivity value for water in the calculation might be questioned.1 C. Nielsen M. Goulian and O. S. Andersen Biophys. J. 1998 74 1966; D. R. Fattal and A. Ben-Shaul Biophys. J. 1993 65 1795. Prof. Lee responded Since anionic phospholipids show stronger binding to the positively charged peptides than do zwitterionic phospholipids the obvious interpretation is in terms of charge»charge interactions. The simple calculation included in the paper shows that relative binding constants of the kind we –nd can be obtained in this way and that they require close contact between the negatively charged headgroup of the phospholipid and the positively charged Lys on the peptide. The observation that a small number of anionic phospholipids interact with the peptide with high affinity with a large number interacting with the same affinity as phosphatidylcholine is easy to explain if charge»charge interactions are important; I do not see any obvious explanation for this in terms of bending energies.However we could test the idea of an involvement of bending energies by measuring binding constants to phosphatidylethanolamines but this we have not yet done. (1) Indeed the data in Fig. 3 –t equally well to eqn. (3) (assuming a single class of binding site) with the parameters in Table 2 or eqn. (6) (assuming two classes of binding site) with the parameters in Table 3. However the point is that the –t to a single class of binding site gives diÜerent relative affinities for the anionic and zwitterionic lipids depending on which one is brominated; Faraday Discuss. 1998 111 137»157 154 22 this is physically impossible»if there was truly only a single class of binding site quenching curves such as those shown in Fig.3 would be symmetrical being the same independent of which lipid was brominated. The fact that the quenching curves are not symmetrical shows that the anionic lipids can bind to sites where they have high affinity but the zwitterionic lipids do not. That is the basic for the two-class of model presented in the paper and the basis for eqn. (6). (2) I am not sure that I understand the question. The data in Tables 2 and 3 show that for the short peptide L there are a small number of ì sites œ where anionic phospholipids bind with 16 relatively high affinity whereas as shown by the data in Table 1 L such ì sites œ do not appear to be present.(3) As described above the physical properties of brominated and non-brominated phospholipids are very similar. Quenching observed in mixtures of brominated and non-brominated phosphatidylcholine or in mixtures of brominated and non-brominated phosphatidylserine are the same and all –t assuming that the brominated and non-brominated lipids bind with equal affinity to the peptide (Table 1). I know of no information that suggests that a dipole in the chain region would aÜect interaction with the hydrophobic region of a peptide or protein. (4) We know the eÜects of hydrophobic mismatch in these systems from our previous studies (ref. 11). When the hydrophobic thickness of the bilayer is more than 10 Aé greater than the hydrophobic length of the peptide the peptide does not incorporate.For hydrophobic mismatch less than this the peptide does incorporate but with a weaker binding to the mismatched lipid. However the diÜerence in binding is no greater than a factor of 2 much less than the diÜerences we observed here. Any diÜerences in hydrophobic lengths between dioleoylphosphatidylcholine and for example dioleoylphosphatidic acid will be very small and will certainly be much too small to account for the eÜects of anionic phospholipids we see here. Further of course it is not at all obvious how a diÜerence in eÜective chain length could give rise to two classes of binding site of the type necessary to explain our data whereas two classes of site –t naturally with what would be expected for binding of a charge lipid to a charged peptide.(5) It is hard at this stage to be certain about what structure will be adopted by a tilted peptide in a lipid bilayer. There might well be some unfolding of the a-helix at the ends of the tilted peptide. In a tilted peptide it is also possible that the charged residues will be located higher above the lipid/water interface than for a peptide oriented parallel to the bilayer normal. I think what we want here is more experimental data what is the orientation of the tilted peptide in a bilayer ? ; does the a-helix unfold at all ? ; how does the charged lipid headgroup interact with the charged groups on the peptide ? and is the structure of the charge residue important? We do not yet know the answer to any of these questions.(6) Yes indeed. The calculation simply shows that the interaction is likely to be a close one. The reason for using the relative permittivity of water in these calculations is that necessarily the anionic headgroup»Lys pair will be surrounded by water and that this surrounding water will have a dominating eÜect whatever the nature of the ìmediumœ actually between the headgroup and the Lys. Prof. Laggner commented A quasi-ternary system (two lipid species and one peptide) should be described in terms of a phase triangle which implies that the interactions also depend on the peptide lipid ratio. How is that situation in your systems? Prof. Lee responded In these experiments we have to have a lipid peptide molar ratio of at least 15 1 to ensure total incorporation of the peptide into the lipid bilayer.We can then increase the lipid peptide molar ratio up to 1000 1 with no further change in the system. For convenience all the experiments shown here were performed with a molar ratio of 100 1. Because of the positive charge on the peptides we assume that the peptides will be simply dispersed in monomeric form in the bilayers although we have no direct evidence for this. Prof. Klein asked Do you have evidence on the state of aggregation of the peptides particularly those with Trp and Tyr? Prof. Lee responded No we have no direct evidence. However we assume that because of the positive charge on the peptides they will be present in monomeric form. 155 Faraday Discuss. 1998 111 137»157 Dr Gon8 i commented Your results with Tyr residues at the ends of the peptide remind me of the ììmysteryœœ of Trp residues at the ends of many transmembrane peptides.Could you share with us your ideas on this point ? Have you detected any signs of hexagonal phase formation with the short peptides/long lipids ? H phase would mean a solution for the problem. II Prof. Lee responded The idea of the experiments with the Tyr containing peptides was to see if we could get any information about the role of these residues in membrane proteins. What we found as shown in Fig. 5 was that Tyr-containing peptides could incorporate into thicker bilayer than corresponding peptides containing only Leu as the hydrophobic residues. The explanation could be that the Tyr residue simply acts as a ì larger œ residue than Leu so that the eÜective hydrophobic length of the Tyr-containing peptides is greater than for the peptides lacking Tyr.If this were the case Phe should have the same eÜect as Tyr; this is something we will test. The alternative possibility is that insertion of the Tyr residues in the hydrophobic region of the bilayer with the OH group oriented to take part in hydrogen bonding gives Tyr (and by analogy Trp) unique properties at the lipid/water interface. To answer your second question we have not looked speci–cally at this. However at the high lipid protein ratio we use we expect the lipids to be purely bilayer ; other groups have observed hexagonal and other non-bilayer phases in mixtures with peptides but only at very low molar ratios of lipid to peptide.Prof. Barclay said I have two queries concerning the implications of your data on the association of speci–c lipids to proteins. First if for example a protein associates preferentially with a long chain fatty acid would you expect these fatty acids to preferentially associate to build up an enriched zone or domain? Secondly I realise you havenœt any data on the other common method of membrane integration namely through glycosylphosphatidylinositol (GPI)/anchors. However there are a large number of data on GPI anchored proteins and association with lipids and formation of ìì raftsœœ»would you like to comment on these ? Prof. Lee responded To answer your –rst point in previous studies we have shown that phosphatidylcholines interact most strongly with peptides when the thickness of the lipid bilayer matches the hydrophobic length of the peptide (ref.11). However the eÜects are relatively small at most a factor of 2. Thus whilst for example there will be more phospholipids with long chains than short chains around a long peptide the eÜect will not be sufficient to build up a macroscopic domain or ì raft œ of the type thought to be important in cell membranes. In response to your second point we have done no work on lipid-modi–ed peptides and so cannot usefully comment on this. lipid ? Prof. Tiddy asked How does bromination modify lipid behaviour e.g. phase transition temperature. What is the phase transition temperature of the C24 Prof. Lee responded Bromination of an unsaturated lipid such as dioleoylphosphatidylcholine has very little eÜect on its physical properties.The low transition temperature of dioleoylphosphatidylcholine follows from the difficulty in packing the ìbentœ cis unsaturated chains into the gel phase. Similarly the two bulky bromines in each chain give a low transition temperature for the brominated analogue; we have shown that the transition temperature is below 5 °C. In that paper we also showed that order parameters for a variety of spin-labelled fatty acids were the same in bilayers of brominated and non-brominated phosphatidylcholines. We also showed that enzyme activities for a transport protein (Ca2`-ATPase) were the same in bilayers of brominated and non-brominated phosphatidylcholines. 1 J. M.East and A. G. Lee Biochemistry 1982 21 4144. Prof. Bayerl asked Could the morphology of the diÜerent vesicles preparation lead to diÜerent curvatures which may explain part of the diÜerences you observed regarding peptide»lipid interaction ? Did you have any control on morphology in your experiments? Faraday Discuss. 1998 111 137»157 156 Prof. Lee responded All the experiments we reported here were with multilamellar vesicles prepared by hand-shaking. The large size of the multilamellar vesicles means that bilayers will be essentially —at. We have also carried out some studies in which lipid and peptide were mixed in cholate as detergent followed by dilution into buÜer; using this procedure membrane fragments are formed containing the peptide. Using both procedures we obtained the same results.I might add that in experiments with long peptides which did not incorporate into thin lipid bilayers the peptide and lipid clearly formed separate phases as the lipid could be separated from the peptide on Sephadex columns. Dr P. N. Edwards commented Anionic phosphate di-esters have very strong hydrogen bond acceptor properties and would wish to be solvated by many more water molecules (acting as H-bond donors) than can physically be accommodated at a lipid interface. Thus protein residues (Tyr ; Trp) with H-bond donor ability and much weaker acceptor ability are especially favoured at the polar/apolar interface. Prof. Lee responded I am not sure about that. Anionic phospholipids will form very stable lipid bilayers under the right conditions.So the phosphatidylserines and phosphatidic acids used here at neutral pH in the absence of divalent cations all form bilayers in which presumably all hydrogen bonding needs are met either by other lipid headgroups or by water. Indeed crystal structures of some of these molecules show extensive hydrogen bonding networks in the headgroup region. However it is certainly true that Tyr and Trp have unique properties amongst the ìhydrophobicœ amino acids and their hydrogen bonding requirements are best met by a location at the lipid/water interface where the polar groups can be involved in hydrogen bonding with the hydrophobic rings still immersed in a hydrophobic environment. Prof. Deber communicated Your Tyr-containing peptide suggests interfacial involvement of Tyr residues. Can you speculate on the nature of the interactions that attract Tyr speci–cally to the interfacial location ? If certain peptides adopt a tilted or slanted orientation as they cross the lipid bilayer what are the implications for local lipid packing and —uidity ? Can such eÜects be measured? Prof. Lee communicated in response Your –rst comment was answered in reply to the previous question. With regard to your second point it is a very good question to which I have no clear answer. One would imagine that the eÜect of a tilted peptide would be signi–cantly diÜerent to one oriented parallel to the bilayer normal. For a peptide oriented parallel to the bilayer normal it would seem likely that most lipids whose fatty acyl chains are in signi–cant contact with the peptide are also in contact in the headgroup region ; this is the classic idea of a shell of ìboundaryœ or ìannularœ lipids surrounding a peptide or protein. For a peptide with a large tilt the arrangement will be more complex and lipids whose headgroups interact strongly with the charged groups on the peptide may well have chains not interacting with the hydrophobic part of the peptide. Conversely lipids whose chains interact with the hydrophobic part of the peptide may not have headgroups interacting with the charged part of the peptide. I suspect that this kind of level of detail may only be obtainable by molecular dynamics simulations. The experiments would then hopefully give sufficient thermodynamic data on strengths of binding etc. to provide suitable tests of the models. 157 Faraday Discuss. 1998 111 137»157

ISSN:1359-6640    

DOI:10.1039/a901093i

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Phospholipid chain length alters the equilibrium between pore and channel forms of gramicidin Toby P. Galbraith and B. A. Wallace* Department of Crystallography Birkbeck College University of L ondon L ondon UK W C1E 7HX Received 26th October 1998 Gramicidin is an excellent model system for studying the passage of ions through biological membranes. The conformation of gramicidin is well de–ned in many diÜerent solvent and lipid systems as are its conductance and spectroscopic properties. It is a polymorphic molecule that can adopt two diÜerent types of structure the double helical ììporeœœ and the helical dimer ììchannelœœ. This study investigated the in—uence of the acyl chain length of membrane phospholipids on the conformations adopted by gramicidin. We used circular dichroism spectroscopy to examine the conformational equilibrium between the pore and channel forms in small unilamellar vesicles of phosphatidylcholine with acyl chain lengths of 18 20 and 22 carbons.Our results show that in C18 and C20 lipids almost all the gramicidin is in the channel form while in the longer C22 lipids the equilibrium shifts in favour of pore conformations such that they form up to 43% of the total population. This change is attributed to the ability of the double helical conformation to tolerate more hydrophobic mismatch than the helical dimer perhaps due to the greater number of stabilising intermolecular hydrogen bonds. Introduction Gramicidin is a linear polypeptide antibiotic synthesised by the soil bacterium Bacillus brevis during sporulation.1 It is hydrophobic and forms ion channels speci–c for passage of monovalent cations across biological membranes.2 The primary structure of gramicidin includes an unusual sequence of alternating L- and D-amino acids.3 The N- and C-termini are capped with formyl (HCO) and ethanolamine (-NHCH2CH2OH) groups respectively and its sequence is as follows formyl-L-Val-Gly-L-Ala-D-Leu-L-Ala-D- Val-L- Val-D- Val- L-Trp-D-Leu-L-Trp-D-Leu-L-Trp-D-Leu- L-Trp-ethanolamine.The peculiarities of its primary structure have important implications for its three-dimensional structure and function. Firstly all the residues are hydrophobic with no charged or hydrophilic side chains. Secondly because both the N- and C-termini are blocked this prevents the formation of a zwitterion or net charge at any pH reinforcing the hydrophobic nature of the peptide and causing it to partition strongly into the hydrophobic core of the phospholipid bilayer.Thirdly the sequence of alternating L- and D-amino acids allows gramicidin to adopt conformations which would be unacceptable for an all L-peptide. In this case the b-sheet-like motif present in gramicidin means that all the side chains protrude from one side of the sheet rather than alternate sides as is the case in all L-peptides. When the sheet is rolled up to form a helix the hydrophobic side chains project from the outside of the helix to interact with the surrounding solvent or lipid while the central lumen is left unobstructed to allow the passage of ions. 159 Faraday Discuss.1998 111 159»164 Physical characteristics of pores and channels Gramicidin is highly polymorphic being able to adopt a wide range of structures with diÜerent topologies orientations and hands. The two most common forms of gramicidin are the double helix (DH) a dimer in which two monomers are intertwined [Fig. 1(a)],4 and the helical dimer (HD) a dimer in which two helical monomers are joined end-to-end [Fig. 1(b)].5h7 Both the double helix and helical dimer forms have b-sheet-like hydrogen bonding patterns diÜering in the number of intra- and inter-molecular hydrogen bonds and the helical rise per residue. The HD form is commonly known as the ììchannelœœ whereas the DH structures are often referred to as ììporesœœ. In ion-free methanol solution gramicidin has been shown to exist as four interconvertible double helical conformers referred to as species 1 2 3 and 4 diÜering in helical sense and chain orientation.8 The components of this equilibrium mixture can be separated using chromatographic techniques allowing characterisation of the individual species by circular dichroism (CD) spectroscopy.Veatch and co-workers8 proposed that species 1 and 2 are left-handed parallel DHs diÜering in the staggers between the ends of their chains species 3 is a left-handed antiparallel DH and species 4 is a right-handed parallel DH. The relative abundance of these four species in a given solution depends on the nature of the solvent.9 In—uences on the equilibrium between DH and HD forms The equilibrium between the DH and HD forms of gramicidin can be shifted by a number of physical conditions.The principle in—uence appears to be the chemical nature of the environment. Whilst in organic solvents the DH forms are almost exclusively found (with the relative proportions of the four DH species present determined by the nature of the solvent),9 in most lipids the HD form predominates. However if the membranes are prepared from lipids with polyunsaturated fatty acids,10 this has been shown to bias the equilibrium towards the DH forms in small unilamellar vesicles (SUVs) and conformational transitions between the DH and HD forms appear to occur faster in bilayers with polyunsaturated acyl chains.11 In addition in SUVs the balance between DH and HD forms can be aÜected by the chirality of the lipid molecules12 used to produce the bilayers.It has also been shown that the relative proportions of DHs and HDs present depend on the solvent and temperature used during the preparation of the SUVs (the so-called ììsolvent-history dependenceœœ). The conformation which is initially adopted when gramicidin is incorporated into membranes is determined by the solvent in which it was dissolved during sample preparation.13h16 This is not necessarily the conformation that is –nally adopted at thermodynamic equilibrium however. Sychev et al.10 assert that while the initial conformational state of gramicidin in hydrated lipid bilayers is aÜected by the solvent the equilibrium conformational state (after heat incubation) is determined by the lipid structure rather than the nature of the solvent.When taken altogether these studies show that considerable variation in the relative proportions of the HD and DH forms can be achieved in SUVs. Quite a diÜerent picture is seen for gramicidin in bilayer lipid membranes (BLMs) used for conductance studies. Under ordinary conditions essentially only HDs are detected in these Fig. 1 Schematic representations of the (a) double helical (DH) and (b) helical dimer (HD) forms of gramicidin. Faraday Discuss. 1998 111 159»164 160 samples.17 While occasionally channels with very long mean channel lifetimes and lower conductance are seen»which are properties attributed to DHs»the proportion of them is so low that it is difficult to quantitate. It has been estimated that approximately one in 10 000 conducting species found in GMO (glycerol monooleate) membranes may be DHs.This very low proportion suggests that the HD form is considerably more stable in these types of bilayers perhaps by about 6 kcal mol~1. In actuality this would correspond to a very small diÜerence in interaction energies on the order of say one additional hydrogen bond formed. It has been difficult to –nd conditions in BLMs that shift the equilibrium towards the DH forms. Unlike in SUVs neither the lipid chirality18 nor the degree of fatty acid unsaturation19 appears to have any eÜect»with only HDs detected in either type of lipid. Furthermore there appears to be no solvent dependence»under similar conditions which favour DH formation in SUVs only HDs are detected.20 Although supposed DH forms can be produced using certain mixtures of gramicidin analogues,17 it has otherwise been difficult to produce conditions in BLMs which even slightly shift the equilibrium towards DH formation.The most striking aspect of these results is the diÜerence seen in the two types of membrane samples SUVs and BLMs examined by circular dichroism spectroscopy and conductance measurements respectively. It is interesting to consider what properties of the bilayers (or of the methods used to examine them) may have resulted in the diÜerent observed behaviours. Lipid vesicles and planar bilayers diÜer amongst other properties in their radii of curvature the ratio of polypeptide-to-lipid molecules present the packing of lipid molecules and solvent content.Hence these features must also be taken into account when considering in—uences on the DH»HD propensities. An alternative explanation may be that conductance measurements only detect open channels which may represent only a small proportion of the total population in some environments or that the CD experiments cannot distinguish between membrane-associated and membrane-spanning conformers. The question then arises whether conditions can be found that will in—uence the equilibrium in both SUVs and BLMS. The one membrane property that may do this is the chain length of the phospholipid molecules. Very long unsaturated lipids,21 which result in a mismatch of lengths between gramicidin and the bilayer seem to produce some (but not exclusively) DH forms in BLMs.An early study in SUVs22 had also suggested that although the lipid phase state in short lipids did not have an in—uence on the balance between the forms produced in lipids composed of saturated fatty acids hydrocarbon chain length did in—uence the balance between DH and HD forms. Therefore in order to examine for a parallel in—uence in both environments this study used the same type of unsaturated lipids as were used in the conductance studies. We present data on CD studies of gramicidin incorporated into mono-unsaturated phosphatidylcholine vesicles of varying acyl chain length and analyse the spectra using a reference database derived from the individual DH and HD structures to estimate the proportion of molecules in each sample in the DH and HD forms.Materials and methods Gramicidin D (the commercially available mixture of D80% gramicidin A 5% gramicidin B and 15% gramicidin C) was obtained from ICN Biochemicals (Aurora OH USA). The phospholipids di-oleoylphosphatidylcholine (di-C18>1-PC) di-eicosenoylphosphatidylcholine (di-C20>1-PC) and di-erucoylphosphatidylcholine (di-C22>1-PC) were obtained from Avanti Polar Lipids (Alabaster AL USA). Gramicidin and lipid (molar ratio 1 20) were initially dissolved in 300 ll of a 2 1 (v/v) CHCl3»CH3OH mixture dried by rotary evaporation and hydrated in 300 ll de-ionised Milli-Q water (Millipore Corp. Bedford MA USA). The samples were mixed thoroughly and sonicated for 10 min at around 50 °C and then incubated overnight (approximately 16 h) at 55 °C.Prior to examination the samples were sonicated using a Soniprep probe sonicator for 6»10 bursts of 10 s. The resulting samples were spun at 12 000g for 4 min and the supernatant then loaded into a 0.02 cm pathlength Suprasil cell. Gramicidin in acetic acid samples were prepared using 15 mg ml~1 in glacial acetic acid (BDH Analar grade). Spectra for these samples were obtained using a 0.001 cm pathlength cell. CD spectra were collected at 0.2 nm intervals over the wavelength range from 190 to 300 nm at 50 °C on an Aviv 62DS spectropolarimeter. Five repeat scans were collected for each sample preparation. 161 Faraday Discuss. 1998 111 159»164 Fig. 2 Reference circular dichroism spectra for the three DH species (»») and the HD form (» » ») of gramicidin.The spectra were averaged and baseline corrected using similarly prepared samples without gramicidin present as the baselines. Individual spectra derived from experiments in propan-2-ol8 provided the reference data for the double helical species 1 3 and 4 (species 1 and 2 were represented by a single spectrum as their spectra are virtually identical) as described in Chen and Wallace,9 and a spectrum from the di-C18>1-PC preparation in this study was used as the reference spectrum for the helical dimer form (Fig. 2). The sample spectra were analysed using a least-squares –tting algorithm23 over the wavelength range from 198 to 250 nm. A –t parameter the normalised root mean square deviation (NRMSD) which is a measure of the correspondence between the calculated results and the experimental data,23 was determined for each analysis.Table 1 Estimated contents of double helical and helical dimer conformers in lipids of diÜerent acyl chain lengths Proportion of conformer (%) NRMSD HD DH Species 4 DH Species 3 DH Species 1 Acyl Chain 0.00 0.09 0.17 1.00 0.99 0.57 0.00 0.00 0.16 0.00 0.00 0.00 0.00 0.01 0.27 CC 18 C20 22 Results In a range of phospholipids with saturated acyl chains of lengths ranging from C12 to C18 gramicidin was found to have a distinct CD spectrum which is attributable to the b6.3 HD structure.13,14,22,24 In this study in di-C18>1-PC the CD spectrum of gramicidin obtained was essentially identical to the spectra previously reported in saturated lipids and thus indicates it too is of a HD form (Fig.3). In di-C20>1-PC the spectrum obtained was very similar (Fig. 3) to the C18 spectrum and analyses indicated that 99% of the molecules were in the HD form with only 1% in DH conformations (Table 1). This was consistent with conductance measurements which showed only one short-lived channel type in this lipid corresponding to the b6.3 helical dimer.21 In di-C22>1-PC however a signi–cantly diÜerent spectrum was obtained (Fig. 3). It shows a marked decrease in the peak at approximately 219 nm and a slightly negative peak at D230 nm where the HD spectrum exhibits a positive ellipticity. According to the analysis (Table 1) the HD conformation only contributes 57% of the components present with DH conformers making up the rest 27% was found to correspond to species 1 with the remaining 16% being attributable to species 4.This is consistent with conductance studies that detected both HD and other (presumably DH) structures in the C22 lipids.21 It has recently been suggested25 that acetic acid is a solvent which can produce the ììchannelœœ form of gramicidin in contrast to all other organic 162 Faraday Discuss. 1998 111 159»164 20>1-PC Fig. 3 Circular dichroism spectra of gramicidin in SUVs composed of lipids with fatty acid chains of diÜerent (» » » ») di-C and (… … … … … … …) di-C22>1-PC. lengths (»»») di-C18>1-PC solvents which have been found to produce ììporeœœ forms. To examine this we obtained the CD spectrum of gramicidin in this solvent (Fig.4). The spectrum in acetic acid clearly diÜers from the signature channel spectrum in both shape and peak positions. When calculations were done to determine the proportions of channel and pore components present the very high NRMSD –t parameter (0.30) suggested that not only was it not the channel structure but also that a diÜerent conformation was present in this solvent. If the reference database was then altered to also include the spectrum of an ion-containing pore form26 the –t improved slightly (to 0.20) but was still poor. Therefore the structure in this solvent is not well represented as any one of the standard helical dimer or double helical structures. Clearly then acetic acid is not a membrane-mimetic solvent for gramicidin and thus despite the authorsœ claims,25 the crystal structure in this solvent is not of the channel form.To date no organic solvent has been found to give rise to the ììchannelœœ structure. lipids (» » »). Discussion In this study we investigated the acyl chain length dependence of the various DH and HD conformations of gramicidin in SUVs formed from lipids with long fatty acid chains which result in a Fig. 4 Circular dichroism spectrum of gramicidin (15 mg ml~1) in glacial acetic acid (»»») compared with the channel spectrum of gramicidin in C18 163 Faraday Discuss. 1998 111 159»164 mismatch between the length of the gramicidin dimer and its surrounding bilayer. These studies can be directly compared with conductance studies on the same types of lipids.Previous studies with other lipid types have not quanti–ed the proportions of the various forms present under diÜerent conditions in SUVs. In this study we were able to do so by augmenting the existing spectral databases produced for analysing the proportion of gramicidin DH species with a gramicidin HD reference spectrum and using these with an established algorithm for secondary structure analysis. Previously Mobashery et al.21 had carried out conductance measurements to assess the eÜect of bilayer thickness on the conformation of gramicidin. They observed that under most experimental conditions gramicidin A adopts the b6.3 HD conformation and that normally DH structures do not form membrane-spanning channels unless there is a considerable mismatch of lengths.In di-C18>1-PC and di-C20>1-PC BLMs the conductance was shown to be of a single type corresponding to the b6.3 HD. However in di-C22>1-PC BLMs signi–cant numbers (although still a minority of the total) of conducting forms were of a diÜerent type assumed to be DH structures. The present CD study suggests the same situation is true in SUVs. In these samples the HD is still the predominant form but a signi–cant proportion of DHs is also formed. Thus it appears that in and di-C18>1-PC di-C20>1-PC the energetically favourable conformation for gramicidin is the b6.3 helical dimer channel. However in the thicker bilayer formed by di- C the hydrophobic mismatch created by the diÜerence in length of the b6.3 channel and 22>1-PC the membrane means that the equilibrium is shifted in favour of double helical conformations.This has been explained21 in terms of the contributions to the free energy diÜerence between gramicidin conformations from gramicidin itself and from its interactions with its environment the free energy diÜerence between the DH and HD forms increases as the hydrophobic mismatch increases such that the equilibrium shifts signi–cantly for DH species to be observed both by CD (this study) and conductance studies.21 This may be attributable to the greater dimeric stability of the double helices which are held together by 28 intermolecular hydrogen bonds compared to only 6 intermolecular bonds in the channel conformation. Paper 8/08270G References 1 R.D. Hotchkiss and R. J. Dubos J. Biol. Chem. 1940 132 791. 2 S. B. Hladky and D. A. Haydon Biochim. Biophys. Acta 1972 274 294. 3 R. Sarges and B. Witkop J. Am. Chem. Soc. 1965 87 2011. 4 W. R. Veatch and E. R. Blout Biochemistry 1974 13 5257. 5 D. W. Urry Proc. Natl. Acad. Sci. U.S.A. 1972 68 672. 6 G. N. Ramachandran and R. Chandrasekaran Indian J. Biochem. 1972 9 1. 7 A. S. Arseniev I. L. Barsukov V. F. Bystrov A. L. Lomize and Y. A. Ovchinnikov FEBS L ett. 1985 186 168. 8 W. R. Veatch E. T. Fossel and E. R. Blout Biochemistry 1974 13 5249. 9 Y. Chen and B. A. Wallace Biopolymers 1997 42 771. 10 S. V. Sychev L. Barsukov and V. Y. Ivanov Eur. Biophys. J. 1993 22 279. 11 K. J. Cox C. Ho J. V. Lombardi and O. D. Stubbs Biochemistry 1992 31 1112.12 J. D. Callahan R. Bittman and B. A. Wallace unpublished results. 13 J. A. Killian K. U. Prasad D. Hains and D. W. Urry Biochemistry 1988 27 4848. 14 P. V. LoGrasso F. Moll III and T. A. Cross Biophys. J. 1988 54 259. 15 M. C. Bano L. Braco and C. Abad Biochemistry 1991 30 886. 16 M. Bouchard and M. Auger Biophys. J. 1993 65 2484. 17 R. E. Koeppe II and O. S. Andersen Annu. Rev. Biophys. Biomol. Struct. 1996 25 231. 18 L. L. Providence O. S. Andersen D. V. Greathouse R. E. Koeppe II and R. Bittman Biochemistry 1995 34 16404. 19 J. Girshman D. V. Greathouse R. E. Koeppe and O. S. Andersen Biophys. J. 1997 3 1310. 20 D. B. Sawyer R. E. Koeppe II and O. S. Andersen Biophys. J. 1990 57 515. 21 N. Mobashery C. Nielsen and O. S. Andersen FEBS L ett. 1997 412 15. 22 B. A. Wallace W. R. Veatch and E. R. Blout Biochemistry 1981 20 5754. 23 B. A. Wallace and C. L. Teeters Biochemistry 1987 26 65. 24 D. V. Greathouse J. F. Hinton K. S. Kim and R. E. Koeppe II Biochemistry 1994 33 4291. 25 B. M. Burkhart N. Li D. A. Langs W. A. Pangborn and W. L. Duax Proc. Natl. Acad. Sci. U.S.A. 1998 95 12950. 26 B. A. Wallace Biophys. J. 1984 45 114. Faraday Discuss. 1998 111 159»164 164

ISSN:1359-6640    

DOI:10.1039/a808270g

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Protein inclusion in lipid membranes A theory based on the hypernetted chain integral equation Patrick Lagué e,a Martin J. Zuckermannb and Benoï� t Rouxa a Departments of Physics and Chemistry Universiteç de Montreç al C.P. 6128 succursale Centre-V ille Montreç al (Queç bec) H3C 3J7 Canada b Physics Department McGill University Montreç al (Queç bec) Canada Receiøed 11th September 1998 A theory for describing the structure of the hydrocarbon chains around a protein inclusion embedded in a lipid bilayer is developed on the basis of the hypernetted chain integral equation formalism for liquids. The exact lateral density»density response function of the hydrocarbon core which is extracted from a molecular dynamics simulation of a pure lipid bilayer is used as input to the theory.Numerical calculations show that the average lipid order is perturbed over a distance of 25 to 30 ” around a hard repulsive cylinder of 5 ” radius representing an a-helical polyleucine protein inclusion. The lipid-mediated protein»protein interaction is calculated and is shown to be non-monotonic being repulsive at an intermediate range but attractive at short range. It is found that the lipid matrix contributes a free energy well of 8 kBT to the association of two cylindrical inclusions. 1 Introduction The microscopic factors which drive the assembly of membrane-bound proteins are of considerable interest to both biophysicists and biologists since the energetics of this process is a key element in understanding membrane protein stability.It has been demonstrated that speci–c protein»protein interactions play a crucial role in the stability of protein aggregates and that the physical state of the bilayer modulates the activity of membrane-bound proteins and aÜects the lateral distribution of protein in the membrane surface.1h3 For example the energetics of inclusion-induced bilayer deformation has been investigated using the gramicidin A channel as a molecular probe.4 The membrane spanning portions of many integral membrane proteins consist of one or a number of transmembrane a-helices. Interactions between transmembrane helices contribute to the energetics of folding and oligomerization in a manner that could be highly speci–c in some cases but relatively non-speci–c in others.3 Since the presence of proteins perturbs the structure of bilayer membranes (see ref.5) it seems reasonable that transmembrane helices may also interact with one another via some non-speci–c lipid-packing eÜects. This concept raises important questions concerning the nature of lipid-mediated driving forces for protein aggregation and assembly. In other words what is the in—uence of the average structure of the core of the membrane a liquid-crystalline —uid environment of partially ordered hydrocarbon chains on protein»protein association ? Theoretical investigations of this problem were initiated by Marcelja in 1976 who proposed a mean-–eld model of a lipid bilayer based on order parameters related to lipid chain conformational states.6 His theory described the eÜects resulting from a non-speci–c interaction between 165 Faraday Discuss.1998 111 165»172 membrane integral proteins and the surrounding lipids. The model assumed that the most important change in lipid structure was restricted to the annulus of those lipid chains which are in direct contact with the protein. However at temperatures above the main gel/liquid crystal phase transition it was found that the disturbance caused by the protein inclusion extended to the second or third neighboring chains. Marcelja showed that the change in lipid order gave rise to an indirect lipid-mediated interaction between membrane integral proteins leading to a monotonically attractive potential between two proteins embedded in a membrane in the —uid state with a free energy well of 2»3 kBT at contact.protein interaction which decays monotonically as a function of distance for two proteins embedded in a membrane in the —uid state (though the interaction was often found to depend on the state of the bilayer see ref. 12 14). Given that the in—uence of a protein inclusion is to perturb the natural state of the bilayer the indirect interaction between membrane proteins was obtained under the assumption that —uctuations are suppressed in the vicinity of the proteins. The lipidmediated protein»protein interaction is then caused by the overlap of the annuli and its range was found to increase with increasing correlation length. Typically the magnitude of this interaction was on the order of a few DiÜerent approaches again using mean-–eld theories were due to Schroé der,7 Owicki et al.8,9 and Pearson et al.10 In these theories the state of the lipid bilayer is characterized by several spatially inhomogeneous ììorder parametersœœ which are directly related to —uctuations in the lateral density of the lipid chains.The equation for the spatial variation of the order parameter –eld is derived from a Landau»de Gennes free energy functional with a limited gradient expansion. 11 Interaction strengths and correlation lengths for the pure bilayer are described in terms of phenomenological parameters in these theories. Similar ideas were used in more recent theoretical work which incorporates the in—uence of membrane stretching bending moduli and the spontaneous curvature.12h14 These earlier theoretical treatments generally predicted an attractive lipid-mediated protein» kBT at protein contact.One important limitation of these previous studies is the signi–cant amount of simpli–cation that has to be introduced in order to construct tractable analytical theories. The bilayer membrane a complex macromolecular assembly of amphiphilic phospholipid molecules is thus described in terms of a necessarily limited phenomenological free energy functional. In contrast a realistic approach for the study of the structure and dynamics of biological membranes is the use of molecular dynamics simulations based on detailed inter-atomic potentials.15 In the last few years studies of pure lipid bilayers16,17 and protein»membrane systems18h20 have demonstrated the feasibility and success of such detailed simulations (see also ref.15 and references therein). In principle free energy molecular simulations and perturbation techniques could be used to calculate the solvation free energy of inclusions.21 However molecular dynamics simulations are computationally intensive and cannot be used to address such aspects of membrane structure. It is at present not feasible to examine long-range protein»protein interactions embedded in a membrane with molecular dynamics simulations due to the very long time scales involved in the relaxation of the lipids. Progress on general questions about protein»protein interactions clearly requires a diÜerent approach. Recently Sintes and Baumgaé rtner22 examined the problem of lipid-mediated protein interactions by using Monte Carlo computer simulations based on a simple model of the lipid membrane.The model represented the bilayer with 2]500 lipid molecules and each molecule was modeled by a —exible chain with –ve monomers; two hard cylinders represented the proteins. The simulations gave a depletion-induced attraction between proteins lying closer than one lipid diameter and a —uctuation-induced attraction for larger inter-protein displacements. The latter has a correlation length of about three lipid diameters. However the main limitation of the approach was again that the description of the membrane lipids was necessarily simpli–ed to decrease the computational cost. In this short paper we propose and develop a diÜerent approach to examine the in—uence of lipid chains on protein»protein interactions.The present approachs based on statistical mechanical theories involving integral equations which were developed for the study of liquids23,24 and which use recent computational advances for multi-dimensional systems.25,26 The integral equation theory presented here is constructed as a hypernetted chain (HNC) equation projected in the two-dimensional space of the lipid membrane plane. The exact lateral density»density response function of the hydrocarbon core is used as an input to this theory. This response function is Faraday Discuss. 1998 111 165»172 166 closely related to the membrane structure factor which can be extracted from X-ray or neutron scattering measurements.23 However since such experimental data are not presently available the response function was calculated from the con–gurations of a molecular dynamics simulation of a lipid membrane.17 Our HNC integral equations then allowed us to calculate the perturbation of the hydrocarbon core around a protein inclusion and the lipid-mediated protein»protein interaction based on the information extracted from the molecular dynamics simulation of the lipid membrane.From this point of view the present theory oÜers an intermediate approach combining aspects of both mean-–eld theories and fully detailed atomic simulations. Our integral equation for the average structure of membranes under the in—uence of protein inclusion is similar in spirit to the theory developed by Pratt and Chandler to describe the hydrophobic eÜect.27 In their theory the experimental oxygen»oxygen pair correlation function was used as an input to calculate the free energy of non-polar solutes in water.To illustrate the present approach the lateral perturbations on the membrane structure as well as the lipid-mediated protein»protein interaction were calculated for a 5 ” radius cylinder representing an a-helical polyleucine. In Section II we develop the theoretical formulation of our approach and preliminary results for a dipalmitoyl phosphatidylcholine (DPPC) bilayer are described in Section III. The paper is concluded with a brief summary and a discussion of future work. (1) [ 1 2 Pdr Pdr@ *o(r)Cm( o r[r@ o )*o(r@) (2) (3) (4) 2 Integral equation theory We consider isolated protein inclusions embedded in a uniform membrane in the —uid state.It is assumed that the carbons along the lipid acyl chain are aÜected by a repulsive potential u(r) due to the presence of the protein inclusions. For the sake of simplicity we assume that the protein inclusions are hard repulsive objects which only interact with the hydrocarbon chains and that the polar headgroups are not directly aÜected by the protein. Furthermore we assume that the perturbation acts only on the lateral positions of the lipids. In this case we require an integral equation which allows us to calculate the average carbon density projected in the two dimensional membrane plane So(r)T where r4(x y). We begin by writing an expression for the free energy density functional theory for non-uniform liquids in the HNC approximation,23,24 A[So(r)T]\Pdr So(r)TlnCSo(r)T [*o(r)]bu(r)So(r)T D o 6 where o 6 is the density of the hydrocarbon chains in the uniform two-dimensional membrane plane *o o (r)\So(r)T[o 6 is the deviation from the uniform density 6 .The direct lipid»lipid correm( o r[r@ o ) C lation function sm(r) the equilibrium carbon»carbon density is de–ned in terms of susceptibility of the uniform unperturbed membrane Cm(r)\(o 6 )~1d(r)[sm~1(r) sm(r) is a response function which is related to the density —uctuations of carbon pairs in the unperturbed bilayer (see below). According to the free energy variational principle,24 the average density is obtained by minimization of the functional A with respect to the functions So(r)T.This leads to the HNC-like integral So(r)T\ equation o 6 expC[bu(r)]Pdr@ Cm( o r[r@ o ) *o(r@)D which must be solved self-consistently. The integral equation can be rewritten in a form more suitable for numerical calculations as a pair of coupled 2d-HNC equations c(r)\exp[[bu(r)]h(r)[c(r)][h(r)]c(r)[1 and (5) o 6 h(r)\Pdr@ c( o r[r@o)sm(r) 167 Faraday Discuss. 1998 111 165»172 where h(r)4*o(r)/o 6 is the protein»lipid correlation function and c(r) is the direct protein»lipid correlation function. Eqn. (5) is the well-known Ornstein»Zernike equation23 for an isolated impurity in an in–nite bulk system. The excess Helmholtz free energy due to the protein inclusion is obtained by substituting the self-consistent solution to eqn.(3) into the free energy functional A of eqn. (1). This leads to the closed form expression (6) b*A\o 6 PdrM12 [h(r)]2[12 h(r)c(r)[c(r)N This expression is used to calculate the lipid-mediated protein»protein interaction. Pair correlation function of the unperturbed membrane A central quantity in the present theory is the response function of the uniform unperturbed membrane sm(r) de–ned in eqn. (2). Functions such as sm(r) play a central role in the response of the average structure of an equilibrium system to a small perturbation.23 The response function s is related to lipid density»density —uctuations of carbon pairs in the unperturbed bilayer at m equilibrium sm( o r[r@ o )\S(o(r)[So(r)T)(o(r@)[So(r@)T)T (7) \So(r)o(r@)T[So(r)TSo(r@)T where o(r) is the density of the ensemble of carbon atoms comprising the lipid chains (8) o(r)\; ; n d(ra ( i)[r) i a/1 Here i is the index of the lipid molecules and a which goes from 1 to n is the index of the carbon atom along the lipid chains.In the uniform unperturbed system the average of o(r) is the average carbon density per unit area So(r)T\; a ( i)[r)T ; n Sd(r i a/1 (9) \o 6 The average density o 6 is equal to 2]n]the surface density of lipid molecule per lea—et where n is the number of carbon atoms in the hydrophobic moiety of one DPPC molecule (the factor 2 appears because of the upper and lower lea—ets of the bilayer). Density»density —uctuations of carbon pairs can also be expressed in terms of the radial intramolecular and intermolecular s ; n ; ; n d(r U )[ ( r a m(r[r pair @)\ correlation T; functions (i)[ of )d the rc ( j pure r@) unperturbed [o 6 o 6 membrane (10) i a/1 j c/1 \o 6 [d(r[r@)]Sm(r[r@)]Hm(r[r@)o 6 ] m(r[r@) represents the carbon»carbon intermowhere the function Sm(r[r@) represents the carbon»carbon intramolecular pair correlation within a given lipid molecule (i\j) while the function H lecular pair correlation between distinct lipids (iDj).By symmetry the pair correlation functions depend only on the distance r projected in the x,y plane with r\J(x[x@)2](y[y@)2. In the present study the pair correlation functions were calculated from the con–gurations generated by molecular dynamics simulations of a detailed atomic model of a pure DPPC bilayer at 323.15 K performed by Feller et al.17 The function S (11) S U m(r) was calculated as m(r)\T1 n ; Nintra a a(r ; (r r ; ] r] *r * ) r) a a where Nintra(r ; r]*r) is the total number of carbons from a given lipid found within the two-dimensional annulus going from r to r]*r centered around carbon a and Faraday Discuss.1998 111 165»172 168 (12) U a(r ; r]*r)\2n[(r]*r)2[r2] is the area of the annulus. The function Hm(r) was calculated as Hm(r)\T1 n ; N a inter a (r ; ( r r ] ; r] *r) * o 6 r) [1 a a where Ninter(r ; r]*r) is the number of carbons from the other lipids found within the twodimensional annulus going from r to r]*r centered around carbon a of a lipid molecule.At large values of r intermolecular correlations vanish and the function Hm(r)]0. All the heavy atoms from the acyl chains and glycerol backbone were counted in the calculation of the pair correlation function for a total 39 particles per lipid. The polar headgroup was not included. Since the average cross-sectional area is 62.9 ”2 per DPPC,17 the average density per unit area o 6 is equal to 1.24 ”~2. Computational details The 2d-HNC eqn. (4) and (5) were solved for two diÜerent systems. First a single isolated a-helical protein inclusion modeled as a hard repulsive cylinder was considered. Secondly two identical cylindrical protein inclusions were examined at various separations and the lipid-mediated protein»protein free energy was calculated using eqn.(6). The radius of the hard cylinder was chosen to be 5 ” corresponding closely to that of a polyleucine a-helix. Eqn. (4) and (5) were solved numerically by a method used previously to solve HNC integral equations.25,26 It involves a mapping of all functions onto a two-dimensional discrete grid e.g. u(x y)]u(i j) and h(x y)]h(i j). Two grid dimensions were used depending on the number of proteins inserted in the system. A discrete grid with N\1024]1024 with a spacing d of 0.10 ” was used with single protein systems and N\2048]1024 with the same d spacing was used for two protein systems. The two-dimensional convolution in eqn. (5) was calculated using a numerical two-dimensional fast Fourier transform (FFT) procedure.28 The convolution was calculated directly without zero-padding.This corresponds to a periodic system in the x and y directions. An iterative scheme with simple mixing was used to solve eqn. (4) and (5) self-consistently. In this scheme the mth iteration is obtained from (13) c(m`1)\jMexp[[bu]h(m)[c(m)][1[h(m)]c(m)N](1[j)c(m) Approximately 50 iterations were necessary for convergence. Numerical solution takes 5 min on a Pentium II 400 mHz. 3 Results and discussion correlation function Fig. 1 gives the results for the carbon-carbon distribution function as extracted from the molecular dynamics simulations of Feller et al.17 The –gure shows that the carbon»carbon intramolecular Sm(r) is positive de–nite. This correlation function has a large contribution followed by a peak up to 3” ” and then a slow decay over a distance of 10-15 .The short range contribution to the intramolecular correlation arises from nearest neighbor carbons along the acyl chains (i.e. carbon i with carbon i[1 and i]1). Such a peak is an indication of a signi–cant amount of short range order in the lipid chains perpendicular to the plane of the bilayer. The long range contribution which extends to 15 ” is due to carbons located in diÜerent acyl chains of a single lipid molecule. The carbon»carbon intermolecular correlation function Hm(r) has a strong negative contribution at short distances due to the lipid»lipid core repulsion. Interestingly the negative contributions are much less important at distances greater than 5 ”.As a result the response function sm(r) exhibits a strong peak for distances up to 3 ” arising from the intramolecular correlations and a second strong peak at a distance of 4 ” »5 arising from the intermolecular correlations. The response function decays in an oscillatory manner with small positive peaks appearing around 9 and 14 ”. The observed structure of the correlation functions of a pure DPPC membrane suggests that the expected response to perturbations may be quite complex. To assess the lateral response of the bilayer quantitatively the average density around an a-helical protein inclusion modeled as a hard cylinder was examined. The pair correlation h(r) for the distribution of carbon atoms around the cylinder as calculated using the integral eqn.(4) and (5) of Section II is shown in Fig. 2. It can be 169 Faraday Discuss. 1998 111 165»172 Fig. 1 Carbon»carbon density»density response function sm(r) (solid line) extracted from molecular dynamics simulations and used as input in the integral equation. The intramolecular Sm(r) (dashed line) and intermolecu- H lar m(r) (dash-dotted line) pair correlation functions extracted from the molecular dynamics simulations of Feller et al.17 are shown. (To match the dimensions of ”~2 of Sm(r) the function sm/o 6 and the function Hm(r)]o 6 are shown.) seen from the –gure that the perturbation of the membrane structure extends from 5 to 25 ” from the center of the cylinder with strong oscillations in the correlation function separated by approximately 4 ” ” .Furthermore the correlation function is negative between 5 and 15 from the center of the cylinder. This indicates that the average lipid density next to the cylinder is lower than its bulk value. This region next to the cylinder can thus be regarded as a depletion layer with respect to the lipid molecules. It should be noted that no lipid can be closer than 5 ” to the center of the cylinder because of the hard core repulsive potential and the protein»lipid correlation function is therefore ” [1 inside the cylinder. The pair correlation function is positive from 15 to 25 which indicates that the density of carbon atoms is higher than the uniform bulk value in this interval. The integral equation theory and the free energy expression of eqn.(6) can be used to calculate the lipid-mediated free energy between two protein inclusions as a function of their separation. The results are given in Fig. 3. The –gure shows that there is a free energy barrier at a distance of 20 ” between the two proteins followed by an attractive free energy well for distances less than Fig. 2 Perturbation of the average structure of the hydrocarbon core around a hard repulsive cylinder of 5 ” radius. Faraday Discuss. 1998 111 165»172 170 k ” BT ) between two hard repulsive cylinders of 5 radius as a Fig. 3 Lipid-mediated free energy (in units of function of their separation distance. ”. The repulsive barrier is approximately 3.5 15 kBT and extends from 15 to 30 ”. At separations greater than 30 ” the protein»protein potential is very small and oscillatory.Finally at protein» protein contact the magnitude of the lipid-mediated potential is [8 kBT . This value is in good agreement with (though somewhat more negative than) previous estimates.6h10,22 However in contrast to previous studies the results based on the integral equation theory show that the lipid-mediated protein»protein interaction is both attractive at short distances between proteins and repulsive at large distances. The attractive part of the interaction is clearly due to the presence of a depletion layer of lipid molecules close to the embedded protein. This is in good agreement with the results of Sintes and Baumgaé rtner22 mentioned above. The repulsive part of the interaction has not been observed previously and has important consequences for protein association in biological membranes.The free energy maximum at 20 ” corresponds to the distance at which it is increasingly more difficult to –t one last lipid molecule between the cylinders. At 20 ” separation there is a 10” ” free space between the two 5 radius cylinders. According to the crosssectional area per lipid (62.9 ”2) the radius of one DPPC is approximately 4.5 ”.17 This is basically the spatial range of the dominant contribution to the membrane response function sm(r) observed in Fig. 1. Thus a reasonable lengthscale for protein»protein interactions arises naturally in the present integral equation theory. The present model accounts only for steric excluded volume interactions of the protein with the hydrocarbon chains.In particular electrostatic interactions were ignored. Ben-Tal and Honig29 calculated the electrostatic contribution to helix»helix interactions using a continuum electrostatic model.29 Their results led to the conclusion that there is a non-speci–c attractive interaction on the order of 1 to 2 kcal mol~1 between transmembrane helices in an anti-parallel con–guration. This suggests that the magnitude of our lipid-mediated potential is thus on the order of the electrostatic interaction between a-helices embedded in a membrane. 4 Conclusion A theory for examining the structure of the hydrocarbon chains around a protein inclusion embedded in a lipid bilayer was developed on the basis of the hypernetted chain integral equation theory for liquids.The exact lateral density»density response function of the hydrocarbon core which is extracted from a molecular dynamics simulation of a pure lipid bilayer was used as an input in the calculations. Although the integral equation resembles that of a simple HNC theory for a two-dimensional liquid both intramolecular and intermolecular correlations of the lipid molecules were included through the response function s which was calculated using con–gu- m rations taken from the molecular dynamics simulation of a DPPC lipid bilayer membrane by Feller et al.17 171 Faraday Discuss. 1998 111 165»172 To illustrate our new approach we –rst examined the structure of the DPPC bilayer in the neighborhood of a hard 5 ” radius cylindrical inclusion which modeled an a-helical polyleucine protein.The results showed that the average lipid order is perturbed over a distance of 25 to 30 ” from the center of the protein. We then calculated the lipid-mediated protein»protein interaction and the results suggest that there is a free energy barrier for protein separations between 15 and 30 ” which inhibits protein»protein association. In contrast the interaction is attractive at distances between proteins which are less than 15 ”. In particular the calculations indicate that the lipid matrix gives rise to a free energy well of 8 kBT for the association of two hard cylindrical inclusions of 5 ” radius. It should be stressed that this result was obtained using a two-dimensional discrete grid of approximately 204” ” by 102 . More than 600 DPPC molecules with 24 000 water molecules would have been required to assemble an atomic model of a bilayer system of corresponding size for molecular dynamics simulations (i.e.more than 150 000 atoms). Therefore it is clear that the present approach permits the investigation of factors that are not readily accessible by straight molecular dynamics simulations. In the future the dependence of the lipid-mediated interaction upon the protein size will be investigated. In addition other familiar closures such as the Percus»Yevick (PY) integral equation, 23 will be examined. Finally further developments of the integral equation theory to include the lateral and transversal response of the bilayer are in progress. Acknowledgements We are grateful to S.Feller R. M. Venable and R. W. Pastor for making their trajectory of a DPPC bilayer membrane available. Financial support from NSERC (Canada) and FCAR (Queç bec) is acknowledged. B.R. is a research fellow of the Medical Research Council of Canada. M.J.Z. is an associate of the Canadian Institute of Advanced Research. Paper 8/07109H References 1 W. Kleeman and H. M. McConnell BBA 1974 345 220. 2 B. A. Lewis and D. M. Engelman J. Mol. Biol. 1984 166 203. 3 M. A. Lemmon and D. M. Engelman Quart. Rev. Biophys. 1994 27 157. 4 C. Nielsen M. Goulian and O. S. Andersen Biophys. J. 1998 74 1966. 5 R. B. Gennis Biomembranes Molecular Structure and Function Springer-Verlag New York 1989. 6 S. Marcelja Biochim. Biophys. Acta 1976 455 1. 7 H. Schroé der J.Chem. Phys. 1977 67 1617. 8 J. C. Owicki M. W. Springgate and H. M. McConnell PNAS 1978 75 1616. 9 J. C. Owicki and H. M. McConnell PNAS 1979 76 4750. 10 L. T. Pearson J. Edelman and S. I. Chan Biophys. J. 1984 45 863. 11 P. G. deGennes T he Physics of L iquid Crystals Oxford University Press London 1974. 12 M. Goulian R. Bruinsma and P. Pincus Europhys. L ett. 1993 22 145. 13 P. A. Kralchevsky V. N. Paunov and N. D. Denkov J. Chem. Soc. Faraday T rans. 1995 91 3415. 14 H. Aranda-Espinoza A. Berman N. Dan P. Pincus and S. Safran Biophys. J. 1996 71 648. 15 Biological Membranes. A molecular perspective from computation and experiment ed. K. M. In Merz and B. Roux Birkhauser Boston 1996. 16 E. Egberts and H. J. C. Berendsen J. Chem. Phys. 1988 89 3718. 17 S. E. Feller R. M. Venable and R. W. Pastor L angmuir 1997 13 6555. 18 T. B. Woolf and B. Roux Proteins Struct. Funct. Genet. 1996 24 92. 19 L. Shen D. Bassolino and T. Stouch Biophys. J. 1997 73 3. 20 D. P. Tieleman and H. J. C. Berendsen Biophys. J. 1998 74 2786. 21 J. P. Postma H. C. Berendsen and J. R. Haak Faraday Symp. Chem. Soc. 1982 17 55. 22 T. Sintes and A. Baumgaé rtner Biophys. J. 1997 73 2251. 23 J. P. Hansen and I. R. McDonald T heory of Simple L iquids 2nd edn. Academic Press Inc. San Diego 1986. 24 D. Chandler J. D. McCoy and S. J. Singer J. Chem. Phys. 1986 85 5971. 25 D. Beglov and B. Roux J. Chem. Phys. 1995 103 360. 26 D. Beglov and B. Roux J. Phys. Chem. 1997 101 7821. 27 L. W. Pratt and D. Chandler J. Chem. Phys. 1977 67 3683. 28 M. Frigo and S. G. Johnson ICASSP Conference Proceedings 1998 3 1382. 29 N. Ben-Tal and B. Honig Biophys. J. 1996 71 3046. Faraday Discuss. 1998 111 165»172 172

ISSN:1359-6640    

DOI:10.1039/a807109h

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Lipid packing stress and polypeptide aggregation alamethicin channel probed by proton titration of lipid charge Sergey M. Bezrukov,a R. Peter Rand,b Igor Vodyanoyc and V. Adrian Parsegiand a NICHD National Institutes of Health Bethesda MD 20892-0924 USA and St. Petersburg Nuclear Physics Institute Gatchina 188350 Russia b Biological Sciences Brock University St. Catharines Ontario L 2S 3A1 Canada c Office of Naval Research Europe L ondon UK NW 1 5T H d NICHD National Institutes of Health Bethesda MD 20892-5626 USA Receiøed 20th August 1998 Lipid membranes are not passive neutral scaÜolds to hold membrane proteins. In order to examine the in—uence of lipid packing energetics on ion channel expression we study the relative probabilities of alamethicin channel formation in dioleoylphosphatidylserine (DOPS) bilayers as a function of pH.The rationale for this strategy is our earlier –nding that the higher-conductance states corresponding to larger polypeptide aggregates are more likely to occur in the presence of lipids prone to hexagonal H -phase formation II (speci–cally DOPE) than in the presence of lamellar L -forming lipids (DOPC). In low a ionic strength NaCl solutions at neutral pH the open channel in DOPS membranes spends most of its time in states of lower conductance and resembles alamethicin channels in DOPC; at lower pH where the lipid polar groups are neutralized the channel probability distribution resembles that in DOPE. X-Ray diÜraction studies on DOPS show a progressive decrease in the intrinsic curvature of the constituent monolayers as well as a decreased probability of H -phase formation when the charged lipid fraction is II increased.We explore how proton titration of DOPS aÜects lipid packing energetics and how these energetics couple titration to channel formation. Introduction The evidence mounts. Membrane lipids are not just a –ller or an inert solvent for membrane proteins ; they are functionally involved. Interactions between lipids and embedded proteins control conformational equilibrium between diÜerent functional states of proteins. Among natural membrane proteins one clear example is the shift between the Meta-I and Meta-II forms of Rhodopsin with varied lipid species in the host membrane. The amount of the Meta-II form increases with the increase in phosphatidylcholine acyl chain unsaturation,1 thus demonstrating the important role of lipids in modulating membrane-signaling systems.Many other examples of the crucial role of lipid»protein interactions in enzymatic reactions and receptor regulation can be found in a recent review.2 It is well known that the conductance lifetime and formation ìon-rateœ of the channel-forming drug gramicidin A3 depend on host-lipid species. In addition to electrostatic eÜects of surface charge4,5 and poorly understood eÜects of neutral lipids6h10 on channel conductance extensive systematic studies of the gramicidin channel in diÜerent host-lipid compositions have shown that 173 Faraday Discuss. 1998 111 173»183 its properties can be controlled by purely mechanical parameters.11 It was shown that gramicidin channel life-time and free energy of dimerization are modi–ed by bilayer curvature stress12 and membrane tension13 in a quantitatively predictable way.Channels formed by the 20-amino acid peptide alamethicin14h16 also show properties that depend on membrane lipid composition17h20 or on tension applied to the bilayer.21 A clear correlation between the tendency of lipids to form the inverted hexagonal phase and the expression of higher-conductance states of alamethicin peptide channels in those lipids was demonstrated in studies with dioleoylphosphatidylethanolamine (DOPE)/dioleoylphosphatidylcholine (DOPC) mixtures.22 Alamethicin was inserted into bilayer membranes composed of lipids of empirically determined inverted hexagonal phase spontaneous radii.Lipids with diÜerent spontaneous radii form planar membranes with expectedly diÜerent degrees of stress of forcing lipids into a planar structure. It was found that this mechanical parameter of the host-lipid bilayer plays a crucial role in alamethicin channel formation. In particular states of higher conductance were found to be much more probable in DOPE a lipid of high curvature than in DOPC lipid of low curvature. In the case of mixtures the relative probability of states was a monotonic function of the DOPE/ DOPC ratio.22 In this paper whose preliminary version was reported elsewhere,23 we demonstrate that a continuous variation of factors that stress membrane structure can direct the conformational equilibrium of channel-forming peptides.Speci–cally we insert alamethicin into DOPS bilayers and change the bilayer surface charge and lipid head group electrostatic interactions with varied pH and varied salt concentration. Dramatic changes in relative probabilities of channel conductance states observed as a result of such manipulations provide further evidence of the importance of host-membrane mechanical parameters for channel protein function. By X-ray diÜraction we show that the decreased electrostatic energy of the polar surface (that goes with the decreased charge at low pH values) shifts DOPS from the purely lamellar form seen at neutral pH to an H phase of ever-higher spontaneous curvature. This shift agrees well with our II transport measurements performed on the same lipids.In 0.1»0.3 M sodium chloride solutions at neutral pH alamethicin channels exhibit the DOPC-like pattern expected for lamellar lipids ; in acidic solutions they show the DOPE-like pattern expected for H -prone lipids of high sponta- II neous curvature.22 Speci–cally we –nd that the higher conductance states of the channel are expressed much more at pH 2.0 than at pH 6.0. The corresponding observed 50-fold change in the relative probability of a particular state vs. the adjacent state suggests that there is a change of D4 kT in the diÜerence between free energies of adjacent states. As expected a qualitatively similar change was observed when salt concentration was increased to 1 or 2 M. The energies of the diÜerent functional states are the most important factors in protein regulation.24 For membrane proteins these energies depend on the lipid molecules outside the protein. We see now that it is possible to regulate the energies of these lipids themselves to modulate their in—uence on proteins. Materials and methods Alamethicin channels were inserted into ì solvent-free œ planar lipid bilayer membranes that had been formed by apposition of two phospholipid monolayers spread on aqueous solutions of sodium chloride (Baker Analyzed grade Baker Phillipsburg NJ USA). The monolayers were prepared from 10% DOPS or DOPE (Avanti Polar Lipids Alabaster AL USA) in pentane (HPLC grade Burdick and Jackson Muskegon MI USA). The Te—on chamber25 (after Montal and Mueller26) with two compartments of 1 ml was divided by 15 lm thick Te—on partition (CHEMFAB Merrimack NH) with a 60 lm diameter aperture.The aperture was pretreated with 1% solution of hexadecane (Aldrich Milwaukee WI USA) in pentane and dried during 10 min prior to monolayer opposition. The same partition was used throughout all measurements reported in this paper. Natural alamethicin (Sigma St. Louis MO USA) was added only to one side of a membrane from 10~5 M stock solution in ethanol to a –nal concentration of (1»3)]10~8 M. All experiments were done at 150 mV positive from the side of alamethicin addition and at a room temperature of (23^1) °C. Alamethicin concentration was adjusted to a concentration that gave –rst current bursts in about 20 min after peptide addition ; in this way we were able to monitor single- Faraday Discuss.1998 111 173»183 174 channel activity (no channel overlapping) for about 10 min. Ion currents ampli–ed with an Axopatch 200A integrating patch clamp ampli–er (Axon Instruments Foster City CA USA) were recorded with a sampling rate of 50 kHz into computer memory and simultaneously onto recordable compact discs. Statistical analysis of state probabilities was performed using direct comparison of the time spent by a channel at diÜerent conductance states (levels). First current histograms were plotted and appropriate windows around each state were determined. Second the total numbers of points within each such window were calculated and their ratios were taken to represent the relative probabilities of corresponding states.Each point in a relative probability graph represents averaging over more than 100 channels that were obtained typically from one membrane. A new membrane was formed for every pH or salt concentration. Results Typical recordings of alamethicin-induced currents in DOPS bilayers at diÜerent pH are shown in Fig. 1. They demonstrate that the probabilistic character of a conductance burst corresponding to a single alamethicin channel is very sensitive to membrane-bathing solution acidity. At relatively high acidity (pH 2.0) when lipid charge of the membrane is mostly neutralized by protons a typical channel undergoes many transitions between diÜerent conductance states. Higher conductance states (labeled 4 5) are well-expressed and are typically observed in every current burst.Increased pH and presumably increased lipid charge progressively suppress higher states. At pH 5.0 a typical channel goes only to Level 0 and back to the closed state. Note that the current burst at pH 2.0 represents a single ion channel of —uctuating size. Conductance increments corresponding to channel transition to the next higher-conductive state increase with the level number from 0.104 nS (background to Level 0 transition) to 0.51 nS (Level 4 to Level 5 transition). If the burst were representing several identical channels occurring at the same time the increments would be equal or would decrease with level number due to interference of ion currents in access areas. While the probabilistic character of a single-channel burst changes dramatically with pH (Fig.1) the acidity of the medium only slightly in—uences the channel conductance itself. Fig. 2 shows that at pH 2.0 all levels exhibit a conductance increase. This conductance increase is several times higher than the corresponding increase in solution speci–c conductivity (data not shown). The Fig. 1 Typical current bursts representing alamethicin channels in DOPS membranes bathed by 0.3 M NaCl at three diÜerent pH values. Current is displayed with a 50 ls resolution. Horizontal dotted lines with numbers show conductance statesœ (levelsœ) notation used in this paper. The pH-dependent character of current bursts corresponding to single alamethicin channels is clearly seen. At pH 2.0 the current burst always appears through the lowest conductance state Level 0 —uctuates between several higher conductance states (Level 1 to Level 5 or even higher) and then disappears.At pH 5.0 a typical channel is seen at Level 0 only. 175 Faraday Discuss. 1998 111 173»183 Fig. 2 Conductance of alamethicin channel levels in DOPS membranes bathed by 0.3 M NaCl as a function of pH. Due to the strong dependence of level probability on acidity of the medium (Fig. 1) conductance of higher levels could be measured only low pH. At pH above pH 3.5 levels higher than Level 1 were virtually nonexistent. Attempts to increase the number of channels per unit time to resolve these levels led to channel overlapping and smearing of current histograms. disparity is probably related to preferential transport of protons vs.sodium cations. At pH 3.0 lower levels show a small dip that re—ects titration of the membrane surface charge and corresponding depletion in counterion concentration. This eÜect is similar to the recently reported titration of gramicidin channel conductance.5 The ìsmoothœ dependence of channel conductance on the acidity of the medium is helpful for statistical analysis of relative probabilities. Fig. 3 demonstrates the results of such an analysis of relative probabilities to quantify the pH-dependence clearly seen ìby naked eyeœ in Fig. 1. The change in pH from 2.0 to 6.3 changes the relative probability of Level 1 vs. Level 0 observation (–lled symbols) by a factor of e4B50. Most of the probability change occurs between pH 2.0 and pK 4.0 that is within the range that includes the lipidœs hydroxy group a .5 Relative probability of Level 2 vs.Level 1 (open symbols) changes similarly and for some reason is very close to that of Level 1 vs. Level 0 in its absolute value. Fig. 4 shows that substituting 0.1 M NaCl for 0.3 M NaCl does not have any statistically signi–cant eÜect on the structure of channel probabilities. Within error bars the quantitative Fig. 3 Relative probability of channel levels in DOPS membranes bathed by 0.3 M NaCl as a function of pH. Filled symbols probability of Level 1 vs. Level 0; open symbols probability of Level 2 vs. Level 1. Higher-level relative probabilities decrease with pH increase ; most of the probability change (about four orders of natural logarithm base) occurs between pH 2.0 and 4.0.Faraday Discuss. 1998 111 173»183 176 Fig. 4 Relative probability of channel levels in DOPS membranes in 0.1 M NaCl shows behavior much like that in 0.3 M NaCl (Fig. 3). Filled symbols probability of Level 1 vs. Level 0; open symbols probability of Level 2 vs. Level 1. behaviour of levelsœ relative probability as a function of pH is the same at these salt concentrations. As shown the relative probability of Level 2 vs. Level 1 (open symbols) in 0.1 M NaCl closely follows the relative probability of Level 1 vs. Level 0 (closed symbols). To check for a possible direct in—uence of acidity or salt concentration on the probabilistic character of the alamethicin channel we ran a series of control measurements with neutral DOPE bilayers. Fig. 5 displays a typical channel in 2 M NaCl exhibiting well-de–ned conductance levels.It is seen that the higher levels are quite probable. This result agrees with earlier observations22 though in the present study we use a neutral form of alamethicin that by Glu18 to Gln18 substitution, 15,16 diÜers from peptide used earlier. Fig. 6 shows that conductance of channel levels is a monotonic function of salt concentration. Comparison to solution speci–c conductivity (solid line) indicates that channel conductance grows slower than does the solution conductivity. Similar to alamethicin channels in DOPS (Figs. 1 and 2) diÜerences between conductance levels diverge with level number (although the levels themselves are sublinear in salt concentration). In particular for 2 M NaCl these increments (Level 0 through Level 5 measured in nS) are 0.27 1.10 1.59 1.80 2.13.This means that the current burst shown in Fig. 5 does represent a single ion channel of —uctuating size and not a random overlap of several identical channels. Fig. 7 demonstrates that the probabilistic character of alamethicin channel reconstituted into uncharged lipid does not depend on salt concentration or on the shift of pH from a neutral to acidic value. Within experimental error changing sodium chloride concentration from 0.1 to 2.0 M or acidity from pH 6.2 to 2.5 does not in—uence levelsœ relative probabilities. This suggests that Fig. 5 Typical current burst of alamethicin-induced conductance obtained from a DOPE bilayer bathed by 2.0 M NaCl at pH 6.3.Channel ìswitches onœ through Level 0 and then —uctuates between diÜerent wellde –ned conductance states reaching Level 5 several times during its lifetime. The probabilistic character of the channel is very close to that of the alamethicin channel in DOPS at pH 2.0 (Fig. 1). 177 Faraday Discuss. 1998 111 173»183 Fig. 6 Conductance of alamethicin channel states as a function of sodium chloride concentration measured on DOPE bilayers at pH 6.3. The solid line gives bulk solution speci–c conductivity scaled in such a way that Level 5 conductance and solution conductivity coincide at 0.1 M concentration to facilitate comparison. The comparison shows that conductance of Level 5 is a weaker function of salt concentration than bulk conductivity. Fig. 7 Relative probability of channel conductance states in DOPE bilayers at diÜerent pH and sodium chloride concentrations.Within experimental error the relative probability of higher states does not depend on salt concentration or acidity of the medium if the channel is inserted into neutral lipid. Whatever pH or salt concentration relative probabilities in DOPE stay close to those in DOPS at pH 2.0»2.5 (Figs. 3 and 4). Fig. 8 Typical current burst of alamethicin-induced conductance obtained from a DOPS bilayer bathed by 2.0 M NaCl at pH 6.2. It is clearly seen that the high salt concentration increases the probability of higher conductance states (compare to the current recording of Fig. 1 for pH 5.0). However due to the appearance of strange conductance substates (shown by tilted arrows) it was impossible to quantify the salt eÜect in the alamethicin/DOPS system as done for DOPE (Fig.7). Faraday Discuss. 1998 111 173»183 178 the dramatic pH eÜects shown in Figs. 1 3 and 4 are related to lipid charge titration that aÜects channel behavior via lipid»channel interactions. In addition to lipid charge titration by proton one can think about the analogous action of high salt concentration. As clearly seen in Fig. 8 illustrating a typical current burst obtained from a DOPS bilayer bathed by 2 M NaCl at neutral pH we do observe a strong increase in higher level probabilities at high sodium chloride concentrations (compare to Fig. 1 recording for pH 5.0). Unfortunately the eÜect of salt could not be quanti–ed reliably because of additional strange (compared to the channel in Fig.5) sublevels in the alamethicin/DOPS system at high salt concentrations. Three of these sublevels are marked by tilted arrows (Fig. 8). Appearance of sublevels prohibited clear determination of level positions that is crucial to the subsequent statistical analysis. It should be noted that the bursts an example of which is presented in Fig. 8 were rare enough (separated on average by 10-fold longer periods of ì silent œ background recording) to exclude a trivial reason of several channels overlap. Discussion Many parameters in—uence the conformational equilibrium of membrane proteins. It is well known that a conformational transition can be triggered by a pH shift or by ligand and multivalent cation binding.The number of appropriate examples is overwhelming since these reactions constitute a basis for diverse physiological regulation at the cellular level.24,27 Much less studied are mechanisms of protein regulation by membrane lipids that act either directly or via the mechanical properties of host bilayers. Recent progress in this –eld demonstrates the possible ubiquity of such regulation. It has been shown that lipids modulate catalytic activity and binding properties of integral membrane proteins that include Insulin-R Na`/K`- ATPase Ca2`-ATPase GABA transporter acetylcholine receptor (Table 1 in ref. 2) and Rhodopsin, 1 to name just a few. Nevertheless the nature of the physical forces underlying lipid action is still a subject of considerable controversy.28 Membrane protein function can be modi–ed by changes in the mechanical properties of a host membrane e.g.by changes in spontaneous curvature of membrane lipids.12,22,29h33 There has been corresponding theoretical eÜort (e.g. ref. 11 34 35). The conformational equilibrium of a protein between diÜerent functional states is governed by the total free energy diÜerences between these states. Clearly if a conformational transition between states involves a change in the shape or length of the protein surface that is exposed to lipids this transition has to be sensitive to membrane mechanics. Although the idea is general the particular approaches permitting a quantitative description are model-speci–c. A correlation between packing stress and conformational equilibrium of a single membranebound polypeptide structure was –rst observed with alamethicin channels in planar bilayer membranes made from DOPE/DOPC mixtures.22 It was shown that an increase in the mole fraction of DOPE which favors a highly curved H -phase shifts the distribution of conductance levels II towards those of higher conductance.The following relationship between spontaneous curvature of the lipid and polypeptide aggregation in the membrane was established higher curvature stress promotes larger alamethicin aggregates. Now we demonstrate that with respect to the probabilistic character of the alamethicin channel the same charged lipid species can be made equivalent to DOPC or DOPE by changing bathing solution pH or salt concentration. Again we correlate ion channel function with the stress of forcing lipids of a given spontaneous curvature into a planar membrane form.The cartoon in Fig. 9 illustrates how a change in the charge of lipid head groups is able to change lipid spontaneous curvature. A useful notion in description of mechanical properties of a bilayer is the eÜective ìshapeœ of the membrane molecules. A cylindrical molecular shape (when the cross-sectional area of the polar head group is similar to the cross-sectional area of the acyl chains) will correspond to lamellar phases and stress-free packing into the bilayer form. At pH 6 and small enough salt concentration DOPS molecules have an approximately cylindrical eÜective shape because of repulsion of fully charged neighboring head groups.At pH 2 proton binding titrates out the head group charge; lipid shape is conical. Correspondingly DOPS hexagonal phase can go from a rather low spontaneous curvature at neutral pH to a high spontaneous curvature in an acidic environment. 179 Faraday Discuss. 1998 111 173»183 Fig. 9 Cartoon illustrating the proton-induced increase in spontaneous curvature of DOPS. Neutralization of the lipid head group charge by proton binding reduces head»head repulsion ; it thus eÜectively changes lipid molecule ìshapeœ. At a fully deprotonated charged state repulsion between DOPS head groups drives system into a lamellar structure ; however at high proton concentrations this repulsion is ìswitched oÜ œ so that a preferred packing is an H -phase of high curvature. II Strong X-ray diÜraction evidence Fig.10 supports this interpretation. Samples with an excess of water solution were used in diÜraction measurements and solution pH was adjusted before and checked after sample equilibration. It is seen that the Bragg repeat spacing for the DOPS hexagonal phase changes sharply between pH 2.5 and 4.0 in excellent agreement with transport measurements (Figs. 3 and 4). For samples above pH 4 X-ray scattering showed disorder characteristic of highly and irregularly separated bilayer membranes. The common but puzzling coexistence of hexagonal and lamellar phases at low pH was probably related to the small free energy diÜerence between these two lipid assemblies.36,37 A general thermodynamic analysis permits us to quantify the energetics of alamethicin channel regulation.Indeed because the diÜerent conductance levels of the channel are well de–ned states Fig. 10 Bragg repeat spacing of hexagonal and lamellar DOPS phases in excess solution. A sharp decrease in hexagonal spacing (increase in spontaneous curvature) occurring between pH 2.0 and 4.0 is clearly seen. The structure above pH 4.0 (that is designated as ìdisorderedœ) most probably corresponds to charged bilayers with irregular separation.40 The reasons for coexistence of hexagonal and lamellar phases at low pH are not yet clear. Faraday Discuss. 1998 111 173»183 180 it is possible to speak of the chemical potentials (or free energies) of the individual states ki\ k (pH kNaCl). The relative probability of two adjacent levels p(i i[1)\P i p(i i[1)\exp[[(ki[ki~1)/kT ] where k and T have their usual meaning of Boltzmann constant and absolute temperature.A change in the relative probability induced by a pH shift can be expressed as the ratio of relative probabilities at the two pH values and equals exp[[*(k pH-induced change in the statesœ chemical potential diÜerence. This ratio re—ecting total change in relative probability caused by acidity shift from pH 2 to 6 (Figs. 3 and 4) is measured to be close to 50. Thus i/Pi~1 is given then by (1) i[ki~1)/kT ]. Here *(ki[ki~1) is the (2) *(ki[ki~1) opH 2FhpH 6+4 kT The rate of change in relative probability with pH is an indication of diÜerence in the number of protons associated with the surface. By standard Gibbs»Duhem reasoning assuming that the active factors are proton and sodium activity the changes in state energy go as (3) dki\[nNaCl i dkNaCl[nH i dkH n where the functions NaCl i and nH i themselves depend on NaCl and H activity.These are the Gibbs excess numbers of sodiums or protons that are associated with the membrane»channel system when the channel is in state i. The change in the relative probability of two conductance levels i and i[1 depends on the diÜerence in these ns (4) d(ki[ki~1)\[(nNaCl i [nNaCl i~1 )dkNaCl[(nH i [nH i ~1)dkH When salt concentration is kept –xed but pH varied the change in relative probabilities gives us nH i [nH i ~1 a diÜerence in the number of protons associated with the system upon a transition from conductance state i[1 to conductance state i.From these equations and kH\kT ln[H] pH\[log[H] we obtain (5) H i [nH i ~1\[ L(ki[ Lk ki~1) \ L ln L p ln[H] (i i[1) \[ L ln ln 10 p(i L (pH) i[1) n H Analysis of the data for the Level 0 to Level 1 and Level 1 to Level 2 transitions around pH 3.0 (which corresponds to the maximum slope of the relative probability dependence on pH Fig. 3) gives nH i [nH i ~1\1.1^0.1. The change in the number of associated protons upon transition to a higher conductance state is thus positive. Higher conductance states by rearranging the whole channel/bilayer system accommodate more protons so that the increase in proton chemical potential makes these states more favorable. As a result higher conductance states are more expressed at high proton concentrations.The thermodynamic description shown above is very useful for quantifying the eÜect of pH on channel function and for restricting the number of possible models. However as usual thermodynamics does not elucidate a speci–c physical mechanism. To approach the mechanism of channel regulation by pH more structural knowledge of the conformation transformations of channel opening/closing and on lipid»peptide interactions is needed. Such knowledge is necessary to discriminate between stress of packing surface tension or other mechanical factor contributions. As recently pointed out on the basis of a careful study of diÜerent mechanical contributions to the energetics of protein inclusions into lipid bilayers,11 the increase in hydrophobic mismatch between the protein and the lipid by a mere 0.3 ” can change the equilibrium distribution between the corresponding protein states by a factor of 10.Also the energy of the protein-induced bilayer deformation can be as high as 2»3 kT per one lipid molecule. In this case the 4 kT eÜect reported in the present study can be explained by perturbation of a few lipid molecules only. It is worthwhile to compare the pH-shift-induced energy change found in our study to other characteristic energies in the system. The electrostatic energy of recharging of a single lipid head group in the fully charged DOPS bilayer (e.g. at pH 6.0) can be easily found as a product of the membrane surface potential t (e.g.ref. 38) and the elementary charge e (6) et 0 0\2 kT sinh~1M[8 kT ee0(Na)]~1@2pN 181 Faraday Discuss. 1998 111 173»183 Here p is the lipid charge surface density e is the dielectric constant and e is the permittivity of 0 free space. Taking p\0.25 C m~2 (one elementary charge per 64 ”2) and [Na]\6.02]1025 m~3 (0.1 M NaCl) we obtain et0\5.2 kT . This energy compares well with the 4 kT change found for the energy of the alamethicin state to state transitions [eqn. (2)]. The work of forcing one lipid molecule from the hexagonal H -phase into the lamellar phase II can be estimated as36,39 (7) E\ akc 2R0 2 where k is the monolayer bending modulus (about 10 kT 37) a is the area per lipid molecule and c R is the radius of spontaneous curvature. Taking a\64 ”2 and deducing the radius of sponta- 0 neous curvature from the Bragg spacing for the hexagonal phase at pH 2.0 (Fig.10) we get EB0.3 kT . This energy is about an order of magnitude less than the acidity-induced change in the energy of alamethicin state-to-state transitions. However many lipid molecules are in direct contact with the alamethicin aggregate. If in addition the lipids that are to be perturbed by protein conformational change extend over distances of many lipid molecules,11 this estimate is reasonable. To conclude varying the pH changes the probabilistic character of alamethicin channels in a way that correlates with the pH-induced changes in the nonlamellar tendency of the host-lipid. This correlation made for the charged lipid DOPS agrees with the logic of our conclusions drawn from earlier observations on alamethicin in mixtures of neutral lipids.22 The channelœs response to pH is all the more impressive given the fact that the present measurements were made on a neutral form of alamethicin that does not possess any residues that can be titrated in this pH range.Still there is a pronounced 50-fold eÜect on channel state-to-state transitions from a pH shift of about two units. Whatever the mechanism lipid charge titration modi–es channel structural equilibrium. This –nding probably unveils an additional previously unrecognized way of pH regulation in membrane transport. R.P.R. acknowledges the –nancial support of the Natural Sciences and Engineering Research Council of Canada and the expert assistance of Mrs.Nola Fuller. References 1 B. J. Litman and D. C. Mitchell L ipids 1996 31 S193. 2 A. Bienvenue and J. S. Marie Curr. T op. Membr. 1994 40 319. 3 R. E. Koeppe II and O. S. Andersen Annu. Rev. Biophys. Biomol. Struct. 1996 25 231. 4 H. J. Apell E. Bamberg and P. Lauger Biochim. Biophys. Acta 1979 552 369. 5 T. K. Rostovtseva V. M. Aguilella I. Vodyanoy S. M. Bezrukov and V. A. Parsegian Biophys. J. 1998 75 1783. 6 E. Bamberg and P. Lauger Biochim. Biophys. Acta 1974 367 127. 7 E. Neher and H. Eibl Biochim. Biophys. Acta 1977 464 37. 8 V. Fonseca P. Daumas L. Ranjalahy-Rasoloarijao F. Heitz R. Lazaro Y. Trudelle and O. S. Andersen Biochemistry 1992 31 5340. 9 J. A. Killian Biochim. Biophys. Acta 1992 1113 391. 10 J. Girshman D.V. Greathouse R. E. Koeppe II and O. S. Andersen Biophys. J. 1997 73 1310. 11 C. Nielsen M Goulian and O. S. Andersen Biophys. J. 1998 74 1966. 12 J. A. Lundbaek and O. S. Andersen J. Gen. Physiol. 1994 104 645. 13 M. Goulian O. N. Mesquita D. K. Fygenson C. Nielsen O. S. Andersen and A. Libchaber Biophys. J. 1998 74 328. 14 M. S. P. Sansom Prog. Biophys. Mol. Biol. 1991 55 139. 15 G. A. Wooley and B. A. Wallace J. Membrane Biol. 1992 129 109. 16 D. S. Ca–so Annu. Rev. Biophys. Biomol. Struct. 1994 23 141. 17 G. Boheim J. Membrane Biol. 1974 19 277. 18 R. Latorre and J. J. Donovan Acta Physiol. Scand. (Suppl.) 1980 481 37. 19 J. E. Hall I. Vodyanoy T. M. Balasubramanian and G. Marshall Biophys. J. 1984 45 233. 20 S. Stankowski U. D. Schwarz and G.Schwarz Biochim. Biophys. Acta 1988 941 11. 21 L. R. Opsahl and W. W. Webb Biophys. J. 1994 66 71. 22 S. L. Keller S. M. Bezrukov S. M. Gruner M. W. Tate I. Vodyanoy and V. A. Parsegian Biophys. J 1993 65 23. Faraday Discuss. 1998 111 173»183 182 23 S. M. Bezrukov I. Vodyanoy P. Rand and V. A. Parsegian Biophys. J. 1995 68 A341. 24 B. Alberts D. Bray J. Lewis M. RaÜ K. Roberts and J. D. Watson Molecular Biology of the Cell Garland Publishers New York 1994. 25 S. M. Bezrukov and I. Vodyanoy Biophys. J. 1993 64 16. 26 M. Montal and P. Mueller Proc. Natl. Acad. Sci. USA 1972 65 3561. 27 Cell Physiology ed. N. Sperelakis Academic Press San Diego 1988. 28 O. G. Mouritsen and M. Bloom Annu. Rev. Biophys. Biomol. Struct. 1993 22 145. 29 S. M. Gruner Proc. Natl. Acad. Sci. USA 1985 82 3665. 30 C. D. McCallum and R. M. Epand Biochemistry 1995 34 1815. 31 C. D. Stubbs and S. J. Slater Chem. Phys. L ipids 1996 81 185. 32 R. M. Epand Chem. Phys. L ipids 1996 81 101. 33 J. A. Lundbaek A. M. Maer and O. S. Andersen Biochemistry 1997 36 5695. 34 N. Dan A. Berman P. Pincus and S. A. Safran J. Phys. II France 1994 4 1713. 35 N. Dan and S. A. Safran Isr. J. Chem. 1995 35 37. 36 M. M. Kozlov S. Leikin and R. P. Rand Biophys. J. 1994 67 1603. 37 S. Leikin M. M. Kozlov N. L. Fuller and R. P. Rand Biophys. J. 1996 71 2623. 38 S. McLaughlin Curr. T op. Membr. T ransport. 1977 9 71. 39 W. Helfrich Z. Naturforsch. 1973 28C 693. 40 N. Fuller personal communication. Paper 8/06579I 183 Faraday Discuss. 1998 111 173»183

ISSN:1359-6640    

DOI:10.1039/a806579i

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Structure-based prediction of the conductance properties of ion channels Oliver S. Smart,*a Guy M. P. Coates,a Mark S. P. Sansom,b Glenn M. Alderc and C. Lindsay Bashfordc a School of Biochemistry T he University of Birmingham Edgbaston Birmingham UK B15 2T T . E-mail o.s.smart=bham.ac.uk b L aboratory of Molecular Biophysics T he Rex Richards Building University of Oxford Oxford UK OX13QU c Division of Biochemistry St. Georgeœs Hospital Medical School Cranmer T errace L ondon UK SW 17 0RE Received 1st September 1998 The HOLE procedure allows the prediction of the absolute conductance of an ion channel model from its structure. The original prediction method uses an empirically corrected Ohmic method. It is most successful with predictions being reliable to within a factor of two.A new modi–cation of the procedure is presented in which the self-diÜusion coefficients of water molecules from molecular dynamics simulation are used to replace the empirical correction factor. A ììpredictionœœ of the conductance for the porin OmpF by the new method is made and shown to be very close to the experimental value. HOLE also allows the prediction of the eÜect that the addition of non-electrolyte polymers will have on channel conductance. The method has great potential to yield structural information from data provided by single channel recordings but needs further validation by making measurements on channels of known structure. Preliminary results are given of single channel records establishing the eÜects of non-electrolytes on the conductance of gramicidin D channels.As an example of the potential uses of the procedure application is made to examine the oligomerization of a-toxin (a-hemolysin) channels. A model for the a-toxin hexamer based on the crystal structure for the heptamer is generated using molecular mechanics methods. The compatibility of the structures with single channel conductance data is assessed using HOLE. 1 Introduction Ion channels are an important class of proteins allowing the passage of ions through the hydrophobic barrier presented by lipid bilayers.1 They are commonly involved in communication and regulation processes and are the target for many drugs. Because of the difficulties in experimentally determining the three dimensional structures of membrane proteins it is common to propose channels models on the basis of low resolution data based on techniques such as mutagenesis2 or electron microscopy data.3 An important part of such a modelling procedure is to validate the –nal models.Even where a high-resolution structure for a channel is available it is not always immediately obvious as to whether this represents the conducting form. A reliable tool for the prediction of the conductance properties of a channel model on the basis of its structure would be invaluable in addressing these questions. This paper describes in Sections 185 Faraday Discuss. 1998 111 185»199 2 and 3 steps taken in the development of such a tool. Section 4 describes attempts to extend the method to allow the prediction that the addition of non-electrolyte molecules has on channel conductance including preliminary results for gramicidin D.Finally the method is applied to examine the complex question of the stoichiometry of a-toxins in lipid bilayers. 2 The HOLE method The HOLE method4 has become widely used in the analysis of ion channel structures. It uses Monte Carlo simulated annealing to maximize the radius of a sphere that is forced to squeeze through the van der Waals surface of a channel. The user de–nes a channel direction vector that is used to de–ne a stack of planes with a uniform separation s. A separate optimization is conducted for each plane maximizing the radius of the probe sphere while avoiding any overlap with the atoms of the channel and keeping the centre of the sphere on the plane.Although the use of a spherical probe is reasonable for narrow channels it becomes problematic for larger anisotropic pores. For this reason an adaptation was made,5 in which the probe sphere is replaced by a capsule (more accurately termed a sphereocylinder). Although this feature allows the measurement of anisotropy the routine suÜers from stability problems in the irregular channels that result from molecular dynamics simulations. For this reason work in progress aims to adapt the Connolly procedure6 to more accurately delimit larger channels. (1) (2) 3 Predicting absolute conductance properties using HOLE The method5 is based on simple Ohmic considerations. The HOLE program can be thought of as measuring the cross-sectional area A(z) of a pore as a function of distance along the channel direction vector z.Consider the pore to be –lled with an electrolytic solution of resistivity o. A reasonable approximation of the conductance (inverse resistance) of such a column of —uid is given by G z/high macro ~1 \ ; os/A(z) z/low where s is the width between parallel planes used in the HOLE method. This assumes that the conductivity of an ionic solution within the channel is equal to that of bulk solution. The assumption would be true if ion channels had macroscopic dimensions (much larger than a water molecule). In practice it is found that real channels have a conductance which is around –ve times lower than that expected from the above equation giving the macroscopic limit Gmacro .In order to make a reasonable estimate of the conductance of a channel an empirically based correction factor is used Gpred\Gmacro/C(m) where m is a characteristic parameter for the channel e.g. minimum radius. Although a number of characteristic parameters have been tested,5 a constant correction equal to 5.6 performs well. The prediction routine was tested on all channel-forming proteins and peptides where both a highresolution structure and conductance data are available. Results are given in Fig. 1. Overall the algorithm yielded good results with predictions accurate to within an average factor of 1.8 to the experimental values. This accuracy is sufficient to make the method a useful part in validating model structures.3.1 Using molecular dynamics simulation data to improve predictions The use of empirical correction factors to Ohmic predictions of absolute conductance although remarkably successful is limited reducing the physical insights5 available and is unlikely to be able to account for the subtle diÜerences between diÜerent channels. Analysis of the motion of water molecules during molecular dynamics (MD) simulations provide an alternative approach while avoiding the necessity of adopting a full scale microscopic simulation of the complete ion translocation process inherent in an ab initio prediction of conductance.7 Faraday Discuss. 1998 111 185»199 186 Fig. 1 Prediction of channel conductance using the 1st generation empirical correction function (adapted from Smart et al.5).Filled circles mark results for experimentally determined models. Open circles mark results for model channels. MD methods use the numerical solution of Newtonœs equations of motion to follow the course of the motion of a protein and its surroundings as a function of time under physically realistic conditions. Breed et al.8 used methods to analyse the restriction on the motion of water molecules within simpli–ed ion channel models. The self-diÜusion coefficient of water molecules from within the pore was reduced by around a factor of –ve compared to the value in bulk.8 The similarity of this reduction to the empirical correction factor used in HOLE led to the proposal that data from MD could be used instead.5 We replace the assumption that the resistivity of the ionic solution within the pore is equal to that of bulk solution (used to date with the consequent necessity of empirical correction functions).Instead the resistivity is made to be a function of z the coordinate along the channel vector and is assumed to vary in inverse proportion to the change in the water self-diÜusion coefficient derived by MD simulation (3) oMD(z)\obulk D D bulk z(z) where D is the self-diÜusion coefficient in the direction z averaged over all water molecules at a z particular z and over the MD run. Dbulk is the water self-diÜusion coefficient derived from bulk phase using the same water model as the ion channel run. The approximation in this procedure is that the reduction in the self-diÜusion coefficients of ions within the channel will be equal to that of water molecules.A MD-corrected conductance prediction GMD can then be made (4) H low) ] GMD ~1\Gz/ ; z high oMD A( ( z z ) )s ] o 4 MD r@( ( z zlow) o 4 MD r@( ( z zhigh) z/zlow high) where s is the width of the slabs being considered. The two last terms represent access resistance9 where r(zlow) and r(zhigh) are eÜective pore radii at the ends of the channel. The access resistance which typically contributes around 10% of the total resistance was previously ignored in eqn. (1) being eÜectively absorbed by the large empirical correction factor.5 To gauge the eÜectiveness of the modi–ed procedure a ììpredictionœœ is made for the E. coli porin OmfP based on a MD simulation performed by Tieleman and Berendsen,10 using the X-ray 187 Faraday Discuss.1998 111 185»199 (D Fig. 2 Using MD simulation data to make a prediction of the absolute conductance of porin OmpF. The solid line and left-hand ordinate show the eÜective pore radius pro–le of the crystal structure11 found by the capsule option of HOLE. The dashed line and right-hand ordinate show the reduction in the water selfz/ Dbulk) diÜusion coefficient for this structure obtained by Tieleman and Berendsen.10 crystal structure determined by Cowan and co-workers.11 The simulation was performed on the complete OmpF trimer in POPE bilayer/water box with data being collected over a 1 ns period.10 Peter Tieleman kindly supplied the self-diÜusion coefficient of the water molecules in the direction of the channel vector averaged over the trimer.Fig. 2 shows the marked reduction in the mobility of water molecules within the pore. The data on the reduction in the mobility of water molecules were combined with the eÜective pore radius pro–le for the crystal structure measured by the capsule option of HOLE (Fig. 2) and used in eqn. (3) and (4). This results in an expected conductance of 850 pS for 1 M KCl solution which is encouragingly close to the experimental value5,12,13 of 700 pS. Work in progress aims at extending the technique to be generally applicable. The method will be tested by application to all the systems illustrated in Fig. 1. Short molecular dynamics runs are desirable to allow general applicability.At present a conductance prediction can be made in a few tens of minutes processing time on a modern workstation. Although including molecular dynamics in the process will by necessity increase the requirement a reasonable limit is a few tens of hours of processing time. The use of an explicit lipid environment in the simulation is excluded by this limit. Although the explicit modelling of lipid increases the realism of the simulation it can be difficult to set up the initial ensemble as demonstrated by the variety of approaches adopted by diÜerent workers.10,14,15 As the simulations only require the water dynamics to be realistic it is expected that excluding the relatively distant lipid will not be too severe an approximation. This is borne out by a comparison of results obtained for water behaviour from alamethicin simulations using an explicit lipid representation with earlier applications without.16 Without an explicit lipid representation restraints on protein backbone atoms will be necessary.17 An issue to be explored is the eÜect of the imposition of restraints during simulation. Restraints are desirable in maintaining the structure close to the starting conformation as all MD simulations tend to suÜer from problems of drift. However restraints on the protein may seriously aÜect the dynamic mobility of the water within the pore particularly for narrow channels such as gramicidin. 4 Structure-based prediction of the eÜect of adding non-electrolytes on conductance Measurements of the eÜect of the addition of uncharged polymers to the medium in single channel conductance measurements can be used to derive information about the pore dimensions of a channel.5,18,19 The standard experimental protocol is to add 15 or 20% w/v polyethylene glycol Faraday Discuss.1998 111 185»199 188 (PEG) to the conducting medium. The variation in channel conductance is measured as a function of the molecular weight of the PEG used. The bulk conductivity of an ionic solution is reduced in the presence of PEG the magnitude of the reduction being broadly independent of the molecular weight provided a constant weight fraction is maintained.18 A recent development of the procedure allows the determination of the eÜect of adding diÜerent molecular weights on either side of the lipid bilayer,20 which allows more information to be derived.When a low molecular weight PEG is added to a wide channel the conductance of the channel will drop by the same factor as the change in bulk conductivity. But as the molecular weight of the PEG is increased the polymer will be progressively excluded from the channel. This will result in a progressive increase in the conductance measured. Results are normally interpreted using the hydrodynamic radii of the PEG or other polymer used. In the limit of very large molecular weight PEGs it is possible to observe an increase in conductance.21 This has been proposed to be because PEG is dehydrating and increases the water activity with the pore from which it is excluded. The increase in conductance depends on the ratio between the channel and access resistance and will be larger for thinner channels.5 The HOLE conductance prediction method has been adapted to predict the eÜect that PEGs will have on conductance.Full details are described in another publication,5 but broadly a conductance prediction is made using HOLE and following eqn. (1) with the modi–cation that the channel is divided into areas which are accessible to probe spheres of a radius RPEG which is set equal to the hydrodynamic radius of the polymer being considered and areas which are not. The former are assigned a resistivity of the PEG bulk solution and the latter a lower resistivity. The conductance prediction is then made as before but no empirical correction factor is required because results are expressed as a ratio of the polymer free case.An application5 of the procedure has been made to assess the eÜect of PEG on the conductance of cholera toxin B by comparing the value computed from the X-ray structure,22 with that 5 obtained experimentally.23 The results are most encouraging with the prediction falling within experimental error (Fig. 3). A further application was comparing results found for models of diÜerent alamethicin conducting states,24 with experimental observations21 arriving at the conclusion that the experiments were consistent with the barrel-stave hypothesis.5 Encouraged by these two successes we have decided to make a systematic study. The aim is to collect polymer-addition data for every ion channel for which a high resolution structure is available to validate the techniques and give real con–dence in the interpretation of data for biologically important channels.The next section describes preliminary results for the –rst study undertaken in this scheme. Fig. 3 Comparing the eÜect of non-electrolytes on experimental conductance23 of cholera toxin B5 with the expected result using HOLE (solid line) based on the X-ray crystal structure.22 189 Faraday Discuss. 1998 111 185»199 4.1 The eÜect of non-electrolytes on the conductance of gramicidin D An important step in the validation of the non-electrolyte addition technique is to apply the method to a narrow channel of known structure. Gramicidin D is the archetypal model ion channel25 with a high resolution solid-state NMR structure available for the channel forming conformation.26 The conductance pathway is narrow with a minimum HOLE pore radius of 1.15 ”,4 which contains a single –le of water molecules.We have started to make measurements of the eÜect of non-electrolyte on gramicidin conductance. Preliminary results are reported here. Gramicidin D (ICN) was incorporated into bilayers of diphytanoylphosphatidylcholine (Avanti Polar Lipids) across a 10»20 lm aperture separating two Te—on chambers as described previously. 19 Current was recorded in voltage-clamp mode via Ag/AgCl electrodes in each chamber. Fig. 4 shows typical traces obtained from membranes containing a few gramicidin channels in the absence and presence of 20% v/v PEG 10 000. Fig. 5 shows the results found and provides a comparison with the results expected using the prediction routine described above with the experimental channel structure.26 The results are most interesting.It is encouraging to observe that the addition of high molecular weight PEG results in an increase in channel conductance. This counterintuitive eÜect can be ascribed to an increase in the water activity within the channel5,21 and has previously been observed for the low conductance states of alamethicin.21 The value found is in approximate accordance with that expected on the basis of the structure. This early result encourages our forecast5 that the asymptotic value may be the most useful parameter in obtaining structural information. More detailed experiments are required to test this assertion.The eÜect that low molecular weight polymers have on conductance (Fig. 5) is more surprising. Both glucose and PEG300 reduce channel conductance by around half the maximum (bulk) eÜect. On the basis of the ìì resting œœ structure of gramicidin in lipid it can be expected that a molecule as large as glucose would not be able to penetrate the channel at all. We know that gramicidin readily conducts cations up to the size of formamidinium27 but is blocked by guanidinium.27 However these ions are still considerably smaller than glucose or PEG300. This makes it most unlikely that gramicidin would be —exible enough to allow the passive diÜusion of glucose into the channel. If the eÜect of glucose and PEG300 is not due to non-electrolyte permeation what is its cause? A possibility is that it may be due to a much larger than expected access resistance at the channel mouths.Gramicidin is known to readily form channels in lipids with thicknesses of up to 40 ”,28 despite having a length of only 26 ”.4 It has been proposed that the channel may cause ììcrimpingœœ of the lipid.29 This local thinning would create ìì vestibules œœ which would be accessible Fig. 4 Gramicidin channels in diphytanoylphosphatidylcholine bilayers in the presence of PEG. Gramicidin D (\10~12 M) treated bilayers in 1 M KCl 0.005 M Hepes containing 20% v/v PEG10000 pH 7.4 at room temperature. The voltage applied is indicated by the lower trace. Faraday Discuss. 1998 111 185»199 190 Fig. 5 The eÜect of non-electrolytes on the single channel conductance of gramicidin.The results are preliminary. Panel (A) shows the eÜect of adding PEGs and panel B sugars. Circles mark individual experimental results and the solid lines show the eÜect expected using the HOLE prediction routine5 on the basis of the solid state NMR structure of Ketchem et al.26 In each case HPEG and HNON are the theoretical limits to the ratio.5 to small polymers but not large ones. The magnitude of this eÜect is testable by making measurements with lipid bilayers of diÜerent thickness. This eÜect may contribute to the large impact of smaller polymers but is unlikely to be the principal cause. This can be seen by noting that PEG300 has approximately 50% of the result expected if it were able to permeate the whole channel.It is difficult to imagine access resistance ever having such a signi–cant impact for a channel as narrow as gramicidin. The likely source is the breakdown of the simplistic model of PEG-accessible regions of high resistivity switching instantly to inaccessible regions with low resistivity. It is likely that nonelectrolytic polymers have signi–cant eÜect in the region around them. Following the philosophy described in Section 3 we plan to undertake molecular dynamics simulations to gauge this eÜect. We hope this will provide data to produce a reliable prediction method which would provide a very valuable aid to modelling studies. In the meantime we urge caution in the interpretation of polymer addition experiments particularly for narrow channels. Using the simple interpretation used by most authors to date the data presented here would likely be interpreted as representing a channel with a characteristic radius above 4 ”.This clearly contradicts a wealth of evidence on gramicidin from other sources. 191 Faraday Discuss. 1998 111 185»199 5 Examination of the oligomerization of the a-toxin channel 5.1 Background seven vs. six a-Toxin (also known as a-hemolysin) is one of a number of toxins secreted by the bacterium Staphylococcus aureus.30 a-Toxinœs cytotoxicity arises primarily from its ability to form pores in cell membranes.30 The crystal structure of a-toxin in what appears to be a pore forming conformation has recently been solved by X-ray crystallography,29 at a resolution at 1.9 ”. This structure31 consists of seven identical subunits each of which consists of a large globular bsandwich domain and a 42 amino acid b-hairpin loop.The subunits are arranged in a ììbunch of tulips œœ pattern where the head domains lie on the outer surface of the membrane and the bhairpin loops bunch together to form a 14 strand b-barrel ììstemœœ which penetrates through the membrane (Fig. 6). The mechanism of a-toxin channel assembly and pore formation has been extensively studied.32h35 A number of separate stages have been identi–ed.32,34 Initially a-toxin monomers in a pre-pore conformation aggregate on the surface of the target membrane.34 A number of the monomers then assemble to form a pre-pore oligomer.34 The oligomer then undergoes a conformational change to the active channel form with little change in the overall proportions of secondary structure.32 This conformational change involves the loop of 42 amino acids inserting into the membrane and adopting a b-hairpin structure.31 It is likely that in the pre-pore oligomer these residues adopt Greek-key motif similar to that found in anthrax protective antigen.36,37 The bhairpins from each monomer come together to form a transmembrane b-barrel which provides a path for ion conductance in a similar manner to the porins.Although the crystal structure of a-toxin showed it to be heptameric,31,33 previous biochemical studies including cross-linking studies32 and electron microscopy (EM)38 indicated that the assembled conformation in lipid vesicles was hexameric. A recent atomic force microscopy (AFM) Fig.6 The construction principles of the a-toxin heptamer structure.31 Produced using the molscript program.52 Faraday Discuss. 1998 111 185»199 192 study39 of a-toxin in lipid bilayers under physiological conditions has revealed that a-toxin can form long lifetime hexamers but no evidence was found of the existence of heptamers. This is in direct contrast to Yang and co-workers40 who applied AFM to a mutant form of the protein which is trapped in the pre-pore oligomer and found it to be heptameric. In view of the current debate on the stoichiometry of a-toxin it was decided to construct a model for a hexameric form based on the ììdesign principles œœ revealed by heptameric crystal structure. The compatibility of the experimental heptamer structure and the model hexamer with the single channel conductance data available is then assessed using the HOLE techniques described above.5.2 Producing a model for a-toxin hexamer 5.2.1 The pore domain. In the heptameric structure the transmembrane domain is formed by each subunit donating an amino acid loop to form a 14 stranded b barrel. It was assumed that a hexameric toxin would form a pore along similar principles forming a 12-stranded barrel. A great deal is known about the geometric factors that constrain the structures of b barrels.41,42 The overall structure of a b barrel can be described by two parameters the number of strands (N) and the stagger of the barrel (S).41 The stagger number S is a measure of how far the b barrel strands tilt relative to axis of the barrel.Certain combinations of S and N are favoured in b barrels as they maximize inter-strands hydrogen bonding and favourable sidechain contacts on the inside of the barrel but do not unduly strain the backbone conformation. Murzin et al.41 showed theoretically that for large b barrels with a central water-–lled cavity S\N]4 resulted in the best geometry. This combination has been found for the E. coli porins OmpF,11 PhoE11 and Rhodobacter capsulatas porin,43 which crystal structures reveal to have N\16 and S\20. The structure of maltoporin44 has a larger barrel with N\18 and S\22. The a-toxin heptamer structure is an exception to this rule with N\14 and S\14.31 To explore the possibility that the hexamer may adopt a diÜerent stagger from the heptamer two separate 12 stranded b barrels were constructed.One was made to have S\12 following the example of the heptamer and the other was made to adopt a great stagger with S\16. Initial coordinates for each conformation were generated using existing models of alanine decamers b barrels with (N\12 S\12) and (N\12 S\16) produced by Sansom and Kerr45 by molecular dynamics simulated annealing. To produce barrels of the required length (20 amino acids) the model barrels were duplicated and joined end to end. Loops of three amino acids in a b turn conformation were added between alternate strands. Individual amino acids were then ììmutatedœœ from alanine to the a-toxin pore domain sequence using the QUANTA package (Molecular Simulations Inc).This resulted in starting conformations for regular (N\12 S\12) and (N\12 S\16) b barrels with good hydrogen bond geometry. The models consist of six separate polypeptide strands with the same sequence and broad topology as the a-toxin pore domain of the heptamer crystal structure. Each of the models was then re–ned using the X-PLOR package.46 The PARAM19 extended atom parameter set was used. Non-crystallographic symmetry was imposed between the six subunits. This forces the individual monomers to adopt a similar conformation resulting in a regular geometry. The models were subjected to 10 000 steps of Powell energy minimization,47 to remove bad contacts. Following this the non-crystallographic restraints were removed and a further 5000 steps of minimization were undertaken.The (N\12 S\16) more staggered conformation was found to have a larger potential energy than the (N\12 S\12) (Table 1). A more detailed comparison between the diÜerent terms contributing to the potential energy function shows that of the bonded terms only the bond angle is markedly diÜerent. The absolute values of the energy show that both conformations are relatively strained and the diÜerence in bond angle energy that the more staggered conformation is more strained than the less. Most of the potential energy diÜerence lies in the non-bonded terms. In particular a large diÜerence in the van der Waals term reveals poorer side chain packing. The smaller diÜerence in electrostatic energy shows that the hydrogen bonds between main chain groups within the barrel have a worse geometry in the more staggered conformation.193 Faraday Discuss. 1998 111 185»199 Table 1 XPLOR potential energy in kcal mol~1 for a-toxin hexamer models Electrostatic Van der Waals Total energy Model Bond angle term Bonds and torsions 308 379 465 466 [12 119 [11 681 [1759 [1590 [13 104 [12 424 (N\12 S\12) (N\12 S\16) The relative stability of the two diÜerent staggers is in contrast to the results found for the model polyalanine barrels45 used in constructing the domain. This eÜect can be seen to arise from two principal sources. An important diÜerence is that the polyalanine barrel is constructed from individual unconnected strands whereas the a-toxin barrel is composed of two-stranded covalently linked loops.For the barrel to be regular (i.e. the monomers to be quasi-equivalent) geometrical considerations necessitate that S\cN where c is an even integer. Other geometries such as S\N]4 can only be achieved if at least one of the loops (i.e. two strands of the barrel) are translated along the barrel axis relative to the other strands. Such a distortion is observed in the (N\12 S\16) structure modelled here and leads to the strain discussed above. While such an arrangement may be allowed in soluble proteins such a distortion would be highly unfavourable in the case of a-toxin. The ends of the barrel are clearly delimited by a ring of charged residues at one end (Asp127 Asp128 and Lys131) and a ring of aromatic residues at the other (Tyr118 and Phe120).Any distortion of the protein loops would mean that the charged residues would penetrate into the core of the membrane or the aromatic residues would be exposed to the extra-membrane environment. It is possible to exclude the S\2N and S\4N as the barrels would be too short to span the membrane. The sequence of the pore domain provides an additional feature which favours the S\12 form. One of the major factors in restricting the possible geometries of b barrels is the packing of the amino acids sidechains on the inside of the barrel. The S\N]4 stability rule arises partly from this eÜect.41 Unusually the pore domain is rich in glycine residues which are generally positioned towards the barrel centre31 (Fig. 7). This in eÜect releases the geometric constraints on the barrel allowing it to adopt other conformations.By these considerations the (N\12 S\12) b barrel was taken as the model for the pore domain of the putative a-toxin hexamer and attention turned to constructing the full oligomer from this core. 5.2.2 Adding the head group. The starting assumption was that the individual head group domains in the hexamer would adopt similar conformations to those of the heptameric crystal structure. The six individual head domains were then packed together using a geometrical procedure coded as a CHARMm48 script. The starting point was the structure for a single head domain (residues 1»108 and 152»293) taken from the crystal structure.31 Five further copies of this structure were generated on top of each other ; with each copy rotated 60° with respect to the previous copy around their common centroid.The six head groups were progressively moved outwards from the centroid in 0.5 ” steps. The direction of movement was such that the orientation of the monomers with respect to the centre of the channel was kept the same as that found in the crystal structure. The potential energy of complex was evaluated using the CHARMm potential energy function at each step during the transformation. The process resulted in a clear potential energy minimum at channel centre»centroid distance of 30 ” (data not shown). This conformation was adopted as a starting structure for the hexameric head group. To remove local bad contacts between the monomers the structure was subjected to energy minimization using the steepest descents and adopted basis set Newton»Raphson techniques implemented with CHARMm.48 Visual comparison between the model for the hexameric head group model developed here and the model hexameric b barrel stem described in Section 5.2.1 showed that they were compatible.The head and stem were joined together manually using the quanta package. The complete molecule was then subject to energy minimization. The local geometry of the resulting molecule was then checked using the PROCHECK package.49 A few improbable or unphysical side chain Faraday Discuss. 1998 111 185»199 194 Fig. 7 The eÜects of changing the stagger number for N\8 b barrel. The diagrams on the left show the topology of the barrel when it is unrolled.In each case strand b1 is shown twice. Empty circles mark residues whose side chains lie above the page whereas –lled dots show side chains below the page. Panels A and B show the (N\8 S\8) structure. Panels C and D show the more staggered (N\8 S\12) structure. Figure adapted from Sansom and Kerr.45 geometries were found. There were corrected using manual repositioning (with the exception of those that were present in the heptameric crystal structure). 5.3 Analysis of a-toxin structures 5.3.1 Local geometry. An important issue is the compatibility of this model structure with the images of a-toxin hexamers obtained by EM38 and AFM.39 The direct comparison of the images with the structure is the best way of resolving this.However in the absence of this information some idea can be gained by examination of dimensions quoted in the works (although these must be treated with caution as they are dependent on interpretation). Table 2 shows that the hexamer model is compatible with the outer diameter for hexamers measured by EM and AFM. Interestingly the quoted outer diameter for heptameric pre-pores obtained from AFM40 is larger and in accord with the crystal structure31 (Table 2). An analysis of the protein geometry revealed that there were no overall changes in the structure of the protein. However several residues have poor stereochemistry. Ser 262 has a forbidden conformation in all of the subunits in the hexameric model but also appears strained in the original heptameric crystal structure.Ser 262 lies at the apex of a b-turn at the end of the barrel domain. The hydrogen-bonding network between residues in the turn especially Thr 261 and Asn 264 forces Ser 262 into its poor conformation. It is interesting to note that these residues are located at the membrane/protein interface. It is likely that the polar surface of the membrane stabilizes the turn by interactions with the polar residues in the loop. 5.3.2 Analysis using HOLE. The dimensions of pore in the hexamer are similar to the heptamer (Fig. 8 and 9). The overall pro–le of the channel is similar to that of the heptamer (Fig. 9). The length of the channel is the same for the hexamer and the heptamer while the diameter of the channel is narrower for the hexamer. The minimum radius of the hexamer pore is 2.5 ” compared to 5.4 ” for the heptamer.This minimum occurs at a constriction caused by an annulus of charged residues that lie towards the distal end of the b barrel region of the pore. 195 Faraday Discuss. 1998 111 185»199 Table 2 Comparing the outer diameter for the a-toxin head group measured by imaging methods and from structural models Outer diameter/” Stoichiometry Method Ref. 70»80 89 76 38 39 40 31 6mer 6mer 7mer (pre-pore) 7mer 6mer EM AFM AFM X-ray crystallography Modelling 90a 75b This work a a a Distance between atoms identi–ed in the experimental structure as C A 71 and C D 270 (from the point outer point of one subunit to the —at face of the opposite).b Distance between the C C 276 and C F 276 between the opposing —at faces of the hexamer. 2a a Fig. 8 Orthogonal views of the HOLE surface of the heptamer (panels A and B) and the hexamer (panels C and D). Produced using the molscript program.52 Faraday Discuss. 1998 111 185»199 196 Fig. 9 Comparing the pore radius for the X-ray crystal structure31 for a-toxin heptamer (solid line) and model hexamer (dotted line). The HOLE procedures described in Sections 3 and 4 were used to predict both the absolute conductance and expected eÜect of the addition of PEG for both the experimental heptamer and model hexamer structures. Table 3 compares the absolute conductances and Fig. 10 shows the eÜect of PEG. Fig. 10 EÜect of PEG on the conductance of a-toxin.Points marked by circles and triangles show the eÜect found experimentally by Korchev et al.19 for the high and low conductance states respectively. Points marked by dotted squares show the experimental observations for the high conductance state by Krasilnikov et al.18 The thick solid line shows the eÜect predicted for the heptameric crystal structure31 and the dashed line for the model hexamer. Table 3 HOLE prediction of the conductance of a-toxin Conductance/pS (0.1 M KCl) 90 \10 52 44d 18 18d Experimental»high conductance statea Experimental»low conductance stateb Prediction heptamer crystal structurec Prediction for model hexamer structure a Ref. 51. b Ref. 19. c Ref. 31. d 1st and 2nd generation predictions.5 197 Faraday Discuss.1998 111 185»199 The data suggest that it is possible that both hexameric and heptameric forms may be seen in conductance measurements. The low conductance state would correspond to the hexamer and the high conductance state to a heptamer. This provides an explanation why the two forms have been observed in imaging techniques. However it leaves a rather large problem namely how can the channel rapidly switch between the two forms. This switching would entail the making and breaking of a large number of hydrogen bonds. However the insertion and pore-forming processes will also involve such changes. The hypothesis is worthy of further consideration. Work in progress seeks to establish the compatibility of the hexamer with AFM and EM data. We also plan to undertake electrostatics calculations to explain the pH sensitivity of the channel and the diÜerence in ion selectivity observed between the high and low conductance states.50 6 Conclusion This work shows the importance of systematically linking the structure of ion channels to their structure.The approach has great promise in aiding both modelling studies and the interpretation of experimental data. An important aim is to provide useful tools for the general scienti–c community working on ion channels. For this reason the program suite HOLE2 is freely available to all non-pro–t organizations ; for further details see http ://www.biochemistry.bham.ac.uk/hole or e-mail o.s.smart=bham.ac.uk. This work was supported by the Wellcome Trust through the provision of a Career Development Fellowship (042889 to O.S.S.) the UK Medical Research Council (grant G4600017) and the Cell Surface Research Fund.We are most grateful to Drs Peter Tieleman and Herman Berendsen of the University of Groningen for the provision of data,10 used in Section 3.1. We thank Joe Neduvelil and Xiaonan Wang for their contributions to the HOLE package. HOLE was originally developed at Birkbeck College with the support of Dr Bonnie Wallace and Prof. Julia Goodfellow. References 1 B. Hille Ionic Channels of Excitable Membranes 2nd edn. Sinauer Associates Inc. Sunderland MA 1992. 2 P. D. Adams I. T. Arkin D. M. Engelman and A. T. Brunger Nature (L ondon) Struct. Biol. 1995 2 154. 3 M. S. P. Sansom C. Adcock and G. R. Smith J. Struct. Biol.1998 121 246. 4 O. S. Smart J. M. Goodfellow and B. A. Wallace Biophys. J. 1993 65 2455. 5 O. S. Smart J. Breed G. R. Smith and M. S. P. Sansom Biophys. J. 1997 72 1109. 6 M. L. Connolly Science 1983 221 709. 7 B. Roux in T heory of transport in ion channels from molecular dynamics simulations to experiments ed. J. M. Goodfellow Weinham 1995. 8 J. Breed R. Sankararamakrishnan I. D. Kerr and M. S. P. Sansom Biophys. J. 1996 70 1643. 9 J. E. Hall J. Gen. Physiol. 1975 66 531. 10 D. P. Tieleman and H. J. C. Berendsen Biophys. J. 1998 74 2786. 11 S. W. Cowan T. Schirmer G. Rummel M. Steiert R. Ghosh R. A. Pauptit J. N. Jansonius and J. P. Rosenbusch Nature (L ondon) 1992 358 727. 12 R. Benz A. Schmid and R. E. W. Hancock J. Bacteriol. 1985 162 722.13 B. K. Jap and P. J. Walian Q. Rev. Biophys. 1990 23 367. 14 T. B. Woolf and B. Roux Proteins 1996 24 92. 15 L. Y. Shen D. Bassolino and T. Stouch Biophys. J. 1997 73 3. 16 M. S. P. Sansom personal communication. 17 M. Watanabe J. Rosenbusch T. Schirmer and M. Karplus Biophys. J. 1997 72 2094. 18 O. V. Krasilnikov R. Z. Sabirov V. I. Ternovsky P. G. Merzliak and J. N. Muratkhodjaev Fems Micro- 19 Y. E. Korchev C. L. Bashford G. M. Alder J. J. Kasianowicz and C. A. Pasternak J. Membr. Biol. 1995 biology Immunology 1992 105 93. 147 233. 20 O. V. Krasilnikov J. B. DaCruz L. N. Yuldasheva W. A. Varanda and R. A. Nogueira J. Membr. Biol. 1998 161 83. 21 S. M. Bezrukov and I. Vodyanoy Biophys. J. 1993 64 16. 22 E. A. Merritt S. Sarfaty F. Vandenakker C.Lhoir J. A. Martial and W. G. J. Hol Protein Sci. 1994 3 166. 23 O. V. Krasilnikov J. N. Muratkhodjaev S. E. Voronov and Y. V. Yezepchuk Biochim. Biophys. Acta 1991 1067 166. Faraday Discuss. 1998 111 185»199 198 24 J. Breed P. C. Biggin I. D. Kerr O. S. Smart and M. S. P. Sansom Biochim. Biophys. Acta Biomembranes 1997 1325 235. 25 B. A. Wallace J. Struct. Biol. 1998 121 123. 26 R. R. Ketchem W. Hu and T. A. Cross Science 1993 261 1457. 27 B. Turano M. Pear and D. Busath Biophys. J. 1992 63 152. 28 V. B. Myers and D. A. Haydon Biochim. Biophys. Acta 1972 274 313. 29 J. A. Killian Biochim. Biophys. Acta 1992 1113 391. 30 S. Bhakdi and J. Tranumjensen Microbiol. Rev. 1991 55 733. 31 L. Z. Song M. R. Hobaugh C. Shustak S. Cheley H. Bayley and J.E. Gouaux Science 1996 274 1859. 32 N. Tobkes B. A. Wallace and H. Bayley Biochemistry 1985 24 1915. 33 J. A. Gouaux G. Braha M. R. Hobaugh L. Z. Song S. Cheley C. Shustak and H. Bayley Proc. Natl. Acad. Sci. USA 1994 91 12828. 34 A. Valeva M. Palmer and S. Bhakdi Biochemistry 1997 36 13298. 35 A. Valeva J. Pongs S. Bhakdi and M. Palmer Biochim. Biophys. Acta Biomembranes 1997 1325 281. 36 C. Petosa R. J. Collier K. R. Klimpel S. H. Leppla and R. C. Liddington Nature (L ondon) 1997 385 833. 37 E. Gouaux Curr. Opinion Struct. Biol. 1997 7 566. 38 R. J. Ward and K. Leonard J. Struct. Biol. 1992 109 129. 39 D. M. Czajkowsky S. Sheng and Z. Shao J. Mol. Biol. 1998 276 325. 40 Y. Fang S. Cheley H. Bayley and J. Yang Biochemistry 1997 36 9518. 41 A. G. Murzin A. M. Lesk and C. Chothia J. Mol. Biol. 1994 236 1382. 42 A. G. Murzin A. M Lesk and C. Chothia J. Mol. Biol. 1994 236 1369. 43 M. S. Weiss and G. E. Schulz J. Mol. Biol. 1992 227 493. 44 T. Schirmer T. A. Keller Y. F. Wang and J. P. Rosenbusch Science 1995 267 512. 45 M. S. P. Sansom and I. D. Kerr Biophys. J. 1995 69 1334. 46 A. T. Brunger A. Krukowski and J. W. Erickson Acta Crystallogr. Sect. A 1990 46 585. 47 R. Fletcher Practical Methods of Optimization V olume 1 Unconstrained Optimization John Wiley & Sons Chichester 1980. 48 B. R. Brooks R. E. Bruccoleri B. D. Olafson D. J. States S. Swaminathan and M. Karplus J. Comput. Chem. 1983 4 187. 49 R. A. Laskowski M. W. Macarthur D. S. Moss and J. M. Thornton J. Appl. Crystallogr. 1993 26 283. 50 Y. E. Korchev. C. L. Bashford G. M. Alder P. Y. Apel D. T. Edmonds A. A. Lev K. Nandi A. V. Zima and C. A. Pasternak Faseb. J. 1997 11 600. 51 G. Menestrina J. Membr. Biol. 1986 90 177. 52 P. J. Kraulis J. Appl. Crystallogr. 1991 24 946. Paper 8/06771F 199 Faraday Discuss. 1998 111 185»199

ISSN:1359-6640    

DOI:10.1039/a806771f

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Molecular dynamics simulation of a hydrated diphytanol phosphatidylcholine lipid bilayer containing an alpha-helical bundle of four transmembrane domains of the In½uenza A virus M2 protein Received 25th August 1998 An a-helical bundle composed of four transmembrane portions of the M2 protein from the In—uenza A virus has been studied in a hydrated diphytanol phosphatidylcholine bilayer using molecular dynamics (MD) calculations. Experimentally the sequence utilized is known to aggregate as a four-helix bundle and act as a pH-gated proton-selective ion channel which is blocked by the drug amantadine hydrochloride. In the presented simulation the ion channel was initially set up as a parallel four-helix bundle. The all-atom simulation consisted of almost 16 000 atoms described classically using a force–eld from the CHARMM22 database.Bilayers with and without the bundle were shown to be stable throughout the nanosecond timescale of the MD simulation. Structural and dynamical properties of the bilayer both with and without the transmembrane protein are reported. I. Introduction The ion channel is understood to play a key role in a wide range of biological phenomena such as nerve conduction heartbeat toxic response and disease and it can be a site for attack by drug molecules. Due to the difficulty in crystallizing membrane proteins relatively few ion channel structures are known. Thus in spite of their importance many key aspects of ion channels are ill understood at the present time. However just recently the structure of the potassium channel KscA has been reported.1,2 The main structural motif in this channel consists of an alpha helix bundle.The alpha helix bundle is an important structural element in other channel forming proteins such as acetylcholine receptors.3,4 Thus the system of an alpha-helical bundle in a lipid bilayer is of general importance in the context of ion-channel function. The present article is concerned with an alpha helix bundle whose structure is not yet characterized at the atomic level. Speci–cally we focus on the membrane protein termed M2 which is associated with the In—uenza A virus.5 The M2 protein is known to form an ion channel and play an essential role in infection. It is gated by hydrogen ion concentration (pH). The channel is the site of attack by anti-—u drugs.5 As already mentioned the structure of the channel formed by the 96-residue M2 protein is not yet well characterized.Accordingly we focus on a simpler system Thomas Husslein,a,b Preston B. Moore,a Qingfong Zhong,a Dennis M. Newns,b Pratap C. Pattnaikb and Michael L. Kleina a Centre for Molecular Modelling and Department of Chemistry University of Pennsylvania Philadelphia Pennsylvania 19104-6323 USA b T. J. Watson Research Center International Business Corporation P.O. Box 218 Yorktown Heights NY 10598 USA Faraday Discuss. 1998 111 201»208 201 Fig. 1 Snapshots of the peptide bundle in a lipid bilayer taken from MD simulation at 300 ps. The four peptide backbones atoms have been drawn with Van der Waals radii. The lipids have been rendered in a ball and stick representation.The lipid head groups nitrogen and phosphorus atoms are drawn as balls. The hydrogens have been omitted for clarity. namely the fully functional channel formed by a bundle of four 25-residue peptides with the M2 transmembrane sequences.6h8 This model ion channel is also blocked by the drug amantadine hydrochloride.6,7 The restricted objective in this paper is to understand how the insertion of this transmembrane peptide bundle aÜects the structure and the dynamics of a typical lipid bilayer. The lipid we have chosen for study is diphytanol phosphatidylcholine (DPhPC) which has been successfully used in many investigations of protein»bilayer interactions especially in experiments on channel-forming proteins.Its high resistance to proton and other ion —ow account for the popularity of DPhPC in ion-conduction measurements.9 Additionally there is a general belief that this particular lipid has a high bilayer stability and acyl chain packing comparable to that of naturally occurring lipid bilayers. Furthermore it provides a well de–ned single-component lipid for experiments whereas the above-mentioned characteristics are traditionally obtained by a using a mixture of diÜerent lipids. In the next section we outline the simulation techniques and the way we set up the system. Then we describe the results of the MD simulations and address the impact of the insertion of the peptide bundle to the lipid bilayer. Finally we present our conclusions. II. Methods and materials Diphytanol phosphatidylcholine is a rather typical phospholipid.It consists of a zwitterion phosphatidylcholine head group linked to a glycerol ester and two hydrocarbon chains. In the case of DPhPC the hydrocarbon chains are branched. The main chain contains 16 carbons but at the positions 15 11 7 and 3 there are methyl groups attached so each chain consists of 20 carbon atoms. Technically the atoms 11 7 and 3 are chiral centres. These have been treated as having an all R con–guration in agreement with ref. 10. For the peptide bundle we used the truncated sequence of the M2 protein (M2-DA) described by DuÜ and Ashley,7 which was shown to function as a pH gated ion channel that is also blocked by the drug amantadine hydrochloride. The Faraday Discuss.1998 111 201»208 202 Ac- sequence is from Ser22-Leu46 with standard capping residues Ac and NH2 SSDPLVVAASIIGILHLILWILDRL-NH2 . Both the DPhPC bilayer and the M2-DA protein were simulated using an all atom description speci–cally the recently developed PARAM22b4b database of the CHARMM22 program package.11 For the histidine (His37) in the M2-DA protein only the NE nitrogen in its aromatic ring was protonated which is consistent with a pH of 7. The water was simulated with the —exible TIP3P model.12 The simulation system consisted of 4 M2-DA proteins 50 DPhPC lipids and 1996 water molecules. The M2-DA proteins were set up as straight a-helices not tilted against each other to form a four-fold symmetric bundle with the residues Ser31 His37 and Trp41 pointing towards the centre of the pore.This bundle was then embedded in the DPhPC lipid bilayer. To do so we used a well-equilibrated structure from a simulation on the pure lipid in which we created a hole by deleting 14 lipid molecules. The water of hydration in the pure lipid simulation was retained and an additional 170 waters were added to –ll the void in the head group region a further 36 waters were used to –ll the interior of the bundle. Fig. 1 shows a typical con–guration taken from the MD simulation. The phase behaviour and dimensions of lipid bilayers are well known to be sensitive to temperature pressure hydration and ionic concentration.9,13,14 We therefore employed a constant NPT simulation of a fully —exible box using the MTK extended system approach as described in ref.15 and 16. We used a chain length of 3 for the thermostats. The cut-oÜ for the interactions was set to 10 ”. Additionally we employed a RESPA scheme to perform the calculation of the forces on their appropriate timescales.17 The time-step we used for the update of the long-range Coulombic and non-bonded forces was 3 fs the update of the short-range non-bonded forces was 1.5 fs whereas the update of the bond stretch bend and dihedral interactions occurred every 0.375 fs. We used a Ewald summation technique for the long range forces. The reciprocal space summation was truncated at kx\ky\kz\10. III. Structural eÜects In a recent publication we compared an MD simulation on a fully hydrated DPhPC bilayer with available experimental data18 and with MD results for DPPC.19 Here we investigate the in—uence of the inserted peptide bundle on the DPhPC bilayer structure.Speci–cally we focus on the structure of the lipid both with and without the M2-DA peptide bundle. Fig. 2 shows the average separation along the bilayer normal between the peaks associated with the distributions of the phosphate atoms on each side of the bilayer. This measure of the bilayer thickness reveals no signi–cant drift suggesting that the overall system dimensions have converged. The lipid phosphate group separation across the bilayer for the case of the embedded peptide bundle is about 33 ” ” which is 5 less than in the pure lipid case. Assuming a constant packing density for the lipid molecules the surface area per lipid should be increased correspondingly.Accordingly we estimate that the surface area per lipid should likely increase from 74.5 to ca. 85 ”2. This increase in area per head group manifests itself in the changes seen in Fig. 3 where the maximum for the P-P in-plane radial distribution function is at ” ca. 6.1 in the case of the pure DPhPC compared to a broad peak at ” ca. 6.7 for the bilayer with the peptide bundle. In a more detailed analysis of the structure we carried out a Voronoi-type tessellation in which individual lipid phosphate groups were assigned their own area.20 To do so we –rst construct a Wigner-type cell around each lipid to encompass the area that is closest to it. This construction is space –lling and takes into account the periodic boundaries of the simulation box.The analysis was performed on systems both with and without the M2-DA peptide bundle in the lipid bilayer. The surface roughness of the bilayer was ignored. The surface area covered by the peptide bundle is not estimated easily. In particular the fact that the pro–le of the bundle can be convex demands a rather re–ned method. In the present case for the calculation of the area of the bundle we have only taken into account those atoms that are submerged in the lea—et of the bilayer that is under consideration. Their positions are projected onto the plane of the bilayer with each atom assigned its van der Waals radius. The area occupied by the bundle is then measured by dividing the bilayer plane into a dense grid.For each element 203 Faraday Discuss. 1998 111 201»208 Fig. 2 Time evolution of the average peak-to-peak separation of phosphate group distributions across the bilayer for DPhPC without (thin line) and with the M2-DA peptide bundle (bold line). of the grid we then decide whether it is within the outline of the protein or outside. When it is within we –nd the corresponding Wigner-type cell which was previously constructed and subtract the area from there. We utilized a cylindrical coordinate system for this analysis slicing the area like a pie. We move the origin of the coordinate system to the centre of mass. Radiating out from the origin we draw a line at an angle h from the x-axis. Starting from well outside the boundaries of the peptide bundle we follow the line until we –nd a peptide atom de–ned by its van der Waals radius.After incrementing the h by dh we draw another line and repeat the above procedure. The two radii and the origin de–ne a triangle that gives an estimate of the area covered by the peptide in this Fig. 3 The in-plane phosphorus-phosphorus radial distribution functions g(r) for DPhPC without (thin line) and with (dashed line) the M2-DA peptide bundle. Faraday Discuss. 1998 111 201»208 204 segment. We now sub-grid this triangle and –nd the Wigner-type cell of the lipid it falls into. We checked this procedure for several sizes of the segments dh and diÜerent sub-grids for the triangle until the area became independent of the choice of these parameters. For our simulation of the M2-DA peptide bundle in the lipid bilayer we –nd an area per lipid molecule of 85 ”2 and without a peptide bundle we reproduce the area per lipid as 74.6 ”2.Fig. 4 shows snapshots of the lipid bilayer at diÜerent times. It illustrates the behaviour of the Wignertype cells for a lipid with and without a peptide bundle. In the case without the peptide the smaller area per lipid is clearly evident. The comparison of the NMR order parameter SCD for DPhPC with and without the peptide bundle is shown in Fig. 5. In contrast to the terrace-like structure that is characteristic for the pure DPhPC case the order parameter for the system with the peptide bundle falls oÜ smoothly with increasing position along the chain indicating a more —uid-like behaviour. Furthermore the overall magnitude is also reduced hence the chain order is decreased.Additionally we have investigated the orientation of the head-groups and the lipid hydrocarbon chains with respect to the bilayer normal. We measured the angle U the angle between the vector connecting the –rst (C-1) and the last chain carbon atom (C-15) and the bilayer normal. Fig. 6 compares the probability distribution for U with and without the bundle in the bilayer. Upon insertion of the peptide bundle the peak value for U increases from ca. 20° to larger values. This behaviour is consistent with the observation that the bilayer thickness is reduced (recall Fig. 2) when the peptide bundle is inserted and hence the chains are not as extended as before. We observe that the probability for a chain to be almost parallel to the bilayer surface is also reduced.We have also investigated the probability distribution of the angle W which is the angle between a Fig. 4 Snapshots of one leaf of the lipid bilayer where we indicate the area per lipid by a Wigner-type cell construction for diÜerent simulation times. (a)»(c) show the bilayer with the M2-DA peptide bundle within whereas (d)»(f ) show the pure lipid. 205 Faraday Discuss. 1998 111 201»208 Fig. 5 Hydrocarbon chain deuterium order parameters SCD as a function of the carbon position along the main chain for DPhPC both with (K) and without (Ö) the peptide bundle. vector connecting P and N of the head-group and the plane of the lipid bilayer. Here we –nd no major diÜerences between the two cases.Fig. 7(a) gives the average projected structure of the a-carbon backbone of the M2-DA peptide in the lipid. On the abscissa we show the average position of the a-carbon atoms with respect to the bilayer normal. On the ordinate we plot the mean distance r(”) of these a-carbon atoms from the channel axis. The data points are an average over the four individual peptides and the whole trajectory. From this –gure one can easily pick the pore-forming residues Ala-30,Ser-31,Gly-34 Fig. 6 Distribution of the angle U between the vector connecting the –rst (C-1) and the last (C-15) main chain carbon atom and the bilayer normal for pure DPhPC (»»») and DPhPC with the M2-DA peptide (» » » »). The inset shows the angle W of the vector connecting the head-group P and N atoms and the plane of the lipid bilayer for both cases.(The scale of the y-axis of both graphs is identical.) Faraday Discuss. 1998 111 201»208 206 Fig. 7 Positions of important parts of the system with respect to the bilayer normal (a) shows the average separation of the individual a-carbon atoms from the channel axis only the pore facing residues are indicated ; (b) shows the contributions to the electron density distributions associated with phosphorus (»»») and nitrogen (» » » ») of the lipid bilayer head-groups. His-37 and Trp-41. Both ends of the peptide are not a-helical but are more extended. The pro–le of the averaged peptide backbone reveals a twist. The position where the backbone is closest to the pore axis is at the Gly-34 position. The pro–le shape should not misinterpreted.Individual peptides from the bundle do not have a kink. The curved pro–le arises solely from the fact that the individual peptides are tilted and coiled with respect to each other giving rise to this hourglass shape.21 In Fig. 7(b) we show the average positions of the lipid head groups with respect to the channel axis. Here the channel does not seem to reside symmetrically in the bilayer. However this only re—ects the fact that at the top of the bilayer the polar Ser residues are rather short whereas the Arg residues at the bottom have rather long side chains with charges at the very end. IV. Conclusion We have reported on the structural consequences of inserting a four-helix peptide bundle into a DPhPC bilayer and compared our –ndings with a previous simulation of a pure DPhPC bilayer.Both simulations were performed in the NPT ensemble and covered the nanosecond timescale. We did not apply any arti–cial structural constraints to these systems. The focus of our investigation was on the impact of the insertion of the M2-DA peptide bundle on a DPhPC bilayer. The most striking eÜect of the insertion is a decrease for the bilayer thickness and a concomitant increase in the surface area per lipid. Furthermore we observe a more —uid-like behaviour of the hydrocarbon chains indicated by a lower deuterium order parameter SCD. Such observations should be amenable to experimental veri–cation.8 The overall stability of our all-atom simulation reinforces the notion that the study of small membrane proteins within lipid bilayers is a promising –eld for further studies.14 Unfortunately much longer MD simulations will likely be required to probe the structure and function of the ion channel itself.22 207 Faraday Discuss.1998 111 201»208 Paper 8/06675B V. Acknowledgements We thank Tarek Mounir Alex MacKerell Bill DeGrado Tim Cross Larry Pinto and Jim Lear for many stimulating discussions. This research was supported by the National Institute of Health via grant GM 40712 and a collaborative research project with IBM. References 1 D. A. Doyle J. M. Cabral R. A. Pfuetzner A. L. Gulbis S. L. Cohen B. R. Chait and R. MacKinnon Science 1998 280 69. 2 R. MacKinnon S. L. Cohen A. L. Kuo A. Lee and B. T. Chait Science 1998 280 106. 3 J.Changeux Sci. Am. 1993 58»62. 4 N. Unwin J. Struct. Biol. 1998 121 181. 5 L. H. Pinto L. J. Holsinger and R. A. Lamb Cell 1992 69 517. 6 L. H. Pinto G. R. Dieckmann C. S. Gandhi C. G. Papworth J. Braman M. A. Shaughnessy J. D. Lear R. A. Lamb and W. F. DeGrado Proc. Natl. Acad. Sci. USA 1997 94 11301. 7 K. C. DuÜ and R. H. Ashley J. V irol. 1992 190 485. 8 F. A. Kovacs and T. A. Cross Biophys. J. 1997 73 2511. 9 C-H. Hsieh S-C. Sue P-C. Lyu and W-G. Wu Biophys. J. 1997 73 870. 10 L. R. Sita J. Org. Chem. 1993 58 5285. 11 A. D. MacKerall Jr. D. Bashford M. Bellott R. L. Dunbrack Jr. J. D. Evanseck M. J. Field S. Fischer J. Gao H. Guo S. Ha D. Joseph-McCarthy L. Kuchnir K. Kuczera F. T. K. Lau C. Mattos S. Michnick T. Ngo D. T. Nguyen B. Prodhom W.E. Reiher III B. Roux M. Schlenkrich J. C. Smith R. Stote J. Straub M. Watanabe J. Wiorkiewicz-Kuczera D. Yin and M. Karplus J. Phys. Chem. B. 1998 102 3586. 12 W. L. Jorgensen J. Chandesekhar J. D. Madura R. W. Impey and M. L. Klein J. Chem. Phys. 1983 79 926. 13 R. W. Pastor Curr. Opin. Struct. Biol. 1994 4 486. 14 D. P. Tieleman S. J. Marrink and H. J. C. Berendsen Biochim. Biophys. Acta 1997 1331 235. 15 G. J. Martyna M. L. Klein and M. Tuckerman J. Chem. Phys. 1992 97 2635. 16 G. J. Martyna M. E. Tuckerman D. J. Tobias and M. L. Klein Mol. Phys. 1996 87 1117. 17 M. Tuckerman B. J. Berne and G. J. Martyna J. Chem. Phys. 1992 97 1990. 18 T. Husslein D. M. Newns P. C. Pattnaik Q. F. Zhong P. B. Moore and M. L. Klein J. Chem. Phys. 1998 109 2826. 19 K. Tu D. J. Tobias J. K. Blasie and M. L. Klein Biophys. J. 1996 70 595. 20 W. Shinoda and S. Okazaki J. Chem. Phys. 1998 109 1517. 21 Q. Zhong Q. Jiang P. B. Moore D. M. Newns and M. L. Klein Biophys. J. 1998 74 3. 22 P. B. Moore Q. Zhong T. Husslein and M. L. Klein FEBS L ett. 1998 431 143. Faraday Discuss. 1998 111 201»208 208

ISSN:1359-6640    

DOI:10.1039/a806675b

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Alamethicin channels in a membrane molecular dynamics simulations D. Peter Tieleman,a Jason Breed,b,§ Herman J. C. Berendsena and Mark S. P. Sansom*b a BIOSON Research Institute and Department of Biophysical Chemistry University of Groningen Nijenborgh 4 9747 AG Groningen T he Netherlands b L aboratory of Molecular Biophysics T he Rex Richards Building Department of Biochemistry University of Oxford South Parks Road Oxford UK OX1 3QU. E-mail mark=biop.ox.ac.uk Receiøed 7th August 1998 A Alamethicin (Alm) is a 20 residue peptide which forms a kinked a-helix in membrane and membrane-mimetic environments. Ion channels formed by intramembraneous aggregates of Alm are thought to be formed by bundles of approximately parallel Alm helices surrounding a central bilayer pore.DiÜerent channel conductance levels correspond to diÜerent numbers of helices per bundle ranging from N\5 to N[8. Calculation of the predicted pK values of the ring of Glu18 sidechains at the C-terminal mouth of the pore suggests that at neutral pH most or all of these sidechains will remain protonated. Nanosecond molecular dynamics (MD) simulations of N\5 6 7 and 8 bundles of Alm helices in a POPC bilayer have been run corresponding to a total simulation time of 4 ns. These simulations explore the stability and conformational dynamics of these helix bundle channels when embedded in a full phospholipid bilayer in an aqueous environment. The structural and dynamic properties of water in these model channels are examined. As in earlier in vacuo simulations (J.Breed R. Sankararamakrishnan I. D. Kerr and M. S. P. Sansom Biophys. J. 1996 70 1643) the dipole moments of water molecules within the pores are aligned antiparallel to the helix dipoles. This helps to contribute to the stability of the helix bundles. Introduction Ion channels are formed in lipid bilayers by integral membrane proteins. Channels enable selected ions to move rapidly (ca. 107 ions s~1 channel~1) and passively (i.e. down their electrochemical gradients) across membranes. Ion channels are important in numerous cellular processes principally electrical signalling,1 but also include diverse processes such as facilitating the uncoating of viral genomes.2 In order to understand the physical events underlying the biological properties of channels one must characterise both their structures and their dynamic behaviour.This is far from easy. Because ion channels are membrane proteins we remain relatively ignorant of their three dimensional structures. Indeed a high resolution structure is known for only one ion channel a bacterial K` channel.3 This re—ects a more general problem for membrane proteins. Although integral membrane proteins comprise ca. 20 to 30% of most genomes,4,5 high resolution structures § Present address Fakultaé t fué r Biologie Universitaé t Konstanz Postfach 5560 M656 78434 Konstanz Germany. 209 Faraday Discuss. 1998 111 209»223 have been solved for only a small handful of these. In this context model membrane proteins have much to oÜer to studies of membrane protein structure and dynamics.For example studies of a channel-forming peptide gramicidin A have provided profound insights into the structural basis of channel function.6h9 However the structural idiosyncrasies of gramicidin (which is made up of alternating L- and D-amino acids) direct oneœs attention to other peptide models which more closely mimic ion channel proteins. Many channel proteins contain a central pore lined by a bundle of approximately parallel a-helices.10 Such channels range in complexity from the M2 protein of in—uenza A (ca. 100 amino acids per subunit)11 to the nicotinic acetylcholine receptor (ca. 500 amino acids per subunit).12 Given the importance of this structural motif in a number of channel proteins it is important to have a simple yet detailed model system for channels formed by a-helices.Such a system is provided by alamethicin. Alamethicin (Alm) is a largely hydrophobic 20 residue peptide which forms ion channels in lipid bilayers. The channel structural and spectroscopic properties of Alm have been studied in considerable detail.7,13,14 The structure of Alm in a non-aqueous environment has been determined by X-ray diÜraction15 and by NMR.16,17 The crystal and solution structures are strikingly similar. The largely a-helical conformation of Alm is stabilised by the presence of a large number of Aib (a-amino isobutyric acid i.e. a-methyl alanine) residues in its sequence Ac-Aib-Pro-Aib-Ala-Aib-Ala-Gln7-Aib-Val-Aib-Gly-Leu-Aib-Pro14-Val-Aib-Aib-Glu18-Gln-Phol The presence of Pro14 induces a central kink in the helix.NMR amide exchange data demonstrate that the largely a-helical conformation of Alm when dissolved in methanol18 is retained when it interacts with lipid bilayers.19 Amide exchange data also suggest that Alm in methanol undergoes hinge-bending motion about the central proline-induced kink.20 Alm forms multi-conductance channels (Fig. 1) in a voltage-dependent manner. The multiconductance behaviour of Alm channels is generally explained in terms of the barrel stave (i.e. helix bundle) model15,21h23 in which several (from ca. 5 to ca. 10) Alm helices form a bundle surrounding a central pore. DiÜerent conductance levels correspond to diÜerent numbers (N) of helices per bundle. This model is in accord with a large body of experimental data (reviewed in ref.13) including neutron scattering studies from Alm in lipid bilayers.24 A key feature of this model is that the helices are oriented parallel (rather than antiparallel) to one another their helix dipole repulsions being overcome by their favourable interaction with the electrostatic –eld across the bilayer and with the water inside the transbilayer pore.25,26 A parallel orientation of the helices is supported by the asymmetry of voltage-activation of Alm channels. Thus when Alm is added to Fig. 1 Recording of ionic currents —owing through a single Alm channel. This recording32 was obtained from a diphytanoyl phosphatidylcholine bilayer exposed to 0.25 lM Alm. The electrolyte was 0.5 M KCl and the transbilayer potential was 125 mV.The arrows and numbers on the right give the presumed number of helices per bundle (N) for each conductance level. Faraday Discuss. 1998 111 209»223 210 one face (the cis face) of a bilayer only cis-positive voltages will induce channel formation. This argues for a structure which is asymmetric with respect to the bilayer plane i.e. a parallel helix bundle. Such a structure is supported by the work of Woolley and colleagues27,28 who have shown that channels formed by covalently coupled pairs of Alm helices (which are physically constrained to be parallel to one another) strongly resemble those formed by unmodi–ed Alm. Molecular models of channels formed by Alm helix bundles have been generated by restrained molecular dynamics (MD) simulations in vacuo and these models have been re–ned by short MD simulations with water molecules included within and at each mouth of the pore.29 These models can explain the change in stability of Alm channels when the Gln7 sidechain is replaced by a smaller polar residue.30,31 The pore dimensions of such models correlate well with experimentally observed conductance values of Alm.13,32,33 Similar models of channels formed by Alm analogues when used in continuum electrostatics calculations predict the ionic-strength dependent non-linear current»voltage curves observed experimentally.28 Models of Alm helix bundles have been used in MD simulations to demonstrate that water within channels formed by parallel helix bundles is ordered and shows reduced translational and rotational mobility relative to bulk water.25 Finally models of Alm channels have been used to explore changes in translational mobility of Na` ions when within narrow pores.34 Thus a-helix bundle models of Alm channels have been studied in some detail and have been shown to correspond well with experimental data.So it is reasonable to use them as the basis of prolonged MD simulations of a-helix bundles in a lipid bilayer. It has recently become feasible to run multi-nanosecond duration MD simulations of Alm channel models with full bilayer models in which the lipid molecules and the water on either face of a membrane are represented explicitly. Such simulations exploit the considerable progress of the past few years in simulations of lipid bilayers per se.35h38 In particular simulations on pure bilayers have explored the dependence of simulation behaviour on the methods employed and have established rules-of-thumb for physically realistic simulations both in terms of suitable parameters for inter-atomic interactions and in terms of optimal simulation protocols.Alm channel simulations also build upon full bilayer simulations of single transmembrane a-helices 39h41 which have shown that such simulations enable one to characterise the structural dynamics of membrane-spanning a-helices in a realistic model of their environment. In previous simulations we have investigated isolated Alm helices in a bilayer and in solution42 and an N\6 model of an Alm helix bundle in a bilayer.26 Here we extend the latter simulations to Alm helix bundles containing from N\5 to 8 helices paying particular attention in the initial setup of the simulation to the ionisation state of the Glu18 sidechains.pKA INTRINSIC\pKA MODEL[ 2.303 [**GBORN]**GBACK] 1 Methods Generation of starting models Initial models of Alm helix bundles with from N\5 to 8 helices per bundle were generated using restrained MD in vacuo as described in ref. 29 and 43. During the –nal stage of the simulated annealing protocol used to build the model the restraints applied were (i) intra-helix restraints to maintain H-bonding of the backbone of each Alm monomer; (ii) inter-helix restraints between the N-terminal segments of adjacent monomers of the bundle in order to maintain the integrity of the bundle; and (iii) non-crystallographic symmetry restraints to maintain the approximate rotational symmetry of the helix bundles.For each value of N an ensemble of 25 structures was generated. The most symmetrical member of each ensemble was used in the bilayer MD simulations. pK calculations A The ionisation states of the Glu18 sidechains in the Alm bundle models were estimated via pKA calculations as described in detail in ref. 44. This proceeded in two stages. Intrinsic pK values A were calculated using the A pK **G where pK of an isolated amino acid is the is the solvation contribution to A MODEL A pK shift and **GBACK is the contribution due to the interaction of the residue with non- BORN 211 Faraday Discuss. 1998 111 209»223 titrating charges.45,46 Absolute pK values were obtained via calculation of titration curves.The A latter were obtained via calculation of p(x)Pexp[[ln 10 ; ci(pKA INTRINSIC i[pH)[b ; ; **Gi k] i k:i i where p(x) is the probability of a residue existing in its ionised state and x is an N-element state vector whose elements are 0 or 1 depending on whether the residue is un-ionised or ionised respectively c\[1 for a basic residue c\]1 for an acidic residue and **Gi k is the screened Coulombic interaction energy between pairs of ionisable residues i and k.45,47 System setup for bilayer MD The setup of the simulation systems was essentially as described in ref. 26. An equilibrated POPC bilayer with 128 lipid molecules was used. For each bundle model a cylindrical hole was made in the centre of the bilayer by removing selected lipid molecules and running a short MD simulation with a radially acting repulsive force to drive any remaining atoms out of the cylinder into the bilayer.The Alm bundle models (with a single ionised Glu18 see below) were inserted into the cavities thus created. These systems (helix bundle plus POPC) were solvated with SPC waters (ca. 30 waters per lipid molecule) and a single Na` ion replacing a water molecule at the position of lowest Coulomb potential. The resultant systems (see Table 1) were simulated for 25 ps (50 ps for the N\8 bundle) with positional restraints on the peptide atoms relative to the starting bundle structure with constant surface area and with a constant pressure of 1 bar in the z (i.e. bilayer normal) direction.The resultant systems were used as the starting point for the 1 ns production runs. MD simulation details Molecular dynamics simulations were run using GROMACS.48 A twin range cut-oÜ was used for longer-range interactions 1.0 nm for van der Waals interactions and 1.7 nm for electrostatic interactions. The timestep was 2 fs using LINCS49 to constrain bond lengths. NPT conditions (i.e. constant number of particles pressure and temperature) were used in the simulation. A constant pressure of 1 bar independently in all three directions was used with a coupling constant of qT\0.1 ps. qP\1.0 ps.50 This allowed the bilayer/peptide area to adjust to its optimum value for the force- –eld employed. Water lipid and protein were coupled separately to a temperature bath at 300 K using a coupling constant Lipid parameters were as in previous MD studies of lipid bilayers,51,52 and as in our previous papers on MD simulations of Alm.26,42 These lipid parameters give good reproduction of the experimental properties of a DPPC bilayer.The water model used was SPC,53 which has been shown to be a reasonable choice for lipid bilayer simulations.54 Longitudinal diÜusion coefficients of water molecules within the pore were determined from their mean square displacement along the pore (z) axis over a period of 5 ps as described by.55 The diÜusion coefficient was assigned to the local region on the pore axis corresponding to the position of the water molecule at the start of the 5 ps period. Computational details Simulations were carried out on a 195 MHz R1000 Origin 2000 and took ca.8 days per processor per 1 ns of simulation time (for about 17 000 atoms). Analysis was performed using facilities within Table 1 Simulation details Final Ca RMSD/ nm Duration of simulation/ps Total number of atoms Number of waters Number of phospholipids N 0.22 0.20 0.24 0.22 1015 1015 1025 1010 16 729 16 893 16 740 16 928 3511 3527 3524 3548 103 102 96 95 5678 Faraday Discuss. 1998 111 209»223 212 GROMACS and with code written speci–cally for this project. Secondary structure analysis employed the DSSP algorithm.56 Initial models were generated using X-PLOR.57 Structures were examined using Quanta (Biosym/MSI) and Rasmol and the diagrams were drawn using MolScript.58 Electrostatics calculations (for pK calculations) were performed using UHBD version A 5.159 (with some local modi–cations) and partial atomic charges from the Quanta/Charmm22 parameter set. Pore radius pro–les were determined using HOLE.60 Results Ionisation state of Glu18 sidechains Examination of the in vacuo generated models of Alm helix bundles reveals a ring of Glu18 sidechains at the C-terminal mouth of the channel [see Fig. 2A]. In this respect the Alm channel model is similar to that of the pore-lining M2 helix bundle of the a7 nicotinic receptor (nAChR) channel.44 Studies of the ionisation state of the ring of glutamate residues at the C-terminal mouth of the nAChR suggest that their location at the C-terminus of the a-helix dipole and in proximity to one another results in a shift of their pKAs such that at neutral pH they are not fully ionised.Previous MD simulations of N\6 Alm helix bundle models have suggested that the integrity of the helix bundle during the simulation is dependent upon the ionisation state of the Glu18 sidechains. Thus it was judged to be important to determine the average ionisation state of the Glu18 sidechains before running the bilayer MD simulations. Note that this follows the approach of Fig. 2 A model of the N\7 Alm bundle generated by restrained MD in vacuo. The ring of Glu18 sidechains is indicated. B calculated titration curves for the seven Glu18 residues of the model shown in A. 213 Faraday Discuss. 1998 111 209»223 Table 2 Calculated pKAs of Glu18 residues in Alm helix bundles Net charge at pH 7 N pKAs of Glu18 residues 7 6 5 8 6.5 7.6 11.3 [14.0 [14.0 6.7 [10.0 13.0 [14.0 [14.0 [14.0 9.4 10.7 11.3 12.2 13.0 13.8 14.0 5.6 7.4 7.5 8.3 8.5 9.2 10.3 [14.0 [0.80 [0.66 [0.69 [1.69 Tieleman and Berendsen55 who employed the ionisation states estimated by KarshikoÜ et al.46 in their bilayer simulations of the porin OmpF.The results of the estimation of pK values for the Glu18 residues are summarised in Table 2 (pKA\4.4). Model calculations in which the pKA value of a Glu sidechain as a A a-helix indicate that this shift is a result of the unfavourable A and an example of the calculated titration curves on which these are based is provided in Fig.2B. Note that the calculated pK values for the N Glu18 sidechains of a bundle are not identical. This A is because the bundle models do not have exact rotational symmetry. Thus in Fig. 2B one may see that three of the Glu18s titrate with pKAs of 9.4 to 11.3 whereas the remaining four titrate pK with somewhat higher However a consistent picture does emerge across the diÜerent N As. values. Overall the Glu18 residues exhibit much higher pK values than that of an isolated glutamate sidechain function of its position along an Ala21 interaction of the C-terminal end of the a-helix dipole with the ionised state of the sidechain (Adcock Smith and Sansom unpublished results) a result which is also supported by experimental data.61,62 One may estimate the consequence of this shift in pK values for Glu18 by calculating the net A charge expected on each helix bundle at pH 7.As can be seen from Table 2 this varies from [0.66 (for N\6) to [1.7 (for N\8). The average across all four models is [0.96. Thus choosing the nearest integer for the bilayer MD simulations each bundle was modelled as having (N[1) Glu18 residues in their protonated state and one Glu18 ionised. In each case that Glu18 sidechain with the lowest calculated pK was chosen as the ionised one. A Progress of the simulations The progress of the simulations was monitored via inspection of the Ca RMSD trajectories (not shown) of the Alm bundles in order to ascertain how far these drifted from their initial structures. For each N value the Ca RMSD rose during the –rst 250 to 500 ps reaching a plateau value of ca.0.2 nm. This is similar to the value obtained in previous simulations of an (electrically neutral) Alm N\6 helix bundle model26 and in MD simulations of an OmpF porin trimer (the latter simulations starting with an X-ray structure). Thus the drift from the initial structure in all four of the current simulations is similar to that observed in other simulations of a protein in a bilayer. Examination of the Ca RMSD values at the end of the 1 ns MD run (Table 1) shows no trend in the RMSD value with the number of helices per bundle. Thus the overall drift in structure from the starting model does not appear to be greater for the large helix bundles (at least not over a period of 1 ns). Integrity of helix bundles In Fig.3 the N\7 helix bundle is shown at the end of the 1 ns simulation. It can be seen that the helix bundle is intact that the central pore remains open and that the helix bundle remains with its pore axis perpendicular to the plane of the bilayer. Examination of the superimposed Ca traces at the start and end of the simulation (Fig. 4) suggests that for all four N values the integrity of the helix bundle is maintained during the simulation. There is some slight degree of expansion of the bundles. This is evident from the Ca traces and also from analysis of the radii of gyration (not shown) of the bundles as a function of time. However this appears to be simply a relaxation of the bundle geometry in the presence of water and lipid (remembering that these simulations started from in vacuo models) and does not appear to disrupt helix»helix packing.Such bundle expansion appears to be a little greater at the C-terminal mouth of the pore which would suggest a limited Faraday Discuss. 1998 111 209»223 214 Fig. 3 Images of the N\7 Alm simulation system taken at t\1 ns. Water molecules are shown in cyan phospholipid molecules in green and the model pore in standard CPK colours. The Alm molecules and the carbonyl oxygens of the fatty acids are shown in space-–lling format. A View down the pore (z) axis with the C-terminal mouth of the pore (z ca. ]4 nm) towards the reader. B View perpendicular to the pore axis with the C-terminal mouth of the pore uppermost. degree of conformational change of the constituent helices (see below).Inspection of the helix bundles plus the water molecules within the pore at the start (not shown) and end (Fig. 5) of the simulation suggests that such expansion corresponds to entry of additional water molecules into the pore. This is most evident for the N\5 helix bundle for which at t\0 ps there were ca. 55 water molecules within the pore whereas after 1 ns this number had risen to 94. Internal motions of the a-helices It is also of interest to examine the internal motions of the helices as these will in turn in—uence the dynamic properties of the transbilayer pore. Experimental20 and simulation42 studies of Alm helices suggest that the Gly-X-X-Pro sequence motif allows for —exibility in the central hinge region of the molecule.Furthermore comparison of the dynamic behaviour of isolated Alm molecules in solution in water and in methanol and in a transbilayer environment,42 suggests that the C-terminal half of the molecule becomes disordered when in an aqueous environment but not when in a non-aqueous solvent or within a bilayer. Analysis of the RMS —uctuations from the average structure during the MD trajectories (Fig. 6) indicates that the C-terminal half of the Alm molecule within a bundle undergoes greater —uctuations than the N-terminal half although these —uctuations are less marked than those for a single Alm molecule in water. This con–rms the impression arrived at from visual comparison of the initial and –nal Ca traces (see above). Thus it seems that the N-terminal halves of the Alm helices which are more tightly packed together in the bundles —uctuate less than the corresponding C-terminal halves.One may examine the consequences of these —uctuations in terms of the secondary structure of the helices analysed using DSSP56 (Fig. 7). This reveals that (i) the majority of residues remain in an a-helical conformation throughout the simulation ; (ii) loss of helicity is transient and tends to be in the C-terminal half of the Alm molecules; and (iii) patterns of loss of helicity vary between the diÜerent constituent Alm molecules of a bundle. As in earlier N\6 simulations26 the overall 215 Faraday Discuss. 1998 111 209»223 Fig. 4 Ca traces of the four Alm helix bundles taken at the start (t\0 ns ; grey lines) and end (t\1 ns; black lines) of each simulation.In each case the view is down the pore axis. degree of helicity is slightly less than that of an isolated Alm molecule in a hydrophobic environment but signi–cantly greater than that of an isolated Alm molecule in water.42 This presumably re—ects the intermediate environment in which an Alm molecule within a bundle –nds itself with its more apolar face towards the lipid but its more polar face towards the water molecules within the pore. Water within the pore As seen above (Fig. 5) there is a well de–ned column of water within the lumen of the each of the four Alm helix bundles. As the water within a pore plays an important role in the permeation of ions through a pore the nature of the pores and their water has been examined in some detail.Pore radius pro–les were determined every 50 ps. The resultant time-averaged radius pro–les are compared in Fig. 8. The N\5 bundle is more constricted at its C-terminal mouth at z ca. 4.5 nm i.e. in the vicinity of the Glu18 sidechain ring yielding a minimum pore radius of ca. 0.15 nm with a lesser degree of constriction in the vicinity of the ring of Gln7 sidechains at z ca. 2.8 nm. This pattern is switched in the N\6 7 and 8 bundles with the narrower constriction in the region of the Gln7 ring and a lesser degree of constriction at the C-terminal mouth. Thus the smallest helix bundle which is thought to correspond to the lowest conductance level of the channel (Fig. 1) diÜers somewhat in its pore geometry from the others. Faraday Discuss.1998 111 209»223 216 Fig. 5 Ca traces of the four Alm helix bundles taken at the end (t\1 ns ; black lines) of each simulation superimposed upon those water molecules within and close to either mouth of the pore (grey bonds). On the basis of these pore radius pro–les one may obtain approximate predictions of the pore conductances using the methods of Smart et al.33 For 1 M KCl this yields a predicted single channel conductance of 170 250 310 and 400 pS for N\5 6 7 and 8 helices per bundle respectively. The N\6 value is in agreement with the corresponding experimental value of 280 pS.33,63 The predicted conductance values for N\7 and 8 are too low. This may re—ect difficulties in applying a simple geometry based prediction method to these large helix bundles (see below).The longitudinal (i.e. along the pore z axis) diÜusion coefficient of water molecules has been examined for all four simulation systems plotting water diÜusion coefficients as functions of their z coordinates [Fig. 8B]. Note that those waters within the pore are located between z\2 and z\5 nm. The diÜusion coefficients of waters within the pore are markedly reduced relative to those of waters in the bulk region. Note that the bulk diÜusion coefficient of SPC water is ca. 5]10~9 m2 s~1. In the narrowest regions of all four pores the water diÜusion coefficients fall to less than 0.5]10~9 m2 s~1 i.e. less than a tenth of the bulk water value. Thus the substantial reduction in the translational motion of water within narrow pores seen in simpler (no bilayer) simulations25 is reproduced in the current more realistic study.Interestingly although there is some dependence of the local diÜusion coefficient on the pore radius this is not as strong as might 217 Faraday Discuss. 1998 111 209»223 Fig. 6 Residue-by-residue Ca RMS —uctuations about their average coordinates for the Alm N\7 model. The vertical broken lines delineate the extents of helices H1 (residues 0 to 20) to H7 (residues 126 to 146). Note that in our residue numbering scheme residues 0 21 42 64 84 105 and 126 correspond to the N-terminal acetyl groups of the Alm molecules. 10 Fig. 7 Secondary structure as de–ned by DSSP,56 as a function of time for the Alm N\7 model. The greyscale is black\a-helix ; dark grey\3 -helix ; pale grey\turn ; and white\coil.H1 to H7 denote the constituent helices of the bundles. Faraday Discuss. 1998 111 209»223 218 Fig. 8 A Pore radius pro–les evaluated using HOLE,60 for the four Alm simulations. The pro–les were averaged from structures saved every 50 ps. The N-termini of the helices are at z ca. ]2 nm and the C-termini at z ca. ]5 nm. B Water diÜusion coefficients (Dz) as a function of position along the pore (z) axes. C Projection of water dipole moments onto the pore axes. For each graph the following convention is used solid black line for N\5; broken black line for N\6; solid grey line for N\7; and broken grey line for N\8. have been expected (compare the coefficients for the four bundles close to the Glu18 rings at z ca. 4.5 nm). The dipoles of the water molecules within all four pores are oriented by the surrounding parallel helix dipoles [Fig.8C]. Thus within the pore the mean z component of the water dipoles is nearly 1.8 Debye which should be compared with a dipole moment of 2.3 Debye for a single SPC 219 Faraday Discuss. 1998 111 209»223 water. There is some dependence of the degree of water dipole orientation on the number of helices per bundle. The observed degree of orientation of the water dipoles may be employed to calculate the local kz\1.79 1.77 1.61 1.60 0\2.3 Debye k is the dipole moment of water and k is its projection along the z (pore) 0 –eld experienced by water molecules within the pore. Thus the values of Debye averaged along the pore for N\5 to 8 respectively when compared with k for SPC water [where z axis] can be used to estimate the –eld strengths within each pore due to the aligned helix dipoles.Using the Langevin equation k B BD z\k0Ccoth Ak k 0 Ez [Ak kB T 0 Ez B T E k is the z-component of the electrostatic –eld due to the helix dipoles and and T are the B Ez\2.4 2.3 1.8 and where z Boltzmann constant and temperature respectively ; this yields –elds of 1.8]109 V m~1 respectively for the N\5 6 7 and 8 pores. (See ref. 64 for a more detailed description of the theory of interaction of a-helix bundle dipoles with pore water molecules.) These ìobservedœ –elds may be compared with approximate predictions of the electrostatic –elds generated by the peptide backbones of the corresponding bundles of aligned a-helices.Estimates of the latter –elds were obtained by numerical diÜerentiation of the electrostatic potential energy along the pore axes of initial Alm helix bundles in vacuo. Note that this approximate prediction is a simple Coulombic –eld and does not take into account the environment surrounding the helix bundle. The –elds estimated in this fashion had maximum values in the centres of the pores of ca. 3.6 3.0 2.1 and 1.2]109 V m~1 respectively for N\5 6 7 and 8 helices per bundle. These values are given the approximations involved in good agreement with the ìobservedœ –elds calculated from the Langevin equation. Such agreement con–rms that there are strong interactions between water dipoles and aligned helix dipoles which will contribute to the stability of the helix bundles.Discussion Simulation methodology The simulations in this paper employ a similar methodology to that employed in a number of recent MD simulations of membrane proteins.26,42,55,65 However as all simulations remain an approximation to the true properties of a system it is useful to consider possible limitations of the approach. In particular 1 ns is a relatively short period of time both in the context of lipid motions37 and of the mean time it takes an ion to move through a channel (ca. 10 to 100 ns). Extension of the simulations is in progress. Furthermore our simulations have been conducted in the absence of a transbilayer voltage diÜerence. This is perhaps not too problematic as the work of He et al.24 suggests that Alm helix bundles may form in the absence of a voltage diÜerence.However a number of recent simulations66h68 and theoretical69 studies have shown that a transbilayer voltage diÜerence may be included in MD simulations. It will be interesting to perform Alm bundle simulations in the presence of an external electrostatic –eld. A further limitation to our simulations is the absence of electrolyte. Experimental studies of Alm channels are conducted using e.g. 1 M electrolyte corresponding to ca. 130 ions in the simulation box used in our simulations. This is an area which future studies will have to address. In particular extended simulation times will be required to allow efficient equilibration of such a system. Another limitation is the relatively simple treatment of long-range electrostatic interactions in our current study.The simulation protocol we have used gives reasonable agreement with experimental results for pure bilayer simulations37 and yields stable simulations for the porin OmpF in a POPE bilayer.55 However a number of studies have been concerned with the eÜects of diÜerent treatments of long-range electrostatic interactions.37,38 It is likely that in the presence of explicit charges longrange interactions should be included by proper lattice sums so as to avoid artefacts. Finally although we have tried to take into account the most likely ionisation state of the Glu18 sidechains one must recognise that this ionisation state may be in—uenced by the headgroups of the Faraday Discuss. 1998 111 209»223 220 lipid molecules and also that the pKAs of the Glu18 sidechains may change dynamically with respect to time.70 Such eÜects will also have to be taken into account in future simulation studies.Biological signi–cance Despite the limitations of the simulations the dynamic behaviour of the Alm N\5 to 8 bundles and their interactions with water and lipids are of biological signi–cance. In particular if Alm helix bundles are thought of as paradigms of other more complex ion channels formed by helix bundles the dynamic behaviour of Alm bundles may have relevance to channels in general. In this respect it is of interest to note that the Alm bundle simulations reveal that a stable helix bundle within a bilayer may exhibit dynamic —uctuations in its constituent helices.This is of interest in the context of K` channels. The recently determined X-ray structure of a bacterial K` channel4 reveals a pore based upon an a-helix bundle motif into which the pore-lining P-domain is inserted in order to confer greater ion selectivity. Site-directed spin labelling studies indicate that subtle changes in TM helix packing of the KcsA channel may be related to its gating. Our studies suggest that MD simulations in a lipid bilayer may be able to reveal details of such changes. This is also of relevance in the context of K` channels. Interestingly recent MD simulations of isolated pore-lining (S6 and M2) helices of K` channels spanning a POPC bilayer71 suggest that hingebending motions occur within these helices. Thus K` channels may exhibit —uctuations in their pore dimensions similar to those seen in Alm channels.A second signi–cant result which emerges is that for Alm bundles with diÜerent numbers of helices are approximately equally stable in MD simulations within a phospholipid bilayer. This correlates nicely with the experimental observation of multiple conductance levels for the Alm channel. Other channel-forming peptides do not show such a wide multiplicity of conductance levels.32,72 It would be of interest to see whether for such peptides helix bundle models with diÜerent N values showed similar stability in bilayer simulations. A general property which is revealed in the current study is the altered dynamics of water molecules within Alm pores. This is seen regardless of the number of helices per bundle.Altered water dynamics have also been seen in comparable simulations of the bacterial porin OmpF.55 Almost identical eÜects on dynamics of water with Alm and other model pores were observed in simulations of a pore plus water system in the absence of a bilayer model i.e. essentially in vacuo,25 and in simulations of peptide channels in which the bilayer was mimicked by an octane slab.73 This is important in that it suggests that simpli–ed simulations may capture the essence of the dynamics of water (and by extension of ions34,74) in models of transbilayer pores. The other unusual property of water within Alm channels is the high degree of alignment of water dipoles along the pore axis. This was also observed in the earlier simulations.25,75 The current study demonstrates that water dipoles are aligned by the electrostatic –eld created by the parallel dipoles of the constituent a-helices of the bundle.Such alignment of intra-pore water dipoles has important consequences. Alignment of water dipoles by parallel helix dipoles means that if one should attempt to model the water within a pore as a continuum a lower dielectric (relative permittivity) than that of bulk water should be used.64 However there are rather more fundamental problems in treating nearly dielectrically saturated water as a simple dielectric continuum. Furthermore as discussed above helix»water dipolar interactions contribute to the stabilisation of parallel a-helix bundles. Future directions These studies have shown that MD simulations of Alm helix bundles including a lipid bilayer plus water reveal important aspects of the dynamic behaviour of both the peptide and water molecules.There are a number of ways in which this work may be extended. Firstly free energy pro–les9,76h78 for the ion as it moves along the pore should be calculated in order to understand better the energetics of pore»ion interactions. Secondly longer MD simulations may reveal details of slower changes in conformation of Alm helix bundles and slower movements of helix bundles relative to their bilayer environment. Thirdly the eÜects on bundle stability of changing the environment e.g. by including a transbilayer voltage diÜerence67,68 or by using other phospholipids should be explored.In this way it will be possible to build up a more complete picture of the behaviour of a simple ion channel at atomic resolution. 221 Faraday Discuss. 1998 111 209»223 Acknowledgements Work in MSPSœs laboratory is supported by The Wellcome Trust. DPT was supported by the European Union under contract CT94-0124. References 1 B. Hille Ionic Channels of Excitable Membranes Sinauer Associates Inc. Sunderland MA 2nd edn. 1992. 2 M. S. P. Sansom L. R. Forrest and R. Bull Bioessays 1998 in press. 3 D. A. Doyle J. M. Cabral R. A. Pfuetzner A. Kuo J. M. Gulbis S. L. Cohen B. T. Cahit and R. MacKinnon Science 1998 280 69. 4 D. Boyd C. Schierle and J. Beckwith Protein Sci. 1998 7 201. 5 E. Wallin and G. von Heijne Protein Sci. 1998 7 1029. 6 O. S. Andersen Annu.Rev. Physiol. 1984 46 531. 7 G. A. Woolley and B. A. Wallace J. Membr. Biol. 1992 129 109. 8 B. A. Wallace Prog. Biophys. Mol. Biol. 1992 57 59. 9 B. Roux and M. Karplus Annu. Rev. Biophys. Biomol. Struct. 1994 23 731. 10 S. Oiki V. Madison and M. Montal Proteins Struct. Func. Genet. 1990 8 226. 11 M. S. P. Sansom I. D. Kerr G. R. Smith and H. S. Son V irology 1997 233 163. 12 M. S. P. Sansom C. Adcock and G. R. Smith J. Struct. Biol. 1998 121 246. 13 M. S. P. Sansom Qt. Rev. Biophys. 1993 26 365. 14 D. S. Ca–so Annu. Rev. Biophys. Biomol. Struct. 1994 23 141. 15 R. O. Fox and F. M. Richards Nature (L ondon) 1982 300 325. 16 G. Esposito J. A. Carver J. Boyd and I. D. Campbell Biochemistry 1987 26 1043. 17 J. C. Franklin J. F. Ellena S.Jayasinghe L. P. Kelsh and D. S. Ca–so Biochemistry 1994 33 4036. 18 C. E. Dempsey J. Am. Chem. Soc. 1995 117 7526. 19 C. E. Dempsey and L. J. Handcock Biophys. J. 1996 70 1777. 20 N. Gibbs R. B. Sessions P. B. Williams and C. E. Dempsey Biophys. J. 1997 72 2490. 21 G. Baumann and P. Mueller J. Supramol. Struct. 1974 2 538. 22 M. K. Mathew and P. Balaram FEBS L ett. 1983 157 1. 23 G. Boheim W. Hanke and G. Jung Biophys. Struct. Mech. 1983 9 181. 24 K. He S. J. Ludtke H. W. Huang and D. L. Worcester Biochemistry 1995 34 15 614. 25 J. Breed R. Sankararamakrishnan I. D. Kerr and M. S. P. Sansom Biophys. J. 1996 70 1643. 26 D. P. Tieleman H. J. C. Berendsen and M. S. P. Sansom Biophys. J. 1999 in press. 27 S. You S. Peng L. Lien J. Breed M. S. P. Sansom and G.A. Woolley Biochemistry 1996 35 6225. 28 G. A. Woolley P. C. Biggin A. Schultz L. Lien D. C. J. Jaikaran J. Breed K. Cowhurst and M. S. P. Sansom Biophys. J. 1997 73 770. 29 J. Breed P. C. Biggin I. D. Kerr O. S. Smart and M. S. P. Sansom Biochim. Biophys. Acta 1997 1325 235. 30 G. Molle J. Y. Dugast G. Spach and H. Duclohier Biophys. J. 1996 70 1669. 31 J. Breed I. D. Kerr G. Molle H. Duclohier and M. S. P. Sansom Biochim. Biophys. Acta 1997 1330 103. 32 M. S. P. Sansom Prog. Biophys. Mol. Biol. 1991 55 139. 33 O. S. Smart J. Breed G. R. Smith and M. S. P. Sansom Biophys. J. 1997 72 1109. 34 G. R. Smith and M. S. P. Sansom Biophys. J. 1998 75 2767. 35 K. M. Merz and B. Roux Biological Membranes A Molecular Perspective from Computation and Experiment Birkhaé user Boston 1996.36 K. M. Merz Curr. Opin. Struct. Biol. 1997 7 511. 37 D. P. Tieleman S. J. Marrink and H. J. C. Berendsen Biochim. Biophys. Acta 1997 1331 235. 38 D. J. Tobias K. C. Tu and M. L. Klein Curr. Opin. Coll. Interface Sci. 1997 2 15. 39 K. Belohorcova J. H. Davis T. B. Woolf and B. Roux Biophys. J. 1997 73 3039. 40 L. Shen D. Bassolino and T. Stouch Biophys. J. 1997 73 3. 41 T. B. Woolf Biophys. J. 1997 73 2376. 42 D. P. Tieleman M. S. P. Sansom and H. J. C. Berendsen Biophys. J. 1999 in press. 43 I. D. Kerr R. Sankararamakrishnan O. S. Smart and M. S. P. Sansom Biophys. J. 1994 67 1501. 44 C. Adcock G. R. Smith and M. S. P. Sansom Biophys. J. 1998 75 1211. 45 D. Bashford and M. Karplus J. Phys. Chem. 1991 95 9556. 46 A. KarshikoÜ V.Spassov S. W. Cowan R. Ladenstein and T. Schirmer J. Mol. Biol. 1994 240 372. 47 C. Lim D. Bashford and M. Karplus J. Phys. Chem. 1991 95 5610. 48 H. J. C. Berendsen D. van der Spoel and R. van Drunen Comput. Phys. Commun. 1995 95 43. 49 B. Hess H. Bekker H. J. C. Berendsen and J. G. E. M. Fraaije J. Comput. Chem. 1997 18 1463. 50 H. J. C. Berendsen J. P. M. Postma W. F. van Gunsteren A. DiNola and J. R. Haak J. Chem. Phys. 1984 81 3684. 51 O. Berger O. Edholm and F. Jahnig Biophys. J. 1997 72 2002. 52 S. J. Marrink O. Berger D. P. Tieleman and F. Jahnig Biophys. J. 1998 74 931. Faraday Discuss. 1998 111 209»223 222 53 J. Hermans H. J. C. Berendsen W. F. van Gunsteren and J. P. M. Postma Biopolymers 1984 23 1513. 54 D. P. Tieleman and H. J. C.Berendsen J. Chem. Phys. 1996 105 4871. 55 D. P. Tieleman and H. J. C. Berendsen Biophys. J. 1998 74 2786. 56 W. Kabsch and C. Sander Biopolymers 1983 22 2577. 57 A. T. Brué nger X-PL OR V ersion 3.1. A System for X-Ray Crystallography and NMR Yale University Press New Haven CT 1992. 58 P. J. Kraulis J. Appl. Crystallogr. 1991 24 946. 59 M. E. Davis J. D. Madura B. A. Luty and J. A. McCammon Comput. Phys. Commun. 1991 62 187. 60 O. S. Smart J. M. Goodfellow and B. A. Wallace Biophys. J. 1993 65 2455. 61 D. SitkoÜ D. J. Lockhart K. A. Sharp and B. Honig Biophys. J. 1994 67 2251. 62 B. M. P. Huyghues-Despointes J. M. Scholtz and R. L. Baldwin Protein Sci. 1993 2 1604. 63 W. Hanke and G. Boheim Biochim. Biophys. Acta. 1980 596 456. 64 M. S. P. Sansom G.R. Smith C. Adcock and P. C. Biggin Biophys. J. 1997 73 2404. 65 L. R. Forrest and M. S. P. Sansom Biochem. Soc. T rans. 1998 26 S303. 66 P. C. Biggin and M. S. P. Sansom Biophys. Chem. 1996 60 99. 67 P. Biggin J. Breed H. S. Son and M. S. P. Sansom Biophys. J. 1997 72 627. 68 Q. Zhong P. B. Moore D. M. Newns and M. L. Klein FEBS L ett. 1998 427 267. 69 B. Roux Biophys. J. 1997 73 2980. 70 K. R. Ranatunga I. D. Kerr C. Adcock G. R. Smith and M. S. P. Sansom Biochim. Biophys. Acta 1998 ”kerfeldt J. D. Lear Z. R. Wasserman L. A. Chung and W. F. DeGrado Acc. Chem. Res. 1993 26 1370 1. 71 I. H. Shrivastava C. Capener L. R. Forrest and M. S. P. Sansom unpublished work. 72 K. S. 191. 73 Q. Zhong Q. Jiang P. B. Moore D. M. Newns and M. L. Klein Biophys. J. 1998 74 3. 74 G. R. Smith and M. S. P. Sansom Biophys. J. 1997 73 1364. 75 P. Mitton and M. S. P. Sansom Eur. Biophys. J. 1996 25 139. 76 B. Roux and M. Karplus Biophys. J. 1991 59 961. 77 B. Roux Biophys. J. 1996 71 3177. 78 V. Dorman M. B. Partenskii and P. C. Jordan Biophys. J. 1996 70 121. Paper 8/06266H 223 Faraday Discuss. 1998 111 209»223

ISSN:1359-6640    

DOI:10.1039/a806266h

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

General Discussion Dr Duax opened the discussion of Prof. Wallaceœs paper It is not clear whether your references to the CD spectra of gramicidin in acetic acid in your contribution to this meeting are meant to question the accuracy of the X-ray structure determinations reported in our PNAS paper.1 One criticism of X-ray studies of gramicidin has been that crystals were grown from alcohols which have relative permittivities near 50. This is very high compared to the relative permittivities of lipids of 2 to 4. To address this problem we grew crystals from glacial acetic acid which has a relative permittivity much closer to that of lipids (5»6). We did not claim that acetic acid resembled lipids in other ways. Our crystal structures of the Cs` complex of gramicidin from methanol and the hydronium ion complex of gramicidin from glacial acetic acid reveal the same right handed antiparallel double helix.These structures agree in all details with the model –rst proposed by Arseniev et al.2 to interpret the NMR Noesy spectra of the Cs` complex of gramicidin in a methanol»chloroform mixture. Our structure is also compatible with 15N NMR data on gramicidin in planar lipid bilayers.3 Our structure is consistent with the standard characterization of the membrane active channel of gramicidin based upon CD spectra as being right handed. While the signi–cance and implications of an X-ray structure may be open to debate the accuracy of a high resolution X-ray structure determination can and should be unambiguous. As you know the results of our X-ray analysis of Cs` and K` complexes of gramicidin are completely diÜerent from yours diÜerent in molecular hand ion content nature of ion coordination channel shape and every single one of the 30 hydrogen bonds present in the antiparallel dimers.On the basis of published results we conclude that your reported structural models have chemical and stereochemical anomalies and crystallographic limitations not present in our re–ned structures. The disparities between our structures and yours are all the more puzzling because the crystallization conditions were very similar and all the cell dimensions of the complexes agree to within 1%. In light of the controversy concerning the relevance of the X-ray results to solution spectra transport properties and the mechanism of membrane transport of gramicidin we consider it vitally important to resolve the questions generated by the profound diÜerences between our X-ray crystal structures and yours.There are three possibilities. (1) Our structures are wrong. (2) Your structures are wrong. (3) Both structures are right. Which of these possibilities is in fact the case can be quickly resolved by comparing the diÜraction intensities that were used in our determinations and yours. We deposited the atomic coordinates and all of the intensities for our Cs` complex in the Protein Data Bank (PDB) when the paper was accepted in PNAS (two months before it was published in November 1998). You told me this morning that you attempted to access our intensities –les through the PDB yesterday and could not.I am sorry to hear that the PDB is unable to cope with deposition and distribution in a timely fashion. We will gladly provide you with the intensity data on our Cs` complex directly. In the spirit of fairness and in the interest of science we would request that you reciprocate by sending us the intensity data for your Cs` complex within 10 days of receipt of our intensities data. We will be happy to do the same with our data on the K` Rb` and other complexes. Will you do this ? We hope that you will agree that it is important to resolve this ambiguity. If your structures and ours are correct it would be the –rst example that I know of where isomorphous forms of a structure with identical cell dimensions have entirely diÜerent conformations.Given the existence of 160 000 structures in the Cambridge Structural Database such an unusual phenomena is certainly worth characterizing carefully. 1 B. M. Burkhart N. Li D. A. Langs W. A. Pangborn and W. L. Duax Proc. Natl. Acad. Sci. USA 1988 95 12950. 2 A. S. Arseniev I. L. Barsukov and V. F. Bystrov FEBS L ett. 1985 180 33. 225 Faraday Discuss. 1998 111 225»246 3 L. K. Nicholson F. Moll T. E. Mixon P. V. LoGrasso A. L. Lay and T. A. Cross Biochem. 1987 26 6621. Dr Wallace replied You have raised a number of points in your comments so I will try to answer them one by one (1) The CD studies in acetic acid were done to address the statements in your PNAS paper that this form represented the conducting form in lipid membranes.They clearly show that the acetic acid form is not the conducting form found in phospholipids but they in no way question the accuracy of your X-ray structure merely the interpretation of it. (2) Some of the other evidences you cite for this being the conducting form are also controversial. While I am not an expert in solid state NMR I understand that a paper disputing your claims that this structure is compatible with the spectroscopic data has been submitted1 that will address that issue better than I can in this limited space. (3) I agree that our X-ray structure of the CsCl form is completely diÜerent from yours. However the crystals examined in each case are diÜerent and given the polymorphic nature of this molecule which can demonstrably form structures of diÜerent handedness diÜerent stagger between the chains and diÜerent hydrogen bonds in solution it does not seem particularly surprising that a similar polymorphism is seen in the crystals.1 T. A. Cross A. Arseniev B. A. Cornell J. H. Davis J. A. Killian R. E. Koeppe L. K. Nicholson F. Separovic and B. A. Wallace submitted. Dr Duax further commented You did not respond to the most important question. Will you release the intensities from your published structures to the PDB as we have? Dr Wallace responded We are currently re-re–ning our CsCl structure1 using improved geometric parameters as you have recently done2 with the earlier BuÜalo structure.3 Once done we intend to publicly release our structure via the PDB in the standard fashion. 1 B. A.Wallace and K. Ravikumar Science 1988 241 182. 2 B. M. Burkhart R. M. Gassman D. A. Langs W. A. Pangborn and W. L. Duax Biophys. J. 1998 75 2135. 3 D. A. Langs Science 1988 241 188. Dr Burkhart asked How much interpretation can you make about the eÜect of side chain conformational change on the CD spectrum diÜerences ? Dr Wallace responded The CD spectra1 do not show signi–cant diÜerences between gramicidin in diÜerent solvents. I know that your crystal structures2 do show a diÜerence in the orientations of the tryptophans between the methanol and ethanol or propanol structures. This may be due to the diÜerences between solution and crystal structures especially any constraints on the tryptophan side chains which are located at the periphery of the moecule due to packing in the crystals.That the two crystal structures crystallised in the same space group (ethanol and propanol) have similar tryptophan orientations and the one crystallised in another space group (methanol) is most diÜerent might tend to support the crystal packing as a source of these diÜerences between solution and solid state studies. 1 W. R. Veatch E. T. Fossel and E. R. Blout Biochemistry 1974 13 5249; Y. Chen and B. A. Wallace Biopolymers 1997 42 771. 2 B. M. Burkhart R. M. Gassman D. A. Langs W. A. Pangborn and W. L. Duax Biophys. J. 1998 75 2135. Prof. Holzwarth asked Why did you use a mixture of gramicidin A B and C and not a pure sample? I would expect diÜerences between a mixture and pure samples of gramicidin. Dr Wallace responded We used the commercially available mixture of gramicidin A B and C.Veatch had previously shown1 that the mixture behaves spectroscopically in a very similar manner to pure gramicidin A. 1 W. R. Veatch PhD Thesis Harvard University 1974. Faraday Discuss. 1998 111 225»246 226 Prof. Laggner commented Could it be that both structures (Wallace and Duax) are right but not really relevant to the structure in the bilayer system. Current ideas in this –eld suggest that the structure of peptides in the membranes may not be unique but rather dependent on various variables e.g. the amount of peptide incorporated to the bilayer. In BLMs monomer it is rather difficult to control the amount of peptide actually present in the ììblackœœ part of the –lm. Dr Wallace replied I agree entirely.From all the physical and chemical studies done on gramicidin over the past 20 years or more it is clear that the principle conducting form in membranes is the helical dimer not any of the double helices seen in either Duaxœs and our structures. The studies presented in my paper at this meeting con–rm this and suggest there are only very speci–c conditions under which double helices may form conducting molecules such as when there is a very severe mis-match of the lipid fatty acid chain length with the gramicidin size. I should point out that the Duax CsCl crystals and ours are prepared under very diÜerent conditions diÜerent peptide and ion concentrations and ratios and diÜerent temperatures (which our recent CD studies1 have shown produce very diÜerent spectra and by implication structures in solution).The crystals have slightly diÜerent unit cell dimensions (but small variations in unit cell dimensions have signi–cant consequences for this molecule.2 Given the polymorphism seen for this molecule in so many diÜerent environments (solution membranes etc.) it is not surprising to –nd polymorphic crystal structures. It is very clear from our anomalous Patterson maps for instance that the Duax structure is not consistent with our data and so must represent a diÜerent crystal form. 1 T. P. Galbraith and B. Wallace unpublished results. 2 D. A. Doyle and B. A. Wallace J. Mol. Biol. 1997 266 963. Prof. Lee said My question is about the eÜect of membrane thickness on the state of gramici- PC is gramicidin present as a b6.3 dimer or monomer or will the CD spectra of din.In di-C22>1 these two forms be indistinguishable ? Second I thought that the thickness of a solvent-free bilayer and the equivalent solvent containing bilayer were very diÜerent so that it seems surprising that chain length eÜects are the same in the two systems. Dr Wallace responded We cannot clearly distinguish b6.3 monomers from b6.3 dimers by CD spectroscopy. Both have similar backbone folds (which is what CD detects). CD studies we did some time ago1 on gramicidin in which the N-terminal formyl group was replaced with an Nterminal acetyl group which tends to destabilise the dimer and thus produce monomers showed very little diÜerence between the spectra so I donœt think we can say for certain whether the structures in C lipids are monomers or dimers from our work.However the conductance 22 studies of Mobashery et al.2 suggest that a signi–cant proportion of the population in this lipid is dimeric and conducting. I agree with you that the bulk thicknesses of bilayers with and without solvent are quite diÜerent. But what I donœt think we know is the actual thickness of the bilayer immediately surrounding and in contact with the gramicidin. In fact I have to say that I was surprised that the ììchain length eÜectœœ we are seeing only starts at C22 . I would have thought that much shorter lipids (even as short as C18) would be a mis-match for gramicidin. But that is again based on thicknesses measured for bulk lipids in the absence of peptide and the lipid surrounding the gramicidin may have substantially diÜerent properties.1 B. A. Wallace W. R. Veatch and E. R. Blout Biochemistry 1981 20 5754. 2 N. Mobashery C. Nielsen and O. S. Andersen FEBS L ett. 1997 412 15. Prof. Roux asked It is not clear to my why the DH form gets stabilized in thick membrane relative to the HD form. Which one is stabilized or perhaps more justly which one is destabilized ? Dr Wallace responded I would think that really we are talking about the absence of stabilizing eÜects as much as anything. Numerous studies have suggested that interactions between tryptophan side chains and the interfacial region of the bilayer may be important structurally. In thin 227 Faraday Discuss. 1998 111 225»246 lipids the HD with its tryptophans near the ends of the molecule has them in a position that could interact favourably with the lipid headgroup region which would have a stabilising in—uence; in the DHs the tryptophans are spread along the length of the molecule and so some of them would have to be buried deep in the bilayer a much less favourable disposition.Thus in this case the equilibrium may be shifted towards the HD. However in thick lipids even the HD would have to bury some of its tryptophans below the bilayer interface therefore this factor would be less likely to shift the equilibrium towards the HD. The other factor in the mismatch case could be the number of intermolecular hydrogen bonds holding the dimers together which is 6 for the HD and between 26 and 30 for the DH.Dr Burkhart commented Given your own admission that gramicidin adopts many diÜerent structural forms and that the in—uences of Trp residue side chain conformation cannot be discerned is it not dangerous to choose only four structural forms to describe each CD spectrum. Dr Wallace responded The control we have for whether the CD data is well represented by the reference data spectra used in the analyses is the NRMSD parameter. It is much like an R-factor in crystallography in that it is a measure of the correspondence between the experimental data and the best –t to the reference data. A low value for the NRMSD indicates the reference data set re—ects well the population of conformers present in the sample. Dr Deber communicated What are the linear (vertical) dimensions of the DH vs.the HD forms of gramicidin ? Dr Wallace ” communicated in response The ion-containing DH is D26 long whilst the HD form is ” D32 (see ref. 1). 1 O. S. Smart J. M. Goodfellow and B. A. Wallace Biophysical J. 1993 65 2455. Dr Okazaki opened the discussion of Prof. Rouxœs paper Within my knowledge the interaction potential between protein cylinder and lipid chain carbon u(r) is essential for this kind of calculation. The results must be very sensitive to the potential function between unlike particles. If you assume the strongly attractive potential between them homogeneous mixing will be obtained but if you set the weakly attractive interaction you will obtain a kind of phase separation i.e. protein aggregation as you show in your work.You might get even a double-well-like free energy pro–le for a particular potential function. The results including the free energy pro–le must change dramatically as a function of the protein»lipid interaction. Now my question is what kind of protein»lipid potential function did you assume or in other words how did you determine the interaction function ? pp(r)]*W (r) where Upp(r) is the microscopic protein» Prof. Roux responded The protein»protein potential of mean force W (r) depends on several factors. One usually writes that W (r)\U protein potential energy and *W (r) is the lipid-mediated free energy potential. What you are saying is that the latter depends on the protein»lipid interactions which is absolutely correct.In this preliminary study we were mostly interested in exploring the magnitude of the forces arising from the in—uence of excluded volume of the hydrocarbon chains by the protein inclusion. Therefore in the present calculation the protein»lipid potential was simply chosen as a repulsive hard cylinder. Of course more realistic interactions could be used (see for instance our answer to Prof. Smith below) but they would not allow an investigation of the excluded volume eÜect on the protein»protein lipid-mediated forces. Dr Smart asked Do your results have similarities to the results of Huang and coworkers1 who have determined the 2D radial distribution function for peptides in lipid bilayers. 1 L. Yang T. A. Harroun W. T. Heller T. M. Weiss and H. W.Huang Biophys. J. 1998 75 641. Prof. Roux responded Huang has measured the in-plane distribution pair correlation function for the gramicidin channel. We intend to compare these data to the calculated lateral packing of cylinders corresponding to the size of a gramicidin channel. Faraday Discuss. 1998 111 225»246 228 Prof. Haymet commented The level of theory seems nicely appropriate for the problem. The difficulty arises with the new calculations which you have discussed now and are not in the printed copy of the paper. The approximation for hydrophobicity which inspired your approach has long been known to break down the size of the solute increases beyond the solvent diameter. Hence calculation of the interaction of two cylinders are likely to be similarly qualitatively incorrect as the cylinder radium increases beyond the characteristic length-scale of the medium in which they are immersed.This is due to the fact that the solvent response-function is not allowed to relax in the presence of the solute. Prof. Roux responded In the paper were described preliminary results obtained for a hard repulsive cylinder of 5 ” diameter. We observe that the lipid-mediated potential of mean force has a complex structure with an attractive well at contact and a repulsive barrier at a cylinder» cylinder separation distance of 20 ”. It is of interest to examine what is the dependence of this result on the size of the protein inclusion and that is the reason why we are investigating this matter. Nevertheless I agree with you that the HNC integral equation theory for simple liquids formed by spherical particles has problems in dealing with very large diÜerences in particle size though I would add that the present case is more complex since there are several lengthscales in the lipid»lipid pair correlation function smm(r) whereas the situation with monoatomic liquids is much simpler.Comparison with molecular dynamics simulations of atomic models will be done in order to assess the range of validity of the current theory for lipid bilayers. The extensive simulations performed in Prof. Kleinœs group will provide a good basis for that. Nevertheless whether the theory is yielding semi-quantitative or qualitative results is secondary at this point. One must realize that an integral equation theory such as described here provides a unique route to gain some insights into lipid-mediated potential of mean force between protein inclusions and the in—uence of the hydrocarbon chains on the protein»protein interactions in bilayer membranes.There is presently no other theoretical approach to gain such information. Prof. Holzwarth asked Are you in a position to predict how far the in—uence of proteins could reach into the membrane; especially how many lipid layers around the protein will be in—uenced? We found experimentally for bacteriorhodopsin that as many as –ve to six layers of surrounding lipids are in—uenced.1 1 A.Boé ttcher N. Dencher R. Groll F. Meyer and J. F. Holzwarth Reactions in Compartmentalized L iquids ed. W. Knoche and R.Schomaé cker Springer Verlag Berlin 1989 pp. 105»115. Prof. Roux responded In Fig. 2 we show that the density of the hydrocarbon core is perturbed over a distance of 30 ” ” ” around a 5 cylinder. The radius of one DPPC being roughly 4.5 based on a surface area per molecule of 64 ”2 (using nr2) this implies that a layer of 2 to 3 lipids are perturbed. The diÜerence may be due in part to the fact that bacteriorhodopsin is much larger (it is a bundle of 7 transmembrane helices). Furthermore in reality there is a direct attractive dispersion interaction between the protein and the lipids hydrocarbon chains whereas here we consider only the excluded volume eÜect. This may amplify the eÜect. Lastly one should keep in mind that our theory oÜers an equilibrium statistical mechanical view of the perturbed density around an impurity.Ultimately the number of observed lipids may depend upon the experimental method used to characterize the system (e.g. magnetic resonance —uorescence infrared spectroscopy differential calorimetry etc.). Prof. Klein commented A liquid hydrocarbon against a hard wall will exhibit layering of methylene groups that extends 4 layers or so into the bulk liquid. Thus the oscillation you observe in the hydrocarbon core around the 5 ” cylinder is reminiscent of this eÜect. It would be interesting to compare the DPPC lipid with for example hexadecane bulk liquid to see if there is any speci–c eÜect of the lipid. The main input in your calculations is the simulation data for the carbon»carbon density»density response function.These data were generated with a relatively small sample. How con–dent are you that the values are reliable in the range 15 ” »20 and will uncertainties in this asymptotic region in—uence your results ? Prof. Roux responded Fig. 1 shows that the carbon»carbon intramolecular correlation function Smm(r) has a large contribution followed by a peak up to 3 ” and then a slow decay over a 229 Faraday Discuss. 1998 111 225»246 distance of 10»15 ”. Clearly the intramolecular correlation contributes signi–cantly to the response function of the lipid bilayer the short range structure in the intramolecular correlation arises from nearest neighbor carbons along the acyl chains. The peak at r\0 is indicative of the signi–cant amount of short range order in the lipid chains perpendicular to the plane of the bilayer.The oscillations you are referring to arise from the intermolecular contributions in the response function. Those correspond partly to carbon»carbon contacts and are reminiscent of liquid hydrocarbon. You are raising a very interesting question to what extent does the response of the hydrocarbon diÜer from that of an isotropic liquid hydrocarbon? We will try to address that in the future by using a response function extracted from a liquid hydrocarbon simulation. Concerning your second question the pair correlation function was calculated from a molecular dynamics trajectory of a lipid bilayer generated by Feller et al.1 The atomic system that they simulated consisted of 72 DPPC molecules (36 in each lea—et).The physical length of the periodic box is approximately 48 ”. Thus since the the correlation function decays almost to zero over a distance of 10 to 15 ” it seems reasonable to assume that the dominant packing structure was captured by the molecular dynamics simulation. 1 S. E. Feller R. M. Venable and R. W. Pastor L angmuir 1997 13 6555. Prof. Petersen asked Please speculate on the interactions among several cylinders. Will there be an optimum size ? Will a bundle of 3 ” ” »5 cylinders of 5 radius behave as a single cylinder of D9 and hence there might be no further aggregation ? Prof. Roux responded The current calculations correspond to the in–nite diluted limit of protein inclusion in a lipid bilayer. This assumption is necessary since we possess only the response function smm(r) of the unperturbed bilayer from the simulation of Feller et al.1 Nevertheless one could try to investigate –nite concentration and aggregation eÜects with the current theory keeping in mind its limitations due to the response function.1 S. E. Feller R. M. Venable and R. W. Pastor L angmuir 1997 13 6555. Prof. Smith asked What are the prospects for evolution of this approach towards an all-atom description of the helices ? Prof. Roux responded The current theory is designed to address the in—uence of the lateral packing of the lipid hydrocarbon chains on the protein»protein potential of mean force. For this purpose the theory was kept as simple as possible. In particular we considered only hard cylindrical protein inclusions with no details.Nevertheless it is possible to construct a more sophisticated integral equation theory in which the detailed atomic structure of a membrane-bound protein will be used (e.g. transmembrane helix or an amphipatic helix associated at the membrane/solution interface). This extended integral equation theory would require a more complete response function than the simple smm(r) used in the present theory and which characterizes the lateral —uctuations. The extended approach would be the equivalent of integral equation theories such as those described in ref. 26. We are currently working on the development of this extended theory. Dr Gilbert opened the discussion of Dr Bezrukovœs paper. In the experiments on alamethicin reported larger pores appear to be favoured in their formation by the presence of lipids possessing a propensity for the formation of H phases (induced by pH modi–cation).Does alamethicin itself II promote H formation? Is this relevant to consideration of these data? Secondly do the tempo- II rally co-existing hexagonal (HII) and lamellar phases also coexist within a single lipid body»i.e. within the surface of a liposome? If so what form would the interface between the lamellar and H phases take ? II Dr Bezrukov responded Yes not only lipid monolayer spontaneous curvature modi–es alamethicin channel behavior but in turn alamethicin itself in—uences lipid curvature properties. According to an X-ray and NMR study by Keller et al.,1 addition of as little as 1% of alamethicin to 1,2-dielaidoyl-sn-lycero-3-phosphoetanolamine introduces a large region of cubic phase into the Faraday Discuss.1998 111 225»246 230 thermal phase diagram. This observation could be important for the molecular model of the phenomenon. To answer your second question in the X-ray measurements that are reported in our paper the samples were prepared in excess water. Both lamellar and hexagonal phases are independent three-dimensional structures. So necessarily they exist separately and not on a twodimensional surface but within the same bulk sample. We do not know the structural nature of the interface between them. 1 S. L. Keller S. M. Gruner and K. Gawrisch Biochem. Biophys. Acta 1996 1278 241. Dr Gon8 i asked To what extent do your ììchange in pHœœ experiments answer the criticism raised in relation to the PC/PE data that you were changing the chemistry of the system? Your results look reasonable to me but in my view protonation is also changing the chemistry of the host lipid towards alamethicin.Have you thought of changing curvature of PC bilayers by adding lyso PC? Or conversely adding polyunsaturated PC to saturated PC? Very little changes in the ììchemistryœœ would take place in such an experiment. Dr Bezrukov responded The answer probably depends very much on what you mean by ììchemistryœœ. If changing head group interaction is chemistry then yes we are changing the chemistry of host lipid. In our opinion however changing the pH means less chemical modi–cation of the system than admixing/substituting one lipid species by a distinct second lipid species.In response to your second point in principle this is a very good suggestion. In practice it may be difficult to design a reliable experiment and to rationalize obtained results due to relatively high water solubility of lyso PC. Besides an additional worry in experiments with the lipid mixtures is a possible ììdemixingœœ of lipids in the vicinity of the channel. Prof. Laggner asked Two related questions What are the relative populations of lamellar and H phases in the coexistence region between pH 1.5 and 3.5. Why do you refer to these two states II as ìì interstable œœ ? Dr Bezrukov responded We can only estimate the relative quantities based on relative X-ray intensities. Qualitatively the lamellar phase appears to be maximum between pH 2.3 and 2.9 where it coexists with the hexagonal phase and appears to involve less than 50% of the lipid.The lamellar phase is nearly but not quite absent at pHs above and below 2.3 and 2.9. As for your second question the X-ray scattering patterns are stable over periods of at least weeks. Prof. Neumann asked Does the dependence of the current intensity *i(t) on the electrolyte concentration re—ect increasing charge screening of the phosphatidylserine (PtdSer) groups and thereby aÜect the spontaneous curvature ? Dr Bezrukov responded The increase of single-channel current with the electrolyte concentration is mostly related to the increase in the average channel occupancy by ions. As seen in Fig. 6 of our paper this increase mostly follows the bulk solution conductivity.The eÜect of increasing lipid charge screening that you mention is very small because the initial contribution from lipid charge to the open channel conductance is tiny. This is not always the case though. For ììsmallœœ channels it can be quite pronounced. Recently we published a study where the lipid charge eÜect on channel conductance was studied over varied charge densities. 1 T. K. Rostovtseva V. M. Aguilella I. Vodyanoy S. M. Bezrukov and V. A. Parsegian Biophys. J. 1998 75 1783. Dr L. Fisher commented I totally agree that membrane lipids must be functionally involved and have a controlling role in membrane protein behaviour. My question is whether we are looking at the right end of the lipid.Lipid packing studies to date have tended to focus on non-glycosylated lipids. It is a fact of nature though that the outer membrane lipids of mammalian cells are heavily glycosylated and it may be that interactions between the headgroups rather than the chains of such lipids dominate their function. Would you care to comment on this and on whether chain and charged headgroup interaction studies are of relevance for such lipids and the membranes containing them? 231 Faraday Discuss. 1998 111 225»246 Dr Bezrukov responded Yes this is very interesting question though I do not think that we or anybody else for that matter have enough evidence to state that interactions between the headgroups are more important than chain interactions. Many factors contribute to the balance of energies in channel conformational equilibrium.Glycosylation is likely to be important; at least the cells seem to think so. Whether head or tail ììdominatesœœ I would rather not decide in general. Prof. Roux said I fail to see a relationship between the propensity of forming the H phase and II the stabilization of alamethicin multimers which are a bundle of transmembrane helices. Could you comment on that please ? Dr Bezrukov responded If you ask about the empirical relationship it is in Fig. 3 vs. Fig. 10 of our paper. As for a detailed molecular mechanism we are still far from any good idea. One possibility could be the ììshapeœœ of the channel to explain the increase in the number of stressrelieved lipid molecules as the channel goes to a state of higher conductance.Prof. Lee commented The difficulty you have as you have said is in separating eÜects of the chemical structure changes from eÜects on curvature. The transition temperature into the hexagonal H phase is higher for POPE than for DOPE. Have you compared the eÜects of POPE and II DOPE on alamethicin channel formation? Dr Bezrukov responded No we have not. But this is de–nitely a good suggestion. Dr Smart asked Can your results be explained if the alamethicin molecule adopts a wedgeshaped bundle (and that this varies with bundle size) ? Can Dr Sansom comment whether the models he has generated support this ? Dr Bezrukov responded Yes the shape of the channel exterior facing lipid matrix could be a key factor. I will leave this question for Dr Sansom who did extensive work modeling alamethicin channel.Dr Sansom responded The models of alamethicin helix bundles we have generated are to a limited extent wedge-shaped (if one takes a cross section down the pore i.e. bundle axis). This may provide a partial explanation of the result of Bezrukov and colleagues. However to be certain of this probably requires a systematic series of simulations with e.g. diÜerent lipid headgroups or headgroup protonation states. I donœt think ììback of an envelopeœœ approaches are going to work here. Dr Kakorin asked Did you change the pH value and NaCl concentration equally on the two sides of the lipid membrane? Dr Bezrukov responded Yes we have varied the pH value and NaCl concentration equally on both sides of the membrane.In the future we are going to change the pH value on one side. Dr Kakorin responded If there were not a diÜerence in pH value or NaCl concentration on the two sides of the membrane then you cannot rationalise the variation of the conductance level of alamethicin channels with pH value or [NaCl] by the proton-induced increase in spontaneous curvature. The cartoon illustration of the curved monolayer (Fig. 9) is not relevant to the data. Actually according to Helfrichœs de–nition the spontaneous curvature re—ects a possible asymmetry in the bilayer,1 not in a monolayer as in Fig. 9. The physical»chemical origin of the asymmetry could be either a diÜerent chemical environment on both sides of membrane or a diÜerent chemical composition of the two monolayers.2 Indeed the variation of pH value and NaCl concentration may change only the Debye screening length and thereby the electrostatical interaction between charged lipid headgroups in two monolayers.This can lead to a change in the lateral membrane tension which may be operative to modulate the conductance level of alamethicin channels. 1 W. Helfrich Z. Naturforsch. C 1973 28 693. 2 U. Seifert Adv. Phys. 1997 46 13. Faraday Discuss. 1998 111 225»246 232 Dr Bezrukov responded While we cannot rule out other factors we correlate our –ndings exactly with what you suggest with the pH-induced change in lateral pressure. As it is well known lateral pressure in a monolayer of a symmetric bilayer is distributed diÜerentially and changes along monolayer depth (e.g.see Fig. 1 of the paper by Templer et al. in this volume). Therefore every attempt to reduce actual distribution with a single number is a simpli–cation of the actual situation. We correlate probabilities of alamethicin channel states with the spontaneous monolayer curvature1 which is related to the –rst moment of the lateral pressure.2 The pH variation changes lipid charge thus changing lipid head»head interactions ; the resulting modi–cation in lateral pressure distribution is manifested by the change in lipid spontaneous curvature (Fig. 10). In our earlier work with alamethicin channel in PE»PC mixtures we were –rst to relate changes in the single-channel expression with lipid spontaneous curvature.3 Recently Olaf Andersen and his colleagues4 have shown that this parameter is also very important for the gramicidin A channel activity.Moreover they showed that at least in the case of gramicidin A and solvent-free bilayers the channel expression is much more sensitive to changes in the –rst moment of lateral tension than to changes in its average value at lipid substitution.5 Thus to say that the cartoon in Fig. 9 is not relevant to our channel data is to put it mildly an exaggeration. For further reading I would recommend the special issue of Chemistry and Physics of L ipids,6 devoted to the functional role of non-lamellar lipids. 1 S. M. Gruner Adv. Chem. Ser. 1994 235 129. 2 R. H. Templer S. J. Castle A. R. Curran G. Rumbles and D. R. Klug Faraday Discuss. 1998 111 41; M.M. Kozlov and V. S. Markin J. Chem. Soc. Faraday T rans. 2 1989 85 261. 3 S. L. Keller S. M. Bezrukov S. M. Gruner M. W. Tate I. Vodyanoy and V. A. Parsegian Biophys. J. 1993 65 23. 4 J. A. Lundbaek A. M. Maer and S. O. Andersen Biochemistry 1997 36 5695 and references cited therein. 5 C. Nielsen M. Goulian and O. S. Andersen Biophys. J. 1998 74 1966. 6 Chem. Phys. L ipids Special Issue 1996 81(2). Dr Templer commented The curvature elastic stress that is stored in a lipid bilayer is in general non-zero. For a —at bilayer which has the same lipid composition on either side the Helfrich Hamiltonian as expressed in terms of the bilayer would indeed be zero. However if it is expressed in terms of each constituent monolayer it becomes evident that each monolayer wishes to bend to the same degree but in opposite directions.Helfrich has called this the torque tension and the frustrated curvature energy stored in each monolayer per unit area is proportional to the square of the spontaneous curvature of the monolayer and the bending modulus of the monolayer. Since the monolayer is held —at this curvature elastic energy must be stored in some way. This is done by increasing the area dilation at the interface. Dr Sansom commented Alamethicin helices are linked (by the glycine-X-X-proline motif) and thus to a crude approximation helix bundles are somewhat ìì hourglassœœ shaped in cross-section. Thus lipids which prefer the H phase may pack better around the helix bundle. II Dr Bezrukov responded Yes I agree.This could be one of the possibilities for explaining the positive correlation between probabilities of higher conductance channel states and H lipid pro- II pensity. II Prof. Evans asked Is the eÜect of ìspontaneous curvatureœ on channel activity altered by organic solvents that partition in the bilayer interior (e.g. as your group has shown for the H phase transition of DOPC)? Dr Bezrukov responded We did not study that. The present paper reports results for ììdryœœ membranes obtained by Montal»Mueller monolayer opposition technique. No solvents were used in the X-ray liquid-crystal phase preparations. Dr Bezrukov opened the discussion of Dr Smartœs paper What you are trying to do is very important for the central issue in molecular biophysics the issue of structure»function relationship.For ionic channels their transport properties in particular conductance are of prime interest. However the predictive power of your approach is not clear. Consider the following example. It is 233 Faraday Discuss. 1998 111 225»246 well known (structural data molecular models permeation results) that the pore radius of the channel formed by gramicidin A is about three times smaller than the pore radius of amphotericin B channel. Using your approach we would expect about ten times higher conductance for amphotericin B channels. The single-channel measurements show that the actual situation is reversed amphotericin B channel conductance is about three times smaller than the conductance of gramicidin A channel. Are you not worried about this 3000% discrepancy with your model prediction ? Dr Smart responded The main point to emphasize is that the predictive power of the technique has been tested on all ion channels where an experimental high resolution structure was available.1 In these tests prediction to within an average factor of 1.6 with a predictive r2 (ref. 2) of 0.9 (6 systems tested). Extending the test set to include model structures with a reasonable certainty produced results to within a factor of 1.8 and a predictive r2 of 0.46. In comparison to methods of forecasting binding affinities to enzymes on the basis of structure these –gures are good particularly for a –rst attempt. The method has not yet been tested on amphotericin B as the structure of the channel conformation has not been experimentally determined.Although models have been proposed3,4 they are by no means certain in particular the stoichiometry and role of sterol molecules is still unclear.4 The pore dimensions of the models have been based on the fact that the antibiotic allows the diÜusion of xylose or ribose but not larger sugars through cell walls.5 Experience has shown that such an identi–cation is good for rigid molecules such as porins (see ref. 5 of our paper) but it is quite possible that the —exibility and/or changes of stoichiometry of amphotericin B may be involved in xylose transport. If the discrepancy is real then it may be due to the fact that amphotericin B is a polyene rather than peptide and is proposed to have a channel lumen lined by hydroxy groups which makes it diÜerent in character from the channels used to parameterize the HOLE conductance prediction.In conclusion the discrepancy may not exist but if it does then this is worthy of further investigation as this would mean that the amphotericin B has a structure activity relation which is 30-fold diÜerent from the many channels which –t within the HOLE procedure. 1 G. R. Smith and M. S. P. Sansom Biophys. J. 1998 75 2767. 2 R. D. Cramer D. E. Patterson and J. D. Bunce J. Am. Chem. Soc. 1988 110 5959. 3 M. Bonilla-Marïç n M. Moreno-Bello and I. Ortega-Blake Biochim. Biophys. Acta 1991 1061 65. 4 M. Baginski H. Resat and J. A. McCammon Mol. Pharmacol. 1997 52 560. 5 B. de KruihÜ W. J. Gerritsen A. Oerlemans R. A. Demel and L.L. M. van Deenen Biochim. Biophys Acta 1974 339 30. Prof. Roux asked Ion channels generally exhibit saturation properties as a function of permeant ion concentration e.g. the conductance increases linearly at low concentration and then reaches a plateau at higher concentration. Sometimes it even decreases as the concentration is raised further. How is this Ohmœs law approximation dealing with such phenomena? Dr Smart responded In short HOLE does not attempt to deal with these phenomena. To date concentration eÜects have been ignored. In the original paper we state the desirability of making predictions in the low concentration range you mention. However we are at present limited to the concentrations at which experimental data is available. The method is successful despite the problem of the diversity of concentration (for details see my earlier reply to Dr Bezrukov).Hopefully in the future we will acquire data under consistent conditions for all channels with known structure. In many respects although a prediction of absolute conductance in the expected range can provide a useful guide in model validation the PEG addition experiment has the greatest potential to be of use yielding much more direct information. However at present there is not a sufficiently large amount of data to make con–dent interpretations. This is why we are extending the method by collecting data for channels of known structure. In passing the situation is even more complex than you state. The surface conductance eÜects can cause higher than expected conductances at low ion concentrations (see ref.50 of our paper) as the eÜective concentration of ions within a pore is increased with respect to bulk. Faraday Discuss. 1998 111 225»246 234 Dr Sansom asked As you suggest the 6-meric and 7-meric a-toxin pores correspond to low and high conductance states what is the timescale for switching between these two forms? Are you suggesting that even for more complex membrane proteins there is a similar problem to that discussed earlier for gramicidin i.e. that of relating diÜerent structure of a channel protein (static) to its biological function (dynamic)? Dr Smart responded Yes the problem of relating structure to function for ion channels is particularly acute. Even for gramicidin where not withstanding recent controversy we know the structure for the conducting form there is the problem of understanding the closure event.This is known to involve dimer breakdown into a monomeric form. However the exact monomeric conformation adopted is still unknown. For more complex behaviours this problem is more acute. It must be remembered that in single channel recordings we are watching a single molecule or molecular assembly in action. Very often there is a massive excess of ìì silent œœ molecules present. In these cases most other experimental techniques can be expected to yield information on the conformation of the closed states. The high and low conductance states may correspond to a diÜerence in oligomerization but this identi–cation is put forward as a working hypothesis worthy of further study rather than a –rm conjecture.It must be remembered that a major piece of evidence is the –t between the structurebased HOLE prediction of the eÜect of PEG on conductance and that further work is required to make identi–cation more –rm (the data reported here on gramicidin is the –rst part of this process). A coauthor of the paper has an alternative explanation that the change in conductane is principally due to an alteration in the ionization state of the channel (see ref. 50 of our paper). The switch between the states is rapid in the timescale of single channel recordings. However closure events are only very rarely observed in the absence of divalent ions. To further complicate the matter single channel recordings suggest that there are two low conductance states which diÜer in their own selectivity (see ref.50 and 51 of our paper). Prof. Smith asked How was the diÜusion coefficient calculated ? Was it from the time dependence of the mean-square displacements and if so did this function exhibit the required linearity ? Finally in which way is the timescale of the motion thus quanti–ed relevant to conductance? Dr Smart responded The diÜusion coefficients were calculated by Tieleman and Berendsen from a molecular dynamics simulation of OmpF. A full description of the method used is given in ref. 10 of our paper but the diÜusion coefficients D are derived from the mean square displace- z ment of atoms using limSMz(t)[z(0)N2T\6Dz t t?= Each water molecule was assigned to belong to a slice of ” z coordinate space (1.2 thick) and its displacement measured over the next 5 ps.At the end of this time interval a reassignment was made. The diÜusion coefficient was an average of all molecules within a slice. No information is given about the linearity you refer to. The relevance of the timescale of the motion to conductance is an interesting question. Smith and Sansom1 have shown that the diÜusion coefficients of ions within channels are aÜected by factors of the same magnitude. My interest in using the data is in taking an empirical approach. A bene–t of using the diÜusion coefficient correction is that it results in the correct ììboundary conditionœœ. As a channel gets sufficiently large to be regarded as macroscopic the diÜusion coefficients of the water within it will tend to bulk values and therefore the correction will tend to 1 and the predictions will reduce to Ohmœs law.The validity of this adaptation can only be proven by extending to all the systems analyzed with the original purely empirical correction and seeing whether it leads to improved predictions. 1 G. R. Smith and M. S. P. Sansom Biophys. J. 1998 75 2767. Dr Sansom responded For the alamethicin simulations and for OmpF (ref. 1) the water diÜusion coefficient was calculated from the mean square displacement over a 5 ps period. In an earlier study without a bilayer2 we have looked at ion diÜusion within pores and shown that diÜusion 235 Faraday Discuss. 1998 111 225»246 coefficients again on a ca.5 ps timescale are similarly reduced. However clearly these timescales are short relative to that of ion permeation. The relevance of such motions to conductance will depend inter alia on the strength of direct ion»pore interactions. 1 D. P. Tieleman and H. J. C. Berendsen Biophys. J. 1998 74 2786. 2 G. R. Smith and M. S. P. Sansom Biophys. J. 1998 75 2767. Prof. Klein commented With regard to diÜusion of water molecules through ion channels I mention that our MD study of LS2 suggested two types of water molecules. The majority diÜuse more or less unhindered through the pore but at about one-third of the bulk liquid value. But a few waters are hydrogen bonded to the inner wall of the channel with residence times of hundreds of ps. In layer channels with more pore water these long-lived bound waters may be less important.Dr Smart responded Presumably the diÜusion coefficient I have used here re—ects an average over these two types of water molecule. I think that the presence of tightly bound water may on occasions be important. Given a tight enough binding the water molecule may in eÜect become part of the channel. This could have some eÜect on the overall conductance but be very important in size selectivity. A possible example is the nicotinic acetyl choline receptor channel which appears from electron microscopic results to have a pore much larger than one would expect from its size selectivity. Dr Gilbert asked Does HOLE take account of the possibility of both surface and bulk conductance phenomena within a channel such as that formed by staphylococcal a-toxin.1 Would the surface/bulk conductance dichotomy be relevant in seeking to explain high and low (and intermediate) conduction states of such channels (especially considering the expected dynamic nature of their eÜective diameters) ? 1 Y.E. Korchev C. L. Bashford G. M. Alder P. Y. Apel D. T. Edmonds A. A. Lev K. Nandi A. V. Zima and C. A. Pasternak FASEB J. 1997 11 600. Dr Smart responded As discussed in my earlier response to Dr Sansom a-toxin has a diÜerence in charge selectivity between the high and low-conductance states. This has implications for interpreting the results of the PEG addition experiments in terms of a small pore size. As discussed in detail elsewhere (in ref. 5 of our paper) this interpretation is thrown into question by data of PEG addition to other channels.For example the diÜerence in the PEG addition curves observed between the C\1 and C\3 states of alamethicin which diÜer roughly ten fold in conductance is smaller than that between the ìhighœœ and ìlowœ conducting forms of a-toxin. It is almost universally agreed that the diÜerent conductance states for alamethicin re—ect diÜerent oligomerization numbers of the peptide rather than a diÜerence in the charge state of the peptide. The assertion therefore that the data for a-toxin is incompatible with a diÜerence in the size of the pore beween the states can be seen to be weak. The HOLE calculations provide an explanation for the diÜerence without the need to invoke surface eÜects.It is quite possible that both eÜects contribute in reality. HOLE takes such eÜects into account in an average fashion when making predictions of absolute conductance as they presumably play a role in the training set used for the derivation of the correction factors. Given a much larger set of data to work with it may be possible to explicitly include a correction function which incorporates the number of ionizable charges on the channel. However at present I have no way of representing the diÜerences of such an eÜect between two systems. You state in passing that a-toxin is expected to have a dynamic eÜective diameter. Given the fact that the channel lumen is provided by a b-barrel with strong hydrogen bonding it is likely that the channel is one of the least dynamic in terms of pore dimensions.Prof. Klein commented Our molecular dynamics studies1 of small peptide bundles suggest that the dynamical behaviour of the bundle itself may play a role in conduction. That is it could be important to take an ensemble average over the modes of vibration of the bundle. Radial breathing and torsional motion may be coupled to the passage of water molecules. This issue is not Faraday Discuss. 1998 111 225»246 236 usually discussed. In a recent paper involving a collaboration with the DeGrado group we draw attention to the possible role of dynamical —uctuations2 at least for small bundles. Do you think this eÜect could have wider implications ? 1 T. Husslein P. B. Moore Q. Zhong M. L. Klein D. M. Newns and P.C. Pattnaik Faraday Discuss. 1998 111 201. 2 G. R. Diekmann J. D. Lear Q. Zhong M. L. Klein W. F. DeGrado and K. A. Sharp Biophys. J. 1999 76 618. Dr Smart responded You make a good point. We know for instance that a caesium ion cannot –t through the gramicidin channel without a marked change in structure. However the HOLE method does not try to account for the real behaviour of channels during the conductance of an ion. Rather an empirical approach is taken in which many factors are incorporated in an average way in the –tting process. A problem behind taking an empirical approach is that it does not lead to physical insights as to the actual processes involved. But it does have other advantages. In the area of prediction of binding affinities of ligands to receptors the ììab initio œœ approach whereby you start with a model some interaction potential energy function and use chemical physics to understand the system has been shown to be of limited use.Rather Marshall and co-authors have shown that calculating relatively simple physicochemical properties for a set of complexes with known binding energies and applying a –tting procedure can result in high reliability methods for making predictions. There is not yet enough data to be able to take such a rigorous approach for channel conductance but HOLE is an attempt to start the process. To date results are good. 1 R. D. Head M. L. Smythe T. I. Oprea C. L. Waller S. M. Green and G. R. Marshall J. Am. Chem. Soc. 1996 118 3959. Prof. Roux commented I agree with Prof.Kleinœs comment about the important structural —uctuations of channels formed by bundles of helices. But I would take this further by questioning the signi–cance of the statistical —uctuations observed in MD in the absence of ions in the channel. The presence of ion in the channel may very much aÜect the structure of those —exible channels. Dr Smart responded As the questions in this discussion have revealed there are very many processes which aÜect the conductance of channels. The presence of an ion can be expected to have a marked eÜect which will vary as the ion moves through the channel. To understand and con–dently predict this eÜect may eventually be possible. But my approach is to avoid it at present. It may be that solvated dynamics runs do not provide an improvement over a purely empirical approach because of the point you make.Prof. Holzwarth asked Can you predict dynamic changes in the micro- to millisecond time range caused by the mobility of membrane lipids next to channel forming peptides like gramicidin ? We investigated the in—uence of peptides like gramicidin and an arti–cial peptide of 30 amino acids (2lys-gly-24leu-2lys-ala-amide) on the dynamics of the main phase transition of bilayer vesicles1 and found very pronounced eÜects on the mobility and structure of the lipids near the peptides ; I wonder how the mobility of the lipids might be re—ected in the transport properties of the channel forming peptides. 1 A. Genz T. Y. Tsong and J. F. Holzwarth Structure Dynamics and Equilibrium Properties of Colloidal Systems NAT OASI Series C ed.D. M. Bloor and E. Wyn-Jones Kluwer Dordrecht 1990 vol. 324 pp. 493»515. Dr Smart responded It may be possible to use molecular dynamics techniques to predict the eÜects that peptides have on membrane lipids and vice versa. However at present simulation practicalities limit the area of the lipid bilayer considered. I have no direct experience in the area. Dr Burkhart commented Crystallographic structures are a space and time average of the molecule and contain in their B-factors many of these static and dynamic features inherent in the molecular motion. 237 Faraday Discuss. 1998 111 225»246 Dr Smart responded Temperature factors provide some information as to the ensemble average dynamics of the molecule within the crystalline environment.How relevant this is to the lipid bound form of a channel is a debatable point. However the high resolution crystal or NMR structure of a channel in its active conformation provides information which is unobtainable in any other way. The recent excitement over the structure for the KcsA potassium channel disproves the view that as ion conductance is a dynamic phenomenon that ìì static œœ structures are of limited importance. Prof. Roux opened the discussion of Prof. Kleinœs and Dr Sansomœs papers Most of the helices forming bundle channels are amphipathic which could presumably be lying parallel to the membrane/bulk interface. The application of a transmembrane potential together with the presence of permeant ions could be the microscopic factor driving the formation of the bundle.This could suggest that the current MDs that you described correspond to a metastable state of the channel. Could you comment please ? Dr Sansom responded The application of a transbilayer voltage is needed to induce helix bundle formation both for alamethicin and for other peptides (e.g. the LS peptide1). However once open these channels are metastable on a ca. 10 ms timescale (i.e. 106 times the simulation timescale) even at zero mV transbilayer potential as shown by open-channel current»voltage curves (e.g. Woolley et al.2 and Kienker et al.3). The eÜect of the presence of permeant ions is more difficult to comment on»certainly there is evidence for increased channel open times at elevated ionic strengths.4 1 J.D. Lear Z. R. Wasserman and W. F. DeGrado Science 1988 240 1177. 2 G. A. Woolley P. C. Biggin A. Schultz L. Lien D. C. J. Jaikaran J. Breed K. Crowhurst and M. S. P. Sansom Biophys. J. 1997 73 770. 3 P. K. Kienker W. F. DeGrado and J. D. Lear Proc. Natl. Acad. Sci. USA 1994 91 4859. 4 W. Hanke C. Methfessell H. U. Wilmsen E. Katz G. Jung and G. Boheim Biochim. Biophys. Acta. 1983 727 108. Prof. Klein responded The metastability referred to by Prof. Roux certainly applies to our MD calculations and those of all other workers in the –eld.1h3 The present situation has been likened to trying to ascertain how the human body functions by carrying out an autopsy on a cadaver.4 Great strides were made in the early days of anatomy by studying non-functioning beings.Similarly the study of an assembled bundle might be expected to yield clues as to the key elements of the functioning channel. Naturally we look forward to the day when larger systems with transmembrane potentials and ions will be amenable to study. For the present we have a more modest aim and indeed focus on essentially pre-assembled bundles»with all of the many limitations this entails. Prof. Roux is correct in pointing out that many helices that form channels are amphipathic in character and would thus likely prefer to be lying parallel to the membrane/bulk interface as isolated monomers. We have recently investigated the behavior of such a monomer»the DuÜ» Ashley M2 peptide and indeed –nd that on the nanosecond timescale this 25-residue a-helical peptide is content to remain parallel to the interface albeit with some speci–c peptide»lipid anchoring interactions (see Fig.1). 1 T. B. Woolf and B. Roux. Proc. Nat. Acad. Sci. USA. 1994 91 11631. 2 T. B. Woolf and B. Roux. Proteins 1996 24 92. 3 M. S. Sansom et al. Biochem. Soc. T rans. 1998 26 438. 4 R. S. Eisenberg 1998 personal communication. Dr Bezrukov said From your Fig. 2B I gather that you took great care over the ionization state of Glu18 sidechains and concluded that the net charge is rather small. What would be the eÜect of the higher charge on the channel structure ? Also what potential do you use to describe interactions between charges ? Dr Sansom responded We have run simulations for N\6 alamethicin bundles with either one Glu18 ionised1 or with zero or six Glu18s ionised.2With six Glu18s ionised the bundle ìì falls Faraday Discuss.1998 111 225»246 238 Fig. 1 The DuÜ»Ashley 25 residue peptide at the lipid/bulk water interface taken from an MD simulation initiated with the peptide lying parallel to the interface. a apartœœ during the simulation. With zero or one Glu18 ionised the bundle remains intact. This suggests that suppression of ionisation of sidechains may play a role in channel stability. In studying the interactions between charges in the pK calculations we used a screened interaction (by including a Debye length equivalent to 1 M KCl). But in the MD simulations a simple unscreened Coulombic interaction was used. Clearly this is an approximation and may be a problem as it is known that e.g.melittin channels are stabilised by high ionic strength. 1 D. P. Tieleman J. Breed H. J. C. Berendsen and M. S. P. Sansom Faraday Discuss. 1998 111 209. 2 D. P. Tieleman H. J. C. Berenden and M. S. P. Sansom Biophys. J. 1999 76 in the press. Prof. Neumann commented. There is no doubt about the importance of charged groups for the structure and for structural changes of macromolecules. Yet for channel proteins the actual transport passage is usually hydrophobic without charged groups at some distance away from the channel part. See for instance Kukol and Neumann.1 1 A. Kukol and E. Neumann Eur. Biophys. J. 1998 27 618. Dr Bezrukov responded Sometimes this is the other way around. See for example Forst et al.1 1 D.Forst W. Welte T. Wacker and K. Diederichs Nat. Struct. Biol. 1998 5 37. Prof. Klein responded Current generation computers allow only ca. 5»10 nanosecond length trajectories in most cases. However there is one reported study by the Kollman group1 spanning 239 Faraday Discuss. 1998 111 225»246 1 ls. The next 3»5 years should see supercomputers reaching peak performance around 100 Tera- —ops which should allow microsecond trajectories for membrane proteins. This will allow us in special cases to follow modest structural changes caused for example by the passage of ions along channels. For less detailed models the projected increase in CPU performance will enable coarsegrained models to study the assembly of model peptides into bundles and the response of the host membrane to external probes.2 1 Y.Duan and P. A. Kollman Science 1998 282 740. 2 R. Lipowsky Progr. Colloid Polym. Sci. 1998 111 34. Dr Sansom said I wish to comment on MD simulations of the tetrameric transmembrane (TM) helix bundle of the M2 channel protein from in—uenza A. We also have run such simulations,1 albeit using a diÜerent lipid (POPC instead of diPhyPC) from that used by Klein and coworkers. 2 However in contrast to the situation with model peptide channels (e.g. alamethicin) there is a problem with simulations of TM helix bundles from larger proteins namely that of the exact extent of the helices. In the case of in—uenza M2 DuÜ and co-workers3,4 have shown that a 25-residue peptide forms channels and is largely a-helical.However it is not certain that all of the residues in the peptide form an a-helix and to what extent the peptide mimics the intact M2 protein. We have used multi-nanosecond simulations of single TM helices of diÜerent lengths (from 18 26 or 34 residues) in a phospholipid bilayer in order to determine the optimum length of the helix from M2. These simulations suggest that a region of length ca. 22 residues forms a helix stable throughout the simulation.5 The length of helix has a profound eÜect on the behavior of four TM helix bundle models in bilayer simulations. We have compared simulations with a bundle of 18-residue helices and with a bundle of 22-residue helices.5 In both cases the helix bundles retained their left-handed supercoil structure and gave a Ca rmsd of ca.0.25 nm after a 4 ns simulation. However the 18-residue bundle contained only 3 water molecules which did not exchange with bulk water. Thus the 18-residue bundle looked like a ììclosedœœ channel. In contrast the 22-residue bundle contained ca. 10 waters and looked like an ììopenœœ channel. What is clear from these simulations is that one has to be a bit cautious in the choice of initial helix bundle model as this may have a profound in—uence on the results of any subsequent bilayer simulation. 1 M. S. P. Sansom D. P. T. Tieleman L. R. Forrest and H. J. C. Berendsen Biochem. Soc. T rans. 1998 26 438. 2 T. Husslein P. B. Moore Q. Zhong D. M. Newns P. C. Pattnaik and M. L. Klein Faraday Discuss. 1998 111 201. 3 K. C. DuÜ and R. H. Ashley V irology 1992 190 485.4 K. C. DuÜ S. M. Kelly N. C. Price and J. P. Bradshaw FEBS L ett. 1992 311 256. 5 L. R. Forrest D. P. Tieleman and M. S. P. Sansom Biophys. J. 1999 76 in the press. 6 L. R. Forrest and M. S. P. Sansom in preparation. Prof. Okazaki said It is interesting for me to see a decrease of membrane thickness and a decrease of alkyl chain order by the inclusion of the peptide. If you have reached some conclusion please let me know about the mechanism. Are there any particular sites in the peptide which interact directly with the lipid molecule? Prof. Klein responded I am con–dent that our results are correct for the peptide concentration used in the MD simulation. Unfortunately the ratio of peptide to lipid is only 1 8 in each lea—et of the bilayer.I am therefore concerned that the observed ììthinningœœ of the membrane is related to channel»channel repulsions arising from interactions between the large helix dipoles. It would be useful to run more dilute samples with say 128 and 256 lipids to quantify this eÜect more precisely. Prof. Klein said Did you –nd that the M2 (22-residue) peptide yielded a stable 4-helix bundle with a water pore or was it blocked at the His as suggested by various experiments and MD calculations1 on a simpler system? 1 Q. Zhong T. Husslein D. Newns and M. L. Klein FEBS L ett. 1998 434 265. Faraday Discuss. 1998 111 225»246 240 Fig. 2 A con–guration taken from a multi-nanosecond MD simulation of the lipid»DNA complex. The two distinct lipids (DMTAP and DMPC) are drawn with light and dark shading.Dr Sansom responded The M2 (22-residue) helix bundle was ìì stable œœ in the sense that the left-handed supercoil was maintained and the Ca rmsd at the end of 4 ns was ca. 0.25. Up to ca. 2 ns there was not a continuous water pore. Instead there was a water-–lled pore which was open to the surrounding environment at the N-terminal mouth but which was occluded towards the C-terminal mouth by the ring of His37 sidechains. However after ca. 2 ns this pore opened up at the C-terminal as well. We suspect that this sort of behaviour might be rather sensitive (at least on a 1 to 10 ns timescale) to the starting model used in the simulation and so we are now exploring diÜerent starting structures. Dr Amblard commented Modelling the dynamics of macromolecules in solution or in membranes by coupling the elasticity of the macromolecule and viscous dissipation by the small surrounding molecules taken as a continuum could be a way of simulating molecular dynamics at much smaller frequency than the limit of MD classical simulation.On the other hand Prof. Smith suggested earlier by MD simulation of bacteriorhodopsin that this membrane protein could undergo dynamical transitions at well de–ned temperatures revealed by the non-linearity of the crystal temperature factor with temperature. This suggests that bacteriorhodopsin does not behave as a set of harmonic oscillators but in a much more complex and likely non-harmonic way. To what extent could the complex elastic behavior of proteins be inferred from temperaturefactor analysis at least to provide a basis for modelling the dynamics at longer time-scales than with MD simulations ? Dr Sansom responded Iœm really not sure about using temperature-factor analysis for membrane proteins.My reservations are (i) for a number of membrane protein structures the resolution is a bit low and so the B-factors may not be that reliable ; (ii) most membrane proteins 241 Faraday Discuss. 1998 111 225»246 are crystallised without lipids and so the relevance of the motions in the crystal to motions in the bilayer is uncertain. Prof. Bayerl asked Does the MD simulation of DNA in DMPC/DMTAP give any indication of a spatial con–nement of the cationic lipids motion due to the electrostatic interaction with the DNA? Prof.Klein responded Our MD simulations1 have been run long enough to begin to quantify the nature of the lipid»DNA interactions. The zwitterionic headgroup of DMPC competes eÜectively with the cationic lipid DMTAP in ììneutralizingœœ the DNA phosphates. Fig. 2 and 3 show the overall structure of the complex and the distribution of nitrogen atoms around the DNA phosphates.1 1 S. Bandyopadhyay M. Tarek and M. L. Klein. in preparation. Prof. Holzwarth commented The molecular dynamics simulations provide very interesting information about structural changes on an atomic level for times from femto- to several nanoseconds but are not able to reach longer times. If one inspects the energy changes connected with membrane processes it can be shown that most of the important changes in biological systems are occurring at much longer times.1 (1) Is there any chance in the near future to reach 100 nanoseconds or better the microsecond time range with MD simulations ? (2) How do you rate the chances for approaches which avoid the enormous computer power needed for full nanosecond simulations by starting with some reasonable assumptions Could this be an acceptable approach and how can Monte-Carlo simulations be included into the solution ? My personal ideas are circulating in a triangle of dynamic structural and thermodynamic information trying to connect all three types of available results to construct a simpler basis for MD simulations.1 J. F. Holzwarth in T he Enzyme Catalysis Process ed. A. Cooper J.L. Houben and L. C. Chen Plenum London 1989 pp. 383»412. Dr Sansom responded There is every chance in the near future of reaching 100 ns simulations given the increasing performance of computers and the development of ììsmarterœœ MD algorithms. Indeed we have run one alamethicin helix bundle simulation for nearly 20 ns (ref. 1). In the non-membrane –eld simulations of the folding of a small protein fragment (ca. 40 amino acids) have been run for ca. 1 ls (ref. 2) although this is still out of the feasible range for membrane simulations. Fig. 3 Distribution of water molecules (green) and lipid N-atoms (red and blue) around a representative DNA phosphate group. Faraday Discuss. 1998 111 225»246 242 1 M. S. P. Sansom and D. P. Tieleman unpublished data.2 Y. Duan and P. A. Kollman Science 1998 282 740. Prof. Klein responded. As mentioned earlier the Kollman group has reported on two microsecond long trajectories for a protein in water.1 This was possible because of the availability of 256 CRAY T3E processors dedicated for a few months. The relentless progress in CPU performance should allow similar capability on a more routine basis in the 3»5 year time frame. The likely availability of these resources does not obviate the need to develop alternative approaches. 1 Y. Duan and P. A. Kollman Science 1998 282 740. Prof. Roux said It is attractive to use a simpli–ed description of the membrane environment in order to gain in computational efficiency and reach out to longer simulation times. However it is important to keep in mind that one still doesnœt know which details of the bilayer are going to be important in investigating the function of a membrane protein.Dr Sansom responded I fully agree with this. I think progress will be made by running more explicit bilayer simulations on a range of diÜerent membrane proteins and then attempting a general analysis of which interactions are most important at the same time as developing simpli- –ed descriptions as exempli–ed in the paper by Roux and co-workers.1 1 P. Lagué e M. J. Zuckermann and B. Roux Faraday Discuss. 1998 111 165. Dr Amblard commented When reading Dr Sansomœs paper I was quite surprised by the fact that results of MD simulation are more often compared with results of other simulations than with experimental results.This is obvious in the ìBiological relevanceœ section of his paper. In the discussion of the last few papers dealing with MD simulation not much was said about the connection with experimental data about the kind of predictions generated by MD simulations at diÜerent time-scales and about their experimental ìì testability œœ and the experimental models of interest. Could you and Prof. Klein comment on these points and help non-specialists like me to grasp what the important questions and limitations are in this –eld beyond the technicalities of MD simulations. Dr Sansom responded Maybe I was a bit too cautious in the ìBiological relevanceœ section of my paper but I deliberately did not wish to over-interpret our results. I think some of the connections with experimental data for alamethicin have been described in Dr Smartœs paper,1 and so I wonœt duplicate them here.As I suggested in my paper I think the motion of the pore-lining helices of alamethicin may have some relevance to gating of e.g. potassium channels such as KcsA. Indeed recent spin-label studies of KcsA2 provide experimental evidence of such helix motions albeit on a much longer timescale. 1 O. S. Smart G. M. P. Coates M. S. P. Sansom G. M. Alder and C. L. Bashford Faraday Discuss. 1998 111 185. 2 E. Perozo D. M. Cortes L. G. Cuello Nat. Struct. Biol. 1998 5 459. Prof. Klein responded You raise an excellent point. It is to be regretted if the results of a MD simulation are inaccessible to experimentalists. The predictive capability of the MD simulations is only useful to the extent that a range of experimental data can be accounted for.Alas typically for membrane proteins little is known beyond the basic structure. Even then data is mostly related to the crystalline environments. The situation may change as modern NMR methods achieve increasing success. Also neutron and X-ray synchrotron experiments are likely to be important complements to modern NMR studies. I agree that it would be unfortunate if the focus shifted solely to technicalities of MD simulations. Prof. Smith commented Prof. Klein provides evidence here for a more —uid-like behaviour of the lipid in the presence of the protein and larger area per lipid relative to the pure lipid. In preliminary results with Dan Mihailescu on gramicidin S binding to a DMPC bilayer we see the 243 Faraday Discuss.1998 111 225»246 opposite eÜect in the lipid —uidity. Would you like to comment on whether –ndings similar to yours1 have been observed elsewhere and on the physical origin of these eÜects ? 1 T. Husslein P. B. Moore Q. Zhong D. M. Newns P. C. Pattnaik and M. L. Klein Faraday Discuss. 1998 III 201. Prof. Klein responded I am sorry to say that I cannot comment on the gramicidin/DMPC system. Prof. Okazaki has been interested in this system for some time. Dr Sansom responded Tieleman and co-workers1 have analysed lipid properties simulations of six-helix bundles of alamethicin. He saw a decrease in order parameters corresponding to increased tilt of acyl chains of lipids close to the helix bundle.I suspect diÜerent results may be found for diÜerent systems (see response to next comment) and so we need to be cautious in making generalisations at this stage. 1 D. P. Tieleman L. R. Forrest M. S. P. Sansom and H. J. C. Berendsen Biochem. 1998 37 17554. Prof. Roux commented The presence of protein in a membrane has been shown to increase the ordering of the acyl chains of the lipids. Rice and Old–eld showed that the deuterium quadrupolar splitting of speci–cally labelled DMPC was increased in the presence of gramicidin. Our results from molecular dynamics simulations are in good agreement with this observation.2 However it should be stressed that averaging from –ve independent trajectories was required to get convergence.1 Rice and Old–ed Biochemistry 1979 18 3272. 2 Woolf and B. Roux Prot. Struct. Funct. Gen. 1996 24 92. Prof. Klein responded Prof. Roux brings out an important point The issue of convergence of NMR order parameters obtained from MD simulations. It is also our observation that these are difficult quantities to obtain reliably. Unfortunately my group has no speci–c data for the gramicidin/DMPC system. Dr Sansom responded Also I think it is important to note that diÜerent results may be obtained for diÜerent proteins. Tieleman and co-workers have analysed lipid properties in simulations of several diÜerent systems. For single transmembrane helices the eÜects on surrounding lipids are relatively minor. For four helix (in—uenza M2) and six-helix (alamethicin) bundles and for the porin OmpF (a large b barrel) a decrease in order parameters corresponding to increased tilt of acyl chains of nearby lipids was seen.This suggests that we need to gather data from a wider range of simulations before we can attempt any generalisations. However I agree that to obtain reliable statistics either long simulations or multiple independent trajectories are needed. 1 D. P. Tieleman L. R. Forrest M. S. P. Sansom and H. J. C. Berendsen Biochem. 1998 37 17554. Prof. Okazaki commented First the calculated order parameter of gramicidin was absolutely dependent upon the initial con–guration. No conformational changes were observed for the gramicidin molecule within our preliminary ns order MD calculation. On the other hand the calculated order parameter of the DPPC alkyl chain in the pure DPPC bilayer in the liquid crystal phase was larger than the experimental one.But the cumulative average of the order parameter did not converge within our several ns order simulation. The convergence is very slow. The order parameter is thus very difficult to evaluate from the limited simulation time of the MD calculation. Dr Sansom responded I agree that longer simulations are needed to get proper estimates of order parameters. Also one suspects careful attention to the setup of the simulation (i.e. lipid» protein packing) is needed. Dr Bezrukov commented I want to highlight the importance of molecular dynamics simulations in ion channel studies. Unfortunately analytical methods for description of channel transport Faraday Discuss.1998 111 225»246 244 properties are far from satisfactory. For example even a ìsimpleœ question of how ionization of a single sidechain facing channel lumen in—uences channel conductance and selectivity is very hard to answer. Even in the case where all structural data is available (nothing to guess about here) I would not be surprised if the sign of the eÜect is difficult or just impossible to predict. Not only electrostatic vs. structural issues are involved here ; the electrostatics itself at such short distances and high –elds diÜers considerably from the continuous classic formulation. For this reason molecular dynamics simulations both of channel structure and its transport properties are much needed at present.Dr Sansom responded I completely agree that even ììsimpleœœ questions are not resolved by simply looking at an X-ray structure. I feel one should be a bit more guarded about how quickly MD studies will progress. In principle one should run a simulation which allows for dynamic protonation/deprotonation as an ion passes. I confess to not knowing how to do this at the moment but Iœm sure it will need long simulation times. Given the diÜerence in timescale between current MD simulations (ca. 10 ns) and the permeation time of an ion (ca. 1 ls) I think we need to develop a hierarchy of theoretical descriptions of a channel in order to relate atomic resolution structures to physiological data from patch clamp experiments. Prof. Holzwarth replied Unfortunately the experimental techniques available are not speci–c enough to tackle the questions posed by Dr Bezrukov.On the other hand molecular dynamic simulations are not able to cover the time range beyond 10 ns. In summary I believe that the present experimental techniques are good in respect of the important time range from nanosecond to second but often lack molecular speci–city ; molecular simulations are –ne for times shorter than a nanosecond but are not able to cover the really important ììlong timeœœ phenomena. The future should improve this situation. Dr Deber communicated In the simulation of the four-helix bundles of M2 what were the features/motifs of interaction of the inter-helical faces/residues within the bundle (as depicted in Fig. 1 of your paper) ? Are the interfaces predictable from primary sequence? Prof.Klein communicated in response To some extent this issue has been discussed by my colleague Prof. DeGrado and his collaborators,1 in their analysis of mutagenesis data. Our MD study was designed to complement this experimental work and their analysis of the bundle structure. Although I am reluctant to speak for my colleagues on this issue it is my impression that the inter-helical faces/residues do correspond to predictions based on the primary sequence of residues. 1 L. H. Pinto et al. 1997 Natl. Acad. Sci. USA 1997 94 1130. Prof. Deber communicated Is it possible to determine how helix»helix interactions vary (energetically occurrences of residues at interfaces) as a function of alamethicin helical bundle size ? A systematic approach here could enhance our understanding of the folding of polytopic membrane proteins.Dr Sansom communicated in response A very perceptive question ! We are working on this but havenœt done such analysis as yet. One of the attractive features of these sorts of MD simulations is that they should allow analysis of helix»helix interactions in various ììsimpleœœ bundles of transmembrane helices. Dr Bhakoo communicated There are a vast array of isolated biomembrane systems and their interactions are being studied by numerous methods by a number of excellent research teams. One particular worry I had is the lack of correlation of data between groups and between techniques utilised. Perhaps this could be addressed by having closer collaboration between research groups.Another key issue that requires addressing is relating the simple isolated membrane systems to real biological membranes. A number of times the detailed understanding of model systems will not lead to the understanding of biological membranes. I believe that there is an urgent need for a 245 Faraday Discuss. 1998 111 225»246 meeting between physical chemists chemical physicists biophysicists biochemists biologists physiologists and microbiologists to discuss this issue. Thus a related question I have is as follows How far do model membrane systems aid in providing understanding of real biological systems they are apparently mimicking? Dr Sansom communicated in response I guess this meeting is intended to address such worries.However I think from a simulation perspective it is important that diÜerent groups work on similar systems so that one can get a feel for the extent to which the results converge. For example going back to in—uenza M2 we have shown recently1 that two independent modelling studies converge on a similar structure which in turn is in agreement with experimental data.2 This is encouraging. In answer to your second question clearly simple systems do not hold all of the answers. But considering ion channels gramicidin provided what turned out to be a good model for channel» ion interactions in the bacterial potassium channel. 1 L. R. Forrest W. F. DeGrado G. R. Dieckmann and M. S. P. Sansom Folding Design 1998 3 443. 2 F. A. Kovacs and T. A. Cross Biophys. J. 1997 73 2511. Prof. Roux also communicated in reply to Dr Bhakoo An important aspect of biophysics (experimental and theoretical) is the ability to use a reductionist approach in order to highlight the most important factors responsible for a phenomena. In doing so it is useful to throw away as much detail as possible as long as the essential elements are preserved. While I do not doubt that the complexity of biological membranes is much beyond our simple models I think there is much to be gained by investigating simple model systems. Nonetheless the intrinsic limitations of simple models should always be kept in mind. In answer to your second question characterizing the magnitude of lipid-mediated protein» protein interactions is only a modest though an important contribution to our understanding of the function of biological membranes. Faraday Discuss. 1998 111 225»246 246

ISSN:1359-6640    

DOI:10.1039/a901091b

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Kinetics of the competitive response of receptors immobilised to ion-channels which have been incorporated into a tethered k3D ~12 for digoxin to completely reverse the bilayer Gillian E. Woodhouse Lionel G. King Lech Wieczorek and Bruce A. Cornell Cooperative Research Centre for Molecular Engineering and T echnology Australian Membrane and Biotechnology Research Institute 126 Greville Street Chatswood NSW 2067 Australia Received 8th December 1998 A competitive ion channel switch (ICS) biosensor has been modelled yielding ligand mediated monomer»dimer reaction kinetics of gramicidin (gA) ion-channels within a tethered bilayer lipid membrane. Through employing gramicidin A functionalised with the water-soluble hapten digoxigenin it is possible to cross-link gramicidin to antibody fragments tethered at the membrane/aqueous interface.The change in ionic conductivity of the channel dimers may then be used to measure the binding kinetics of hapten»protein interactions at the membrane surface. The approach involves measuring the time dependence of the increase in impedance following the addition of a biotinylated antibody fragment (b-Fab@) which cross-links the functionalised gramicidin monomers in the outer layer of the lipid bilayer to tethered membrane spanning lipid. The subsequent addition of the small molecule digoxin (M 781 Da) competes with and reverses this interaction. r The model provides a quantitative description of the response to both the cross-linking following the addition of the b-Fab@ and the competitive displacement of the hapten by a water-soluble small analyte.Good agreement is obtained with independent measures of the cross-linking reaction rates of the gramicidin monomer»dimer and the b-Fab hapten complex. The rate and amplitude of the competitive response is dependent on concentration and provides a fast and sensitive detection technique. Estimates are made of the concentration of gramicidin monomers in both the inner and outer monolayer lea—ets of the membrane. This is used in the calculation of the gramicidin K monomer/dimer equilibrium constant 2D3 . Other considerations include the membrane impedance limit set by the membrane leakage which is also a function of the concentration of the gA monomer concentration and the two-dimensional kinetic association constant k2D2 of the hapten b-Fab@ complex.The gA dimer concentration is dependent on both the concentration of gA-dig and of the tethered streptavidin b-Fab@ complexes. k2D3 ~1 must be at least 10 times The model shows that the 2D dissociation constant faster than the 3D dissociation constant cross-linked hapten»receptor interaction at the membrane interface. Introduction Tethered bilayer lipid membranes (t-BLMs) oÜer improved robustness and simpler assembly pro- 247 Faraday Discuss. 1998 111 247»258 Fig. 1 (a) Chemical components for the competitive ICS biosensor. The immobilised components of the membrane spanning phytanyl lipids (DLP) full membrane spanning lipids (MSL) and biotinylated MSL (bmembrane comprise a mixture of tethered gramicidin A gAYYSSBn (gAT) double length reservoir half MSL).The surface density of these tethered species is controlled by dilution with the small hydrophilic mercaptoacetic acid disul–de (MAAD). The outer mobile membrane lea—et consists of a 7 3 mole ratio mix of cedures compared with solvent formed bilayer lipid membranes (BLMs) patch clamping techniques Langmuir»Blodgett layers or supported bilayer lipid membranes (s-BLMs). They have many potential applications related to bioelectronics biosensing pharmaceutical screening and physiological transduction studies. Tethered bilayers provide a novel tool for studying the molecular dynamics of channels incorporated into the membrane and of the binding kinetics of proteins Faraday Discuss. 1998 111 247»258 248 diphytanyl phosphatidylcholine (DPEPC) glycerodiphytanyl (DPG) which is mixed with a fraction of 1 10 000 mole ratio of digoxingenin derivatised gramicidin A (gA-dig).(b) Schematic of competitive ICS biosensor and its simple RC equivalent circuit. The equivalent circuit for the sensor may be approximated as an eÜective Helmholtz capacitance (Ch) in series with the capacitance of the membrane (Cm) and which is bypassed by the ionophore conduction (Gm). at the membrane surface. The self-assembly mechanism of the tethered bilayer results in the formation of an ionic reservoir between the gold interface and the membrane. In the presence of conductive channels within the membrane ions —ow between the reservoir and the external solution driven by an externally applied potential.Raguse et al.1 have reported the use of valinomycin to characterise the ionic conductivity and capacity of the reservoir. As described by Cornell et al.,2 the competitive biosensor consists of a tethered bilayer incorporating a reservoir mobile and tethered ion channels and a family of molecular tethers. Fig. 1(a) shows the inner layer species of the membrane which comprises –ve components. Four of these components are amphiphilic having hydrophobic membrane forming end groups which attach to the gold surface via a disul–de foot and an intervening hydrophilic portion constructed from two repeated units of tetraethyleneglycol succinic ester (TEGsuc). These are tethered gramicidin A 1 (gAT) membrane spanning lipid 2 (MSL) biotinylated membrane spanning lipid 3 (b-MSL) and a half membrane spanning lipid 4 (DLP).The –fth component is the small hydrophilic disul–de molecule dithiodiglycolic acid 5 which acts as a spacer on the surface. SA b-Fab@). The outer lipid layer is composed of predominantly untethered lipids capable of translational diÜusion within the 2D plane of the membrane. The outer layer lipids are 3-di-O-phytanyl-snglycerol 6 (DPG) and 2,3-di-O-phytanyl-sn-phosphatidylcholine 7 (DPEPC). The b-MSL 3 deposited with the –rst layer spans the lipid bilayer and its terminal biotin group extends into the aqueous environment and in the presence of streptavidin forms a complex on the membrane outer surface MSLSA . Subsequent addition of biotinylated anti-digoxin antibody Fab@ fragments (b- Fab@) to MSLSA aÜords the ternary b-Fab@ streptavidin MSL complex (MSL The outer layer membrane also includes digoxigenin derivatised gramicidin A (gA-dig) 8 which can be cross-linked to MSLSA b-Fab@ to form a further quaternary complex (MSLSA b- Fab@ gA-dig).The gA-dig molecules in this quaternary complex are constrained to be out of register with the inner membrane gramicidins (gAT) and thereby prevented from forming the ion-conducting gramicidin dimers (gADIMER). Addition of digoxin (M 781 Da) reverses this r binding releasing the gA-dig and allowing the channels to conduct. Experimental Materials The structure of the membrane components have been described elsewhere DLP,2 MAAD,2 DPG,2 DPEPC,2 MSL,3 b-MSL,3 gA-dig,4 and gAT .4 All chemicals were puri–ed by HPLC and their purity was shown to be better than 99% by MALDI mass spectroscopy TLC and 1H NMR.Analytical grade ethanol (Ajax) was used in preparing all the lipidic solutions. The components were stored for up to 6 months in solution at 4 °C. Digoxin (Sigma Chemical Co. St. Louis USA) 249 Faraday Discuss. 1998 111 247»258 was dissolved in ethanol to 10 mg mL~1 and dilutions were made from this stock with phosphate buÜered saline (50 mM PO4~ pH 7.2) (PBS). Streptavidin solutions were prepared from a lyophilised powder (Boehringer Mannheim Germany) and dissolved in PBS to prepare stock solution concentrations of 1 mg mL~1. Anti-digoxin monoclonal antibody was obtained from Biodesign Ltd. (Maine USA). (Fab@)2s were prepared by enzymic fragmentation which were further reduced to Fab@s.Free cysteine exposed at the hinge region was biotinylated using iodoacetyl-LC-biotin (Pierce Rockford USA) to give a mono-biotinylated Fab@ (b-Fab@). Purity was con–rmed to be better than 98% (SDS-PAGE). The procedure has been described previously.5 Biosensor preparation Clean room facilities were used for the preparation of the gold –lms and the deposition of the initial monolayer formation. All procedures were carried out at 21»22 °C unless otherwise speci- –ed. Glass microscope slides (Objekttrager HD Scienti–c Germany) were cleaned by soaking with detergent (Extran 300 BDH) for 2 h and rinsing thoroughly with milli-Q water. They were subsequently dried under a stream of dry nitrogen gas.A chromium adhesion –lm of 20 nm in thickness was evaporated at 0.1 nm s~1 onto each microscope slide followed by a 100 nm –lm of gold at 2.5 nm s~1 using an Edwards High Vacuum Coating Unit (18SE4/S53) operating at 5]10~7 Torr. Monolayers were chemically adsorbed onto the freshly prepared gold –lms by incubating the slides for 1 h in an ethanolic solution of 300 lM DLP 150 lM MAAD 0.043 lM gAT 4.5 lM MSL 0.225 lM b-MSL. The slides were then rinsed thoroughly with ethanol and stored under ethanol at 4 °C until required. Monolayer coated glass slides were removed from the storage solution and dried under a high pressure stream of dry boil-oÜ nitrogen gas. Slides were placed on a brass support (approximately 10 cm]5 cm]7 mm) and a second brass piece containing eight holes spaced on a standard 96-well ELISA plate grid pattern (internal diameter 6 mm) into which machined Te—on tubes (3.6 mm internal diameter) had been inserted was clamped onto the microscope slide and supporting brass block to form sealed discrete cells.The area of electrode exposed within the well had an internal area of 0.10 cm2. A 10 lL volume of a 10 mM solution of DPG DPEPC (3 7) containing 0.1 mM gA-dig was dispensed into each cell sequentially immediately followed by 100 lL phosphate buÜered saline (pH 7.2 50 mM). Excess lipid was washed out of the cell through exchanging the buÜer volume at least 4 times ensuring that the membrane interface was not exposed to air. A 5 lL volume of a streptavidin solution at 0.05 mg mL~1 was injected into the cell to give a –nal cell concentration of 40 nM and incubated on the membrane for 10 min.The excess material was removed by copiously washing with PBS. A 5 lL volume of a b-Fab@ solution at 1 lM (PBS) was added to the cell incubated for 10 min and rinsed by exchanging the volume with PBS. Impedance spectra were obtained using a custom built 32-channel impedance spectrometer (Associative Measurements Pty. Ltd. Sydney Australia) employing data analysis software commissioned for our laboratory (Aguilla Holdings Sydney Australia). Frequencies were scanned between 1 Hz and 1 kHz over a period of 16 s. A bias of [300 mV was applied between the sensor electrode and a Pt counter-electrode while an ac ion excitation signal of 50 mV was applied between the measurement electrode and a Ag reference electrode.Up to 32 independent impedance trials could be conducted simultaneously. After collecting a baseline of at least 10 spectra test analyte was introduced to the cell by adding 100 lL into the 100 lL volume of PBS existing in the well and mixing thoroughly. Modelling of competitive response This paper describes the modelling of the kinetics of two reactions the addition of b-Fab to a membrane containing mobile gA-dig and tethered MSLSA to form the complex MSLSA b- Fab@ gA-dig [eqn. (1)»(5) and (7)] and the disruption of this complex by solution digoxin to form MSLSA b-Fab@ dig]gA-dig [eqn. (6)]. Assuming adequate mass transport of analyte to the membrane the behaviour of the system may be modelled by the following seven equilibrium reactions Faraday Discuss.1998 111 247»258 250 (1) gAT]gA-dig kA8B 2D1 gADIMER where K2D1\k2D1/k2D1 ~1 k~1 2D1 (2) gA-dig b-Fab@]gAT kA8B 2D2 b-Fab@ gADIMER where K2D2\k2D2/k2D2 ~1 k~1 2d2 (3) MSLSA b-Fab@]gA-dig kA8B 2D3 MSLSA b-Fab@ gA-dig where K2D3\k2D3/k2D3 ~1 k~1 2D3 (4) b-Fab@]MSLSA kA8B 3D1 MSLSA b-Fab@ where K3D1\k3D1/k3D1 ~1 k~1 3D1 (5) gA-dig]b-Fab@ kA8B 3D2 gA-dig b-Fab where K3D2\k3D2/k3D2 ~1 k~1 3D2 (6) MSLSA b-Fab@]dig kA8B 3D3 MSLSA b-Fab@ dig where K3D3\k3D3/k3D3 ~1 k~1 3D3 (7) gADIMER]b-Fab@ kA8B 3D4 gADIMER b-Fab where K3D4\k3D4/k3D4 ~1 k~1 3D4 2D1 K2D2 K2D3 The constants k2D1 k2D2 k2D3 k3D1 k3D2 k3D3 k3D4 k2D ~11 k~1 2D2 k~1 2D3 k~1 3D1 k~1 3D2 k~1 3D3 k~1 3D4 are the two- and three-dimensional association and dissociation reaction rates and K K3D1 K3D2 K3D3 K3D4 are the equilibrium constants for reactions (1)»(7).The model assumes the binding of streptavidin to b-MSL to be irreversible. Eqn. (3) describes the two dimensional reaction in which the complex MSLSA b-Fab@ crosslinks to mobile gA-dig via two dimensional diÜusion to form the ternary complex MSLSA b- Fab@ gA-dig. In the presence of solution digoxin mobile gA must compete for binding to the complex MSLSA b-Fab@ with the analyte in solution [eqn. (6)]. The competition between reactions (3) and (6) will determine the reaction kinetics and equilibrium of reaction (1).The measurable result of this series of interactions is the concentration of gramicidin dimers which is proportional to the admittance at minimum phase (Y’ min). (8) Y’ minP[gADIMER] Electrical properties of the competitive ICS biosensor The equivalent electrical circuit of the tethered membrane is given in Fig. 1(b) together with a schematic diagram of the competitive ICS biosensor. Through modelling the spectra (C (Fig. 2) values for membrane capacitance C m) (h) have been derived and Helmholtz capacitance and are in agreement with values reported by others.1,6,7 The modulus of the admittance Fig. 2 (a) Impedance spectra and (b) phase vs. frequency plot for competitive ICS biosensor. Impedance and phase vs. frequency data of a tethered membrane with MSL (L).This shows the and gA-dig is shown SA impedance and phase characteristics of the same membrane after 50 nM digoxin b-Fab@ has been incubated for 10 min (|) and after excess b-Fab@ has been removed and 0.38 ng digoxin has been added (K). Faraday Discuss. 1998 111 247»258 251 Table 1 Capacitance and conduction values of t-BLM before and after b-Fab@ gA-dig cross-linking After b-Fab@ addition Before b-Fab@ addition 0.50 lF cm~2 0.52 lF cm~2 Membrane capacitance (Cm) 7.6 lF cm~2 7.8 lF cm~2 420 lS cm~2 850 lS cm~2 Helmholtz capacitance (Ch) Membrane conduction (Gm) lF cm~2 depending on electrolyte where C is is the capacitance for GC a Compared with CGC[20 lF cm~2 and CSAMB2 the classical interfacial capacitance described by Guoy and Chapman8 and C a self-assembled monolayers described by Ulman.9 T] ( for [gA-dig]\7]109 K) 2.8]109 (L) 1.4]109 is –xed and [gA-dig] is systematically increased (c) the concentration of T K2D1 .The actual surface gA K2D1 values are derived from Fig. 3 (a) Titration of gA inner and gA outer in the bilayer. The data shows a titration of admittance at phase Y minimum Y (lS) as a function of the nominal concentration of gA-dig and gAT . The graph shows ’ min ’ min 7]108 as a function of [gA 0 (]) molecules (|) ()) cm~2. When [gA-dig]\0 and the admittance of gA is increased in the bottom layer very little change in conduction occurs demonstrating that gA is not transferring into the outer bilayer lea—et to form conducting T T gA dimers.The data suggest that the admittance given for [gA-dig]\0 is due to conduction of monmeric gA. (b) and (c) When gA-dig is at a –xed concentration and gA is systematically increased in the inner T gAT layer (b) and conversely where [gA the bilayer increases to a point until a concentration which is equivalent to T] concentrations are not known. The nominal 2D1 \1/[gA] for eqn. (1) 2D1(gAvdig)\1]109 ’ minP[gADIMER] Y 2D1(gAT)\6]109 K assuming that K giving an apparent molecules cm2. The ratio of K2D1(gAvdig)/K2D1(gAT) allows the ratio of actual [gA] to be determined. Faraday Discuss. 1998 111 247»258 252 Literature After digoxin addition 0.52 lF cm~2 0.68 lF cm~2 (ref. 6) 0.72 lF cm~2 (ref. 7) 0.52 lF cm~2 a 5 lF cm~2 1 7.8 lF cm~2 670 lS cm~2 SAM K and apparent o Y o\uChCm/(Ch]Cm) is obtained over a range of frequencies (l) typically from 1 kHz to 1 Hz.A measure of the Helmholtz capacitance may be obtained from the modulus of the admittance o Y o at 1 Hz and the membrane capacitance from the modulus of the admittance o Y o at 1 kHz. An approximate measure of the channel conductance may be obtained from the admittance o Y o taken where the slope dY /du is minimum. Given an electrode area of 0.10 cm2 C and C were h m calculated and are shown in Table 1. Upon the addition of the biotinylated anti-digoxin Fab@ there is little change in either the membrane or Helmholtz capacitance by contrast the conduction G decreases by a factor of 2. m Similarly upon subsequent addition of digoxin the capacitance remains virtually unchanged whereas the conductance increases.These electrical changes correspond to those expected for a bilayer containing the reversible ionic switches as depicted schematically in Fig. 1(b). of 6]109 molecules cm~2 respectively. gA dimerisation in terms of the apparent concentration of gA provided [gA] in the titrated layer is when [gA] at the saturating admittance\(K2D1 )~1. Fig. 3(b) and (c) A titration of the membrane admittance as a function of the [gA-dig] and [gAT] Fig. 3(b) and (c) has been used to estimate the apparent monomer»dimer gA equilibrium constant K2D1 . The nominal concentration of gA molecules incorporated into each monolayer was initially assumed to be proportional to the ratio of the concentrations of gramicidin to those of the membrane forming lipids in the deposition solution.The total lipid surface concentration was taken as approximately 1.4]1014 cm~2.10 Monomers of gA are seen to cause leakage due to conduction T through the outer sealed lipid monolayer [Fig. 3(a)]. This leakage is approximately 5»8% of the total conduction for a bilayer with a nominal [gA-dig] of 1.4]109 molecules cm~2. When one gramicidin population is –xed and the second titrated the number of dimers reaches a saturating concentration demonstrating that the amount of —ip-—op is small. Furthermore it is possible to estimate K2D1 much greater than [gADIMER] show the graphical determination of a nominal K2D . The leakage component in the outer layer was accounted for by subtracting the conduction values for nominal [gA-dig] from the data with various values of [gA K2D1 (gA-dig) of T].Fig. 3(b) and (c) give estimates of nominal values for 1]109 molecules cm~2 and K2D1 (gAT) K By taking the ratio of the apparent 2D1 s in the two monolayers it is possible to cancel the uncertainties in actual concentration and obtain a ratio of the inner to outer layer gramicidin concentrations. This ratio may be used to constrain the modelling of the biosensor response. following the SA. ( In contrast the introduction of biotin Mr Electrical response to analytes Y was calculated using a spline –t of phase vs. frequency data. Fig. 4(a) shows the kinetic ’ min response at to the b-Fab@ addition. The time-course of the change in Y’ min Y’ min addition of the b-Fab@ and the digoxin were well approximated by the functions Y’ min\Y0(1[e~t@q)]Y1 and Y’ min\Y0 e~t@q]Y1 respectively.In addition measures were obtained of the gating (%) given by 100]Y0/(Y0]Y1) and the normalised maximum slope (NMS) given by (Y0/q)/(Y0]Y1) where Y0\the amplitude of the exponential q\time constant and Y1\the calculated admittance at in–nity. The response elicited from b-Fab@ addition to the membrane arises from cross-linking the gA-dig molecules to the tethered MSL complex. The dependence of the admittance change on the presence of the SA MSL complex in the sensor con–rms this mechanism. Omission of streptavidin replacing the SA biotinylated MSL with a non-biotinylated MSL or replacing the biotinylated Fab@ with a nonbiotinylated Fab@ eliminates the response.Fig. 4(c) shows the time-course of the Y for a range ’ min of concentrations of the b-Fab@. Fig. 4(b) shows the subsequent competitive response elicited by digoxin addition to the bilayer. Digoxin (M 781 Da) causes an increase in the admittance as it r competes with the b-Fab@ immobilised on the MSL 244 Da) has very little eÜect on the membrane conduction. This re—ects the high binding constant of MSLSA b-Fab@ (1]1015 M~1),11,12 compared with that for the gA-dig b-Fab@ complex 253 Faraday Discuss. 1998 111 247»258 ’ min Y’ min response to b-Fab@ for membranes with and without MSL incorporated. SA Y upon addition of 50 nM digoxin b-Fab@. The data show b-Fab@ and without MSL SA ()) SA (L) incorporated. b-Fab@ was added to the response for 50 mM biotin (|) and 10 nM digoxin Fig.4 (a) Comparison of The data shown are for the change in responses for membranes with MSL sensor at t\100. The b-Fab@ yielded a signi–cant response with MSL and a negligible response for the membrane without MSL SA SA . This shows the MSLSA is necessary for cross-linking. Similar controls have shown that the response is negligible if gA is used instead of gA-dig if digoxin Fab@ is used instead of digoxin b-Fab@ and if thyroxin b-Fab@ is used instead of digoxin b-Fab@. (b) Speci–city of competitive response to digoxin. Y Data show the (K) demonstrating that the cross-linked complex gA-dig b-Fab@ MSLSA is not disrupted with biotin but is with 7.8 ng mL~1 digoxin re—ecting the ’ min high binding constant of b-Fab MSLSA the relative stability of gA-dig b-Fab@ MSLSA in PBS and the fast apparent dissociation of gA-dig b-Fab@ MSLSA to gA-dig and b-Fab@ MSLSA .(c) Kinetics of response for titration of digoxin b-Fab@ shown as Y’ min vs. time. The initial Y for each concentration is taken as the ’ min Y mean of at t\0. Digoxin b-Fab@ was introduced at t\150 s into a phosphate buÜered saline electrolyte ’ min Caption continued on next page. Faraday Discuss. 1998 111 247»258 254 k k2D1 and 2D2 \1]10~11 molecules cm~2 s~1; k2D2 ~1 \100 s~1; k2D3 \1]109 molecules cm~2 s~1; k~1 2D3 \1]10~4 molecules cm~2 s~1; k3D1 \1 3D4 \1]104 M~1 s~1; k3D2 ~1 k3D3 ~1 k3D2 k3D3 and k k3D4 ~1 \5 and is illustrated. 50 mM PO4~pH 7.2 22 °C.For the low concentrations the –rst part of the response shows the k2D1 and the reaction is limited by the b-Fab@ concentration. At higher concentrations the initial rate is very fast and reaches a constant value when the response is limited by the equilibrium of the gA monomer»dimer reaction. A decrease in amplitude of response is seen when increasing b-Fab@ concentration above 625 nM as the excess of b-Fab@ results indirectly in gA-dig b-Fab@ and b-Fab@ MSLSA as well as gA-dig b-Fab@ MSLSA . (d) Modelled kinetic responses for digoxin b-Fab@. The experimental data shown given in part (e) have been modelled and are shown here. The input parameters were k2D1 ~1 and ]105 M~1 s~1; s k ~1; 3D1 ~1 \1]10~6 ]10~4 s~1. [MSL]\8]108 molecules cm~2 [gAT]\1]109 molecules cm~2 and [gA-dig]\5]108 molecules cm~2.(e) Kinetics of competitive response for titration of digoxin analyte shown as Y as a ’ min function of time. Digoxin b-Fab@ introduced at t\200 s into a phosphate buÜered saline electrolyte 50 mM PO4~ pH 7.2 22 °C. The rate dependence of the competitive kinetic response which is a function of the surface concentrations of digoxin and MSLSA (2]108 M~1).13 Fig. 4(e) shows the concentration dependence of the time course for the competitive digoxin response. This response is the basis of a fast and sensitive competition assay for digoxin. Modelling kinetic and equilibrium data Fig. 5(a) and (b) show a superposition of the experimental and modelled kinetic and equilibrium responses. In this model it is assumed that membrane leakage is negligible and that the conduction is dominated by gramicidin dimers.Also it is assumed that the solution volume above the membrane is completely mixed at all times and that mass transfer eÜects do not limit the kinetics. It is further assumed that the streptavidin»biotin interaction is essentially irreversible. A software package developed in this laboratory was used to numerically generate kinetic responses for the formation of the digoxin b-Fab@ complexes and the competitive response upon the addition of digoxin. The values for the following parameters were derived from independent kinetic analysis of receptor analyte interactions. k k3D2\2]104 M~1 s~1 IAsys (aÜinity sensors) data13 3D2 ~1 \5]10~4 s~1 IAsys 13 k k3D1\1]105 M~1 s~1 surface plasmon resonance (SPR) data12 3D1 ~1 \\1]10~5 s~1 SPR data12 Other input parameters were as follows K k2D3\1]108»1]1010 molecules cm~2 s~1 2D3\7]108»1]109 molecules cm~2 s~1 k k2D1\1]10~10»1]10~11 molecules cm~2 s~1 2D1 ~1 \0.033»0.1 s~1 The ratio [gA-dig]/[gAT] was estimated as discussed previously.The adopted values of [gAT] [gA-dig] [MSLSA b-Fab@] and K2D2 were varied in the model until a best –t was obtained. Fits to the data shown in Fig. 5(a) and (b) were obtained for speci–ed values given in the –gure caption. This model alone can not determine absolute concentrations of the bilayer components because the surface concentration of accessible binding sites can not de–nitively be correlated to the concentration of membrane components discussed previously.The model also does not provide an adequate description of the gating amplitude and other factors such as the membrane leakage level which can account for up to 10»20% conduction under certain conditions. It is observed experimentally that the kinetic response (NMS) increases linearly with [b-Fab@] up to 5 and 30 nM for [gA-dig] 5]108 and 1]109 molecules cm~2 respectively [Fig. 5(a)]. At this [b-Fab@] dependent portion of the titration curve the binding of b-Fab@ to MSLSA is the rate determining step. Above these concentrations the rate of response is [b-Fab@] independent. The 255 Faraday Discuss. 1998 111 247»258 k2D2 ~1 \100 s~1; k2D3 \5]108; s~1; 5]109 and k2D3 \5]108 molecules cm~2 s~1; Fig.5 (a) Digoxin b-Fab@ response as normalised maximum slope (NMS). The graph shows the experimental and modelled titration curves for the ICS biosensors to digoxin b-Fab@ as (Y0/q)/(Y0]Y1) (NMS). The nominal concentrations for gA-dig are 1.4]109 (K) and 2.8]109 (|). The concentrations used in the model were 5]108 and 1]109 for gA-dig and 1]109 for gA and 8]108 for MSL. The values for k2D3 were varied in T the model as follows for [gA-dig] at 1]109 molecules cm~2 5]10~8 (Ö) 5]10~9 (=) 5]10~10 (>) and for [gA-dig] at 5]108 molecules cm~2 5]10~8 (+) 5]10~9 (@) 5]10~10 (]). Other inputs into the 2D2 \1]10~11 molecules cm~2 s~1; k k2D1 and model were k2D1 ~1 and 5]109 and 5]1010 molecules cm~2 s~1; k2D ~1 3 \1]10~3 molecules cm~2 s~1; M k ~1 s~1; 3D1 \1]105 3D1 ~1 \1]10~6 s~1; 3D4 \2]104 M~1 s~1; 0 3D2 ~1 ; k3D3 ~1 3D4 ~1 \5]10~4 k s~1.k3D2 k3D3 and k k and k (b) Digoxin b-Fab@ response as gating ratio. The graph shows the experimental and modelled titration curves for the ICS biosensors to digoxin b-Fab@ as 100]Y0/(Y0]Y1) (gating ratio). The nominal concentrations for gA-dig are 1.4]109 (K) and 2.8]109 (|). The concentrations used in the model were 5]108 and 1]109 for gA-dig and 1]109 for gA and 8]108 for MSL. The values for k2D3 were varied in the model as follows T for [gA-dig] at 1]109 molecules cm~2 5]10~8 (Ö) 5]10~9 (=) 5]10~10 (>) and for [gA-dig] at 5]108 molecules cm~2 5]10~8 (+) 5]10~9 (@) 5]10~10 (]). Other inputs into the model were k2D1 2D2 \1]10~11 molecules cm~2 s~1; k2D1 ~1 k and k2D2 ~1 \100 and 5]1010 molecules cm~2 s~1; k2D ~1 3 \1]10~3 molecules cm~2 s~1; M k ~1 s~1; k 3D1 \1]105 3D1 ~1 \1 3D4 \2]104 M~1 s~1; ~1 3D2 k~1 3D3 ]10~6 s~1; k3D4 ~1 \5]10~4 s~1.(c) Digoxin analyte 3D2 k3D3 and k k k and response as gating ratio. The graph shows the experimental and modelled titration curves for the ICS bio- 0/(Y0]Y1) 2D2 \1]10~11 100]Y molecules cm~2 sensors to digoxin as k (gating ratio). Model inputs were s~1; s~1; 2D2 ~1 \100 2D3 \1]1010 k2D3 ~1 \1]10~3 molecules cm~2 s~1; k 3D1 \1 k k ]106 M~1 s~1; s k3D1 ~1 \1]10~6 ~1; M k3D3 \2]104 ~1 s~1; s k~1 3D3 \5]10~4 ~1. model shows that the transition from [b-Fab@] dependent to [b-Fab@] independent kinetics represents the transition from a three-dimensional to a two-dimensional rate determining step.Eqn. (4) (5) and (7) describe the 3D reactions. Fig. 5(a) shows that the NMS is proportional to i.e. the absolute rate is independent of [gA so it follows that the rate deterk the [gADIMER] DIMER] mining step is the binding of b-Fab@ to MSLSA [eqn. (4)]. From Fig. 5(a) the gradient of NMS/[b- Fab@] (apparent on rate per dimer) gives 8]104 and 16]104 M~1 s~1 for 1]109 and 5]108 gA-dig molecules cm~2 respectively. The ratio of MSL to gA-dig can be derived from these apparent\k3D1 [MSLSA]/[gADIMER]. Using the independently mea- values using the relationship SA k sured value for 3D1\1]105 M~1 s~1 the ratio of [MSLSA]/[gADIMER] has been derived and shown to be 0.8 and 1.6 for 1]109 and 5]108 gA-dig molecules cm~2 respectively.These Faraday Discuss. 1998 111 247»258 256 values compare with ratios of 25 and 50 for nominal [MSL ]/nominal [gA-dig]. While the low SA ratio may re—ect actual ratios of deposited components (b-MSL/gA-dig) it may alternately be accounted for by lower than expected ratios of or MSLSA b-Fab@/gA-dig MSLSA/gA-dig. In turn this low ratio may re—ect either functionality or accessibility of the species. For the portion of the titration curve for which the response is 2D rate limited either the cross-linking of gA-dig to MSLSA b-Fab@ [eqn. (3)] or the dissociation of gA-dig or gA-dig b- Fab@ from gADIMER [eqn. (1) and (2)] may be the rate determining step. If the rate limiting step is the cross-linking of b-Fab@ to gA-dig [eqn.(3)] –tting the model to the experimental data [Fig. 5(a)] shows the sensitivity of the response to the rate of surface analyte cross-linking. Alternatively if the rate-limiting step is the dimer dissociation k2D1 ~1 can be determined from the k reciprocal of NMS in the 2D limited part of the curve [Fig. 5(a)]. By this analysis 2D1 ~1 equates to 0.003 s~1. This value is comparable to the independently measured gA dimer life-time reported by Anderson and co-workers.14 It is not possible to discriminate which of these 2D reactions are limiting from this data set. The equilibrium response (gating ratio) also shows the 3D limiting region of the curve up to 5 and 30 nM for [gA-dig] of 5]108 and 1]109 molecules cm~2 respectively [Fig.5(b)] through the dependence of the response on surface or solution analyte concentration. Two-dimensional rate constants K DIMER . k2D3 between 1]10~7 and 5]10~9 cm2 mol~1 [Fig. 5(b)] do not impact the gating response signi–cantly. However for rates of less than approximately 5]10~9 cm2 molecules~1 s~1 the response K becomes limited by 2D3 . When is limited the gating ratio is reduced and there is a greater 2D3 ììhookingœœ eÜect. The data suggests that the kinetics is not strongly limited by K2D3 . Above approximately 200 nM b-Fab@ the response decreases. This phenomena provides evidence for the binding of b-Fab@ onto gA-dig and gA The binding of the b-Fab@ onto the gA inhibits its ability to cross-link to the MSLSA b-Fab@. If the 2D limiting is surface analyte crosslinking then the equilibrium response at this region will be dependent upon [MSLSA] providing a test for the nature of the 2D limiting process.The –tted data is sensitive to the number of accessible functional binding sites presented on in that a b-Fab@ response is not observed when the concentration of b-MSL is decreased MSLSA by a factor of 10. The model successfully predicts this and supports the low ratio of [MSL ]/[gA- SA dig] actual discussed previously. The competitive kinetic response is rate and amplitude dependent which arises from the combined 2D and 3D kinetics. The rate is a function of the surface concentration of digoxigenin [gAdig] as free digoxin will bind directly to MSLSA b-Fab and compete for complexed b-Fab.When using the same parameters for modelling the competitive experimental gating response there is again good agreement between the data and the model. The model is constrained to the values shown in Fig. 5(c). To obtain the –nal –t to the data the k2D ~12 [eqn. (5)] must be equal to or less than 1]10~3 s~1. [gA-dig] and gA-dig and MSL can be estimated. Conclusion A biosensor based on a functional synthetic bilayer membrane has been constructed. The speci–c b-Fab@ response has been characterised as a kinetic process occurring in both two- and threedimensional planes involving the capture of analyte onto the membrane surface and the crosslinking of the b-Fab. The modelling of the competitive ICS has demonstrated that the experimental data is in good agreement with our mechanistic model and has enabled the parameters within the model to be constrained.By titrating gA in the inner and outer layer the relationship between [gA K T] 2D1 has been determined. Using this relationship in the model and using independent measures of 3D receptor binding constants the actual concentrations of gAT SA Acknowledgements We would like to thank Denise Thomas for her technical contributions and advice. We are grateful to Dr Frank de Hoog and Huu-Nuynh CSIRO-Mathematical and Information Sciences for 257 Faraday Discuss. 1998 111 247»258 helping to construct the mathematical model which was used for the analysis of the data reported herein. This work has been supported by the Cooperative Research Centre (CRC) program.The partner organisations within the CRC for Molceular Engineering and Technology are the Commonwealth Scienti–c and Industrial Research Organisation (CSIRO) the University of Sydney and the Australian Membrane and Biotechnology Research Institute (AMBRI). We thank all those within the CRC who have contributed to the invaluable discussions and who have supplied us with the quality materials that enabled the study to be undertaken. Paper 8/09608B References 1 B. Raguse V. Braach-Maksvytis B. A. Cornell L. G. King P. D. J. Osman R. J. Pace and L. Wieczorek 2 B. A. Cornell V. Braach-Maksvytis L. G. King P. D. J. Osman B. Raguse L. Wieczorek and R. J. Pace 3 K. Raval P. Culshaw J. Prashar and B. Raguse Synthesis of functionalised membrane spanning lipid L angmuir 1998 3 648.Nature (L ondon) 1997 387 580. derivatives manuscript in preparation. 4 S. Kim D. Bali and L. G. King Synthesis of functionalised gramicidin A derivatives manuscript in preparation. 5 G. W. Oddie L. C. Gruen G. A. Odgers L. G. King and A. A. Kortt Anal. Biochem. 1997 244 301. 6 A. L. Plant M. Gueguetchkeri and W. Yap Biophys. J. 1994 67 1126. 7 R. Benz O. Frohlich P. Lauger and M. Montal Biochim. Biophys. Acta 1975 394 323. 8 P. Delahay Double L ayer and Electrode Kinetics Wiley-Interscience New York 1965. 9 A. Ulman Introduction to T hin Organic Films From L angmuir»Blodgett to Self-Assembly Academic Press Boston 1991. 10 A. Lewis and D. M. Engleman J. Mol. Biol. 1983 166 211. 11 N. M. Green Advances in Protein Chemistry Academic Press New York 1975 pp. 29»85. 12 S. Martin and G. Woodhouse binding kinetics derived using surface plasmon resonance to determine binding rates of streptavidin binding to a monolayer of 2000 1 MSL b-MSL molecules cm~2 unpublished data. 13 G. Krishna and L. G. King Binding kinetics of digoxin»b-Fab@ obtained through analysis of binding rates of digoxin b-Fab@ to NHS-xx-digoxigenin coupled to the carboxydextran matrix of an IAsys cuvette (Affinity Sensors) unpublished data. 14 D. B. Sawyer S. Oiki and O. S. Andersen Biophys. J. 1990 57 100a. Faraday Discuss. 1998 111 247»258 258

ISSN:1359-6640    

DOI:10.1039/a809608b

出版商:RSC    

年代:1999

数据来源: RSC