| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Biophys J, February 2001, p. 801-811, Vol. 80, No. 2
and
*Department of Physiology and Biophysics and the Program in
Macromolecular Structure, University of California at Irvine,
Irvine, California 92697-4560 USA and Department of
Biochemistry, School of Medical Sciences, University of Bristol,
Bristol, BS8 1TD, United Kingdom
 |
ABSTRACT |
|---|
 TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
RESULTS AND DISCUSSION
CONCLUSIONS
APPENDIX
REFERENCES
|
|---|
Melittin is arguably the most widely studied amphipathic,
membrane-lytic 
-helical peptide. Although several lines of evidence suggest an interfacial membrane location at low concentrations, melittin's exact position and depth of penetration into the
hydrocarbon core are unknown. Furthermore, the structural basis for its
lytic action remains largely a matter of conjecture. Using a novel
x-ray absolute-scale refinement method, we have now determined the
location, orientation, and likely conformation of monomeric melittin in oriented phosphocholine lipid multilayers. Its helical axis is aligned
parallel to the bilayer plane at the depth of the glycerol groups, but
its average conformation differs from the crystallographic structure.
As observed earlier for another amphipathic -helical peptide, the
lipid perturbations induced by melittin are remarkably modest. Small
bilayer perturbations thus appear to be a general feature of
amphipathic helices at low concentrations. In contrast, a dimeric form
of melittin causes larger structural perturbations under otherwise
identical conditions. These results provide direct structural evidence
that self-association of amphipathic helices may be the crucial initial
step toward membrane lysis.
| |
INTRODUCTION |
|---|
|
TOP
ABSTRACT
INTRODUCTION
MATERIALS AND METHODS
RESULTS AND DISCUSSION
CONCLUSIONS
APPENDIX
REFERENCES
|
|---|
Detailed structural and thermodynamic information
about amphipathic -helical peptides in membranes is important
because of the widespread occurrence of the motif in host-defense
peptides and membrane proteins. The most extensively studied of such
peptides is undoubtedly melittin (MLT) (Habermann, 1972
), the
26-residue membrane-lytic (Sessa et al., 1969) peptide isolated from
the venom of the European honeybee Apis mellifera (reviewed
in Dempsey, 1990). Nevertheless, little structural information about
MLT in membranes is available beyond the basic facts that it forms, at low membrane concentrations, monomeric (Altenbach and Hubbell, 1988;
John and Jähnig, 1991) -helices in the membrane interface arranged parallel to the bilayer plane (Frey and Tamm, 1991; John and
Jähnig, 1991; Dempsey and Butler, 1992; Okada et al., 1994; White
et al., 1998; Ladokhin and White, 1999). The exact location, however,
of MLT in the interface has remained a matter of speculation. The lack
of a precise location is evident in recent molecular dynamics
simulations in which the starting configurations place MLT in positions
ranging from outside the headgroup region (Lin and Baumgärtner,
2000) to positions deep within the interface, with a variety of
orientations (Bernèche et al., 1998; Bachar and Becker, 1999,
2000). We present here x-ray diffraction results that
precisely pinpoint the location and orientation of MLT in a membrane interface.
At higher concentrations, MLT permeabilizes membranes by
forming ~25-Å diameter pores (Katsu et al., 1988; Ladokhin et al., 1997a) thought to consist (Vogel and Jähnig, 1986) of
transmembrane helices in a barrel-stave arrangement (Baumann and
Mueller, 1974; Fox and Richards, 1982; Hall et al., 1984), possibly
containing interspersed lipids (Matsuzaki et al., 1997). The helical
structure of MLT in fluid membranes is frequently assumed to be
identical (Bernèche et al., 1998) with the crystallographic
structure (Terwilliger and Eisenberg, 1982a,b), although nuclear
magnetic resonance (NMR) studies suggest otherwise (Dempsey and Butler,
1992; Okada et al., 1994). The structure of the MLT monomer in
tetrameric crystals, grown from high ionic strength aqueous solutions
(Terwilliger and Eisenberg, 1982a,b), consists of two -helical
segments separated by a characteristic kink at
Pro14. For MLT in membranes, Terwilliger et al.
(1982) conjectured that the crystal structure is largely preserved and
arranged with the plane of the kink perpendicular to the membrane.
Although this structure is suggestive and forms the basis for
barrel-stave pore models (Vogel and Jähnig, 1986), MLT-induced
membrane permeabilization is only vaguely understood. Kinetic studies
suggest helix association on the membrane as a precursor of pore
formation, with dimerization being the rate-limiting step in many
(DeGrado et al., 1982; Schwarz and Beschiaschvili, 1989; Schwarz et
al., 1992), but not all (Rex and Schwarz, 1998), cases. Recently, Takei
et al. (1999) provided direct evidence for the kinetic importance of
MLT association in pore formation through studies of
cysteine-substituted MLT pairs linked by disulfide bridges. We examine
here one such pair formed from a substitution of Cys for Gln at
position 25.
We consider four questions: Does the conformation of MLT in fluid
bilayers differ from what is observed in the crystal structure, as
suggested by NMR studies? What is the location and orientation of MLT
in bilayers? To what extent is the bilayer perturbed by MLT monomers?
Does dimerization increase the perturbations? We have sought answers to
these questions using a novel absolute-scale x-ray diffraction method
previously applied to another amphipathic helical peptide,
Ac-18A-NH2 (Hristova et al., 1999). Specifically, we have examined MLT and a cysteine-linked dimeric MLT (Q25C) in
oriented multilamellar bilayer arrays formed from
dioleoylphosphatidylcholine (DOPC) at 66% relative humidity (RH). The
results show, as for Ac-18A-NH2, that monomeric
MLT causes only minor changes in bilayer structure. This suggests
that a general feature of amphipathic helices in membrane interfaces at
low concentrations is small bilayer perturbations. In contrast,
several small peptides and Q25C cause significantly larger perturbations.
| |
MATERIALS AND METHODS |
|---|
|
TOP
ABSTRACT
INTRODUCTION
MATERIALS AND METHODS
RESULTS AND DISCUSSION
CONCLUSIONS
APPENDIX
REFERENCES
|
|---|
Materials
DOPC and
1-oleoyl-2-(9,10-dibromostearoyl)-sn-glycero-3-phosphocholine
(OBPC) were purchased from Avanti Polar Lipids (Alabaster, AL). Purity
of OBPC was determined by elemental analysis to be >99.9% (Microlit
Laboratories, Madison, NJ). MLT, HPLC purified, was purchased from
Sigma (St. Louis, MO). This grade of MLT, though not listed in the
Sigma catalog, can be ordered using catalog number M1407. Q25C was
synthesized and characterized as described elsewhere (Takei et al.,
1999).
Membrane composition and oriented circular dichroism
Water contents of oriented multilayers with peptides were
determined gravimetrically as described previously (Hristova et al.,
1999). The nature and number of counterions in peptide samples were
determined after lyophilizing solutions of the peptide and pumping
under high vacuum, followed by solution NMR in
D2O to identify and quantitate counterions by
integration. Oriented circular dichoroism (OCD) was carried out on
samples prepared in exactly the same manner as those used for
diffraction studies, as described previously (Hristova et al., 1999),
following the approach of Huang and colleagues (Wu et al., 1990).
X-ray diffraction
Lipid and MLT (1 mol%) were codissolved in methanol, and then
deposited on a curved glass substrates in the manner previously described (Wiener and White, 1991c; Hristova and White, 1998). RH was
maintained at 66% with a saturated solution of
NaNO2. The sample was placed in a custom-made
humidity chamber with thin x-ray-transparent beryllium windows. The
curved surface of the substrate was placed in the x-ray beam in a
manner that permitted all lamellar diffraction orders to be recorded in
a single experiment. In this geometry, most of the wide-angle
scattering, due, for example, to the lipid acyl chains, was absorbed by
the glass substrate (Wiener and White, 1991c). Because of the thermal
motion of the bilayer and the presence of the peptide, only five orders
of diffraction data were observable, compared to the eight orders
observed in the absence of MLT. The peptide, in essence, smoothes out
the more rugged peptide-free bilayer profile. Five orders of
diffraction thus yield fully resolved peptide-bilayer profiles, as
discussed below and more extensively elsewhere (Hristova et al., 1999).
"Fully resolved" means that five orders of diffraction
(h = hmax) is the
maximum number observable. That is, the shape of the profile is such
that no more than five orders can be produced by diffraction, even
though the lattice is nearly perfect. The confusing issue of resolution
in lamellar diffraction has been discussed in detail by Wiener and
White (1991a). Particularly confusing is the meaning of the canonical
resolution, defined as
d/hmax where d
is the Bragg spacing (unit cell dimension). With d ~ 50 Å, the canonical resolution is ~10 Å in these experiments. This
number is a measure of the thermal disorder of the system in the
absence of lattice disorder, as in our experiments. In other words, the
scattering atoms are spread out over a significant fraction of the unit
cell, on the time scale of the diffraction experiment. The result is a
broad, smooth electron density envelope that requires no more than
hmax terms for a faithful Fourier
reconstruction from the diffracted intensities. This canonical
resolution is often confused with what White and Wiener (1995,
1996) have dubbed "resolution precision," which is the precision of
determining the position and width of the smooth envelope of electron
density within the unit cell. The resolution precision depends upon the relative scattering strength of the envelope and its position in the
unit cell, but is typically smaller than 0.5 Å (Wiener and White,
1991a), and can be as small as 0.01 Å. Consequently, we were able to
determine the positions and widths of the MLT envelope and Br-labeled
double-bonds with excellent precision.
Sample degradation was monitored by thin layer chromatography (TLC) and
high performance liquid chromatography (HPLC). For typical exposure
times of 8-10 h, no degradation was detected. Furthermore, no
systematic differences in the line widths or integrated intensities
were observed. X-ray diffraction measurements were performed with
CuK
radiation using a rotating anode x-ray generator (Bruker AXS (formerly Siemens), Madison, WI) operated at 38 kV and 30 mA. Double-focusing optics (Charles Supper, Nattick, MA) were
used to focus the beam at the detector. The diffraction pattern was
recorded on a Siemens X-1000 xenon-filled area detector with
position-decoding circuit and real-time data display. The collection of
x-ray data, peak integration, and absorption corrections were performed
as described extensively elsewhere (Hristova and White, 1998).
Experimental uncertainties for each peak were obtained from the
statistical uncertainties of the integrated intensities of the
diffraction peaks taken as (peak area + background)1/2. As discussed below, complete
x-ray data sets were collected using six different OBPC:DOPC mole
fractions ranging from 0 to 0.50. This procedure, used for both MLT and
Q25C, is equivalent to six experimental determinations of structure
factors and Bragg spacing.
Absolute-scale structure refinement
Lamellar diffraction is the only method available for the direct
determination of the structure of lipid membranes in their natural
fluid state (Franks and Levine, 1981). Fourier inversion of the
diffracted intensities of x-rays or neutrons obtained from multilamellar arrays of fluid bilayers yields one-dimensional scattering-length density "profiles" (e.g., Fig.
1) that represent the projection of the
highly thermally disordered contents of the unit cell onto the bilayer
normal (Wiener and White, 1991a). The profiles are usually determined
on a relative scale, i.e., the amplitudes of the electron density
fluctuations around the mean density are displayed on an arbitrary
scale. Although these relative-scale profiles provide useful
rudimentary information about bilayer structure, we have demonstrated
(Hristova et al., 1999) that they provide little or no insight into the
disposition of peptides within bilayers. The usefulness of lamellar
diffraction experiments is improved dramatically by determining
profiles on an absolute scale, as first shown by MacNaughtan et al.
(1985). Importantly, absolute scaling ties profiles directly to the
composition and physical density of the bilayer (Franks et al., 1978;
Wiener and White, 1991a; Hristova and White, 1998; Hristova et al.,
1999). Although absolute-scale profiles are often presented in units of
electrons per unit volume (electron density), we generally use
scattering-length density units because they are more appropriate when
x-ray data and neutron data are combined in so-called composition space
refinement (Wiener and White, 1991b). As discussed in the Appendix,
scattering-length density is obtained by multiplying electron density
by
mc2/e2.
The most convenient scattering-length absolute scale is the so-called
per-lipid scale (Hristova et al., 1999), which has the advantage of not
requiring direct knowledge of the area per lipid (see the Appendix).
Regardless of the units used, it is the differences in densities
between profiles in the presence and absence of peptide that matter.
The computation of the positions and widths of Gaussian profile
features is unaffected by the choice of units.
|
Wiener and White (1991a,b) showed that combined x-ray and neutron
experiments using heavy-atom specific labeling not only allow profiles
to be placed on an absolute scale, but permit their decomposition into
a collection of Gaussian component distributions as well. Each
distribution represents the time-averaged projection of the motions of
the water and principal lipid structural (component) groups (carbonyls,
double-bonds, phosphate, etc.) onto the bilayer normal. The
distributions taken as a group account for the contents of the unit
cell and comprise the structure of a fluid bilayer (e.g., Fig.
2 A).
The positions and widths of the Gaussians are obtained by joint
refinement of x-ray and neutron data using a crystallographic approach,
referred to as liquid-crystallography (see reviews in White and Wiener,
1995, 1996). The absolute-scale x-ray refinement method (Hristova et
al., 1999) we employ here uses, as a starting point, the complete
liquid-crystallographic structure of a peptide-free fluid bilayer, in
this case DOPC at 66% RH (Wiener and White, 1992).
|
In the presence of a peptide, the structure of the peptide-bilayer
complex is given by the superposition of the transbilayer distribution
of the peptide with the set of component distributions of the
peptide-perturbed lipid bilayer. The object of absolute-scale refinement is to determine these distributions. As described in detail
elsewhere (Hristova et al., 1999), refinement involves four steps: 1)
Determination of the x-ray scattering-length density profiles of the
bilayer with and without peptide on the per-lipid absolute scale. This
is accomplished by means of an isomorphous marker-lipid (OBPC)
equivalent to DOPC brominated at the 9,10-double-bond of the
sn-2 chain through an addition reaction (Wiener and White, 1991c). This step also yields the transbilayer double-bond distribution in the presence and absence of peptide, as shown by Wiener and White
(1991c). Importantly, this distribution serves as a general measure of
the state of the bilayer hydrocarbon (HC) core (Hristova and White,
1998). 2) Construction of peptide-perturbed bilayer model structures
derived from changes in Bragg spacing and the double-bond distribution.
This step is essential because the peptide causes the bilayer structure
and the contents and dimensions of the unit cell to change. 3)
Determination of the best transbilayer Gaussian distribution for the
peptide. We have shown by molecular dynamics simulations that a
Gaussian distribution accurately describes the transbilayer
distribution of peptides in highly thermally disordered bilayers
(Hristova et al., 1999). 4) Determination, by model building, of the
range of peptide conformations, positions, and orientations that best
satisfy this distribution. The successful application of our approach
depends, however, on the peptide-induced bilayer structural changes
being small enough to allow the perturbed bilayer to be modeled from
the unperturbed bilayer in Step 2 by a simple scaling procedure (below)
(Hristova et al., 1999). We found that the method could be applied
successfully for monomeric MLT, but not dimeric Q25C. This means that
the perturbations caused by the dimeric form are substantial.
Refinement procedures
Absolute-scale bilayer profiles
All bilayer profiles were placed on the absolute per-lipid scale using mixtures of DOPC and an isomorphous variant (OBPC) brominated at the 9,10 position of the sn-2 chain, as described in detail elsewhere (Wiener and White, 1991c; Hristova and White, 1998). Briefly,
the experimental structure factors f(h) from a given experiment depend upon the amount of sample in the beam, precise
geometry of the sample-beam interaction, x-ray beam intensity, and
other experimental conditions. The true (absolute) structure factors,
F(h), are determined solely by the scattering
factor of the unit cell. The experimental structure factors are related to the true structure factors by f(h) = KF(h) in which K is the instrumental
constant. Fourier reconstructions of bilayer scattering-length or
electron density profiles yield only arbitrary fluctuations of
scattering density along the bilayer normal if the average scattering
length density of the unit is not accounted for, and whether
f(h) rather than F(h) is
used. Determination of the instrumental constant allows one to relate
the scattering profiles obtained in diffraction experiments to the
actual contents and molecular packing of the bilayer unit cell. To do
this, one must determine the true mean value of the scattering profile
using the composition of the unit cell and calibrate the fluctuations
around this mean value based upon difference structures of Br-labeled
and unlabeled DOPC bilayers (Franks et al., 1978; Wiener and White,
1991c; Hristova and White, 1998). (It is these difference structures
that disclose the transbilayer distribution of Br-labeled
double-bonds.) The absolute scattering lengths are easily computed from
knowledge of the mole fraction of Br-labeled DOPC in the bilayer and
known scattering length of Br. This serves to calibrate the amplitude of the fluctuations on the absolute scale.
Data scaling and phasing
We scaled the diffracted intensities by gathering complete x-ray data sets for at least six OBPC:DOPC mole fractions ranging from 0 to 1. This procedure, used for both MLT and Q25C, is equivalent to six experimental determinations of the structure factors, but with several bonuses. It reduces experimental uncertainties, averages out random error, and assures that OBPC is isomorphous with DOPC in the presence and absence of the peptide. Isomorphous behavior is proven if the measured structure factors are linear functions of the mole fraction of OBPC. Wiener and White (1991c) have described in detail a procedure for
scaling multiple data sets that involves, in simple terms, rescaling
the structure factors so that the data sets are described by a set of
internally consistent experimental constants. One result of the
procedure, automated by means of computer programs, is good estimates
of the experimental uncertainties of the structure factors, which is
necessary for establishing the value of the so-called self-R-factor
(see below). Another result is that the phases of the structure
factors, at all OBPC mole fractions, are established unequivocally
(Wiener and White, 1991c; Hristova et al., 1999).
Perturbed-bilayer models
The structure of the perturbed bilayer in the experiments with MLT was estimated (Hristova et al., 1999) from the measured changes in
the parameters of the bromine distribution
ABr and ZBr and the Bragg spacing
d. The bromine distribution reports the distribution of
double bonds in DOPC bilayers and provides information about the
physical state of the hydrocarbon core (Hristova and White, 1998). The
comparison of the bromine distributions with and without peptide
reveals the changes in the hydrocarbon region that occur due to peptide
insertion (Hristova et al., 1999). We created several bilayer models
based upon the measured changes in d-spacing or
ZBr, and judged their adequacy against
the experimental data in the structure refinement process. Only those
models with R < Rself
(see below) were retained. We found that satisfactory models could be
produced in only two ways. In the first, called Model A, only the
positions of the lipid components were scaled according to the measured
change in Bragg spacings (see Hristova et al., 1999). That is, the
Gaussian positions of the Wiener and White (1992) structure were simply
scaled by
dDOPC+MLT/dDOPC. The widths were not changed. In the second, called Model B, good fits
were obtained by additionally scaling the widths in accordance with the
slight increase in the bromine width distribution in the presence of
MLT. The methods for evaluating the quality of the structure refinement
have been described in detail (Hristova et al., 1999).
The number of structure factors required to describe the perturbed
bilayer profile is determined by the sum of the structure factors
resulting from the Gaussian distributions, and also careful consideration of experimental uncertainty and signal-to-noise ratios.
We typically consider values as high as h = 8, the
number observed for the neat bilayer. But an interesting thing happens when the structure factor contributions of the peptide are added to
obtain the model peptide + bilayer structure factors for comparing to
the observed structure factors: The number of structure factors observable, hmax, decreases because of
the smoothing effect of the broad peptide distribution! We have
discussed this phenomenon in quantitative detail elsewhere (Hristova et
al., 1999). Because of it, the difference in number of structure
factors between the neat and peptide-containing structure factors is
automatically reconciled, physically. A practical problem remains,
however. Upon adding the structure factors of the MLT Gaussian
distribution (below) to the perturbed-bilayer structure factors, we
have only five experimental peptide + bilayer structure factors for
carrying out the optimization of the MLT Gaussian. We resolve the
problem by simply using the first five structure factors of the model bilayer, and ignoring the higher ones. When we follow this protocol, we
invariably find that the higher-order structure factors produced by the
model bilayer + MLT Gaussian are predicted to be below the limits of
experimental detectability, as discussed by Hristova et al. (1999).
Determination of Gaussian parameters for the melittin distribution
The Gaussian parameters of the MLT transbilayer distribution were determined by nonlinear least-square fitting of the calculated structure factors to the experimental ones (Hristova et al., 1999). For
a particular model bilayer, the refinement computation finds the
optimal position ZMLT and width
AMLT of the peptide transbilayer distribution. For each cycle of the computation, the structure factors
of the peptide distribution were added to the fixed structure factors
of the bilayer model. These summed structure factors were then compared
to the observed structure factors of the DOPC/MLT complex by means of
the crystallographic R-factor.
Peptide modeling
The most realistic approach to modeling the peptide would be to produce an ensemble of conformations in a bilayer environment by molecular dynamics simulations, but this approach is presently impractical. We therefore adopted a simpler method that allowed us to explore a reasonable range of peptide backbone and sidechain conformations. For a particular peptide conformation, each atom (a) was represented in the z-axis projection by a Gaussian scattering distribution whose 1/e-halfwidth Aa was related to the atom's B-factor. All atoms were assigned the same B-factor during the refinement procedure, for two reasons. First, we did not have a sufficient number of structure factors to vary the B-factors of, say, individual residues. Second, because of the very high thermal disorder of the bilayer, there is no reason to expect significant differences in B-factors among the sidechains. As discussed below, we accept only those models whose average B-factor is within the range of B-factors of the lipid components. The peptide model-building step was implemented by generating a library of peptide structures using molecular dynamics simulations whose positions and orientations in the bilayer were optimized by refinement of the calculated structure factors of the bilayer/peptide complex against the observed structure factors. The primary refinement variables used were the position and tilt of the peptide axis and the average crystallographic Debye temperature factor (B) of the peptide's atoms. The B-factor is a measure of the amplitude of the thermal fluctuations of an atom around its mean position (Warren, 1969). By "average" B, we mean that a single
B-factor was applied to all atoms (above). We assumed that
the transbilayer Gaussian envelope of the whole peptide could be
obtained by summing the Gaussian scattering-length densities of the
individual atoms and that the most likely atomic B-factors
would be those that were close to the B-factors of lipid
component groups. The latter assumption is reasonable because the
conformational flexibility of the peptide must surely reflect the
thermal motion of its surroundings, i.e., the fluid bilayer. This
assumption provided the basis for choosing the most likely peptide
conformations from the library of peptide structures.
The Gaussian obtained in the determination stage was thus used in a
second minimization routine to obtain the optimum average B-factor for a particular structural model. If the
calculated B-factors were of the order of the
B-factor for the glycerol moiety (Hristova et al., 1999) (78 Å2) or higher (but <200
Å2), the conformation was considered consistent
with the data. A library of MLT conformations was created with the
software package Insight II (Biosym Technologies, San Diego, CA). After
minimization, 1 ps molecular dynamics simulations were carried out at
300 K to produce a range of sidechain conformations. The projections of
the atom coordinates along the bilayer normal defined the centers of
the transbilayer distributions of the atoms. The average of these
projections defined the center of the helix transbilayer distribution.
The widths of the atom distributions were characterized by a Gaussian
with 1/e-halfwidth A, given by
A2 = Ac2 + Aa2,
where Ac is the covalent radius of the
atom, and Aa is the thermal atomic
1/e-halfwidth, which is a measure of the thermal motion of
the atom in the fluid bilayer. The value of
Aa is related to the crystallographic
B-factor according to B = 4
2Aa2.
Refinement computations
The computations used nonlinear minimization of R using the standard Levenberg-Marquardt algorithm (Bevington, 1969;
Press et al., 1989). Only those solutions were accepted whose
R-factors were smaller than the so-called "self"
R (Rself) of the observed structure factors. Rself is defined
(Wiener and White, 1991b) as Rself2 = 
h (
(h))2/h (F*(h))2,
where the (h) are the uncertainties in the determination
of the structure factors F*(h). Rself
measures, in essence, the total experimental uncertainty of the
observed structure factors after scaling (Wiener and White, 1991b). The
value of Rself was 6.0 × 10
2 for the experiments
reported here.
Error analysis
The robustness of the fits and the uncertainties in ZMLT and AMLT for the MLT distribution (Step 3) were determined using the Monte Carlo sampling procedure of Wiener and White (1991a,b). This procedure is based upon the fact that each
structure factor has an experimental uncertainty (h) that
can be used to define a normal distribution for each structure factor
F(h). In simple terms, a Box-Muller algorithm
(Ross, 1989) seeded with a random number is used to generate sets of
"mock" structure factors from the observed
F(h) and (h). The mean and standard
deviation of these mock sets will match those of the observed data.
Although each set of mock data represents a statistically acceptable
combination of structure factor amplitudes, each set will yield
slightly different values for the parameters obtained in the
refinement. The mean values and standard deviations of the collection
of parameters describe the most likely values of the parameters and
their uncertainties. If all of the sets of mock data lead to a
convergence of the refinement, the fits can be considered robust. The
error ranges noted in the figures and tables were obtained in this way.
| |
RESULTS AND DISCUSSION |
|---|
|
TOP
ABSTRACT
INTRODUCTION
MATERIALS AND METHODS
RESULTS AND DISCUSSION
CONCLUSIONS
APPENDIX
REFERENCES
|
|---|
Disposition of melittin in DOPC bilayers
The absolute per-lipid scale bilayer profiles determined for pure
DOPC (Wiener and White, 1992; Hristova et al., 1999) and DOPC + 1 mol%
MLT are shown in Fig. 1 A and the corresponding structure
factors in Table 1. The half-unit cell
contents, required for the absolute scaling, are 5.4 waters per lipid
in the absence of MLT and 5.6 waters, 0.01 MLT, and 0.06 acetyl
counterion per lipid in the presence of MLT. (Six acetyl counterions
remain associated with MLT after freeze-drying, as determined by NMR
[data not shown]). On a volume-fraction basis, MLT plus the acetyl
counterions accounted for 2.75% of the unit-cell volume. Even casual
examination of the two profiles in Fig. 1 A discloses a
highly significant increase in scattering density in the headgroup
region of the bilayer when MLT is present. In the central region of the
DOPC/MLT profile, corresponding to the bilayer HC core, the changes
caused by MLT are barely outside experimental uncertainty (dotted
lines). The MLT-induced bilayer perturbations were small enough to
allow models for the perturbed bilayer to be generated from the
unperturbed bilayer structure (Wiener and White, 1992) by the simple
scaling procedure (Hristova et al., 1999) described in Methods. The
profile of one model is shown in Fig. 1 C (blue
curve) and a partial set of component distributions in Fig.
2 A. The packing constraints on the bilayer are not
apparent in such distributions because the contents of the unit cell
are projected onto the bilayer normal, causing the component groups to
overlap. In three-dimensions, however, the component groups are
excluded from each others' volumes (discussed in Hristova et al.,
1999).
|
Starting with the perturbed-bilayer models and superposed Gaussians for the MLT transbilayer distribution, the positions (±ZMLT) and widths (AMLT) of MLT distributions were obtained by nonlinear least-squares fitting in reciprocal space. Parameters resulting from the two different bilayer models, A and B (above), are shown in Table 2. Fig. 1 C shows the five-order Fourier reconstruction of the profiles of the model bilayer (blue curve), MLT (violet curve), and the composite of the two (red curve). The excellent quality of the refinement is apparent from the tight overlap of the composite with the experimentally determined profile (black dashed line). The best estimates for the MLT Gaussian parameters are ZMLT = 17.5(±0.2) Å and AMLT = 4.3(±0.4) Å, obtained by averaging the values in Table 2. The MLT distribution is superimposed on the component distributions of the perturbed-bilayer model in Fig. 2 A. The center of the distribution is very nearly the same as that of the glycerol moiety of the DOPC bilayer. OCD measurements on MLT in oriented DOPC multilayers obtained under experimental conditions identical to those of the diffraction experiments showed that MLT was helical and oriented parallel to the bilayer plane (below). The center of the observed distribution therefore corresponds to the location of the helix axis and the width of the distribution approximately to the average diameter of the helix, ~9 Å. Figure 2 A shows, then, that the surface of the MLT helix extends to about the depth of the DOPC double bonds.
|
The most likely helical conformations of MLT that account for the
observed distribution were determined by first creating a library of
helical conformations derived from the NMR models of Dempsey and Butler
(1992) and Okada et al. (1994) using methods previously described
(Hristova et al., 1999). Because both studies indicated that the last
four residues (KRQQ) lack secondary structure in membranes, we included
this feature in all models. The structure factors of these models were
evaluated against the structure factors of the experimentally
determined MLT Gaussian (above) by determining the average
B-factors required for a satisfactory fit. Acceptable models
were those with average B-factors falling between 70 and 200 Å2, which are characteristic of fluid bilayers
(Hristova et al., 1999). One of many possible models that is consistent
with the experimental data is shown in Fig. 2 B[1]
(red dashed curve) along with the experimentally determined
distribution (solid black curve). For this model,
B = 140 Å2. The excellent
quality of the model is obvious from the fact that its distribution is
well within the experimental uncertainty of the observed distribution
(black dashed curves). Only those conformations that were
highly helical and oriented parallel to the bilayer surface satisfied
the experimental distribution, consistent with OCD measurements (below).
We specifically examined the possibility that MLT in the conformation
observed in the crystal could explain our diffraction results. The
crystallographic model of MLT oriented with the plane of the
Pro14 kink normal to the membrane plane, as
proposed by Terwilliger et al. (1982), did not provide a satisfactory
fit to the observed distribution (Fig. 2 B[2]) with
B = 140 Å2. Nor did orientation
of the kink parallel to the bilayer plane (Fig. 2 B[3]).
The kink-parallel crystallographic conformation is thus, at best, a
minor constituent of the ensemble of conformations represented by the
observed MLT distribution. But we could not arrive at satisfactory fits
of the crystallographic structures unless the average
B-factors of the structures were fixed at nearly 0. This in
itself shows that the crystallographic models do not describe the
conformation of MLT in fluid bilayers.
Bilayer perturbations due to monomeric mellittin
The introduction of MLT into the DOPC bilayer caused the Bragg
spacing d to decrease and the brominated double-bond
position ZBr to shift toward the
bilayer center with a slight increase of its 1/e-halfwidth
ABr (Table 1). The distributions of
the DOPC double bonds in the presence and absence of MLT, as reported by Br at the 9,10 positions of the sn-2 chain of OBPC, are
shown in Fig. 3 A (blue
curve and dashed black curve, respectively). These
changes show that monomeric MLT has a remarkably modest effect on the
structure of the DOPC bilayer at low concentrations. And so does the
amphipathic helical peptide Ac-18A-NH2 (Hristova et al., 1999), a model for apolipoprotein A-I that is known for its
ability to completely solubilize bilayer phases at high concentrations (Mishra et al., 1994). The MLT and 18A results together thus raise the
possibility that very small changes in
ZBr and
ABr may be peculiar to amphipathic
helices at bilayer interfaces, at least a low concentrations.
|
Figure 3 B shows the notably small effect of 5 mol%
Ac-18A-NH2, corresponding to a unit-cell volume
fraction of 10.6%. The surprising noninvasive behavior of monomeric
amphipathic helices becomes apparent when the helix-induced
perturbation of the double-bond distribution is compared to the changes
induced by two, small unstructured peptides: the pentapeptide (Wimley
and White, 1996) AcWLWLL (10 mol%; 6.7% unit-cell volume-fraction)
and the 13-residue antimicrobial peptide indolicidin (Selsted et al.,
1992; Ladokhin et al., 1997b, 1999) (5 mol%; 7.6% unit-cell
volume-fraction). The most significant features of the perturbations in
these two cases are the broadening and the substantial shift of the
double-bond distribution toward the bilayer center (Fig. 3,
C and D). These bilayer perturbations are so
large that the construction of models from the component distributions
of the neat DOPC bilayer are precluded. It is thus clear that, at low
concentrations, amphipathic helices that have the ability to solubilize
bilayers at high concentrations fit nicely into the bilayer interface
with few bilayer perturbations beyond those expected from slight
increases in the area per lipid.
Bilayer perturbations due to dimeric mellittin
OCD measurements of dimeric Q25C in DOPC bilayers gave spectra virtually identical with those of MLT under otherwise identical conditions, as shown in Fig. 4. Q25C therefore had high helicity and was oriented parallel to the bilayer plane. Nevertheless, the structure of DOPC/Q25C bilayers is quite different from DOPC/MLT bilayers. The profiles for DOPC with and without dimeric Q25C are shown in Fig. 1 B (red and blue curves, respectively). The effect of Q25C on the double-bond distribution as reported by the OBPC bromine labels is shown in Fig. 3 A (red curve). At the same monomer-per-lipid concentration used for monomeric MLT, the Q25C dimer caused a significant increase in ABr compared to the monomer. Table 1 shows that the values of ZBr and ABr are statistically different for the monomer and dimer, even though not visually striking in Fig. 3 A. Although the shift in ZBr toward the bilayer center was smaller for Q25C than for MLT, the overall perturbation was quite large. The Q25C perturbation is so large that it entirely precludes successful modeling of the perturbed bilayer. That is, no simple rescaling procedure ever resulted in models whose structure factors were within experimental uncertainty. More visually revealing is the dramatic change in the bilayer profile near the center of the bilayer. A comparison of Fig. 1, A and B shows that the Q25C perturbations affect the entire bilayer rather than just the double-bond distribution. Differences in the bilayer profiles caused by added peptide arise from both the added scattering density of the peptide and the scattering-density changes of the bilayer due to lipid rearrangements. Studies using specific deuteration and neutron diffraction will be necessary to understand the nature of the perturbations. We expect the perturbations to depend strongly on concentration. The results reported here for MLT and Q25C at low concentrations provide a reference point for such measurements.
|
| |
CONCLUSIONS |
|---|
|
TOP
ABSTRACT
INTRODUCTION
MATERIALS AND METHODS
RESULTS AND DISCUSSION
CONCLUSIONS
APPENDIX
REFERENCES
|
|---|
Our results show that MLT in fluid bilayers is unlikely to prefer
the conformation observed in the crystal structure. One model that is
consistent with our data and NMR data (Dempsey and Butler, 1992; Okada
et al., 1994) is a generally helical conformation encompassing the
first 22 residues and an extended conformation running roughly parallel
to the bilayer plane for the last four residues. Of course, MLT must
adopt many different conformations represented as an ensemble average
in the diffraction experiment. The image obtained in our x-ray
experiments is one-dimensional, along the bilayer normal, and thus does
not contain information about the organization of the peptides in the
bilayer plane, especially the nature of the association of the monomers
comprising Q25C. Also, without specific heavy-atom labeling of the
peptide, we can say nothing about the relative depths of different MLT
residues in the bilayer, because orientations obtained via rotation of the molecule around the helix axis are indistinguishable. In the model
of Dempsey and Butler (1992), Pro14 is positioned
roughly at the same depth in the bilayer as the helix axis. A kink may
exist in the plane of the bilayer, but an orientation of the kink-plane
normal to the bilayer surface (Fig. 3 B) is extremely unlikely.
Although the disposition of MLT can change with hydration, monomeric
MLT in fully hydrated bilayers is parallel to the bilayer plane
(Altenbach and Hubbell, 1988; Altenbach et al., 1989; Frey and Tamm,
1991; Dempsey and Butler, 1992) as observed here for lower hydrations.
We have already shown that the structure of the pure DOPC bilayer does
not change substantially over a wide range of hydrations (Hristova and
White, 1998). The results reported here are thus very likely to be true
at full hydration, especially because the conformation of MLT is
consistent with that determined by NMR at full hydration. They also
demonstrate that absolute-scale lamellar diffraction is an important
tool for studying the disposition of peptides in bilayers.
Considering the highly destabilizing effect of amphipathic helices at
high concentrations, we expected both MLT and
Ac-18A-NH2 to have a far greater effect on
bilayer structure, even at the low concentrations used here. The
changes in bilayer structure seen, however, are stunningly small. This
observation alone strongly implicates helix association as a necessary
precursor to bilayer destabilization. The increased bilayer
perturbation of Q25C at low concentration is apparently a harbinger of
destabilization. As MLT or Q25C concentrations are increased, we expect
to observe major changes in bilayer organization and, ultimately,
reorientation of the helices into a transmembrane configuration.
Indeed, early neutron diffraction studies (Strom et al., 1983; Bradshaw
et al., 1994) on MLT at concentrations of 5-10 mol% suggested such an orientation, and more recent NMR studies (Smith et al., 1994; Naito et
al., 2000) have proven it. In the course of the present experiments, we
found by OCD that MLT began reorienting at ~4 mol%. Crucial
questions that remain to be answered are how the lipids adapt to
helices as peptide concentration increases and how these adaptations
may lead toward destabilization of the membrane-parallel orientation of
helices, and ultimately to complete destabilization of the bilayer at
high peptide concentrations.
Besides being important in pore formation, helix association also
appears to be important in membrane fusion. This is because the helical
bundle has emerged as the principal motif of fusion peptides. These
bundles are generally envisioned as having a membrane-parallel surface-bound state near the time of the membrane-destabilizing fusion
event (Skehel and Wiley, 1998). A general principle arising from
studies of both MLT and fusion peptides is that helix multimers have a
destabilizing effect on membrane integrity. The structural approach
used here may thus provide important insights into membrane fusion when
applied to helical-bundle fusion peptides.
| |
APPENDIX: THE PER-LIPID ABSOLUTE SCALE |
|---|
|
TOP
ABSTRACT
INTRODUCTION
MATERIALS AND METHODS
RESULTS AND DISCUSSION
CONCLUSIONS
APPENDIX
REFERENCES
|
|---|
An x-ray profile on an absolute scale is a measure of electron
density (number n per unit volume,
n/cm3) as a function of position in
the unit cell. For small-angle diffraction, the "strength" of the
scattering of X-rays by electrons is measured by the scattering length
b (cm), which is simply proportional to n:
b = n · mc2/e2,
where c is the speed of light, m is electron
mass, and e is electron charge (Woolfson, 1970). One can
thus also describe x-ray profiles in terms of scattering-length density
, which has units of b/cm3 = cm2. As noted in
Materials and Methods, we have historically used scattering-length
density rather than electron density because it is more convenient when
combining x-ray and neutron diffraction data.
On the true absolute scale, the scattering-length density is given by
|
(A1) |
0 is the average
scattering-length density of the unit cell given by
|
(A2) |
*(z) = (z) · S,
which is dimensionless (see Jacobs and White, 1989).
The per-lipid scale has two advantages. First, the experimental
uncertainties of per-lipid scale profiles are reduced compared to
profiles on the true absolute scale because the uncertainties of
S are avoided. Inclusion of the uncertainty of S
will significantly broaden the error bands shown in Fig.
1 A, causing the position of the peptide to be less
certain. Second, *(z) is more sensitive to the presence
of the peptide than (z); its amplitude is determined almost entirely by the number of electrons in the unit cell, rather than the number per unit volume. This is true because the Bragg spacing
typically changes by only about 2% upon the addition of peptide (Table
1), causing *(z) ~ constant × buc. This can be seen more clearly by
considering the effect of MLT in our experiments.
Based upon partial specific volumes of MLT, computed from the data of
Makhatadze et al. (1990), and the hydrated phosphocholine group (Wiener
and White, 1992), the electron densities are (1536/3736 Å3) = 0.41 e/Å3 and (218/498
Å3) = 0.44 e/Å3, respectively. This is
consistent with the findings of MacNaughtan et al. (1985), who observed
a drop in electron density relative to the lipid headgroups for
cytochrome c bound to the surface of cerebroside sulfate/cholesterol
bilayers. It is apparent that 1 mol% of MLT in the DOPC headgroup
region would be difficult, if not impossible, to detect using an
e/Å3 scale. The situation for the
per-lipid scale is dramatically different. The addition of 0.01 MLT
increases the number of electrons in the headgroup region from 218 to
233, an increase of ~7%. Our results show that this increase is
easily detected.
In earlier papers from our laboratory, we used the nomenclature
"Scattering Density: (z) · S · 104" to describe scattering density on
bilayer-profile plots. But we now use "Scattering Density per
Lipid × 104," introduced in our paper
concerned with the disposition of the 18A peptide in lipid bilayers
(Hristova et al., 1999). The factor of 104
applied to a dimensionless quantity at first sight seems meaningless. Here is the explanation. We use Ångströms for lengths and square Ångströms for areas, so that, in centimeters,
0 · S is on the order of
1012/108 = 104 (dimensionless).
Thus, the per-lipid scattering density values shown in our plots are
multiplied by 104 to indicate that the numbers
read from the ordinate should be multiplied by the reader by
104 for use in
computations. Because the values of the structure factors we report
take these issues into account, we include this factor of
104 in bilayer profiles.
The need for the 104
factor is also apparent from the scattering length conversion factor
mc2/e2
(above), which has numerical value 0.282 × 1012 cm, or 0.282 × 104 Å. From Eq. A2, for
a unit cell measured in Å containing
nuc electrons, the scattering density
per lipid is
nuc(mc2/e2)/d
or
0.282(nuc/d) × 104.
| |
ACKNOWLEDGMENTS |
|---|
This research was supported in part by a grant from the National Institutes of Health (GM-46823). We thank Drs. William Wimley and Alexey Ladokhin for many helpful conversations.
| |
FOOTNOTES |
|---|
Received for publication 14 July 2000 and in final form 17 November 2000.
Address reprint requests to Stephen H. White, Department of Physiology and Biophysics, Med. Sci. I, D346, University of California at Irvine, Irvine, CA 92697-4560. Tel.: 949-824-7122; Fax: 949-824-8540; E-mail: shwhite{at}uci.edu.
| |
Abbreviations used: |
|---|
Abbreviations used: MLT, melittin: GIGAVLKVLTTLPALISWIKRKRQ25Q-amide; AC-18A-NH2, acetyl-DWLKAFYDKVAEKLKEAF-amide; Q25C, a disulfide-linked dimer of MLT Cys-substituted at position 25 (GIGAVLKVLTTLPALISWIKRKRC25Q-amide).
| |
REFERENCES |
|---|
|
TOP
ABSTRACT
INTRODUCTION
MATERIALS AND METHODS
RESULTS AND DISCUSSION
CONCLUSIONS
APPENDIX
REFERENCES
|
|---|
-helix at a membrane interface: a structural study using a novel x-ray diffraction method.
J. Mol. Biol.
290:99-117-helices on membranes: energetics of helix formation by melittin.
J. Mol. Biol.
285:1363-1369
