2012 Biophysical Journal Volume 109 November 2015 2012–2022

Computational Tool

CHARMM-GUI HMMM Builder for Membrane Simulations with the HighlyMobile Membrane-Mimetic Model

Yifei Qi,1 Xi Cheng,1 Jumin Lee,1 Josh V. Vermaas,2,4,7 Taras V. Pogorelov,3,5,6,7 Emad Tajkhorshid,2,4,7

Soohyung Park,1 Jeffery B. Klauda,8 and Wonpil Im1,*1Department of Molecular Biosciences and Center for Computational Biology, The University of Kansas, Lawrence, Kansas; 2Department ofBiochemistry, 3Department of Chemistry, 4Center for Biophysics and Computational Biology, 5School of Chemical Sciences, 6National Centerfor Supercomputing Applications, and 7Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign,Urbana, Illinois; and 8Department of Chemical and Biomolecular Engineering and the Biophysics Program, The University of Maryland, CollegePark, Maryland

ABSTRACT Slow diffusion of the lipids in conventional all-atom simulations of membrane systems makes it difficult to samplelarge rearrangements of lipids and protein-lipid interactions. Recently, Tajkhorshid and co-workers developed the highly mobilemembrane-mimetic (HMMM) model with accelerated lipid motion by replacing the lipid tails with small organic molecules. TheHMMMmodel provides accelerated lipid diffusion by one to two orders of magnitude, and is particularly useful in studying mem-brane-protein associations. However, building an HMMM simulation system is not easy, as it requires sophisticated treatment ofthe lipid tails. In this study, we have developed CHARMM-GUI HMMM Builder (http://www.charmm-gui.org/input/hmmm) to pro-vide users with ready-to-go input files for simulating HMMM membrane systems with/without proteins. Various lipid-only andprotein-lipid systems are simulated to validate the qualities of the systems generated by HMMM Builder with focus on the basicproperties and advantages of the HMMM model. HMMM Builder supports all lipid types available in CHARMM-GUI and alsoprovides a module to convert back and forth between an HMMM membrane and a full-length membrane. We expect HMMMBuilder to be a useful tool in studying membrane systems with enhanced lipid diffusion.

INTRODUCTION

One of the challenging aspects in molecular simulations oflipid membranes is predicting how chemical compoundsand macromolecules (peptides, proteins, and polymers)bind to and insert into these membranes and how themembranes respond to the binding (1–3). Even for shortpeptides, studying their interactions with membranes canbe challenging (2) due to the relatively slow diffusion(~10�7 cm2/s) of glycerophospholipids (4,5). As a result,lipid mixing and membrane-induced peptide responsesoccur over long timescales. Peptide secondary structurescan be influenced by the membrane environment, as exem-plified by the folding and unfolding of peptides at the mem-brane interface by antimicrobial peptides (6) or intrinsicallydisordered proteins like a-synuclein (7). Long timescalesare also needed for antimicrobial polymers that disturb thelipid membrane (1).

The long timescales required present unique challenges instudying both the folding and insertion of transmembraneand peripheral membrane proteins using traditional molecu-lar dynamics (MD) simulations (8–11). Since membranebinding of these proteins occurs on microsecond or longertimescales, the access to specialized high-performance com-puters, such as the Anton supercomputer (12), is one

Submitted August 5, 2015, and accepted for publication October 9, 2015.

*Correspondence: wonpil@ku.edu

Editor: Scott Feller.

� 2015 by the Biophysical Society

0006-3495/15/11/2012/11

approach that permits detailed studies of these proteins.For example, binding of the oxysterol binding protein ofyeast (Osh4) to model membranes on a microsecond time-scale was studied by MD simulation using Anton (11).Although this is useful and serves as a standard to comparealternative methods, limited access to such resources makesit intractable to fully probe many different protein-mem-brane systems and conditions. Most MD simulation studiesof membrane anchoring domains have starting models thatare fully or partially embedded and thus require a prioriknowledge of approximate membrane binding regions andtheir orientation and insertion depth with respect to themembrane (8).

To overcome the timescale and sampling issues, alterna-tive methods have been used to improve sampling beyondthe capability of unbiased standard MD. Replica-exchangeMD (REMD) (13) has been extensively used for studyingpeptide folding in a water solvent (14). The key to thismethod is using replicas at different temperatures to ex-change conformations and probe peptide conformations.Temperature replicas must be used with caution in mem-brane systems, as membranes undergo phase transitionsnear physiological conditions and are only stable over asmall range of temperatures. Therefore, more recently, theuse of a random walk in surface tension (not temperature)has been shown to enhance sampling of peptides in thebilayer while maintaining membrane structure (15). Other

http://dx.doi.org/10.1016/j.bpj.2015.10.008

CHARMM-GUI HMMM Builder 2013

methods, such as accelerated MD (AMD) (16) and replicaexchange with solute tempering (REST) (17), have alsoshown promise in accelerating timescales in membrane sys-tems. Accelerated MD adds a boost potential to flatten theenergy surface and allow for easy barrier crossing (16).This has been successfully applied to membranes with aspeedup of two- to threefold compared to unbiased MD(18). REST is used for membranes with tempering of thelipid, and the solvent temperature is fixed. This methodwas shown to increase the diffusion rate of lipids by an orderof magnitude compared to unbiased MD (17).

Another approach is to study a model system where thecomposition of the membrane accelerates lipid sampling.The highly mobile membrane-mimetic (HMMM) model(19) utilizes a biphasic setup (8) by using HMMM lipidswith truncated acyl chains (generally three to five acylcarbons) and an organic solvent to represent the hydropho-bic membrane core. The solvent used to date, 1,1-dichloro-ethane (DCLE), is liquid at standard temperatures andpressures, hydrophobic enough to mimic the membraneinterior, and also has a small molecular weight that maxi-mizes its diffusion constant (20). This made DCLE a naturalchoice among many other alternative solvents tested for thebiphasic model that evolved into the HMMM model (8).

Free-energy calculations have shown that this modelaccurately represents interactions between the membraneand peripheral membrane proteins (21). The main advantageof the HMMM model is that the lipid diffusion is enhancedby one to two orders of magnitude, depending on the densityof the HMMM lipids at the interface (19). Similar to sur-face-tension-based REMD, the increased surface tension re-sults in a decrease in lipid packing and an increase in lipiddiffusion (15). However, the HMMM model does notrequire use of REMD but instead uses unbiased conven-tional MD, allowing for accelerated sampling. One of themain applications of the HMMM model is to study how pe-ripheral membrane proteins or their membrane-binding do-mains bind to the membrane interface (20). For example,binding of the g-carboxyglutamic-acid (GLA)-rich domainto an HMMM dioleoyl-phosphatidylcholine (DOPC) mem-brane was found to occur in 50-ns MD simulations (19),where the membrane simulations of full-length (not trun-cated) lipids required substantially longer timescales evenwith an applied force (22). The HMMM model has beenused to explore a wide variety of membrane-associatedproblems that require enhancement in lipid diffusion (20),e.g., peripheral protein membrane anchoring (23), TM helixorientation (24), and protein (25–29) and lipid (30) insertioninto biological membranes.

In this work, to facilitate wider use of the HMMMmodel,we describe HMMM Builder, a new tool developedin CHARMM-GUI (http://www.charmm-gui.org) (31).CHARMM-GUI provides a web-based graphical user inter-face to generate various molecular simulation systems andinput files (for CHARMM (32), NAMD (33), GROMACS

(34), AMBER (35), OpenMM (36), and CHARMM/OpenMM) to facilitate and standardize the usage ofcommon and advanced simulation techniques (37–39).Within the framework of Membrane Builder (37,40,41) inCHARMM-GUI, HMMM Builder (http://www.charmm-gui.org/input/hmmm) facilitates easy system building withthe HMMM lipids by automatic conversion of all full-lengthlipid types available in Membrane Builder to the corre-sponding HMMM lipids. The input files generated are readyfor use with CHARMM, NAMD, and GROMACS.

In the following, we describe the setup of HMMMBuilder and the steps required by users to generate theirown systems in detail. To illustrate that HMMM Builderis robust and well suited for various types of membranesimulations, we provide results for a series of example sys-tems, including single-lipid and mixed-lipid bilayers andpeptide-membrane systems.

MATERIALS AND METHODS

System building

The construction of membrane systems in HMMM Builder follows the five

general steps in CHARMM-GUI Membrane Builder (31,37,40,41). In

step 1, a protein structure (for protein-membrane systems) can be read in

through Protein Data Bank (PDB) Reader (42) either by uploading a

PDB structure file containing single or multiple chains or by typing a

PDB ID from the PDB (43)/Orientations of Proteins in Membranes

(OPM) (44) database. In step 2, if necessary, the protein can be aligned

to and/or translated along the z axis (by definition, a bilayer is centered

at z ¼ 0), and pore water can be generated in the case of a protein with a

channel or more extensive lumen. In step 3, the system size is determined

based on the user-specific input, and lipid-like pseudoatoms are packed to

determine the initial positions of lipid headgroups in the xy plane. In this

step, as the HMMM-related options, a user can choose the position to

trim an acyl chain (the default is to use acyl carbon number 6 as the terminal

carbon, and sterol tails are not trimmed) and the per-lipid area ratio of an

HMMM lipid with respect to the corresponding full-length lipid (the default

is 1.1). In step 4, system components including full-length lipids, ions, and

water box are generated. A DCLE box is also generated by duplicating a

preequilibrated cubic DCLE box whose density is 0.0073 molecule/A3.

In step 5, the acyl tails of full-length lipids are trimmed based on the option

in step 3; an example of a full-length and a trimmed palmitoyl-oleoyl-phos-

phatidylcholine (POPC) is shown in Fig. S1 in the SupportingMaterial. The

DCLE box from step 4 is overlaid on the trimmed membrane, and the over-

lapping DCLE molecules are removed. All the components are assembled

into a final system, and the simulation input files for NPT (constant particle

number, pressure, and temperature), NPAT (constant particle number, pres-

sure, surface area, and temperature), or NPgT (constant particle number,

pressure, surface tension, and temperature) ensembles are generated.

Conversion between HMMM and full-lengthmembranes

In addition to HMMM system building, HMMM Builder supports conver-

sion between HMMM and full-length lipid systems. The full-to-HMMM

conversion is identical to the procedures to build HMMM systems. How-

ever, HMMM-to-full is more complicated because HMMM lipid tails are

more flexible than the same atoms in full-length lipids. To obviate this dif-

ficulty, before the conversion to a full lipid, the position of the last carbon

atom in an HMMM lipid is changed to align the last carbon-carbon vector to

Biophysical Journal 109(10) 2012–2022

2014 Qi et al.

the z axis. The missing coordinates in the converted full-length lipids are

then copied from a library of lipid structures with an additional check to

avoid bad contacts and penetration of lipid tails into ring structures in lipids

and proteins.

Simulation protocol

All equilibration and production simulations were performed using NAMD

2.9 (33) with the NPT ensemble and the CHARMM36 force field (45–48).

The van der Waals interactions were smoothly switched off at 10–12 A by a

force-based switching function (49). The particle-mesh Ewald algorithm

was used to calculate electrostatic forces (50). A time step of 2 fs was

used in all simulations. Temperature was maintained at 310 K using Lange-

vin dynamics with a coupling coefficient of 1 ps�1. The Nose-Hoover Lan-

gevin-piston method (51,52) was used to maintain constant pressure at 1 bar

with a piston period of 50 fs and a piston decay of 25 fs. To limit fluctuation

of short-tailed lipids along the membrane normal and to prevent their occa-

sional diffusion into the bulk solution (i.e., surfactant-like behavior), the

HMMM lipids were restrained using weak harmonic potentials (with a

force constant of 0.5 kcal/mol/A2 for systems with cholesterols (CHOLs)

and 0.05 kcal/mol/A2 for CHOL-free systems) applied to the z-positions

of C21 and C31 (carbonyl carbon) atoms (even for production) (20). In

addition, several restraints were applied to lipid tails, water, and DCLE

molecules during equilibrations (see Table S1).

Trajectory analysis

Trajectory analysis was performed with CHARMM (32), VMD (53), and

MDAnalysis (54). The area per lipid was calculated using the Voronoi dia-

gram (see Table S2 for the atoms used to represent each lipid). The diffusion

constant was calculated from the slope of the lateral mean-square displace-

ment (MSD),

MSDðtÞ ¼ 4Dt ¼ � jrðt þ t0Þ � rðt0Þj2�; (1)

whereD is the diffusion constant, r is the x-y positional vectors, t is the time

interval ranging from 0 to 0.8 � tsim, t0 is the starting point for MSD calcu-

lations, and tsim is the total simulation time. The MSD value was averaged

over all lipids from both leaflets with corrections to remove the leaflet drift.

The lateral pressure profile from NAMD output was extracted with the

Harasima contour (55), in which the normal component, pNðzÞ, in the pres-

sure, pðzÞ ¼ pLðzÞ � pNðzÞ, cannot be obtained (56). Because the bilayer

systems in this study are in equilibrium, the normal component, pNðzÞ,was calculated from

pN ¼ L�1z

ZLz=2

�Lz=2

pLðzÞ dz; (2)

where Lz is the box size in the z direction. The first moment of the pressure

profile for each leaflet was calculated as

F0ð0Þupper ¼ dF

dR�1

����R�1 ¼ 0

¼ �ZN

0

z pðzÞ dz (3)

Z0

F0ð0Þlower ¼

�N

z pðzÞ dz; (4)

where F is the free energy of the membrane per unit of lipid area, R�1 is the

membrane curvature, and p(z) is the lateral pressure profile. The first

moment of the pressure profile for the whole bilayer was calculated as

Biophysical Journal 109(10) 2012–2022

F0ð0Þbilayer ¼ F

0ð0Þupper � F0ð0Þlower: (5)

The mean residence time in the mixed-lipid systems was calculated ac-

cording to the method described in Impey et al. (57). For two lipid types A

and B, whose lipid numbers are NA and NB, a contact time series between

lipid i from type A and lipid j from type B, pijðt0; t0 þ t; t�Þ, is set to 1 if

these two lipids are in contact at time t0 and t0 þ t and are separated for no

longer than the tolerance time, t* (20 ps in the current study), and 0 other-

wise. Here, two lipids are in contact if the minimal distance between the

heavy atoms of these lipids is <4.5 A. To make a fair comparison between

HMMM and full-length lipids, only the heavy atoms shared by the

HMMM and full-length lipids were used in the contact calculations.

The average number of contacts between type A and type B as a function

of the time interval, t, is

nðtÞ ¼ 1

tmax � t

Xtmax�t

t0 ¼ 1

XNA

i¼ 1

XNB

j¼ 1

pijðt0; t0 þ t; t�Þ; (6)

where tmax is the total simulation time. When t0 ¼ 0, n(0) is the average

number of contacts between types A and B during the simulation. The

mean residence time of two lipid types can be estimated by calculating

n(t)/n(0) from the trajectory and fit the data to an exponential decay func-

tion, exp(�t/t), where t is the mean residence time. In this study, we fitted

n(t)/n(0) up to t ¼ 10 ns for HMMM lipids and 30 ns for full-length lipids.

The fitting is an approximate method, but it is enough for the purpose of

comparing the behavior of HMMM and full-length lipids. The mean and

standard errors were obtained by dividing the whole trajectory into five

segments and repeating the calculation on each segment. We also calcu-

lated the contact time series, pijðt0; t0 þ t; t�Þ, based on the neighboring

pairs in the Voronoi diagram, but the main conclusion did not change

(data not shown).

RESULTS AND DISCUSSION

To illustrate membrane simulations with HMMM Builder,we tested several systems including three single-lipid sys-tems, four mixed-lipid systems, and two peptide-membranesystems (Table 1). Replacing the lipid tails with DCLE mol-ecules in the membrane hydrophobic core makes the mem-brane more fluid. For a given number of lipids, however, it isnot easy to determine the optimal number of DCLE mole-cules, because the system behavior depends not only onlipid type, but also on other factors, such as temperatureand lipid chain saturation. To consider these facts,HMMM Builder allows the user to change the per-lipidarea ratio (RSA) of the HMMM lipids relative to the corre-sponding full-length lipids. As the thickness of the DCLElayer is roughly constant, varying the per-lipid area ratiois equivalent to varying the volume that DCLE moleculeswill occupy in the final system, thereby changing the num-ber of DCLE molecules in the system. In other words, if RSA

> 1, the initial system will have a larger xy system size (thanthe system based on the area per full-length lipid) and thusmore DCLE molecules will be filled in the membranehydrophobic core, likely leading to faster lipid diff-usion. To show the effect of using different RSA valuesand to determine a reasonable default value, we tested six

TABLE 1 Lipid components and scaling factors of the example systems

Number of Lipids in Each Leaflet RSA

Single-Lipid System

POPC 64 POPC 0.8, 0.9, 1.0, 1.1, 1.2, 1.3

POPE 64 POPE 0.8, 0.9, 1.0, 1.1, 1.2, 1.3

POPS 64 POPS 0.8, 0.9, 1.0, 1.1, 1.2, 1.3

Mixed-Lipid System

MIX1 40 POPC/40 POPE/20 POPS 0.8, 0.9, 1.0, 1.1, 1.2, 1.3

MIX2 41 POPC/41 POPE/16 POPS/1 POPI24/1 POPI25 0.8, 0.9, 1.0, 1.1, 1.2, 1.3

MIX3 40 POPC/40 POPE/20 TOCL2 0.8, 0.9, 1.0, 1.1, 1.2, 1.3

MIX4 34 POPC/33 PSM/33 CHOL 0.8, 0.9, 1.0, 1.1, 1.2, 1.3

MARCKS-ED Peptide Systems

MARCKS1 60 POPC/60 POPE/30 POPS 1.1

MARCKS2 61 POPC/61 POPE/24 POPS/2 POPI24/2 POPI25 1.1

POPC, palmitoyl-oleoyl-phosphatidylcholine; POPE, palmitoyl-oleoyl-phosphatidylethanolamine; POPS, palmitoyl-oleoyl-phosphatidylserine; POPI24,

palmitoyl-oleoyl-phosphatidylinositol-(4,5)-bisphosphate with protonation on P4; POPI25, palmitoyl-oleoyl-phosphatidylinositol-(4,5)-bisphosphate with

protonation on P5; TOCL2, tetraoleoyl cardiolipin; PSM, n-palmitoyl sphingomyelin; CHOL, cholesterol;

CHARMM-GUI HMMM Builder 2015

different RSA values from 0.8 to 1.3 for the lipid-only sys-tems (Table 1).

Single-lipid systems

The three single-lipid bilayers (POPC, POPE, and POPS)were simulated for 100 ns to investigate the basic propertiesof HMMM membranes. As a control, we also simulated thecorresponding full-length lipid systems for 150 ns. The per-lipid area of the HMMM lipid is larger than the area of thecorresponding full-length lipid in all three bilayer systems(Fig. 1 A), which is accompanied by a decrease in theHMMM bilayer thickness (compared to the thickness ofthe full-length membranes) at a given RSA (Figs. 1 B andS2). As a balance between per-lipid area and bilayer thick-ness, we use RSA ¼ 1.1 as the default value in HMMMBuilder; this value can be changed by the user in step 3(see Materials and Methods). The HMMM lipids are ex-pected to diffuse much faster than full-length lipids. To es-timate the diffusion constant, we performed linear fitting ofthe mean-square displacement as a function of time interval(Fig. S3). The HMMM lipids show a diffusion constant be-tween 2 � 10�7 and 10 � 10�7 cm2/s, which is ~2–10 timesfaster than the diffusion constants of the full-length lipids,

A B

FIGURE 1 Single-lipid HMMM membrane properties: (A) surface area per

and POPS HMMM membranes as a function of the per-lipid area ratio (RSA)

are presented as the ratio of an HMMM membrane to a full-length membrane,

color, go online.

which are 1.02 � 10�7 (POPC), 0.43 � 10�7 (POPE), and0.32 � 10�7 (POPS) cm2/s (Fig. 1 C). In general, theHMMM lipids diffuse faster at larger RSA, which is due toless packing between the lipids at larger RSA.

As the HMMM membrane expands in the xy plane andshrinks along the z axis, it is important to make sure thatthe DCLE density is still reasonable. In all three systems,the DCLE densities at the bilayer center are very close toeach other, regardless of RSA (Fig. 2). The density is~0.0068 molecule/A3, somewhat lower than the initial den-sity (0.0073 molecule/A3) in the DCLE box to build the sys-tem. The thickness of the DCLE layer at the half-maximumdensity increases with increasing RSA, in accordance withthe change in bilayer thickness (Fig. S4 A). The numberof DCLE molecules per trimmed acyl tail carbon atomranges from 0.15 to 0.25 at different RSA values(Fig. S4 B). When RSA is ~1.0–1.1, the number is close tothe value used in a previous study, in which ~14 acyl tailcarbons are replaced with three DCLE molecules(i.e., 0.21 DCLE molecules per trimmed carbon atom)(20). In addition, ~1% of the DCLE molecules escaped tothe bulk during the simulation, which we believe is negli-gible and is due to the natural partition coefficient ofDCLE between lipid and solution.

C

lipid, (B) bilayer thickness, and (C) diffusion constant of POPC, POPE,

and in the full-length membrane. In (A) and (B), the area and thickness

with the standard deviations shown by the error bars. To see this figure in

Biophysical Journal 109(10) 2012–2022

A B C

FIGURE 2 Density profiles of DCLE molecules in pure (A) POPC, (B) POPE, and (C) POPS HMMMmembranes with different RSA. To see this figure in

color, go online.

2016 Qi et al.

Lateral pressure in membranes plays an important role inthe regulation of protein function (58,59). Due to the liquidnature of DCLE, an HMMM membrane is not expected tohave the same pressure profile as in a full-lipid membrane.To quantify the difference, we calculated the pressure pro-files of the HMMM and full-length POPC, POPE, andPOPS membranes (Fig. 3). All HMMM membranes featurea profile similar to that of the full-length membrane outsidethe DCLE layer, and a neutral pressure at the membranecenter (due to the liquid nature of DCLE). In contrast, thefull-length membranes show a positive peak at the mem-brane center, which arises from the repulsive interactionsbetween the lipid tails. This behavior is expected from thenature of HMMM, where the small DCLE molecules actas a bulk organic solvent that would have a vanishing shearmodulus (analogous to bulk water). In HMMM membranes,the positive pressure at the lipid headgroup and tail interface(~z¼512 A) is higher than the pressure in the correspond-ing full-length membranes, which compensates for theneutral pressure at the membrane center and keeps the inte-gration of the lateral pressure profile along the z axis at zero(i.e., zero surface tension) under the NPT condition.

The first moment of the pressure profile, F0ð0Þ, provides

the free-energy derivative at planar curvature (60,61), witha positive/negative value indicating a preference to bendtoward the headgroup/tail direction. In all HMMM systems,the monolayers (upper leaflet or lower leaflet) show a posi-tive F

0ð0Þ around 0.1 at RSA ¼ 0.8, suggesting that the lipids

A B

FIGURE 3 Pressure profiles of HMMM and full-length (A) POPC, (B) POPE,

not shown. To see this figure in color, go online.

Biophysical Journal 109(10) 2012–2022

are packed too compactly and the monolayer would bend inthe headgroup direction (i.e., positive curvature) at this RSA

(Fig. S5). With increasing RSA, the F0ð0Þ of the monolayer

approaches zero in POPC and POPE but remains negativein POPS, which likely arises from the intrinsic curvaturepreference of the POPS lipid. Not surprisingly, the F

0ð0Þof the whole bilayer is close to zero in all HMMM andfull-length membranes due to the imposed symmetry ofthe membrane and the z-position restraints in the HMMMmembrane. Although the full-length membrane’s F

0ð0Þvalues are not reproduced exactly in the HMMM mem-branes, our results suggest that using RSA > 1.0 is stillreasonable for NPT simulations.

Mixed-lipid systems

HMMM Builder supports all lipid types available inCHARMM-GUI. To demonstrate the simulations ofdifferent lipid mixtures, we tested four mixed-lipid systems(Table 1). The per-lipid areas in systems MIX1–3 are largerthan the full-lipid areas (Fig. S6), similar to the results forsingle-lipid systems. However, in MIX4, POPC, PSM, andCHOL have smaller areas than the full-length lipids, withRSA values around 0.9 (Fig. S6). A possible cause is thefavorable interactions between PSM and CHOL makingthem pack more compactly in the HMMM membrane.The HMMM lipids in MIX1–3 show a speedup in diffusionsimilar to that observed in the single-lipid systems (Figs. 4 A

C

and (C) POPS bilayers. For clarity, the profiles at RSA ¼ 0.8, 1.0, and 1.2 are

A B C

FIGURE 4 Mixed-lipid HMMMmembrane properties: (A) diffusion constant, (B) mean residence time, and (C) pressure profile at different RSA. In MIX2,

the residence times between PIP2 and other lipid types are not calculated due to lack of data. To see this figure in color, go online.

CHARMM-GUI HMMM Builder 2017

and S7). In MIX2, POPI24 and POPI25 show a large varia-tion in diffusion constant, because only two POPI24/POPI25 molecules were present in each leaflet, so that thedata obtained in the simulations are not sufficient to makea good estimate. In MIX4, due to the decreased fluiditycaused by CHOL, POPC diffuses more slowly than it doesin CHOL-free HMMM systems (MIX1-3), but still muchfaster than the full-length POPC lipids in the control systemwith the same lipid composition (Fig. S7). To compare thelipid mixing in HMMM and full-length membranes, wecalculated the mean residence time, t, between lipid types,which characterizes the persistence of one lipid type in con-tact with another one (see Materials and Methods). In allfour systems, the full-length lipids have larger t and thusslower mixing than the corresponding HMMM lipids(Figs. 4 B and S8). Moreover, the HMMM lipids show adecrease of t at large RSA, which is in agreement with thefaster diffusion of the HMMM lipids at larger RSA.

The pressure profiles in systemsMIX1–4 are similar to theprofiles in the single-lipid systems, all showing a neutral pres-sure at the HMMMmembrane center and a positive pressurepeak at the full-lipid membrane center (Fig. 4 C). In MIX4,the pressure in the HMMM membrane is larger than that inthe full-lengthmembrane, which is likely due tomore rigiditycaused by the large restraint force constant (0.5 kcal/mol/A2

inMIX4 and 0.05 kcal/mol/A2 inMIX1–3; seeMaterials andMethods) on the lipid carbonyl atoms in the HMMM mem-brane. Due to this large restraint force, the membrane thick-ness in MIX4 does not change with respect to RSA and themembrane is thinner than the corresponding full-lengthmem-brane at all RSAvalues (Fig. S9A). The CHOL in the HMMMmembrane has slightly wider tilt-angle distributions thanCHOL in the full-length membrane due to the fluidity ofthe DCLE molecules (Fig. S9 B). However, the densities ofthe CHOL heavy atoms and O3 atoms match those in thefull-length membrane well (Fig. S9, C and D).

Biophysical Journal 109(10) 2012–2022

2018 Qi et al.

As expected, the first moment of the pressure profiles forthe whole bilayer is close to zero in all mixed-lipid systems(Fig. S10). The monolayer F

0ð0Þ decreases as RSA increases,similar to the trend observed in the pure-lipid systems. InMIX4, the full-length F

0ð0Þ values for the monolayers donot agree with each other, which is likely due to insufficientsampling caused by the slow diffusion in this system.

MARCKS-ED membrane association

As a practical application of HMMM Builder, we conduct-ed a simulation of peptide-membrane association. The 25-amino-acid peptide derived from the effector domain ofmyristoylated alanine-rich C kinase substrate protein(MARKCS-ED) was chosen as a model system (62). Thispeptide is enriched with both basic and hydrophobic resi-dues (sequence: KKKKK RFSFK KSFKL SGFSFKKNKK), and is able to sequester anionic lipids such asPIP2 through nonspecific electrostatic interactions(63,64). To compare the effect of PIP2 lipids, two simula-tion systems were set up: system MARCKS1, consistingof POPC, POPE, and POPS, and system MARCKS2, con-sisting of POPC, POPE, POPS, and PIP2 (Table 1). Twocopies of the peptide were placed 45 A above and belowthe membrane center, respectively (Fig. 5 A), and each sys-tem was simulated for 300 ns with five replicates. InMARCKS-ED simulations, the center of mass of the pep-tides was restrained to be between �48 and 48 A alongthe z axis.

In both systems, the peptide binds quickly to the mem-brane within 50 ns, initially driven by favorable electro-static interactions between the peptide and the membrane(Fig. 5, B and C). To check the convergence of each resi-due’s burial depth, we calculated the z-positions of eachside chain relative to the lipid phosphate groups in 50-nsblocks (Fig. 6). MARCKS1 shows good convergencewith the phenylalanine (residues 7, 9, 13, 18, and 20)and leucine (residue 15) residues buried deeply in themembrane, whereas the lysine residues at the termini

FIGURE 5 MARCKS-ED-membrane association systems. (A) An initial str

spheres with lysine colored in blue, phenylalanine and leucine in orange, and oth

POPS in purple sticks, POPE in silver sticks, and PIP2 in red sticks; ions and w

MARCKS-ED in (B) MARCKS1 and (C) MARCKS2 simulations (each with fiv

simulation, and the black lines mark the average position of the lipid phosphate

Biophysical Journal 109(10) 2012–2022

remain on the membrane surface. This binding mode isin agreement with the previous results showing that thepeptide adopts a U-shape when it binds to membranes(65). However, in the MARCKS2 system, the z-positionsshow large variations between independent simulations.A clustering analysis shows that the MARCKS-ED struc-tures are concentrated on fewer clusters in MARCKS2than in MARCKS1 (Fig. S11 A). It is likely that PIP2 bindsto the peptide and slows down the conformational sam-pling of the peptides in MARCKS2. In addition, the pep-tide structure is more compact in terms of radius ofgyration in MARCKS2, suggesting that the peptide isless extended in the presence of PIP2 (Fig. S11 B). Theseresults suggest that PIP2 has a large impact on the pep-tide’s conformation.

The burial depth of 12 out of the 25 residues has beendetermined experimentally by replacing the correspondingresidue with a spin label and measuring the electron para-magnetic resonance spectrum (66). For both MARCKS1and MARCKS2, the depths of only a few residues agreewith the experimental estimates with bulky spin labels.A likely reason for the MD/experiment disagreement isthat the experiment measures the depth of the spin labelattached to a cysteine single mutant, not the wild-typeamino acid. The S-(1-oxyl-2,2,5,5-tetramethyl-2,5-dihy-dro-1H-pyrrol-3-yl)methyl methanesulfonothioate (MTSL)spin label is 8–10 A from the Ca and likely interactswith the membrane (67). Therefore, further simulationswith spin label at specific residues are needed to clarifythis discrepancy and better quantify interactions betweenMTSL and the membrane.

Adding phosphatidylserine and PIP2 lipids to PC vesiclesenhances the binding between MARCKS-ED and mem-branes (68). To compare the sequestration effect on differentlipids, we calculated the percentage of lipids in contact withMARCKS-ED (Fig. 7). The sequestration effect is strongestto PIP2, followed by POPS, and then POPE and POPC (seeTable S3 for the two-sample t-test results), in accordancewith the amount of negative charge in each lipid type. It is

ucture of the MARCKS2 simulation system. MARCKS-ED is shown in

er residues in white; DCLE is shown in blue spheres, POPC in orange sticks,

ater molecules are not shown for clarity. (B and C) The center of mass of

e replicates). Each colored line represents one peptide in each independent

groups. To see this figure in color, go online.

A

B

FIGURE 6 Relative z-positions of each residue in

50-ns blocks in (A) MARCKS1 and (B) MARCKS2.

The lipid phosphate groups are recentered to be

located at z ¼ 0 and the membrane center is located

at z < 0. Each line represents one peptide; there are

10 lines from the five independent simulations. The

stars indicate the experimental estimates from Qin

et al. (66). To see this figure in color, go online.

CHARMM-GUI HMMM Builder 2019

worth noting that the most probable number of PIP2 in con-tact with MARCKS-ED is 3, which is close to the experi-mental estimation (69).

HMMM to full-lipid conversion

HMMM Builder provides a facility to convert back andforth between an HMMM system and a full-length lipid

FIGURE 7 Percentages of lipids in contact with MARCKS-ED and a snapsho

PIP2. A contact is counted when the minimal distance between the heavy atom

peptide. The averages and standard errors of percentages from 150 ns to 300 n

spheres with lysine colored in blue, phenylalanine and leucine in orange, and oth

POPS in purple sticks, POPE in silver sticks, and PIP2 in red sticks. To see this

system. To illustrate this function, we converted the lastsnapshot of one of the MARCKS1 system to a full-lengthlipid system and simulated the full-length lipid system for100 ns (Fig. 8 A). As expected, the box size of the full-lipid system extends along the z axis and shrinks alongthe xy plane (Fig. 8 B). The peptides remain associatedwith the membrane (Fig. 8 C), as shown in the z-positiontime series for each residue (Fig. S12).

t of the MARCKS2 system showing a MARCKS-ED in contact with three

s from a lipid and the peptide is <4.5 A. Each colored line represents one

s are shown in blue numbers. In the snapshot, MARCKS-ED is shown in

er residues in white; DCLE is shown in blue spheres, POPC in orange sticks,

figure in color, go online.

Biophysical Journal 109(10) 2012–2022

FIGURE 8 Simulation of the converted full-length lipid system. (A) The last snapshot of the full-length lipid MARCKS1 simulation. Lipids are shown in

sticks; peptides are shown in spheres, with lysine colored in blue, phenylalanine and leucine in orange, and other residues in white. (B) The time series of the

system box size (x and y are constrained to be identical in simulations). (C) Time series of the center of mass of the peptides. The black lines indicate the

average position of the lipid phosphate groups. To see this figure in color, go online.

2020 Qi et al.

CONCLUSIONS

In this study, we present HMMM Builder, a new module inCHARMM-GUI to build membrane simulation systems us-ing the HMMM model. To show the capability of HMMMBuilder, several single-lipid and mixed-lipid bilayers wereset up to investigate the basic properties of HMMM mem-branes under the NPT ensemble. Due to the liquid nature ofthe DCLE molecules used to represent the membrane core,the HMMM membranes show larger per-lipid area, fasterlipid diffusion and mixing, and neutral pressure in theDCLE layer (with close resemblance of the full-lengthmembrane pressure profile outside the DCLE layer). TheMARCKS-ED peptide is simulated as an example to studypeptide-membrane association using mixed-lipid mem-branes generated by HMMM Builder. The simulationsshow good convergence in the POPC/POPE/POPS bilayer,highlighting the advantage of using the HMMM model.However, in the POPC/POPE/POPS/PIP2 bilayer, differentreplicates of the peptide show some variations in the resi-due burial depth due to strong interactions betweenMARCKS-ED and PIP2. It should be emphasized that theHMMM model accelerates diffusion of lipids and othermolecular species within the lipids, but not much the inter-nal degrees of freedom of the peptide. Finally, HMMMBuilder provides conversions between HMMM and full-length lipid systems, allowing the user to easily take advan-tage of both HMMM and full-length membranes for thesystem of interest.

SUPPORTING MATERIAL

Twelve figures and three tables are available at www.biophys.org/biophysj/

supplemental/S0006-3495(15)01047-4/.

AUTHOR CONTRIBUTIONS

Y.Q., J.B.K., andW.I. designed the research; Y.Q., X.C., and J.L. performed

the research; Y.Q., X.C., J.L., J.V.V., T.V.P., E.T., and S.P. contributed ana-

lytic tools; Y.Q., J.B.K., and W.I. analyzed data; Y.Q., J.V.V., E.T., J.B.K.,

and W.I. wrote the article.

Biophysical Journal 109(10) 2012–2022

ACKNOWLEDGMENTS

This work was supported in part by National Science Foundation grants

DBI-1145987 (W.I.), MCB-1157677 (W.I.), DBI-1145652 (J.B.K.),

MCB-1149187 (J.B.K.), and XSEDE TG-MCB070009 (W.I.), and by Na-

tional Institutes of Health grants P41-GM104601 (E.T.) and U54-

GM087519 (E.T. and W.I.).

REFERENCES

1. Tew, G. N., R. W. Scott, ., W. F. Degrado. 2010. De novo design ofantimicrobial polymers, foldamers, and small molecules: from discov-ery to practical applications. Acc. Chem. Res. 43:30–39.

2. Bolintineanu, D. S., and Y. N. Kaznessis. 2011. Computational studiesof protegrin antimicrobial peptides: a review. Peptides. 32:188–201.

3. Bennett, W. F., and D. P. Tieleman. 2014. The importance of membranedefects-lessons from simulations. Acc. Chem. Res. 47:2244–2251.

4. Vaz, W. L., R. M. Clegg, and D. Hallmann. 1985. Translational diffu-sion of lipids in liquid crystalline phase phosphatidylcholine multibi-layers. A comparison of experiment with theory. Biochemistry.24:781–786.

5. Gaede, H. C., and K. Gawrisch. 2003. Lateral diffusion rates of lipid,water, and a hydrophobic drug in a multilamellar liposome.Biophys. J. 85:1734–1740.

6. Sinthuvanich, C., A. S. Veiga, ., J. P. Schneider. 2012. Anticancerb-hairpin peptides: membrane-induced folding triggers activity.J. Am. Chem. Soc. 134:6210–6217.

7. Eliezer, D., E. Kutluay, ., G. Browne. 2001. Conformational proper-ties of a-synuclein in its free and lipid-associated states. J. Mol. Biol.307:1061–1073.

8. Arcario, M. J., Y. Z. Ohkubo, and E. Tajkhorshid. 2011. Capturingspontaneous partitioning of peripheral proteins using a biphasic mem-brane-mimetic model. J. Phys. Chem. B. 115:7029–7037.

9. Stansfeld, P. J., and M. S. Sansom. 2011. Molecular simulation ap-proaches to membrane proteins. Structure. 19:1562–1572.

10. Ulmschneider, J. P., M. Andersson, and M. B. Ulmschneider. 2011.Determining peptide partitioning properties via computer simulation.J. Membr. Biol. 239:15–26.

11. Rogaski, B., and J. B. Klauda. 2012. Membrane-binding mechanism ofa peripheral membrane protein through microsecond molecular dy-namics simulations. J. Mol. Biol. 423:847–861.

12. Shaw, D. E., M. M. Deneroff, ., S. C. Wang. 2008. Anton, a special-purpose machine for molecular dynamics simulation. Commun. ACM.51:91–97.

13. Sugita, Y., and Y. Okamoto. 1999. Replica-exchange molecular dy-namics method for protein folding. Chem. Phys. Lett. 314:141–151.

CHARMM-GUI HMMM Builder 2021

14. Ostermeir, K., and M. Zacharias. 2013. Advanced replica-exchangesampling to study the flexibility and plasticity of peptides and proteins.Biochim. Biophys. Acta. 1834:847–853.

15. Mori, T., J. Jung, and Y. Sugita. 2013. Surface-tension replica-ex-change molecular dynamics method for enhanced sampling of biolog-ical membrane systems. J. Chem. Theory Comput. 9:5629–5640.

16. Hamelberg, D., J. Mongan, and J. A. McCammon. 2004. Acceleratedmolecular dynamics: a promising and efficient simulation method forbiomolecules. J. Chem. Phys. 120:11919–11929.

17. Huang, K., and A. E. Garcıa. 2014. Acceleration of lateral equilibrationin mixed lipid bilayers using replica exchange with solute tempering.J. Chem. Theory Comput. 10:4264–4272.

18. Wang, Y., P. R. Markwick,., J. A. McCammon. 2011. Enhanced lipiddiffusion and mixing in accelerated molecular dynamics. J. Chem. The-ory Comput. 7:3199–3207.

19. Ohkubo, Y. Z., T. V. Pogorelov,., E. Tajkhorshid. 2012. Acceleratingmembrane insertion of peripheral proteins with a novel membranemimetic model. Biophys. J. 102:2130–2139.

20. Vermaas, J. V., J. L. Baylon, ., E. Tajkhorshid. 2015. Efficient explo-ration of membrane-associated phenomena at atomic resolution.J. Membr. Biol. 248:563–582.

21. Pogorelov, T. V., J. V. Vermaas, ., E. Tajkhorshid. 2014. Partitioningof amino acids into a model membrane: capturing the interface. J. Phys.Chem. B. 118:1481–1492.

22. Ohkubo, Y. Z., and E. Tajkhorshid. 2008. Distinct structural and adhe-sive roles of Ca2þ in membrane binding of blood coagulation factors.Structure. 16:72–81.

23. Arcario, M. J., and E. Tajkhorshid. 2014. Membrane-induced structuralrearrangement and identification of a novel membrane anchor in talinF2F3. Biophys. J. 107:2059–2069.

24. Blanchard, A. E., M. J. Arcario, ., E. Tajkhorshid. 2014. A highlytilted membrane configuration for the prefusion state of synaptobrevin.Biophys. J. 107:2112–2121.

25. Baylon, J. L., I. L. Lenov,., E. Tajkhorshid. 2013. Characterizing themembrane-bound state of cytochrome P450 3A4: structure, depth ofinsertion, and orientation. J. Am. Chem. Soc. 135:8542–8551.

26. Vermaas, J. V., and E. Tajkhorshid. 2014. Conformational heterogene-ity of a-synuclein in membrane. Biochim. Biophys. Acta. 1838:3107–3117.

27. Wu, Z., and K. Schulten. 2014. Synaptotagmin’s role in neurotrans-mitter release likely involves Ca2þ-induced conformational transition.Biophys. J. 107:1156–1166.

28. Baylon, J. L., and E. Tajkhorshid. 2015. Capturing spontaneous mem-brane insertion of the influenza virus hemagglutinin fusion peptide.J. Phys. Chem. B. 119:7882–7893.

29. Rheault, J. F., E. Gagne,., P. Lague. 2015. Molecular model of hemo-globin N from Mycobacterium tuberculosis bound to lipid bilayers: acombined spectroscopic and computational study. Biochemistry.54:2073–2084.

30. Vermaas, J. V., and E. Tajkhorshid. 2014. A microscopic view of phos-pholipid insertion into biological membranes. J. Phys. Chem. B.118:1754–1764.

31. Jo, S., T. Kim, ., W. Im. 2008. CHARMM-GUI: a web-based graph-ical user interface for CHARMM. J. Comput. Chem. 29:1859–1865.

32. Brooks, B. R., C. L. Brooks, 3rd,., M. Karplus. 2009. CHARMM: thebiomolecular simulation program. J. Comput. Chem. 30:1545–1614.

33. Phillips, J. C., R. Braun,., K. Schulten. 2005. Scalable molecular dy-namics with NAMD. J. Comput. Chem. 26:1781–1802.

34. Van Der Spoel, D., E. Lindahl,., H. J. Berendsen. 2005. GROMACS:fast, flexible, and free. J. Comput. Chem. 26:1701–1718.

35. Case, D. A., T. E. Cheatham, 3rd, ., R. J. Woods. 2005. The Amberbiomolecular simulation programs. J. Comput. Chem. 26:1668–1688.

36. Eastman, P., M. S. Friedrichs, ., V. S. Pande. 2013. OpenMM 4:a reusable, extensible, hardware independent library for high perfor-mance molecular simulation. J. Chem. Theory Comput. 9:461–469.

37. Jo, S., J. B. Lim,., W. Im. 2009. CHARMM-GUI Membrane Builderfor mixed bilayers and its application to yeast membranes. Biophys. J.97:50–58.

38. Cheng, X., S. Jo,., W. Im. 2013. CHARMM-GUI micelle builder forpure/mixed micelle and protein/micelle complex systems. J. Chem. Inf.Model. 53:2171–2180.

39. Qi, Y., X. Cheng,., W. Im. 2014. CHARMM-GUI PACE CG Builderfor solution, micelle, and bilayer coarse-grained simulations. J. Chem.Inf. Model. 54:1003–1009.

40. Jo, S., T. Kim, and W. Im. 2007. Automated builder and database ofprotein/membrane complexes for molecular dynamics simulations.PLoS One. 2:e880.

41. Wu, E. L., X. Cheng, ., W. Im. 2014. CHARMM-GUI MembraneBuilder toward realistic biological membrane simulations. J. Comput.Chem. 35:1997–2004.

42. Jo, S., X. Cheng, ., W. Im. 2014. CHARMM-GUI PDB manipulatorfor advanced modeling and simulations of proteins containing nonstan-dard residues. Adv. Protein Chem. Struct. Biol. 96:235–265.

43. Berman, H. M., J. Westbrook,., P. E. Bourne. 2000. The Protein DataBank. Nucleic Acids Res. 28:235–242.

44. Lomize, M. A., A. L. Lomize, ., H. I. Mosberg. 2006. OPM: orienta-tions of proteins in membranes database. Bioinformatics. 22:623–625.

45. Huang, J., and A. D. MacKerell, Jr. 2013. CHARMM36 all-atom addi-tive protein force field: validation based on comparison to NMR data.J. Comput. Chem. 34:2135–2145.

46. Venable, R. M., A. J. Sodt,., J. B. Klauda. 2014. CHARMM all-atomadditive force field for sphingomyelin: elucidation of hydrogenbonding and of positive curvature. Biophys. J. 107:134–145.

47. Klauda, J. B., R. M. Venable, ., R. W. Pastor. 2010. Update of theCHARMM all-atom additive force field for lipids: validation on sixlipid types. J. Phys. Chem. B. 114:7830–7843.

48. Lim, J. B., B. Rogaski, and J. B. Klauda. 2012. Update of the choles-terol force field parameters in CHARMM. J. Phys. Chem. B.116:203–210.

49. Steinbach, P. J., and B. R. Brooks. 1994. New spherical-cutoff methodsfor long-range forces in macromolecular simulation. J. Comput. Chem.15:667–683.

50. Essmann, U., L. Perera, ., L. G. Pedersen. 1995. A smooth particlemesh Ewald method. J. Chem. Phys. 103:8577–8593.

51. Martyna, G. J., D. J. Tobias, and M. L. Klein. 1994. Constant-pressuremolecular-dynamics algorithms. J. Chem. Phys. 101:4177–4189.

52. Feller, S. E., Y. H. Zhang, ., B. R. Brooks. 1995. Constant-pressuremolecular-dynamics simulation: the Langevin piston method.J. Chem. Phys. 103:4613–4621.

53. Humphrey, W., A. Dalke, and K. Schulten. 1996. VMD: visual molec-ular dynamics. J. Mol. Graph. 14:33–38, 27–38.

54. Michaud-Agrawal, N., E. J. Denning, ., O. Beckstein. 2011. MDA-nalysis: a toolkit for the analysis of molecular dynamics simulations.J. Comput. Chem. 32:2319–2327.

55. Harasima, A. 1958. Molecular theory of surface tension. Adv. Chem.Phys. 1:203–237.

56. Sonne, J., F. Y. Hansen, and G. H. Peters. 2005. Methodological prob-lems in pressure profile calculations for lipid bilayers. J. Chem. Phys.122:124903.

57. Impey, R. W., P. A. Madden, and I. R. McDonald. 1983. Hydration andmobility of ions in solution. J. Phys. Chem. 87:5071–5083.

58. Cantor, R. S. 1997. Lateral pressures in cell membranes: a mechanismfor modulation of protein function. J. Phys. Chem. B. 101:1723–1725.

59. Cantor, R. S. 1999. The influence of membrane lateral pressures onsimple geometric models of protein conformational equilibria. Chem.Phys. Lipids. 101:45–56.

60. Szleifer, I., D. Kramer,., S. A. Safran. 1990. Molecular theory of cur-vature elasticity in surfactant films. J. Chem. Phys. 92:6800–6817.

Biophysical Journal 109(10) 2012–2022

2022 Qi et al.

61. Sodt, A. J., and R. W. Pastor. 2013. Bending free energy from simula-tion: correspondence of planar and inverse hexagonal lipid phases.Biophys. J. 104:2202–2211.

62. McIlroy, B. K., J. D. Walters,., J. D. Johnson. 1991. Phosphorylation-dependent binding of a synthetic MARCKS peptide to calmodulin.J. Biol. Chem. 266:4959–4964.

63. Rauch, M. E., C. G. Ferguson, ., D. S. Cafiso. 2002. Myristoylatedalanine-rich C kinase substrate (MARCKS) sequesters spin-labeledphosphatidylinositol 4,5-bisphosphate in lipid bilayers. J. Biol.Chem. 277:14068–14076.

64. Wang, J., A. Gambhir, ., S. McLaughlin. 2002. Lateral sequestrationof phosphatidylinositol 4,5-bisphosphate by the basic effector domainof myristoylated alanine-rich C kinase substrate is due to nonspecificelectrostatic interactions. J. Biol. Chem. 277:34401–34412.

65. Ellena, J. F., M. C. Burnitz, and D. S. Cafiso. 2003. Location of themyristoylated alanine-rich C-kinase substrate (MARCKS) effectordomain in negatively charged phospholipid bicelles. Biophys. J.85:2442–2448.

Biophysical Journal 109(10) 2012–2022

66. Qin, Z., and D. S. Cafiso. 1996. Membrane structure of protein kinase Cand calmodulin binding domain of myristoylated alanine rich C kinasesubstrate determined by site-directed spin labeling. Biochemistry.35:2917–2925.

67. Kemmerer, S., J. C. Voss, and R. Faller. 2013. Molecular dynamicssimulation of dipalmitoylphosphatidylcholine modified with a MTSLnitroxide spin label in a lipid membrane. Biochim. Biophys. Acta.1828:2770–2777.

68. Arbuzova, A., L. Wang, ., S. McLaughlin. 2000. Membrane bindingof peptides containing both basic and aromatic residues. Experimentalstudies with peptides corresponding to the scaffolding region of caveo-lin and the effector region of MARCKS. Biochemistry. 39:10330–10339.

69. Wang, J., A. Arbuzova, ., S. McLaughlin. 2001. The effector domainof myristoylated alanine-rich C kinase substrate binds strongly to phos-phatidylinositol 4,5-bisphosphate. J. Biol. Chem. 276:5012–5019.