Dynamic Mechanism of Fatty Acid Transport across Cellular

Sep 24, 2008 - Drug Discovery and Design Center, State Key Laboratory of Drug Research, Shanghai Institute of Materia Medica, Shanghai Institutes for ...
3 downloads 14 Views 1MB Size
13070

J. Phys. Chem. B 2008, 112, 13070–13078

Dynamic Mechanism of Fatty Acid Transport across Cellular Membranes through FadL: Molecular Dynamics Simulations Hanjun Zou,† Mingyue Zheng,† Xiaomin Luo,† Weiliang Zhu,† Kaixian Chen,† Jianhua Shen,*,†,‡ and Hualiang Jiang*,†,‡ Drug DiscoVery and Design Center, State Key Laboratory of Drug Research, Shanghai Institute of Materia Medica, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, and Graduate School of the Chinese Academy of Sciences, 555 Zuchongzhi Road, Shanghai 201203, China, and School of Pharmacy, East China UniVersity of Science and Technology, Shanghai 200237, China ReceiVed: NoVember 17, 2007; ReVised Manuscript ReceiVed: May 17, 2008

FadL is an important member of the family of fatty acid transport proteins within membranes. In this study, 11 conventional molecular dynamics (CMD) and 25 steered molecular dynamics (SMD) simulations were performed to investigate the dynamic mechanism of transport of long-chain fatty acids (LCFAs) across FadL. The CMD simulations addressed the intrinsically dynamic behavior of FadL. Both the CMD and SMD simulations revealed that a fatty acid molecule can move diffusively to a high-affinity site (HAS) from a low-affinity site (LAS). During this process, the swing motion of the L3 segment and the hydrophobic interaction between the fatty acid and FadL could play important roles. Furthermore, 22 of the SMD simulations revealed that fatty acids can pass through the gap between the hatch domain and the transmembrane domain (TMD) by different pathways. SMD simulations identified nine possible pathways for dodecanoic acid (DA) threading the barrel of FadL. The binding free energy profiles between DA and FadL along the MD trajectories indicate that all of the possible pathways are energetically favorable for the transport of fatty acids; however, one pathway (path VI) might be the most probable pathway for DA transport. The reasonability and reliability of this study were further demonstrated by correlating the MD simulation results with the available mutagenesis results. On the basis of the simulations, a mechanism for the full-length transport process of DA from the extracellular side to the periplasmic space mediated by FadL is proposed. Introduction Exogenous long-chain fatty acids (LCFAs) are sources of energy and carbon for many biological functions and biomolecule syntheses for both microorganisms and mammals.1-4 In addition, the uptake of LCFAs might also be important to some pathophysiological processes such as bacterial infection.5 Therefore, the transport of LCFAs across cellular membranes represents a fundamental biological process.6 LCFAs can cross membranes spontaneously because of their hydrophobic nature.6 However, in many prokaryotic and eukaryotic cells, the transport of exogenous LCFAs is effectively mediated by specific membrane-bound proteins.7 So far, several distinct membranebound and membrane-associated fatty acid transport proteins have been identified and characterized in a number of different systems, such as FadL,8,9 fatty acid translocase (FAT, CD36),10 fatty acid binding protein-plasma membrane bound (FABPpm),11 and fatty acid transport protein (FATP) in adipocytes and yeast.3 A wealth of structural and functional information is available for the transport proteins of hydrophilic molecules, such as ionconducting properties of porins12,13 and aquaporins.14,15 Unfortunately, the transporters of hydrophobic molecules across cellular membranes are not as clear. FadL is an important outer membrane-bound protein for fatty acid binding and transport, which was first found in Escherichia coli (E. coli).16 In addition, * To whom correspondence should be addressed. E-mail: jhshen@ mail.shcnc.ac.cn. † Chinese Academy of Sciences. ‡ East China University of Science and Technology.

many systems that biodegrade synthetic industrial chemicals such as xenobiotics and polyaromatic hydrocarbons involve proteins that are FadL orthologs.17 Accordingly, FadL has been taken as a model for investigating the transport mechanisms for hydrophobic molecules such as fatty acids and xenobiotics.17 The transport mechanism for LCFAs across FadL has been investigated extensively using both biochemical and biophysical approaches.18-20 Recently, the X-ray crystal structures of E. coli FadL in two crystal forms (monoclinic and hexagonal) were determined [Protein Data Bank (PDB) entries 1T16 and 1T1L],21 which provided insight into the structural foundation underlying the binding and transport of LCFAs to the transporters. The structures of FadL also support the hypothesis that ligandinduced conformational changes are involved in the transport process. The structure of FadL can generally be divided into three domains as demonstrated in Figure 1: Fourteen antiparallel β-strands line a long (∼50 Å) β-barrel, which is designated as the transmembrane domain (TMD) because half of the barrel spans the membrane (Figure 1A); a small domain consisting of a folded R-helix (L3) and a flexible loop (L4) is located at the top of TMD; and three short helices (H1, H2, and H3) connected by two short loops (L1 and L2) fold into a hatch domain, which plugs the β-barrel (Figure 1A). On the small extracellular domain, a solvent-exposed hydrophobic groove is located between the L3 and L4 loops in which fatty acid analogs (C8E4) are bound, as illustrated by the crystal structure.21 Accordingly, this hydrophobic groove is assigned as the low-affinity site (LAS) for fatty acid binding during the initial interaction of FadL with LCFAs (Figure 1A).21 Inside the FadL barrel, a prominent hydrophobic pocket is found on the extracellular side

10.1021/jp710964x CCC: $40.75  2008 American Chemical Society Published on Web 09/24/2008

LCFA Transport across Cell Membranes through FadL

Figure 1. (A) Three-dimensional representation of the transmembrane transporter, FadL. The ligand, LDAO, is represented by spheres. The TMD and hatch domains are colored and labeled in yellow and blue, respectively. The β-strand S3 and kink are colored in salmon and aqua, respectively. The image on the right is an enlargement of the hatch domain. The seven N-terminal residues in the monoclinic (green) and hexagonal (orange) crystals are represented as loops. The different conformations of Phe3 in the two crystal structures are shown in sticks and colored in green and orange, respectively. The putative locations of the channel are indicated with asterisks and marked with CH1, CH2, and CH3. (B) Overview of the FadL/DMPC/water simulation system. The protein is shown in cartoon format, and the P atoms are shown as salmon spheres. The side chains of Trp and Tyr are shown as blue spheres, which are mainly located in the lipid headgroup layer. These residues interact with the phospholipids head groups of the lipids. The pore axis scale along the channel is also shown.

of the membrane where the lauryldimethylamine-N-oxide (LDAO) molecule resides in Figure 1A; this site is thought to be the high-affinity site (HAS) for fatty acids.21 The two crystal structures determined in two different space groups display the significant difference for the conformations of the seven N-terminal residues (Figure 1A), reflecting the flexibility of the N-terminus. Based on the X-ray crystal structures together with mutagenesis and biochemical results, van den Berg et al. proposed a mechanism for LCFA transport mediated by FadL.21 In general, the transport of LCFAs through FadL is a spontaneously diffusive process that is associated with the diffusion of substrate from the LAS to the HAS and the conformational changes of the N-terminus, the hatch domain, and possibly the S3 kink. Although this mechanism has provided a structural model for LCFA transport across the outer membrane through FadL, several fundamental questions are still open. In particular, how does FadL behave dynamically in an environment mimicking the cell membrane? Is there any other role that the small outside domain of FadL plays for LCFA transport in addition to attracting substrates from the external medium? How do LCFAs overcome the barrier caused by the hatch domain when they cross the barrel? How do the conformational changes of FadL correlate with the transport of LCFAs? To reach the answers of these questions, we have performed a series of conventional molecular dynamics (CMD) and steered molecular dynamics (SMD) simulations on E. coli FadL, FadL mutants, and their complexes with lauryldimethylamine-N-oxide (LDAO) and dodecanoic acid (DA). Herein, we report on the MD results for the dynamic and kinetic mechanism of the transport of LCFAs crossing the outer membrane through FadL. Materials and Methods Simulation Systems. The X-ray crystal structures of the FadL-LDAO complex at 2.6-Å resolution (PDB entries 1T16 and 1T1L)21 were used as starting structures for simulations. The structures of FadL mutants were constructed using Insight

J. Phys. Chem. B, Vol. 112, No. 41, 2008 13071 II (Accelrys, San Diego, CA). The pKa was calculated to determine any of the binding site residues in the proteins that were likely to adopt nonstandard ionization states. The surfaceaccessibility-modified Tanford-Kirkwood (TK) method22,23 encoded in MacroDox version 3.2.124 was employed to determine the protonation status of each titratable residue in the protein models at pH 6.8. The structure of DA was generated by modifying the structure of LDAO using Insight II. The molecular topology files for ligands were generated with the program PRODRG (http://davapc1.bioch.dundee.ac.uk/programs/ prodrg/prodrg.html).25 The atomic charges of ligands were assigned using the CHELPG procedure at the Hartree-Fock level with a 6-311G* basis set.26 The structure models of FadL-DA in the LAS were constructed by using the docking program AutoDock 3.0.27 The girdles of aromatic residues such as Trp and Tyr on the surface of membrane proteins have been used as a criterion to choose the suitable lipid for the simulations of β-barrel membrane proteins.28-33 Such aromatic residues were also found on the interfacial surface of FadL. As shown in Figure 1B, the dimyristoylphosphatidylcholine (DMPC) bilayer is thick enough for FadL to embed in it. Moreover, DMPC has been widely used in the simulations of other bacterial outer membrane proteins.28-32 Therefore, DMPC was employed to mimic the membrane environment of FadL in this study. A preequilibrated DMPC bilayer with 128 lipid molecules was obtained from SoftSimu (http://www.apmaths.uwo.ca/∼mkarttu/ downloads.shtml), and then a large DMPC bilayer containing 255 lipid molecules was constructed. Twenty-four lipid molecules were subsequently removed from the constructed bilayer to generate a suitable membrane system into which the β-barrel of FadL could be embedded (Figure 1B). The lipid topology file was provided by Gurtovenko et al.34 Each protein/DMPC system was then solvated in a bath of SPC water molecules. The resulting system was subjected to energy minimization to remove unfavorable contacts and equilibrated by a short MD simulation with positional constraints on the protein atoms. A 100 mM NaCl solution was also used, along with a number of counterions, to neutralize the total charge of the system (all ions were placed initially at random). The entire system was then energy minimized with steepest descent for some steps until reaching the convergence value of 100 kJ · mol-1 · nm-1. CMD Simulations. All CMD simulations were performed using the GROMACS 3.1.4 package35,36 with the GROMACS force field.37 The linear constraint solver (LINCS) method38 was used to constrain bond lengths, allowing an integration step of 2 fs. The electrostatic interactions were calculated using the Particle Mesh Ewald (PME) method39,40 with a 1.0-nm cutoff. NPT conditions (constant number of particles, pressure, and temperature) were used in the simulations, which allowed the bilayer/protein area to adjust to its optimum value for the force field employed. Water, lipids, and protein (or protein-ligand complex) were coupled separately to a temperature bath at 310 K using a coupling constant of 0.1 ps.41 SMD Simulations. In SMD simulations, an external force is applied to permit a system to overcome barriers in a shorter time than in CMD simulations.42-44 To map the possible pathways of FadL for fatty acid transport, a soft spring rather than stiff spring was assigned to the ligand. The force constant for the spring and pulling velocity were set to 500 kJ · mol-1 · nm-2 and 0.01 Å · ps-1, respectively. The pulling direction was along the Z axis. Analyses were performed using the facilities within the GROMACS package. The binding free energies of DA to FadL were calculated using the scoring function of AutoDock 3.0.27

13072 J. Phys. Chem. B, Vol. 112, No. 41, 2008 Structural diagrams were prepared using PyMOL (http:// pymol.sourceforge.net).

Zou et al. TABLE 1: Summary of 11 CMD and 25 SMD Simulationsa model trajectory duration (ns) method I

A1 A2

30 10

CMD CMD

II

B1

15

CMD

B2

10

CMD

C1

20

CMD

C2

20

CMD

C3

10

CMD

C4

10

CMD

D1

1.8

SMD

D2

∼6

SMD

D3

∼5

SMD

V

E1-E22

∼5.8

SMD

VI

F1

10

CMD

F2

10

CMD

F3

10

CMD

Results and Discussion As mentioned above, the transport mechanism of LCFAs and other hydrophobic molecules through FadL is very complicated and involves several steps: the binding of LCFAs to the lowaffinity site (LAS); the diffusion of the LCFAs to the highaffinity site (HAS); and conformational changes in the N-terminus, the hatch domain, and S3 kink that create channel(s) through which the LCFAs can pass. The main goal of this study was to investigate the dynamic/kinetic mechanism of LCFA transport across FadL. To this end, six models for CMD and SMD simulations were designed. In model I, two CMD simulations on apo-FadL in the monoclinic form and hexagonal form (A1 and A2, respectively) were conducted to explore the dynamics of the protein. In model II, another two CMD simulations (B1 and B2) were carried out to probe the binding and behavior of DA in the LAS. In model III, four CMD simulations (C1-C4) were carried out on FadL in complexes with ligands to investigate the conformational change process induced by the ligands. Because the structure model of the DA-FadL complex was constructed based on the crystal structure of the LDAO-FadL complex, CMD simulations were performed on both models of DA-FadL and LDAO-FadL complexes at this stage to test the reliability of the simulation procedure. In model IV, three SMD simulations (D1-D3) were performed to investigate the diffusion process of the ligand from the LAS to the HAS and provide some insight into how fatty acids transport from the HAS to the periplasm. In model V, 22 SMD simulations (E1-E22) were carried out to further explore the transport mechanism of ligand from the HAS to the periplasm in detail. Finally, three CMD simulations were performed on three mutants of FadL (F1-F3) in model VI to confirm the reliability of MD simulations against the biological evidence. In total, 11 CMD and 25 SMD simulations were performed in this study. More detailed information on the CMD and SMD simulations is listed in Table 1. Dynamics of FadL. To quantify the dynamic behavior of FadL, we monitored the root-mean-square deviation (RMSD) of the protein in simulation A1 (Figure 2A). The RMSD profiles indicate that the TMD equilibrated after ∼2 ns and that the overall simulation system converged after ∼8 ns (simulation A1). Accordingly, we only performed 10 to 20-ns simulations for other systems (simulations A2, B1-B2, C1-C4, and F1-F3) to save computing expenses. The results indicate that the simulations are long enough for the systems to achieve equilibrium (Figure S1 in the Supporting Information). From the RMSD profiles in Figure 2, we can address the dynamics of FadL, i.e., the rigid and flexible parts. Inclusion of all atoms yields a relatively high RMSD of about 5.0 Å, which mainly comes from the extracellular loop (L3) of the small domain (Figure 2A). Normally, the crystal structure of the L3 segment would not be resolved if the segment underwent such large motions. However, the crystal structures of FadL were determined with a detergent molecule (C8E4) binding to the LAS,21 which can stabilize the structure of the L3 segment. To verify this hypothesis, we compared the average structures of the L3 segment in the dynamics trajectories of apo-FadL (simulation A1) and the FadL-DA complex (simulations B1 and B2). The results indicated that the binding of ligand within the LAS can constrain the motion of the L3 segment (Figure S2 in the Supporting Information). The normal-mode analysis also revealed that the first mode corresponds to the swing motion of

III

IV

brief description apo-FadL protein apo-FadL protein in the hexagonal form DA bound to the LAS of FadL (binding model I) DA bound to the LAS of FadL (binding model II) LDAO bound to the HAS of FadL DA bound to the HAS of FadL LDAO bound to the HAS of FadL in the hexagonal form DA bound to the HAS of FadL in the hexagonal form diffusion of DA from the LAS to the HAS starting from the snapshot at 1070 ps in trajectory B2 extended simulation of D1 from the HAS to the periplasm transport of DA from the HAS to the periplasm starting from a snapshot of trajectory C4 transport of DA from the HAS to the periplasm starting from different snapshots from trajectory C2 simulation of FadL mutant His83Ala simulation of FadL mutant obtained by insertion of Glu and Phe between Ala211 and Gly212 simulation of FadL mutant obtained by insertion of Glu and Phe between Val383 and Asp384

a All of the simulation models were constructed based on the crystal structure of FadL in the monoclinic form (PDB entry 1T16) unless otherwise specified.

the L3 segment toward the L4 segment (Figure S3 in the Supporting Information). This indicates that the large structural flexibility of the L3 segment might be its intrinsic property, which might be one of the driving forces for the diffusive process of LCFAs from the LAS to the HAS. The structure of the β-barrel is an important characteristic of membrane channels.45 As expected, the transmembrane β-barrel is quite stable, exhibiting deviations significantly smaller than those of all atoms (Figure 2A). However, a breathing motion, though small, was observed for the TMD, as seen from the profile of the radius of gyration (Rg) (Figure 2B). Unlike other membrane transport proteins such as OmpF46 and OmpA,47 FadL has an internal hatch domain. RMSD analysis indicated that the conformation of this domain is also stable inside the central β-sheet core (Figure 2A). However, Rg of the hatch domain revealed a more significant fluctuation around 10.7 Å compared to the TMD (Figure 2B). To further quantify the breathing motion, the time dependences of the volume changes for the hatch domain (without the seven N-terminal residues) and the corresponding segment of the TMD were monitored during simulation A1 by using an in-house program developed based on the algorithm of Laskowski et al.48 (Figure 2C). Indeed, the breathing motion was also observed from the profiles of volume changes. As for the profiles of Rg, more significant breathing motion was also found for the hatch domain, especially when the total volume was taken into account (about 1.5% variation for the TMD and about 5% for the hatch domain). To qualify the breathing motion from the viewpoint of structure, we carefully examined the structures in trajectory A1. Two typical snapshots from simula-

LCFA Transport across Cell Membranes through FadL

Figure 2. (A) RMSDs of the conformation of FadL with respect to the starting conformation throughout the simulation in trajectory A1. In all cases, to remove the rigid-body translational and rotational protein motions, the simulation snapshots were superimposed referring to the TMD. (B) Time dependence of the radius of gyration (Rg) of the TMD and hatch domains in trajectory A1. (C) Time dependence of the volume of the hatch domain (without the seven N-terminal residues) and the corresponding segment of the TMD in trajectory A1. All curves were obtained as averages over 100 ps.

tion A1 were superimposed, and their surfaces are shown in Figure S4 in the Supporting Information, which suggests that the surfaces of the hatch domain and the TMD move in opposite directions; i.e., the hatch domain undergoes a contraction, whereas the TMD tends to spread. These results together indicate that a spontaneous “breathing” motion might occur inside the TMD, as van den Berg et al. proposed,21 and that the intrinsic breathing motion mainly coming from the hatch domain is beneficial to the transport of LCFAs across FadL. This might also be an implication of spontaneous conformational changes occurring during the transport process. van den Berg et al. proposed that a ligand-induced conformational change might be involved in the transport process.21 Therefore, a comparison between the dynamic behaviors of the apo and holo forms of FadL should be necessary. The RMSDs of the hatch domain in both simulations B1 and B2 were smaller than that in simulation A1, indicating that the binding of a fatty acid with the LAS might stabilize the structure of the hatch domain (data not shown). However, a large structural drift of the small domain was also observed in the simulations on FadL in complexes with DA and LDAO in the HAS (simulations C1-C4), suggesting that the binding of a ligand with the HAS does not influence the flexibility of the small domain (Figure S5 in the Supporting Information). Structurally, the ligands in simulations C1 and C2 bind shallowly with the HAS and are

J. Phys. Chem. B, Vol. 112, No. 41, 2008 13073 thus somewhat far from the hatch domain, whereas the ligands in simulations C3 and C4 bind with the pocket deeply so that they might disturb the hatch domain and further facilitate transport. Indeed, the RMSD values of these simulations also imply that the deep binding of fatty acids might affect the structure of the hatch domain. The average RMSDs of the hatch domain in simulations A1, C1, and C2 were 2.2, 2.2, and 1.8 Å, respectively, and the average RMSDs of the hatch domain in simulations A2, C3, and C4 were 1.6, 2.2, and 2.0 Å, respectively. Diffusion of DA from the LAS to the HAS. No experimental structure is available for the complex of a fatty acid binding with the LAS pocket of FadL. Docking simulations were therefore performed to construct the three-dimensional structure of such a complex. Two major binding models of DA in the LAS of FadL were addressed by docking simulations, as shown in Figure S6 of the Supporting Information. For model I, DA lies in the LAS with its carboxylate headgroup toward the HAS. For model II, DA is located in the LAS with its orientation opposite to that in model I. Based on these two binding models, we performed two CMD simulations (simulations B1 and B2). The distance between the centers of mass of the L3 segment and the L4 segment was monitored to describe the swing motion of the L3 segment, and the distance between the centers of mass of DA and the TMD was measured to illustrate the diffusion of DA toward the HAS. Intuitively, the carboxylate group of DA should point toward the HAS because of the electrostatic attraction with two positively charged residues (Arg157 and Lys317) when DA enters the LAS. However, DA was quite stable within the LAS in model I during the 15-ns simulation (Figure 3A), whereas the hydrophobic tail of DA in binding mode II shifted quickly from the LAS to the HAS after ∼1 ns of simulation (Figure 3B). This indicates that the hydrophobic tail of DA enters the LAS first. This phenomenon can be explained by the following points: (1) The LAS is hydrophobic in nature; thus, the hydrophobic interaction of the tail of DA with the pocket is a driving force for DA to enter the LAS. (2) The structures of the LAS and HAS determines that the carboxylate group of DA could not enter the HAS first because the electrostatic interaction between the carboxylate group and the two positively charged residues (Arg157 and Lys317) would constrain DA from moving further into the HAS, as indicated in Figure S6A in the Supporting Information. On the contrary, the hydrophobic tail pointing toward the HAS is beneficial for DA entering the HAS. Moreover, the electrostatic interaction between the carboxylate group and the two positively charged residues becomes another driving force for the diffusion of DA when the tail of DA moves further into the HAS. These simulation results are consistent with the hypothesis proposed by Dirusso et al.49 that the hydrophobic tail of the fatty acid enters the HAS first. Detailed analysis of the simulation results reveals that the motion of DA toward the HAS is accompanied by the swing movement of the L3 segment (Figure 3B). It seems that the swing motion is an intrinsic property of FadL because this motion was also observed in the simulations of apo-FadL (simulations A1 and A2). In addition, the L3 segment moves faster than DA within 1 ns (Figure 3B). Accordingly, in addition to the attractive force of the HAS, the extrusion action of the L3 segment also contributes to the diffusion. Thus, the hydrophobic nature and the intrinsic swing motion of the L3 segment dominate the diffusion process of DA from the LAS to the HAS. Even though the hydrocarbon tail enters the binding pocket first, it cannot fully arrive at the HAS guided by the tail within

13074 J. Phys. Chem. B, Vol. 112, No. 41, 2008

Figure 3. Time-dependent distances between the centers of mass of the L3 and L4 segments and between the centers of mass of DA and the TMD in trajectories (A) B1 and (B) B2. All curves were obtained as averages over 100 ps. (C) Force required to pull DA from the LAS to the HAS in simulation D1. The pore axis is defined as that in Figure 1B.

such a short period of time. Moreover, DA has to adjust its orientation and conformation within the HAS for further transport. A period of 10 ns is not long enough for simulation of this process. Accordingly, simulation D1 was conducted using the SMD approach to pull DA into the HAS starting from the snapshot at 1070 ps of trajectory B2. The forces along the diffusion process are shown in Figure 3C. The maximum force was about 450 pN in the simulation, indicating a relatively easy diffusion of DA from the LAS to the HAS. The resulting binding model of DA with the HAS is similar to that of simulation C2 based on the DA-FadL complex structure in the monoclinic form. Transport from the HAS to the Periplasm. To explore how DA transports from the HAS to the periplasm, we extended simulation D1 by pulling DA into the periplasmic space (simulation D2). The force and free energy along the transport pathway were monitored, as shown in Figure S7 of the Supporting Information. Interestingly, this SMD simulation revealed a completely different transport pathway from that proposed by van den Berg et al.21,50 DA was found to transport through the N-terminus without elimination of the kink and further to cross the gap circled by the L2 segment and the TMD (the CH3 region in Figure 1A). We performed an additional SMD simulation on the X-ray crystal structure of the DA-FadL complex in the hexagonal form (simulation D3), in which the conformational change of the N-terminus was considered as having paved the way for the transport of fatty acids. Simulation

Zou et al. D3 revealed a transport pathway that was the same as that proposed by van den Berg et al.21 These two different pathways for DA transport indicate that multiple pathways might exist for fatty acids permeating the barrel. To map the possible transport pathways and to observe the relevant structural transition during the transport of DA across the barrel, 22 SMD simulations (E1-E22) were conducted starting from different snapshots isolated from trajectory C2. Structurally, the headgroup of DA faced the hatch, and the tail pointed toward the barrel (Figure 1A). Dirusso et al.49 suggested that the transport of fatty acids from the HAS to the periplasm was guided by the headgroup. Also, the previously described CMD simulations indicated that the polar head of either LDAO or DA advanced toward the hatch. Therefore, the potential was applied to the carbonyl carbon of DA for all of the SMD simulations. Detailed information on the 22 SMD simulations is listed in Table S1 of the Supporting Information. The conformational response of FadL to DA transport can be directly examined by comparing the RMSDs between the starting structures and snapshots isolated from the trajectories. The RMSDs of the TMD and hatch domains with respect to their starting structures for the SMD simulations and to the X-ray crystal structure of FadL (starting structure for simulation C2) are listed in Table S1 of the Supporting Information. As shown in this table, the RMSDs of the TMD with respect to the X-ray crystal structure (data in parentheses) in all simulations were greater than 2.0 Å. The average maximum RMSD of the TMD for all 22 trajectories was 2.2 Å, which is slightly higher than that measured from the CMD trajectories (e.g., 2.0 Å for simulation C2). This indicates that the conformation of the TMD does not change much during fatty acid transport. However, the RMSDs of the hatch domain in most of the SMD simulations were over 2.5 Å, and the average maximum RMSD was as much as 2.9 Å, which is much higher than the maximum RMSD observed in simulation C2 (2.3 Å). Accordingly, the SMD simulations demonstrate that the conformational change of the hatch domain of FadL contributes more to the transport of DA. To address the relationship between the transport of DA from the HAS to the periplasm and the conformational change of the hatch domain, the Rg profile of the hatch domain was calculated over the 22 SMD trajectories (Figure S8 in the Supporting Information). Rg was found to decrease as DA approached the hatch domain and increase again as DA moved into the periplasm. This result indicates that the hatch domain might also undergo a breathing motion during the transport of DA, further confirming the breathing motion addressed by the above-described CMD simulations (Figure 2B and C and Figure S4 in the Supporting Information). The forces were monitored during the SMD simulations (Figure 4A), as were the maximum forces and integrated forces (Table S1 in the Supporting Information). The maximum forces to pull DA from the barrel ranged from 700 to 1200 pN, which implies a difficult transport in comparison with the diffusion process from the LAS to the HAS (maximum force of ∼450 pN). Clearly, there are two force peaks in almost all of the SMD trajectories, suggesting that the overall transport of DA from the HAS to the periplasm might consist of two phases. Interestingly, the two phases exactly matched the two-part model of the channel structure. As indicated in Figure 5A, the upper part (U) of the channel consisted of the segment L1 and the kink, and the lower part (L) was mainly composed of the L2, H2, and H3 segments. The first force peak was caused by breaking of the H-bonds formed by DA with Arg157, Ser360, and Thr121 in the HAS. The second peak resulted from the

LCFA Transport across Cell Membranes through FadL

J. Phys. Chem. B, Vol. 112, No. 41, 2008 13075 TABLE 2: Summary of Four Possible Routes for the Transport of DA computed quantities occurrence average maximal force (pN) integrated force (pN · ns) average maximal RMSD (Å): TMD average maximal RMSD (Å): Hatch the segment with the largest RMSF in the hatch domain

path I

path IV

path VI

path IX

3 742

3 920

6 842

3 892

2,315

2,279

2,377

2,437

1.8

1.6

1.7

1.7

(2.3)‡ 2.5

(2.1)‡ 2.2

(2.3)‡ 2.8

(2.2)‡ 2.7

(2.8)‡ L1

(2.6)‡ H2

(2.9)‡ L2

(2.7)‡ L2

‡ The RMSD in parenthesis in rows five and six is computed with respect to the starting structure in the simulation C2.

Figure 4. (A) Pulling force profiles of the 22 SMD simulations (E1-E22). The average force profile is shown as a thick black line. The two major force peaks are highlighted. (B) Average pulling force profiles along pathways I, IV, VI, and IX. The pore axis is defined as that in Figure 1B.

Figure 5. (A) Nine possible pathways revealed by the 22 SMD simulations (E1-E22). The overall transport process is divided into two phases according to force profiles, labeled with U and L. The split point is indicated with an asterisk. The upper part of channel and three major branches in the lower part are highlighted. (B) Cationic track along path VI. The positively charged residues and several fatty acid snapshots in this pathway are represented by sticks. The pathway is highlighted in gray.

strenuous squeezing of DA through the lower part of the channel (CH1, CH2, or CH3); in addition, rupturing of the H-bonds between DA and the lower part of the channel also contributes to the second peak, as DA most frequently forms H-bonds with FadL around this position (Figure S9 in the Supporting Information). The average maximum forces and integrated forces of the four major pathways are listed in Table 2. There is no correlation between these two kinds of forces; further analysis is therefore necessary to characterize each pathway (see the section Energy Landscape of Transport). The 22 SMD simulations addressed nine distinct transport pathways, as shown in Figure 5A. Paths I, IV, VI, and IX showed higher probabilities (number of occurrences g 3) than the other pathways (Table 2 and Table S1 in the Supporting

Information). Therefore, only these four pathways are discussed in detail here. Path VI appears to be the most probable pathway for DA to pass through the membrane protein because it was observed most frequently in the SMD simulations (number of occurrences ) 6; see Table 2). Moreover, simulation D2 also addressed this pathway. Along this pathway, the transport of DA from the HAS to the periplasm is initiated by breaking of the H-bonds formed by DA with Thr121, Ser360, and Arg157. Meanwhile, the aromatic ring of Phe3 flips by ∼180°, covering the kink and exposing the backbone atoms of Phe3 to form H-bonds with the carboxylate headgroup of DA. In this scenario, it is not necessary to destroy the H-bond between Gly103 and Phe3,21 resulting in a well-kept kink. As shown in Figure 6B, several positively charged residues line this pathway that can form H-bonds with DA (Figure S9 in the Supporting Information). These residues might function as a “conveyer belt” for DA transport: (1) The electrostatic attractions of the negatively charged head of DA with the positive side chains of Arg326, Arg228, Arg32, and Arg15 initiate the transport of DA. (2) Afterward, a slight expansion of L2 further makes way for the transport of DA. (3) Finally, another four positively charged residues, Arg342, Arg222, Arg282, and Lys219, drive DA to finish the transport toward the periplasmic space. Meanwhile, the hydrophobic tail of DA does not like to stay in the hydrophilic channel, which is also a driving force for DA to move forward. Unlike path VI, path IX is initiated by the electrostatic attraction of Arg366. Therefore, the transport of DA starts from the backside of the N-terminus of FadL. Except for the initial stage, this pathway adopts the same route as path VI after the split point (Figure 5A). Path I is the same as that proposed by van den Berg et al.,21 and three of the 22 SMD trajectories follow this pathway. Within this pathway, the breakup of the H-bond between Phe3 and Gly103 eliminates the kink and generates a channel for DA to move into the periplasm. Therefore, the largest fluctuation in the hatch comes from segment L1. Three trajectories go along path IV. In fact, path IV is a combination of the upper part of path VI and the lower part of path I. The interaction profile is therefore also the combination of those of paths VI and I. Energy Landscape of Transport. To gain insight into the transport process in terms of the energy landscape, binding free energy profiles of DA to FadL along with SMD trajectories were calculated. To fully address the energy landscape of the overall

13076 J. Phys. Chem. B, Vol. 112, No. 41, 2008

Figure 6. (A) Binding free energy profile of DA with FadL calculated using the AutoDock scoring function in trajectories B2, D1, D3, and E1-E22. The black curve is the binding free energy profile from the LAS to the HAS calculated based on trajectories B2 and D1. The thick magenta line is the average binding free energy profile over the 22 SMD simulations. The free energy profile in simulation B2 (0-1070 ps) is shaded in light gray. The energy barriers are highlighted in gray. The dissociation of DA to solvent is shaded in dark gray. (B) Average binding free energy profiles from the HAS to the periplasm along paths I, IV, VI, and IX and along the trajectory of simulation D3. The energy barriers are highlighted in gray, and the dissociation of DA to solvent is shaded in dark gray. The pore axis is defined as that in Figure 1B.

transport process, the energy profiles for the diffusion process were also calculated. The results are shown in Figure 6, where the binding free energy between DA and FadL can be seen to decrease sharply as DA goes along the diffusion pathway until it arrives at the HAS. This is in good agreement with the hypothesis that fatty acids bind to the low-affinity site first and then can automatically move to the high-affinity site.21 Remarkably, the lowest predicted binding free energy of DA to FadL (from -27 to -33 kJ/mol) is fully consistent with that derived from the experimental binding affinity (from -30 to -34 kJ/ mol),51 suggesting the reasonability of the AutoDock scoring function in calculating the binding free energy between DA and FadL. The 22 SMD simulations based on the monoclinic form (E1-E22) and the one on the hexagonal form (simulation D3) produced similar profiles in terms of the binding free energy from the HAS to the periplasm: after overcoming a very low barrier, which corresponds to the first rupture force peak (Figure 4A), the binding free energy successively decreases. The energy barrier shaded in dark gray near the exit is caused by the dissociation of DA to solvent. As periplasmic tail-specific protease (Tsp) has been implicated in subsequent LCFA binding in the periplasmic space,50 the huge energy barrier near the exit to the periplasmic side does not reflect the true situation. Unlike in the rupture force profile, there is no energy barrier at the position of the second force peak. As mentioned above, the positively charged residues along the pathways can form H-bonds with the negatively charged head of DA (Figure S9 in the Supporting Information), which might eliminate the energy barrier. This supports the hypothesis that the substrate transport of FadL adopts a process similar to passive diffusion through spontaneous breathing of the barrel rather than an active

Zou et al. substrate release process.20 The free energy profiles indicate that the profile of path I agrees well with that of simulation D3, demonstrating again that path I is the same as that proposed by van den Berg et al. (Figure 6B).21 A more detailed analysis for the free energy profiles of the four possible pathways reveals that the energy barriers of paths IV and VI are lower than those of other paths and the overall binding free energy profiles of paths VI and IX are lower than those of paths IV and I (Figure 6B). Accordingly, path VI might be the most probable pathway for DA transport, not only thermodynamically but also kinetically. Correlation between Mutagenesis and Simulation. Mutagenesis analysis indicates that lower incidence of binding was observed in His83Ala mutant.20 van den Berg et al.21 speculated that this mutation might break the H-bond between His83 and Phe8, thereby increasing the local mobility of the hatch domain and subsequently disturbing the high-affinity binding pocket through the displacement of Phe3. Therefore, the H-bonds formed between these residues were monitored during the CMD simulations. For apo-FadL (simulation A1), the occupancies of the Phe3 · · · Gly103 and His83 · · · Phe8 H-bonds were 71.6% and 54.2%, respectively (Table S2 in the Supporting Information). To understand the role of His83 to the binding of substrates, simulation F1 was carried out on the His83Ala mutant of FadL (Table 1). The corresponding occupancy of the Phe3 · · · Gly103 H-bond in the mutant decreased to 13.2%, and all segments in the hatch except the H3 segment of the mutant revealed higher deviations than those of the wild type (Table S2 in the Supporting Information). This indicates that the disruption of the His83 · · · Phe8 H-bond might destabilize the Phe3 · · · Gly103 H-bond, distort the structure of the HAS, and thus decrease the binding affinity of substrates to FadL. Furthermore, we also monitored the profiles of the His83 · · · Phe8 and Phe3 · · · Gly103 H-bonds for the FadL-DA (simulation C2) and FadL-LDAO (simulation C1) complexes (Table S2 in the Supporting Information). Interestingly, substrate binding destabilized both of these H-bonds: The occupancies of these two H-bonds for FadL-DA decreased to 48.3% and 21.4%, respectively, and the two H-bonds in Fad-LDAO almost disappeared, their occupancies falling to 0.6% and 0.4%, respectively. This is in agreement with the conclusion derived from the structural dynamics of FadL, as discussed above, that substrate binding might disturb the structure of the hatch domain. The consistency between the simulations and the mutagenesis data demonstrates that our simulation models and results are reasonable. Experimental data have also shown that the insertion of Glu and Phe between Ala211 and Gly212 results in low fatty acid binding but has only a moderate effect on the transport, suggesting that this mutant has an open channel.4 Insertion of Glu and Phe between Vla383 and Asp384 does not affect the binding of oleate but reduces the rate of oleate transport,4 indicating that this mutant disrupts the transport channel. To verify the reasonability of our simulation models and results, we conducted two simulations on these two mutants (F2 and F3, respectively, in Table 1). The average structures of the hatch domain in simulations F2 and F3 were superimposed on that of the wild type to observe the conformational changes (Figure S10 in the Supporting Information). The average radii of the CH1 and CH3 regions measured by the HOLE program52-54 for simulation F2 went from 1.04 to 1.17 Å and from 0.91 to 1.01 Å, respectively (Table S3 in the Supporting Information), whereas these radii for simulation F3 showed an apparent contraction, decreasing to 0.72 and 0.80, respectively. However, the radii of the CH2 region in simulations F2 and F3 decreased almost to zero (channel not found). Furthermore, this is also

LCFA Transport across Cell Membranes through FadL

Figure 7. Postulated mechanism for the transport of an LCFA across FadL. The LCFA is shown in red stick form. The LAS, HAS, and hatch domain are represented by cyan, yellow, and gray surfaces, respectively. Two positively charged residues (Arg157 and Lys317) lining the HAS are represented by magenta spheres. (a) LCFA initially binds to the LAS between extramembrane loops L3 and L4. (b) Subsequent diffusion of the LCFA from the LAS to the HAS takes place as a result of the intrinsic swing motion of the L3 segment (the shift from the cyan surface to green one) and hydrophobic attractions. (c) LCFA then becomes bound in the high-affinity site that is linked to the hatch domain, and a ligand-induced perturbation of the hatch domain follows, as can be seen from the watermarked surface of the hatch domain. (d) Conformational change within the N-terminus of FadL facilitates the opening of fatty-acid-specific channels and transport to the periplasmic space. Nine pathways were found to be involved in this process for DA. The three bars with fatty acids in the lower part represent three major channel directions, which correspond to the CH1, CH2, and CH3 regions, respectively, in Figure 1A.

supported by the H-bond and distance between Gln33 and Ala21 in the CMD simulations and occurrences in the SMD simulations. In simulation A1, this H-bond existed all along the simulation with an occupancy of 81%. Similar results were found in simulations A2 and C1-C4. Moreover, the distance between Gln33 and Ala21 was also very stable (data not shown). In the above SMD simulations, 10 of the 22 SMD simulations were through CH3, 9 were through CH1, and only 2 were through CH2. The agreement between the results from experimental mutation and those from simulations F2 and F3 suggests that transport through either CH1 or CH3 is possible. This demonstrates again that multiple pathways exist for FadL transport of fatty acids and different pathways dominate the transport through different substrates; for example, pathway VI seems to be the most probable one for DA permeating FadL. Possible Transport Mechanism of FadL. Based on our simulation results and analyses, we can propose an atomic mechanism for the transport of fatty acids (taking DA as an example) through FadL as indicated in Figure 7. (1) One fatty acid molecule attaches to the highly hydrophobic groove between the L3 and L4 segments with the hydrophobic tail pointing toward the LAS, and the hydrophobic attraction from the LAS steers the fatty acid to move to the LAS of FadL. (2) Subsequently, the fatty acid diffuses to the HAS pocket as a result of the pushing force from the intrinsic swing motion of the L3 segment, and then, the fatty acid reverses its orientation within the HAS. (3) Meanwhile, the impact force of the fatty acid perturbs the stability of the hatch domain, which makes passageways for the fatty acid to cross the barrel. (4) For transport from the HAS to the periplasm, the fatty acid goes through the gap between the hatch domain and the barrel wall through several routes, i.e., a multiple-pathway mechanism. For example, nine pathways exist for DA permeating the barrel, and path VI might be the major pathway (Figure 5). Conclusions The present MD simulations have provided new insight into the transport mechanism of fatty acids across FadL at the atomic level (Figure 7), extending our understanding of the membrane-

J. Phys. Chem. B, Vol. 112, No. 41, 2008 13077 protein-mediated transport of small organic molecules. The MD simulations clearly addressed the intrinsically dynamic behavior of FadL (Figure 2). In particular, a spontaneous breathing motion mainly from the hatch domain was observed in the CMD simulations (Figure 2B and C), indicating that the transport of FadL accompanies a spontaneous conformational change. A ligand-induced conformational change of FadL was also observed in the CMD simulations. The binding of fatty acid with the LAS might stabilize the hatch domain, whereas binding with the HAS increases the flexibility of the hatch domain. Both the CMD and SMD simulations indicated that a fatty acid can diffusively move to the HAS from the LAS. During this process, the swing motion of the L3 segment and the hydrophobic interaction between the fatty acid and FadL might play important roles. The 22 SMD simulations revealed that fatty acids can pass through the gap between the hatch domain and the inner wall of the barrel through different pathways (e.g., nine pathways are possible for DA; Figure 5). However, for each fatty acid, one dominant pathway exists (e.g., path VI for DA; Figure 5B). The binding free energy profiles between DA and FadL along the MD trajectories indicated that all of the possible pathways are energetically favorable for fatty acids going through the barrel and also revealed that paths I, IV, VI, and IX are the most probable pathways (Figure 6). The reasonability and reliability of this study were further demonstrated by correlating the MD simulation results with the available mutagenesis results. Finally, a mechanism for the full-length transport process of DA from the extracellular side to the periplasmic space mediated by FadL was proposed (Figure 7). Acknowledgment. This work was supported by the State Key Program of Basic Research of China (Grant 2002CB512802). Shanghai Supercomputing Center and Computer Network Information Center are acknowledged for the allocation of computing time. Supporting Information Available: Additional computational results in figures and tables are included. This material is free of charge via the Internet at http://pubs.acs.org. References and Notes (1) Das, U. N. J. Assoc. Phys. Ind. 2006, 54, 309–319. (2) Stubbs, C. D.; Smith, A. D. Biochim. Biophys. Acta 1984, 779, 89–137. (3) Hirsch, D.; Stahl, A.; Lodish, H. F. Proc. Natl. Acad. Sci. U.S.A. 1998, 95, 8625–8629. (4) Kumar, G. B.; Black, P. N. J. Biol. Chem. 1991, 266, 1348–1353. (5) Auvin, S.; Collet, F.; Gottrand, F.; Husson, M. O.; Leroy, X.; Beermann, C.; Guery, B. P. Pediatr. Res. 2005, 58, 211–215. (6) Dutta-Roy, A. K. Cell. Mol. Life Sci. 2000, 57, 1360–1372. (7) Abumrad, N.; Coburn, C.; Ibrahimi, A. Biochim. Biophys. Acta 1999, 1441, 4–13. (8) Black, P. N.; Said, B.; Ghosn, C. R.; Beach, J. V.; Nunn, W. D. J. Biol. Chem. 1987, 262, 1412–1419. (9) Black, P. N. J. Bacteriol. 1988, 170, 2850–2854. (10) Harmon, C. M.; Abumrad, N. A. J. Membr. Biol. 1993, 133, 43– 49. (11) Stremmel, W.; Strohmeyer, G.; Borchard, F.; Kochwa, S.; Berk, P. D. Proc. Natl. Acad. Sci. U.S.A. 1985, 82, 4–8. (12) Berneche, S.; Roux, B. Nature 2001, 414, 73–77. (13) Doyle, D. A.; Morais Cabral, J.; Pfuetzner, R. A.; Kuo, A.; Gulbis, J. M.; Cohen, S. L.; Chait, B. T.; MacKinnon, R. Science 1998, 280, 69– 77. (14) de Groot, B. L.; Grubmuller, H. Science 2001, 294, 2353–2357. (15) Tajkhorshid, E.; Nollert, P.; Jensen, M. O.; Miercke, L. J.; O’Connell, J.; Stroud, R. M.; Schulten, K. Science 2002, 296, 525–530. (16) Nunn, W. D.; Simons, R. W. Proc. Natl. Acad. Sci. U.S.A. 1978, 75, 3377–3381. (17) Nikaido, H. Microbiol. Mol. Biol. ReV. 2003, 67, 593–656. (18) Cristalli, G.; DiRusso, C. C.; Black, P. N. Arch. Biochem. Biophys. 2000, 377, 324–333.

13078 J. Phys. Chem. B, Vol. 112, No. 41, 2008 (19) Kumar, G. B.; Black, P. N. J. Biol. Chem. 1993, 268, 15469–15476. (20) Black, P. N.; Zhang, Q. Biochem. J. 1995, 310, 389–394. (21) van den Berg, B.; Black, P. N.; Clemons, W. M., Jr.; Rapoport, T. A. Science 2004, 304, 1506–1509. (22) Matthew, J. B. Annu. ReV. Biophys. Biophys. Chem. 1985, 14, 387– 417. (23) Matthew, J. B.; Gurd, F. R. Methods Enzymol. 1986, 130, 413– 436. (24) Northrup, S. H.; Laughner, T.; Stevenson, G. MacroDox, version 3.2.1; Tennessee Technological University: Cookeville, TN, 1999. (25) van Aalten, D. M.; Bywater, R.; Findlay, J. B.; Hendlich, M.; Hooft, R. W.; Vriend, G. J. Comput. Aided Mol. Des. 1996, 10, 255–262. (26) Breneman, C. M.; Wiberg, K. B. J. Comput. Chem. 1990, 11, 361– 373. (27) Morris, G. M.; Goodsell, D. S.; Halliday, R. S.; Huey, R.; Hart, W. E.; Belew, R. K.; Olson, A. J. J. Comput. Chem. 1998, 19, 1639–1662. (28) Faraldo-Go’mez, J. D.; Smith, G. R.; Sansom, M. S. P. Biophys. J. 2003, 85, 1406–1420. (29) Domene, C.; Bond, P. J.; Deol, S. S.; Sansom, M. S. P. J. Am. Chem. Soc. 2003, 125, 14966–14967. (30) Carpenter, T.; Khalid, S.; Sansom, M. S. P. Biochim. Biophys. Acta 2007, 1768, 2831–2840. (31) Im, W.; Roux, B. J. Mol. Biol. 2002, 319, 1177–1197. (32) Bond, P. J.; Derrick, J. P.; Sansom, M. S. P. Biophys. J. 2007, 92, L23–L25. (33) Yau, W. M.; Wimley, W. C.; Gawrisch, K.; White, S. H. Biochemistry 1998, 37, 14713–14718. (34) Gurtovenko, A. A.; Patra, M.; Karttunen, M.; Vattulainen, I. Biophys. J. 2004, 86, 3461–3472. (35) Berendsen, H. J. C.; van der Spoel, D.; van Drunen, R. Comput. Phys. Commun. 1995, 91, 43–56. (36) Lindahl, E.; Hess, B.; van der Spoel, D. J. Mol. Model. 2001, 7, 306–317.

Zou et al. (37) van Gunsteren, W. F.; Berendsen, H. J. C. GROMACS Force Field; University of Groningen: Groningen, The Netherlands, 1987. (38) Hess, B.; Bekker, H.; Berendsen, H. J. C.; Fraaije, J. G. E. M. J. Comput. Chem. 1997, 18, 1463–1472. (39) Darden, T.; York, D.; Pedersen, L. J. Chem. Phys. 1993, 98, 10089– 10092. (40) Essmann, U.; Perera, L.; Berkowitz, M. L.; Darden, T.; Lee, H.; Pedersen, L. G. J. Chem. Phys. 1995, 103, 8577–8592. (41) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; DiNola, A.; Haak, J. R. J. Chem. Phy. 1984, 81, 3684–3690. (42) Kosztin, D.; Izrailev, S.; Schulten, K. Biophys. J. 1999, 76, 188– 197. (43) Isralewitz, B.; Gao, M.; Schulten, K. Curr. Opin. Struct. Biol. 2001, 11, 224–230. (44) Izrailev, S.; Stepaniants, S.; Balsera, M.; Oono, Y.; Schulten, K. Biophys. J. 1997, 72, 1568–1581. (45) Koebnik, R.; Locher, K. P.; Van Gelder, P. Mol. Microbiol. 2000, 37, 239–253. (46) Lou, K. L.; Saint, N.; Prilipov, A.; Rummel, G.; Benson, S. A.; Rosenbusch, J. P.; Schirmer, T. J. Biol. Chem. 1996, 271, 20669–20675. (47) Pautsch, A.; Schulz, G. E. Nat. Struct. Biol. 1998, 5, 1013–1017. (48) Laskowski, R. A. J. Mol. Graph. 1995, 13, 323–330. (49) Dirusso, C. C.; Black, P. N. J. Biol. Chem. 2004, 279, 49563– 49566. (50) van den Berg, B. Curr. Opin. Struct. Biol. 2005, 15, 401–407. (51) Black, P. N. Biochim. Biophys. Acta 1990, 1046, 97–105. (52) Smart, O. S.; Goodfellow, J. M.; Wallace, B. A. Biophys. J. 1993, 65, 2455–2460. (53) Smart, O. S.; Neduvelil, J. G.; Wang, X.; Wallace, B. A.; Sansom, M. S. P. J. Mol. Graph. 1996, 14, 354–360. (54) Smart, O. S.; Breed, J.; Smith, G. R.; Sansom, M. S. P. Biophys. J. 1997, 72, 1109–1126.

JP710964X