|Home | About | Journals | Submit | Contact Us | Français|
Simulations of antimicrobial peptides in membrane mimics can provide the high resolution, atomistic picture that is necessary to decipher which sequence and structure components are responsible for activity and toxicity. With such detailed insight, engineering new sequences that are active but non-toxic can, in principle, be rationalized. Armed with supercomputers and accurate force fields for biomolecular interactions, we can now investigate phenomena that span hundreds of nanoseconds. Although the phenomena involved in antimicrobial activity, (i.e., diffusion of peptides, interaction with lipid layers, secondary structure attainment, possible surface aggregation, possible formation of pores, and destruction of the lipid layer integrity) collectively span time scales still prohibitively long for classical mechanics simulations, it is now feasible to investigate the initial approach of single peptides and their interaction with membrane mimics. In this article, we discuss the promise and the challenges of widely used models and detail our recent work on peptide–micelle simulations as an attractive alternative to peptide–bilayer simulations. We detail our results with two large structural classes of peptides, helical and beta-sheet and demonstrate how simulations can assist in engineering of novel antimicrobials with therapeutic potential.
The goal of research in the area of antimicrobial peptides (AMPs) is to reduce the host-cell toxicity levels of active peptides by intelligent mutations to facilitate the design of novel antibiotic molecules. Despite considerable progress and advancements in the use of various experimental and computational techniques, the scope of rationalizing the design of effective AMPs remains limited . The key bottleneck in the process is the lack of availability of molecular details of peptide–lipid interactions that eventually drive cell-death. It has been oft-repeated that experiments have not been able to resolve the fast time scales of peptide–membrane interactions. Molecular dynamics simulations of AMPs with model membranes have led to useful insights into the types of biophysical interactions that drive peptide–membrane association . Although the length and time-scales presently accessible to these simulations limit the amount of useful molecular information that could be of use to design efforts, the future of such methods is looking brighter. In the current article, we will discuss some of our recent work on AMP simulations in micellar environments. The initial part of the paper will be devoted to drawing a short summary of the limitations and scope of some of the recent progress made in all-atom molecular dynamics simulations of AMP–membrane systems. For a review of advances made before 2001, please refer to La Rocca et al. . Please note that this article is not a comprehensive review of peptide–membrane simulations in general; the focus is ultimately on all-atom molecular dynamics simulations of AMP–membrane systems.
The mechanism of action of cell-death induced by AMPs putatively involves initial binding to the membrane, induction of peptide secondary structure (in most cases), local aggregation of peptides at the membrane, and subsequent cell lysis caused by membrane destabilization . Unfortunately, limitations on current computing capacity still preclude the investigation of peptide aggregation and pore/carpet formation on the membrane. Thus, one of the key goals of simulations is to shed light on the initial events involved in the binding of an AMP to a membrane. This involves deduction of the important biophysical interactions between the AMP and the membrane, and the quantification of the effect of peptide binding on membrane dynamics and structure and vice versa. Based on these observations, it is possible to identify the role of individual amino acids in the peptide’s interaction with the membrane. In principle, this information can then be used to design mutants of the peptide to improve activity against microorganisms, or to attenuate toxicity against host cells.
Hydrated phospholipid bilayers, surfactant micelles, and phospholipid monolayers have been used to model membrane interfaces. Typically, the simulation of a single AMP near a hydrated model membrane is started with the peptide placed either in the aqueous phase  or positioned in the membrane in an orientation suggested by experiments . We will refer to these two types of simulations as type 1 and type 2 respectively. The former type of simulations have broader scope because the binding dynamics of the peptide to the membrane can be investigated. In either case, experiments provide the initial structure of the peptide. The simulations are usually implemented in the constant-temperature, constant-pressure (NPT) statistical ensemble. When lipid bilayers are used as membrane models, the NPT ensemble must be used in order to allow the area per lipid in the membrane to adjust to the insertion of the peptide. However, type 1 simulations of peptides in hydrated phospholipid bilayers run into convergence problems [5,7]. The same is true when monolayers are used . One key objective of an AMP–membrane simulation is to correctly identify the final binding state of the peptide near the membrane. In order to do so with a high degree of confidence, it must be shown that different initial conformations of the peptide lead to a similar final ensemble of states. Kandasamy et al. found that convergence was not achieved within time scales of ~20 ns for magainin peptides in POPC bilayers . We have carried out simulations of the helical AMP ovispirin and the β-sheet peptide protegrin-1 near lipid micelles [9–12]. When the peptide starts out in the aqueous phase, we have found that the final binding orientation of the peptide depends upon which face (hydrophobic or hydrophilic) of the amphipathic peptide is initially oriented towards the micellar interface. On the other hand, Lensink et al.  were able to demonstrate convergence to the same final state from different starting conformations of the cell penetrating peptide penetratin in POPC and POPG bilayers. To our best knowledge, this is the only case of type 1 simulations in literature where such convergence has been established in NPT simulations of AMPs near lipid bilayers. Unfortunately, the timescales of convergence these simulations were as high as 290 ns (in POPC) and 150 ns (in POPG). Although such simulations are feasible with current computing power, the very long time scales of convergence set limits on the number of such simulations that one can implement in a given time frame. Typically, one wants to evaluate and compare the membrane interaction properties of several different related AMPs in order to identify the role of specific mutations which differentiate the microbicidal and toxic properties of the peptides. For the reasons noted above, all-atom hydrated lipid bilayers are perhaps not the ideal model to implement a large number of simulations.
In order to overcome some of these shortcomings, we have implemented simulations of several sets of related AMPs in zwitterionic dodecylphosphocholine (DPC) and anionic sodium dodecylsulphate (SDS) micelle systems. As models of membrane interfaces, micelles offer several advantages over lipid bilayers. Micelles are useful systems to study the influence of interfacial membrane electrostatics on the structural properties of small peptides. The primary advantage of using micelles as opposed to lipid bilayers is the faster time scales of motion of DPC and SDS molecules. It has been shown both experimentally  and by simulations [15–18] that the slowest relaxation times of lipids in micellar solutions are of the order of 500–2000 ps. On the other hand, the area-per-lipid in bilayers relaxes on timescales of the order of ~20000 ps . The micelle contains about 75% the number of atoms of a typical 128-lipid peptide–water-bilayer simulation cell. This allows much longer simulations and permits monitoring of biological phenomena of longer time scales. In our recent work, we have shown that the micelle model is successful in capturing experimentally-observed binding states of small peptides [9–11]. DPC lipids carry a phosphocholine head group like zwitterionic membrane phospholipids, and the DPC micelle is thus well suited as a model to study peptide–membrane interactions on a zwitterionic interface. The head group of the SDS micelle is very different from anionic phospholipids, but still offers an anionic lipid–water interface at which peptide binding, and changes in peptide structure can be monitored. The most significant advantage of using spherical micelles as opposed to planar lipid bilayers is that the need for implementing several different simulations with different initial conformations is precluded. We will discuss this in more detail in the methods section. Indeed, there are several limitations in the use of micelles as models of membrane interfaces; and these are pointed out in the conclusions section.
We have carried out simulations for several different cathelicidin antimicrobial peptides. Cathelicidins are one of the two major families of mammalian AMPs, and cathelicidin peptides of all three principal structural classes (α-helical, β-sheet, and unstructured) have been characterized. We have implemented simulations for each structural class.
Helical peptides are the most abundant class of antimicrobial peptides, and the mechanism of action of various such peptides has been thoroughly investigated in literature. For representative examples, please refer to [20,21]. Ovispirin-1 (OVIS) is an 18-amino acid (KNLRR IIRKI IHIIK KYG) helical AMP, and has significant antimicrobial activity. It is unsuitable for therapeutic use owing to its hemolytic properties . Novispirin G10 (KNLRR IIRKG IHIIK KYG) and Novispirin T7 (KNLRR ITRKI IHIIK KYG) are single residue mutants of OVIS which retain the antibacterial activity, but are less toxic. The three dimensional structures of all three peptides were evaluated in presence of tri-fluoro-ethanol (TFE) (Fig. 1). The point mutations led to significant structural changes in the peptides, inducing a helix kink in G10 and N-terminal flexibility in T7 . The structural and functional properties of these three peptides make them well suited to carry out investigations of the interplay between membrane-influenced peptide structure and peptide–membrane interactions.
The protegrins are a family of five potent, naturally occurring cationic antimicrobial peptides that were originally purified from porcine leukocytes [24,25]. They have a β-hairpin structure that is held in place by two disulfide bonds. Protegrin-1 [PG-1, RGGRL CYCRR RFCVC VGR] can launch a rapid response to infection by diverse bacterial species  including Escherichia coli, Candida albicans, and Listeria monocytogenes . We have obtained data from our co-workers at UCLA, who have created sixty protegrins based on the sequence of PG-1. They have provided us with data for the activity for these protegrins against various microbial species and also cytotoxicity and hemolysis data. As of now, we have completed simulations of 8 protegrins in SDS and DPC micelles. For the protegrins we will be discussing here, these data are shown in Table 1.
We have also carried out simulations of indolicidin (ILPWK WPWWP WRR), for which no well-defined structure has been isolated yet. The results from these simulations will be reported elsewhere.
We have implemented simulations of AMPs in zwitterionic DPC and anionic SDS micelles. In the case of DPC micelles, the starting coordinates of the micelle–water complex were obtained from simulations carried out by Wymore et al. . This structure was obtained after extensive minimization and dynamics of about 1 ns in a cubic simulation cell. In the case of SDS micelles, the starting coordinates of the micelle–water complex were obtained from simulations carried out by MacKerell . This structure was obtained after extensive minimization and 120 ps of NVT and NPT simulation in a cubic simulation cell. SDS and DPC were both parameterized using the CHARMM force field. In either case, the micelle was placed in a cubic simulation box of cell size 56.15 Å. The cell dimensions were setup so as to obtain the equilibrium bulk water density (0.033/Å3) far away from the interface. Water was modeled using the TIP3P potential . 5 Na+ and Cl− ions were added as 0.15 mM electrolyte. In SDS, chloride counterions were distributed with random coordinates in the aqueous phase to keep the system electrostatically neutral.
In principle, to observe the folding of the peptide on the membrane interface, the simulation should start with a random coil peptide structure. However, the very long time scales of peptide folding on membrane interfaces prohibit such simulations. The peptide structure obtained from experiments, as reported in the pdb data bank, were used to start the simulations. For protegrins, only the structure of PG1 is available in the pdb data bank. Homology modeling was used to obtain the structures for the other peptides.
Solid-state NMR experiments of OVIS in lipid bilayers  suggest that the majority of the OVIS helix is oriented parallel to the interface, with the hydrophobic face embedded in the lipid core. We did not place the peptide at the micelle–water interface to avoid biasing final conformation of the simulation. Instead, each peptide was placed in the simulation box with its center of mass coinciding with that of the micelle. In this conformation (CONF1), the peptide helix lay along one of the diameters of the micelle, with only its termini exposed to the solvent interface (Fig. 3A and C). Owing to the spherical symmetry of the micelle, the orientation of the peptide is unimportant, and thus the need to start the simulation from different starting orientations with respect to the membrane interface is eliminated (Fig. 2). Simulations were also run with the peptide initially placed in the aqueous phase (CONF2). The final conformations obtained from unbiased simulations of type CONF1 were used to bias the initial orientation of the peptide w.r.t. the membrane interface in simulations of type CONF2. Thus, for each helical peptide in each micelle, we carried out 2 simulations. One where the peptide is initially placed along a micelle diameter (unbiased); and the other where the peptide is initially placed in the aqueous phase, but the initial orientation of the peptide is guided by the results of the first simulation.
To remove initial bad contacts between the peptide and the micelle core, and prevent penetration of water, the peptide and bulk water were kept under weak harmonic constraints with spring constants of 10 and 5 kcal/mol Å, respectively during equilibration. The constraints were gradually removed in 20,000 steps of minimization (steepest descent method). The entire system was then minimized for 20,000 more steps without any constraints. Thereafter, the system consisting of about ~16,000 atoms was gradually heated to 303 K. The entire assembly was subjected to NPT dynamics at pressure P=1 atm and temperature T =303.15 K for 40 ns. The constant pressure–temperature module of CHARMM  was used for the simulation with a leap-frog integrator. A time step of 2 fs was used. The temperature was set at 303.15 K using Nose–Hoover temperature control . For the extended system pressure algorithm employed, all the components of the piston mass array were set to 500 amu . The electrostatic interactions were simulated using the particle mesh Ewald (PME) summation  without truncation and a real space Gaussian width of 0.25 Å−1, a β-spline order of 4, and a FFT grid of about one point per Å. The non-bonded van der Waals interactions were smoothly switched off over a distance of 3.0 Å, between 9 Å and 12 Å. SHAKE was used to eliminate the fastest degrees of freedom involving bonds with hydrogen atoms. The simulations were carried out using CHARMM version c30b2 with the all atom param22 parameter set. We have shown earlier that unconventional π-helices are not formed with the param22 parameter sets in peptide-micelle simulation setups . For calculation of most dynamical properties, trajectories were sampled every 10 ps.
Simulations with the protegrins were carried out using the same methodology. For some smaller protegrins, the steady state conformations were achieved within 10 ns, and these simulations were run for a total of ~25 ns. Most of these simulations were carried out using only the CONF1 conformation.
All three peptides converge to the interfacial-bound steady state in both types of micelles in nearly 40 ns of simulations (Fig. 3). In either micelle, the hydrophobic surfaces of the peptides are embedded in the micellar core, while the hydrophilic face remains exposed to water and the lipid head groups. The long axis of the peptide lies parallel to the micelle surface tangent. For OVIS, these observations are in good agreement with NMR experiments carried out in lipid bilayers . At the time this article was being prepared, there was no experimental data available for G10 and T7.
Fig. 4 shows two indicators of the convergence of the CONF1 and CONF2 simulations. In Fig. 4a, we have plotted the distance between the center of mass of the micelle and the peptide from the two different starting conformations of OVIS in DPC micelles. In Fig. 4b, the orientation of the peptide w.r.t. the micelle is plotted for the same set of simulations. The orientation of the peptide w.r.t. the micelle was measured by calculating the angle between the peptide helical axis, and the vector from the center of mass of the micelle to the center of mass of the peptide. Throughout the simulations, the peptide remains more or less parallel to the surface of the micelle. In either case (CONF1 or CONF2), the peptides diffuse to the interface bound conformation within 20–25 ns. The simulations were run for another ~20 ns, during which the position and structure of the peptide did not change significantly. The last 10 ns of each simulation were used to calculate ensemble averaged properties. Most of the results presented for this section will be from the more recent CONF2 simulations.
Importantly, simulations of type CONF1 and CONF2 both lead to the same final peptide conformations and orientations. The convergence has similar trends for all three peptides, in both types of micelles. There are only slight differences in the backbone dihedral angle values, and these do not significantly affect the position of each residue in the Ramachandran plot (data not shown).
The binding depths of the peptides are similar in SDS, which correlates with their equally potent antibacterial properties. However, OVIS, the most toxic peptide, is embedded deepest into the DPC micellar interface (Fig. 5). G10 and T7, on the other hand, are less buried in the DPC micelle. The relative depths of insertion in DPC micelles correlate well with the more toxic properties of OVIS, and the lower toxicity of G10 and T7.
A clustering algorithm of peptide backbone dihedral angles (,ψ) was implemented in order to check if new peptide conformations were observed in the production period. Time series of the peptide dihedral angles were obtained from different initial timepoints tini in the trajectory. tini was varied from tini =0 ns to tini =33.6 ns with a 2.4 ns interval. Thus, each set of time series contained the dihedral angle values for the peptide for a trajectory window starting at time tini, till the end of the simulation. Each set of time series was clustered using the ART-2 clustering algorithm in CHARMM. The number of clusters thus obtained is a good measure of the number of different peptide conformations observed during a trajectory window. The results of this clustering analysis for the CONF2 DPC simulations are shown in Fig. 6. The simulation was run for 44 ns in this case. No new peptide conformations are sampled during the production period of the simulation for any of the three peptides. The peptide G10 does form 2 clusters at tini =31.2 ns. However, the cluster disappears if a slightly larger cluster radius is chosen, indicating that the 2 clusters are very close to each other, and represent very similar peptide conformations. We believe that such a calculation is an excellent indicator of the approach of a simulation towards peptides’ conformational equilibrium. Additionally, at tini =0, T7 and G10 have a higher number of clusters than OVIS, indicating that these two peptides sample a much larger variety of conformational states during the full course of the simulation, which correlates with their greater flexibility.
The helical structure of OVIS was further stabilized compared to its experimental structure in TFE, as the peptide diffused to the interface. All the (i,i+4) backbone–backbone hydrogen bonds were satisfied, leading to a compact interface-bound helix which lay parallel to the interface (Figs. 3, ,7).7). On the other hand, both G10 and T7 became less helical in DPC micelles. The experimental TFE-induced structure of G10 has a bend as positions 12 and 13 induced by inclusion of a glycine at position 10 . The bend in the G10 helix in DPC extended from positions 7 through 11. This results in partial unfolding of the second N-terminal helix turn. The flexibility is enhanced in presence of the DPC micelle. Residue 6 (Ile) also acquires non-native dihedral angles.
The structures of the three peptides are as helical in SDS micelle simulations as they are in the experimentally obtained structures in TFE. The SDS lipids do not induce any unfolding of the helices. In fact, the helices in SDS are tighter and conform better to the native helical (ϕ,ψ) pair value of (−50, −60) (data not shown).
The structure of OVIS remains amphipathic in the presence of either DPC or SDS micelles. However, the less toxic G10 and T7 become less amphipathic in DPC micelles. The reduced amphipathicity of G10 and T7 leads to weaker interaction of hydrophobic residues with the DPC micellar core, (as revealed by radial distribution functions), eventually leading to weaker binding to the DPC micelle. The implications of these subtle changes in secondary structure are expounded upon in the next section.
Interestingly, the binding behavior of charged residues of OVIS to the DPC vs. the SDS micelles was qualitatively similar. In either case, radial distribution functions (rdfs) revealed formation of hydrogen bonds between the positively charged amino acid side chains and either the phosphate oxygens in DPC, or the sulfate oxygens in SDS. However, the relative heights of the peaks indicate stronger H-bonds and interactions in SDS. This is to be expected from the differences in the electrostatic surfaces of the two micelles. In DPC, the choline group can screen the electrostatic interaction between the charged residues of the peptide and the phosphate groups both by steric hindrance and by the electrostatic repulsion from the cationic nitrogen. Similar results are obtained for G10 and T7 as well. The possible implications of these observations are discussed in the following section.
We have implemented MD simulations of an active and toxic helical AMP and its two active, but non-toxic analogues in zwitterionic micelles. Simulations were also carried out in SDS micelles to compare the influence of membrane electrostatics on peptide secondary structure and toxicity. The goal of the study was to dissect the differences in the peptides that make the analogues (G10 and T7) less toxic compared to the native peptide (OVIS). The simulations converged to their final equilibrium state with respect to the distance between the center of masses of the peptide and the micelle, as well as with respect to the orientation of the peptide w.r.t the micelle. We also confirmed that no new secondary structural features were appearing during the ensemble-sampling period of the simulations. Moreover, simulations were started from two very different conformations and converged to the same final state. Despite these substantiations to argue for the convergence to the equilibrium conformation in phase space, there is always a doubt as to whether more sampling is required to guarantee that equilibrium has been achieved, and that the conformations observed in the current simulations represent true free energy minima. The results in this work should be interpreted in this light.
There are no significant differences in the three peptides with respect to interaction of polar residues with the DPC micelle. The simulations explicitly demonstrate that hydrophobic interactions drive the strong association of the OVIS peptide with the DPC lipids. This observation has been widely documented in literature for various helical AMPs. The key phenomenon that the simulations indicate is that the most toxic peptide, OVIS, remains helical, while the less toxic G10 and T7 lose helicity in presence of DPC. The stabilization of the helical structure of OVIS in the presence of DPC lipids allows the hydrophobic residues of the peptide to bind cooperatively to the micelle. The reduced helicity induced in G10 and T7 by DPC leads to lower amphipathicity resulting in a lower depth of insertion into the micelle. For this reason, G10 and T7 are suggested to be less toxic than OVIS. Both G10 and T7 have one less hydrophobic residue than OVIS, and indeed, this also contributes to the lower toxicity, but ramifications of the point mutations in G10 and T7 extend well beyond the reduction of total hydrophobic content. The mutations induce critical changes in peptide secondary structure that prevents the cooperative isolation of hydrophobic residues into the membrane core. The situation is different in SDS micelles, where all three peptide retain their secondary structure, and remain equidistant from the micellar center of mass. The comparable depth of binding of all three peptides to SDS, and the lesser binding depth of G10 and T7 in DPC correlate perfectly with the high antimicrobial activity of all three peptides, and the reduced toxicity of G10 and T7. We still do not have a clear answer to the question “why are the structures of G10 and T7 more helical in SDS, than in DPC ?”. It is known that electrostatic interactions between cationic sidechains of unstructured peptides and the anionic lipid headgroups drive the initial binding of AMPs to bacterial membrane interfaces. The rdfs described in Fig. 8 explicitly quantify the stronger attraction to anionic interfaces, compared to zwitterionic interfaces. It is possible that a greater degree of adsorption to anionic interfaces (driven by the strong initial electrostatic force) allows the peptide to take a more amphipathic helical form. However, such hypothesis can only be justified when the folding events of unstructured peptides near interfaces are investigated in simulations, or in experiments. Unfortunately, these simulations are still not computationally tractable.
It is believed that the electrostatic differences in the composition of bacterial and mammalian plasma membranes are responsible for selective toxicity of AMPs. However, besides attracting cationic AMPs, the anionic surface of bacterial membranes might influence other processes which affect selectivity. It is known that most AMPs are unstructured in solution, and achieve their helical form after binding to the membrane. The extent of helical content induced in an AMP in the presence of a membrane interface can influence its hemolytic and antibacterial properties. Generally, no comprehensive attempt has been made to correlate peptide toxicity with the differences in the abilities of anionic and zwitterionic membranes to induce secondary structure in AMPs. Our work tries to fill this void. For a more detailed discussion of these arguments, please refer to . Based on our simulations and the above discussion, we suggest that mutations which marginally lower peptide helical content should reduce toxic properties by lowering overall peptide amphipathicity. Interestingly, we found at least two such examples in literature. These are the helical AMPs pleurocidin [34,35] and IsCT and their analogues . In both cases, less helical mutants were less toxic to mammalian cells.
The study of protegrin-1 (PG-1) and its mutants have shown promise for the potential development of a protegrin-derived antibiotic peptide. We are currently in the process of simulating the sixty protegrin mutants for which activity and toxicity data are available. We have completed simulations of a range of peptides with varying levels of activity and toxicity and have begun to determine rules for the design of new protegrins. In Fig. 6, we show an example of the initial and final configurations for PG-1 in an SDS micelle.
In our original work with protegrin-1 (PG-1), we found that Leu-5 plays an integral role in the peptide’s interactions with SDS, but does not interact with the DPC micelle . We saw that the other side of the β-hairpin, mainly Phe-12 and Val-14 are the residues most involved in the disruption of the DPC micelle. We have designed a peptide based on these results that we hope will have a lower toxicity and are presently testing it.
In our work with the much smaller protegrin mutants PC72 and PC73, we found that the presence of the N-terminal leucine residue on PC72 is responsible for the activity of this peptide against microbial species (unpublished results). This was the expected conclusion simply based on the sequences, but we further examined the individual residues’ roles in the toxicity of PC72 and have determined that the region near Tyr-3, should have its hydrophobic content reduced to create a less toxic version of this peptide. We are currently testing peptides based on this finding.
We have additionally examined several other protegrins . PC101, PC104, and PC107 have the same sequence as protegrin-1 for the first 15 residues but have different types of amino acids at the C-terminal end. We have determined that removing the positive charge from the C-terminus has a significant impact on the modes of peptide–micelle interactions. For PC101 and PC104, the second strand of the hairpin inserts into the SDS micelle, suggesting that these residues play a role in the antimicrobial activity of these peptides. We have also examined PC94 and PC98, peptides with have mutations at their N-termini.
Comparing all of the peptides for which we have completed simulations in both types of micelles, we see some trends in the positions affecting activity and toxicity. In Fig. 7, the residues most deeply inserted into the micelle are colored red and the residues farthest from the micelle are colored blue, with colors in between representing distances in between 15 Å and 25 Å. From the sequence alignment, we can see some trends in the residues that interact most strongly with the interior of the micelle. Mutations at the N- and C-termini cause changes in the interaction patterns with the SDS micelle. The wild-type protegrin-1 inserts strongly at Leu-5, and we see similar interactions for PC107. PC107 has the C-terminal positive charge found on PG-1 replaced with a negatively charged glutamate residue. PC101 and PC104 have sequences similar to PC107; however, these two peptides insert their C-terminal strand residues into the SDS micelle. This is due to the difference in charge of the C-termini. The negative charge on PC107 is repelled from the anionic micelle surface causing Leu-5 to be inserted into the micelle interior. PC101 and PC104, on the other hand are neutral at this region, and allow the hydrophobic residues Phe-12, Val-14, and Ile-16 on PC104, to insert into the micellar core.
It is also apparent that the mutations on the N-terminus require these peptides to shift their interactions with the SDS micelle to the second strand of the hairpin. The removal of two positive charges from the N-termini of PC94 and PC98 leave the peptide with no way to “anchor” this side of itself to the micelle surface. For these peptides, the phenylalanine and valine residues seem to have important interactions with bacterial membranes.
Simulations in SDS indicate that the residues on the second strand of the peptide are the ones that are causing toxicity. Fig. 8 shows the distances from the center of the DPC micelle for each residue. Red is most close to the micelle center and blue is outside of the micelle in the bulk water. Upon comparing Figs. 7 and and8,8, we see there is a distinct difference in the interaction patterns for these peptides. Mainly, the first half of the hairpin does not seem to be involved in the toxicity of the protegrins. The valine residue analogous to PG-1’s Val-14 is deeply inserted for each peptide, except for the significantly smaller PC72 and PC73 (Fig. 9).
Our attempts to design new peptides will be based on comparing the results of the interactions of the peptides in both SDS and DPC micelles. From the results presented here, a few rules can be determined. Of note is that there PC94 and PC98 have similar interactions in DPC versus SDS. This suggests that there may not be a straightforward way to reduce the toxicity of these peptides while retaining their antimicrobial activity. The removal of the positive charges from the N-terminal side of the hairpin thus may not be a beneficial mutation. Instead, we will focus on mutations on the C-terminus of the peptide. It is of interest that PC101 and PC104 have strong hydrophobic interactions in both DPC and SDS for Phe-12, Val-14 and the 16th residue, threonine or isoleucine, while PG-1 does not. The presence of a positive change at Arg-18 on PG-1 anchors the C-terminus to the SDS micelle surface, and prevents Phe-12, Val-14 and Val-16 from inserting too deeply into the micelle. PC101 and PC104 do not carry the positive charge at the C-terminus, thus allowing strong hydrophobic interactions between the C-terminal strand and the micelle. PC107 does not insert into this region because of the negatively charged glutamine residue on the C-terminus. For this reason, this sequence seems the most promising for further mutations to lower toxicity but retain activity (Fig. 10).
From the above observations, it is apparent that there are differences in the ways that different PG-1 mutants interact with the different types of micelles. These differences will allow us to modulate sequences to reduce the toxicity of the peptides and increase or maintain their activity. We will continue to gather information about the protegrins from the simulations of other protegrins analogues. We will also continue to test the sequences we have designed in the laboratory (Fig. 11).
Although the simulations of peptides with micelles successfully capture experimentally observed binding states, and explain biophysical peptide–membrane interactions, there are certain caveats which should not be overlooked. The geometry of the spherical lipid micelle is very different than planar lipid bilayers, and the interfacial curvature can potentially induce peptide structures which may never be actually observed near realistic membranes. A gradient of order parameters along the length of the hydrocarbon lipid tails is present in real lipid vesicles and atomistic models of lipid bilayers. This gradient alters the flexibility of the lipid chains along the bilayer normal, and might affect the depth of insertion of small peptides. This structural feature is not well-modeled in micelles. Although the DPC molecule has a headgroup similar to that of real zwitterionic phospholipids, the SDS molecule is a relatively poor representation of anionic phospholipids which carry a phosphatidylglycerol or a phosphatidylserine headgroup. Additionally, it is not possible to investigate the simultaneous binding of multiple peptides in micelles, something which can be done by increasing the size of bilayer systems. Despite these limitations, the micelles still offer efficient models of interfaces where simulations with many different peptides can be carried out with a high degree of confidence with respect to the sampling of the free energy minimum conformations. In the simulations reported in the current work, we have tried to address some of the problems associated with peptide–membrane simulations. We have also successfully developed elementary design rules for better antimicrobial peptides of two key structural classes of peptides: α-helical and β-turn. Of course, only testing those experimentally, will provide us with the ultimate confidence in the models, and we are working in this direction. Specifically, data on binding energies of peptides to micelles/bilayers can provide definitive proof of the conclusions drawn in the current work. Meanwhile, simulations will continue to provide us with the much-required atomistic level descriptions of peptide–membrane interactions.
In 2001, Tieleman and Sansom, two pioneers in the area of protein–membrane simulations, summarized that: “reliably simulating binding to membranes, insertion into membranes and aggregation of peptides both in and at the surface of membrane protein remain challenging problems” . Surprisingly, the situation has not changed drastically since. However, as methods for calculating free energies improve in accuracy, and computers keep getting cheaper and faster, our ability to address the time scales and length scales of the relevant peptide–membrane interaction phenomena can only improve. The future for simulations is looking fairly bright.
This work was supported by grants from NIH (GM 070989). Computational support from the Minnesota Supercomputing Institute (MSI) is gratefully acknowledged. This work was also partially supported by National Computational Science Alliance under MCB030027P and utilized the marvel cluster at the Pittsburgh Supercomputing Center. We thank Prof. Alan Waring for useful discussions. We also thank Prof. Ramamoorthy for inviting us to write a paper for this special issue.