Antimicrobial Peptides in the Cross Hairs of Computer Simulations

Main Text

Antimicrobial peptides occur in a staggering variety throughout all organisms and range from just a few residues and no well-defined secondary structure to small proteins like human defensins (1, 2). They are both of major biological interest in the chemical warfare between species and as possible new antibiotics, in their native forms or as a source of inspiration for the design of new compounds. Antimicrobial peptidelike peptides include cell-penetrating peptides, which may be useful in drug delivery and peptides targeted at other cell types, including cancer.

We now know that antimicrobial peptides act through a host of different mechanisms, including membrane perturbation, inhibition of cell division or cell-wall biosynthesis, and intercalation in ribosomes, but a common step is the requirement to be taken up by a cell through a cell membrane. Interactions with lipids also determine the selectivity of peptides, so that peptides can be targeted specifically to bacterial cells, with their negative surface charge and phosphatidylethanolamine and phosphatidylglycerol lipids as major components, over mammalian cells, rich in phosphatidylcholine lipids and cholesterol. However, peptides vary widely in charge, from close to neutral to highly cationic, distribution of hydrophobic residues along the sequence, secondary structure, the presence of aromatic residues, and a range of other chemical factors. Thus, there may not be a single mechanism of interaction with membranes. Many different mechanisms have been proposed, including various types of pores ranging from well-defined structures with reproducible conductance levels to disordered surfactant like disruption of membranes, as well as silent translocation of charged peptides without measurable pore formation.

In principle, computer simulations are an obvious method to address the disordered nature of lipids and peptide-lipid interactions, and indeed some of the earliest lipid simulations in the 1990s looked at lipid-peptide interactions (3). However, until recently, the timescales involved in the reorganization of the membrane have been a major challenge for computer simulations. To a first order of approximation, longer biophysical times translate into higher computational cost, in practice quickly exhausting the power of even the largest super computers.

Despite some early successes on a timescale of hundreds of nanoseconds (4), recent articles have established that timescales required to simulate basic processes, involving peptide binding, typically readily approach tens of microseconds and may go orders of magnitude beyond this. An important article on this topic by Neale et al. (5) looked carefully at statistical errors associated with a particular type of calculation aimed at the translocation of the 13-residue peptide Indolicidin with a total of 1.3 ms of simulation. In practice, simulation times have increased from hundreds of picoseconds in the early 1990s to of the order of tens of microseconds in the article by J. P. Ulmschneider in this issue (6). This timescale is becoming readily accessible on relatively cheap computational resources, which allows investigating a range of parameters, multiple copies to establish statistical errors, and in general enables a high degree of reproducibility.

In this issue, Ulmschneider (6) discusses simulations with very few assumptions that directly show translocation of the highly charged antimicrobial peptide PGla through model membranes, without pore formation. A common assumption is that such peptides translocate by forming transient pores, and indeed for many peptides such pores can be observed by electrophysiology experiments. However, for a large number of peptides (including PGLa), translocation, leakage of vesicles, and antimicrobial activity have been observed, but no pores. In this article, simulations explain why. Translocation of individual peptides is spontaneous and involves transient metastable states in which a peptide is inserted in the membrane, despite its high charge. Charged residues interact with either side of the bilayer where possible, and are stabilized by interactions with other peptides, still bound to the surface of the bilayer or partially inserted, and water defects that lower the energetic cost of inserting charged or polar side chains in the membrane interior. From these metastable states, peptides can either return to their original side or end up on the other side, providing a mechanism for silent translocation without stable pores. The kinetics of this process will depend strongly on the structure of the peptide, including the distribution of charged residues along the sequence and the propensity of the peptide to form a helical structure, so that unsatisfied backbone hydrogen bonds do not pose an unsurmountable barrier to translocation. In addition, the simulations show how Na⁺ ions and lipids can use the inserted peptide state to translocate across the bilayer. The overall picture defies a simple cartoon, as different pathways are observed, characterized by a short-lived intermediate state.

This work follows several other recent articles that demonstrate dramatic progress in both our understanding of the interactions between antimicrobial peptides and our ability to use advanced computer simulations to probe these interactions in great detail. Perrin et al. (7, 8) compared in two articles the interactions of piscidin and alamethicin with model lipid bilayers, also using computer simulations. Piscidin appears to behave similar to PGLa. In this study, piscidin was modeled in a number of different states, starting from regular barrel-stave pores in which the peptides form a regular bundle with transmembrane peptides. These pores were unstable and reverted through a more disordered toroidal pore structure, with one peptide causing a pore defect back to a surface-bound state for all peptides. This is consistent with solid-state NMR measurements on this peptide, which show predominantly surface-bound orientations. Thus, piscidin may also translocate silently, without forming pores, in a mechanism similar to that of PGLa.

In contrast to these two peptides, two other peptides showed evidence of forming relatively stable pore structures. Alamethicin is a hydrophobic peptide with low charge that is an archetypal barrel-stave pore. Simulations on a microsecond timescale showed no insertion at normal temperatures and without electric field, but at higher temperatures and with an applied electric field, peptides were inserted. When initially modeled as a barrel-stave pore, the structure remained stable for 14 μs, in agreement with experiments that show stable conductance levels for alamethicin (8). The overall story here is somewhat more complicated due to the role of folding at the membrane surface and the also experimentally observed requirement for a transmembrane electric field, but the main point is that this article gives detailed insight into another class of peptides with a substantially different mechanism.

A final example is the peptide maculatin. Wang et al. (9) showed by extensive equilibrium simulations that these peptides form a range of regular pore structures that continuously form and dissociate. It is quite surprising that the simulations yield regular hexa-, hepta-, and octomers in addition to smaller aggregates that appear to be the main pathway for transport of water, ions, and dyes. These pores form by adding helices to a single helix or helix bundle, which has been suggested for alamethicin and related peptides to match the different observed conductance levels of alamethicin.

These studies represent major progress, although they also highlight several limitations. One concern in general for antimicrobial peptides is that membrane interactions and translocation are a necessary step in peptide uptake and play a key role in selectivity, but the most effective antimicrobial peptides typically have other targets inside the cell. Thus, progress in understanding cell entry and membrane interactions does not directly translate into more effective antimicrobial peptides.

More technically, the simulations in the articles discussed here help identify current simulation challenges. First, these simulations, by and large, identify minimum simulation times of tens of microseconds for substantial rearrangements of peptides, which is also plausible given the relaxation times of lipids in modern lipid simulations and experiments. Although these timescales are now accessible, they are a lower bound and the kinetics of insertion or the formation of larger structures can be orders-of-magnitude slower for other sequences or likely other lipids. Of note, several articles used higher than normal temperatures and shorter than common lipids. There is experimental support for the relevance of the results in both approximations and extrapolation to lower temperatures is reasonable, but this will require further work. An alternative for direct simulation, where the timescale is limited by computer power, is the use of enhanced sampling methods. In a typical approach, a specific reaction coordinate, which can be multidimensional, is identified that describes the process of interest. In the case of peptide insertion this could be the location of the center of mass of the peptide, the angle between a helical axis and membrane normal, or a more complicated coordinate. Along this coordinate, barriers that are much higher than can be sampled by direct simulation can be accurately calculated, in principle giving access to truly macroscopic timescales like hours for lipid flip-flop (10). However, the choice of such coordinates is far from trivial, and this work will help identify reasonable choices and validate them for cases where both direct simulation and enhanced sampling approaches are feasible. Lastly, these simulations highlight the need for accurate force-field parameters, as relatively small errors in solubility of side chains in different environments may translate in significantly different populations of inserted versus surface-bound structures as well as in large differences in kinetic parameters.

After a slow start, it is encouraging to see the tremendous progress simulations are making in understanding the varied mechanisms of antimicrobial peptides interacting with lipid bilayers. A major challenge now will be to link a growing number of mechanisms observed in simulations, with themselves often variable pathways, to the chemical structures of the peptides and the specifics of membrane composition to predict and design properties of these peptides rationally. These simulations also point at future directions in testing and improving force fields and open up new routes to the clever use of enhanced sampling algorithms that allow access to much longer timescales but require avoiding biases in the design and setup of simulations. With the type of simulations in this article now accessible on relatively common hardware, we can explore the diversity of structures and mechanics and replace some of the current rather cartoonish mechanisms with high-resolution physical models with details that depend on a realistic description of specific peptides and lipids.

Editor: Claudia Steinem.