Molecular Dynamics Simulations Are Redefining Our View of Peptides

Dec 8, 2017 - dream to study all complex biological phenomena in silico, simply ... yet been met and that the dream will be realized, if at all, only ...
1 downloads 0 Views 3MB Size
Download PDF
Article Cite This: Acc. Chem. Res. 2018, 51, 1106−1116

pubs.acs.org/accounts

Molecular Dynamics Simulations Are Redefining Our View of Peptides Interacting with Biological Membranes Jakob P. Ulmschneider*,† and Martin B. Ulmschneider*,‡ †

Institute of Natural Sciences and School of Physics and Astronomy, Shanghai Jiao Tong University, Shanghai 200240, China Department of Chemistry, King’s College, London SE1 1DB, U.K.

‡

CONSPECTUS: Ever since the first molecular mechanics computer simulations of biological molecules became possible, there has been the dream to study all complex biological phenomena in silico, simply bypassing the enormous experimental challenges and their associated costs. For this, two inherent requirements need to be met: First, the time scales achievable in simulations must reach up to the millisecond range and even longer. Second, the computational model must accurately reproduce what is measured experimentally. Despite some recent successes, the general consensus in the field to date has been that neither of these conditions have yet been met and that the dream will be realized, if at all, only in the distant future. In this Account, we show that this view is wrong; instead, we are actually in the middle of the in silico molecular dynamics (MD) revolution, which is reshaping how we think about protein function. The example explored in this Account is a recent advance in the field of membrane-active peptides (MAPs). MD simulations have succeeded in accurately capturing the process of peptide binding, folding, and partitioning into lipid bilayers as well as revealing how channels form spontaneously from polypeptide fragments and conduct ionic and other cargo across membranes, all at atomic resolution. These game-changing advances have been made possible by a combination of steadily advancing computational power, more efficient algorithms and techniques, clever accelerated sampling schemes, and thorough experimental verifications. The great advantage of MD is the spatial and temporal resolution, directly providing a molecular movie of a protein undergoing folding and cycling through a functional process. This is especially important for proteins with transitory functional states, such as pore-forming MAPs. Recent successes are demonstrated here for the large class of antimicrobial peptides (AMPs). These short peptides are an essential part of the nonadaptive immune system for many organisms, ubiquitous in nature, and of particular interest to the pharmaceutical industry in the age of rising bacterial resistance to conventional antibiotic treatments. Unlike integral membrane proteins, AMPs are sufficiently small to allow converged sampling with the unbiased high-temperature sampling methodology outlined here and are relatively easy to handle experimentally. At the same time, AMPs exhibit a wealth of complex and poorly understood interactions with lipid bilayers, which allow not only tuning and validation of the simulation methodology but also advancement of our knowledge of protein−lipid interactions at a fundamental level. Space constraints limit our discussion to AMPs, but the MD methodologies outlined here can be applied to all phenomena involving peptides in membranes, including cell-penetrating peptides, signaling peptides, viral channel forming peptides, and fusion peptides, as well as ab initio membrane protein folding and assembly. For these systems, the promise of MD simulations to predict the structure of channels and to provide complete-atomic-detail trajectories of the mechanistic processes underlying their biological functions appears to rapidly become a reality. The current challenge is to design joint experimental and computational benchmarks to verify and tune MD force fields. With this, MD will finally fulfill its promise to become an inexpensive, powerful, and easy-to-use tool providing atomic-detail insights to researchers as part of their investigations into membrane biophysics and beyond.

■

INTRODUCTION

molecular biology is shaped by structural methods, which have revealed snapshots of proteins that carry out the key metabolic functions of cells. However, these methods can reveal only stable structures that persist over long time scales. Functional structures or complexes that form only transitorily have posed a formidable and to date unconquered challenge to these methods, biasing our view of protein function. This is

Many biological processes involve the interaction of peptides or small proteins with lipid bilayer membranes, to which they adsorb or insert or across which they translocate, lysing or permeabilizing the membrane, attaching to larger membrane proteins, or self-assembling into oligomeric structures and forming pores or other aggregates. These processes often involve transient structures and mechanisms that are notoriously difficult to study experimentally because of the fluidic nature of the lipid bilayer. Our understanding of © 2018 American Chemical Society

Received: December 8, 2017 Published: April 18, 2018 1106

DOI: 10.1021/acs.accounts.7b00613 Acc. Chem. Res. 2018, 51, 1106−1116

Article

Accounts of Chemical Research

Figure 1. Development timeline of models we have used to study peptide partitioning in silico. (a) Limited computer performance (pre-2008) led to the development of numerous implicit membrane models based on the generalized Born equation.12−27 Excellent predictions of the key configurational minima in the membrane, transmembrane (TM)-inserted as well as surface-adsorbed, are possible with such models, as shown here for a synthetic WALP sequence. (b) The first attempt to apply full-atomic-detail models revealed incorrect unfolding in the membrane,28 in direct contradiction to experimental circular dichroism (CD) measurements that indicate a perfectly helical confirmation for WALP sequences up to 95 °C.29 Parametrization and force field changes were required to identify the cause, resulting in (c) a folding−partitioning simulation revealing the correct pathway and a stable TM helix as the native state of hydrophobic peptides, consistent with experiments.29−33 Panel (b) adapted from ref 28. Copyright 2008 American Chemical Society. Panel (c) adapted from ref 31. Copyright 2010 American Chemical Society. Panel (a) adapted with permission from ref 26. Copyright 2008 Taylor & Francis.

are poorly understood.5,7−10 This has impeded the development of AMPs for biomedical applications,11 which requires a detailed understanding of the dynamic interactions between these peptides and microbial membranes. However, as we summarize here, AMP simulations can now capture the entire process of membrane binding, peptide aggregation, pore formation and conduction, and pore dissolution multiple times, revealing an ensemble of functional structures and an unprecedented wealth of information on how temporarily active structures carry out their function.

particularly true for membrane biology, which has focused on very stable structures that are amenable to crystallographic techniques. A vast range of membrane biology yet to be properly chartered instead relies on transient structures formed by peptides and proteins that inhabit both aqueous and membrane domains. To fully understand these systems and their dynamic processes, atomic-detail molecular mechanics simulations currently present a powerful method. Molecular simulations are coming of age as a result of hardware advances, algorithm improvements, and more accurate parametrizations of the chemical interactions involved (i.e., force fields). In this Account, we summarize the developments that have enabled direct simulation of peptide partitioning, folding, and insertion in realistic lipid bilayers. We further show that this methodology is suitably advanced and accurate to reveal how temporary pores are formed by membrane-active peptides (MAPs) and to describe their lifetimes and functions.1,2 This is demonstrated for the important class of sequences known as antimicrobial peptides (AMPs).3−5 AMPs are an essential part of the nonadaptive immune system for many organisms and play an important role in containing human microbiota.6 While pore-forming AMPs are known to preferentially target and cause leakage in microbial membranes, no conclusive evidence links poration to antimicrobial activity, and the mechanisms of membrane binding and pore formation

■

EARLY DEVELOPMENTS: IMPLICIT MEMBRANE MODELS More than 10 years ago, it was computationally unfeasible to directly attempt explicit molecular dynamics (MD) simulations of processes such as polypeptide adsorption, folding, insertion, and translocation, let alone events on much longer time scales such as peptide aggregation and assembly into functional pores that permeabilize the membrane by conducting solutes and water. As a consequence, alternative implicit membrane models (IMMs) were developed, extending the commonly used “generalized Born” theory of solvation to represent the membrane environment implicitly by introducing a planar hydrophobic (i.e., solvent-excluded) zone.12−27 The resulting reduction in the number of atomic interactions that needed to be evaluated vastly increased the simulation time scales. Such 1107

DOI: 10.1021/acs.accounts.7b00613 Acc. Chem. Res. 2018, 51, 1106−1116

Article

Accounts of Chemical Research

Figure 2. Unbiased folding−partitioning of peptides into lipid bilayers using atomic-detail simulations. (a) Theoretical cartoon of the equilibrium of an α-helical membrane-active peptide in a lipid bilayer. By means of high-temperature direct partitioning simulations, this equilibrium between surface-adsorbed (S) and transmembrane (TM)-inserted states can be directly calculated.31−33,41 No soluble state (W) exists for most peptides hydrophobic enough to insert, which is consistent with experiments showing that these peptides fold and aggregate in water and rapidly precipitate out of solution.31 (b) The partitioning equilibrium of peptides in a bilayer consists of flipping back and forth between folded interfacial (S) and TMinserted orientations, with the secondary structure remaining fully helical (sequence: LLLLWLLLL). The three states have characteristic insertion depths and tilt angles.41 (c) The transition rates strongly increase with temperature: Arrhenius plots of the insertion and expulsion rates for several synthetic peptides (L7, L8, and GL8 (GGPG-(L)8-GPGG)) all show first-order kinetics with TM/S barriers of 5−24 kcal/mol. Extrapolated transition times at 30 °C would be τS→TM ≈ 100−200 ms, many orders of magnitude larger than the time scales that can currently be reached in an MD simulation.32 (d) Key to the partitioning simulations is the temperature independence of the S-to-TM free energy surface, as shown for a polyleucine with eight residues (L8), with no change of ΔGS→TM over the range 30−160 °C.32 The thermostability of the inserting sequences is a key reason for the isentropic (ΔSS→TM = 0) nature of partitioning and allows the convenient estimation of ΔGS→TM from high-temperature simulations (simulations using helical restraints are denoted with [R]). Panels (a) and (c) reprinted from ref 32. Copyright 2011 American Chemical Society.

In moving to a more accurate model, several strategies may be followed. One popular approach is the application of coarsegrained, hybrid, and multiscale methods to membrane phenomena.34,35 In these methods, the reduced representation allows for much-extended sampling time scales, yet all of the system components (i.e., water, ions, lipid bilayer, proteins, and other solutes) are explicitly present, a marked improvement over IMMs. Instead of this path, since 2008 we have focused on using all-atomistic models because we did desire to retain from our IMM the full atomistic representation of the protein and wished to have the same level of detail for the lipid molecules, similar to the approach used in some recent membrane mimetic models.36 Early attempts with umbrella sampling or other potential of mean force (PMF) techniques revealed enormous convergence and hysteresis problems in studying peptide equilibria in membranes, precluding quantitatively accurate predictions. More importantly, such techniques are biased by the assumptions of the scientist constructing the reaction coordinate or transition pathway. They do not allow for blind, ab initio channel assembly predictions. We thus needed a completely new, preferably simple simulation strategy to overcome the sampling challenge of fully explicit equilibrium MD.

methods proved very successful, allowing the prediction of the partitioning and insertion of membrane-associated peptides of known orientation correctly, with an average tilt deviation of 100 times more efficient.18,26−29 However, the limitations of IMMs quickly became apparent: all important details of protein−lipid and protein−water hydrogen bonding are missingfor example, how could an anionic bilayer be distinguished from a zwitterionic one, or how could complex membranes with a wide range of lipid molecules, cholesterol, lysolipids, lipoproteins, or glycosylated lipids and other structural components be modeled to, e.g., distinguish a bacterial membrane from a eukaryotic one? It is obvious that selective peptide partitioning, adsorption, aggregation, and insertion depend on these details. Most importantly, the spontaneous formation of water-filled peptide-lined pores and the determination of their leakage and conductance are by design very difficult with IMMs. 1108

DOI: 10.1021/acs.accounts.7b00613 Acc. Chem. Res. 2018, 51, 1106−1116

Accounts of Chemical Research

■

THE DIRECT FOLDING, INSERTION, AND PARTITIONING METHODOLOGY

Article

■

EQUILIBRIUM PARTITIONING

■

EXPERIMENTAL VALIDATION

Sequences that are hydrophobic enough to insert into lipid bilayers in a TM orientation are predicted to partition to the membrane according to a two-state model: adsorption to the membrane is very strong, so a hydrated state is not usually populated in equilibrium and in experiments is not observed because of precipitation of the peptides out of solution. In the membrane, there is either a surface-bound (S) or TM-inserted state (Figure 2a).32,33,41 By means of HT-MD, this partitioning equilibrium is directly probed: peptides flip back and forth between folded interfacial and TM-inserted orientations, with the secondary structure remaining fully helical (Figure 2b).32 The transition rate increases exponentially with temperature. Arrhenius plots show insertion barriers of 5−24 kcal/mol for numerous sequences (Figure 1c). This allows extrapolation to 30 °C, typically showing τS→TM > 100 ms, which is impossible to reach using conventional MD simulations at present. Because there are so many transitions under hot conditions, ΔGS→TM can be precisely determined by simple population statistics. How useful is the measured ΔGS→TM at high temperature? A notable discovery was that the equilibrium has no entropic component (ΔSS→TM = 0).32,33 Thus, ΔGS→TM is temperatureindependent, and the value obtained from HT-MD is the same as for a putative simulation at physiological temperature. This is likely due to several reasons. First, the S state is found to be very deeply buried, at and below the glycerol linker groups of the lipid molecules. Thus, hydration, and consequently the temperature-dependent hydrophobic effect, play only a minor role in the S → TM equilibrium. Second, the peptides remain folded, and the bilayer does not change its chemical cross section: there is no unfolding or a shift in relative position/ orientation in the membrane, so only orientational entropy contributes. Since a rigid helix is roughly equally restrained in its motion in both the S state and the TM-inserted state, the orientational entropies are equal and ΔSS→TM = 0. Essentially, high temperatures have no effect on the partitioning thermodynamics (Figure 2d)only the kinetics changes.32,41 Extension of these simulations to common natural MAPs, such as AMPs and cell-penetrating peptides, revealed that these peptides also tend to be extremely thermostable in and on the membrane and can be accurately studied using HT-MD.1,2,42,43 An important fact that makes HT-MD simulations possible is that in the computational model it is possible to retain liquidstate conditions above the boiling point of water. When using periodic boundary conditions with simple water models such as SPC or TIP3P, phase transitions are not possible using conventional pressure and temperature coupling algorithms for the liquid state (i.e., a gas phase will not spontaneously form). Instead, the system resembles a superheated liquid, allowing reliable simulations up to 200 °C. The membrane remains intact at these temperatures.

The result of these developments was the elevated-temperature MD folding−partition simulation method, a useful tool for revealing the atomic-detail mechanisms of the interactions of MAPs with bilayers. In this method, peptides are initially placed fully extended in water above a lipid bilayer and are then allowed to freely fold, bind, insert, and exit the bilayer, akin to a real experiment. No restraints or biasing potentials of any kind are applied, and the results are independent of the initial conditions. One of the key strengths of this approach is that it allows precise quantification of folding−partitioning kinetics as well as membrane transfer free energies. Key to this method is the exceptional stability of peptides bound to membranes against thermal denaturation, which we have verified experimentally.31,32,37,38 This allows for accelerated partitioning kinetics via high-temperature simulations.15,16 In contrast to a soluble protein, where heating quickly results in denaturing, the membrane environment acts as a conformational restraint on the peptide structure that can be exploited to vastly speed up the kinetics. At first, it looked like this approach would not work: a 3 μs (for 2008 a very long time scale) high-temperature MD (HT-MD) simulation of a hydrophobic WALP sequence seemingly revealed both unfolded deep bilayer insertion and thermodynamically unstable α-helices (Figure 1b).29 This contradicted established theory, which dictates that only helical conformers can insert and reside stably in the membrane because of the huge ∼4 kcal/mol penalty to break a backbone hydrogen bond in the hydrophobic core.39,40 Figure 1b can serve as a warning to the danger of blindly trusting MD simulations, especially those that have not been properly parametrized and experimentally validated. The results are clearly incorrect, because these hydrophobic peptides can be experimentally shown to be thermostable. Circular dichroism (CD) measurements of the same peptide (WALP) in DPPC and DOPC vesicles (peptide/ lipid ratio = 1/100) over a temperature range of 45−90 °C showed that the peptides remain helical over the entire temperature range, independent of lipid type, peptide/lipid ratio, and peptide length. Similar results were seen for other hydrophobic peptides (Figure 1b). Thorough investigation led to the realization that the erroneous equilibrium MD results originated from inherent deficiencies in the underlying united-atom protein force field at elevated temperature.29 Our approach was to change to an allatom protein force field, in combination with new lipid force field parameters to ensure that the strengths of backbone hydrogen bonds and protein−lipid interactions were accurately reproduced.29,30 After this was accomplished and the simulation was repeated, the results agreed to a high degree with the experiments, as shown in Figure 1c. A plot of the peptide insertion depth versus helicity reveals that the general pathways taken by membrane-inserting peptides consist of three steps: (I) absorption, (II) interfacial folding, and (III) folded transmembrane (TM) insertion. The nonequilibrium phase (stages I and II) is usually completed in μmol).44 Final comparison of the results from all three methods for the insertion propensity PTM and ΔGS→TM revealed a remarkable agreement, indicating the reliability of the simulation technique in reproducing what is measured experimentally.

■

MEMBRANE INSERTION AND OLIGOMERIZATION OF A TYPICAL PORE-FORMING AMP After the demonstration that the simulation methodology can accurately partition peptides, the real utility lies in the application of this approach to study peptide aggregation and pore formation of biologically relevant sequences, such as AMPs, for which no experimental structural information exists to date. Pore prediction for individual AMPs is shown in Figure 4 for a typical pore former, the AMP maculatin from the Australian tree frog Litoria genimaculata. All of the peptides initially reside in their stable surface-adsorbed state,46 consistent with the −4.6 ± 0.8 kcal mol−1 partitioning free energy determined for wild-type (WT) maculatin via CD titration. Figure 4 shows that individual peptides rapidly TMinsert via a range of different mechanisms, despite theoretical membrane translocation barriers of ∼15−20 kcal mol−1.47 These barriers are overcome by cooperative insertion involving two peptides in a head-to-tail arrangement in combination with a water defect. TM-inserted maculatin can rapidly catalyze additional TM insertions through membrane defects induced by its charged and polar side chains. At equilibrium, the 1110

DOI: 10.1021/acs.accounts.7b00613 Acc. Chem. Res. 2018, 51, 1106−1116

Article

Accounts of Chemical Research

Figure 4. Channel assembly simulation protocol, here for the AMP maculatin. (a) Unbiased equilibrium MD simulations reveal that maculatin rapidly binds to POPC and crosses the membrane to populate both interfaces.2 The equilibrium is a dynamic swapping between transmembrane (TM)-inserted and surface-bound (S) helical conformations. (b) Maculatin forms an ensemble of structurally different channels that continually form and disband at equilibrium. Several channels form multiple times during the simulations. Only small pores consisting of up to eight peptides are observed. (c) CD spectra of maculatin in oriented DMPC and POPC bilayer stacks show that the experimental equilibrium orientation is ∼40% TMinserted and 60% surface-bound, consistent with the simulation average. The gray TM and S spectra shown are theoretical reference spectra used to determine the average TM population via linear deconvolution. Scans at 222 nm (helicity) show that these data are temperature-independent up to 95 °C (inset). (d) Plot of oligomer populations over time; the overall occurrence of each oligomer (color-coded, S = surface, 1 = single TM, 2 = TM dimer, ...) is shown in the distribution at the right. Panel (a−d) adapted from ref 2.

peptides are symmetrically distributed along the bilayer normal, continually changing between marginally stable TM oligomeric assemblies and surface-bound states on both interfaces. Overall, the surface-bound states slightly dominate the equilibrium with 53 ± 7% occupation (averaged over all simulations, ΔGS→TM = 0.1 kcal mol−1), in agreement with the 50 ± 15% interfacial population determined from oriented circular dichroism (OCD) measurements in aligned POPC bilayer stacks (Figure 4d). Time scales of 10−20 μs capture numerous spontaneous pore formation events. Rather than a single, well-defined, and stable pore structure, as would be expected for a channel protein, maculatin forms an ensemble of conformationally diverse channel-like pores, which continually assemble and disband. Permutational clustering software allows classification of the dominant oligomeric TM assemblies (Figure 4c): only small pores with three to eight peptides were observed, while

larger aggregates were found only rarely. Maculatin clearly prefers antiparallel peptide arrangements (90%), with the base structural motif being an antiparallel dimer offset by eight to 10 residues, with tight interhelical packing of the C-terminal moieties of the peptides. Larger oligomers are symmetric combinations of this basic dimer motif. Remarkably, single (P15A; E19Q) and double (P15A, E19Q) mutants, which show little or no difference in experimental leakage and pore size, revealed different ensemble weightings, as did changes in membrane acyl chain length. This shows that while maculatin forms proper channels in the membrane rather than disorded pores or detergent-like holes, the channel equilibrium is influenced by the environment or minor mutations. Indeed, the heterogeneous nature of the channellike structures precludes their classification into the generally assumed barrel stave or toridal models suggested for AMP pore 1111

DOI: 10.1021/acs.accounts.7b00613 Acc. Chem. Res. 2018, 51, 1106−1116

Article

Accounts of Chemical Research

Figure 5. Pore assembly mechanisms and experimental size validation. (a) Assembly process of AMP pores in the bilayer, shown for maculatin, which assembles by C-terminal-first “surface scavenging” catalyzed by existing TM helices. (b) Experimental pore sizing provides an estimate of the number of peptides making up the largest conducting pores. Maculatin-induced leakage of fluorescent dyes of different sizes from large unilamellar vesicles shows that the pores are small, allowing leakage of only 350 Da dyes (ANTS/DPX). (c) Simulation of an octameric pore identified during the assembly simulations shows spontaneous conduction of ANTS and DPX dyes, confirming that pores formed by simulations are consistent with the leakage data. Adapted from ref 2.

structures.9,48,49 The diversity of pores observed and their variation upon minor perturbation of the peptide sequence or membrane lipids explain why no clear pore-forming motif has been found in AMP sequences and why bacterial resistance against AMPs is remarkably low despite continuous exposure over millions of years. How do these pores assemble? For peptides with a large surface concentration (S state), the dominating assembly process observed (maculatin, Figure 5a) is a C-terminal-first surface insertion (“surface scavenging”) catalyzed by existing TM helices, which facilitates translocation of the polar side chains. For other AMPs that are much more inserted (pTM ≫ pS), direct TM−TM oligomerization is preferred. In both cases, larger oligomers form by addition of individual helices to an existing cluster rather than fusing of two larger oligomers. This is due to the specific contact interface that required the AMPs to orient their polar faces to the pore lining.

peptides can be estimated by assuming simple channel geometries. This revealed that only the smallest dyes encapsulated (ANTS/DPX), with molecular weights of ∼400 Da, are able to leak out.2 The consistency of the pore structures observed by the assembly simulation with the pore size determined from the dye leakage experiments can in turn be validated by simulating conduction of the largest dye known to be transported experimentally (Figure 4c). Dye-conductance simulations reveal that the octameric pores efficiently conduct ANTS/ DPX fluorophores, confirming that these are indeed the largest pores that maculatin can form.2

■

TRANSLOCATION WITHOUT PORATION The HT-MD protocol allows prediction of the equilibrium of a large number of membrane-active peptide sequences. This includes AMP sequences that behave very differently from pore formers, instead permeabilizing membranes without forming pores at all (Figure 6). Indeed, pores need not be invoked to explain both leakage and peptide translocation. Even a highly charged cationic peptide (+5, PGLa from Xenopus laevis) can spontaneously translocate across a hydrophobic lipid bilayer despite the prohibitive energetic penalty (Figure 6b).1 The simulation shows that peptides spontaneously translocate across the membrane individually on a time scale of tens of microseconds, with both surface-bound peptides and lipids assisting in the one-by-one translocation of the charged side

■

EXPERIMENTAL PORE SIZING The size of the predicted poresin the case of maculatin only eight peptides at maximumcan be experimentally validated using a pore-sizing assay (Figure 5b), which determines the leakage of fluorescent dyes of increasing molecular weight from liposomes treated with peptide. Varying the molecular weight of the dye from ∼350 Da (ANTS/DPX) to ∼40 kDa (biotinylated Alexa) provides a crude measure of the pore size, from which the approximate number of participating 1112

DOI: 10.1021/acs.accounts.7b00613 Acc. Chem. Res. 2018, 51, 1106−1116

Article

Accounts of Chemical Research

Figure 6. Monomeric peptide translocation mechanism, as observed for the AMP PGLa (charge +5). (a) Some AMPs do not form pores but greatly prefer the interfacial S state (PGLa), whereas others are pure channel formers (alamethicin, Alm), leading to very distinct oligomerization histograms (blue = S state).1 (b) Total peptide mass density, showing equilibration of the peptides across the bilayer normal to occupy both membrane interfaces. (c) Helix tilt angles of all PGLa peptides showing the preferred interfacial orientation. (d) Translocation events usually involve two or three peptides in mutual contact at their C-termini. Mutual deep insertion opens a transient water bridge through which the charged lysines are translocated, facilitated via cotransport of anionic lipids or chloride anions. The whole process occurs in 50 ns, after which the peptides return to their S state and the water bridge dissolves. No stable pores are formed, but one peptide has been translocated across the membrane. Events are rare, occurring every 10−20 μs. (e, f) The short-lived water bridges allow for ion leakage through a variety of mechanisms, usually involving anionic lipid flip-flop. Panel (b−f) adapted from ref 1. Copyright 2017 Cell Press.

chains. Single peptides can remain in a TM orientation for many microsceconds, snorkelling some charged residues to one

interface and some to the opposite, but without inducing a water channel (Figure 6d). Instead of stable pores, short-lived 1113

DOI: 10.1021/acs.accounts.7b00613 Acc. Chem. Res. 2018, 51, 1106−1116

Article

Accounts of Chemical Research

be possible via fully automated in silico computational processing, requiring little or no interaction and supervision. The vision of computer simulations to provide reliable, quantitative, and mechanistic predictions of peptide−membrane phenomena is rapidly becoming a reality and will transform the way peptide sequences are designed, selected, and optimized for subsequent biomedical applications. The elevated-temperature accelerated sampling technique is obviously restricted to the specific case of thermostable MAPs and cannot easily be extended to membrane proteins with large extra-membraneous domains or even to solvated proteins, which denature at much lower temperatures. However, even if no accelerating techniques are employed, current computational progress is so rapid that most biological phenomena will come within reach of all-atom MD in the near future.52 Proper force field parametrization via experimental verification will be the next big challenge, as we enter an era where limited sampling time is no longer a barrier, but solely the accuracy of the computational model. Comparison to in vitro experiments, as illustrated here, is only the first step. Matching in vivo measurements is much more challenging and will be the ultimate test of the utility of MD as a general research tool alongside structural, spectroscopic, and in vitro assays. The next years will no doubt become the golden age of MD simulations, as more and more systems become computationally tractable and advances in methodologies like the ones outlined here will transform MD into a powerful laboratory tool that can be applied by anyone to their arsenal of techniques.

water bridges occur when two or three peptides connect at their termini, allowing both ion translocation and lipid flip-flop via a brushlike mechanism usually involving the C-terminus of one peptide. The mechanism can explain why for many AMPs no channel structures have been determined experimentally, despite clear experimental evidence of membrane leakage and antimicrobial activity. Some AMPs are thought to translocate into cells and act on intracellular targets in bacterial cell death processes that are not caused by membrane permeabilization. For a complete understanding, in addition to translocation simulations, the intracellular target will need to be known.

■

CONCLUSION AND OUTLOOK The progress reported in this Account demonstrates that the ultimate promise of running atomistic MD simulationsto allow direct observation of complex biological phenomena in silicocan really be achieved. It is now possible to directly predict the insertion mechanism, native state structures, and conduction properties of MAPs at all-atomic resolution. We have limited our discussion here to AMPs, for which atomicdetail equilibrium molecular dynamics simulations can accurately reproduce experimental ensemble averages and partitioning data, revealing how AMPs target microbial membranes and spontaneously assemble into channels and how these channels conduct. Many AMPs have been known for decades, and countless experimental studies have been performed to decipher how they work. Nevertheless, because of experimental difficulties, there has been a lack of mechanistic understanding in how they target, lyse, porate, disrupt, or otherwise interact with membranes. Few have succeeded to being clinically approved, despite the fact that their application could greatly assist in countering the alarming rise of drugresistant and multidrug-resistant bacteria. Obtaining detailed structural information greatly enhances our understanding of biological systems, and in this case knowledge of the role of AMPs in innate immune defense will ultimately enable rational design and optimization of new AMPs for clinical, bioengineering, and agricultural applications. As AMPs are likely to play a major role in maintaining the balance of power between constituent microbes that make up the human microbiome, knowledge of their structure and function is also of great biomedical importance. In addition to channel-forming peptides, simulation studies can also reveal the mechanism and kinetics of membrane-translocating AMPs that subsequently act on other inner-cellular targets. The wider applicability of the methods outlined here to membrane protein biology beyond AMPs is vast. Studies on the targeting of particular membranes and the proteins they contain will provide the foundation for future investigations into the molecular interaction of other classes of biologically active peptides with membranes (e.g., cell-penetrating peptides and toxins). In the simulations, simple model membranes are increasingly being replaced by natural membranes containing the full plethora of different lipid types, proteins, and auxiliary molecules, resulting in increasingly realistic models of whole bacterial cell envelopes.50,51 The main applications will be the large-scale computational screening of a virtually infinite amount of newly designed sequences, the determination and classification of the mechanisms of MAPs, such as AMPs, cellpenetrating peptides, viral channel formers, fusion peptides, toxins, and signaling sequences, and the tuning of known sequences for potency optimizations. All of this will ultimately

■

AUTHOR INFORMATION

Corresponding Authors

*E-mail: [email protected]. *E-mail: [email protected]. ORCID

Jakob P. Ulmschneider: 0000-0003-0937-6963 Martin B. Ulmschneider: 0000-0001-8103-0516 Notes

The authors declare no competing financial interest. Biographies Jakob P. Ulmschneider received his Ph.D. in physics from Yale University in 2004. Since 2011 he has been a professor at the Institute of Natural Sciences and School of Physics and Astronomy at Shanghai Jiao Tong University, and in 2012 he was a recipient of the China 1000 Plan’s Program for Young Talents. Martin B. Ulmschneider received his Ph.D. in physics from Oxford University in 2002. He was an assistant professor at the Institute for NanoBioTechnology at Johns Hopkins University before joining the Chemistry Department of King’s College London as a reader in 2017.

■

ACKNOWLEDGMENTS This work was supported by the China 1000 Plan’s Program for Young Talents (13Z127060001) and a grant from the National Natural Science Foundation of China (91230105) to J.P.U. and by a grant from the U.S. Defense Threat Reduction Agency (HDTRA1-15-1-0046) to M.B.U.

■

REFERENCES

(1) Ulmschneider, J. P. Charged Antimicrobial Peptides Can Translocate across Membranes without Forming Channel-like Pores. Biophys. J. 2017, 113, 73−81.

1114

DOI: 10.1021/acs.accounts.7b00613 Acc. Chem. Res. 2018, 51, 1106−1116

Article

Accounts of Chemical Research

generalized Born membrane. Proteins: Struct., Funct., Genet. 2007, 69, 297−308. (24) Bu, L.; Im, W.; Brooks, C. L., III. Membrane Assembly of Simple Helix Homo-Oligomers Studied via Molecular Dynamics Simulations. Biophys. J. 2007, 92, 854−863. (25) Ulmschneider, J. P.; Ulmschneider, M. B. Folding simulations of the transmembrane helix of virus protein U in an implicit membrane model. J. Chem. Theory Comput. 2007, 3, 2335−2346. (26) Ulmschneider, M. B.; Ulmschneider, J. P. Membrane adsorption, folding, insertion and translocation of synthetic transmembrane peptides. Mol. Membr. Biol. 2008, 25, 245−257. (27) Ulmschneider, J. P.; Ulmschneider, M. B. Sampling efficiency in explicit and implicit membrane environments studied by peptide folding simulations. Proteins: Struct., Funct., Genet. 2009, 75, 586−597. (28) Ulmschneider, M. B.; Ulmschneider, J. P. Folding peptides into lipid bilayer membranes. J. Chem. Theory Comput. 2008, 4, 1807− 1809. (29) Ulmschneider, J. P.; Doux, J. P. F.; Killian, J. A.; Smith, J. C.; Ulmschneider, M. B. Peptide partitioning and folding into lipid bilayers. J. Chem. Theory Comput. 2009, 5, 2202−2205. (30) Ulmschneider, J. P.; Ulmschneider, M. B. United atom lipid parameters for combination with the Optimized Potentials for Liquid Simulations all-atom force field. J. Chem. Theory Comput. 2009, 5, 1803−1813. (31) Ulmschneider, M. B.; Doux, J. P. F.; Killian, J. A.; Smith, J.; Ulmschneider, J. P. Mechanism and kinetics of peptide partitioning into membranes. J. Am. Chem. Soc. 2010, 132, 3452−3460. (32) Ulmschneider, J. P.; Smith, J. C.; White, S. H.; Ulmschneider, M. B. In silico partitioning and transmembrane insertion of hydrophobic peptides under equilibrium conditions. J. Am. Chem. Soc. 2011, 133, 15487−15495. (33) Ulmschneider, J. P.; Andersson, M.; Ulmschneider, M. B. Determining Peptide Partitioning Properties via Computer Simulation. J. Membr. Biol. 2011, 239, 15−26. (34) Khalid, S.; Bond, P. J. Multiscale Molecular Dynamics Simulations of Membrane Proteins. In Biomolecular Simulations: Methods and Protocols; Monticelli, L., Salonen, E., Eds.; Humana Press: Totowa, NJ, 2013; pp 635−657. (35) Saunders, M. G.; Voth, G. A. Coarse-Graining Methods for Computational Biology. Annu. Rev. Biophys. 2013, 42, 73−93. (36) Ohkubo, Y. Z.; Pogorelov, T. V.; Arcario, M. J.; Christensen, G. A.; Tajkhorshid, E. Accelerating Membrane Insertion of Peripheral Proteins with a Novel Membrane Mimetic Model. Biophys. J. 2012, 102, 2130−2139. (37) Ulmschneider, M. B.; Ulmschneider, J. P.; Schiller, N.; Wallace, B. A.; von Heijne, G.; White, S. H. Spontaneous transmembrane helix insertion thermodynamically mimics translocon-guided insertion. Nat. Commun. 2014, 5, 4863. (38) Chen, C. H.; Wiedman, G.; Khan, A.; Ulmschneider, M. B. Absorption and folding of melittin onto lipid bilayer membranes via unbiased atomic detail microsecond molecular dynamics simulation. Biochim. Biophys. Acta, Biomembr. 2014, 1838, 2243−2249. (39) Bennett, W. F.; Hong, C. K.; Wang, Y.; Tieleman, D. P. Antimicrobial Peptide Simulations and the Influence of Force Field on the Free Energy for Pore Formation in Lipid Bilayers. J. Chem. Theory Comput. 2016, 12, 4524−33. (40) White, S. H.; von Heijne, G. How translocons select transmembrane helices. Annu. Rev. Biophys. 2008, 37, 23−42. (41) Gumbart, J. C.; Ulmschneider, M. B.; Hazel, A.; White, S. H.; Ulmschneider, J. P. Computed Free Energies of Peptide Insertion into Bilayers are Independent of Computational Method. J. Membr. Biol. 2018, DOI: 10.1007/s00232-018-0026-y. (42) Upadhyay, S.; Wang, Y.; Zhao, T.; Ulmschneider, J. Insights from Micro-second Atomistic Simulations of Melittin in Thin Lipid Bilayers. J. Membr. Biol. 2015, 248, 497−503. (43) Ulmschneider, M. B.; Ulmschneider, J. P.; Freites, J. A.; von Heijne, G.; Tobias, D. J.; White, S. H. Transmembrane helices containing a charged arginine are thermodynamically stable. Eur. Biophys. J. 2017, 46, 627−637.

(2) Wang, Y.; Chen, C. H.; Hu, D.; Ulmschneider, M. B.; Ulmschneider, J. P. Spontaneous formation of structurally diverse membrane channel architectures from a single antimicrobial peptide. Nat. Commun. 2016, 7, 13535. (3) Epand, R. M.; Vogel, H. J. Diversity of antimicrobial peptides and their mechanisms of action. Biochim. Biophys. Acta, Biomembr. 1999, 1462, 11−28. (4) van’t Hof, W.; Veerman, E. C. I.; Helmerhorst, E. J.; Amerongen, A. V. N. Antimicrobial peptides: properties and applicability. Biol. Chem. 2001, 382, 597−619. (5) Phoenix, D. A.; Dennison, S. R.; Harris, F. Antimicrobial Peptides; Wiley-VCH: Weinheim, Germany, 2013. (6) Cullen, T. W.; Schofield, W. B.; Barry, N. A.; Putnam, E. E.; Rundell, E. A.; Trent, M. S.; Degnan, P. H.; Booth, C. J.; Yu, H.; Goodman, A. L. Gut microbiota. Antimicrobial peptide resistance mediates resilience of prominent gut commensals during inflammation. Science 2015, 347, 170−5. (7) Wang, G.; Li, X.; Wang, Z. APD3: the antimicrobial peptide database as a tool for research and education. Nucleic Acids Res. 2016, 44, D1087−D1093. (8) Pasupuleti, M.; Schmidtchen, A.; Malmsten, M. Antimicrobial peptides: key components of the innate immune system. Crit. Rev. Biotechnol. 2012, 32, 143−171. (9) Wimley, W.; Hristova, K. Antimicrobial Peptides: Successes, Challenges and Unanswered Questions. J. Membr. Biol. 2011, 239, 27− 34. (10) Strandberg, E.; Ulrich, A. S. AMPs and OMPs: Is the folding and bilayer insertion of β-stranded outer membrane proteins governed by the same biophysical principles as for α-helical antimicrobial peptides? Biochim. Biophys. Acta, Biomembr. 2015, 1848, 1944−1954. (11) Galdiero, S.; Falanga, A.; Berisio, R.; Grieco, P.; Morelli, G.; Galdiero, M. Antimicrobial Peptides as an Opportunity against Bacterial Diseases. Curr. Med. Chem. 2015, 22, 1665−1677. (12) Feig, M.; Im, W.; Brooks, C. L. Implicit solvation based on generalized Born theory in different dielectric environments. J. Chem. Phys. 2004, 120, 903−911. (13) Lazaridis, T. Effective energy function for proteins in lipid membranes. Proteins: Struct., Funct., Genet. 2003, 52, 176−192. (14) Efremov, R. G.; Nolde, D. E.; Vergoten, G.; Arseniev, A. S. A solvent model for simulations of peptides in bilayers. I. Membranepromoting α-helix formation. Biophys. J. 1999, 76, 2448−2459. (15) Efremov, R. G.; Nolde, D. E.; Vergoten, G.; Arseniev, A. S. A solvent model for simulations of peptides in bilayers. II. Membranespanning α-helices. Biophys. J. 1999, 76, 2460−71. (16) Im, W.; Feig, M.; Brooks, C. L., III. An implicit membrane generalized born theory for the study of structure, stability, and interactions of membrane proteins. Biophys. J. 2003, 85, 2900−2918. (17) Spassov, V. Z.; Yan, L.; Szalma, S. Introducing an implicit membrane in generalized Born/solvent accessibility continuum solvent models. J. Phys. Chem. B 2002, 106, 8726−8738. (18) Ulmschneider, M. B.; Ulmschneider, J. P.; Sansom, M. S. P.; Di Nola, A. A generalized Born implicit membrane representation compared to experimental insertion free energies. Biophys. J. 2007, 92, 2338−2349. (19) Tanizaki, S.; Feig, M. A generalized Born formalism for heterogeneous dielectric environments: application to the implicit modeling of biological membranes. J. Chem. Phys. 2005, 122, 124706. (20) Tanizaki, S.; Feig, M. Molecular Dynamics Simulations of Large Integral Membrane Proteins with an Implicit Membrane Model. J. Phys. Chem. B 2006, 110, 548−556. (21) Im, W.; Brooks, C. L. Interfacial folding and membrane insertion of designed peptides studied by molecular dynamics simulations. Proc. Natl. Acad. Sci. U. S. A. 2005, 102, 6771−6776. (22) Im, W.; Chen, J. H.; Brooks, C. L., III. Peptide and protein folding and conformational equilibria: Theoretical treatment of electrostatics and hydrogen bonding with implicit solvent models. Adv. Protein Chem. 2005, 72, 173−198. (23) Ulmschneider, J. P.; Ulmschneider, M. B.; Di Nola, A. Monte Carlo folding of trans-membrane helical peptides in an implicit 1115

DOI: 10.1021/acs.accounts.7b00613 Acc. Chem. Res. 2018, 51, 1106−1116

Article

Accounts of Chemical Research (44) Wu, Y.; Huang, H. W.; Olah, G. A. Method of oriented circular dichroism. Biophys. J. 1990, 57, 797−806. (45) Hessa, T.; Meindl-Beinker, N. M.; Bernsel, A.; Kim, H.; Sato, Y.; Lerch-Bader, M.; Nilsson, I.; White, S. H.; von Heijne, G. Molecular code for transmembrane-helix recognition by the Sec61 translocon. Nature 2007, 450, 1026−1030. (46) Wang, Y.; Zhao, T.; Wei, D.; Strandberg, E.; Ulrich, A. S.; Ulmschneider, J. P. How reliable are molecular dynamics simulations of membrane active antimicrobial peptides? Biochim. Biophys. Acta, Biomembr. 2014, 1838, 2280−2288. (47) MacCallum, J. L.; Bennett, W. F.; Tieleman, D. P. Distribution of amino acids in a lipid bilayer from computer simulations. Biophys. J. 2008, 94, 3393−3404. (48) Matsuzaki, K.; Murase, O.; Fujii, N.; Miyajima, K. An antimicrobial peptide, magainin 2, induced rapid flip-flop of phospholipids coupled with pore formation and peptide translocation. Biochemistry 1996, 35, 11361−8. (49) Baumann, G.; Mueller, P. A molecular model of membrane excitability. J. Supramol. Struct. 1974, 2, 538−557. (50) Boags, A.; Hsu, P.-C.; Samsudin, F.; Bond, P. J.; Khalid, S. Progress in Molecular Dynamics Simulations of Gram-Negative Bacterial Cell Envelopes. J. Phys. Chem. Lett. 2017, 8, 2513−2518. (51) Patel, D. S.; Qi, Y.; Im, W. Modeling and simulation of bacterial outer membranes and interactions with membrane proteins. Curr. Opin. Struct. Biol. 2017, 43, 131−140. (52) Shaw, D. E.; Grossman, J. P.; Bank, J. A.; Batson, B.; Butts, J. A.; Chao, J. C.; Deneroff, M. M.; Dror, R. O.; Even, A.; Fenton, C. H.; Forte, A.; Gagliardo, J.; Gill, G.; Greskamp, B.; Ho, C. R.; Ierardi, D. J.; Iserovich, L.; Kuskin, J. S.; Larson, R. H.; Layman, T.; Lee, L.-S.; Lerer, A. K.; Li, C.; Killebrew, D.; Mackenzie, K. M.; Mok, S. Y.-H.; Moraes, M. A.; Mueller, R.; Nociolo, L. J.; Peticolas, J. L.; Quan, T.; Ramot, D.; Salmon, J. K.; Scarpazza, D. P.; Schafer, U. B.; Siddique, N.; Snyder, C. W.; Spengler, J.; Tang, P. T. P.; Theobald, M.; Toma, H.; Towles, B.; Vitale, B.; Wang, S. C.; Young, C.: Anton 2: raising the bar for performance and programmability in a special-purpose molecular dynamics supercomputer. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis; IEEE Press: New Orleans, LA, 2014; pp 41−53.

1116

DOI: 10.1021/acs.accounts.7b00613 Acc. Chem. Res. 2018, 51, 1106−1116