Model
To support our wetlab strategy of fusing antimicrobial peptides (AMPs) to endolysins, we developed AMP-MD, a molecular dynamics (MD) platform to study how AMPs interact with the outer membrane of Gram-negative bacteria.
Unlike most studies that focus on the inner membrane, our simulations uniquely address the outer membrane, a critical but understudied barrier due to its structural complexity. By combining Umbrella Sampling, unbiased simulations, and analysis tools we investigated peptide–membrane interactions and demonstrated that the destabilization of our model membrane through peptide insertion follows the “self-induced uptake” mechanism proposed by Hancock (Hancock, 1997), a mechanism that has so far not been analyzed by any molecular dynamics simulation study before.
Through targeted case studies, we identified why some AMPs are more effective than others to enhance the membrane permeability when fused to endolysins.
The general workflow for our simulations is shown in the graphical abstract below and the scripts used for these simulations with a short tutoial are uploaded on the iGEM Münster Gitlab Repository.
The outer membrane of Gram-negative bacteria is located between the peptidoglycan cell wall and the extracellular environment. It consists of phospholipids in the inner leaflet and lipopolysaccharides (LPS) on the outer leaflet. The structure of an LPS (E. coli R3 O111) is shown in Figure 1 and consists of three different parts.
Lipid A: A phosphorylated glucosamine disaccharide attached to fatty acids, forming the hydrophobic membrane anchor.
Core sugar region: A polysaccharide chain with unusual sugars such as Kdo (3-deoxy-D-manno-octulosonic acid) and phosphorylated sugars. This region is negatively charged and stabilized by divalent cations like Ca²⁺ and Mg²⁺, which bind to phosphate groups (Erridge, Bennett-Guerrero and Poxton, 2002), providing structural stability to the outer membrane.
O-antigen: A highly variable, repeating polysaccharide chain extending from the bacterial surface. LPS lacking this region are called “rough” (e.g., A. baumannii), whereas those with it are “smooth” (e.g., E. coli).
Figure 1: Schematic illustration of LPS structure.
A molecular dynamics simulation simulates a classical system of N atoms. It tries to determine the position and velocity of every atom in the system by solving Newton’s equations of motion (Vollmayr-Lee, 2020),
\begin{equation} F_\mathrm{i} = m_\mathrm{i} a_\mathrm{i} . \end{equation} This equation is solved using the leap-frog integrator (Hockney, Goel and Eastwood, 1974), which is the standard integrator in GROMACS (Abraham et al., 2015). By solving these equations of motion iteratively with the leap-frog integrator, the position and velocity of all atoms in the system for the next time step are determined. To solve these equations the forces acting on each atom have to be calculated at each time step.There are two types of interactions between atoms that determine the force acting on an atom, nonbonded and bonded interactions.
Atoms connected by chemical bonds behave like springs and are approximated by a harmonic oscillator,
\begin{equation} V_\mathrm{ij}^{\text{bond}} = k_\mathrm{ij}\,(r - r_{0,\mathrm{ij}}). \end{equation}For different atom pairs ij in different molecules, the parameters \(k_\mathrm{ij}, r_\mathrm{0,ij}\) are defined by the CHARMM36m force field (Huang et al., 2017).
Atoms not directly bonded still influence each other. The repulsive interactions due to Pauli repulsion (Pauli, 1925) between two atoms and the attractive interactions (van-der-Waals (Langbein, 1974)) are summarized using the Lennard-Jones Potential (Jones, 1924),
\begin{equation} V_\mathrm{ij}^{\text{LJ}} = 4\epsilon \left[ \left( \frac{\sigma}{r_\mathrm{ij}} \right)^{12} - \left( \frac{\sigma}{r_\mathrm{ij}} \right)^6 \right]. \end{equation}Lennard Jones Potentials are short-ranged potentials, that means in practice these are only calculated for atoms that are in a certain cutoff radius. Coulomb potentials, which are long-range potentials, are calculated through the Particle Mesh Ewald summation (Ewald, 1921). All forces are parametrized, enabling fast computation, which is crucial for membrane simulations spanning long time scales.
A simplified MD algorithm can be shortly described by:
For our mechanistic studies of AMP insertion into membranes, we employed three different bacterial membrane models. The membrane compositions are summarized in Table 1.
| Membrane | Lower Leaflet | Upper Leaflet | Literature |
|---|---|---|---|
| E. coli R3 O111A | 54 POPE / 3 POPG / 3 PVCL | 20 LPS | (Sharma and Ayappa, 2022) |
| A. baumannii | 56 POPE / 12 POPG / 9 PVCL | 25 LPS | (Jiang et al., 2020) |
| E. coli K-12 | 90 POPE / 5 POPG / 5 PVCL | 33 LPS | (Sharma et al., 2024) |
Atomistic membranes were built with CHARMM-GUI (Jo et al., 2008) and equilibrated following the standard six-step CHARMM-GUI protocol, followed by 400 ns equilibration in the NPT ensemble (constant temperature and pressure). The equilibrated membranes for E. coli R3 O111A and A. baumannii are shown in Figure 2.
E. coli R3 O111
A. baumannii
Figure 2: Equilibrated membranes of E. coli R3 O111 (left) and A. baumannii (right). The O-antigen region is colored in yellow, the Core Saccharide region in red, and the hydrophobic region in grey. The Ca2+ ions are portrayed as black beads. For illustration purposes, the H atoms and water molecules are not displayed. All MD snapshots are visualized with VMD (Humphrey, Dalke and Schulten, 1996).
We used 2 different antimicrobial peptides in our simulations (CM15 and SMAP29). The structures of CM15 (RCSB: 2JMY) (Respondek et al., 2007) and SMAP29 (RCSB: 1FRY) (Tack et al., 2002) were observed as random coils in experiments. Therefore, the structures from the Protein Data Bank needed to be turned into random coil structures before simulating peptide membrane interactions. The peptides were simulated for 1 µs in water with a salt concentration of 0.15 mol KCl, and the final structure was taken and inserted into the membrane system by replacing solvent molecules on top of the outer membrane. After insertion, the system needed to be energy minimized and equilibrated again. The MD production runs were performed with the following settings.
The MD simulations were performed using the GROMACS MD engine (Abraham et al., 2015) and the CHARMM-36m forcefield (Huang et al., 2017). Periodic boundary conditions were applied in all 3 directions, and the leapfrog integrator with an integration step of 2 fs was chosen to solve the equations of motion. To maintain a constant temperature (310 K) and pressure (1 bar), the v-rescale thermostat (Bussi, Donadio and Parrinello, 2007) was used with the Parrinello-Rahman barostat (Parrinello and Rahman, 1981). The LINCS (Linear constraint solver) algorithm (Hess et al., 1997) was applied to constrain hydrogen bonds, and the Particle Mesh Ewald algorithm to compute long-range interactions (Darden, York and Pedersen, 1993). In the unbiased simulations, to ensure the peptide doesn’t diffuse to the other side of the membrane through the pbc (periodic boundary condition), an energy barrier was employed at the end of the simulation box, which applied a harmonic potential on the peptide if it was in the process of diffusing to the other side.
The MD simulations were performed on the Palma II server of the Universität Münster. We used two different GPUs for our MD simulations, Nvidia A40 and Nvidia RTX 4090. Performing MD simulations with GPU accelerations increases the time simulated per day enormously, and without using GPUs, it would be impossible to achieve timescales on µs scale with our membrane systems that contain 60k to 100k atoms (400 ns per day).
All the MD settings used for simulations are uploaded on the iGEM Münster Gitlab Repository.
Umbrella Sampling is a computational technique used to calculate the potential of mean force (PMF) along a chosen reaction coordinate (ζ). In our project, we initially aimed to quantify the binding affinities of different AMPs to bacterial membranes, in order to select peptides most suitable for coupling to endolysins.
(The PMF describes the free energy landscape of peptide insertion into the membrane. Previous work, e.g., Yong et al. (Deylami, Chng and Yong, 2024), used it to study antibiotic insertion through lipid bilayers.)
Figure 3: Schematic illustration of Umbrella Sampling (Lemkul, 2024) . The arrow in the top is the reaction coordinate (z direction) and the filled red circle represents the atom group (CM15 peptide) moving along this reaction coordinate. The dotted arrows illustrate the extraction of starting windows along this reaction coordinate. After extracting the starting windows, a biased simulation (constraint along the reaction coordinate with a harmonic potential) is started for each window. The biased distribution of these simulations are illustrated at the bottom of this illustration.
There was only one work from Sharma et al. (Sharma and Ayappa, 2022) that we found during the research phase that was applying Umbrella Sampling on peptides going through the membrane. We tried to replicate these results first, but with a few changes: different thermostat because the Nose-Hoover thermostat doesn’t work well with GPU acceleration, slightly different force fields (instead of CHARMM-36, the improved CHARMM-36m forcefield), different membranes and different peptide starting configurations because there was no structure file for the initial membranes and peptides. The membrane used was the E. coli R3 O111 membrane, and the peptide used was the CM15 peptide.
The membrane peptide system was initialized as described in Chapter Two. The reaction coordinate was defined as the distance between the COM (Center of Masse) of the peptide and the COM of the membrane in z direction. The peptide was pulled into the z-direction through the membrane to create the starting windows. The pull process is shown in Figure 4.
Figure 4: The CM15 peptide is pulled through the E. coli R3 O111 membrane to generate windows for Umbrella Sampling. The N-terminus of the peptide is marked in green, and the C-terminus is marked in pink.
We created 23 starting windows from the pull process and equilibrated each of these windows for 20 ns before starting the 100 ns production runs. During the production runs, the peptide was restrained in z direction by a harmonic potential with a force constant of k = 1000 kJ mol-1 nm-2. The probability distributions along the z-axis of these production runs are shown in Figure 5.
Figure 5: Histogram of the biased production runs of 23 windows that are constrained in the z direction by a harmonic potential.
The free energy/PMF was calculated using the weighted histogram analysis method (WHAM) (Hub, de Groot and van der Spoel, 2010) in GROMACS, which transforms the biased distributions to unbiased distributions in an iterative algorithm. The unbiased distribution along the z-axis can then be used to receive the free energy \(\Delta G\).
>
Figure 6: Insertion free energy \(\Delta G\) of peptide into the outer membrane of E. coli R3 O111. The interfaces of extracellular phase, O-antigen (OA) and Core-Saccharide (CS) are marked with dotted lines.
The free energy profile shows that the insertion has an energy minimum of \(\Delta G\) = -12 kJ mol-1 at z = 3.5 nm which corresponds to the OA/CS interface. After that, the energy increases quite drastically in the Core Saccharide region. The reason for that could be the repulsive forces between the divalent Ca2+ cations and the 5 times positively charged peptide. The final configurations of a few windows are shown in Figure 7.
Figure 7: Final configurations of the peptide membrane system after 100 ns simulation time at z = 6.0, 4.8, 3.5 and 1.8 nm.
Compared to the work of Sharma et al (Sharma and Ayappa, 2022), the minimum energy was smaller by one magnitude, while the overall shape of the PMF curve was similar. This highlights that quantifying AMP efficacy solely based on insertion free energy is unreliable. Umbrella Sampling is too sensitive to small changes like changes in thermostat, pull process or starting structure, which makes it not fitted to quantify energies at the system scale of 60-100k atoms (Javed, Kapakayala and Nair, 2024). The energies also highly depend on the position the peptide is placed above the membrane.
Because of these limitations, we shifted our focus from comparing insertion energies of different AMPs to understanding the mechanistic details of peptide insertion and membrane disruption. This motivated the use of unbiased MD simulations, allowing us to study the spontaneous insertion of peptides and identify which regions of the peptide are critical for insertion and how their insertion affects membrane properties.
In the previous chapter, we determined the insertion free energy (\(\Delta G\)) using Umbrella Sampling. These results showed that insertion into the OA region and up to the CS interface is an energetically favourable process., However, Umbrella Sampling relies on externally applied forces and therefore cannot fully capture the natural dynamics of peptide–membrane interactions. To gain a deeper mechanistic understanding of how peptides insert into membranes and how these interactions contribute to membrane permeabilization, we turned to unbiased molecular dynamics simulations. Unlike Umbrella Sampling, this approach allows peptides to insert spontaneously without artificial pulling, providing a more realistic representation of the insertion process.
In the first unbiased simulation, we examined the E. coli R3 O111/CM15 system, which was simulated for t = 4 µs. 6 CM15 peptides were randomly placed on top of the membrane as shown in Figure 8.
Figure 8: Equilibrated configuration of E. coli R3 O111 membrane with 6 CM15 peptides placed on top of the membrane. The N-terminus of the peptide is marked in green, and the C-terminus is marked in pink.
The first step in the analysis of MD simulations is typically to examine the trajectories, as shown in Figure 9. A complete insertion of one CM15 peptide into the OA region was observed after 1 µs, followed by spontaneous translocation into the CS region between 1 and 4 µs.
Figure 9: : Trajectory of the E. coli R3 O111 system from 1 µs to 4 µs. For visualization purposes, only the CM15 peptide that inserted into the membrane is shown.
To understand which interactions between the membrane and the peptide led to the insertion into the CS region, we analyzed the interaction energies Eint between the peptide and the CS/OA region. Figure 10 shows the short-range Lennard-Jones (LJ) and Coulomb (Coul) interactions between the peptide and the CS/OA region. As a reminder, negative energies mean attractive interactions. From t = 0 to 600 ns, the LJ and Coul interactions between the peptide and the OA region are at around Eint ≈ -200 kJ. The Coul interaction between the positively charged CM15 peptide and the negatively charged CS region starts to increase at t = 800 ns until it reaches a minimum of Eint ≈ -650 kJ at around t = 2000 ns. Between t = 2000 ns and t = 3500 ns, the Coulomb interaction fluctuates around an interaction of Eint ≈ -400 kJ, exhibiting two maxima at t = 2320 ns and t = 3400 ns, which correspond to weakened or even repulsive interactions of up to Eint ≈ +60 kJ between the CS region and the peptide.
Figure 10: : Short-range interaction energies, LJ (dotted lines) and Coul (solid lines) between CM15 peptide and CS region (red) and between CM15 peptide and OA region (blue).
The configurations corresponding to these maxima are shown in Figure 11. They also indicate that reaching certain confirmations, for example the one at t = 4000 ns, requires sufficient conformational freedom of the peptide in the z direction. This freedom is strongly restricted in biased simulations such as Umbrella Sampling, which can lead to inaccurate sampling within specific windows. Overall, the observed maxima suggest that the insertion mechanism of the CM15 peptides is not a simple stepwise process but rather occurs by a more complex mechanism.
Maximum at t = 2320 ns
Maximum at t = 3400 ns
Figure 11: Configurations where the Coul interaction between peptide and CS region reaches a maximum (least attractive energy between peptide and CS region).
To further elucidate this mechanism, we investigated the average Coul interactions of each individual peptide residue with the CS region within the time window of 2 µs to 3 µs which are shown in Figure 12.
Figure 12: Average Coulomb interaction energies
The positively charged lysin residues show the strongest attractive Coul interaction, especially Lys residues 7 and 13. This was expected since Lys 1 and Lys 3 interact with the OA region during the analyzed time window. The relatively low RMSD values of 25 kJ mol-1 and 13 kJ mol-1 support this observation. The Lys 6 residue moved along the interface between the OA and the CS region during the simulation window. The high RMSD values of Lys 6,7 and 13 indicate that the peptide is not statically bound to one positive phosphate group but instead explores the CS region with considerable mobility. The only repulsive interaction observed during this interval occurred between Leu 15 at the negatively charged C-terminus and the CS region.
The repulsive interactions and the high RMSD values of the Lys 13 and 7 explain the Coul interaction maximum at t = 2320 ns. To confirm this interpretation, we plotted the Coul interaction between individual residues and CS region for the interval t = 2200 - 2400 ns. As we can see in Figure 13, the attractive interactions between the Lys residues and the CS region are almost zero, only the Lys 13 has an attractive interaction of Eint ≈ -20 kJ mol-1 whereas the repulsive interactions between C-terminal Leu 15 and CS region is at Eint ≈ 80 kJ mol-1 which leads to an overall Coul interaction energy between the whole peptide and the CS region of Eint ≈ -60 kJ mol-1 (Figure 10 and Figure 13).
Figure 13: Coulomb interaction between different charged residues and CS region from t = 2200 – 2400 ns. The dashed line marks the maximum of interaction energy between the peptide and the CS region of the membrane at t = 2320 ns.
In previous chapters, we explored the energy-free landscape of the peptide in the OA and the CS region using umbrella sampling. We then investigated the insertion mechanism of the antimicrobial peptide (AMP) as it migrated from the OA region into the CS region of the outer membrane of Gram-negative bacteria.
The remaining unexplored part is the hydrophobic region of the outer membrane, which is composed of Lipid A in the upper leaflet and phospholipids in the lower leaflet. In the following section, we focus on the properties of Lipid A/CS (Lipid A/water) interfaces and the insertion of the peptide into the hydrophobic part of the Lipid A layer.
Sharma et al. showed in their studies of the energy-free landscape of the peptide through the membrane (Sharma and Ayappa, 2022) that the Lipid A/CS interface forms a pronounced energetic barrier preventing peptide penetration. Our unbiased simulations confirm this result, as the peptide did not cross the Lipid A/CS interface. This barrier can also be illustrated by plotting the average density of the membrane and peptide during the simulation time frame of t = 3 - 4 µs, which is shown in Figure 14. The density of the peptide goes to zero at exactly the start of the Lipid A region at z = 10 nm.
Figure 14: Density profile for components of the OM, peptide (only inserted one) and water from a simulation time frame of t = 3 – 4 µs.
Lipid packing defects are defined as regions of loose lipid organization on the membrane surface that generate cavities where aliphatic lipid carbons are exposed to the aqueous environment. These cavities are essential for the binding of peripheral proteins since many of these proteins contain hydrophobic moieties that insert into the membrane (Gautier et al., 2018). This is particularly relevant for our AMP CM15, in which the C-terminus is strongly hydrophobic.
This led us to an initial hypothesis: the energetic barrier at the Lipid A/CS interface arises because the hydrophobic C-terminus of the peptide cannot insert into the Lipid A hydrophobic region, potentially due to lipid-packing defects being too small.
To study these lipid-packing defects, we used the “PackMem” Python package (Gautier et al., 2018) that detects two types of lipid defects, which are shown in Figure 15. This package was developed and optimized by Antonny and co-workers. It uses a 2D Cartesian grid-based method that scans the lipid to water interface to identify aliphatic lipid tails that are exposed to the surface.
Figure 15: Schematic representation of the two different types of lipid-packing defects. The first line separates the polar head group from the aliphatic lipid tails, and the dashed line separates deep lipid packing defects from shallow ones.
The method was originally developed for phospholipid bilayers, but for our study we adapted it to Lipid A by generating a list of Lipid A atoms, their van der Waals radii, and their classification as aliphatic or non-aliphatic.
The van der Waals radii were calculated using half the minimum distance of the Lennard-Jones potential, corresponding to the point where two identical atoms experience neither attraction nor repulsion.
\begin{equation} r_{\mathrm{vdW}} = \tfrac{1}{2} R_{\min} = 2^{\tfrac{1}{6}} \, \sigma \end{equation}We automated this process with a custom script that extracts the radii directly from GROMACS topology files (available on the iGEM Münster Gitlab Repository).
To analyze lipid packing defects, we need a bigger membrane system with at least 100 phospholipids to get statistically reasonable results. We therefore built a new membrane with the standard membrane building and equilibration protocol described in Chapter 2. The composition for this membrane is shown in Table 2.
| Membrane | Lower Leaflet | Upper Leaflet | Literature |
|---|---|---|---|
| E. coli R3 O111A | 90 POPE / 5 POPG / 5 PVCL | 32 LPS | (Sharma et al., 2024) |
Figure 16: a) shows the E. coli Lipid A membrane in VDW representation with the cartesian grid on the upper surface. b) shows the lipid packing defects (top-down view), deep defects in red and shallow defects in blue. The polar atoms (N, O and P) are marked in orange.
Figure 16 a) illustrates the Cartesian grid on top of the membrane used to analyze lipid packing defects, where each grid point is classified as either no defect, a shallow defect, or a deep defect. Figure 16 b) depicts the distribution of deep (red) and shallow (blue) lipid packing defects in the E. coli lipid A membrane after a simulation time of t = 1000 ns. The differences in packing defects between the upper leaflet and the lower leaflet are shown in Figure 18. The lipid A (upper leaflet) defect areas are much bigger than the phospholipids (lower leaflet) defect areas.
The differences in packing defects between the upper leaflet and the lower leaflet are shown in Figure 18. The lipid A (upper leaflet) defect areas are much bigger than the phospholipids (lower leaflet) defect areas.
Figure 17: a) lipid packing defects of the lower leaflet compared to b) lipid packing defects of the upper leaflet. The areas where aliphatic tails are exposed to the water surface are much larger in the upper leaflet.
To quantify the size and longevity of the appearing lipid packing defects, we analyzed 3000 frames of the E. coli Lipid A membrane during a simulation time of t = 700 - 1000 ns. The probability of the occurrence of a defect with an area A was fitted to a mono-exponential decay (Cui, Lyman and Voth, 2011; Vamparys et al., 2013) ,
\begin{equation} P(A) = b \cdot e^{-A / \pi}, \end{equation}where P(A) is the probability of finding a lipid packing defect among all packing defects with an area of A, b is the preexponential factor (useless value for analysis), and π is the packing defect constant. The packing defect constant π corresponds to the surface area of the lipid packing defect. The higher the packing defect π, the more abundant are the large packing defects. Figure 18 shows the probability of finding a packing defect with area A on a logarithmic scale. The data points are fitted to equation 5, and the packing constants π for deep, shallow, and all packing defects were calculated.
Figure 18: The probability of finding a lipid packing defect P(A) as a function of the packing defect area A for “deep”, “shallow”, and “all” packing defects. The data points for each packing defect type, which are larger than A = 15 Å2 and for probabilities of P(A) > 10-4, are fitted to equation 5. The packing defect constant π was calculated from the slopes of the fitted functions and visualized in the bottom right diagram.
The packing defect constant π of the E. coli Lipid A membrane is two to three times larger than values reported for phospholipid bilayers (Gautier et al., 2018). In the studies of Ma (Ma et al., 2021), they observed a penetration and destruction of their DPPG phospholipid membrane with the AMP CM15.
Since we observed relatively large lipid packing defects of Lipid A, our results — together with insights from the cited studies — do not support our initial hypothesis. The size of the lipid packing defects of the Lipid A leaflet does not explain the energy barrier at the Lipid A/CS interface prohibiting the translocation from the CS region into the Lipid A region (Figure 14).
To ensure that the size of the lipid packing defects is large enough for a peptide to insert into the hydrophobic region, we did an unbiased simulation with the same E. coli Lipid A membrane (no CS/OA region) and the CM15 peptide. After 1000 ns we observed an insertion onto the surface of the hydrophobic region of the Lipid A, and after 8000 ns, the whole peptide was completely inserted into the membrane, which is visualized in Figure 19. We can also observe a helical type of structure, which is classified as an α-Helix by the VMDs STRIDE algorithm (Frishman and Argos, 1995). This α-helical structure motif was also observed in NMR experiments (Respondek et al., 2007).
Figure 19: Configuration of the E. coli Lipid A and CM15 peptide after a simulation time of 8 µs. The CM15 peptide forms an α-helical structure in the hydrophobic Lipid A region.
In the previous section, we characterized the mechanism of membrane insertion of the CM15 peptide into the E. coli R3 O111 membrane. Ma (Ma et al., 2021) demonstrated that the CM15 peptide inserted and destroyed a homogenous DPPG membrane (model membrane for the inner membrane of Gram-negative bacteria) by inserting into the membrane and disrupting the interactions between the phospholipids. In contrast, much less is known about the mode of action of CM15 on the outer membrane (OM). Most all-atom simulations and experimental studies of antimicrobial peptides (AMPs) have focused on their role in compromising the inner membrane. But for our project idea, which is to bind AMP to endolysins, it would be crucial to understand the mechanism of outer membrane disruption through AMPs leading to a higher permeability of the outer membrane to other molecules.
As outlined in Chapter 1, the outer membrane of Gram-negative bacteria is stabilized by extensive interactions between divalent cations (Ca²⁺, Mg²⁺) and negatively charged phosphate groups in the lipopolysaccharide (LPS) core–sugar (CS) region. These bridging interactions provide structural integrity to the membrane. Disrupting this stabilizing role of Ca²⁺ is therefore a promising strategy for increasing outer membrane permeability. Hancock (Hancock, 1997) proposed the most prominent mechanism for positively charged AMPs against the Gram-negative OM, described as the self-promoted uptake mechanism. In this model, the affinity of cationic peptides for LPS is estimated to be roughly three times stronger than that of Ca²⁺. As a result, AMPs competitively displace Ca²⁺ ions, weaken LPS cross-linking, and create transient structural defects or “cracks” that facilitate peptide translocation. This mechanism was later discussed in reviews of endolysin-based antibacterial strategies (Gerstmans et al., 2016) and in the Artilysin study (Briers et al., 2014).
To assess whether CM15 follows a similar mechanism, we analyzed Coulomb interaction energies between Ca²⁺ ions and the CS region during a 4 µs simulation (Figure 20). At 1750 ns, coinciding with peptide insertion, the interaction energy increased from −40,000 kJ mol⁻¹ to −37,000 kJ mol⁻¹. This reduction in attractive Coulomb energy indicates weakened Ca²⁺–LPS interactions, consistent with partial displacement of Ca²⁺ by CM15. This finding supports the hypothesis that CM15 promotes local destabilization of the OM through electrostatic competition. Based on these results, we designed wetlab experiments to test whether CM15 can enhance OM permeability and facilitate endolysin passage.
Figure 20: Coulomb interaction Eint between Ca2+ ions and CS region during a simulation time of t = 4 µs. The configuration when the peptide inserts into the CS region of the outer membrane is shown at t = 1750 ns.
Hancock’s mechanism describes peptides creating transient OM “cracks”; however, this term lacks structural definition. To translate it into a quantifiable metric, we define such cracks as lipid packing defects or abnormal increases in packing defect size. We therefore analyzed lipid packing defects in two systems: (1) an E. coli R3 O111 membrane alone, and (2) an E. coli R3 O111 membrane with CM15. For this analysis, we used 3000 frames from the interval between 3700–4000 ns. Because of the limited system size, we could not reliably calculate the defect constant π (as in Gautier et al., 2018), but the qualitative differences remain revealing.
As shown in Figure 21, the control system without peptide (red points) displayed defect areas mostly below 130 Ų. In contrast, the CM15-containing system (blue points) exhibited frequent occurrences of large defects between 150–300 Ų. These extended defects represent significant disruptions to lipid packing, which likely correspond to transient openings through which peptides—and even coupled proteins such as endolysins—can cross the OM. This observation directly supports Hancock’s self-promoted uptake model: insertion of a cationic peptide into the CS/O-antigen region generates structural perturbations that promote further uptake.
Figure 21: Comparison of “deep” and “all” lipid packing defects of the E. coli R3 O111 (red) only membrane system and the E. coli R3 O111/CM15 (blue) system.
Although our current system size limits the statistical robustness of the packing-defect analysis, the findings strongly suggest that CM15 destabilizes LPS–Ca²⁺ interactions and creates lipid packing defects consistent with "self-promoted uptake". Future simulations with larger membranes will allow calculation of the defect constant π, thereby quantifying this effect more rigorously.
The goal of these simulations was to gain more insights into the mechanism of peptide insertion and membrane disruption, and to understand why peptide insertion could lead to higher membrane permeability. Clarifying this mechanistic link can guide the choice of which antimicrobial peptides (AMPs) can be effectively coupled with which endolysins.
First, we characterized the insertion of CM15 into the outer membrane of E. coli R3 O111. By analyzing the interaction energies between the CM15 and different membrane regions, we found that Coul interactions between the positively charged Lys residues of CM15 and the negatively charged CS region of the membrane act as the main driving force. At the same time, the negatively charged C-terminus of CM15 was found to experience repulsive Coulomb interactions with the CS region, which can destabilize the peptide–membrane interaction, even after partial insertion.
Next, we examined how peptide insertion affects the membrane. An important work for this question was the proposed “self-uptake” mechanism of Hancock (Hancock, 1997). It describes that the positively charged peptide competitively displaces the Ca2+ ions, leading to the membrane developing transient “cracks” which permit passage to a variety of molecules, and most importantly to the perturbing peptide itself, thus “self-uptake”. We showed through our simulations that the attractive interaction between the Ca2+ and the membrane decreases during the insertion of the peptide and that the lipid packing defects, which can be defined as “cracks” in the membrane, are much larger in the simulation with the inserted peptide than in the other simulation. These larger packing defects lead to increasing membrane permeability towards the peptide itself and other molecules. This also explains why coupling endolysins to an AMP is such a promising strategy to increase their effectivity against Gram-negative bacteria.
In previous chapters, we studied the mechanism by which the AMP CM15 inserts into a model E. coli R3 O111 membrane and destabilizing it. This chapter focuses on two additional membrane/peptide systems. The first involves the AMP (SMAP29), which has already been used as a peptide for Artilysins (Briers et al., 2014). The second investigates the E. coli K12 membrane, allowing us to compare our simulations with wetlab data.
In the work of Lavigne and coworkers, they designed the Artilysin Art-175, which consists of endolysin KZ144 with an N-terminal fusion to SMAP29. The Artilysin was found to be effective against Gram-negative bacterial strains like Acinetobacter baumannii (Defraine et al., 2016). A distinctive feature of the A. baumannii membrane is the absence of Ca²⁺ ions within the CS region; they are present only at the CS/lipid A interface, which is important to notice when analyzing the insertion of peptides into the A. baumannii (more to that in Case Study of E. coli K12).
From experimental data, we know that the LPS affinity of the more flexible N-terminus of SMAP29 has higher affinity for LPS than its more hydrophobic C-terminus (Tack et al., 2002). The same study also reported no intermolecular positive cooperativity. Thus, we only inserted one SMAP29 peptide into our simulation system.
The equilibrated configuration of the A. baumannii/SMAP29 system is shown in Figure 22. The exact number of lipid A and phospholipid used in this system can be found in Table 1 .
Figure 22: Equilibrated configuration of A. baumannii membrane with one SMAP29 peptide placed on top of the membrane. The N-terminus of the peptide is marked in green, and the C-terminus is marked in pink.
The system was simulated for 4 µs, and the trajectories are shown in Figure 23. The SMAP29 peptide inserted into the membrane with the N-terminus, consistent with its stronger affinity for LPS. We could also observe that the C-terminus, after a simulation time of 4 µs, was located in the extracellular space.
Figure 23: Trajectory of the A. baumannii/SMAP29 system from 1 µs to 4 µs.
After analyzing the interaction energies between the CS region of the membrane and the SMAP29 peptide, we could observe a minimum at t = 400 ns, during which the C-terminus briefly inserted into the CS region before diffusing back into the extracellular space (Figure 24). This observation supports the experimental results of Tack et al., confirming the lower affinity of the C-terminus. Overall, insertion into the A. baumannii outer membrane proceeds preferentially via the N-terminus.
Figure 24: Interaction energies Eint as a function of the simulation time t. The configuration of the system at t = 400 ns is shown, where the C-terminus inserts into the CS region of the outer membrane.
Compared with CM15, SMAP29 consistently inserts with the N-terminus, whereas the C-terminus stays in the extracellular region (Figure 23). This suggests that, for the design of Artilysins, SMAP29 should be fused via its C-terminus to the N-terminus of an endolysin. It also suggests that SMAP29, or a longer peptide, may be more suitable for fusion than shorter AMPs such as CM15, since coupling SMAP29 to the endolysin at the C-terminus is less likely to interfere with the peptide’s mode of action compared to cases where the C-terminus itself inserts into the membrane.
The E. coli K12 outer membrane contains rough LPS. The key difference from the A. baumannii membrane is the Ca2+ amount in the CS region. The partially phosphorylated polysaccharides of E. coli K12 allow Ca²⁺ binding to the negatively charged phosphate groups (Figure 25). In contrast, Ca²⁺ is absent from the CS region of A. baumannii.
Figure 25: Configuration of outer membrane of a) A. baumannii and b) E. coli K12. The Ca2+ ions are represented by black spheres.
We performed unbiased simulations of both SMAP29 and CM15 with the E. coli K12 membrane. For each peptide, three initial orientations were tested: one oriented parallel to the membrane, one with the N-terminus perpendicular to the membrane, and one with the C-terminus perpendicular to the membrane. This resulted in three independent simulations for each peptide.
After analyzing the trajectories, we could not observe an incomplete insertion into the membrane in each of the simulations after a simulation time of t = 4000 ns. Figure 26 shows the configuration of the two systems with a parallel peptide starting configuration after a simulation time of t = 4000 ns. Also compared to the A. baumannii system and the E. coli R3 O111 system, the insertion into the outer membrane is much slower.
Figure 25: Configuration of E. coli K12 membrane system with CM15 on the left and SMAP29 on the right side. The peptide is localized at the CS/water interface.
The reason for the improper insertion of the peptide into the CS region is the repulsive interactions between the positively charged peptides and the Ca2+ cations. Because the competitive displacement of Ca²⁺ by the peptides occurs too slowly to be simulated within our timescales, the peptide does not advance into the Lipid A/CS interface after t = 4000 ns. This also explains why the calculated free energy along the z-axis in our umbrella sampling shows an energy barrier at the middle of the CS interface, where the Ca2+ cations are located.
Our wetlab performed assays with E. coli K12 biolfilms by evaluating a combinatorial approach involving bacteriophages, antimicrobial peptides (AMPs), and antibiofilm peptides, including K14 — a modified analogue of CM15 with reduced hydrophobicity due to the substitution of a valin residue by lysin. The K14 exhibited moderate biofilm reduction at low concentrations, whereas higher concentrations unexpectedly promoted biofilm formation relative to the buffer control.
A possible explanation — derived from the drylab case studies to E. coli K12 to explain this phenomenon of K14 losing effectiveness when increasing the concentration of the peptide in biofilm — is related to the high charge density of K14. We showed in our modeling approach that the insertion into the E. coli K12 is a much slower process than the insertion into a “smooth” membrane or A. baumannii. When increasing the concentration of the K14, the intermolecular repulsive interactions between the peptides could disrupt the insertion process of the K14 into the membrane. An alternative explanation with a more detailed explanation of biofilms can be found on the wetlab 2 page.
With our AMP-MD platform, we developed a framework to build, equilibrate and simulate a wide variety of membrane/peptide systems in silico. In addition, it provides analysis tools to study these systems. All the scripts (with MD inputs) used for initiating, running, and analyzing the membrane/peptide simulations are uploaded on the iGEM Münster Gitlab Repository.
By utilizing this AMD-MD setup, we first studied the interaction free energy through Umbrella Sampling, which showed that the insertion into the CS region is an energetically favored process. This motivated us to start unbiased simulations which enabled us to analyze the spontaneous insertion into the outer membrane. The next step was to investigate the effects on the membrane caused by this insertion. We showed that the insertion of the AMP into the CS region already increases the membrane permeability (increase in size of lipid packing defects) which is a crucial process to understand especially for our wetlab strategy of combining AMPs and endolysins to target Gram-negative bacteria. To further help with the decision process, we started two case studies which explained why the SMAP29 should be fused at the C-terminus and why a longer peptide (30 aa) is more suited for such a fusion.
We hope that our modeling project encourages other teams to use the AMP-MD setup to either advance our studies on AMP insertion or expand this platform to protein/membrane systems which could lead to a variety of new applications such as the adsorption of peripheral protein on membranes.
Abraham, M.J. et al. (2015) “GROMACS: High performance molecular simulations through multi-level
parallelism from laptops to supercomputers,” SoftwareX, 1–2, pp. 19–25. Available at:
https://doi.org/10.1016/j.softx.2015.06.001.
Briers, Y. et al. (2014) “Engineered Endolysin-Based ‘Artilysins’ To Combat Multidrug-Resistant Gram-Negative Pathogens,”
mBio, 5(4), p. 10.1128/mbio.01379-14. Available at:
https://doi.org/10.1128/mbio.01379-14.
Bussi, G., Donadio, D. and Parrinello, M. (2007) “Canonical sampling through velocity rescaling,”
The Journal of Chemical Physics, 126(1), p. 014101. Available at:
https://doi.org/10.1063/1.2408420.
Cui, H., Lyman, E. and Voth, G.A. (2011) “Mechanism of Membrane Curvature Sensing by Amphipathic Helix Containing Proteins,”
Biophysical Journal, 100(5), pp. 1271–1279. Available at:
https://doi.org/10.1016/j.bpj.2011.01.036.
Darden, T., York, D. and Pedersen, L. (1993) “Particle mesh Ewald: An N⋅log(N) method for Ewald sums in large systems,”
The Journal of Chemical Physics, 98(12), pp. 10089–10092. Available at:
https://doi.org/10.1063/1.464397.
Defraine, V. et al. (2016) “Efficacy of Artilysin Art-175 against Resistant and Persistent Acinetobacter baumannii,”
Antimicrobial Agents and Chemotherapy, 60(6), pp. 3480–3488. Available at:
https://doi.org/10.1128/aac.00285-16.
Deylami, J., Chng, S.S. and Yong, E.H. (2024) “Elucidating Antibiotic Permeation through the Escherichia coli Outer Membrane: Insights from Molecular Dynamics,”
Journal of Chemical Information and Modeling, 64(21), pp. 8310–8321. Available at:
https://doi.org/10.1021/acs.jcim.4c01249.
Erridge, C., Bennett-Guerrero, E. and Poxton, I.R. (2002) “Structure and function of lipopolysaccharides,”
Microbes and Infection, 4(8), pp. 837–851. Available at:
https://doi.org/10.1016/S1286-4579(02)01604-0.
Ewald, P.P. (1921) “Die Berechnung optischer und elektrostatischer Gitterpotentiale,”
Annalen der Physik, 369(3), pp. 253–287. Available at:
https://doi.org/10.1002/andp.19213690304.
Frishman, D. and Argos, P. (1995) “Knowledge-based protein secondary structure assignment,”
Proteins: Structure, Function, and Bioinformatics, 23(4), pp. 566–579. Available at:
https://doi.org/10.1002/prot.340230412.
Gautier, R. et al. (2018) “PackMem: A Versatile Tool to Compute and Visualize Interfacial Packing Defects in Lipid Bilayers,”
Biophysical Journal, 115(3), pp. 436–444. Available at:
https://doi.org/10.1016/j.bpj.2018.06.025.
Gerstmans, H. et al. (2016) “From endolysins to Artilysin®s: novel enzyme-based approaches to kill drug-resistant bacteria,”
Biochemical Society Transactions, 44(1), pp. 123–128. Available at:
https://doi.org/10.1042/BST20150192.
Hancock, R.E. (1997) “Peptide antibiotics,”
The Lancet, 349(9049), pp. 418–422. Available at:
https://doi.org/10.1016/S0140-6736(97)80051-7.
Hess, B. et al. (1997) “LINCS: A linear constraint solver for molecular simulations,”
Journal of Computational Chemistry, 18(12), pp. 1463–1472. Available at:
https://doi.org/10.1002/(SICI)1096-987X(199709)18:12%253C1463::AID-JCC4%253E3.0.CO;2-H.
Hockney, R.W., Goel, S.P. and Eastwood, J.W. (1974) “Quiet high-resolution computer models of a plasma,”
Journal of Computational Physics, 14(2), pp. 148–158. Available at:
https://doi.org/10.1016/0021-9991(74)90010-2.
Huang, J. et al. (2017) “CHARMM36m: an improved force field for folded and intrinsically disordered proteins,”
Nature Methods, 14(1), pp. 71–73. Available at:
https://doi.org/10.1038/nmeth.4067.
Hub, J.S., de Groot, B.L. and van der Spoel, D. (2010) “g_wham—A Free Weighted Histogram Analysis Implementation Including Robust Error and Autocorrelation Estimates,”
Journal of Chemical Theory and Computation, 6(12), pp. 3713–3720. Available at:
https://doi.org/10.1021/ct100494z.
Humphrey, W., Dalke, A. and Schulten, K. (1996) “VMD: Visual molecular dynamics,”
Journal of Molecular Graphics, 14(1), pp. 33–38. Available at:
https://doi.org/10.1016/0263-7855(96)00018-5
.
Javed, R., Kapakayala, A.B. and Nair, N.N. (2024)
“Buckets Instead of Umbrellas for Enhanced Sampling and Free Energy Calculations,”
Journal of Chemical Theory and Computation, 20(19), pp. 8450–8460. Available at:
https://doi.org/10.1021/acs.jctc.4c00776
.
Jiang, X. et al. (2020) “Molecular dynamics simulations informed by membrane lipidomics reveal the structure–interaction relationship of polymyxins with the lipid A-based outer membrane of Acinetobacter baumannii,”
Journal of Antimicrobial Chemotherapy, 75(12), pp. 3534–3543. Available at:
https://doi.org/10.1093/jac/dkaa376.
Jo, S. et al. (2008) “CHARMM-GUI: A web-based graphical user interface for CHARMM,”
Journal of Computational Chemistry, 29(11), pp. 1859–1865. Available at:
https://doi.org/10.1002/jcc.20945
.
Jones, J.E. (1924) “On the determination of molecular fields.—I. From the variation of the viscosity of a gas with temperature,”
Proceedings of the Royal Society of London. Series A, Containing Papers of a Mathematical and Physical Character, 106(738), pp. 441–462. Available at:
https://doi.org/10.1098/rspa.1924.0081.
Langbein, D. (1974) “Theory of Van der Waals attraction,” in D. Langbein (ed.) Springer Tracts in Modern Physics. Berlin, Heidelberg: Springer, pp. 1–139. Available at:
https://doi.org/10.1007/BFb0042407.
Lemkul, J.A. (2024) “Introductory Tutorials for Simulating Protein Dynamics with GROMACS,” The Journal of Physical Chemistry B, pp. 9418–9435. Available at:
https://doi.org/10.1021/acs.jpcb.4c04901.
Ma, L. et al. (2021) “Interaction between Antimicrobial Peptide CM15 and a Model Cell Membrane Affected by CM15 Terminal Amidation and the Membrane Phase State,”
Langmuir, 37(4), pp. 1613–1621. Available at:
https://doi.org/10.1021/acs.langmuir.0c03498.
Parrinello, M. and Rahman, A. (1981) “Polymorphic transitions in single crystals: A new molecular dynamics method,”
Journal of Applied Physics, 52(12), pp. 7182–7190. Available at:
https://doi.org/10.1063/1.328693.
Pauli, W. (1925) “Über den Zusammenhang des Abschlusses der Elektronengruppen im Atom mit der Komplexstruktur der Spektren,”
Zeitschrift für Physik, 31(1), pp. 765–783. Available at:
https://doi.org/10.1007/BF02980631.
Respondek, M. et al. (2007) “Mapping the Orientation of Helices in Micelle-Bound Peptides by Paramagnetic Relaxation Waves,”
Journal of the American Chemical Society, 129(16), pp. 5228–5234. Available at:
https://doi.org/10.1021/ja069004f.
Sharma, P. et al. (2024) “Structure of the Bacterial Cell Envelope and Interactions with Antimicrobials: Insights from Molecular Dynamics Simulations,”
Langmuir, 40(15), pp. 7791–7811. Available at:
https://doi.org/10.1021/acs.langmuir.3c03474.
Sharma, P. and Ayappa, K.G. (2022) “A Molecular Dynamics Study of Antimicrobial Peptide Interactions with the Lipopolysaccharides of the Outer Bacterial Membrane,”
The Journal of Membrane Biology, 255(6), pp. 665–675. Available at:
https://doi.org/10.1007/s00232-022-00258-6.
Tack, B.F. et al. (2002) “SMAP-29 has two LPS-binding sites and a central hinge,”
European Journal of Biochemistry, 269(4), pp. 1181–1189. Available at:
https://doi.org/10.1046/j.0014-2956.2002.02751.x.
Vamparys, L. et al. (2013) “Conical Lipids in Flat Bilayers Induce Packing Defects Similar to that Induced by Positive Curvature,”
Biophysical Journal, 104(3), pp. 585–593. Available at:
https://doi.org/10.1016/j.bpj.2012.11.3836.
Vollmayr-Lee, K. (2020) “Introduction to molecular dynamics simulations,”
American Journal of Physics, 88(5), pp. 401–422. Available at:
https://doi.org/10.1119/10.0000654.