Revealing the Specificity of a Range of Antimicrobial Peptides in Lipid Nanodiscs by Native Mass Spectrometry

Abstract

Antimicrobial peptides (AMPs) interact directly with lipid membranes of pathogens and may have the potential to combat antibiotic resistance. Although many AMPs are thought to form toxic oligomeric pores, their interactions within lipid membranes are not well understood. Here, we used native mass spectrometry to measure the incorporation of a range of different AMPs in lipoprotein nanodiscs. We found that truncation of human LL37 increases the lipid specificity but decreases specificity of complex formation. We also saw that reduction of disulfide bonds can have dramatic effects on the ability of AMPs to interact with lipid bilayers. Finally, by examining a wider range of peptides, we discovered that AMPs tend to interact specifically with anionic lipids but form nonspecific complexes with wide oligomeric state distributions. Overall, these data reveal that each AMP has unique behaviors, but some common trends apply to many AMPs.

Graphical Abstract

INTRODUCTION

Antimicrobial peptides (AMPs) are short, mostly cationic, peptides that target bacteria, fungi, and/or viruses by interacting with lipids in their cell membranes. AMPs have been discovered across a wide variety of organisms and play an important role in the innate immune system. Although most antibiotics inhibit cellular processes such as cell wall, protein, or nucleic acid synthesis,¹ AMPs generally interact directly with the lipids in the membrane.² This unique mechanism gives AMPs the potential to treat drug-resistant microbes.^3,4 However, the modes of action of many AMPs are not well understood. Some are thought to form pores in the membrane that allow the efflux of small molecules and ions.^5,6 Other AMPs are thought to cause massive disruption of the cell membrane, allowing small and large molecules to efflux.⁷ Finally, some are thought to pass through the cell membrane and inhibit synthesis of biomolecules.⁸ In all proposed mechanisms, interaction of the peptide with the lipid membrane is a critical step, and the negative charges on the head groups of bacterial lipids are thought to play an important role in driving interactions with cationic peptides.⁹ However, the oligomeric states and lipid interactions of AMPs inside lipid bilayers often remain unclear.

Measuring the stoichiometry of AMP complexes in membranes is challenging for three reasons. First, AMP complexes may be polydisperse, limiting conventional structural biology tools. Second, attaching labels to the small peptides may distort their natural interactions. Finally, lipid membranes are essential for studying their natural interactions, so AMP complexes cannot be effectively studied in detergent micelles. We addressed these obstacles by combining nanodiscs and native mass spectrometry (MS). Nanodiscs are a lipid bilayer encircled by two membrane scaffold proteins.¹⁰ Native MS uses nondenaturing ionization conditions to preserve intact nanodiscs for mass analysis.¹¹ We have previously shown that nanodiscs containing AMPs can also be preserved for mass analysis, which provides a label-free measurement of the numbers of peptides associated with the nanodiscs.¹² Preferences for specific oligomeric states of AMP complexes can be inferred from nonrandom distributions of the number of AMPs incorporated into the nanodiscs. Preferences for specific lipid interactions can be determined by comparing incorporation into nanodiscs composed of different lipids. Thus, native MS reveals the degree to which AMPs form specific oligomeric complexes and how AMPs behave differently in nanodiscs with different lipids that model either bacterial or mammalian membranes.

Our initial study explored three peptides: melittin, gramicidin A, and LL37.¹² Here, we applied our novel method to examine the stoichiometries and lipid preferences for a wider range of different commercially available AMPs. We first explored the effects of truncation on a larger human AMP and then the effects of disulfide reduction on cysteine-containing AMPs. Pooling these results with additional AMPs and the AMPs from our initial study, we discovered that each peptide showed different behaviors, but three clear trends emerged. First, most AMPs showed strong preference for anionic lipids. Second, most AMPs showed relatively minor preference for specific stoichiometries. Finally, the overall level of membrane incorporation was similar. Together, these results suggest that AMPs generally have clear preferences for lipids but weak preferences for oligomeric complex formation.

MATERIALS AND METHODS

Materials.

1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC) and 1,2-dimyristoyl-sn-glycero-3-phospho-(1’-rac-glycerol) (DMPG) lipids were purchased from Avanti Polar Lipids (Alabaster, AL). KR12 peptide was purchased from Bachem (Torrance, CA). Indolicidin, lactoferricin B, magainin 2 (UniProtKB C0HKN6), temporin A (UniProtKB P56917), and temporin L (UniProtKB P57104) were purchased from Anaspec (Fremont, CA). All peptides were at 95% purity or better. Ammonium acetate, Amberlite XAD-2, sodium cholate, and dithiothreitol (DTT) were purchased from Sigma Aldrich.

Nanodisc Assembly.

Nanodiscs were first assembled without peptides as previously described.^13,14 Briefly, either DMPC or DMPG lipids in chloroform were dried under nitrogen and overnight vacuum. Dried lipids were solubilized in 0.1 M sodium cholate. Membrane scaffold protein, MSP1D1(–), was expressed and purified from E. coli and mixed with lipids in 50 mM cholate.^10,14 MSP and lipids were mixed at around a 1:100 ratio with a final cholate concentration between 20–25 mM. In the reconstitution mixture, the MSP concentration was typically 115 μM with a lipid concentration of 11.5 mM. Detergent was removed by addition of Amberlite XAD-2 beads. Nanodiscs were removed from beads and purified using a Superose 6 Increase 10/300 column (GE Healthcare) with 0.2 M ammonium acetate at pH 6.8.¹¹

Mass Spectrometry Sample Preparation.

Nanodiscs were mixed with peptides at specific molar ratios of peptide:nanodisc for mass spectrometry as previously described.¹² Purified nanodiscs were diluted to 2.5 μM in ammonium acetate. Peptide stocks were prepared by dissolving a known mass of peptide in methanol and diluting to working concentrations. Peptides were added at specific molar ratios from 0–18:1 peptide:nanodisc. Lactoferricin B at 18:1 showed only free peptide with no signal for nanodiscs or MSP, so the highest ratio used was 12:1. Imidazole was added to all samples at a final concentration of 25 mM as previously described.¹² Briefly, a solution of 0.4 M imidazole was adjusted to pH 6.8 with acetic acid before being added to the nanodisc-peptide mixture. The final nanodisc concentration in the mixture was 2 μM. The samples were allowed to incubate for 5 minutes at room temperature (21–23 °C) to ensure equilibrium was reached before each measurement. However, the temperature of the measurement in the MS source was around 37 °C.

To reduce the disulfide bonds of lactoferricin B and bactenecin, we tested β-mercaptoethanol (BME), tris-carboxyethyl phosphine (TCEP), and dithiothreitol (DTT) and monitored addition of two hydrogens by MS of the peptide alone. BME did not reduce the disulfide bond. TCEP successfully reduced the disulfide bond, but it led to precipitation of the peptide and nanodiscs. DTT was also able to completely reduce the disulfide bond without precipitation, so we used DTT as the reducing agent. For samples that contained DTT, the peptide was incubated with 5 mM DTT for at least 10 minutes before the peptide was added to the nanodiscs. DTT was not removed or quenched prior to native MS. Control experiments confirmed that DTT did not influence the native mass spectra of nanodiscs in the absence of peptide.

Mass Spectrometry.

Native MS was performed using a Q-Exactive HF quadrupole-Orbitrap mass spectrometer with the Ultra-High Mass Range (UHMR) research modifications (Thermo Fisher Scientific, Bremen). Instrumental parameters were used as previously described.^12,13,15,16 Nano-electrospray ionization (ESI) was used to introduce the sample into the mass spectrometer with a capillary voltage of 1.1 kV and capillary temperature of 200 °C. Nanodiscs were analyzed from 2,000–25,000 m/z at a resolution setting of 15,000 and a trapping gas setting of 7. The source fragmentation was set to 50 V to aid in desolvation, and in-source trapping was set to 0 V. To analyze free peptides, the MS was operated from 500–5000 m/z with the resolution set to 120,000 and a trapping gas setting of 3. Source fragmentation was kept at 50 V to aid in desolvation.

Mass Spectrometry Data Analysis.

Deconvolution of native mass spectra was performed using UniDec and MetaUniDec.^16,17 Deconvolution settings for nanodiscs included a mass range of 20–200 kDa, a charge range of 5–25, and a peak fwhm of 10. Both charge and mass smoothing were used with a width of 1, and the lipid mass was used as the mass difference. Because nanodiscs have a broad mass distribution due to their intrinsic polydispersity, macromolecular mass defect analysis was used following deconvolution to determine the number of peptides incorporated into the nanodiscs.^12,18 In mass defect analysis, we divide the measured mass by the lipid mass and only keep the remainder, which is the normalized mass defect. Because each additional lipid changes the integer part of the quotient but not the remainder, the mass defect is only influenced by non-lipid components: the AMPs or MSP belts. Except where noted, every nanodisc has two MSP belts, so shifts in the mass defects reveal how many peptides have been incorporated into the nanodiscs. For example, KR12 has a mass of 1571 Da, and DMPG has a mass of 667 Da. Thus, the mass defect of KR12 is the remainder of 1571/667 or 0.355. Two MSP belts (22,044 Da each) give a mass defect of 2 × 0.05 = 0.10, so a nanodisc with two MSP belts and two KR12 peptides would have a mass defect of 2 × 0.355 + 0.10 = 0.810. Mass defect values for different levels of incorporated peptides were predicted (Table S1), and assignments were made by comparing the measured mass defect values with this table. In the case of potentially ambiguous mass defect values, we used CID and other peptide concentrations to infer the correct assignments. When two mass defect values were too close to be well resolved (and present simultaneously), we fit the mass defect peaks to overlapping Gaussians to determine how much each state contributed to the mass defect peak. All samples were measured with three replicate nanodiscs that were assembled separately, and error bars are shown as the standard deviation between the three replicates. Representative spectra and mass defects are shown.

RESULTS AND DISCUSSION

Comparing Full Length LL37 with Truncated KR12.

To determine how peptide length affects oligomerization and lipid interactions, we examined the interactions of human AMP LL37 and its truncated variant KR12 with lipid nanodiscs using native MS. As shown in Figure 1, we added AMPs directly to preformed nanodiscs, allowed them to associate with the lipid bilayer, and then analyzed the intact AMP-nanodisc complex with native MS (Figure 1A–C). Small shifts in the mass can be detected and quantified with mass defect analysis (Figure 1D) to determine the relative abundance of different stoichiometries of peptides incorporated in nanodiscs with either DMPC or DMPG lipids (Figure 1E). We chose DMPC and DMPG lipids because they formed very stable nanodiscs and yielded mass defect values that worked well for quantitation of many different peptides. Native MS with nanodiscs thus provides a label-free measurement of the stoichiometry of AMP insertion into different model lipid membranes.¹²

Figure 1.

Schematic showing experimental design and KR12 incorporation into nanodiscs. (A) Nanodiscs were mixed with KR12 at different molar ratios from 0–18:1 KR12:nanodisc and ionized by nano-ESI. (B) Mass spectra were obtained and (C) deconvolved using UniDec. (D) Normalized mass defects were used to quantify how many peptides were incorporated into the nanodiscs and relative intensities were extracted to show the number of peptides incorporated at different molar ratios with (E) DMPG nanodiscs.

KR12 is the shortest LL37 truncation that shows antimicrobial activity, representing the minimum functional unit.¹⁹ When KR12 was added to DMPC nanodiscs at KR12:nanodisc molar ratios from 3:1 to 18:1, we did not observe any significant change in the mass spectra (Figure 2B and Figure S1). Clearly, KR12 does not interact with DMPC lipid bilayers. In contrast, our prior research has shown that LL37 had a small amount of incorporation before the DMPC nanodiscs dissociated at a 9:1 molar ratio of LL37:nanodisc (Figure 2A).¹² These data revealed that interactions with DMPC require the full-length LL37.

Figure 2.

LL37 (A, C) and KR12 (B, D) incorporation in DMPC (A, B) and DMPG (C, D) nanodiscs. Native mass spectra of (A) LL37 and (B) KR12 added to DMPC nanodiscs at concentrations from 0–18:1 peptide:nanodisc show breakdown of the nanodisc by LL37 and no effects of KR12. Small amounts of three KR12 incorporated in B are not statistically significant and are artifacts caused by similar mass defects to empty nanodiscs. Insets show the relative abundance of different stoichiometries of incorporated peptides. The relative abundances of (C) LL37 and (D) KR12 incorporated into DMPG nanodiscs show broader distributions with KR12. Data for LL37 is from Walker et al. (2019),¹² and D is taken from Figure 1. The full sequence of LL37 is shown at the top with the KR12 truncation highlighted in red.

In contrast to DMPC, we saw clear incorporation of KR12 into DMPG nanodiscs, and up to eight KR12 peptides were incorporated (Figure 2D). The overall level of incorporation was similar to LL37 in DMPG nanodiscs (Figure 2C). However, LL37 had a stark preference for specific stoichiometries (such as 4 or 6 peptides per nanodisc), whereas KR12 showed a broader distribution in stoichiometries. Interestingly, these data revealed that KR12 has a clear preference for DMPG lipids and is even more selective than LL37 for lipid head groups despite its shorter length. However, LL37 forms more specific oligomeric complexes than KR12, which appears to incorporate mostly randomly. Overall, the short KR12 segment of LL37 seems to control lipid specificity whereas the rest of the sequence is responsible for controlling the formation of specific oligomeric complexes.

Investigating the effect of disulfide reduction on AMP interactions.

Some AMPs contain multiple cysteine residues that may form disulfide bonds in vivo, including lactoferricin B (LacB) and bactenecin. Previous studies have shown that LacB shows little difference in activity in the disulfide form compared to the reduced form,²⁰ but bactenecin in the reduced form showed greater activity against some bacterial species and lower activity against others.²¹ It is unclear how these differences in activity in the reduced form are dependent on lipid interactions. Here, we investigated the incorporation of LacB and bactenecin into nanodiscs containing DMPC or DMPG lipids and quantified the different levels of incorporation in the absence or presence of a reducing agent.

LacB is a 25-residue cationic peptide derived from the enzymatic digestion of bovine lactoferrin. It has antibacterial, antifungal, antiviral, antiparasitic, and antitumor activity.^22–29 LacB contains two cystine residues at opposite ends of the peptide that form a disulfide bond.²² When disulfide-containing LacB was added to DMPC nanodiscs, no LacB was incorporated, even up to a 12:1 molar ratio of LacB:nanodisc (Figure S2). The lack of incorporation demonstrated that no nonspecific peptide association occurs as a result of the ESI process or binding of LacB to the MSP belt. In contrast, LacB readily incorporated into DMPG nanodiscs (Figure 3A–C). The incorporation of LacB appeared random with no preferences for specific stoichiometries observed (Figure S2). Thus, oxidized LacB showed clear preferences for PG over PC lipids but no specific formation of oligomeric complexes.

Figure 3.

Lactoferricin B incorporation in DMPG nanodiscs without (A–C) and with (D–F) 5 mM DTT reveals that reduced LacB breaks apart DMPG nanodiscs. (A, D) Raw mass spectra of 0–12:1 molar ratio of LacB:nanodisc and (B, E) deconvolved mass spectra. (C, F) Mass defect analysis with the number of peptides incorporated annotated. 0* corresponds to broken nanodiscs around 75 kDa that have lost one MSP belt and have no incorporated LacB.

Similar to oxidized LacB, reduced LacB showed very little incorporation into DMPC nanodiscs (Figure S3). At a 3:1 molar ratio, the oxidized and reduced forms showed similar levels of incorporation into DMPG nanodiscs (Figure 3). However, at higher ratios, reduced LacB destabilized DMPG nanodiscs, and only broken fragments of nanodiscs were observed. Although some minor fragments were observed at high molar ratios, disulfide-containing LacB did not destabilize nanodiscs to the same extent. Together, these data showed that reduced LacB was more disruptive to nanodiscs whereas oxidized LacB is more likely to incorporate into the bilayer without causing the nanodisc to break apart. Because prior reports showed that reduced and oxidized LacB had similar activities,²⁰ these data may indicate that the membrane insertion, which is similar at 3:1 LacB:nanodisc in both forms, is the critical factor for LacB activity.

Bactenecin, a 12-residue cationic peptide from bovine neutrophils³⁰, also has a disulfide bond and shows antimicrobial activity against Gram-negative bacteria.²¹ Like LacB, no interaction of bactenecin was observed with DMPC nanodiscs (Figure S4). Oxidized bactenecin incorporated into DMPG nanodiscs, but no specific stoichiometries were preferred (Figure 4A–D). Interestingly, reduced bactenecin did not incorporate into either DMPC or DMPG nanodiscs, showing that reduction eliminated its ability to interact with DMPG lipids (Figure 4E–F and Figure S5).

Figure 4.

Bactenecin incorporation in DMPG nanodiscs without (A–D) and with (E–H) 5 mM DTT. (A, E) Raw mass spectra of 0–18:1 molar ratios of bactenecin:nanodisc, (B, F) deconvolved mass spectra, and (C, G) mass defect analysis showing incorporation of bactenecin in DMPG nanodiscs. (D, H) The relative amounts of bactenecin incorporated into DMPG nanodiscs.

These data agree with previous results from Wu et al. (1999)²¹ that showed reduced bactenecin having significantly higher minimum inhibitor concentrations against a range of bacteria. Interestingly, Wu et al.²¹ also showed that there are a few Gram-positive bacteria that have a lower minimum inhibitor concentration with reduced bactenecin. These functional data may suggest that our DMPG nanodiscs more closely mimic the different properties of the lipid bilayer of Gram-negative compared to Gram-positive bacteria,² or they may suggest that bactenecin has a different mode of action in Gram-positive bacteria that does not involve membrane interactions.

Exploring a wider range of peptides for nanodisc native MS.

After exploring the effects of truncation and reduction, we next measured the incorporation of a wider range of AMPs in DMPC and DMPG nanodiscs to see if a pattern of behavior emerged. We focused on commercially available AMPs that had been previously studied. Indolicidin is a 13-residue cationic peptide from bovine neutrophils that is rich in tryptophan and proline.³¹ It showed small amounts of incorporation into DMPC nanodiscs, but significant amounts incorporated into DMPG nanodiscs, indicating a preference for DMPG lipid (Figure S6). At low ratios with DMPG nanodiscs, there seemed to be a preference for two incorporated peptides, indicating potential dimer formation. However, the specificity was not observed at higher ratios, and some nanodisc dissociation was observed at an 18:1 molar ratio. Thus, indolicidin has a strong preference for DMPG but only minimal specificity for specific oligomeric states.

Magainin 2 (Mag2), a 23-residue cationic peptide from the African clawed frog, Xenopus laevis³², showed no interaction with DMPC and significant incorporation into DMPG nanodiscs (Figure S7). Interestingly, although the stoichiometry distributions were mostly random, the mass defect value associated with 7 peptides per nanodisc was never observed for Mag2 in DMPG nanodiscs. This missing value indicates that Mag2 might not show totally random incorporation and may form some complexes with partial specificity.

Temporin A (TempA) and temporin L (TempL) are both 13-residue cationic peptides secreted from the skin of the European common frog, Rana temporaria³³. TempA has a +1 charge, and TempL has a +2 charge. TempA showed no interaction with DMPC nanodiscs and only minimal interaction with DMPG nanodiscs (Figure S8). Interestingly, although TempA did not significantly incorporate into DMPC and DMPG nanodiscs, it caused significant loss of lipids from each type of nanodisc, suggesting it may destabilize the membrane in solution or may interact too weakly with the nanodisc to be retained during native MS. TempL showed small amounts of interaction with DMPC nanodiscs and significant interactions with DMPG nanodiscs (Figure S9). Interestingly, TempL had relatively narrow stoichiometry distributions, suggesting potential specificity for complex formation. Thus, TempA had minimal interaction with DMPC and DMPG nanodiscs, but TempL interacted significantly with DMPG nanodiscs with some specificity for complex formation.

Combining these results with our prior work on LL37, melittin, and gramicidin A¹² and the new results above from KR12, LacB, and bactenecin, we observed that AMPs can have diverse behaviors with regards to their preferences for lipid interactions and formation of specific oligomeric complexes (Figure 5). For example, gramicidin A formed highly specific dimers but did not show a strong preference for incorporation into PC or PG bilayers, likely because gramicidin A is highly hydrophobic and uncharged. In contrast, LacB showed a complete preference for incorporation into PG nanodiscs but did not seem to show formation of specific oligomeric complexes in the membrane.

Figure 5.

Summary of AMP lipid and complex specificity. Structures were generated using PDB files: 1MAG (gramicidin A), 1G89 (indolicidin), 2NA3 (KR12), 1LFC (LacB), 5NNT (LL37), 2MAG (Mag2), 6DST (melittin), 2MAA (TempA), 6GS5 (TempL). Residues are colored as either nonpolar (white), polar (green), cationic (blue), or anionic (red). Peptide sequences are provided in Table S1.

Although each peptide showed unique behaviors, we found that most of the ten AMPs in these studies behaved more like LacB than gramicidin A, reinforcing the unique nature of gramicidin A. In other words, most were more specific for model bacterial membranes (DMPG) and showed limited interactions with model mammalian membranes (DMPC). Although some AMPs, such as reduced LacB, caused the nanodisc to break apart, most inserted into the membrane without major disruption of the complex. Interestingly, most of the AMPs tested showed similar levels of overall incorporation (Figure 6). For peptide-peptide interactions, some complexes, such as LL37 and gramicidin A, showed clear preferences for specific oligomeric states, but most AMPs in these studies showed minimal specificity for complex formation. Although each AMP is unique, the overall trends are for stronger lipid specificities and weaker oligomeric state specificities (Figure 5).

Figure 6.

Average incorporation of different peptides in DMPG nanodiscs. The weighted average of the number of peptides incorporated for each peptide is shown at the low (A), intermediate (B), and high (C) peptide concentrations. *LacB high peptide concentration was 12:1 whereas all others were at 18:1. Data for LL37, gramicidin A, and melittin are taken from ref [11]. Peptides are annotated in A and are sorted from lowest to highest incorporation in that panel. Bactenecin and LacB are shown for oxidized forms.

CONCLUSIONS

Here, we explored the lipid and oligomeric state specificity for a wider range of complexes using native MS and nanodiscs. We discovered that truncation of LL37 did not change the overall level of incorporation, but the truncated KR12 showed both increased specificity for DMPG over DMPC lipids and decreased specificity for oligomeric complex formation. We also found that reduction of disulfide bonds caused LacB to break apart DMPG nanodiscs at lower concentrations, but reduction of bactenecin eliminated its interactions with membranes entirely. Combined, these data confirm that the structure of the peptide can have profound impacts on its interactions, and each peptide shows unique behaviors.

Although each peptide is unique, we discovered some common trends among the ten AMPs in these studies. First, most AMPs did not form highly specific complexes, but rather tended to form complexes with a range of oligomeric states. Thus, formation of specific oligomeric complexes may not be required for its toxicity. Second, most AMPs had strong preferences for bacterial PG lipids over mammalian PC lipids, which showed that lipid head groups play an important role in AMP selectivity. Future work will examine how lipid tails affect incorporation of AMPs and explore other lipid subclasses that are commonly found in different bacterial species, such as phosphatidylethanolamine and cardiolipin. Finally, the overall levels of incorporation were similar between many of the AMPs. Although these ten peptides represent only a small subset of the thousands of AMPs that have been discovered, the overall trends we observed shed light on AMP mechanisms more broadly and reveal the important role that lipid head groups play in AMP interactions.