Mechanism of Four de Novo Designed Antimicrobial Peptides*

As pathogenic bacteria become resistant to traditional antibiotics, alternate approaches such as designing and testing new potent selective antimicrobial peptides (AMP) are increasingly attractive. However, whereas much is known regarding the relationship between the AMP sequence and potency, less research has focused on developing links between AMP properties, such as design and structure, with mechanisms. Here we use four natural AMPs of varying known secondary structures and mechanisms of lipid bilayer disruption as controls to determine the mechanisms of four rationally designed AMPs with similar secondary structures and rearranged amino acid sequences. Using a Quartz Crystal Microbalance with Dissipation, we were able to differentiate between molecular models of AMP actions such as barrel-stave pore formation, toroidal pore formation, and peptide insertion mechanisms by quantifying differential frequencies throughout an oscillating supported lipid bilayer. Barrel-stave pores were identified by uniform frequency modulation, whereas toroidal pores possessed characteristic changes in oscillation frequency throughout the bilayer. The resulting modes of action demonstrate that rearrangement of an amino acid sequence of the AMP resulted in identical overall mechanisms, and that a given secondary structure did not necessarily predict mechanism. Also, increased mass addition to Gram-positive mimetic membranes from AMP disruption corresponded with lower minimum inhibitory concentrations against the Gram-positive Staphylococcus aureus.

All multicellular organisms such as animals and plants protect themselves against pathogenic microbes by producing antimicrobial peptides (AMPs) 2 that selectively disrupt bacterial cell membranes (1). Although they are enormously diverse, AMPs are mainly comprised of hydrophobic and cationic amino acids that are spatially organized along the molecule. Because bacteria depend on the integrity of their anionic cell membrane, disrupting their membrane with cationic peptides could offer an alternate strategy to conventional antibiotics for killing pathogenic bacteria (2). As these pathogens become resistant to traditional antibiotics, alternate approaches such as designing and testing new potent selective antimicrobial peptides are increasingly attractive.
The proper composition of amino acids, their sequential arrangement, and peptide length are essential for effective action of AMPs (3). It has been demonstrated that AMP activity is more closely tied to amino acid composition than amino acid sequence or AMP structure (4 -8). However, it has recently been shown that for the ␣-helical class of AMPs, ordering amino acids during AMP design into an imperfect amphipathic ␣-helix, a helix barrel-stave with one hydrophobic and one hydrophilic face where the hydrophobic face is disrupted by one hydrophilic amino acid, is beneficial for increasing AMP activity (9). Understanding the mechanism behind how these peptides disrupt cell membranes could benefit future designs of potent, selective AMPs.
Various mechanisms of interaction between AMPs and microbial lipid membranes include, (i) peptide adsorption to and expansion of the headgroup region (carpet, toroidal pore formation, or sinking raft mechanisms) resulting in positive curvature strain of a supported lipid bilayer (SLB) (10 -12), and (ii) peptide binding without the prerequisite positive strain (barrel-stave pore and electroporation mechanisms) (13). One instrument that can capture these interactions in real time is a quartz crystal microbalance with dissipation (QCM-D). Two main benefits to the QCM-D for studying peptide-lipid interactions over other surface measurement techniques are that (i) in addition to measuring wet-mass binding to the surface layer (i.e. the lipid bilayer) it also quantifies the energy dissipation or rigidity of the lipid bilayer in real time, and (ii) mass and dissipation changes are quantified across different penetration depths (14,15). This information allows for analyzing differential interactions at the lipid-water interface and the internal bilayer, which provides the basis for the determination of the mode of action of AMPs.

Results
Mechanistic information regarding the natural interactions of AMPs with SLBs are determined by examining kinetic (Fig. 1A I-IV ) o , which is the 9th overtone subtracted from the 3rd overtone normalized by the molecular weight of the peptide, and is used to represent the mass added at the lipid-water interface relative to the interior of the SLB. and equilibrium (Fig. 1B I-IV ) changes in the properties of the bilayer during AMP flow and subsequent buffer wash. These changes are quantified by the change in frequency (⌬F) and change in dissipation (⌬D) of an oscillating silica sensor, which is a technique commonly reported in the literature (15,(27)(28)(29)(30)(31)(32). Here, kinetic information is generally used to determine the rates and number of observable stages involved in each mode of action of the AMPs, whereas equilibrium data are related to the final mechanistic changes of the SLB. The equilibrium data, calculated by the changes in ⌬F (⌬⌬F) of the initial SLB (Fig. 1A, t ϭ 0 s) subtracted from the final bilayer properties after a buffer wash following AMP interaction (Fig. 1A, t ϭ 1,700 s), are plotted against overtone number where the overtone with the furthest penetration from the sensor surface, three, is plotted at the top, whereas shallower penetrating overtones are plotted below in order of penetration depth ( Fig. 1B I-IV ) (14). To quantify the change in bilayer properties as a parameter relating distance to the sensor, the difference in ⌬F for the ninth overtone (⌬F 9 ) is subtracted from ⌬F of the third overtone (⌬F 3 ), and that difference is normalized by the molecular mass of the AMP in kDa (⌬⌬F 3-9 o ). These values are overlaid onto their corresponding panels in Fig. 1B.
To understand how ⌬F and ⌬D relate to mass changes, the total mass (m) of the SLB was calculated using the third, fifth, seventh, and ninth overtones of ⌬F and ⌬D and an assumed SLB density, f , of 1,100 kg/m 3 using the Voigt model (14,26). Upon fitting the model, the mean film thickness (␦) was obtained. The total mass was then calculated using ␦, f , and the total surface area of the crystal, 154 mm 2 . ␦ was shown to be linearly dependent on f (R 2 ϭ 0.9687), whereas m was independent of f . Thus results are presented in terms of m and not ␦.

Natural AMPs
Kinetic Behavior-Addition of PG-1 induced a two-step effect on the SLB. In the first 60 s of AMP flow (0.50 nmol AMP dosage), ⌬F 3 decreased from Ϫ26.2 Ϯ 0.39 to Ϫ28.5 Ϯ 0.60 Hz, whereas ⌬D increased to 2.25 Ϯ 0.25 ϫ 10 Ϫ6 , indicating net mass addition to the SLB with decreased rigidity (Fig. 1A I ). Both ⌬F and ⌬D plateaued for an additional 60 s following the initial decrease (1.0 nmol of AMP dosage). Subsequent PG-1 flow resulted in an additional decrease in ⌬F to Ϫ45.1 Ϯ 2.8 Hz and increase in ⌬D to 9.53 Ϯ 0.34 ϫ 10 Ϫ6 (Fig. 1A I ). During the following buffer wash, ⌬F further decreased to Ϫ48.0 Ϯ 2.7 Hz coupled with a ⌬D increase to 12.3 Ϯ 0.26 ϫ 10 Ϫ6 indicating further mass addition to the SLB (Fig. 1A I ).
Applying the Voigt model to the kinetic data resulted in a change in mass (⌬m) once the AMP reached a dose of 0.47 nmol of PG-1 applied to the SLB (⌬m 1 ) to 0.28 Ϯ 0.030 g during the first step of PG-1 (Fig. 2, A and A I ). When the dosage of PG-1 reached ϫ10 the amount of ⌬m 1 , 4.7 nmol (⌬m 2 ), the SLB modeled mass increased to 2.12 Ϯ 0.096 g ( Fig. 2A, A II ), corresponding to significant ⌬F and ⌬D changes observed in Fig. 2A. Finally, during the buffer wash, the SLB mass (⌬m 3 ) ultimately increased to 3.16 Ϯ 0.25 g (Fig. 2, A and A III ). Mechanistically, this shows that PG-1 requires an initial critical concentration of peptide before severe damage occurs to the SLB. Assuming the initial mass change calculated by the Voigt model is due purely to AMP addition, the initial PG-1 amount to reach this critical threshold is 0.13 nmol of PG-1.
Equilibrium Behavior-PG-1 has large negative values of ⌬⌬F (all overtones ϽϪ20 Hz) and a value of ⌬⌬F 3-9 o ϭ 9.1 Hz kDa Ϫ1 , indicating significant frequency decreases, a parameter change commonly related to increasing mass (31), throughout the SLB with a preference for interaction furthest from the sensor surface ( (Fig. 2), it is possible for simultaneous increases in both ⌬F and ⌬D to cause an overall increase in the mass of the adsorbed layer (14). No change in signal was observed for ␣D-1 during the entire AMP flow and buffer wash (Fig. 1, A IV and B IV ).
B2, which comprises a rearrangement of the amino acids in B1, behaves similarly to B1 yet with a more pronounced effect (Fig. 3B I,II ). The values for ⌬⌬F are lower at every overtone indicating greater mass adsorption (⌬⌬F Ͻ Ϫ9 Hz for B2, ϽϪ4 Hz for B1 at every overtone, Fig. 3B I,II ). In addition, the dynamic behavior of B2 (Fig. 3A II ) shows a spontaneous increase in ⌬F and decrease in ⌬D during B2 cross-flow just prior to the buffer wash indicating a natural removal of mass, likely lipid mass, from the surface (32). Furthermore, ⌬F increases and ⌬D decreases after the buffer wash, which is potentially explained by loose lipids being removed from the SLB during the wash (Fig. 3A II ) (32).
The most potent designed AMP against Mycobacterium tuberculosis and many Gram-positive and Gram-negative bacteria, B3 (9), produced a similar ⌬⌬F overtone profile to B2, but with more pronounced dynamic results (Fig. 3, A II,III and B II,III ). The values for ⌬⌬F at every overtone were less than Ϫ8 Hz, which is comparable with B2, but significantly less than PG-1 and IND (Fig. 1B I,II ). However, the dynamic behavior shows B3 first decreases ⌬F and increases ⌬D, but then greatly increases ⌬F, whereas still increasing ⌬D (Fig.  3A III ). This is similar to the effect seen by B2 just prior to buffer wash, but much stronger and occurs much earlier during AMP flow (Fig. 3A II ).
B4, an AMP with nearly identical amino acids as B1-B3 yet ordered in a scrambled amino acid sequence, has a very similar ⌬⌬F overtone profile to B1-B3, but with a moderately weaker signal (all overtones ϽϪ3 Hz, Fig. 3B IV ) (9). With ⌬⌬F 3-9 o ϭ 1.9 for B4 (Fig. 3B IV ), which is close to the values of B1-B3, the  DECEMBER 2, 2016 • VOLUME 291 • NUMBER 49 JOURNAL OF BIOLOGICAL CHEMISTRY 25709 mechanism for B4 is apparently similar to B1-B3, but due to relatively lower ⌬F changes, is evidently not as effective.

Discussion
Previous reports characterizing the amino acid sequence, structure, mechanism of binding, and minimum inhibitory concentrations for four well studied natural AMPs allow comparison of their QCM-D results with those of four de novo designed AMPs (9, 16, 18 -24). This comparison provides a basis for suggesting the mechanism of interaction of the designed de novo AMPs when the AMPs are passed in solution across a deposited SLB consisting of 3:1 POPC:POPG lipids. o , which is the 9th overtone subtracted from the 3rd overtone normalized by the molecular weight of the peptide, and is used to represent the mass added at the lipid-water interface relative to the interior of the SLB.
Natural AMP Mechanisms-In comparison to PG-1, similar ⌬⌬F trends were reported for the AMP sheep myeloid antimicrobial peptide (SMAP-29), where values for ⌬⌬F for all overtones were ϽϪ10 Hz and ⌬⌬F [3][4][5][6][7][8][9] o ϳ4 Hz kDa Ϫ1 (ϽϪ20 Hz and ⌬⌬F 3-9 o ϳ 9.1 Hz kDa Ϫ1 for PG-1) (15). Wang et al. (15) proposed that SMAP-29 had a mechanism of peptide insertion across the bilayer, whereas possibly forming water channels along with adsorption to the SLB surface. Separately, PG-1 has been shown to act through toroidal pore formation where the pores consist of 8 -10 monomer units for lipids with 16 -18 carbon chain lengths, which are the lengths of the POPC and POPG lipid tails examined here (20). Therefore, the signature of ⌬⌬F for PG-1 (Fig. 1B I ) appears to be one for pore formation, likely toroidal pore formation, where there is also a significant amount of AMP adsorbed at the lipid-water interface.
For IND, Wang et al. (15) notably reported very different ⌬⌬F overtone profiles for IND at 5 and 10 M against egg PC lipids. They concluded that IND adsorbs at the lipid-water interface without peptide or water intercalation deep into the bilayer (15). A recent molecular dynamics simulation, which is in agreement with this result, reports IND causes thinning of a POPC lipid bilayer through adsorption at the lipid-water interface (21). In contrast, the ⌬⌬F overtone profiles in Fig. 1B II are much more similar to the pore formation with the surface adsorption mechanism of PG-1 than the profiles reported by Wang et al. (15) for membrane thinning caused by surface adsorption (21). This is likely due to the inclusion of the anionic POPG at a 3:1 mol ratio of PC:PG in our SLBs. IND is known to partition into large unilamellar vesicles (LUVs) composed of POPG with a partition coefficient 2 orders of magnitude higher than for POPC LUVs (22). Also, IND causes complete POPG LUV leakage, whereas it only results in 20 -30% leakage of POPC LUVs at the same peptide concentrations (22). Therefore, at the AMP concentration and lipid composition examined here, IND binds further into the membrane than solely through the surface adsorption mechanism previously observed with a PC SLB, and likely through water channel formation in toroidal pores combined with a large amount of surface adsorption similar to that observed for PG-1 (Fig. 1, B I and B II ).
MG-2, which is known to operate through a pore formation mechanism (34), undergoes a transition in the lipid interaction mechanism from membrane thinning at 4 M to pore formation at 7 M against 1:1 DOPC:DOPG mole ratio lipids (23). Here, at an intermediate peptide concentration (5 M) against a similar lipid system, MG-2 appears to act through a mechanism of peptide insertion across the SLB, a result that is consistent with the interaction of alamethicin, a peptide known to insert into zwitterionic lipid systems, with PC lipids (15). Alamethicin is shown to form cylindrical pores that stabilize the SLB with similar ⌬⌬F overtone profiles at 0.5 and 1 M AMP (15). A water channel, which would result in mass increases similar to those seen for PG-1 and IND, is not likely to form on addition of MG-2. Therefore, the ⌬⌬F overtone profile for MG-2 observed here gives strong support for peptide insertion orthogonal to the SLB plane without water channel formation (Fig. 1B III ).
No change in signal was observed for the ␣D-1-SLB interaction at the experimental conditions examined here. This is in agreement with a previous publication (24) detailing that the antimicrobial mechanism of ␣D-1 involves functional interaction with lipid II, a precursor molecule necessary for cell wall synthesis, and not LUVs. Furthermore, it was also shown that ␣D-1 induced minimal leakage of liposomes with similar lipid compositions at the concentration tested here (24). Thus, ␣D-1 should have a negligible interaction with the SLB under the currently examined conditions. Furthermore, it is shown that the total mass added at the end of AMP flow (⌬m 2 ) rank the natural AMPs in the order of PG-1 Ͼ IND Ͼ MG-2 Ͼ ␣D-1 ( Fig. 2A II ). This ordering is consistent with the natural activity of AMPs against the Grampositive Staphylococcus aureus, in terms of minimum inhibitory concentration ( Designed AMP Mechanisms-The difference in ⌬⌬F values for B1 in contrast to PG-1 indicates B1 has a significantly smaller ratio of mass bound at the lipid-water interface versus inside the SLB compared with PG-1 and IND. The more even distribution of ⌬⌬F with n for B1 suggests barrel-stave pore formation, a mechanism that has comparatively less peptide adsorbed at the lipid-water interface, as opposed to toroidal pore formation (32). Additionally, Wang et al. (15) showed that alamethicin, a pore forming AMP, resulted in uniform ⌬⌬F changes to a PC SLB (⌬⌬F [3][4][5][6][7][8][9] o ϳ 0 Hz kDa Ϫ1 ), supporting the contention here that B1 is a pore-forming AMP. Therefore, the mechanistic inference for B1 is probably a barrel-stave pore formation with relatively less surface adsorption at the lipidwater interface than for a toroidal pore.
Similar to B1, B2, B3, and B4 have minimal preference for adsorption at the lipid-water interface, with ⌬⌬F 3-9 o ϭ 0.94, 1.9, and 2.1 Hz kDa Ϫ1 for B2, B3, and B4, respectively (Fig. 3B), which is well below the values for AMPs with known preferences for interfacial adsorption (⌬⌬F [3][4][5][6][7][8][9] o ϳ 9 Hz kDa Ϫ1 , Fig.  1B). Interestingly, the dynamic behavior shows B3 first decreases ⌬F and increases ⌬D, but then greatly increases ⌬F while still increasing ⌬D (Fig. 3A III ). This is similar to the effect seen by B2 just prior to buffer wash, but much stronger and occurs much earlier during AMP flow (Fig. 3A II ). Therefore, a large amount of mass, relative to B2 and especially the other AMPs, appears to be naturally removed from the SLB, likely due to lipid removal from the surface, during peptide flow.
For B4, the kinetic behavior reveals a slight increase in ⌬F, a feature also observed throughout MG-1 flow (Figs. 1A III and 3A IV ). MG-1 inserts itself into the SLB without forming water channels similar to alamethicin (15). Thus there is evidence that B4 goes through an intermediate state of peptide insertion throughout the bilayer before transitioning to pore formation with water channels. It is possible that because B4 is a less potent AMP against numerous bacteria (9), the AMPs B1-B3 also undergo this transitional state but because they are more effective, the transition state is not observed at the AMP concentration examined here.
The ⌬⌬F overtone profiles indicate that the four designed AMPs follow a very similar mechanism to one of lipid removal with barrel-stave pore formation. However, the dynamic behavior of B1-B4 is quite different: B1 has one observable step to its kinetics, B2 has two stages of ⌬F and decreases with a slight increase prior to buffer wash, B3 has three completely unique stages of ⌬F changes, whereas B4 has a prolonged lag phase prior to a relatively slower one-step kinetic change (Fig.  3A I-IV ). Overall, these behaviors are in line with one all-encompassing four-step mechanism where different stages are observed depending on the antimicrobial activity of these peptides (B3 Ͼ B2 Ͼ B1 Ͼ B4) (9). The four stages are as follows. (i) An initial lag phase of AMP binding/insertion during which a critical AMP concentration must be reached prior to significant bilayer damage. (ii) A primary AMP water-channel formation stage during which there is rapid mass addition to the bilayer to help form the pores. (iii) A secondary mass addition into a pseudo-equilibrium state where there is slower mass addition followed by a plateau in ⌬F, whereas ⌬D continues to increase. (iv) A significant mass removal from the surface where ⌬F increases and ⌬D again continues to increase.
Step i of the overall mechanism is only observed by B4, the least potent designed AMP (Fig. 3A IV ) (9). The first step, which occurred during the first 300 s of B4 flow (2.5 nmol of B2 dosage), involved ⌬F increasing slightly from Ϫ25.5 Ϯ 0.72 to Ϫ24.7 Ϯ 0.86 Hz, whereas ⌬D increased from 0.21 Ϯ 0.068 ϫ 10 Ϫ6 to 0.51 Ϯ 0.080 ϫ 10 Ϫ6 (Fig. 3A IV ). During this period, there was very little net mass change to the SLB (⌬m ϭ 0.081 Ϯ 0.079 g). These signals for ⌬F and ⌬D are indicative of AMP insertion throughout the bilayer without water channels (similar to alamethicin (15)), and not AMP adsorption to only the surface of the SLB (IND (15)).
Step ii occurs with all 4 AMPs, but it is most obvious in B4 and B1. For B4, the primary water-channel formation stage occurred the slowest of the four AMPs, consistent with its lowest antimicrobial activity (Fig. 3A I-IV ) (9). After the initial lag phase of B4, ⌬F decreased to a final value of Ϫ30.9 Ϯ 0.79 Hz, whereas ⌬D increased to 2.39 Ϯ 0.56 ϫ 10 Ϫ6 over the final 300 s of B4 flow (2.5 nmol of B4 dosage) (Fig. 3A IV ). These changes are modeled by an increase in ⌬m to 0.50 Ϯ 0.13 g (Fig. 2B). For B1, these changes are qualitatively similar, but occur more rapidly. In the first 70 s of B1 flow (0.58 nmol of B1 dosage), ⌬F decreased from Ϫ24.8 Ϯ 0.16 to Ϫ32.0 Ϯ 0.39 Hz, whereas ⌬D increased to 1.84 Ϯ 0.29 ϫ 10 Ϫ6 (Fig. 3A I ). The modeled mass associated with these ⌬F and ⌬D changes is 0.60 Ϯ 0.032 g, which is the same as B4 within error (Fig. 2B). For B4 and B1, this is the final stage of the mechanism that these AMPs reach. The water-channel pores are sufficient for antimicrobial activity, as evidenced by their micromolar MICs against a range of bacteria (9). However, the more effective AMPs, B2 and B3, exhibited additional mechanistic stages beyond these initial two.
Step iii involves a slower rate of mass addition followed by a plateau of mass increase to the SLB and is only seen for B2 and B3. For B2, the secondary mass addition occurred when ⌬F decreased from Ϫ35.6 Ϯ 0.45 to Ϫ43.2 Ϯ 0.50 Hz and ⌬D increased from 4.41 Ϯ 0.16 ϫ 10 Ϫ6 to 7.70 Ϯ 0.60 ϫ 10 Ϫ6 over the course of 450 s, whereas the primary mass addition occurred in the first 120 s of B2 flow (Fig. 3A II ). The net mass addition reached a maximum of 1.20 Ϯ 0.16 g in this time period, a higher value than for any other designed AMP (Fig.  2B). B3 underwent a qualitatively similar process. ⌬F decreased from Ϫ31.0 Ϯ 1.73 to Ϫ35.8 Ϯ 1.45 Hz, whereas ⌬D increased from 2.11 Ϯ 0.58 ϫ 10 Ϫ6 to 3.31 Ϯ 0.98 ϫ 10 Ϫ6 over 130 s AMP flow (Fig. 3A III ). During this time, B3 had a maximum net mass addition to the SLB of 0.63 Ϯ 0.015 g while under AMP flow (Fig. 2B).
The final step (step iv) is significant lipid removal from the SLB. Both B2 and B3 have final stages where there is a ⌬F increase coupled with a ⌬D increase. For B2, it occurs in the final 30 s of AMP dosage, where ⌬F increases from a minimum of Ϫ43.2 Ϯ 0.50 to Ϫ41.8 Ϯ 0.48 Hz (Fig. 3A II ). For B3, this stage occurs over a more prolonged period where ⌬F increases from a minimum of Ϫ35.8 Ϯ 1.45 to Ϫ30.5 Ϯ 1.03 Hz and ⌬D increases from 3.31 Ϯ 0.98 ϫ 10 Ϫ6 to 4.45 Ϯ 0.15 ϫ 10 Ϫ6 (Fig.  3A III ). The changes for B3 are coupled with a very slight total mass decrease from a maximum of 0.63 Ϯ 0.015 to 0.58 Ϯ 0.073 g, although these numbers only reflect the wet plus dry mass change to the surface, and not necessarily just the dry lipid mass that remains on the sensor (Fig. 2B). B2 does not result in a significant mass decrease due to the large error associated with the modeling of the B2 mass change. However, the ⌬F and ⌬D changes seen here for B2 and B3 are typical for an AMP with a mechanism of lipid removal from an SLB, as seen for chrysophsin-3 (32). Thus, the four stages of the designed AMP mechanism are observed throughout the dynamic behavior of the four AMPs, where the initial stages are visible for the less potent AMPs, and the more potent AMPs proceed into the final two stages of the mechanism (Fig. 3). Representative schematics for the mechanisms of all natural and designed AMPs are shown in Fig. 4.

Experimental Procedures
AMPs Natural Peptides-The four natural AMPs were synthesized using standard Fmoc/t-butyl chemistry by Dr. Haydn Ball from the University of Texas Southwestern (Dallas, TX) and were purified to over 95% purity using HPLC in Dr. Ball's laboratory. All peptides with the exception of Magainin II were synthesized on a Protein Technologies Inc. Symphony peptide synthesizer using standard N,N,NЈ,NЈ-tetramethyl-O-(1H-benzotriazol-1yl)uronium hexafluorophosphate (HBTU)/hydroxybenzotriazole (HOBt) activation protocols. Magainin II was synthesized on a Applied Biosystems 433A using HCTU activation. Cleavage of the peptides was achieved using 95% TFA with thioanisole and 1,2-ethanedithiol scavengers between 1.5 and 2.5 h at room temperature. The crude peptides were precipitated in cold diethyl ether and the pellet was washed 3 times with fresh ether. The peptides were analyzed and purified by Waters RP-HPLC using Vydac C4 columns. The purified fractions were characterized using either Micromass MALDI-TOF or Agilient ESI-MS mass spectrometers. Peptides were received as lyophilized powders and were dissolved in 10 mM phosphate-buffered saline, pH 7.4 (Sigma, catalog number P4417), prior to use.
Synthetic Peptides-The synthetic AMPs were produced using an automated Multiprep RS synthesizer (Intavis AG, Germany) in the laboratory of Dr. Pankaj Karande. Fmoc solid-phase chemistry was used to synthesize the peptides from their C termini to N termini on a TentaGel rink amide resin (0.25 mmol/g) (Intavis Inc., Chicago, IL). Pre-synthesis, the resin was swollen in a dimethylformamide:dichloromethane (2:1) solution. Post peptide-chain assembly, the resin was washed with dichloromethane and the peptides were cleaved off using a TFA:TIS:H 2 O (88:6:6) mixture. Bulk TFA was removed by precipitating the peptides in ice-cold methyl tert-butyl ether (MTBE) followed by centrifugation and a second MTBE wash. Peptides were air dried and dissolved in acetonitrile:H 2 O (1:5) for lyophilization. Lyophilized peptides were stored at Ϫ20°C.
Mass Spectrometry Analyses-LC-MS/MS experiments were performed on Thermo LTQ XL Orbitrap mass spectrometers (Thermo, Bremen, Germany). Samples were injected using an Agilent 1200 C18-HPLC system (Agilent, Palo Alto, CA) using an Agilent 1200 autosampler. Thermo Biobasic column (150 ϫ 2.1 mm; particle size 5 m) was used for separation. The injection volume was 3 l and a flow rate of the mobile phase was 250 ml/min. The mobile phase consisted of 0.2% formic acid (solvent A) and 0.2% formic acid in acetonitrile (solvent B). The mass spectrometers operated in ESI mode with detection of positively charged ions using the Orbitrap as detector in m/z range 300 -2,000. The resolution was ϳ30,000 and mass accuracy was better than 3 ppm. LC-MS/MS gave the measured masses of (MϩH) ϩ as m/z (observed) 1885.1247, 1885.1212, 1885.1210, and 1812.1253 for B1, B2, B3, and B4, respectively. These results confirm that all four peptides were amidated at the C termini and is in agreement with the theoretical mass, m/z ϭ 1885.1224 (B1-B3) and 1812.1272 (B4). Finally, we also ran tandem MS/MS and all amino acid sequences as listed above.

Lipid Vesicle Preparation
15 mol of total lipids, 75% by mole of POPC (Avanti Lipids, Alabaster, AL, catalog number 850457) (8.55 mg) and 25% by mole of POPG (sodium salt) (Avanti Lipids, catalog number 840457) (2.89 mg), were dissolved in chloroform (Sigma, catalog number 528730) in a round bottom flask. Chloroform was evaporated under a N 2 stream in a 45°C water bath. Lipids were lyophilized overnight to remove residual chloroform. Lipid film was rehydrated in 3 ml of MilliQ TM water. Lipid vesicles were extruded (Mini-extruder kit, Avanti Lipids, catalog number 610000) 21 times through a 30-nm polycarbonate membrane (Avanti Lipids, catalog number 610002) at 42°C. 300 l of lipid vesicle solution was added to 10 ml of phosphate buffer saline, pH 7.4, containing an additional 140 mg of NaCl.

Silica Crystal Preparation
50-nm SiO 2 crystals (Biolin Scientific, Linthicum Heights, MD, catalog number QSX 303) were initially washed with MilliQ TM H 2 O, EtOH (Sigma, catalog number 792780), and dried with N 2 . Atmospheric plasma (ATOMFLO, Surfx Technologies LLC, Culver City, CA) using 30 liters/min of helium, 0.2 liter/min of O 2 , and 120 W power was used to clean the crystal surface. Crystals were then placed inside a flow module for use (Q-Sense Flow Module, QFM 401). After experimentation, crystals were washed with H 2 O, EtOH, dried with N 2 , and stored in air for further use.

QCM-D
Changes in frequency (⌬F) and change in dissipation (⌬D) at overtones 3, 5, 7, and 9 were obtained from the quartz crystal microbalance with dissipation (QCM-D, Biolin Scientific/Q-Sense, Q-Sense E4 Auto, Västra Frölunda, Sweden). These correspond to changes in mass and rigidity, respectively, for a rigid layer above the crystal. Lipid bilayer deposition was performed following a protocol from Cho et al. (25) AMP interaction with the newly formed SLB was carried out as follows: 5 M AMP flow for 10 min at 100 l/min followed by PBS flow for 20 min at 100 l/min. The SiO 2 surface was then regenerated with a wash of H 2 O for 10 min at 500 l/min, 2% SDS (powder purchased from Sigma, catalog number 436143) for 10 min at 500 l/min, H 2 O for 10 min at 500 l/min, and finally PBS for 10 min at 100 l/min to re-equilibrate the sensor.

Model
For non-rigid layers, which the SLBs become after interaction with the all AMPs except MG-2 and ␣D-1, changes in frequency (⌬F) and dissipation (⌬D) are related to changes in mass and viscoelastic changes to the SLB using the Voigt model, which is briefly reviewed here (14). For consistency, the Voigt model is used here to model all AMP-SLB interactions. ⌬F and ⌬D are related to mass and viscoelastic properties by, where L is the liquid viscosity (1,000 kg/m 3 ), ␦ L is the acoustic wave decay length, m f is the film mass per unit area, m q is the mass of the quartz crystal, f is the film density (1,100 kg/m 3 ) (26), GЈ and GЉ are the storage and loss moduli, respectively, and n is the overtone number (14).