Antibacterial isoamphipathic oligomers highlight the importance of multimeric lipid aggregation for antibacterial potency

Cationic charge and hydrophobicity have long been understood to drive the potency and selectivity of antimicrobial peptides (AMPs). However, these properties alone struggle to guide broad success in vivo, where AMPs must differentiate bacterial and mammalian cells, while avoiding complex barriers. New parameters describing the biophysical processes of membrane disruption could provide new opportunities for antimicrobial optimization. In this work, we utilize oligothioetheramides (oligoTEAs) to explore the membrane-targeting mechanism of oligomers, which have the same cationic charge and hydrophobicity, yet show a unique ~ 10-fold difference in antibacterial potency. Solution-phase characterization reveals little difference in structure and dynamics. However, fluorescence microscopy of oligomer-treated Staphylococcus aureus mimetic membranes shows multimeric lipid aggregation that correlates with biological activity and helps establish a framework for the kinetic mechanism of action. Surface plasmon resonance supports the kinetic framework and supports lipid aggregation as a driver of antimicrobial function.

PBS subtracted x-ray scattering curves were reduced as described above. Signal from the oligomers decays broadly into the wide-angle regime reflecting their small size. (b) The radius of gyration was determined by extrapolating to infinite dilution (y-intercept) from the radii at individual concentrations determined by fitting within the Guinier regime (Table S1). Pair-wise distribution generated by GNOM 13 within BioXTAS RAW. 14 Dmax was calculated to be 2.8nm and 3.0nm for the Meta and Para, respectively, based on DATGNOM.
(right) Raw scattering data and the fit solution used in the reconstructed pair-wise distribution. Even with the addition of neutral lipids, the S. aureus mimetic SUVs maintained a strong negative charge. TP and TO refer to the cardiolipin tail lengths explored. TPCL was used because it provided a better quality membrane after inspection by fluorescence microscopy.   There was little remaining signal, but these images were captured by maximizing the fluorescence excitation and using a long exposure time. The scratch was created while the bilayer was present (before removal) to find the focal plane. Microscopy slide was then treated with 5mM Triton-X to remove the R18-labeled lipid bilayer. (left) The bilayer removal was verified visually (right) and also by the lack of a photo-bleached spot by the FRAP laser (100ms exposure, which was located in the black crosshairs). Calculation to estimate the aggregate area percentage of the FRAP laser spot. At this concentration ( Figure S16), approximately 1500 aggregates were counted in the full viewing area of 450 x 335 µm, meaning the aggregate density is on average 1e-2 aggregates per µm 2 . Thus in the ~315µm 2 FRAP spot, ~3 aggregates will be present with variation. In this case, five aggregates were photo-bleached. Assuming each aggregate averages approximately 2 µm 2 in area, the average area of aggregates in the FRAP laser spot would be a total of 6 µm 2 , which is approximately 2% of the FRAP spot area. For the maximal aggregate count observed in this study of ~3000 aggregates (at 10uM), that would double this estimation such that the aggregates would be approximately 4% of the FRAP laser spot size.     In assuming irreversible aggregation (OL*) to be responsible for the elevated baseline during dissociation, the data remaining from that assumption shows a curve that would be challenging for fitting with the two state model with loss on the second step. In particulate, the curvature of downward slope (see arrow), would be difficult to fit.  does not continue increasing as the oligomer concentration decreases. Physically, it is hard to interpret K1 because it is not simply "binding," as it does encompass several physical states as understood in literature (e.g. binding, tighter binding, insertion, molecular aggregation) 30 . Therefore, this data shows there is perhaps a limit to the cooperativity of this process (i.e. the equilibrium decreases as the concentration increases). These rate constants show behavior where the parameter mostly trends with concentration and do not show the expected behavior of a differential at low concentration and similarities at high concentration. Parameters k5 and k6 are interesting because they indicate that the losses incurred by the membrane are functions of concentration. These losses may be representations of complex states that are not properly described by a single parameter, which could be cooperative or competing (aggregates, micelle formation, etc).

Compilation of data on lipid composition of Staphylococcus aureus.
Several articles were gathered on Staphylococcus aureus to understand the lipid composition in growth conditions similar to MIC assay (exponential growth). This includes understanding the amount of neutral lipids present in the membrane, effects of growth phase, and antibiotic resistance mechanism based on the lipid composition. Ultimately, the main goal is to determine a lipid composition to use in membrane mimetic studies. While bacterial membranes are anionic, a simple, fully-charged lipid bilayer would likely be mechanically frustrated from electrostatic repulsion and does not serve as a good mimetic.
Three reports quantified the neutral lipid (NL) component of S. aureus to be 19±7% of the membrane. Five other reports mentioned/showed qualitative support of the neutral lipid fraction but did not quantify it. Phosphatidylglycerol (PG) is a major headgroup component in the S. aureus membrane with eleven reports quantifying it at an average of 57±5% Cardiolipin (CL) was also quantified in similar reporting to be 16±4% of the membrane. Lastly, lysylphosphatidylglycerol (lysyl-PG or PG) was reported at 16±3% of the S. aureus membrane. LPG has an attached lysine via the carboxylic acid, meaning there are two free amines on the structure. The predicted pKa (Marvin Sketch 16.5.23) of these amines is 7.4 (amino group) and 10.21 (functional group), meaning at physiological pH, it has a net +0.5 charge. On a surface, there are additional effects on these pKa numbers and molecular association. LPG has been shown to impart resistance to cationic antimicrobials, including nisin and human defensins in S. aureus. 15 However, LPG is expensive (~$200/mg) and its use was prohibitive for study in the scope of this work, which seeks only to understand the interaction of antibacterial oligoTEAs and an S. aureus mimetic membrane. LPG within the native S. aureus membrane should attenuate the activity of any cationic membrane disrupting agent, but likely does not broadly change the mechanics of interaction.

Supplementary Methods.
Most all tabulated data was processed with GraphPad Prism 7.01. (1x) PBS was pH 6.8 unless otherwise stated. NMR spectra were recorded on an INOVA 600 MHz spectrometer and analyzed by MestReNova (version 10.0.0). 1 H NMR chemical shifts are reported in units of ppm relative to the deuterated solvent.
LCMS experiments were carried out on an Agilent 1100 LCMS system with a Poreshell 120 EC-C18 (3.0x100mm, 2.7um) column monitoring at 210nm in positive mode for detection. Solvents for LCMS were water with 0.1% acetic acid (solvent A) and acetonitrile with 0.1% acetic acid (solvent B). A flow rate of 0.6 mL/min was used with a gradient starting at 5% solvent B, followed by a linear gradient of 5% to 100% solvent B over 10 min, 100% solvent B for 2 min, before returning to 5% solvent B until equilibrated (3 min).
HPLC purification was performed on an 1100 Series Agilent HPLC system using a reverse phase Agilent Eclipse XDB-C18 column (9.4x250mm, 5µm) at 40°C and collected using an automated fraction collector. Solvents for HPLC were water with 0.1% trifluoroacetic acid (solvent A) and acetonitrile with 0.1% trifluoroacetic acid (solvent B). Compounds were eluted with either a 3.17 %B/min gradient or a 1.25 %B/min gradient at 4mL/min.
Pure and chemically confirmed oligoTEAs were quantified by being lyophilized into a tared microcentrifuge tube. Single stock solutions were prepared at either 1 or 5 mM in ultrapure water and used for all experiments.
Statistical analysis was performed on specific data sets as described in the main text.
Means are defined as = Additional methods for oligothioetheramide assembly. General method for fluorous allyl amine synthesis: 2-[2-(1H,1H,2H,2H-Perfluoro-9-methyldecyl)isopropoxycarbonyloxyimino]-2phenylacetonitrile (fluorous tag) was dissolved in THF (15mg/mL). Two equivalents of allyl amine and two equivalents of triethylamine were added and stirred at RT for at least 3 hours. The THF was completely removed and the reaction mixture was fluorous purified to yield fluorous allyl amine as a light yellow solid.

General method for thiol-Michael addition:
Two equivalents of the Boc-amine ethylene N-allyl-N-acrylamide was activated by dimethylphenyl phosphine (DMPP, 5mol% of monomer) and added to the fluorous thiol (50-300mM in methanol or 1:1 methanol:acetone). The reaction completion was tracked by thiol detection via an assay with 2,2' dithiodipyridine (DTDP) as described. 1 Hemolysis Assay. 200 µL of red blood cells (Innovative Research Novi, MI) were washed twice with 500 µL PBS by centrifugation (5 min at 500g) and resuspended in PBS at 4 v/v%. OligoTEA solutions or controls were mixed 1:1 with the RBC solution in a v-bottom 96-well plate (100 µL total). The resulting mixture was incubated at 37°C for 1 h and then centrifuged (Beckman Coulter, 5 min at 1000g) at 4 °C. A total of 75 µL of supernatant was transferred to a new plate. Hemolysis was measured via absorbance of released hemoglobin at 540 nm on a TECAN Infinite M1000 PRO Microplate reader (Männdorf, Switzerland) and normalized to 0.1% Triton-X (100%) or PBS buffer (0%). All experiments were performed in duplicate.

Spin-labeling the oligomer scaffold.
With the use of several primary amine groups in this scaffold, an amine protecting strategy orthogonal to base (synthesis of monomer), nucleophile (phosphine of thiol-Michael), UV of thiolene, and acid (TFA, fluorous tag deprotection) had to be utilized to enable spin labeling. The phthalimide-protected amine N-allyl-N-acrylamide was thus developed ( Figure S3). Generally, monomers are produced by alkylation (allyl group) and then acylation (acrylamide) or vice versa starting with either functional amine or alkyl halide. However, insufficient yield was obtained from the alkylation and then acylation of a phthalimide protected ethylene diamine. Also, the phthalimide was unstable in the presence of NaH, needed to for the acylation-then-alkylation route. Deprotection of the N-Boc ethylene N-allyl-N-acrylamide with TFA and reprotection with phthalic anhydride was not feasible due to immediate polymerization.
Thus, the following scheme was developed ( Figure S3, shown again above): 1) alkylation of N-Boc ethylene bromide with 50 eqv of allyl amine 5eqv of K 2 CO 3 purified by filtration, 2) deprotection with 200 mM 1:1 TFA:DCM stirred at RT for 1 hour, 3) treatment with 1eqv phthalic anhydride 200 mM in acetic acid at 100C for 1 hour, dried, precipitated in water and dried, and 4) acylation as previously described 2 and flash purified, eluting at 60% ethyl acetate in hexanes. In Step 3, the selectivity of the phthalic anhydride to protect primary amines over secondary amines was critical, but was reported in literature before. 3,4 With the N-phthalimide ethylene N-allyl-N-acrylamide, assembly of the antibacterial oligoTEA was completed as previously described. 5 The standard TFA deprotection then yields two primary amines for conjugation with a Proxyl-NHS (Toronto Research Chemicals, North York, Canada) with conditions previously reported. 1 Full deprotection of the oligomer (4 phthalimide groups) was modest with hydrazine, even with heat. Thus, a previously reported route of reduction with sodium borohydride and hydrolysis with acid 6 was optimized to utilize a stronger acid and less heat ( Figure S4, shown again below). The oligomer was purified by HPLC and verified by LCMS ( Figure S34,S35). The paramagnetic label prevents against full 1 H NMR of the compound.

Use of VT-PFG NMR, DEER EPR, and Stokes-Einstein-Sutherland Equation.
This method to measure the size and aspect ratios of flexible macromolecules has previously been discussed extensively 1 and makes use of work from Stokes-Einstein-Sutherland (SES), Chen and Chen 7 , and G. de la Torre 8 , and F. Perrin 9-11 with a review by A. Macchioni 12 .
The SES equation above relates the Boltzmann temperature and the translational molecular diffusion where k B is the Boltzmann constant, T is the temperature in Kelvin, r H is the prospective hydrodynamic radius, η is the dynamic viscosity, c is a size-dependent modification to transition between the slip/no-slip boundary conditions, 7 p is the geometrically defined aspect ratio, and f s is the shape-modified friction factor. [9][10][11] By measuring the diffusion by variable temperature PFG-NMR, the two remaining unknowns in this equation of hydrodynamic radius and shape are constrained. This is especially sensitive with the slope of the normalized temperature (T/η) is used. Then, the end-to-end distance can serve as a prospective diameter and allow the equation to be solvable, revealing the hydrodynamic radius and aspect ratio given a geometric shape, either an ellipsoid 9-11 or rod. 8 For this analysis, a prolate ellipsoid was assumed and the end-to-end distance was assumed to describe the longer dimension of the ellipsoid. dioleoyl-sn-glycero-3-phospho]-sn-glycerol (18:1 cardiolipin, TOCL) were purchased from Avanti Polar Lipids, Inc. Cholesterol (≥99%) was purchased from Sigma Aldrich. SUVs were prepared similar to the Avanti Polar Lipids protocol. Lipids were combined from chloroform stock solutions and dried using dry nitrogen at room temperature and dried for at least 6 hours at ≤10 mtorr. The dried lipids were suspended in 1x phosphate buffered saline (PBS) by vigorous vortexing for 5 minutes. Light sonication was needed for any lipid mixture with TPCL. The suspension was gently mixed overnight to age. The suspension was then sonicated for 10-30 minutes in a water bath (RT) using a QSonica Q125 probe-tip sonicator (20kHz) at 50% power, pulsing for 7 seconds with a 2 second rest. The resulting translucent suspension was then centrifuged at 17k RCF for 7 minutes to remove any metal particles. The solution was then filtered through a 0.2 µm PES filter. Dynamic light scattering (DLS) was then completed on a Malvern Zetasizer XS determining the hydrodynamic size to be 20-80 nm ( Figure S8a). Particle concentrations such that the attenuator was automatically set at 7 or 8, preset parameters of a liposome (RI 1.34) dispersed in PBS (0.9103 cP at 25C) were used. Zeta potential measurement was completed at the same particle concentration ( Figure S7b). With the exception of TPCL lipid mixes (1 week), SUVs were stored at 4C and seen to be stable for approximately 3-4 weeks.

Particle analysis in ImageJ to quantify number and size of aggregates.
Thresholding was performed to select the brightest pixel populations using the "Triangle" method. 29 While ImageJ was set to analyze particles from 0.0-10 µm, the smallest aggregates of ~0.1 µm were at the resolution and threshold limit, limiting the accuracy of their quantification in number and size by generally undercounting. Aggregates 0.2 µm and greater were reliably counted. The upper limit of particle analysis was 10 µm, with most oligomer induced aggregates ranging from 0.2-7.5 µm for the Meta and 0.2-5µm for the Para. For equilibrium and kinetic counts, the total count of particles was subtracted from the number initial defects (e.g. spots) present on the membrane to determine the number of aggregates that formed during oligomer exposure.

Surface plasmon resonance (SPR).
SPR was completed using a Biacore 3000 with an L1 Chip at 25C modified slightly from manufacturer protocol. Before use, desorb, sanitize, and an overnight wash with ultrapure water were completed. 1x pH 6.8 PBS was used throughout in all runs and solutions. Conditioning of the L1 chip was completed at the beginning and end of each run with 7uL 40mM octyl-βglucopyranoside at 10uL/min (Alfa Aesar). Additional manufacturer recommended washes were used. SUV capture was performed at 5 uL/min for 10 minutes to ensure surface saturation. Control runs of 1 minute pulses of 10mM NaOH showed little change in the baseline and ensured minimal formation of multilayer membranes. At an equilibrated flow of 30 uL/min, samples were injected (kinject) for 5 minutes and dissociated for 6 minutes with PBS.
Equations S1 | Ordinary differential equations from kinetic framework. (top) Excerpt of Figure 5a of translated to mass action kinetics rates (Equations 1-5). SPR observes effective refractive index changes near the gold sensor surface, encoding both mass and structural changes. Thus, the SPR cannot observe material lost from the surface seen in Equation 3 and 4 and Equation 5 sums up the combined observations.

Use of MATLAB to model and fit SPR data.
Fitting of the kinetic rates to the SPR sensorgram data was completed in MATLAB R2017a (9.2.0.538062) using lsqcurvefit. The function was two separate nested ODEs, one each for the association and dissociation phases where C oligoTEA was the designated concentration and 0uM, respectively. The ODEs were numerically solved by ode15s with RelTol and AbsTol as 1e-8. Convergence of lsqcurvefit was seen typically in less than 1k function iterations to a resnorm less than 1e6 to capture the behavior of the curve. Kinetic rate parameters k 1 ,k 2 ,k 3 ,k 4 , and k 5 were not bounded, while C Lipid,total was kept constant (3000 ± <0.01%). C Lipid,total should be dependent on the physical number of binding spots of the lipid surface, only changing based on SUV lipid composition, not the oligomer concentration (only the S. aureus lipid composition was fit). C Lipid,total must be greater than the highest RU value in the experimental data. For example, C Lipid,total must be at least ~1650 RU worth of lipid in order to fit the 10µM Para from Figure S19. C Lipid,total scales the curve response, with little change to the curve shape. Since the kinetic rates significantly affect curve shape, C Lipid,total was held constant to allow comparison of kinetic rates.

Supplementary Software. MATLAB Code.
The code is broken into a script that calls a custom function. Within the function are the ODEs for association and dissociation ([Oligomer] = 0) are called by the script to be numerically integrated in a sequential manner.