All peptides were purchased as lyophilized powder and certified >95% purity from GL Biochem (Shanghai) Ltd., China. Monobasic sodium phosphate (NaH₂PO₄), dibasic sodium phosphate (Na₂HPO₄), tribasic sodium phosphate (Na₃PO₄), sodium chloride (NaCl), Corning^® black 384 transparent-bottom plates were purchased from Sigma-Aldrich (Singapore). Deuterated phospholipids and deuterated cholesterol were synthetized by ANSTO’s National Deuteration Facility. Isopropyl alcohol, cholesterol and non-deuterated phospholipids used in QCM-D and SLB experiments were purchased from Sigma Aldrich (Singapore), QCM-D sensors were acquired from Biolin Scientific AB (Gothenburg, Sweden). Silicon wafers for NR were purchased from El-Cat Inc. (NJ, United States).

Stock solutions of HBpep peptide were prepared by solubilizing the lyophilized peptide directly in 10 mM acetic acid to achieve a final peptide concentration of 10 mg/mL. At this low pH (∼3), the peptides remain largely monomeric and do not coacervate. Sodium phosphate buffer solutions (10 mM, 154 mM NaCl final concentration) with pH ranging between 2-11 were prepared by mixing appropriate volumes of NaH₂PO₄, Na₂HPO₄, and Na₃PO₄ stock solutions to achieve the desired pH, which were later confirmed using a pH meter (Mettler Toledo, Singapore) before and after the addition of peptide stock solutions. For determining the concentration range within which HBpep coacervate, 11 different samples were prepared in a 38-well transparent-bottom microtiter plate by diluting the peptide stock solution into sodium phosphate buffer (pH 7.4) such that the final peptide concentration varied between 0 and 1 mg/mL in 0.1 mg/mL increments and a final volume of 100 µL. The volume of 10 mM acetic acid was kept constant in all the above samples. Immediately after the addition of the peptide, the absorbance of the samples were collected between 400 and 800 nm using a Tecan Spark (Männedorf, Switzerland) multimode plate reader. The mean of the absorbance values at 600 nm (A₆₀₀) for each sample, collected from three separate experiments, was plotted against the corresponding peptide concentration and fitted to a dose response curve using OriginPro 2021. Error bars indicate standard error of the mean. For characterizing the pH range within which HBpep coacervate, samples were prepared in a 38-well transparent-bottom microtiter plates by diluting the peptide stock solution into sodium phosphate buffer (10 mM, 154 mM NaCl) of varying pH such that the final pH of the sample varied between 2-11 in small increments and a final volume of 100 µL. Absorbance of the samples was collected and the data plotted as described above.

All DLS measurements were carried out in a Malvern Zetasizer Nano ZS (Malvern Panalytical Ltd., Singapore). HBpep coacervates were prepared in a quartz cuvette (10 mm pathlength) by diluting the peptide stock solution that was previously dissolved in 10 mM acetic acid, into sodium phosphate buffer (pH 6, 7.4 or 8) such that the final peptide concentration was 0.1, 0.25, 0.5 or 1 mg/mL. DLS measurements were conducted at a 173^o detection angle for 60 min at a fixed temperature of 25°C. Data was plotted using OriginPro 2021. Experiments were done in duplicates.

About 20 µL of the peptide solution of various (0.1–1 mg/mL) concentrations prepared in sodium phosphate buffer (10 mM, 154 mM NaCl, pH 7.4) was pipetted on to the cover glass of a MatTek dish (cover glass No. 1.5) and the sample imaged using a Carl Zeiss Axio Observer Z1–Inverted Microscope fitted with a motorized stage and an EC Plan-Neofluar 63x/1.25 oil, FWD 0.1 mm, CG 0.17 mm (DIC) objective. All images were collected using an AxioCam MRm CCD camera, 1,388 × 1,040 pixels, 6.45 µm/pixel. All images were acquired using Zen 2 (blue) software and processed using ImageJ2 (version 2.3.0/1.53f). Experiments were done in duplicates.

Measurements were performed using a Q-SENSE E4 system (Q-Sense, Sweden) connected to an Ismatec peristaltic pump model ISM596D. The 50 nm SiO₂ coated Au sensors were cleaned in 2 v/v% Hellmanex under bath sonication for 10 min, followed by extensive rinsing with MilliQ water and ethanol. The sensors were dried under N₂ flow and subjected to oxygen plasma cleaning for 2 min in a Harrick plasma cleaner (Model No. PDC-002, Harrick Plasma Inc. United States) fitted with a gas flow mixer (PLASMAFLO PDC-FMG, Harrick Plasma Inc. United States) for controlling oxygen flow rate. A lipid vesicle suspension was prepared at 0.2 mg/mL in MilliQ water (hydrodynamic diameter smaller than 50 nm as confirmed by DLS, prepared by tip sonication in H₂O). The composition of the lipid vesicles was 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC), 1-palmitoyl-2-oleoyl-sn-glycero-3-[phospho-rac-(1-glycerol)] (POPG) and cholesterol (POPC:POPG:Cholesterol) at a 7:1:2 molar ratio. This composition was chosen to represent the overall charge of mammalian cells (Virtanen et al., 1998). The lipid films were prepared by mixing appropriate amounts of lipids from chloroform stocks and gently evaporating the solvent under N₂ flow and rotation, and finally subjecting to vacuum for at least 1 h.

Prior to experiments, the fundamental frequency (5 MHz) and five overtones (3rd, 5th, 7th, 9th, and 11th) were found and recorded in MilliQ water. The flow rate was set to 0.1 mL/min, and the temperature was controlled as specified. A baseline was obtained in MilliQ water prior to injection of the vesicle solution (the lipid vesicle suspension was mixed in a 1:1 volume ratio with 4 mM CaCl₂ just prior injection into the solid-liquid cells). A constant flow of vesicles was maintained until stable signals were obtained before rinsing with excess MilliQ water. The supported lipid bilayers (SLBs) gave signals of typically -24-25 Hz and 0.2–0.4 ppm in dissipation, in line with previous results (Clifton et al., 2020). Rinsing with 10 mM phosphate buffer enriched with 154 mM NaCl at pH 7.4 or pH 6 was performed. HBpep coacervates (0.25 mg/mL) were formed and incubated for 5 min prior to addition to the SLBs. QCM-D measurements were performed in duplicates. For measurements in which coacervate sedimentation was avoided, the instrument was set upside down as previously done for other systems (Lind et al., 2019).

The supported lipid bilayers (SLBs) were prepared on a glass coverslip in a microfluidic chip by the solvent-assisted method reported by Ferhan et al. (2019) with slight modifications. To clean and hydrophilize the glass surface, the coverslip was soaked overnight in a 1 v/v% solution of Hellmanex, washed with MilliQ water and ethanol, dried in an oven then subjected to oxygen plasma cleaning for 2 min in a Harrick plasma cleaner (Model No. PDC-002, Harrick Plasma Inc. United States) fitted with a gas flow mixer (PLASMAFLO PDC-FMG, Harrick Plasma Inc. United States) for controlling oxygen flow rate. Immediately after this, the coverslip was attached to a multichannel microfluidic chip (Ibidi sticky-Slide VI0.4), and all microfluidic channels were filled with MilliQ water. The chip has six microfluidic channels, each with a length of 17 mm, a width of 3.8 mm and a height of 400 µm.

To prepare the lipid solution, POPC, cholesterol and two fluorescent markers, namely, 18:1 Liss Rhod PE and 25-NBD cholesterol (that selectively mix with POPC and cholesterol, respectively), were solubilized in isopropanol, while POPG was solubilized in ethanol. Appropriate amounts of each of those five components were mixed and diluted in water:isopropanol with 1:1 volume ratio to create a 0.4 mg/mL POPC:POPG:Cholesterol solution with a 7:1:2 molar fraction. Liss Rhod PE and NBD cholesterol were used at a concentration of 3.4 μg/mL and 1.8 μg/mL, respectively.

To create a SLB, an Ismatec peristaltic pump model ISM596D was used. First, water:isopropanol with a 1:1 volume ratio was pumped for several minutes to ensure complete exchange of the MilliQ water that previously filled the channel. Then, the lipid solution was pumped for 10 min at 50 μL/min, the flow was stopped for 5 min, and MilliQ water was pumped, also for 10 min at 50 μL/min. The resulting SLB homogeneously and completely covered the glass substrate in the microfluidic channel.

The coacervate experiment was then carried out just as in the QCM-D, the only difference being the reduced flow rate of 50 μL/min. Images were collected as the experiment was ongoing in a Carl Zeiss Axio Observer Z1 inverted fluorescence microscope fitted with a motorized stage and a Plan-Apochromat 63x/1.40 oil DIC, FWD 0.19 mm, CG 0.17 objective. All images were acquired using a Carl Zeiss AxioCam 503 Mono camera, 1,936 × 1,460 pixels, 72 nm/pixel using Zen 2 (blue) software and processed using ImageJ2 (version 2.9.0/1.54f).

Experiments were performed on the reflectometer SPATZ at the Australian Nuclear Science and Technology Organisation (ANSTO) using a setup for solid/liquid interfaces (Le Brun et al., 2023). The momentum transfer range and momentum transfer (Q) resolution used was 0.011 Å⁻¹ < Q < 0.273 Å⁻¹ and ΔQ/Q = 5.8%, respectively. The instrument uses the time-of-flight principle and a disc chopper pairing of discs 1 and 2 set 480 mm apart running at 25 Hz was used (2.5 Å < λ < 20 Å). Collimation slits used were set to 2.66 mm and 0.75 mm for an angle of incidence of 0.85° and 10.94 mm and 3.09 mm for 3.5°, providing an illuminated footprint of 55 mm along the beam. Data was reduced with the refnx software package (Nelson and Prescott, 2019) where the time-of-flight is converted to wavelength which is used to calculate Q, accounts for detector efficiency, re-bins the data to instrument resolution, subtracts the background, stiches the data from the two angles of incidence together at the appropriate overlap region, and scales the data so that the reflectivity at the critical edge is equal to 1.

The silicon blocks (100) of 100 mm diameter and 10 mm thickness and O-rings (TEFLON of 100 µm thickness) were cleaned in the following series of solvents: 2% Hellmanex for at least 1 h followed by extensive rinsing with MilliQ water. Before assembling the cells, the blocks were dried with a flow of nitrogen gas and exposed for 30 min to ozone to finally be assembled in dry conditions. During the assembling, the substrate was sealed with the polished surface in contact with the liquid using a thin Teflon O-ring, and the cells were connected to an HPLC pump to allow the exchange of solvent with D₂O, H₂O, and CMSi (38:62 v/v D₂O: H₂O).

The solid/liquid cells were pre-equilibrated to 25°C using a circulating water bath and the surfaces were characterized in H₂O and D₂O to determine the roughness and the thickness of the silicon oxide layer. Then, the SLBs were formed using vesicle fusion as described in Section 2.6, and characterized in three isotropic contrasts composed of H₂O based-, D₂O based, and CMSi-based (38:62 D₂O: H₂O (v/v)) buffer. The composition of the lipid vesicles was POPC-d₆₄:DOPG-d₆₆:Cholesterol-d₄₅ at a 7:1:2 molar ratio. Here, d₆₄:DOPG was used since synthesis of tail deuterated POPG has not been reported or performed to the date of publication. The lipid films were prepared by mixing appropriate amounts of lipids from chloroform stocks and gently evaporating the solvent under N₂ flow and rotation, and finally subjecting to vacuum for at least 1 h.

The SLBs were fitted using a 4-layer model constituted by the oxide layer and three layers that correspond to the heads-tails-heads of the lipid membrane. During the data fitting, the lipid bilayer was constrained to be symmetric, that is, the coverage, the thickness, and the scattering length density (SLD) of the heads were the same for both leaflets. Additionally, the roughness was constrained to be the same across the whole bilayer. The HBpep peptide was slowly introduced via a syringe at concentrations of 0.25 mg/mL in the corresponding buffer, upon re-equilibration of ∼5 min. After roughly 1 h 20 min of incubation, the samples were rinsed with H₂O based buffer and re-characterized in all three contrasts by using 1/10 of the final peptide concentration. Data was collected prior and after buffer exchange in H₂O based buffer and no changes in reflectivity were observed.

To fit the data in the three isotropic contrasts after protein incubation and rinsing, the head-group thickness and coverage was kept constant but the thickness and SLD of the tails were allowed to vary. A layer of protein was added on top of the SLB. The experiments were performed only once due to the restricted availability of beam time. Data fitting was performed using the MOTOFIT package (Nelson, 2006) and to determine the errors of the fits, a Monte Carlo error analysis was performed using genetic optimization.

The structure of the peptide was built using the PEP-FOLD4 server (Rey et al., 2023; Tufféry and Derreumaux, 2023), which has the advantage over other available servers that it embeds a Debye-Hückel formalism to treat pH conditions and salt concentration variations. Peptide conformations are pH- and salt concentration-dependent, and thus, two systems were considered to mimic neutral and acidic conditions. The main input was the sequence in FASTA format (GHGVY GHGVY GHGPY GHGPY GHGLYW). The forward-backtrack sampling algorithm was selected to sample the structural alphabet profile and the number of models generated was 100. Default parameters related to the Monte Carlo protocol (10,000 steps) and the pseudo-random generator seed were adopted to generate the 3D models. The Debye-Hückel contribution was switched on, and the pH was specified as well as an ionic strength of 154 mM NaCl to match the experimental conditions. The use of the zwitterionic forms of the peptides and the default pKa values of 6 for His residues were used.

A membrane containing POPC, POPG and cholesterol molecules in a proportion 70:10:20 (POPC:POPG:Cholesterol) was generated using CHARMM-GUI Membrane Builder (Jo et al., 2008; Jo et al., 2009; Wu et al., 2014). The membrane contained 210 POPC molecules, 30 POPG molecules and 60 cholesterol molecules in the mixed bilayer. Subsequently, the multicomponent assembler (Kern et al., 2023) of CHARMM-GUI facilitated to build two all-atom molecular models at neutral or acidic pH, using the first 15 peptides rendered by PEP-FOLD4 (the more favorable energetically), and the previously built complex membrane. The combined system (bilayer and peptides) was then solvated to produce a rectangular box of dimensions (130 × 130 × 100) ų. Ions were added resulting in a final concentration of 154 mM NaCl (Humphrey et al., 1996). The final system contained over 168,000 atoms. CHARMM22 (Li et al., 2005) parameters (with CMAP correction) were used for the peptides, CHARMM36 for lipids (Klauda et al., 2010), standard parameters for ions (Noskov et al., 2004) and the TIP3P (Jorgensen et al., 1983) model for water. NAMD 3.0alpha GPU (Phillips et al., 2005) and Gromacs 2023 (Berendsen et al., 1995) were employed to calculate the dynamics of the systems throughout (Phillips et al., 2005). Although we started using NAMD, due to performance, we switched to Gromacs. One simulation of the neutral system and one of the charged system were obtained with NAMD, and five replicas per pH were obtained with Gromacs. The total cumulative production time is 12*0.2 μs (2.4 µs).

Initially, 10,000 steps of minimization were performed to remove steric clashes, followed by the progressive removal of constraints at 500 ps intervals to allow for gradual equilibration of the system. Constraints were released in the following order: i) bulk water and lipid tails; ii) lipid head groups; iii) peptide side chains. Unrestrained dynamics was then undertaken in the NPT ensemble. The Particle Mesh Ewald method was used for the treatment of full-system periodic electrostatic interactions; interactions were evaluated every second timestep with a value of 1 Å to determine grid spacing (Darden et al., 1993). Electrostatic and van der Waals forces were calculated every timestep and up to a cutoff distance of 12 Å. A switching distance of 10 Å was chosen to smoothly truncate the non-bonded interactions. Only atoms in a Verlet pair list within a cutoff distance of 13.5 Å were considered, with the list reassigned of reassigned every 20 steps (Verlet, 1967). The SETTLE algorithm was used to constrain all bonds involving hydrogen atoms to allow the use of a 2 fs timestep (Miyamoto and Kollman, 1992). The Nose-Hoover-Langevin piston method was employed to control the pressure with a 200 fs period, 50 fs damping constant and a desired value of 1 atmosphere (Martyna et al., 1994; Feller et al., 1995). The system was coupled to a Nose-Hoover thermostat to sustain a temperature of 298 K throughout.

HBpep is a 26-residue peptide inspired by the Histidine-rich beak proteins (HBPs) of the Humboldt squid (Tan et al., 2015). The peptide sequence (GHGVY GHGVY GHGPY GHGPY GHGLY W) consists of five GHGxY repeats, [where x is either valine (V), proline (P) or leucine (L)], and a C-terminal tryptophan residue, both of which are important drivers of phase separation (Figure 1A) (Wang et al., 2018; Gabryelczyk et al., 2019). HBpep phase separates under physiological conditions into µm-sized coacervate microdroplets that exhibit an extremely high efficiency (>99%) to recruit a wide variety of therapeutics within their dense phase (Lim et al., 2018; Lim et al., 2020). Consequently, HBpep and its variant were developed into an intracellular delivery system capable of delivering a wide variety of therapeutics into different cell types in vitro with little to no cytotoxicity (Sun et al., 2022).

In order to identify the optimal experimental conditions for investigating HBpep-coacervate and model cell membrane interactions, turbidity measurements were carried out with HBpep at different conditions of pH and concentrations (Figures 1A, C). In general, coacervation of phase-separating peptides and proteins can occur over a wide pH range and can be monitored by measuring sample turbidity (Supplementary Figure S1), coacervate size (Supplementary Figure S2), or both. However, turbidity measurements—the standard practice in the field—allow for the determination of “phase-separation (a.k.a coacervation) propensity” when the coacervate size remains more or less similar across a pH range. Turbidity measurements of HBpep prepared in sodium phosphate buffers spanning a pH range of 2–11, indicated that the highest turbidity was observed at pH 7.4 which is close to the peptide’s theoretical pI of 7.96 (Figure 1A). In acidic conditions and at pH values above 9, HBpep displayed minimal turbidity indicating its inability to coacervate and likely remained monomeric or resulted in low-level oligomerization. This pH-dependent behavior is due to the positive charges arising from the five histidine side groups when the pH is below its pK _a2 of 6 (Saurabh et al., 2022), and to the negative charges arising from the tyrosine side groups when the pH is above its pK _a2 of ∼10 (Figure 1B), which result in charge-charge repulsion at these pH ranges. Since HBpep is acetyl-capped at the N-terminus and amidated on the C-terminus, we do not expect any end terminal charge contribution.

Similarly, a concentration-dependent behavior was observed when HBpep stock solutions were diluted into a buffer maintained at pH 7.4 (Figure 1C). An increase in turbidity (OD₆₀₀) correlated with an increase in HBpep concentration, with a detection threshold at 0.2 mg/mL and a peak turbidity signal at 0.6 mg/mL, beyond which the signal plateaued. At pH 6, where the peptide carries a net positive charge, DLS measurements suggest an oligomeric state for the peptide at 1 mg/mL (the oligomeric formation was confirmed by SANS data, data not shown) (Figure 1D). At pH 7.4, HBpep coacervate droplet sizes varied from 400–800 nm (Figure 1D), with size being both time- and concentration-dependent (Figure 1E). Micrometer-sized droplets were observed within 5 min for peptides at a concentration of 1 mg/mL, whereas sub-μm size particles were observed at a lower peptide concentration of 0.25 mg/mL (Figure 1D). Furthermore, μm size droplets sedimented over time, and the coalescing property of the coacervates became more discernible at higher peptide concentrations as seen from DIC microscopy images (Figures 1F–I). Based on the above results, a pH of 7.4 and 6 were chosen as the coacervating and non-coacervating pHs for the peptide, respectively. Since high peptide concentrations (>0.5 mg/mL) led to coacervate coalescence and a concomitant increase in droplet size, we determined 0.25 mg/mL as an appropriate peptide concentration to use for all further experiments as we expect the coacervate size to remain relatively stable under these conditions.

The interactions between HBpep coacervates and model cell membranes were followed via QCM-D on SLBs formed by vesicle fusion using vesicles composed by POPC:POPG:Cholesterol in a 7:1:2 molar ratio as a model of the charge in mammalian cell membranes (Virtanen et al., 1998). HBpep coacervates (0.25 mg/mL) were prepared and incubated for 5 min prior to addition to the SLBs. Analysis of the raw QCM-D data (Figures 2A, B) confirmed HBpep binding at pH 6, a condition where no significant coacervation occurs (Figures 1A, D). The observed QCM-D signal at acidic conditions closely resembles that of cell penetrating peptides such as penetratin (Hedegaard et al., 2018). Upon rinsing with buffer, partial desorption of HBpep occurred, which is also consistent with data observed for cell penetrating peptides (Hedegaard et al., 2018) (Figure 2B). In contrast, at pH 7.4, substantial adsorption of HBpep coacervates on the SLB coated SiO₂ sensor was observed (Figure 2A). The resulting QCM-D signal resembles the adsorption of lipid vesicles on SiO₂ and their eventual rupture, as evidenced by the sharp decrease in frequency and the increase in dissipation with time, followed by an inversion of the signals (increase in frequency and decrease in dissipation). Thus, it is likely that some of the μm-sized coacervates spread out on the SLB-coated sensor, mirroring what was observed by microscopy on glass slides (Figures 1H, I). We note that most of the adsorbed coacervates could be removed upon rinsing with buffer at pH 7.4 suggesting that the adsorption was largely reversible and could be due to simple droplet sedimentation.

In order to decouple the effect of molecular adsorption on the SLB and sedimentation, the QCM-D E4 unit was flipped upside down and the experiment at pH 7.4 was repeated (Figure 2C). In this case, a signal mirroring that measured at pH 6 was detected with full reversibility upon rinsing with a solution containing 1/10 of the original coacervate solution (i.e., 0.025 mg/mL) followed by excess buffer. Taken together, both peptide and coacervate adsorption takes place regardless of pH, but the adsorption is fully reversible only at neutral pH. The latter suggests that no major rearrangement of the lipid bilayer structure takes place due to peptide adsorption at pH 7.4. This was independently verified with SLBs prepared using a solvent-assisted method on glass slides that were imaged using DIC microscopy (Figures 2D–F). The experimental conditions were kept consistent with those used in the QCM-D experiments depicted in Figures 2A–C.

The DIC images reveal that at pH 7.4 in the upward position of the chamber (Figure 2D), more μm-sized coacervates adsorbed to the SLB compared to when the microfluidic chamber was set-up upside down (Figure 2F). At pH 6, no coacervates were observed (Figure 2E) as expected based on turbidity and DLS measurements (Figures 1A, D). A manual count (Supplementary Figures S3, S4) of the number of coacervates in each of the representative DIC images in Figures 2D–F revealed the count to be 228 for pH 7.4, 0 for pH 6 and 30 for upside down at pH 7.4. Since at pH 6 HBpep does not significantly coacervate into large droplets, the black dots observed in Figure 2E do not correspond to coacervates and are most likely pH dependent salt crystallization, or low-level oligomerization below the resolution limit of the microscope objective used.

Neutron reflection data were collected at three isotropic contrasts before and after exposure to 0.25 mg/mL HBpep at pH 6 (Figure 3) or pH 7.4 (Figure 4), and the data were fitted assuming a symmetric lipid bilayer structure (lipid heads–lipid tails–lipid heads). The thickness of the core lipid bilayer, the head groups and the roughness at pHs 6 and 7.4 are summarized in Table 1. The coverage was in both cases above 90%. The values in Table 1 compare well to those reported by Waldie et al (Waldie et al., 2018; Waldie et al., 2019) for which 30.9-33.5 Å thick core was measured for 10–40 mol% cholesterol in various phosphatidylcholine SLBs including POPC. The head group region was found to be 7-8 Å thick for PC-cholesterol (Waldie et al., 2018; Waldie et al., 2019). The larger headgroups [12.9 and 13.3 (±0.3) Å at pH 6 and pH 7.4, respectively] and headgroup hydration is thus expected from the presence of charged phosphatidylglycerol groups (Le Brun et al., 2014).

At pH 7.4 (where HBpep coacervation takes place), there were minimal changes in reflectivity (Figure 4). The small changes observed at low Q are simply due to small differences in the D₂O content in the buffer, typical when there is incomplete exchange upon flushing with buffer. There was though a very small increase in reflectivity mainly observable for the H₂O based buffer. These changes can be interpreted as an increase in the coverage or scattering length density (SLD) of the SLB. Indeed, upon co-refinement of the data using a four-layer model (lipid heads–tails–lipid heads–HBpep), a suitable fit was obtained assuming a constant layer thickness of the SLB but allowing for increase in the coverage of 7%. There was neither any significant peptide adsorption on top of the SLB, nor any significant drop in the SLD of the lipid bilayer region. During SLB formation by vesicle fusion, it is common that a small fraction of vesicles remain bound, these are hardly seen by NR (Lind et al., 2014). Together, these values suggest the rupture of a few vesicles that might have remained bound to the SLB after the rinsing step, and that there was neither significant peptide binding nor integration in the lipid bilayer, consistent with the QCM-D results (Figure 2C).

At pH 6, where HBpep remains mostly in the monomeric or oligomeric state (Figure 1E), significant changes in reflectivity were noticed at all contrasts besides the small changes observed at low Q for D₂O-based buffer due to incomplete exchange upon flushing with buffer. In particular, the cross over in the D₂O based buffer contrast suggest a change in the SLB structure/composition (Lind T. K. et al., 2015). Indeed, upon co-refinement of the data using a four-layer model (lipid heads–tails–lipid heads–HBpep), a suitable fit was obtained with a 1.4 Å increase in the lipid tail layer thicknesses, a lowering of 0.24 × 10⁻⁶ Å⁻² in the lipid tail layer SLD, and a coverage increase by 6%. Together, these values point towards the integration of peptide in the layer. This corresponds to 5 % of the layer being composed of peptide and 95% of lipids. There was some peptide adsorption on top of the SLB, although minimal.

NR indicates that at conditions in which the peptide is cationic and does not form coacervates (i.e., at low pH), the peptide intercalates in the lipid bilayer, partially removing some lipids, a finding in agreement with observations for cationic peptides with antimicrobial capacity (Lind T. et al., 2015; Nielsen et al., 2019; Hedegaard et al., 2020). For neutral pH, where the peptide is largely uncharged and mainly present in the coacervate phase, minimal interaction was detected with the SLB. Instead, a slight increase in the coverage of the SLB was observed without affecting the lipid core SLD. This suggests that eventual non-fused lipid vesicles collapsed due to the presence of coacervates in the bulk phase.

Figures 5A–C presents comparative snapshots of the SLB before and after 200 ns of simulation at both neutral and acidic conditions. At neutral conditions, the HBpeps predominantly form aggregates within the solution, consisting of clusters of 4–6 members, where they remain for the duration of the simulations. In acidic conditions, however, HBpeps display a distinct behavior: they land on the surface of the bilayer either as individual entities or as small aggregates comprised of 2, 3 or 4 peptides, subsequently penetrating into the membrane core. Note that oligomeric formation agrees well with the DLS data presented in Figure 1D at acidic pH. Even though larger aggregates are not observed at neutral pH, all peptides in the simulation box form aggregates while only a few do form oligomers at acidic pH. The primary interactions between the HBpeps and the bilayer is via His and the phosphate group of POPG or the aromatic ring of Tyr, Trp or His and the hydroxyl group of cholesterol, as depicted in Figure 5B. These observations indicate that charged lipids and cholesterol are crucial for facilitating the binding of HBpep to membranes, in agreement with findings from previous cell studies (Sun et al., 2023).