Surface Plasmon Resonance (SPR) characterization
The interactions between AMPs 1–5 and LUVs were first investigated by SPR as a benchmark for the MST measurements. The data was acquired using the same lipid compositions as used in the MST data acquisition. LUVs were immobilized on an L1 chip, and an increasing concentration of AMPs was injected over them. Figure 2 displays typical steady state fits of the SPR measurements, and the disassociation- and partitioning constants, KD and KP, extracted after 180 seconds are listed in Tables 3 and 4. The dissociation rate, koff, was calculated from the dissociation step, using the methodology presented by Figueira et al.21, summarized in Table 2.
Table 2
Summary of koff of AMPs 1–5 evaluated by SPR.
# | Peptide | koff PC (s− 1) | koff PC/PG (s− 1) |
1 | WWWRRR | 0.22 ± 0.02 | 0.19 ± 0.01 |
2 | WRWRWR | 0.87 ± 0.19 | 0.48 ± 0.05 |
3 | WWWKKK | 0.48 ± 0.07 | 0.32 ± 0.05 |
4 | WKWKWK | 0.90 ± 0.24 | 1.32 ± 0.05 |
5 | LWwNKr | 1.76 ± 0.12 | 1.75 ± 0.16 |
SPR showed values for KD, KP and koff in the range of 50–800 µM, 1-150 10− 3 and 0.2-1.0 s− 1 respectively. 1 was the strongest binding compound, followed by 3, 2 and 4, with inactive AMP 5 being the weakest. This trend was also preserved when anionic lipids were present, but as expected, the overall affinities of all compounds was increased. The koff values also followed this trend with the most active AMP, 1, having the slowest dissociation. The overall conclusion from SPR points towards increased affinities of the clustered peptides over alternating ones towards both zwitterionic and ionic lipid bilayers.
Vesicle MST response profiles
For KD evaluation by MST, FHot was registered during the temperature related intensity change (TRIC) phase of the MST trace (see Methods). The advantage of evaluating KD during the TRIC phase is that any potential effects of prolonged heating on the thermostability of the sample that may influence the binding can be avoided.35 The dose-response profile of MST response against lipid concentration for 1–5 largely follows a sigmoidal curve shape (Figs. 3A and 3B). In the cases of 5 and 4, the weaker binding resulted in truncation of the sigmoidal curve, with the asymptotic region at high lipid concentration not being fully sampled.
A blank series consisting of only lipids was collected for each lipid composition, and a weak MST response could be detected at the highest lipid concentrations despite the lack of a fluorophore. The blank response was stable up to concentrations around 1 mM. The two to three highest lipid concentrations measured did however build up a background response above the noise level of the baseline (Figs. 3C and 3D), most likely due to increasing turbidity. The added unspecific response to the last two data points thus imposes a limitation on how weak interactions can be reliably measured (mM lipid concentration range). The turbidity did however not notably impact the extraction of Fnorm in the presence of AMPs (carrying the fluorophore), as the much stronger signal of the fluorophore did not display any signs of inheriting the turbidity contribution to Fnorm from the blank profile (raw data shown in Figure S1). This suggests that W fluorescence was the dominant contribution to the measured response when it was present. It is however not possible to rule out that the light scattering effect may dominate the response for very weak binders. Peptide 5, for example, shows a change in Fnorm for the final two points together with DMPC liposomes, the same points that are significantly affected by light scattering effects in the blank measurement (Fig. 3A). The effect is however absent for the same peptide together with DMPC:PG liposomes even though the two blanks behave similarly, demonstrating that potential light scattering contributions are difficult to predict or compensate for.
Nanodisc MST response profiles
The MST data for two nanodisc types, SMA and SMA-QA, were evaluated using the same methodology as the vesicle data, with FHot taken during the TRIC, and Fnorm plotted against log[lipid] and fit to Eq. 6. The binding profiles of AMPs binding to SMA and SMA-QA nanodiscs show significantly different behaviours (Fig. 4).
The SMA-QA nanodiscs produce a sigmoidal-like response curve (Fig. 4D and 4E), as was observed for the vesicles, though the curve is right-shifted to higher lipid concentrations, i.e., AMPs bind more weakly to the SMA-QA nanodiscs than to the corresponding vesicles. The consequence of the right-shift is that the binding curves of the weaker binders are not fully sampled, and in some cases only a minimum KD can be determined. As with vesicles, at the highest lipid concentrations light scattering can become a factor in the measurement of Fnorm. The result is that it becomes impractical to fully sample the binding curve by further increasing the lipid concentration, as an increased influence of light scattering effects would be expected.
By contrast, the SMA-nanodiscs response is left-shifted to lower lipid concentrations, corresponding to a stronger binding than to the corresponding vesicles (Figs. 4A and 4B). The curves also deviate from the sigmoidal Fnorm profile with a secondary drop in Fnorm after reaching the maximum, in the mM range. Such profiles have previously been observed in the MST response of higher stoichiometric bindings where additional interacting ligands gave rise to a new species of the complex.22 In the case of nanodisc-AMP interactions, the most plausible explanation is that direct interactions between 1–5 and the SMA polymer gives rise to the atypical binding profile. A direct interaction between the free SMA polymer and the AMPs can indeed be observed in the high nM – low µM range in a control experiment without lipids present (Fig. 4C and 4G). The presence of strong interactions to the SMA polymer, and a multiple-phase response profile, indicates that the interaction with the polymer dominates the measurement to the extent that the AMP-lipid interaction cannot be directly measured in SMA nanodiscs (Table 3). The strong interaction to the SMA polymer can be attributed to the rich anionic maleic acid content (deprotonated at pH 7.4), resulting in favourable electrostatic interactions to the cationic AMPs. Consequently, no such binding is observed for the SMA-QA polymer, which is instead rich in cationic moieties, and thus has the attractive electrostatic interaction potential to cationic AMPs replaced by a repulsive potential, which is reflected in the response profiles being shifted towards weaker binding.
KD comparison
Comparison of the MST and SPR derived KD for LUVs shows that the absolute KD obtained are systematically offset by an approximate factor 4 (Fig. 5C-D and Table 3), but the relative values within each dataset result in the same stratification of the peptides binding strength to LUVs as measured by MST and SPR. The inactive peptide 5 had a considerably higher KD compared to 1–4, though the profile of 5 could not be fully sampled. Therefore, only the minimum KD was determined, providing an explanation as to why MST did not identify the same strong dependence on the lipid composition as SPR for this peptide. Consistent with SPR, the clustered sequence peptides 1 and 3 had significantly stronger binding than 4. This is also consistent with the determined MICs, with 1 and 3 being the most active peptides. These observations are in line with previous reports that clustering of W residues is positively correlated with antimicrobial activity.4
The SMA-QA derived KD compare favourably with the SPR results in the instances where the curve is adequately sampled (Fig. 5B, Table 3). However, as the KD approach the mM region, the SMA-QA data set are not sampled sufficiently to produce reliable results. One example of insufficient sampling is shown by the poor reproduction of the PC/PG discrimination observed in SPR for 2. This is however not the case for the more active AMPs where the binding strength is well inside the sampling range. SMA-QA nanodiscs are thus only viable for assessing AMP-lipid interactions stronger than the mM range by MST.
Table 3
Summary of KD determined using SPR and MST. Errors represent the standard deviation of the triplicates.
Peptide | SPR (KDµM) | Vesicle (KDµM) | SMA-QA (KDµM) | SMA (KDµM) |
DMPC | DMPC/PG | DMPC | DMPC/PG | DMPC | DMPC/PG | DMPC | DMPC/PG |
LWwNKr | 2548 ± 493 | 1033 ± 58 | > 670 | > 650 | - | > 1000 | - | - |
WKWKWK | 712 ± 27 | 474 ± 45 | 282 ± 58 | 112 ± 29 | > 713 | - | 8.6 ± 0.5 | 9.8 ± 0.3 |
WRWRWR | 318 ± 62 | 105 ± 7 | 73 ± 53 | 24 ± 7 | > 541 | > 615 | 2.7 ± 0.5 | 5.6 ± 1.1 |
WWWKKK | 302 ± 32 | 112 ± 15 | 28 ± 3 | 17 ± 13 | 267 ± 47 | 145 ± 55 | 1.1 ± 0.3 | 1.1 ± 0.5 |
WWWRRR | 142 ± 35 | 70 ± 1 | 21 ± 3 | 10 ± 5 | 142 ± 17 | 40 ± 5 | 1.2 ± 0.5 | 2.2 ± 0.4 |
The SMA discs exhibit overestimated binding strengths, showing all peptides to have apparent KD of 10 µM or lower (Table 3), and show poor discrimination between PC and PC/PG lipids (Fig. 5A). The lack of discrimination is attributed to the strong interaction with the anionic SMA polymer dominating the response and masking the expected binding enhancement from the addition of the anionic lipids.
In general, the presence of PG lipids leads to a decrease in KD for all peptides by an approximate factor of 2 when the sigmoidal binding curve could be fully sampled. In previously published work, Christiaens et al.36 found that individual peptides had a broad range of binding strengths, from a weak 350 µM towards PC vesicles, to stronger binding in the low µM-nM range to vesicles rich in anionic charges. Analysing MSP-nanodiscs by ITC and NMR, Zhang et al.37 observed binding to anionic lipid nanodiscs in the range of 1–2 µM. The KDs measured by MST are thus in line with results in the literature that shows that AMPs can bind in the low µM range to both vesicles and nanodiscs in the presence of anionic lipids, with an expected weakening of the interaction towards zwitterionic membranes. It is known that cationic AMPs have a selectivity towards bacterial membranes where anionic lipids and LPS are present on the outer membrane over cells with a more neutral surface. While the reduction in KD upon introduction of anionic lipids observed for 1–5 is not as large as described in the above studies, the anionic component introduced in this work (5%) is low in comparison to the 20% used by Zhang et al. and Christiaens et al.
The calculated KD in this work is fit using a two-state model.38 This is however not necessarily expected to be an accurate representation of lipid interactions, where the target has no defined binding site, and it is possible that self-aggregation on the lipid surface, saturation effects, cooperative- or competitive binding will occur that will influence the binding of further AMPs.39–40 For this reason, KD is a useful illustrative and communicable descriptor of AMP-lipid binding, but care would be advised to not overinterpret the absolute value. The measured apparent KD is best considered a composite value representing multiple processes of a complex interaction, that is system- and method dependent – as also highlighted in this work.
Fluorescence intensity and KP
The partition coefficient (KP) describes the preference of compounds for lipid- or aqueous phases, with a high KP indicating a greater preference for the lipid phase. KD in contrast describes the bound state as a molecular complex rather than phase. To explore the possibility to determine the KP of AMPs 1–5 using MST, the single wavelength fluorescence intensities measured by the MST instrument were used. KP was extracted by fitting the data to a hyperbolic partition curve described by Eq. 8, after removal of non-hyperbolic points.41
Lipid-only blanks were collected to assess the effect the turbidity had on the measurements (Fig. 6). For the vesicles, this is a modest signal that increases linearly in the mM lipid concentration range. The increase is consistent with the observed MST response reported above (Fig. 3C and 3D), attributed to light scattering effects. Both types of SMA nanodiscs similarly follow a mostly linear trend but the signals are more intense (Fig. 6).
The background fluorescence for the vesicles is modest at the initial concentrations compared to the measured fluorescence intensities of the AMP and does not have the same profile as lipid concentration is increased (Figure S2). A larger increase is observed for the highest lipid concentrations, showing the influence of the turbidity of the system (Fig. 6). The SMA-QA nanodiscs initial background fluorescence signal is significantly stronger, and a substantial level is maintained over the measured concentration range. As with the vesicles and SMA-nanodiscs, the intensity profile of the SMA-QA sample with AMP present does not match the profile when the AMP is present (Figure S2). For this reason, it is not possible to subtract the blank baseline from the AMP signal. Hence, the final points carry increased uncertainty because of potential, but inconsistent, light scattering contributions (Figs. 7 and S3).
Many AMPs, especially when using vesicles, showed a spike in fluorescence at low lipid concentrations (at high peptide:lipid ratio). This phenomenon has previously been described by Melo and Castanho,42 where they attribute the deviation to the saturation of the bilayer with AMP that prevents the uptake of additional AMPs. It has also been attributed to changes in conformation and peptide-peptide interactions within the bilayer when it is saturated with AMPs.43 To extract KP where deviations are observed, the deviant points are excluded from the fit. At lower lipid concentrations (high peptide:lipid ratio), a critical point is reached where the hyperbolic model is no longer followed and points beyond the critical point cannot be described by Eq. 8. It should be noted that the removal of points introduces uncertainty and increases the error of the extraction of KP; this is particularly problematic as the hyperbolic shape is best described by the initial points along the curve, but these are also the points that are the most affected by the high peptide:lipid ratios.41–42 For both SMA-QA and LUVs, many points deviated from the hyperbolic shape, and a large number of points needed to be removed, typically leaving the final 4–5 points for the final fit.
Table 4
Summary of KP determined using SPR and MST KP x103. Green: good correlation with SPR. Yellow: reasonable correlation. Errors correspond to the standard deviation of the triplicate fits.
Peptide | SPR KP x103 | Vesicle KP x103 | SMA KP x103 | SMA-QA KP x103 |
DMPC | DMPC/PG | DMPC | DMPC/PG | DMPC | DMPC/PG | DMPC | DMPC/PG |
LWwNKr | 0.28 ± 0.01 | 0.40 ± 0.02 | 3.10 ± 2.17 | 1.47 ± 0.07 | 6.98 ± 1.24 | 4.75 ± 0.32 | 17.12 ± 12.90 | 3.35 ± 0.57 |
WKWKWK | 0.53 ± 0.01 | 0.63 ± 0.03 | 8.56 ± 7.41 | 14.90 ± 4.84 | 0.70 ± 0.11 | 1.14 ± 0.12 | 6.15 ± 6.54 | 3.23 ± 1.87 |
WRWRWR | 1.3 ± 0.09 | 3.16 ± 0.15 | 7.08 ± 2.32 | 16.97 ± 3.83 | 9.72 ± 2.69 | 3.57 ± 0.36 | 36.55 ± 11.82 | 12.62 ± 4.16 |
WWWKKK | 2.53 ± 0.08 | 5.16 ± 0.34 | 2.55 ± .89 | 11.05 ± 4.49 | 2.34 ± 1.40 | 4.21 ± 0.43 | 9.16 ± 2.86 | 4.93 ± 0.81 |
WWWRRR | 6.65 ± 0.8 | 12.71 ± 0.16 | 7.24 ± 1.73 | 7.73 ± 1.58 | 10.40 ± 1.52 | 10.27 ± 2.13 | 7.07 ± 1.78 | 7.63 ± 1.89 |
In brief, the SPR determined KP follows the KD trend that 1 > 3 > 2 > 4 > 5 for both lipid compositions, with KP determined in the range of 0.3–7 x103 for DMPC and 0.4–13 x103 for DMPC/PG. The MST derived KP on the other hand are inconsistent with both the SPR results and the MST derived KD, including the expected differences in partitioning to DMPC and DMPC/PG (Table 4 and Fig. 8). The difficulties in the MST KP extraction compared to SPR likely lies in the intrinsic differences in the methods. Label-free MST relies on the intrinsic fluorescence of W and the instrument measures the intensity at a fixed wavelength. The fluorescent intensity of W is influenced by static and dynamic quenching and may experience blue-shifting, processes that differ significantly between different environments, modes of binding and tendency to self-aggregate.44 Significant blue-shifts of the W emission will displace the signal maximum to varying degrees away from the static detection frequency, resulting to a lower signal intensity being detected. Thus, there are additional factors that can negatively affect the detected signal in addition to the phase distribution.
Together, these results showed that the explored MST method was unreliable to extract KP for our panel of AMPs. However, there are some correlations present that suggests that with some further work, particularly around alleviating some issues around the blue-shifting, that MST might become a viable tool to estimate KP in the future. The spectral shift technology employed in newer Nanotemper devices may be ideal to further explore KP extraction by MST.45
Vesicle and nanodisc comparison
The SMA-QA- and vesicle derived KD showed differences between the two lipid systems. The vesicles produced had a diameter of ~ 140 nm and should therefore consist of approximately 200,000 lipids (with a molecular weight of ~ 140 MDa). In comparison the 22 nm DMPG SMA-QA nanodiscs contain approximately 1300 lipids (lipid weight of ~ 760 kDa), but also a substantial fraction of polymer (Table 5). The AMPs used have molecular weights between 884 Da and 1027 Da, therefore when multiple AMPs are binding to a single disc, the relative change in weight, size, and shape of the nanodisc-complex will be different than with a vesicle-complex, such relative changes may impact the Fnorm measurement as MST is sensitive to these properties of the complexes measured.46
Table 5
Comparison of estimated vesicle and nanodisc sizes. *surface area of both sides of the bilayer. ** assuming 100% DMPC composition with head area of 0.6 nm2 per lipid. *** Weight excludes SMA-QA polymer due to the uncertainty of the amount of SMA-QA per disc.
Model | Vesicle | SMA-QA nanodisc DMPC / DMPC/PG |
Radius (nm) | 72 | 6 / 11 |
Surface area (nm2)* | 120000 | 260 / 790 |
Total number of lipids** | 200000 | 430 / 1320 |
Approx. weight** | 140 Mda | 290 / 890 kDa*** |
Another difference between vesicles and nanodiscs, aside from the size, is the planarity of the lipid surfaces. Solubilised as LUVs, vesicles have a slightly convex surface curvature which introduces surface stress.47 Nanodiscs on the other hand have a planar surface48 – like the surface of the cell wall that is planar at a local level. In this context, nanodiscs may be a more representative model system. The difference in curvature between the two systems can affect the lipid phase in the bilayers of the nanodiscs and vesicles. The lipids solubilised as vesicles, have a uniform phase (at 25oC this is near the Tm of DMPC and in the liquid-ordered phase).49 In contrast, the lipids in SMA-nanodiscs are less tightly packed than those solubilised as vesicles and have a reduced melting point,50 and the same characteristics is expected of the SMA-QA nanodiscs. The central lipids of nanodiscs are in a more ordered phase,48 while the outermost lipids, closest to the SMA-belt, are perturbed by the styrene groups of SMA.50 AMPs are known to favour lipids that are in a more disordered phase and therefore one would expect heterogeneous interactions and distributions within the nanodiscs.51
Furthermore, the fractions of the lipids in the models that are accessible by the AMPs differ. In nanodiscs both sides of the bilayer are accessible, and potentially enable cooperative interactions from opposite sides of the discs. In contrast, for vesicles only the outer leaflet of the vesicle surface is initially accessible, with the inner leaflet only accessible to AMPs by first translocating across the bilayer. However, as MST is a steady state measurement, it is unclear if this affects the observed values.
The role and impact of the respective belt polymers net charge is interesting to consider with regards to the two nanodisc systems. The negatively charged SMA will favourably interact with the cationic AMPs through electrostatic interactions, thereby also retaining the peptide in the proximity of the disordered lipid region. In MST, this was observed as a strong binding of the AMPs to both the SMA nanodiscs and the polymer alone. In contrast, the cationic SMA-QA will have a repulsive effect on the AMPs, potentially repelling the AMPs from parts of the most disordered region containing the most favourable interactions on the nanodisc (Fig. 9). The difference in the size of the two nanodisc preparations may also influence the interactions, with the larger SMA-QA PG containing discs having more lipids in an ordered phase than the smaller SMA discs. The role of the net charge of the belt is however expected to be the main driving factor behind the 20–200 times stronger interaction between the SMA nanodiscs and cationic AMPs compared to vesicles and SMA-QA nanodiscs. Surprisingly, while the respective KD of the AMPs towards the two systems showed much more enhanced binding to the anionic SMA nanodiscs, KP appeared to be measurable and consistent with SPR. Further studies are required to establish if the MST measured KP reflects the true partitioning to the lipids in the SMA nanodiscs in this case, or if the apparent correlation is a product of cancellation effects. Accordingly, the interaction between the AMPs and the SMA-QA is approximately 4–10 times weaker than for the vesicles, likely affected by an electrostatic repulsion between the mutually cationic AMPs and SMA-QA. This may in turn reduce the area of accessible lipids to interact with, in particular the lipids near the polymer that are in a less ordered phase.
The differences between the SPR and MST KD could be explained by the experimental differences between the two methods. In MST, peptide response is monitored as a function of lipid concentration, whereas in SPR the added peptide mass to lipids immobilized on a chip is monitored as a function of peptide concentration. By keeping the AMP concentration fixed, any changes in activity due to concentration dependent processes, such as self-aggregation of AMPs, either pre- or post-binding, is not monitored. For the same reason, concentration dependent processes of the lipid system such as fusion, aggregation or turbidity are part of the response profile. Despite these differences, the vesicle MST and SPR produce binding data that are consistent relative to one another with regards to the ranking of the AMPs and the relative differences between the determined KD values.