Correlated protein conformational states and membrane dynamics during attack by pore-forming toxins

Significance

Listeriolysin O (LLO), a pore-forming toxin expressed by Listeria monocytogenes, plays a critical role in assisting bacteria escape the phagocytic vacuole. To evolve a suitable defense mechanism against LLO-mediated infections, a microscopic understanding of the process of vacuole disruption and consequent immune evasion is needed. Using a combination of imaging and dye-leakage kinetics on giant unilamellar vesicles along with atomistic simulations, we illustrate the intimate coupling between membrane-bound LLO conformations, pore formation, and phospholipid reorganization implicated in the leakage kinetics. This study has direct bearing on our understanding of membrane disruption by pore-forming toxins and lipid-reorganization events known to play an important role in signal transduction in cells.

Abstract

Pore-forming toxins (PFTs) are a class of proteins implicated in a wide range of virulent bacterial infections and diseases. These toxins bind to target membranes and subsequently oligomerize to form functional pores that eventually lead to cell lysis. While the protein undergoes large conformational changes on the bilayer, the connection between intermediate oligomeric states and lipid reorganization during pore formation is largely unexplored. Cholesterol-dependent cytolysins (CDCs) are a subclass of PFTs widely implicated in food poisoning and other related infections. Using a prototypical CDC, listeriolysin O (LLO), we provide a microscopic connection between pore formation, lipid dynamics, and leakage kinetics by using a combination of Förster resonance energy transfer (FRET) and fluorescence correlation spectroscopy (FCS) measurements on single giant unilamellar vesicles (GUVs). Upon exposure to LLO, two distinct populations of GUVs with widely different leakage kinetics emerge. We attribute these differences to the existence of oligomeric intermediates, sampling various membrane-bound conformational states of the protein, and their intimate coupling to lipid rearrangement and dynamics. Molecular dynamics simulations capture the influence of various membrane-bound conformational states on the lipid and cholesterol dynamics, providing molecular interpretations to the FRET and FCS experiments. Our study establishes a microscopic connection between membrane binding and conformational changes and their influence on lipid reorganization during PFT-mediated cell lysis. Additionally, our study provides insights into membrane-mediated protein interactions widely implicated in cell signaling, fusion, folding, and other biomolecular processes.

Pore-forming toxins (PFTs) are perhaps the earliest known proteins implicated in bacterial pathogenesis and defense against the host immune system (1). They are cytolytic in nature, with their primary virulence manifesting through unregulated pore formation and cell lysis (2). PFTs are classified as either $\alpha$- or $\beta$-PFTs, based on the secondary structure of their transmembrane domains (3). Listeriolysin O (LLO), a foodborne pathogen and a cholesterol-dependent cytolysin (CDC) secreted by Listeria monocytogenes, is implicated in gastroenteritis as well as meningitis in immune-compromised individuals (4). Pore formation and membrane rupture provide a unique phagosomal escape strategy, a key role in the survival and growth of L. monocytogenes in host cells. Despite their known existence for several decades, the molecular mechanisms for pore-formation and membrane-disruption processes, during PFT-mediated bacterial virulence, are still unclear (1).

Understanding the mechanistic pathways for PFTs has relied largely on structural data gleaned from cryo-electron microscopy (cryo-EM) and X-ray crystallography, where detergents induce oligomerization to stabilize pore structures (5, 6). Since CDCs and LLO in particular act by binding to membrane cholesterol, model lipid membranes have been used to ascertain the presence of prepore and pore structures with atomic force microscopy (AFM) (7–10). The LLO pore states can exist either as partially oligomerized arcs or complete pores, with sizes ranging from 40 to 80 nm (9) and selective influx of dyes into giant unilamellar vesicles (GUVs) exposed to LLO confirm the presence of functional oligomeric intermediates (10). A widely used method to probe protein-conformational states and their transitions is Förster resonance energy transfer (FRET) (11–14). Although FRET has been extensively used to study bulk proteins, fewer reports are available on intermolecular FRET between lipids and proteins (15, 16). Recently, FRET has been used to study oligomeric intermediates of cytolysin A, an $\alpha$-PFT in detergent (17).

To determine the process of cell lysis induced by LLO pore formation, one needs to decipher the lipid-mediated membrane response. Vesicle-leakage experiments can probe the kinetics of pore formation and cellular lysis without revealing direct information about oligomeric intermediates or lipid response (18–22). Recent superresolution-stimulated emission depletion–fluorescence correlation spectroscopy (FCS) experiments (7, 23–25) on supported lipid bilayers (SLBs) have revealed the extent of induced dynamic heterogeneity and protein-mediated lipid reorganization due to pore formation. Deviations from Brownian diffusion are observed at length scales between 100 and 200 nm. Molecular dynamics (MD) simulations, sampling time scales up to a few microseconds, have to a limited extent been used to illustrate changes in lipid dynamics and conformational transitions in the presence of pore-forming proteins (26–28) as well as antimicrobial peptides (29, 30). However, connecting the structure of the membrane-bound oligomeric intermediates to leakage and ensuing lipid reorganization is challenging, as it requires simultaneous monitoring of several quantities to yield a complete picture of pore-formation dynamics.

In this work, we attempt to bridge this gap and provide a microscopic connection between pore formation, lipid dynamics, and leakage indicative of pore function, by using a combination of FRET and FCS measurements on single GUVs exposed to LLO. Using fluorescently labeled protein and lipids with vesicles containing cyanine dye as a leakage readout, we correlate different structural states of LLO as revealed by FRET, with the changes in lipid diffusivities (FCS) and leakage from the GUVs. By using dye leakage as a readout for pore formation, two distinct populations of GUVs with widely different leakage kinetics emerge. We attribute these differences to the existence of oligomeric intermediates, sampling various membrane-bound conformational states of the protein, and connect these with concomitant changes in lipid dynamics. In the absence of leakage, FCS reveals a distinct lowering of lipid diffusivity corroborated by atomistic MD simulations. An increase in lipid diffusivities differentiates the leaked vesicles (LVs) with functional pore states from nonfunctional bound states associated with unleaked vesicles (ULVs). All-atom MD simulations are used to ascertain FRET assignments for different membrane-bound conformational states of LLO.

Results and Discussion

LLO Selectively Causes Leakage.

Although it is well known that LLO induces lysis in erythrocytes (SI Appendix, Fig. S1B) by pore formation (10, 20), the associated microscopic lipid rearrangements and dynamics during this process are unexplored. In order to make this connection, we studied the interaction of LLO with GUVs made up of dioleoylphosphatidylcholine (DOPC):1,2-dipalmitoyl-sn-glycero-3-phosphocholine (DPPC):cholesterol (Chl) (2:2:1) filled with fluorescent dye cyanine-3–N-hydroxysuccinimido (Cy3-NHS ester). Leakage kinetics was found to be a function of membrane composition, and the above composition provided optimal time scales to capture leakage kinetics through imaging and record FCS data on a specific GUV (SI Appendix, Fig. S3A). However, for GUV populations of a given composition, variation in leakage kinetics was not significant. Experiments were carried out by exposing an ensemble of GUVs tethered to SLBs to fixed concentrations of LLO. Data were collected by time-lapse imaging of individual GUVs before and after the addition of LLO. Two distinct populations of GUVs emerged in our observations (Fig. 1).

Fig. 1.

Time-lapse images of leaked (A) and unleaked (B) GUVs [DOPC:DPPC:Chl (2:2:1)] taken on the equatorial plane after exposure to 75 nM LLO toxin added at t = 0. Images illustrate Cy3 leakage (A) and Cy3 encapsulated within the vesicle (B) despite the presence of bound LLO. Additional analysis of the images illustrates that LLO (green) binds specifically the $L_d$ regions of the vesicle (SI Appendix, Fig. S3). Green, LLO; red, lipid bilayer; cyan, Cy3 dye in vesicle. (Scale bars: 10 $\mathrm{\mu }$m.)

In one population, dye leakage was observed within 3 min of LLO addition, while in a second population, leakage was not observed even after a duration of 4 h. We henceforth refer to these disparate populations of leaked and unleaked GUVs as LVs and ULVs, respectively. To rule out the role of intrinsic composition or size-related heterogeneity of GUVs in leading to the observed emergence of two populations of GUVs—LVs and ULVs—we checked for both labeled lipid and cholesterol concentration in GUV populations before incorporation of LLO. We did not find any intrinsic compositional heterogeneity of significance in GUVs, before LLO incubation, which could explain the emergence of LVs or ULVs (SI Appendix, Fig. S2 A–D). In both LVs and ULVs, the images revealed the presence of labeled LLO on the membrane (Fig. 1 and SI Appendix, Fig. S3 D and E). However, the LVs, on average, showed a higher (approximately three times) LLO concentration compared with the ULVs (data shown in SI Appendix, Fig. S3B). The inherent stochastic nature of LLO binding was revealed in the broad distribution of LLO concentration on LV or ULV populations. Coupled with the observation of concentration-dependent lytic activity of LLO on erythrocytes (SI Appendix, Fig. S1B), our time-lapsed images suggest a clear connection between LLO concentration and vesicle leakage or cell lysis. Imaging of individual GUVs revealed that reduced lytic activity of LLO (in particular) and PFTs (in general) is due to heterogeneity of protein concentration on vesicles, resulting in a population of vesicles with subcritical concentration of toxins devoid of lytic activity. These observations suggest the following membrane-bound states for LLO: In the case of the ULVs, LLO is unable to form membrane-inserted pores or arcs capable of leakage, suggesting that the protein remains as membrane-bound monomers or trapped as nonfunctional prepores in various oligomeric states (9, 10). The primary difference between these membrane-inserted or -uninserted states is the proximity of the D3 subunit from the bilayer interface (Fig. 2). Although several AFM experiments have confirmed the presence of different prepore and pore states (7, 9, 10) on supported bilayer platforms, the causal connection between these states and leakage is unclear. To distinguish between these conformational states in the LVs and ULVs, we carried out intermolecular FRET between dyes present on the D1 domain of LLO and lipids on GUVs.

Fig. 2.

(A) Histograms of the FRET efficiency between the donor Alexa 488 located on the D1 domain of LLO and the acceptor DiI present in the lipid bilayer of the GUV for LVs (red) and ULVs (green). The efficiency values of the histograms include the lower and upper limits within each band. (B) Zoomed-in histogram from A for efficiencies between 0.1 and 0.6, indicating a prominent peak at ∼0.25 present only for the LVs ($\mathcal{I}\mathcal{I}$). The higher efficiencies (>0.3) are associated with membrane parallel conformations, $\mathcal{I}\mathcal{I}\mathcal{I}$ (SI Appendix, Fig. S10) present in both LVs and ULVs. (C and D) Shown are 2D histograms of the FRET distance $z$ mapped to efficiency in LVs (C) and ULVs (D) illustrating the dominant contribution from FRET pairs associated with the conformational state $\mathcal{I}\mathcal{I}$ present in oligomers of functional pores or arcs in LVs. Color map indicates normalized occurrences. To rule out unphysical $z$ values, a lower efficiency cutoff at 0.05 is used. (E–G) Snapshots from MD simulations for states $\mathcal{I}$ (E), $\mathcal{I}\mathcal{I}$ (F), and $\mathcal{I}\mathcal{I}\mathcal{I}$ (G) with corresponding FRET distances of $z_{MD}$ = 8.34 $\pm$ 0.94 nm, 6.60 $\pm$ 0.18 nm, and 5.35 $\pm$ 0.56 nm, respectively (SI Appendix, Figs. S10 and S11).

FRET Reveals Presence of Prepores on the ULVs.

Intermolecular FRET efficiencies and the Förster distance, ${\mathrm{R}}_0$, were obtained by using FRET pairs, Atto488DMPE- and DiI-labeled on the lower and upper leaflets of a supported 1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC) lipid bilayer, respectively (SI Appendix, SI Methods and Fig. S5A). Using this value of ${\mathrm{R}}_0$ and the appropriate formalism (SI Appendix, SI Methods and Fig. S6), we extracted the donor–acceptor separation (henceforth denoted as “$z$”) from the measured FRET efficiencies on GUVs. Since the FRET measurements on GUVs were carried after 15 min of exposure to LLO, they do not reflect temporal conformational changes during pore formation, which occur on time scales lower than the leakage timescales of 4–5 min (10) (SI Appendix, Fig. S3A). In addition, these membrane-bound states either in pore or prepore forms are known to be extremely stable (8, 10, 31), and we were unlikely to observe any transitions between these membrane-bound states during our measurements. Hence, assignment of states from our FRET efficiencies was based on conformations reported from AFM and cryo-EM experiments (5, 10), as well as MD simulations carried out in this study.

The distribution of the FRET efficiencies and corresponding 2D histograms of $z$ vs. efficiency for LVs and ULVs are shown in Fig. 2. Although both LVs and ULVs had a high percentage of lower FRET efficiencies ($<$0.1), it was considerably higher for ULVs. The main distinguishing feature was the relatively higher percentage at a FRET efficiency of 0.25 for LVs, which was conspicuously absent in ULVs. Corresponding to an efficiency of 0.25, the distance $z$, between Alexa 488 (donor) on LLO and DiI (acceptor) in the lipid membrane (Fig. 2), was 5.90 $\pm$ 0.11 nm, which we associate with the membrane-inserted state $\mathcal{I}\mathcal{I}$ (Fig. 2). The corresponding $z$ values for the lowest efficiencies ($\approx$0.1) in either LVs or ULVs corresponded to 8.65 $\pm$ 0.8 nm, which we associate with prepore states, $\mathcal{I}$. These two states correspond to the states detected in AFM experiments which distinguish between the uninserted prepore and membrane-inserted pore states (9, 10, 20, 23). The slightly lower $z$ values obtained in our study can be ascribed to the location of the donor, which lies ∼2 nm (6) below the upper surface of LLO which is probed in the AFM experiments. Interestingly, our FRET data also had higher efficiency points, for both LVs and ULVs, albeit at lower occurrences, corresponding to smaller distances ($z$ = 4–5 nm) of the protein from the membrane (Fig. 2 A and C). This suggests that the donor is in close proximity to the membrane, sampling conformations where the monomer is ostensibly oriented parallel to the membrane interface in a nonfunctional state. We assigned this to state $\mathcal{I}\mathcal{I}\mathcal{I}$, present in both LVs and ULVs. Membrane-parallel conformations have also been observed in transmission electron microscopy images of LLO mutants which trap the protein in a monomeric state (31) as well as in recent structure-based MD simulations of a cytolysin A monomer (32).

Since LVs showed pronounced and almost instantaneous leakage while ULVs did not leak within the measured imaging time of 4 h, leakage is correlated with the existence of LLO in membrane-inserted states, $\mathcal{I}\mathcal{I}$. For a leakage time of 50 s, a kinetic analysis (21) estimated (SI Appendix, SI Methods, Eq. S2) the presence of five fully formed LLO pores (radius 20 nm) on a GUV of 10-$\mathrm{\mu }$m diameter, and leakage times were on the order of 5 min for a single pore. Considering that we did not observe any leakage over a period of 4 h for the ULVs, this confirms the absence of any functional LLO pores or arcs on these GUVs. Thus, LLO is likely to exist as prepores consisting predominantly of protein in state $\mathcal{I}$ or other membrane-bound conformations such as state $\mathcal{I}\mathcal{I}\mathcal{I}$. The low concentration (SI Appendix, Fig. S3B) of LLO in the ULVs precludes the transformation from prepores to membrane-inserted arcs which can potentially cause leakage. The data provide a direct correlation between conformational states of membrane-bound LLO and leakage. To connect the different conformational states with the lipid dynamics, we carried out time-resolved and -averaged measurements using confocal FCS on LV and ULVs.

Lipid Dynamics a Signature of LLO Membrane-Bound States.

FCS measurements were carried out to obtain lipid diffusion before and after the addition of LLO. The diffusivities illustrated in Fig. 3 were only obtained from the $L_d$ phase of the vesicles where the LLO predominantly binds (SI Appendix, Fig. S3 D–G), consistent with our recent study (8). The most probable $D$ values were found to be slightly higher for LVs, while a significant reduction was observed for ULVs, compared with the pristine GUVs. It should be noted here that the $D$ values in pristine GUV populations, before LLO incubation (SI Appendix, Fig. S2E), did not display such a large range of variation and, more significantly, did not have $D$ values as low as those observed for ULVs. This clearly suggests that the observed lipid diffusivities after LLO incubation in LVs and ULVs were indeed due to incorporation of LLO onto the bilayer membranes in various oligomeric states. Thus, a clear correlation between leakage and lipid diffusivity emerges. Furthermore, linking this with the FRET analysis, we associate the higher diffusivities in the LVs to the existence of LLO in state $\mathcal{I}\mathcal{I}$ (membrane inserted) while the lowered diffusivity in the ULVs is a consequence of LLO in states $\mathcal{I}$/$\mathcal{I}\mathcal{I}\mathcal{I}$. Conformations associated with these states are likely to result in lower values of $D$ in both LVs and ULVs. As mentioned earlier, LLO concentration is considerably lower in ULVs compared with LVs, suggesting a correlation between LLO concentration, conformational states, and lipid diffusivities, $D$. Is it, therefore, possible that a transition from unleaked to leaked states is triggered by a cooperative transition from states $\mathcal{I}$/$\mathcal{I}\mathcal{I}\mathcal{I}$ to $\mathcal{I}\mathcal{I}$ leaving a footprint in the lipid diffusivities? Indeed, the time evolution of lipid diffusivities (Fig. 3B) revealed a distinct reduction in $D$ after LLO addition, followed by a gradual recovery (in LVs) to a value which was similar to or slightly higher than the $D$ (Fig. 3A) values before the addition of LLO. In contrast, this recovery in $D$ was not observed for the ULVs, and lower values of $D$ were sustained. The time-dependent evolution suggests a cooperative transition of LLO from states $\mathcal{I}$/$\mathcal{I}\mathcal{I}\mathcal{I}$ to $\mathcal{I}\mathcal{I}$ and the corresponding cooperative switch in lipid $D$ to higher $D$ values, triggering the onset of leakage. To obtain microscopic insights into these different states, we performed atomistic MD simulations on LLO–membrane complexes.

Fig. 3.

(A) Histogram of lipid diffusivities, $D$, in bare GUVs and GUVs exposed to 75 nM LLO, classified based on the observed leakage of the vesicles. A shift to lower $D$ values are observed for ULVs, with $D$ increasing for LVs. (B) Time-lapse $D$ values from FCS experiments on a GUV before (black) and after (red for LVs and green for ULVs) addition of LLO. The data indicate a change from high to low $D$ values upon addition of LLO, followed by a recovery to higher $D$ for LVs. This recovery was not observed for the ULVs.

Lipid Dynamics and LLO Conformational States from MD Simulations.

Although all-atom MD simulations have been carried out for smaller PFT pore complexes (27, 28, 32), the large size (40–80 nm), absence of a crystal structure, and uncertainty in the number of protomers precluded carrying out MD simulations for the oligomeric state of membrane-bound LLO. Instead, we carried out all-atom MD simulations of single membrane-bound monomers and membrane-inserted protomer states to connect with the FRET data. Since the experiments revealed that protein binding and pore formation occurred predominantly in the DOPC-rich $L_d$ domains (SI Appendix, Fig. S3), 1-$\mathrm{\mu }$s-long all-atom MD simulations (SI Appendix, SI Methods) were first performed on a single D4 LLO subunit (Fig. 4A) bound to a DOPC–Chl (3:1) membrane, representative of the initial binding state of LLO (6, 10). Simulations with the entire LLO monomer are given in SI Appendix, Fig. S10. It is evident from the values of $D$ (SI Appendix, Table S2) and the MSDs (Fig. 4C and SI Appendix, Fig. S9) that lipid diffusivities in the LLO–membrane complex were lower than in the bare membrane. The diffusion coefficient, $D$, of lipid molecules in the upper (LLO-bound) leaflet of the LLO–membrane complex ($D$ = 3.6 $\pm$ 0.6 $\mathrm{\mu }{\mathrm{m}}^2$/s) was reduced by $\approx$46$\%$ compared with the $D$ values of lipid molecules in the bare bilayer ($D$ = 7.9 $\pm$ 0.6 $\mathrm{\mu }{\mathrm{m}}^2$/s). This quantitatively captured the observed reduction in diffusivities during the early stages of pore formation (Fig. 3B) in the LVs as well as the lowered ensemble averaged peak diffusivities observed for ULVs (Fig. 3A). The corresponding change in the lower leaflet was expectedly smaller (SI Appendix, Fig. S8).

Fig. 4.

(A) Configuration of the LLO–D4–membrane complex after a 1-$\mathrm{\mu }$s all-atom MD simulation. Cholesterol, green; D4 subunit, purple; lipid headgroups, brown. (B) D4 subunit with cholesterol molecules bound to the undecapeptide loops. (C) Lateral mean square displacement (MSD) of lipid molecules (red, upper leaflet; green, lower leaflet) in the D4–membrane complex are compared against the lateral MSD of lipid molecules on the bare membrane. Dashed lines represent data fitted to obtain diffusion coefficients. (D and E) Upper leaflet (LLO-bound) lateral mobility maps of lipid molecules (D) and density maps of cholesterol (E) reveal lowered mobility and enhanced cholesterol density in the vicinity of the D4 domain. The location of the cholesterol density maxima in the upper leaflet is spatially correlated with a weaker density maxima in the lower leaflet (SI Appendix, Fig. S8C).

Simulations with the full LLO monomer resulted in a transition from a membrane-perpendicular orientation to a membrane-parallel configuration (SI Appendix, Fig. S10) over the course of a 1-$\mathrm{\mu }$s simulation. MSDs evaluated from these simulations, despite changes in the protein orientation, revealed a lowering of lipid diffusivities in the upper (LLO-bound) leaflet (SI Appendix, Fig. S9). Mobility maps (Fig. 4D and SI Appendix, Fig. S8) revealed a significant reduction in the lateral mobility of lipids in the vicinity of LLO (${\mu }_i$ = 3.8 Å/ns) compared with the mobility of lipid molecules away from the D4–membrane complex (${\mu }_i$ = 5.0 Å/ns). Simulations with the full monomer (SI Appendix, Fig. S9) showed qualitatively similar trends, albeit with lower spatial variation. Evidence of cholesterol-binding sites in the undecapeptide loops (6) resulted in a strong density maximum in the cholesterol density map (Fig. 4E). Reduction in lipid mobilities is expected to be enhanced when several monomer units (state $\mathcal{I}$) oligomerize before membrane insertion. This would also stabilize the prepore states on the membrane.

Using the crystal structure of the LLO monomer (6) [Protein Data Bank (PDB) ID code 4CDB], we computed the distances corresponding to the FRET pairs, used in experiments, from MD trajectories ($z_{MD}$) of the full monomer initially placed vertically on the membrane (SI Appendix, Fig. S10). The ${\mathrm{z}}_{MD}$ distance from the MD simulations for state $\mathcal{I}$ over the first 100 ns where the protein retains an upright position (Fig. 2E) was 8.34 $\pm$ 0.94 nm, agreeing with the corresponding experimental FRET distance (Fig. 2C). Additionally, the $z_{AFM}$ value of 11.60 $\pm$ 0.94 nm, which is the distance between a plane corresponding to the distal part of the monomer and the phospholipid headgroups of the upper leaflet of the DOPC bilayer (SI Appendix, Fig. S12), compared well with distances reported from AFM measurements (10) for this prepore state. Over the course of a 1-$\mathrm{\mu }$s simulation, the monomer tilted to adopt a membrane-parallel orientation (SI Appendix, Fig. S10). The FRET distances, $z_{MD}$, for the membrane-parallel orientation (state $\mathcal{I}\mathcal{I}\mathcal{I}$) lie between 4 and 6 nm when the monomer intermittently sampled this state for a duration of 50 ns over the course of 1-$\mathrm{\mu }$s MD simulation (Fig. 2E and SI Appendix, Fig. S10).

FRET distances were evaluated from a single membrane-inserted LLO protomer by using the only reported CDC crystal structure for the Pneumolysin pore state (5), which has a 67% sequence similarity (33). By using homology modeling (SI Appendix, Fig. S11) and an MD simulation of 100 ns, the distance for the FRET pairs, $z_{MD}$ for the membrane-inserted state ($\mathcal{I}\mathcal{I}$, Fig. 2F) was 6.6 $\pm$ 0.18 nm, which compares well with the $z$ value of 5.90 $\pm$ 0.11 nm obtained from experimental FRET data (Fig. 2C). A $z_{AFM}$ value of 7.23 $\pm$ 0.11 nm (SI Appendix, Fig. S12) was in good agreement with AFM experiments for fully formed LLO pores (10, 23). Despite small differences between MD simulations and FRET data, the simulations allowed us to distinguish and confirm FRET assignments to the three distinct membrane-bound states.

Summary

The pore-formation mechanism of LLO has been widely studied, with experimental techniques independently revealing various structural aspects of membrane-bound LLO oligomers and ensuing lipid reorganization due to pore formation. However, connecting these observations with leakage kinetics which ultimately determines the cytolytic pathway is challenging. In this work, we explicitly bring out this connection using combined FRET, FCS, and all-atom MD simulations to probe the correlated conformational transitions of the protein on the membrane and the concomitant lipid dynamics which determine leakage kinetics. Our study has broader implications in connecting downstream signaling events modulated by membrane–protein-binding events and lipid reorganization known to play an important role in signal transduction in cells (34).

Materials and Methods

Expression, Purification, Characterization, and Labeling of LLO.

LLO was purified from Escherichia coli BL 21 by using the Ni–nitrilotriacetic acid bead-based affinity chromatography with slight modifications from the procedure as described (10). Details of methods used for protein purity analysis, site-directed mutagenesis, and labeling protocols (35, 36) are provided in SI Appendix, SI Methods.

GUV Preparation, Fixing, and Imaging.

GUVs were prepared by using the electroformation method (37). The GUVs were imaged by using a Leica SP5 microscope system as they settled on the bilayer and bound to the biotin–streptavidin complex. The protocol is described in detail in SI Appendix, SI Methods.

FRET.

To calculate the efficiency of FRET between Alexa 488 (donor) and DiI (acceptor), the GUVs containing LLO were imaged for 10 frames before bleaching the acceptor with a 534-nm laser, and 10 frames of postbleached images were taken by using the photomultiplier tube detector with the above-mentioned channels. Intensities were averaged for five frames, and the FRET efficiency was evaluated by using $E=1-I_{DA}/I_D$, where $I_{DA}$ is the intensity of the donor in the presence of the acceptor (prebleach intensity) and $I_D$ is the intensity of the donor in the absence of the acceptor. The donor and acceptor distance $z$ was calculated by using

$$z=R_0^{3/2}{\left[\frac{{\sigma }_A\pi }{2}\left(\frac{1}{E}-1\right)\right]}^{1/4},$$ [1]

where $R_0$ = 5.414 nm is the Förster distance for Alexa 488 and DiI pairs calculated as described (38). Detailed formalisms and the $R_0$ calculations are provided in SI Appendix, SI Methods.

FCS.

Time-correlated single-photon counts were collected by using the Avalanche Photo Diode detector using two channels with 500- to 550-nm and 640- to 704-nm filters and were correlated with the Piqo-quant Symphotime software. The correlation curves were analyzed and fitted by using a standard 2D autocorrelation equation (SI Appendix, SI Methods) as the data were collected on the polar region of the GUVs. The diffusion coefficient was evaluated by using

$$D=\frac{{\omega }^2}{8{\tau }_{D}ln2},$$ [2]

where $\omega$ is the full width at half maximum of the confocal beam (24).

All-Atom MD Simulations.

All-atom MD simulations were performed on a DOPC with 30$\%$ cholesterol membrane and LLO–membrane complexes. The crystal structure of LLO (PDB ID code 4CDB) was used for the monomer simulations (states $\mathcal{I}$ and $\mathcal{I}\mathcal{I}\mathcal{I}$), and homology modeling (SI Appendix, SI Methods) by using Swiss-Model was carried out with the crystal structure of Pneumolysin (PDB ID code 5LY6) to obtain the membrane-inserted protomer (state $\mathcal{I}\mathcal{I}$). Simulation details for the various systems investigated are given in SI Appendix, SI Methods.

Change History

September 7, 2021: The SI Appendix has been updated to coincide with a formal Correction.