Formation of Large Hypericin Aggregates in Giant Unilamellar Vesicles—Experiments and Modeling

Abstract

The incorporation of hypericin (Hyp) from aqueous solutions into giant unilamellar vesicle (GUV) membranes has been studied experimentally and by means of kinetic Monte Carlo modeling. The time evolution of Hyp fluorescence originating from Hyp monomers dissolved in the GUV membrane has been recorded by confocal microscopy and while trapping individual GUVs in optical tweezers. It was shown that after reaching a maximum, the fluorescence intensity gradually decreased toward longer times. Formation of oversized Hyp clusters has been observed on the GUV surface at prolonged time. A simplified kinetic Monte Carlo model is presented to follow the aggregation/dissociation processes of Hyp molecules in the membrane. The simulation results reproduced the basic experimental observations: the scaling of the characteristic fluorescence decay time with the vesicle diameter and the buildup of large Hyp clusters in the GUV membrane.

Introduction

Hypericin (Hyp; see Fig. 1 a) is a naturally occurring photosensitizer that can be extracted from the plants of genus Hypericum (1, 2). Hypericin has been studied extensively for its photoinduced antiviral and antibacterial properties, and also for its antiproliferative and cytotoxic effects in many tumor lines. All these properties, together with minimal dark toxicity and high clearance rate from the host body, make Hyp a promising agent for photodynamic therapy (see reviews in Falk (1), Agostinis et al. (3), Kiesslich et al. (4), and Karioti and Bilia (5)). Hyp is insoluble in water, where it forms nonfluorescent aggregates; these significantly suppress its photodynamic activity (6, 7). By contrast, it is well dissolved in polar organic substances including acetone, ethanol, and dimethyl sulfoxide (DMSO). In this case, Hyp possess a strong fluorescence and high quantum yield of singlet oxygen formation (8, 9).

Figure 1.

(a) The chemical structure of hypericin. (b) Schematic view of the microfluidic sample cell used in laser tweezers experiments. (Arrow) Motion of the trapped GUV in and out of the microshelter.

Hyp has high affinity to lipid membrane structures, where it is primarily dissolved in the monomer (fluorescent) form. The interaction of Hyp with artificial lipid membranes has been studied by several groups (10, 11, 12, 13, 14, 15). Using giant unilamellar vesicles (GUVs) of ternary lipid mixtures, Ho et al. (10) showed preferential localization of Hyp in the lipid rafts. In addition, Hyp was shown to have increased affinity to cholesterol-rich domains of binary mixture (cholesterol, DPPC) Langmuir monolayers (10). In our previous work, it has been shown that Hyp monomers incorporated into planar DPhPC bilayers are predominantly oriented with the S₁ ← S₀ transition dipole moments perpendicular to the membrane surface (11). The same preferential orientation was predicted by molecular dynamics simulations (12). Binding constants of Hyp partitioning into different liposomes were reported by Ehrenberg et al. (13) and Roslaniec et al. (14).

Weakening of Hyp fluorescence was observed in lipid media (e.g., in DPhPC bilayers) at high local Hyp concentrations (11). In principle, this behavior can be explained by two different processes. The fluorescence deterioration can be caused either by in-membrane Hyp aggregation or by dynamic fluorescence self-quenching of adjacent Hyp monomers. More detailed analysis by Losi (15) revealed that formation of nonfluorescent Hyp clusters dominates in gel-phase DPPC membranes, whereas dynamic self-quenching of Hyp fluorescence is characteristic for the sol-phase membranes (above transition temperature). Weakening of the fluorescence intensity was also observed for Hyp dissolved in low-density lipoprotein (LDL) and high-density lipoprotein (HDL) particles at Hyp/LDL >30:1 and Hyp/HDL >8:1 (16, 17).

As only the monomer form of Hyp can create singlet oxygen (18), aggregation of Hyp in lipid membranes (at high local concentrations) may lead to a decrease in its photodynamic efficacy. Understanding the details of Hyp aggregation in lipid structures is necessary for adjusting optimal Hyp concentration in target cells. In this work, we focused on Hyp incorporation into GUV membranes at high Hyp concentrations, as followed by time-resolved fluorescence detection. The GUVs were examined under a confocal microscope or while being trapped in a laser tweezers apparatus.

Molecular dynamic simulations of Hyp monomers and small aggregates in liposomes were performed by Jämbeck et al. (19). The applied coarse-grained model allowed simulation periods as long as 10 μs. However, the characteristic times of Hyp fluorescence changes in the GUVs studied in this work are in the range of 10–1000 s. To extend the modeling time period and to gain better understanding of the experimental observations, a simple kinetic Monte Carlo (MC) model of Hyp aggregation in GUVs was developed. This model solves the Smoluchowski coagulation equation, extended with injection and fragmentation processes, taking the two-dimensional (2D) character of the GUV surface into account.

The applied experimental techniques are described in Materials and Methods, GUV Preparation, and Fluorescence and Raman Spectroscopy Measurements. The simulation details are given in Numerical Modeling. The experimental and modeling results are presented and compared in Results and Discussion.

Materials and Methods

Materials

1,2-dioleoyl-sn-glycero-3-phosphocholine was obtained from Avanti Polar Lipids (Alabaster, AL). Hypericin, DMSO, chloroform, glucose, sucrose, and phosphate saline buffer (pH 7.4) were purchased from Sigma-Aldrich (St. Louis, MO). Methanol was obtained from Merck (Darmstadt, Germany). Stock solution of Hyp (4 × 10⁻³ M) in DMSO was prepared and was kept in the dark at 4°C.

GUV preparation

GUVs were prepared by electroformation on Pt wires (20). 1,2-dioleoyl-sn-glycero-3-phosphocholine was dissolved in chloroform/methanol (36:1, v/v) mixture at a final concentration of 0.06 mg/mL. Approximately 30 μL of the solution was deposited and dried on two Pt electrodes (40 mm long, 0.3 mm diameter). Solvent traces were removed in vacuum (90 min). Next, the electrodes were connected to 10 Hz, 0.15 V_pp alternating voltage and were placed into a 1 × 10 × 45-mm cuvette filled with a 150 mM aqueous sucrose solution. The distance between the two wires was 4 mm. The voltage amplitude was gradually increased from 0.15 to 2.1 V during 30 min and was left at 2.1 V for another 90 min. Finally, to promote vesicle closure and detachment from the electrodes, the frequency was gradually reduced to 1 Hz in ∼20 min. The described method led to formation of 1–10-μm diameter GUVs that were then stored at 4°C.

Fluorescence and Raman spectroscopy measurements

Confocal fluorescence imaging

The interaction of GUVs with Hyp was investigated first on a confocal microscope (LSM700; Carl Zeiss, Oberkochen, Germany) using a 63× oil immersion objective (N.A. = 1.46). The excitation laser wavelength was 555 nm. The GUV stock solution was mixed with 150 mM aqueous glucose solution (1:1) containing Hyp. The final Hyp concentration was 2 × 10⁻⁶ M (0.05% of DMSO). The obtained GUV-Hyp solution was placed into a 35 mm culture dish with a glass coverslip bottom (MatTek, Ashland, MA). The presence of glucose in the GUV surroundings facilitated the settlement of vesicles (having denser sucrose interior) to the bottom glass. No special effort was made to immobilize the GUVs on the glass surface. To minimize the photodynamic effect of Hyp after the illumination of Hyp-GUV complexes, confocal fluorescence and bright-field GUV images were taken every 5 (or every 10) min, otherwise keeping the sample in the dark.

Laser tweezers measurements

The initial mixing period of GUVs with the Hyp solution (first few minutes) could not be followed with the confocal microscope. To get insight into the early Hyp incorporation to GUVs, the evolution of Hyp fluorescence signal was acquired in a series of laser tweezers experiments. The advantage of convenient buffer exchange in the GUV surroundings while holding a single GUV in an optical trap was used (see Fig. 1 b). Details of the optical setup were given in our previous works (21, 22). The trapping laser was operated at 785 nm (10 mW) while another 488 nm (2 μW) laser was used to excite the fluorescence signal. At the beginning of the experiment, a vesicle (trapped in Hyp-free glucose/sucrose buffer) was transferred into a “microshelter”—a thin, dead-end side arm of the fluidic channel (Fig. 1 b) (21). After buffer exchange in the main channel, the GUV was moved back into the new Hyp (2 × 10⁻⁶ M) solution. Fluorescence spectra were recorded by means of a spectrograph (iHR 550; HORIBA Jobin Yvon, Edison, NJ) equipped with a TE-cooled charge-coupled device camera.

Raman spectroscopy

The modular optical setup (see above) was also used to detect resonance Raman spectra of Hyp in the GUV membrane. The measurements were carried out on GUVs stabilized in dense sucrose/glucose solutions (without trapping). Before Raman detection, a droplet of GUV solution (in the presence of 2 × 10⁻⁶ M Hyp) was air-dried for few hours on a glass coverslip. Raman spectra were recorded using the 488 nm laser excitation (70 μW on the sample); the acquisition time was 60 s.

Numerical modeling

The evolution of Hyp aggregate size distribution in the GUV membrane was followed by kinetic Monte Carlo simulations. We aimed to build a simple model that can reproduce the basic phenomena observed in the experiments. The scheme of the model is shown in Fig. 2 a. The following processes were taken into account: 1) approach of aggregated Hyp molecules present in the buffer solution to the GUV surface by bulk diffusion; and 2) further aggregation (fusion of smaller aggregates) and fragmentation (detachment of Hyp monomers) of Hyp clusters incorporated into the GUV membrane. Both processes were characterized by corresponding rate constants. The model simulated the evolution of Hyp aggregate size distribution, which was used to calculate the time-dependence of the expected fluorescence signal. The obtained modeling results were compared with experimental fluorescence data. In the following, we provide a detailed description of the most important model parts.

Figure 2.

(a) Simplified scheme of the proposed model. (b) Schematic view of Hyp aggregates immersed in the membrane. To see this figure in color, go online.

Incoming flux of Hyp aggregates

Hyp aggregates diffused from the solution to the GUV. Based on our previous measurements (7), the bulk diffusion constant of Hyp aggregates in water was estimated as D_w = 4.3 × 10⁻¹¹ m² s⁻¹. The effective radius of these aggregates (approximated by rigid spheres) was calculated from the Stokes-Einstein relation: R_w = kT/(6πμD_w) = 5 × 10⁻⁹ m, where μ is the water viscosity. Assuming an average volume per Hyp molecule in the aggregate V_m = 5 × 10⁻²⁸ m³, we found that the aggregates are composed of ∼N_arr = 1000 Hyp molecules. It is noted that the value of V_m used here exceeds the van der Waals Hyp volume (23) by ∼30%, accounting for partially disordered packing of Hyp inside of large aggregates.

In our model, it was assumed that all the aggregates in the aqueous phase were of the same size (composed of N_arr molecules) and that all impinging aggregates stick to the GUV surface. For the sake of simplicity, the total flux of aggregates reaching the GUV was taken to be constant (neglecting the early transient period) and was calculated from the steady-state solution of the diffusion equation in spherical coordinates (24):

$${\Phi }_{\text{arr}}=4\pi c_0{R}_{\text{GUV}}{D}_w/N_{\text{arr}},$$ (1)

where c₀ is the total concentration of Hyp molecules in the buffer. The flux Φ_arr determines the rate of arriving aggregates in the Monte Carlo model.

Hyp diffusion in the membrane

The motion of Hyp monomers and aggregates embedded in the GUV membrane is driven by diffusion. As it was shown in Strejčková et al. (11), Losi (15), Weitman et al. (25), and Eriksson and Eriksson (26), Hyp monomers dissolved in the membrane are located below the membrane surface close to the water/lipid interface. The size of larger Hyp aggregates may exceed the bilayer thickness. According to the modeling results of Jämbeck et al. (19), small aggregates composed of a few (up to 10) Hyp molecules stick out of the membrane, and only penetrate the bilayer with the outermost molecules. Our attempt to build a model where Hyp aggregates are tethered to the outer membrane surface through a small number of Hyp molecules failed to reproduce the experimental observations. It was then hypothesized that the large Hyp complexes (composed of hundreds or thousands of Hyp molecules) are embedded in the membrane, forming cross-membrane structures (see Fig. 2 b).

In this numerical model, a unified diffusion scheme was applied for all the different Hyp complexes moving in the 2D membrane. The aggregates composed of i molecules were assumed to be compact and were approximated by spherical particles of radius R_i (see Fig. 2 b) with their volume V_i being proportional to the number (i) of Hyp molecules included: V_i = iV_m = (4/3)πR_i³. As the membrane viscosity significantly exceeds the viscosity of the surrounding medium, it was assumed that the dominant drag on particles comes from the membrane. Accordingly, the in-membrane diffusion coefficient of Hyp particles D_i was approximated with that of a circular disk that spans the membrane (equating the disk radius to R_i; see Fig. 2 b). Indeed, detailed theoretical analysis of the viscous drag showed that at high membrane viscosities, the effect of object protrusion into the surrounding fluid can be neglected when the protrusion is smaller than the viscose length scale l = η/(μ₁ + μ₂) (27). Here, η is the membrane surface viscosity, while μ₁ and μ₂ are the bulk viscosities of the media on the two sides of the membrane. For these experimental conditions (η ≈ 5 × 10⁻¹⁰ Pa.s.m (28), μ₁ ≈ μ₂ ≈ 1 × 10⁻³ Pa.s) l = 250 nm was obtained. The diffusion coefficient of aggregates was calculated by the empirical formula of Petrov and Schwille (29) and Petrov et al. (30), which approximates the modeling results of Hughes et al. (31) derived for cylindrical membrane inclusions. The analytical formula determines the diffusion coefficient as a function of the reduced particle radius ε_i = R_i/l and has a form: D(ε) = k_BT/(4πη) × [ln(2/ε) – γ + 4ε/π – (ε²/2)ln(2/ε)] × [1 – (ε³/π)ln(2/ε) + vε^p/(1 + wε^q)]⁻¹, where v = 0.73761, w = 0.52119, p = 2.74819, q = 0.51465, and γ = 0.577215 is the Euler constant.

The chosen description represented a compromise. In case of small aggregates (or Hyp monomers fully immersed in the membrane), our approach underestimated the diffusion coefficient, while for extremely large Hyp complexes (typically for R_i > l), the applied diffusion coefficient was overestimated. The possible effect of membrane curvature on the diffusion coefficient (32) was neglected.

Aggregate collisions in the membrane

Larger Hyp complexes are formed in the GUV membrane through diffusion-controlled collisions of Hyp aggregates and/or monomers. The total rate B_ij for association of smaller aggregates is given by

$$B_{i,j}=P_S{\beta }_{i,j}\times \{\begin{array}{cc}{n}_i{n}_j& i\ne j\\ \frac{n_i\left(n_i-1\right)}{2}& i=j\end{array},$$ (2)

where n_i and n_j are the numbers of aggregates composed of i and j Hyp molecules in the GUV. It was assumed here that the colliding particles stick together (forming larger aggregates) with a constant probability P_s. The collision kernel β_i,j was calculated from the approximate collision rate constant given by Martins et al. (33) for particles diffusing to randomly distributed traps on a 2D surface. In our model, the less abounded of the colliding aggregates (i or j) were considered as traps, to which the other ones diffuse with the mutual diffusion constant D₂ = D_i + D_j. The used collision kernel is given by

$${\beta }_{i,j}=\frac{D_2}{4R_{\text{GUV}}^2}\frac{1.8}{\text{ln}\left(\raisebox{1ex}{$R_d$}\!\left/ \!\raisebox{-1ex}{$R_e$}\right.\right)},$$ (3)

where R_d is the effective diffusion space radius around the traps (R_d² = 4R_GUV²/n_trap), and R_e = R_i + R_j is the encounter radius of colliding aggregates. Again, the effect of membrane curvature (and the spherical shape of the GUV) on the collision rate was neglected (34, 35).

Dissociation of Hyp aggregates

Dissociation (or fragmentation) of Hyp aggregates in the GUV membrane is of major importance. The Hyp fluorescence signal observed experimentally in GUVs (or other lipid structures) was emitted by Hyp monomers (7, 16). These monomers were formed by fragmentation and dissociation of aggregates in the lipid environment. For the sake of simplicity, only detachment of Hyp monomers from larger aggregates was considered in our model. In agreement with the picture depicted in Fig. 2 b, dissociation happens along the perimeter of the aggregate with its rate $K_i^{\text{diss}}$ being proportional to the aggregate radius R_i:

$$K_i^{\text{diss}}=k_{\text{diss}}2\pi R_i\left(\frac{S_{\text{lip}}}{S_{\text{lip}}+S_{\text{hyp}}}\right)n_i,$$ (4)

where $k_{\text{diss}}$ is the dissociation rate constant (number of events per unit length and unit time) in an empty membrane. Monomer detachment in the presence of other aggregates was viewed as a landing process limited by the available free GUV surface. The corresponding probability of successful detachment was simply calculated through the bracket term in Eq. 4. $S_{\text{lip}}$ is the overall lipid membrane area, while $S_{\text{hyp}}={\sum }_i{n}_i\pi R_i^2$ stands for the area occupied by Hyp aggregates.

The kinetic MC model

The aggregation-dissociation process of Hyp molecules in the GUV membrane (completed by particle injection from the surroundings) can be described by the corresponding generalized system of Smoluchowski equations:

$$\frac{\partial n_k}{\partial t}={\Phi }_k+\frac{1}{2}\sum _{i=1}^{k-1}{B}_{i,k-i}-\sum _{i=1}^{\infty }{B}_{i,k}+K_{k+1}^{\text{diss}}-K_k^{\text{diss}},$$ (5)

$$\frac{\partial n_1}{\partial t}=-\sum _{i=1}^{\infty }{B}_{i,1}+2K_2^{\text{diss}}+\sum _{i=3}^{\infty }{K}_i^{\text{diss}}.$$ (6)

In this model, the injection term Φ_k is only nonzero for k = N_arr. The graphical presentation of the simulated processes is shown in Fig. 3.

Figure 3.

The modeling scheme of aggregation and dissociation processes. To see this figure in color, go online.

The evolution of aggregate population as defined by Eqs. 5 and 6 was followed by a direct-simulation kinetic Monte Carlo model (see e.g., Garcia et al. (36) and Kruis et al. (37)). The applied numerical scheme was very similar to the algorithm used by Garcia et al. (36) and Malyshkin and Goodman (38). In each step, two random numbers X and Y (uniformly distributed between 0 and 1) were generated. X was used to determine the time interval between successive events:

$$t_{S+1}-t_S=-\frac{\text{ln}X}{K_{\text{tot}}},$$ (7)

where $K_{\text{tot}}={\sum }_{i=1}^N{\sum }_{j=\text{i}}^N{B}_{i,j}+{\sum }_{i=1}^N{K}_i^{\text{diss}}+{\Phi }_{\text{arr}}$ is the cumulative rate of all the processes, and N is the largest aggregate size present in the membrane. The event type at t_S+1 (pair aggregation, dissociation or arriving new particle) was chosen from an ordered list R_k of the event rates (B_{i≤ j}, K_i, and Φ_arr) selecting the event number l when

$$\sum _{k=1}^{l-1}{R}_k<K_{\text{tot}}Y\le \sum _{k=1}^l{R}_k.$$ (8)

To optimize the algorithm, only the occupied aggregate sizes (n_i ≠ 0) were stored and followed by the code. The MC model was tested by running the code with a constant kernel β_ij = A (without injection and dissociation) and comparing the simulation results with the analytical solution of the corresponding coagulation problem (37).

Fluorescence calculations

The fluorescence signal was calculated in a separate Monte Carlo process. This additional code simulated the spatial distribution of aggregates on the spherical GUV surface. Hyp aggregates of a given size distribution were successively placed (starting from the larger ones) to random positions taking the already occupied locations into account. Once the aggregates were distributed on the GUV surface the number of fluorescence molecules n_fluor was calculated. It was assumed that only Hyp monomers having no neighbors within a quenching radius R_q (defined as the edge-to-edge distance) are emitting fluorescence. The fluorescence intensity was assumed to be proportional to n_fluor.

Fitting unknown parameters to experimental LDL data

Three modeling parameters—the sticking probability of colliding aggregates in the GUV membrane P_s (see Eq. 2), the dissociation rate constant $k_{\text{diss}}$ (Eq. 4), and the fluorescence quenching radius R_q—were to be estimated without available reference data. To minimize the uncertainty of these parameters, the model was modified to simulate aggregation of Hyp molecules in LDL particles. Comprehensive experimental investigation of Hyp dissolved in LDL has been published (16). The percentage of Hyp monomers and the fluorescence intensity have been reported for different numbers of Hyp molecules (ranging from 0 to 200) in the LDL particle. It was assumed here that Hyp molecules are mostly located in the outer phospholipid shell of the LDL particle (39, 40), and do not enter the highly hydrophobic LDL core. From a modeling point of view, this makes the LDL system very similar to GUVs.

To reproduce the LDL data, the model was run with a fixed number of Hyp molecules looking for steady-state (equilibrated) size distribution of aggregates. To adapt the model to LDL conditions, the diffusion coefficient of the aggregates was modified twofold. First, the viscosity of the inner LDL core (needed to calculate the diffusion coefficient) was taken to be 100 times the water viscosity. Second, the significant part of the LDL surface was occupied by the apoB-100 protein (41). Resonant energy transfer from the apoB-100 tryptophan residues to Hyp in LDL particles has been studied previously by steady-state and time-resolved fluorescence techniques (16, 40). There was no direct experimental evidence of Hyp binding to the apoB-100. Accordingly, it was only assumed in our model that Hyp diffusion is restricted in the LDL shell by apoB-100. The diffusion coefficient was then multiplied by a factor of (1–2c) (42), where c = 0.45 was the estimated percentage of the LDL surface taken by the apoB-100 protein (42). This approach is a rough approximation that needs to be justified by additional experiments and/or simulations.

As both the aggregate sticking probability P_s and the dissociation rate coefficient $k_{\text{diss}}$ affect the equilibrated number of Hyp monomers, only one of these two (interconnected) parameters can be fitted to reproduce the experimental percentage of monomers in the LDL particle (16). It was our choice to set the value of P_s to 0.01 and to calculate the corresponding $k_{\text{diss}}$ by fitting the modeling results to experimental LDL data. One must keep in mind that even if the obtained P_s, $k_{\text{diss}}$ pair is consistent with the experimental LDL results, their absolute values may be erroneous.

The quenching radius R_q was obtained by calculating the fluorescence signal of Hyp aggregates distributed on the LDL surface. The value of R_q was varied until the relative fluorescence intensity (plotted against the number of Hyp molecules per LDL particle) described the experimental data (16) the best.

Results and Discussion

Time-evolution of Hyp fluorescence in Hyp-GUV complexes

Fig. 4 a shows a series of bright-field and confocal fluorescence images of a GUV recorded after adding Hyp (2 × 10⁻⁶ M) to the GUV solution. The most interesting effect observed here was the appearance of dark spots in the GUV membrane. The number of these dark spots increased in time and the spots changed position as they moved on the GUV surface. The presence of dark spots was observed in many independent experiments. The confocal image of the GUV-Hyp complex, taken 5 min after mixing GUV and Hyp solutions (Fig. 4 a), showed a high intensity of Hyp fluorescence that was gradually decreasing in time (see Fig. 4 b). This observation led us to conclusion that the dark spots were formed by large Hyp aggregates, as discussed in detail below.

Figure 4.

(a) Sequence of bright-field and confocal fluorescence images of two adherent GUVs mixed with 2.10^–6 M Hyp. (b) Time dependence of the total fluorescence intensity of Hyp incorporated to the GUVs. To see this figure in color, go online.

The buildup of dark spots was seen also when keeping the GUVs in dark during the whole period of Hyp incorporation into the GUV lipid membrane. This indicated that the observed extensive Hyp aggregation was not conditioned by the possible oxidation of membrane lipids. It is important to note that the size and shape of the GUVs did not change significantly throughout the experiment. More detailed analysis of the fluorescence images revealed the presence of internal protrusions in the GUV membrane, the location of which did not correlate with the position of the dark spots. Inner budding of GUVs was observed by other authors at different experimental conditions (see, e.g., Heuvingh and Bonneau (43) and Kerdous et al. (44), and references cited therein). More detailed analysis is needed to explain the morphological membrane transitions in the studied GUV-Hyp system, which is beyond the scope of this work.

Raman spectroscopy was employed to confirm the presence of Hyp in the dark spots. The studied vesicles were stabilized by partially air-drying a droplet of glucose/sucrose solution containing GUVs and Hyp. During the stabilization process, the dark spots grew further in size, as it can be seen in the inset of Fig. 5. Raman spectra recorded from the dark spot (spectrum A), from the GUV membrane region (spectrum B), and from the vesicle surroundings (spectrum C) are shown in Fig. 5. All the observed Raman peaks coming from the dark spot are attributed to Hyp vibrations (45). Presence of low-intensity Raman bands of Hyp was also observed outside of the dark spot in the GUV membrane (spectrum B). It means that some of the Hyp molecules were not present in the dark spot and were spread on the GUV surface. The most intense bands in the 1250–1400 cm⁻¹ region can be assigned to in-plane ring-stretching modes of Hyp (46). A comparison of the Raman bands in this spectral region with the results published in Raser et al. (45) indicates that the structure of the observed Hyp clusters was very similar to, or only slightly less packed than, the solid form of Hyp.

Figure 5.

Raman spectra detected in the dark spot (A), in the GUV membrane (B), and in the GUV surroundings (C). (Inset) Image of the stabilized GUV used for measurements. To see this figure in color, go online.

Laser tweezers experiments were carried out to improve the time resolution of the fluorescence recording after mixing the GUV and Hyp solutions. The fluorescence spectrum of Hyp recorded from a trapped GUV-Hyp complex has a blue-shifted maximum (at 599 nm) as compared to Hyp dissolved in DMSO (maximum at 603 nm, see Fig. 6 a). This is in good agreement with our previous observations of Hyp fluorescence in lipid structures (11, 16). Typical time-evolution of Hyp fluorescence in trapped GUV-Hyp complex is shown in Fig. 6 b. After positioning the GUV into the Hyp solution, the total Hyp fluorescence signal increased, then reached a maximum and finally decayed. The initial increase of the fluorescence intensity indicates that Hyp aggregates diffusing to the GUV surface from an aqueous solution were dissolved in the membrane and formed fluorescent Hyp monomers. The situation changed after reaching a critical concentration of Hyp in the membrane, when the Hyp fluorescence reached a maximum. As mentioned in the Introduction, this decrease of fluorescence intensity can be explained in two ways: either the number of Hyp monomers decreased due to aggregation, or the fluorescence of monomers was quenched by other Hyp molecules located nearby.

Figure 6.

(a) Fluorescence spectra of Hyp dissolved in a GUV membrane and in DMSO. (b) Time-dependence of the total (integrated) Hyp fluorescence emitted from a GUV-Hyp complex (diameter of 1.3 μm) trapped in laser tweezers. The GUV was positioned to the Hyp solution at t = 50 s and was released from the trap at t = 700 s. The decrease of the Hyp fluorescence signal was fitted by an exponential function. (c) The characteristic decay time of the fluorescence decrease plotted against the squared GUV diameter. The solid point at 25 × 10⁻¹² m² belongs to the fluorescence decrease of Fig. 4b. To see this figure in color, go online.

The decrease of the fluorescence signal was fitted by a single exponential function with an additive constant (Fig. 6 b). The characteristic decay times determined for different-size GUVs are plotted in Fig. 6 c. In principle, there is no direct physical explanation for using the exponential fitting procedure, which is rather to be considered as an empirical description of the observed signal. Still, it is interesting to note that the observed characteristic decay times tend to be proportional to the square of the GUV radius, as shown in Fig. 6 c.

The solid point in Fig. 6 c was obtained by fitting the fluorescence data of nontrapped GUVs from Fig. 4 b. The kinetics of Hyp fluorescence decrease shows the same tendency in both experiments. The estimated total light dose was >100 times lower in the confocal microscope as compared to a typical laser tweezers measurement, which supports the assumption that the observed fluorescence decay was due to Hyp aggregation and self-quenching in the GUV membrane rather than Hyp photobleaching.

Modeling of GUV-Hyp complexes

The kinetic Monte Carlo model was applied first to simulate the equilibrated size-distribution of Hyp aggregates in LDL particles. As mentioned in Numerical Modeling, the experimental data on the percentage of Hyp molecules in monomer state as a function of Hyp/LDL ratio (16) were used to estimate the value of the dissociation rate constant $k_{\text{diss}}$ used in our model (see Eq. 4). The experimental data are plotted in Fig. 7 a as a function of the total number of Hyp molecules per LDL particle, together with the modeling results (red line) corresponding to $k_{\text{diss}}$ = 1.8 × 10¹³ m⁻¹ s⁻¹. The model reproduced the experimental observations very well. One should keep in mind that the obtained value of the dissociation rate constant is consistent with our choice of sticking probability (P_s = 0.01, see the discussion in Numerical Modeling). Higher P_s would result in a higher value of $k_{\text{diss}}$, and vice versa.

Figure 7.

Fitting the modeling parameters to experimental data measured on LDL-Hyp complexes. (a) Measured and calculated percentage of Hyp monomers for different Hyp/LDL ratios. The dissociation rate constant $k_{\text{diss}}$ was adjusted in the model to reproduce the experimental data. (b) Measured and calculated fluorescence intensity of Hyp at different Hyp/LDL ratios (normalized to unity). The quenching radius R_q was tuned in the MC code to fit the experimental data. To see this figure in color, go online.

In the next step, the second MC code calculating the Hyp fluorescence signal (corresponding to the number of unquenched Hyp monomers) was run for the obtained size distributions of Hyp aggregates in LDL particles. The quenching radius R_q was varied, looking for the best overlap between the experimental and simulated fluorescence data (Fig. 7 b). The best agreement was found for R_q = 2.8 nm. This quenching (edge-to-edge) distance can be compared with the critical distance R₀ = 2.7 nm reported for Förster energy transfer from tryptophan residues to Hyp inside of LDL (40).

Finally, the model was applied to simulate the incorporation of Hyp into vesicles of various diameters. Due to computation limitations, we only report simulation results for vesicle diameters smaller than 240 nm, which is approximately one-sixth of the smallest GUV followed experimentally in this work. Typical time dependence of Hyp fluorescence (expressed by the number of unquenched monomers), calculated for a vesicle diameter of 200 nm, is shown in Fig. 8 a. To decrease the effect of stochastic fluctuations, the average of three independent simulations is plotted in the figure. The calculated fluorescence curve resembles the experimental one (see Fig. 6 b). After reaching a maximum, the signal gradually decreased to approximately one-fourth of its maximal value. At the same time, the number of monomers gradually increased, as indicated by the solid line in Fig. 8 a. According to our simulations, only ∼6% of the monomers remained unquenched at 3 s of the simulation time at these conditions.

Figure 8.

(a) The time-dependence of the calculated number of unquenched monomers, which determines the fluorescence signal (left scale) and the number of Hyp monomers (right scale) in a 200-nm diameter vesicle. (b) The characteristic decay time of the simulated Hyp fluorescence decrease (solid squares) and the experimental data from Fig. 6c (open circles). To see this figure in color, go online.

The course of the simulated Hyp fluorescence decrease (Fig. 8 a) is not a single exponential curve. The characteristic decay time (to be compared with the experiment) was obtained by fitting the late decrease (after 1 s in Fig. 8 a) with a single exponential function and an additive constant. The characteristic decay times obtained this way for different vesicle diameters are plotted in Fig. 8 b together with the experimental data of Fig. 6 c. The simulation results can be well extrapolated toward larger GUV diameters to get a good overlap with the experimental data. The obtained exponents of the power-law dependences plotted in Fig. 8 b are: 2.0 ± 0.1 for the simulation results (solid squares) and 2.2 ± 0.3 for the experimental data (open circles).

Our approach to fit only the late decay of the simulated fluorescence signal is purely empirical. The possible reasons of the observed discrepancy between the simulated and experimental fluorescence curves (mostly in the early transient period) can be listed as follows:

• 1.The flux of Hyp aggregates entering the membrane is oversimplified in the model: the size distribution of incoming aggregates is approximated by a monodisperse distribution. In reality, both smaller and larger Hyp aggregates approach the GUV surface, which may influence the concentration changes of Hyp monomers in the membrane.
• 2.The model assumes that only Hyp monomers are detached from larger aggregates in the membrane, while the possibility of aggregate fragmentation is discarded. Tentatively, this assumption may overestimate the number of monomers during transient periods.
• 3.Simulated Hyp aggregates approaching the vesicle surface from the aqueous surroundings are immediately embedded into the membrane. Again, this may overestimate the rate at which Hyp is dissolved into the lipid phase.

To reproduce the appearance of extremely large Hyp aggregates in the vesicle membrane, the model was run for a prolonged time (11 s) on a 100-nm diameter vesicle. The size distributions of Hyp aggregates in the membrane, obtained at 1 and 11 s of the simulation period, are plotted in Fig. 9 a. The figure inset shows a simplified visualization of the aggregates (at t = 11 s) placed on the vesicle surface. The corresponding size distribution of Hyp aggregates (open black circles) can be divided into three groups. First, the number of small aggregates containing up to ∼100 Hyp molecules decreases with the aggregate size. The distribution of this group obeys a power-law dependence on the number of molecules in the aggregate with the scaling exponent of −1.6, as indicated by the blue line in Fig. 9 a. It is interesting to note, that the size-distribution of these small aggregates did not change between 1 and 11 s. Moreover, the scaling exponent was found to be independent of the vesicle size (data not shown). Next, there is a low abundance intermediate group of larger aggregates (100 < i < 10,000). Finally, the third group is represented by the two largest aggregates composed of ∼112,000 and 33,000 Hyp molecules, which represents ∼60 and 18% of all the Hyp molecules in the vesicle, respectively. These huge clusters can be assigned to the dark spots observed experimentally on GUVs.

Figure 9.

(a) Size distribution of Hyp aggregates (composed of i molecules) in a 100-nm-diameter vesicle after 1 s (solid points) and 11 s (open circles). (Straight line) Power-law dependence of the aggregate number on the aggregate size. (Curved line) Guide to the eye. (Inset) Simplified visualization of the aggregates in the membrane after 11 s. (b) Time dependence of the number of Hyp molecules contained in the largest aggregate i_max divided by the total number of molecules in the vesicle. To see this figure in color, go online.

The time dependence of the number of Hyp molecules contained in the largest aggregate i_max divided by the total number of Hyp molecules N_tot in the vesicle is shown in Fig. 9 b. The spikes between 0.1 and 1 s are caused by individual Hyp aggregates (i = 1000) entering the vesicle membrane from the water surroundings. The spike widths indicate the time needed to dissolve these aggregates in the membrane. The largest aggregates started to grow beyond the power-law distribution after ∼1 s. Indeed, the size distribution at 1 s (see the solid blue circles in Fig. 9 a) only contains the low-size part of the distribution.

The presence of large Hyp aggregates in GUV membranes reminds the phenomenon of gelation that occurs in different coagulating systems, as discussed by other authors (38, 47, 48, 49, 50). In the case of gelation, an infinite cluster appears at a critical time. By contrast, multiple large Hyp aggregates are observed in vesicle membranes both experimentally and by simulations in our case.

Conclusions

The main result of this work is the observation that GUV membranes facilitate extensive Hyp aggregation, leading to formation of extremely large Hyp clusters that are not observed in aqueous Hyp solutions. The aggregated form of Hyp is biologically inactive. To the best of our knowledge, large Hyp aggregates were not yet observed inside of living cells. Potentially, our observations may be interesting from drug delivery point of view, when various types of vesicles or other lipid-based carriers (e.g., LDL particles) are loaded with Hyp. These results extend the knowledge about the loading capacities of lipid carriers and the aggregation state of drugs inside these vehicles. All this can be helpful for improvement of the efficacy and safety of the drug transport inside biological organisms.

It would be desirable to get a detailed description of the interaction of large Hyp aggregates with phospholipid bilayers. It is assumed in our present kinetic MC model that these aggregates form transmembrane structures; however, this assumption requires further experimental and/or theoretical (e.g., modeling) elucidation. Despite the limitations of this model, the simulation results reproduce the basic experimental observations well: the scaling of the characteristic fluorescence decay time with the vesicle diameter and the buildup of oversized Hyp aggregates in the GUV membrane. Future experimental work in our group aims to test the modeling results on large unilamellar vesicle systems with particle diameters in the range of 100–500 nm.

Author Contributions

J.J. performed experiments and wrote the article; M.R. performed numeric simulations; A.S. performed experiments and wrote the article; V.H. performed experiments; J.S. designed research; D.J. designed research and wrote the article; P.M. designed research; and G.B. designed research, performed experiments, and simulations, and wrote the article.

Editor: Ana-Suncana Smith.