Structure of Immune Stimulating Complex Matrices and Immune Stimulating Complexes in Suspension Determined by Small-Angle X-Ray Scattering

Abstract

Immune stimulating complex (ISCOM) particles consisting of a mixture of Quil-A, cholesterol, and phospholipids were structurally characterized by small-angle x-ray scattering (SAXS). The ISCOM particles are perforated vesicles of very well-defined structures. We developed and implemented a novel (to our knowledge) modeling method based on Monte Carlo simulation integrations to describe the SAXS data. This approach is similar to the traditional modeling of SAXS data, in which a structure is assumed, the scattering intensity is calculated, and structural parameters are optimized by weighted least-squares methods when the model scattering intensity is fitted to the experimental data. SAXS data from plain ISCOM matrix particles in aqueous suspension, as well as those from complete ISCOMs (i.e., with an antigen (tetanus toxoid) incorporated) can be modeled as a polydisperse distribution of perforated bilayer vesicles with icosahedral, football, or tennis ball structures. The dominating structure is the tennis ball structure, with an outer diameter of 40 nm and with 20 holes 5–6 nm in diameter. The lipid bilayer membrane is 4.6 nm thick, with a low-electron-density, 2.0-nm-thick hydrocarbon core. Surprisingly, in the ISCOMs, the tetanus toxoid is located just below the membrane inside the particles.

Introduction

Immune-stimulating complexes (ISCOMs) were first described more than 25 years ago (1) and represent a well-known type of adjuvant delivery system that has been used extensively in various experimental vaccines (2,3). ISCOMs are nanoparticulate structures composed of the immune stimulating saponin Quil A, cholesterol, and phospholipids (see Supporting Material) mixed at defined ratios along with the vaccine antigen. Both ISCOMs and ISCOM matrices (with and without the antigen, respectively) are self-assembling, cage-like structures that normally carry a net negative charge at physiological pH due to the glucuronic acid component in Quil A. It is difficult to characterize the ISCOM nanoparticle structure in detail, but such a characterization is important to optimize the formulation, the preparation process, and the storage conditions.

The ISCOMs can be prepared by a wide variety of methods, such as ultracentrifugation (1), dialysis (4), lipid film hydration (5), and ethanol injection (6). The different methods all result in a colloidal dispersion, but with differences in homogeneity, occurrence of other particulate structures and the time required to reach equilibrium in the process. The resulting colloidal structures depend not only on the preparation method but also on the ratios of the various components. In this study we chose to use the dialysis method because very pure and homogeneous ISCOM batches can be produced when applied to Quil A/1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC)/cholesterol mixtures, in this specific case at a weight ratio of 5:1:1. Transmission electron microscopy (TEM) is the traditional method used to investigate the structure of ISCOMs and ISCOM matrices in sample preparations, and over the last two decades, analyses have resulted in several different proposed structures. Typically, cage-like structures ∼40–60 nm in diameter are observed regardless of whether an antigen is incorporated or not.

In 1989, based on electron microcopy, Özel et al. (7) suggested a cage-like structure of ISCOM matrix nanoparticles prepared by dialysis. These cage-like structures consisted of Quil A and cholesterol, and they exhibited icosahedral symmetry composed of 20 ring-like morphological subunits with an outer diameter of ∼15 nm and an inner open diameter of ∼7 nm. The subunits were assembled in a pentagonal dodecahedron with a hole on each of the 12 pentagonal faces. All of these suggestions were based on detailed TEM studies of dried and stained ISCOMs and ISCOM matrices as well as freeze-dried ISCOM matrices. Also, freeze fracture EM micrographs of ISCOM matrices composed of Quil A, phosphatidylethanolamine, cholesterol, and the major outer membrane protein of Neisseria gonorrheae revealed spherical structures with surfaces showing small intrusions, possibly pores. In addition, cross sections showed micellar globular subunits with a diameter of ∼10 nm (8). For comparison, the structure of saponin and cholesterol micelles examined as early as 1964 by Lucy and Glauert (9) was proposed to represent a ring structure as an array of small 4 nm globular micelles with a hydrophobic core. In 1991, Kersten et al. (10) suggested that the ISCOM matrices can be considered as rigid, multimicellar structures that are shaped and stabilized by hydrophobic interactions, electrostatic repulsion, steric factors, and possibly hydrogen bonds. In their proposed model, the individual micelles are relatively flat, ring-shaped structures that provide space in the center for a bulky sugar chain from the saponin. The constituents are placed in stacks with the hydrophobic parts of the molecules facing the interior of the structures and the more-hydrophilic parts facing the surrounding aqueous medium (10,11). Lendemans et al. (12) and Myschik et al. (13) further discussed this proposed structure and incorporated a model for the molecular alignment in positively charged ISCOM matrices similar to the one proposed for negatively charged ISCOM matrices and ISCOMs. All previous suggested structures are based on traditional TEM studies carried out on ISCOM matrix batches of various compositions. These techniques remain the major tools to investigate the nanostructures of ISCOMs and related assemblies. However, possible disadvantages of electron microscopy are that the affinity of the different ISCOM structures to the grid can vary, and the structure can change during preparation of the sample. Thus, the structure in suspension may be quite different from the dried or snap-frozen structure on the sample grid. In addition to microscopic techniques, dynamic light scattering (DLS) has also been used to characterize ISCOMs and ISCOM matrices at a more crude level, i.e., in terms of particle size distributions. However, due to the heterogeneity of the structures that often occur in ISCOM preparations, including helical micelles, ring-like micelles, and lamellar sheets as by-products, in addition to the cage-like structures (14,15), the DLS method also has limitations that one must take into consideration when interpreting data.

To our knowledge, no studies have been published on structure determination of either ISCOM matrix particles or ISCOMs in suspension by small-angle scattering techniques. Such studies could provide valuable information about the more detailed dimensions of the ISCOMs, their polydispersity in size, and possibly also the distribution between different morphologies. In addition, localization of the antigen loaded in ISCOMs might also be derived. Up to now, such studies have not been possible because the structures in the batches are quite complex and the models for analyzing the scattering data have to be correspondingly complex, and thus have not been available (16).

Here, we present a new (to our knowledge) analysis approach for analyzing SAXS data from ISCOM particles in suspension. The method represents the structures by sets of points generated by Monte Carlo simulation. The approach is similar to traditional modeling of SAXS data, in which a certain geometric structure is assumed, the scattering intensity is calculated, and parameters describing the structure are optimized by weighted least-squares methods when the scattering intensity of the model is fitted to the experimental data. The main difference of the new method is that Monte Carlo simulation techniques are used for integrating over the volume of the objects in connection with the calculation of the scattering intensity. The Monte Carlo method has the great advantage that one can analyze the scattering data using very complex structural models because one is not limited to structures for which the intensity can be calculated analytically or semi-analytically. In the new approach, a large set of statistically uniformly distributed points are first generated by Monte Carlo simulation within a cubic search volume. For each component of the structure, a subset of the points is selected by geometric conditions for the coordinates. The subset is defined by parameters that can be varied during a least-squares optimization. The parameters also define the various components of the model (e.g., for lipid structures, hydrocarbon cores, and headgroup shell), each of which may have a different scattering length density. Because the numerical calculations involved in the calculation of the models and the SAXS intensities are very extensive, we introduce efficient algorithms and methods to speed up the calculations.

We recorded and analyzed SAXS data on suspensions of ISCOM particles without and with toxoid using the new method. The structures can be described as perforated vesicles with very well-defined structures and a certain number of perforating holes at the surfaces. The SAXS data can be modeled as a polydisperse collection of bilayer vesicles with core and headgroup regions with three types of structures: icosahedral, football (soccer ball), and tennis ball. The analysis shows that the new methods can be used to model the structure of quite complex self-assembled particles.

Materials and Methods

Materials

Cholesterol (>98%) and POPC (99%), dioleoylphosphatidylethanolamine (DOPE) (>99%), and cholesterol (>98%) were purchased from Avanti Polar Lipids (Alabaster, AL). Decanoyl-methylglucamide (Mega-10) was purchased from Bachem (Weil am Rhein, Germany) and Quil A was purchased from Brenntag Biosector A/S (Frederikssund, Denmark). Tetanus toxoid (TT) was purchased from the Statens Serum Institute (Copenhagen, Denmark). All other chemicals were obtained commercially at analytical grade.

Preparation and characterization of the ISCOM matrix and ISCOMs

The ISCOM matrices were prepared by the standard dialysis method first described by Höglund et al. (4) (Supporting Material). The size distribution was measured by DLS in undiluted samples in small volume cuvettes at 25°C using a Zetasizer Nano ZS (Malvern Instruments, Worcestershire, UK) equipped with a 633 nm laser and 173° detection optics. Malvern DTS v.5.10 software (Malvern Instruments) was used for data acquisition and analysis. For viscosity and refractive index, the values of pure water were used. The ζ potential was measured on the same instrument by laser-Doppler electrophoresis (LDE) in 1:10 diluted samples in Milli-Q water. To attach the TT to the ISCOM matrix particles, we attached DOPE to the protein using ethyl(dimethylaminopropyl) carbodiimide (EDC)/N-Hydroxysuccinimide (NHS) chemistry (17) (Supporting Material).

ISCOM matrix structure in suspension as determined by cryo-TEM

The ISCOM matrix samples for electron microscopy were prepared in a controlled environment at 29°C with the relative humidity kept close to saturation to prevent water evaporation from the sample. A 5 μL drop of the sample was placed on lacey carbon filmed copper grids, and excess liquid was removed by careful blotting with absorbent filter paper, leaving thin (<300 nm) biconcave liquid films spanning the holes of the carbon grid. The samples were then rapidly plunged into liquid ethane (−180°C) and cooled by liquid nitrogen to obtain a vitrified film. The vitrified sample was stored under liquid nitrogen and transferred in a cryo holder (Oxford CT3500), and its workstation was used to transfer the specimen into the electron microscope (Philips CM120 BioTWIN Cryo) equipped with a post-column energy filter (Gatan GIF100). The acceleration voltage was 120 kV. The images were recorded digitally with a CCD camera under low electron dose conditions.

SAXS measurements

SAXS data for suspensions were collected on the prototype of the commercially available three-pinhole NanoSTAR camera from Bruker AXS (18). The instrument uses a rotating anode as the source and Göbel mirrors to monochromatize the beam. The divergent beam is reflected by the mirrors and essentially made parallel. The sample–detector distance is 65 cm and the instrument covers scattering vectors moduli q of 0.008 – 0.33 Å⁻¹. The q-value is given by q = 4π sin(θ)/λ, where 2θ is the angle between the incident and scattered beam, and λ is the x-ray wavelength. A semitransparent beam stop of 3.0 mm in diameter was used, which allowed accurate normalization and background subtraction. The samples were kept in reusable quartz capillaries (diameter of ∼1.7 mm) thermostated to a temperature of 25°C. The beam size at the sample position was 1.0 mm in diameter and the sample volumes were ∼40 μL.

Samples with the ISCOM matrix particles without the toxoid, the complete ISCOM particles with toxoid, and a sample with the pure toxoid were measured. The corresponding buffer (phosphate-buffered saline) was measured and used for background subtraction. The SUPERSAXS package was used for background subtraction and all necessary normalizations (C.L.P. Oliveira and J.S. Pedersen, unpublished; available for download from: http://stoa.usp.br/crislpo/files/). Data were normalized to an absolute scale with a sample of pure water as the primary standard. However, the absolute scale could only be exploited in the analysis of the data from the pure TT, where the sample composition is accurately known. The ISCOM particles are relatively large and the scattering is influenced by instrumental smearing. We included this in the analysis by smearing the model intensities by the instrumental resolution function (19) for which the parameters were determined by estimating the width of the direct beam on the detector (20).

SAXS modeling by Monte Carlo integration

The TEM investigations of the ISCOM particles show that the particles are spherical, with holes perforating the surface. Because it is not easy to calculate the scattering of such spherical structures on an analytical form (21,22), we developed and implemented what to our knowledge is a new modeling method. The approach was inspired by the work of Hansen (23) and Spinozzi et al. (24). In their methods, Monte Carlo simulations are used to generate uniformly distributed points in space, and a subset of these points are used to represent the geometric structure of particles. The selected points are then used in a calculation of the small-angle scattering form factor of the particles with random orientation, as they would have in suspension.

In this work, we combined the Monte Carlo method for inhomogeneous particles with least-square methods to optimize the parameters describing the structures. In the implementation (see also Supporting Material for details), initially 10⁷ points, randomly and uniformly distributed in space, are generated within a cubic search volume of volume (2R_max)³, where R_max is an estimate of the largest possible radius of the particles. To save computational time, these points are used throughout the run of the least-squares program and generated once and for all when the program is initiated. A subset of the points is selected by geometric conditions for the coordinates. The subset is defined by parameters that are varied during a least-squares optimization, such as radius and hole size. The parameters are also used to define different phases with different scattering length densities, such as the core and shell. One phase is defined as a reference phase with scattering length density of unity. The scattering length of each point is chosen so that the preset scattering length density of each phase is obtained. To define one structure, a total for 40,000 points are selected. The scattering from randomly oriented particles with the structure given by the points can be calculated by the Debye equation (25):

$$I\left(q\right)=\sum _{i,j=1}^{N,N}{b}_i{b}_j\frac{\sin \left(q{d}_{ij}\right)}{q{d}_{ij}}$$ (1)

where N is the number of points, b_i is the scattering length of the ith point, and d_ij is the distance between the ith and jth points. However, this procedure is very slow because it requires a calculation of N² sine terms. Two features are introduced to speed up the calculations (23): 1), the use of a pair distance distribution p(r), which is a histogram of interpoint distances weighted by the scattering length of the points; and 2), the distribution of the points representing the structure divided into subsets. The 40,000 points used in the implementation are thus divided into 10 subsets with 4000 points in each. A pair distance distribution p(r) is calculated with resolution of 0.1 Å for each subset and added to a total p(r) function p_tot(r). The use of subsets is a fast way to calculate p_tot(r). In principle, the calculation of a p(r) function requires on the order of 160,000,000 distance calculations for 40,000 points, whereas the method with subsets requires only 16,000,000 calculations. The accuracy of the method with subsets is relatively good because the points and distances in the subsets are more independent than in the full set (23). Using p_tot(r), the intensity is given by the Fourier transformation described by

$$I\left(q\right)=\sum _{i=1}^M{p}_{tor}\left(r_i\right)\frac{\sin \left(q{r}_i\right)}{q{r}_i}$$ (2)

The p_tot(r) function used in Eq. 2 was normalized to give I(q = 0) = 1. The term p_tot(r_i = 0) value is omitted from the sum because this term makes I(q) level off at 1/4000 for homogeneous particles (where 4000 is the number of points in each subset), and this would limit the dynamic range of the calculation.

We included the polydispersities of the sizes of the particles in terms of size distributions using a Gaussian function. To achieve fast calculations, we introduced the polydispersity into the p_tot(r) function in an approximate way by redistributing p_tot(r) into a new histogram with the r axis rescaled and using appropriate weightings during the redistribution. Note that all distances within the structure are rescaled by this procedure (affine scaling), so, for example, a vesicle with a constant shell thickness cannot be obtained. However, the procedure is reasonable for low polydispersity values. Because we use normalized p_tot(r) functions, we decided to use a weighting with a scale factor equal to the power of 6 of the radius of the particles when redistributing the functions. This corresponds to intensity weighting for homogeneous particles. Because the particles are shell-like, it might have been better to use a weighting to the power of 4 instead. However, because the polydispersities are small, the differences in the derived polydispersities will be small.

We optimized the parameters for the structure, relative scattering length densities, scale, and background using a standard Levenberg-Marquardt algorithm. It turned out that with the chosen implementation and the number of points set to 40,000, the gradients were well defined and could be calculated for all parameters without any problems. Additional details regarding the implementation and a test of the method on a simulated dataset are described in the Supporting Material.

Geometric structures for the ISCOM matrix

The experimental SAXS data for the ISCOM matrices are shown in Fig. 1. There are pronounced oscillations, as would be observed for relatively monodisperse spherical particles. There is also a broad maximum at large q similar to what is typically observed for vesicles by SAXS as a result of the bilayer cross section, with scattering length densities of opposite signs for the hydrophobic bilayer and the hydrophilic inner and outer leaflets (Fig. S3). It is reasonable to assume that in the ISCOM particles, this feature originates from the combination of a negative excess electron density of the hydrocarbon chains of POPC, the apolar parts of Quil A and cholesterol, and a positive excess electron density originating from the more electron-dense polar parts of the molecules. For brevity, in the following we will refer to the hydrophobic part as the core and the hydrophilic part as the shell. These structural elements are incorporated in the models presented below.

Figure 1.

Experimental SAXS data for the ISCOM matrices. The broken curve is the nonsmeared fit for the model described in the text, which consists of icosahedral, tennis ball, and football ball structures. The full curve is the actual fit to the data including the instrumental smearing.

Icosahedral structure

One of the simplest types of particles with high symmetry that are compatible with the TEM micrographs is an icosahedral structure with holes at the 12 corners around which there is fivefold symmetry. This structure is equivalent to a dodecahedron with holes at the centers of the 12 pentagons. In the modeling presented here, we used an icosahedral structure and first generated an initial vesicle with a radius at the center of the layer equal to the center–corner distance. The 12 holes were made by excluding points closer to the corners than a certain distance. The included points closest to the corners were changed to shell points within a distance from holes corresponding to the shell thickness. Toroidal holes (26) would have been more realistic; however, the SAXS data did not have sufficient resolution to distinguish such details of the structure. Polydispersities were included in the modeling as described above.

The model did not fit the data well (Supporting Material) and there were large deviations, in particular around q = 0.056 Å⁻¹, where the model intensities were lower than the data. We concluded that this was due to the lateral correlations across the holes not occurring at the correct value relative to the overall size of the particles. To move the maximum originating from these correlations to a higher q-value relative to the low-q Guinier region originating from the overall size, we considered larger structures as described below.

Football structure

A structure of a dodecahedron truncated at the corners gives a football structure with 12 pentagons and 20 hexagons. The truncations were made so that regular pentagons and hexagons with identical side lengths would be obtained. The centers of the pentagons and hexagons were determined and projected onto the sphere that goes through all of the corners. Holes were made around the projected centers of the pentagons and hexagons by excluding points close to centers. The holes at the hexagons were taken to be 6/5 times larger than those at the pentagons to obtain connections of the vertices of similar thickness. This choice was made to reduce the number of free parameters. For this structure, the points closest to the holes were changed to obtain the shell scattering length densities, and the polydispersity of the size was included.

The football structure also did not fit the data well. There were large deviations around q = 0.050 Å⁻¹, where the model intensity is too low, and around q = 0.070 Å⁻¹, where the model intensity is too high compared with the data (Supporting Material). To remedy this, we attempted to use a mixture of icosahedral and football structures fitted to the data; however, this also did not provide a satisfactory fit. There were still large deviations around q = 0.050 Å⁻¹, where the model intensity is too low; however, the agreement at low q improved. We concluded that particles with a number of holes between that of icosahedrons and footballs could possibly give a better fit, and therefore we considered the structure described below.

Tennis ball structure

There are no high-symmetry structures between the icosahedron and the football structure; however, a tennis ball structure of lower symmetry with 12 pentagons and eight hexagons can be constructed (27). We determined the approximate positions of the vertex points and optimized the distance between nearest-neighbor points by minimizing the sum of 1/r, where r is the distance between nearest-neighbor points. The resulting points were then projected onto the surface of a sphere, and holes were made on the pentagons and hexagons as described for the football structure. Furthermore, a polydispersity was included. The fit to the data was very good except around q = 0.020 Å⁻¹ (Supporting Material), where the smearing of the data indicates that the polydispersity of overall size was too low in the model. However, the polydispersity was increased, the fit became worse at high q, and we subsequently concluded that icosahedral and football structures might also be present. Therefore, we applied the model with mixtures as described below.

Model with mixtures

The final model that was fitted to the ISCOM matrix SAXS data had a mixture of icosahedral, tennis ball, and football structures. To minimize the number of free parameters, each of these species had the same relative polydispersity of their radii described by a Gaussian distribution. The size of the holes on the surfaces of the particles was the same for all species, with the radius of hexagonal holes being 6/5 of the radius of the pentagons. The widths of the core and shell, as well as the relative scattering length densities of these species were also the same. However, a calculation showed that the relative volumes of core and shell were not the same in the different species, as it should be if the species had the same molecular composition. Because an assumption of similar composition is reasonable, we decided to fit only the size of the tennis ball structure and then use rescaling of the radius of the icosahedral and football structures to ensure that the relative volumes of the core and shell, respectively, would be the same. This was implemented with the use of a simple iterative variation procedure. The choice of having the same polydispersity, composition, and hole sizes of the species was made to reduce the number of free parameters.

The scattering of each of the three species were normalized to unity at q = 0 so that the relative scale factors of the species would be proportional to φ_m M, where φ_m is the mass fraction of the species and M is its mass. The relative masses can be estimated, for example, from the volume of the hydrophobic cores. By exploiting the fact that the sum of the mass fractions has to be equal to unity, we were able to calculate the mass fractions and determine the relative number of the species as φ_m /M.

Modeling SAXS data for ISCOMs with TT incorporated

We used SAXS data recorded for the TT to demonstrate that it is homologous to Clostridium botulinum neurotoxin B (28) (Supporting Material). PDB file 1EPW for this structure was used for further analysis.

The stoichiometry of the ISCOMS with TT, i.e., the number of toxoids associated with the various species of the ISCOMs, is not known. To extract this information from the modeling of the complexes, we had to estimate the scattering contrast of a toxoid relative to that of the ISCOMs. The scattering from the various components is proportional to the total number of excess electrons relative to the solvent of the respective components. The excess number of electrons of TT can be calculated to 17,700 per protein using the scattering contrast factor of Δρ_m = 2.00 × 10¹⁰ cm/g, the mass of the toxoid of ∼150 kDa, the scattering length of an electron (the Thomson radius) r_Thomson = 0.282 × 10⁻¹² cm, and a specific volume of the protein of v = 0.73 cm³/g. If it is assumed that the core of the ISCOMs consists of aliphatic chains, the number of excess electrons of the core can also be estimated using a typical excess electron density of −0.060 e/ų for the chains and the volume of the core. For the tennis ball structure, the volume of the core was determined to be ∼1.60 × 10⁶ ų, resulting in a number of excess electrons of approximately −96,500 e for the core. Because the ratios of the volumes of the core and headgroup, as well as the relative scattering length densities, are known for the model, it is now possible to normalize the scattering of a toxoid relative to the scattering from the ISCOM matrix.

When the scattering was calculated, TT was represented by 257 points placed at some of the C_α positions of the protein. The scattering from TT was normalized as outlined above. Initially, one TT was placed in each of the species; however, we later realized that we could obtain a slightly better fit by placing two TTs in the football structure. This seems reasonable because these structures are much larger than the icosahedral and tennis ball structures. The two TTs were localized at opposite positions, (x, y, z) and (−x, −y, −z), respectively, in the football structure. The distance between the core of the ISCOM shell and TT was the same in all species, and the positions and orientations of the TTs were optimized during the fits.

Results and Discussion

Characterization of the ISCOM matrix

The presence of spherical, cage-like structures 40–60 nm in size was observed by cryo-TEM (Fig. 2). The size of the ISCOM matrices was determined by DLS, which provided a Z-average value based on four measurements (Fig. 3). The reported ζ potential measured by LDE was an average of 10 measurements. The measured size range corresponded to the expected structures with a size of 40–60 nm. The mean size of the ISCOM matrices was 43.9 ± 0.2 nm and the polydispersity index was 0.141, indicating a rather narrow and homogeneous size distribution. The ζ potential was −32.3 ± 10.2 mV, showing negatively charged ISCOM matrices as expected. The reproducibility of the preparation method was confirmed by low interbatch variability as evaluated by DLS size measurements. These observations correspond to data obtained with the same components and preparation procedure (29,30).

Figure 2.

Cryo-TEM image of ISCOMs in suspension visualized by cryo-TEM. Scale bar is 50 nm. For comparison, the ISCOM matrix structures derived from the SAXS data are also shown (nonscaled).

Figure 3.

Size distribution weighted by the intensity of the ISCOM matrices measured with DLS.

ISCOM matrix structure in suspension as determined by SAXS

The various steps involved in developing the structural models to analyze the SAXS data are described in the Materials and Methods section. The best fit to the SAXS data is shown in Fig. 1. The resulting structures are displayed in Fig. 4 and the fit results are given in Table 1. The fit reproduced all features in the experimental SAXS data. The SAXS model structures agree with the structures previously visualized using TEM and cryo-TEM (10,13,30), as well as with the cryo-TEM micrographs in Fig. 2. The final model that was fitted to the data included a mixture of icosahedrons, footballs, and tennis balls (Fig. 4) with the same size of perforating holes, the same width of hydrocarbon core and headgroup shell, and the same composition. Each species also had a polydispersity, which was assumed to be the same for all species. Due to the assumptions made for relating structural parameters, the final model includes only five parameters for describing the structure: the radius of the tennis ball structure, R_tennis (to the center of membrane); half the width of the core, W_core; half of the outer extension of the headgroup W_out; the size of the holes at the pentagons, R_hole; and the relative scattering length density of the core, Δρ_core. There is also a parameter σ that describes the relative polydispersity, three scale factors for the overall scale of the scattering from the three contributions, and a constant that describes the residual background in the data. All parameters are stable in the fit and they come out with reasonable small standard errors (Table 1). This is a result of the relatively high information content in the data, and shows that the fit is sensitive to all parameters and that none of them are redundant.

Figure 4.

ISCOM matrix structure from analysis of SAXS data. (A) Icosahedral structure. (B) Tennis ball structure. (C) Football structure. The tennis ball structure model has 12 pentagons and eight hexagons arranged as sketched in panel D.

Table 1.

Parameters for the ISCOM matrix particles as derived from analysis of SAXS data

Parameters	SAXS results
Outer diameter football (calculated)	48.6 ± 0.4 nm
Outer diameter tennis ball 2 Rtennis + 2 Wout (fitted)	38.1 ± 0.3 nm
Outer diameter icosahedral (calculated)	28.8 ± 0.3 nm
Half thickness of membrane core Wcore (fitted)	1.0 ± 0.2 nm
Total thickness of membrane 2 Wout (fitted)	4.6 ± 0.2 nm
Radius of holes on pentagons Rhole (fitted)	4.8 ± 0.3 nm
Radius of holes on hexagons (calculated)	5.8 ± 0.3 nm
Mass fraction of icosahedrons, tennis balls, and footballs, respectively (fitted)	0.10:0.79:0.11
Relative number fractions of icosahedrons, tennis balls, and footballs, respectively (calculated)	0.18:0.76:0.06
Relative polydispersity σ (fitted)	5.8%
Relative scattering length density of core (fitted)	−1.41 ± 0.42

The structures determined by SAXS look very similar to those observed in the cryo-TEM micrograph of the same batch, which also shows a variation in size and morphology (Fig. 2). The SAXS analysis gives further details on the structure of the particles. The hydrocarbon core of the membrane is relatively thin, with a half thickness of only 1.0 nm. It contains the lipid tails of the POPC and probably the cholesterol-like hydrophobic part of the Quil A molecule (Fig. 4, red). The POPC headgroups and the sugar parts of Quil A, which are hydrophilic, are the main components of the headgroup region that are identified by the modeling of the SAXS data (Fig. 4, green). Cholesterol is expected to be located at the surface of the hydrocarbon core. These interpretations are in agreement with Kersten et al. (10), who suggested that the constituents of the ISCOMs are placed in stacks, with the hydrophobic parts of the molecules facing the interior of the particles and the more-hydrophilic parts facing the buffer. Note that the hydrocarbon cores in the SAXS models have a half width of only ∼1.0 nm, which is very small compared with the hydrocarbon chain length of the phospholipids. The headgroup regions in the SAXS model are ∼1.3 nm thick, which is also quite small considering the structure of the Quil A molecules (Fig. S1 C). The total thickness of the membrane is 4.6 nm (Table 1), which is in good agreement with the generally accepted values for a phospholipid bilayer thickness of ∼5 nm, given that the presence of the other molecules might influence the thickness. The relative polydispersity of the species is 5.8%, which is very low. This shows that the individual species are very well defined. The mass fraction of the species and the relative number of particles were also calculated (Table 1). With 76% tennis ball structures, the sample is completely dominated by this species. This was also confirmed by the fit of data to the tennis ball structure alone that gave a good fit to the data except at low q, where the influence of the larger football structures was significant.

In 1989, Özel et al. (7) suggested a cage-like structure that exhibited icosahedral symmetry and was composed of 20 spherical subunits assembled into ring-like structures with an outer diameter of ∼15 nm and an inner open diameter of ∼7 nm. The overall morphology of the particle was a dodecahedron structure with a hole on each of the 12 pentagonal faces. These structures are similar to the small icosahedral structures that are considered in the SAXS analysis, but they represent only a small component compared with the dominating tennis ball structure. The distinct separation into spherical subunits in the previous model is not present in our SAXS models, and one could speculate that this is an artifact in the work by Özel et al. that resulted from the staining and drying applied in the electron microscopy. However, the spherical subunits observed by Özel et al. could also be Quil A micelles, which are known to form above a critical micellar concentration of ∼0.3 mg/ml (31) and have a size of 12 nm.

Structure of TT-ISCOMs in suspension as determined by SAXS

The SAXS data for the TT-ISCOMs are displayed in Fig. 5. The oscillations in the data are not as pronounced as they are for the pure ISCOM matrix particles (Fig. 1). The model used to fit the SAXS data for the TT-ISCOMs is described in detail in Materials and Methods. From the volumes and scattering length of the hydrocarbon core and the TT, we estimated that the complexes should contain approximately one toxoid each. The fitting was therefore initially done with one toxoid per ISCOM, and the TT was initially placed sticking out like a thorn. The fit significantly improved as the toxoids moved below the surface but remained in contact with the membrane (Fig. 6). The structures shown in the figure are those of the best fit with two toxoids included in the football structure. The fit is of good quality (Fig. 5), similar to the quality of the fit for the ISCOM matrix particles.

Figure 5.

Experimental SAXS data for ISCOM particles loaded with TT. The broken curve is the nonsmeared fit for the model described in the text. The full curve is the actual fit to the data including the instrumental smearing.

Figure 6.

Suggested location of TT in the icosahedral, tennis ball, and football ISCOM structures.

The structural parameters for the best fits are displayed in Table 2. In general, the geometric dimensions of the TT-ISCOMs are very similar to those of the corresponding ISCOM matrix structures. The toxoid is located inside the particle in close proximity to the membrane. This location could potentially explain why toxoids have never been observed by TEM in samples of ISCOMs. The main difference between the results for the matrix particles and those for the TT-ISCOMs is in the distribution between the different species. The TT-ISCOMs have slightly more material in the icosahedral and football structures, resulting in a larger overall polydispersity. In agreement with this, the DLS measurements (data not shown) also clearly showed that the TT-ISCOM preparation was more inhomogeneous than the ISCOM matrix suspensions. The sample used for the SAXS measurements was sufficiently homogeneous for the SAXS analysis to give structural parameters with relatively small standard errors.

Table 2.

Parameters for the TT-ISCOMs as derived from analysis of the SAXS data

Parameters	SAXS results
Outer diameter football (calculated)	48.2 ± 0.8 nm
Outer diameter tennis ball 2 Rtennis + 2 Wout (fitted)	38.2 ± 0.7 nm
Outer diameter icosahedral (calculated)	28.6 ± 0.6 nm
Half thickness of membrane core Wcore (fitted)	1.1 ± 0.2 nm
Total thickness of membrane 2 Wout (fitted)	4.6 ± 0.2 nm
Radius of holes on pentagons Rhole (fitted)	4.9 ± 0.9 nm
Radius of holes on hexagons (calculated)	5.9 ± 1.0 nm
Mass fraction of icosahedrons, tennis balls, and footballs, respectively (fitted)	0.14:0.72:0.14
Relative number fractions of icosahedrons, tennis balls, and footballs, respectively (calculated)	0.24:0.68:0.08
Relative polydispersity σ (fitted)	6.1%
Relative scattering length density of core (fitted)	−1.74 ± 0.32

Summary and Conclusion

In this work, we used cryo-TEM and SAXS to conduct structural investigations of ISCOM matrix particles consisting of a mixture of Quil A, cholesterol and phospholipids, as well as ISCOM particles loaded with TT. In agreement with previous observations, the cryo-TEM micrograph shows ISCOM particles with very well defined structures, with a network organization on the surface and holes with pentagonal and hexagonal symmetry. We analyzed the SAXS data for the ISCOMs using a new modeling method that we developed and implemented to describe the scattering from the complex structures. The approach is similar to traditional modeling of SAXS data, in which a structure is assumed, the scattering intensity is calculated, and parameters describing the structure are optimized by weighted least-squares methods when the scattering intensity of the model is fit to the experimental data. The new (to our knowledge) feature of our modeling method is the use of Monte Carlo simulation techniques for integrating over the volume of the objects in connection with the calculation of the scattering intensity. A finite set of points generated by Monte Carlo methods is used to represent the structure, and this allows the scattering from very complex structural models to be calculated.

The SAXS data obtained from ISCOM matrix suspensions without toxoid, as well as those from complete TT-ISCOMs, can be modeled as a polydisperse collection of perforated bilayer vesicles with core and headgroup regions with icosahedral, football, or tennis ball structures. Both types of samples are characterized by small polydispersities of the three single types of species, and by the predominance of the tennis ball structure. The mean size of the tennis ball ISCOMs is very similar in the two cases, with a structure of 40 nm in outer diameter, 20 holes in the membrane, consisting of, respectively, 12 pentagonal and eight hexagonal holes. The thickness of the membrane is 4.6 nm and it has a central core consisting of hydrocarbons with a lower electron density than water. The core is surrounded by layers of the polar parts of the constituting molecules with an electron density higher than water. For the TT-ISCOMs, the SAXS data could be modeled with one toxoid in the icosahedral and tennis ball structure, and two toxoids in the football structure. The toxoid is located inside the particles adjacent to the membrane.

The application of SAXS and advanced modeling has provided detailed insight into the structure of ISCOM matrices and TT-ISCOMs. Such a study has not been possible up to now, because the small-angle scattering from the complex structures could not be modeled. The new analysis approach applied here, based on Monte Carlo integration methods, makes it possible to perform such modeling and provides a very versatile tool that also can be used to model small-angle scattering data from other complex structures, such as micelles and protein-detergent complexes (32). Although the calculations are quite demanding in terms of computational power, our implementation is efficient enough to allow the least-squares optimization of the fit parameters to be done on a personal computer.