Apolipoprotein A-I structural organization in high density lipoproteins isolated from human plasma

HDL isolation

To study apoA-I in isolated human HDL, we used a one-step gradient ultracentrifugation method to obtain 5 density subfractions from normal human plasma that have been extensively characterized previously ²⁴ (defined here as HDL_2b-HDL_3c). ApoA-I comprises between 60-70% of total protein in HDL, with apoA-II the second most abundant at about 20%. Although the cross-linking technique can provide structural information in complex systems, we felt it was important to simplify the HDL particles as much as possible. Therefore, for each HDL density subfraction, we further isolated particles containing apoA-I but not apoA-II (LpA-I) using sulfhydryl covalent chromatography ²⁵. Briefly, reduced HDL was passed down a sulfhydryl resin. The free disulfides at Cys 6 in human apoA-II interact with the column to retain any particle containing apoA-II (or other disulfide-containing proteins, i.e. apoJ) while non Cys-containing particles pass through. The products, defined as LpA-I_2b – LpA-I_3c, were analyzed by SDS PAGE in Fig. 1a. The apoA-II content of the LpA-I fractions was reduced by 70-80% compared to the original HDL sample. The separation was most effective in the HDL_3c, HDL_3b and HDL_2b fractions but moderately less so in the HDL_3a and HDL_2a fractions where we typically find the most apoA-II ²⁶. Nevertheless, apoA-I represented some 80-90% of the coomassie stainable protein in the LpA-I particles. We recognize that these particles also contained trace levels of other proteins, the majority of which likely include the apoC's ^{27, 28}.

Once separated, we measured the LpA-I particle sizes on a calibrated gel filtration column (Fig. 1b) and by native gel electrophoresis (Fig. 1c). The diameter estimates from both techniques were consistent and ranged from 11.2 to 8.8 nm (Table 1). Native PAGE indicated that the LpA-I_2a fraction contained two major populations, consistent with the non-Gaussian peak shape by gel filtration. The LpA-I particle diameters were similar to the corresponding HDL particles, although these had evidence of at least two populations in all subfractions, consistent with the removal of apoA-II containing particles (Supplementary Fig. 1a). Negative stain electron microscopy (EM) showed that all particles were spheres and confirmed the size trends in Fig. 1 (Supplementary Fig. 1b). However, the measured EM diameters were notably smaller than those from gel filtration and native PAGE. This could be due difficulties in resolving particle edges at this size ²⁹ or deformation of particles during drying ²³. Given that gel filtration and native PAGE offer hydrated diameters that are likely more physiologically relevant than EM, we averaged the diameters from these two techniques for the model building described below. Some albumin contamination was apparent in the LpA-I_3c fraction in Fig. 1a, as a shoulder eluting at 15 ml in the gel filtration trace (Fig. 1b) and as a separate band in the native gel (Fig. 1c). The latter result suggests that albumin is probably not associated with the HDL particles.

LpA-I compositional analysis

We next characterized the particle chemical compositions (Table 1). Supplementary Fig. 2 shows that the predicted particle densities, calculated from the weight percentages of the various components, correlated tightly with experimentally determined particle densities. To estimate the average number of apoA-I molecules per particle we cross-linked the samples with the soluble homobifunctional cross-linker BS³. Supplementary Fig. 3 compares the susceptibility of LpA-I_3a, LpA-I_2b and a control reconstituted discoidal particle to cross-linking as a function of increasing cross-linker to protein ratio. All particles responded identically to the cross-linker, indicating that the extent of intermolecular contacts were similar between all three particles. The cross-linked LpA-I particles were analyzed by SDS-PAGE (Fig. 2a). All subfractions generated a diffuse band that decreased in size only slightly between LpA-I_2b and LpA-I_3c. The diffuse bands are typical of cross-linked proteins as the reagent adds variable masses and the point(s) at which cross-linking occurs can affect migration on SDS gels ¹¹. By measuring the smallest and largest band boundaries, we estimated that LpA-I_2b contains 5-7 molecules of apoA-I, whereas LpA-I_2a through LpA-I_3c contain 4-6, 4-6, 3-5, and 3-5 apoA-I molecules, respectively.

Before building structural models, it is critical to determine the average molar stoichiometry of protein and lipid components within each species. Knowing the molar ratio of each lipid component with respect to apoA-I (calculated from Table 1), we summed the partial specific volumes for each component to derive a predicted diameter for each particle (see Methods). This was calculated for various numbers of apoA-I molecules per particle and then compared to the experimentally measured hydrodynamic diameter averaged from native PAGE and gel filtration in Table 1. An example calculation is shown in Fig. 2b for the LpA-I_2b particle. As the putative number of apoA-I molecules was increased, the predicted volume (and hence diameter) of the particle increased. At 5 apoA-I molecules per particle, the theoretical diameter matched closely with the experimentally determined diameter for LpA-I_2b (dotted line). This also fell within the range of expected apoA-I molecules determined by cross-linking (Fig. 2b, bracket). Performing this calculation for all LpA-I particles, we were surprised that LpA-I_2a-3c all averaged 4 molecules per particle. The concordance between calculated and experimental particle diameters are shown in Supplementary Fig. 4. This suggests that a transition from a diameter of 11.2 nm to 9.8 nm involves the loss of one molecule of apoA-I, on average. However changes in diameter from 9.8 to 8.8 nm can all be accommodated with four molecules remaining on the particle. We point out that these determinations represent averages for each subfraction and do not preclude the possibility that the LpA-I_3c fraction, for example, contains some particles with only three or even two molecules of apoA-I per particle. Based on these calculations, Supplementary Table 1 shows the number of molecules of each major component of the particles studied. It should be noted that our previous estimates of apoA-I molar content in native HDL particles were slightly lower and range from 2.9 in HDL_3c to 4.3 in HDL_2b ³⁰, likely due to the removal of apoA-I-poorer LpA-I:A-II particles by covalent chromatography.

We also analyzed the contributions of lipid and protein to the total surface area of the particles. Using data from surface balance experiments on HDL lipids indicating molecular surface areas of 65 Ų per phospholipid molecule ³¹ and 40 Ų per cholesterol ³¹, total surface lipids accounted for only 33, 26, 23, 18, and 12% of the total surface area of LpA-I_2b through LpA-I_3c, respectively. They accounted for even less area when values from molecular dynamics simulations were assumed (27, 21, 19, 15 and 10%, based on 55 Ų for phospholipid and 27 Ų for cholesterol ³²). Thus, by any measure, the outer shell of these particles is dominated by apoA-I.

Cross-linking and mass spectrometry

Each subfraction was cross-linked, delipidated, subjected to exhaustive trypsin digestion, and then analyzed by ESI-MS. Table 2 summarizes all MS/MS verifiable cross-links found in three separate experiments (see Methods). We identified 39 cross-links among the 5 subfractions. Of these, 25 cross-links had been identified in our previous studies of both discoidal and spherical reconstituted HDL particles. In addition, we identified a set of 14 cross-links that were not found in reconstituted HDL particles. Surprisingly, most cross-links were distributed equally across the 5 subfractions, indicating that apoA-I contacts do not vary substantially among them, despite differences in particle diameter.

Since the gel filtration analysis in Fig. 1b shows considerable size overlap among the LpA-I species, even between LpA-I_2b and LpA-I_3c, we were concerned that the similarity in cross-linking patterns may have occurred because of particle cross-contamination. To test this, we isolated LpA-I_2b and LpA-I_3b fractions by ultracentrifugation and disulfide chromatography as described above and then subjected the fractions to high-resolution gel filtration chromatography using three Superdex 200 columns connected in series ³³. LpA-I_3b was selected rather than LpA-I_3c to avoid albumin contamination (Fig. 1). We then isolated the largest fractions of LpA-I_2b and the smallest fractions of LpA-I_3b. Supplementary Fig. 5a shows a native PAGE analysis of these fractions indicating no size overlap between the samples. Cross-linking of both samples again showed a nearly identical pattern consistent with Table 2 (Supplementary Fig. 5b). Therefore, our data strongly indicate that apoA-I molecular contacts do not vary substantially between the smallest/densest and the largest/lightest HDL particles in human plasma.

ApoA-I conformation in LpA-I particles

We next evaluated apoA-I structure in the subfractions by three independent methods. First, we used circular dichroism (CD) spectroscopy. Ordinarily, one would not attempt this on physiologically isolated HDL particles because the presence of diverse apolipoproteins would provide only an averaged readout. However, given the dominance of apoA-I in these separated particles (Fig. 1), we reasoned that CD should provide a suitable estimate of its secondary structure. The CD spectral shapes (Fig. 3a) were indicative of highly helical proteins with characteristic minima at 208 and 222 nm. LpA-I_2b exhibited a helical content of 76%, consistent with measurements of reconstituted particles ²¹. LpA-I_2b through LpA-I_3a showed no difference in spectral shape indicating similar secondary structures. However, LpA-I_3b and LpA-I_3c exhibited reduced helical content, suggesting that certain apoA-I helical domains are unfolded in the smaller particles. We also evaluated the particles by susceptibility to limited proteolysis (Fig. 3b). ApoA-I molecules on LpA-I_2b, _2a, and _3a were relatively resistant to trypsin under these conditions but were substantially more susceptible in the two densest fractions. Finally, we measured the binding of a monoclonal antibody (A-115) that is specific for the central domain of apoA-I (residues 115-126) ³⁴. Supplementary Fig. 6 shows that the association of this antibody was minimal in the HDL_2b through HDL_3a particles, but increased dramatically in HDL_3c and trended upwards in HDL_3b. Taken together, the CD, proteolysis and antibody binding studies indicate that apoA-I conformation is similar in the LpA-I_2b, LpA-I_2a and LpA-I_3a particles, but localized conformational differences likely exist in LpA-I_3b and LpA-I_3c.

Evaluation of the trefoil model in LpA-I

We next determined the plausibility of all cross-links in Table 2 with respect to the trefoil model of apoA-I in spherical HDL ²¹ (see Introduction). Interestingly, of the 39 identified cross-links, 32 were judged to be consistent with the 5/5 (or 5/2) forms of the double belt/trefoil models, i.e. the side chain - NH₂ groups of the involved lysines could be within the 11.4 Å spacer arm of the cross-linker ¹⁰. These include the critical long range cross-links such as Lys118-Lys140, Lys59-Lys208 and Lys77-Lys195 that distinguish the antiparallel belt model ⁷. The 5/5 trefoil model is shown in Fig. 4 with applicable cross-links from Table 2 shown in red, illustrating the concordance of the model with the data.

To further refine how the base trefoil model might apply to the different sized LpA-I species, we began by considering the largest particle LpA-I_2b. In its fully extended form, the trefoil model defines a spherical particle with a diameter of about 10.8 nm, reasonably close to the experimentally determined diameter of LpA-I_2b of 11.2 nm. However, LpA-I_2b contains, on average, 5 apoA-I molecules vs. three in the trefoil. This discrepancy can be resolved by intercalating two additional apoA-I molecules with a corresponding decrease in the apoA-I bend angles so that each creates a 72° biangle vs. 120° for the trefoil (Fig. 5a,b). This “pentafoil” markedly increases the percentage of the particle surface area occupied by protein, but the particle diameter remains consistent with LpA-I_2b. Importantly, all apoA-I molecules make similar intermolecular contacts, consistent with the cross-linking data. It should be noted that the original trefoil model was built without the N-terminal 43 amino acids of apoA-I. Thus, the diameter predicted by the trefoil is likely smaller than that generated by full-length apoA-I and may account for the small difference in size between the trefoil prediction and the measured LpA-I_2b diameter. We also point out that models related to the double belt such as the ‘belt and buckle’ model ⁶ (in which the N-terminus lays back across the belts) is consistent with this arrangement. Indeed several of the cross-links in Table 2 (see footnotes) support this idea.

In the case of LpA-I_2a, only one molecule needs to be added to the trefoil. However, the predicted diameter is larger (10.8 nm) than the experimentally measured 9.8 nm (Fig. 5c). We can envision two general ways in which the smaller diameter can be accommodated. First, regions of the resident apoA-I molecules could fold off the particle surface to form a hinge domain that reduces the distance between the particle poles ^35,36. However, one expects that these exposed sites should be hypersensitive to proteolysis. Since the experiments in Fig. 3 showed no remarkable differences in proteolytic sensitivity or secondary structure between LpA-I_2b and LpA-I_2a (or LpA-I_3a), we favor a second explanation for varying particle size that involves the twisting of apoA-I molecules on the particle surface (Fig. 5d). One can imagine turning the poles of the particle in opposite directions by an angle Θ, shrinking the particle diameter. Using a mathematical modeling program, we calculated the degree of twist required to account for the diameters of LpA-I_2a through LpA-I_3c (Fig. 6). By inducing moderate degrees of twist we could easily account for the observed experimental diameters of all LpA-I particles while maintaining the required number of apoA-I molecules. This predicts that the LpA-I_3c particle surface is composed almost entirely of protein with only room for a small amount of surface lipids in each biangle, consistent with the relative contributions of protein and PL surface areas measured experimentally. Importantly, this twisting should not result in major changes in the intermolecular registry of apoA-I molecules, consistent with the similarities in cross-linking patterns among the LpA-I species.

Implications of the new models for HDL functionality

The models shown in Fig. 6 clearly lack enough resolution to allow speculation of how apoA-I mediates interactions with other proteins. However, our demonstration that apoA-I overwhelmingly dominates the surface of these HDL particles may provide clues to how other HDL-associated proteins interact with the particle. HDL has been shown to contain some 35-50 minor proteins in addition to apoA-I and apoA-II ⁴² and we recently showed that a majority of these are associated with the smallest ²⁸ or most dense HDL_3c fractions ²⁷. The presumption has been that these proteins associate with the phospholipid surface to coexist with apoA-I. However, it is clear from Fig. 6 and our composition calculations that about 85% of the LpA-I_3c particle surface is covered by apoA-I. Thus, it is difficult to imagine how other proteins can find room to bind. One implication of this is that additional proteins may associate directly with apoA-I and not the lipid surface, or they may form their own separate particles which happen to co-purify in this density range.

Finally, our limited proteolysis, CD and antibody binding data indicate that, although all subfractions exhibit a similar global structure, the smallest/densest particles have localized structural differences vs. the larger/lighter ones. Under high degrees of apoA-I twist, localized areas of apoA-I may be envisioned to unfold without necessarily leaving the particle surface. This concept can be likened to twisting a rubber band until local areas begin to distend. A similar idea was previously proposed by Gu et. al in discoidal particles studied by molecular dynamics simulated annealing ²³. They hypothesized that two localized regions, one near the N-terminus and the other in helix 8 of apoA-I, lose helicity when discs transition between the planar and saddle-shape morphologies. Our data may suggest that a similar transition occurs in physiological spherical particles at the LpA-I_3a and LpA-I_3b boundary. Whether these conformational changes occur at the same locations in physiological spherical HDL and how they might affect HDL metabolism awaits further study.

In summary, we have provided the most comprehensive set of distance constraints for native human plasma HDL particles achieved to date. The results unambiguously show that apoA-I contacts found in model reconstituted particles occur in authentic human plasma HDL particles. We extended models originally derived in reconstituted particles to these native species. Our favored scheme, based on our proposed trefoil arrangement, is a parsimonious solution that is consistent with the majority of experimental, geometric and molecular simulation data from this study and others in which changes in the particle neutral lipid core are mediated by twisting motions of the surface apoA-I molecules. It remains to be seen how these changes affect important physiological interactions with plasma lipid remodeling enzymes and cell surface receptors responsible for HDL metabolism and function.

HDL subfractionation and LpA-I isolation

Normolipidemic human EDTA plasma was subjected to a one-step gradient density ultracentrifugation procedure detailed by Chapman et al. ⁴³. Lipoprotein fractions (VLDL, LDL and HDL) of 0.8 ml each were collected from top to bottom of the tube and the fractions corresponding to the five HDL subclasses ²⁶: HDL_2b (d 1.063 to 1.090 g ml^-1), HDL_2a (d 1.090 to 1.120 g ml^-1), HDL_3a (d 1.120 to 1.150 g ml^-1), HDL_3b (d 1.150 to 1.180 g ml^-1) and HDL_3c (d 1.180 to 1.210 g ml^-1) from multiple centrifuge tubes were pooled, concentrated by ultrafiltration and dialyzed into standard Tris buffer (STB) for further LpA-I HDL isolation. These density ranges were slightly different from previous reports ⁴⁴ and likely represent slight intralab variations between the French and U.S. laboratories. Three individual human plasma preparations were used for the study (n=3).

To isolate LpA-I from LpA-I/A-II HDL density subfractions, we used the sulfhydryl covalent chromatography technique developed by Rosales et al ²⁵ with modifications (details are in Supplementary Methods).

Particle composition and characterization

The protein was determined by Markwell modified Lowry assay ⁴⁵, the phospholipid (PL), total cholesterol (TC), free cholesterol (FC) and triglyceride (TG) contents were determined using enzymatic assay kits from Wako Diagnostics (Richmond, VA). Cholesterol ester (CE) concentrations were calculated by subtraction of FC concentrations from TC concentration and then multiplied by 1.67 (MW of CE divided by MW of FC). Particle size distribution was assessed by gel filtration using a calibrated Superdex 200 column (Amersham). Particle Stokes diameters were measured on cross-linked particles by native PAGE (Phast system, Amersham Pharmacia). Negative stain electron microscopy (EM) (Supplementary Methods) was also performed on LpA-I subfractions as independent approach to assess particle sizes.

Circular dichroism, limited proteolysis and surface plasmon resonance with monoclonal antibody experiments are described in Supplementary Methods.

Cross-linking and MS measurements and data analysis

LpA-I subfractions (1 mg ml^-1) dialyzed into phosphate buffered saline (PBS) were cross-linked with a homo-functional cross-linker, bis(sulfosuccinimidyl) suberate (BS³) at a protein to BS³ molar ratio of 1:50 that was optimized for the maximum formation of the high-order oligomers (Supplementary Fig. 3). We carried out the cross-linking reaction, chloroform/methanol delipidation and trypsin digestion for MS analysis using our optimized protocols (Supplementary Methods).

MS measurements of cross-linked particles were performed on a Sciex/Applied Biosystems QSTAR XL coupled with an Agilent on-line capillary HPLC (Supplementary Methods). The MS data analysis was carried out using the Mascot Script in the instrument software and home built software using our previously described criteria for cross-links (Supplementary Methods).

Building the twisted particle models

The models in Fig. 6 were produced using the Mathematica 7 software package. In each case, the apoA-I proteins are represented by a pair of like-colored tubes lying on the surface of a sphere. The leftmost panel shows 5 molecules of apoA-I lying in a symmetric pentafoil arrangement on the LpA-I_2b particle. The molecules have bend angles slightly less than 72° (the reduction being necessary to enable separation of the molecules from each other). The remaining panels represent twisted quatrefoil arrangements of four apoA-I molecules on the LpA-I_2a, LpA-I_3a, LpA-I_3b and LpA-I_3c particles. The strands have been twisted in such a way that the twist at the poles is ± Θ and the intermediary twisting is a linear function of distance along the polar axis. Thus, for instance, the axis of one strand is given by the vector function

The path of the axes of the remaining strands is obtained by rotating this curve around the polar axis. The twist Θ is determined by requiring that the strands retain a constant length. The length of such a strand as a function of the twist angle is given by

In the models, the helical diameter of the apoA-I proteins is assumed to be 11 Å. Supplementary Table 2 shows the total particle diameter, A; the radius R of the sphere on which the axes of the apoA-I helices lie; and the twist angle Θ required for the proteins to fit on the particle in this fashion.

Particle volume, diameter and surface area calculations

We calculated the particle volume from the experimentally determined molar ratio of various lipid components and derived particle diameter from the calculated particle volume. Particle surface area calculations were taken from compositional data for phospholipids and free cholesterol. Details are described in Supplementary Methods.