If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Department of Cellular and Molecular Medicine, Cleveland Clinic, Cleveland, OHCenter for Cardiovascular Diagnostics and Prevention, and, Cleveland Clinic, Cleveland, OHDepartment of Chemistry, Cleveland State University, Cleveland, OH
Unit for Virus Host Cell Interactions, UMI 3265 (CNRS-EMBL-UJF), BP 181, F38042 Grenoble FranceCEA, CNRS, and Université Joseph Fourier, Institut de Biologie Structurale Jean-Pierre Ebel, UMR5075 Grenoble, France
Department of Cellular and Molecular Medicine, Cleveland Clinic, Cleveland, OHCenter for Cardiovascular Diagnostics and Prevention, and, Cleveland Clinic, Cleveland, OHDepartment of Cardiovascular Medicine, Cleveland Clinic, Cleveland, OH
Small-angle neutron scattering (SANS) with contrast variation was used to obtain the low-resolution structure of nascent HDL (nHDL) reconstituted with dimyristoyl phosphatidylcholine (DMPC) in the absence and presence of cholesterol, [apoA1:DMPC (1:80, mol:mol) and apoA1:DMPC:cholesterol (1:86:9, mol:mol:mol)]. The overall shape of both particles is discoidal with the low-resolution structure of apoA1 visualized as an open, contorted, and out of plane conformation with three arms in nascent HDL/dimyristoyl phosphatidylcholine without cholesterol (nHDLDMPC) and two arms in nascent HDL/dimyristoyl phosphatidylcholine with cholesterol (nHDLDMPC+Chol). The low-resolution shape of the lipid phase in both nHDLDMPC and nHDLDMPC+Chol were oblate ellipsoids, and fit well within their respective protein shapes. Modeling studies indicate that apoA1 is folded onto itself in nHDLDMPC, making a large hairpin, which was also confirmed independently by both cross-linking mass spectrometry and hydrogen-deuterium exchange (HDX) mass spectrometry analyses. In nHDLDMPC+Chol, the lipid was expanded and no hairpin was visible. Importantly, despite the overall discoidal shape of the whole particle in both nHDLDMPC and nHDLDMPC+Chol, an open conformation (i.e., not a closed belt) of apoA1 is observed. Collectively, these data show that full length apoA1 retains an open architecture that is dictated by its lipid cargo. The lipid is likely predominantly organized as a bilayer with a micelle domain between the open apoA1 arms. The apoA1 configuration observed suggests a mechanism for accommodating changing lipid cargo by quantized expansion of hairpin structures.
Cholesterol from peripheral tissues, such as within the artery wall, is transported to the liver by high density lipoprotein (HDL) particles, which are continuously remodeled in a process called reverse cholesterol transport (
). The amphipathic apoA1 chains in all models wrap around a central lipid core, and the orientation of amino acid residues along the apoA1 chains position the hydrophobic side chains toward the lipid surface. Before studies employing contrast variation small-angle neutron scattering (SANS) (see below), the overall conformation of apoA1 within a lipid containing HDL particles was inferred from crystal structure data from lipid free apoA1 mutant forms (
The recent application of contrast variation SANS to the structural interrogation of nHDL and spherical HDL particles has allowed for the first time direct visualization of the low-resolution shape envelope of the protein versus the lipid components of HDL (
). The structures revealed allowed for the possibility of developing models that incorporated the SANS shape constraints, giving a better understanding of how apoA1 and lipids within different forms of HDL are structurally organized. Among all nHDL models, only the Double Super Helix model (
) incorporated direct SANS imaging data of the low-resolution structures of apoA1 and the lipid domain of a recombinant biologically functional nHDL preparation reconstituted with 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) (
)], and thus has been a focus of structural studies. While our initial SANS studies of nHDL focused on use of full length apoA1 and the longer chain and more physiologically relevant molecular species of PC (POPC), because of the dramatic differences in overall particle organization observed with the double super helix (DSH) model versus the earlier discoidal models based on computer simulations, we felt it reasonable to directly interrogate the structure of “DMPC discs” in nascent HDL/dimyristoyl phosphatidylcholine without cholesterol (nHDLDMPC).
In a contrast variation SANS experiment the concentration of D2O in buffer is varied such that the scattering length density of the buffer matches that of a component within the particle (i.e., protein, lipid, or nucleic acid), making that component “transparent”. In addition, by selectively deuterating specific components of a macromolecular assembly (in our case, apoA1, by producing the recombinant protein in Escherichia coli grown in D2O with deuterated nutrients), one can further enhance the contrast between different components within the complex and reduce the interference of the other component (the lipid in this case). The contrast variation experiment thus enables one to “triangulate” different components of a macromolecular complex, making SANS a powerful and unique structural interrogation tool (
Herein we employ contrast variation SANS to directly visualize the individual low-resolution protein and lipid conformations within nHDL reconstituted using apoA1:DMPC (1:80) (nHDLDMPC), as initially employed for the discoidal HDL model (
), and apoA1:DMPC:cholesterol (1:80:10) [nascent HDL/dimyristoyl phosphatidylcholine with cholesterol (nHDLDMPC+Chol)]. Our studies reveal, remarkably, that despite overall “discoidal” particle shapes, apoA1 in both particles is not organized in a circumferential belt conformation, but instead, is open and slightly out of plane. In nHDLDMPC three arms of apoA1 enshrine a central oblate lipid core. One of the arms is a large hairpin, a configuration confirmed by cross-linking mass spectrometry, and further corroborated by H/D kinetic exchange analyses consistent with a large hairpin. On the other hand, in nHDLDMPC+Chol where the lipid core is 20% expanded, apoA1 is closer in conformation to a traditional double belt conformation, enshrining a less symmetric lipid phase, but nonetheless with open apoA1 conformation. The open apoA1 conformation observed coupled with the dimensions observed for the lipid core of both particles reveals a mechanism accounting for quantized particle size expansion frequently observed during nHDL formation, as well as suggests a lipid organization within the particle that possesses a combination of both lamellar (bilayer) and micellar configuration.
MATERIALS AND METHODS
Preparation of nHDLDMPC with and without cholesterol
The particle composition selected for our studies and the methods for particle generation and characterization are the same as those previously published (
). Human apoA1 was isolated from HDL provided by CSL Limited (Victoria, Australia), as well as from plasma obtained from healthy volunteers who gave written informed consent under a protocol approved by the Cleveland Clinic Institutional Review Board. ApoA1 was purified from HDL isolated by ultracentrifugation (density range of 1.07–1.21 g/ml). Lipid-free human apoA1 was recovered by delipidation of the HDL followed by ion exchange chromatography, as previously described (
. The His tag of isolated recombinant apoA1 forms was removed by formic acid cleavage, and subsequent cleanup was done by nickel affinity chromatography and a polishing ion exchange chromatography step (
). SDS-PAGE was used to assess the purity of isolated human and recombinant apoA1.
Because initial efforts to prepare nHDL preparations by DMPC-driven spontaneous particle formation led to inhomogeneous preparations, reconstituted nHDLDMPC with and without cholesterol was prepared using a modified sodium cholate dialysis method (
) at an initial molar ratio of 80:1 of DMPC:apoA1 or 10:80:1 of cholesterol:DMPC:apoA1. Subsequently, the HDL particles were purified by gel filtration chromatography using Sephacryl S300 (GE Healthcare) column. The size of reconstituted nHDL preparations were measured by using dynamic light scattering, and nondenaturing PAGE (
). The stoichiometry of each of the nHDL preparations was determined by quantifying apoA1 phospholipid content using the microphosphorous assay, and cholesterol content using stable isotope dilution mass spectrometry (
). The remaining sodium cholate in reconstituted nHDL preparations was also quantified by HPLC tandem mass spectrometry, which confirmed that no significant residual cholate was present (i.e., cholate content << 0.1 mol%). BS3 was used as a chemical cross-linker at a molar ratio of 200:1 versus apoA1 for 5 min at 37°C, as previously described (
Small-angle neutron scattering experiments were carried out at the instrument D22 of the Institut Laue-Langevin, Grenoble, France, and KWS1 at Jülich Center for Neutron Sciences, Garching, Germany. D22 and KWS1 are classical pinhole cameras that provide a high neutron flux. Data were collected for ∼2 h from two positions of the detector at the D22 instrument (2 and 5.6 m with collimation lengths of 2.8 and 5.6 m), and three positions of the detector at the KWS1 instrument (1, 2, and 8 m with collimation lengths of 4 and 8 m), and covered a range of momentum transfer (q) from 0.008 Å−1 to 0.5 Å−1. The wavelength λ of the neutron beam used was 6 Å on D22 and 4.5 Å on KWS1. SANS with contrast variation was used to individually map the protein and lipid components in nHDLDMPC. The samples were measured at 6–8°C in 12% (q range 0.005–0.304 Å−1) and 42% D2O solution (q range 0.007–0.453 Å−1). The radius of gyration (Rg) was obtained based on the Guinier approximation (
) was used to produce low-resolution models (n = 32) of the protein component from the P(R) function obtained for the sample measured in 12% D2O (q range 0.005–0.256 Å−1). The P(R) function obtained from the 42% D2O scattering intensity curve (q range 0.007–0.453 Å−1) was similarly used to obtain low-resolution structures (n = 16) of the lipid component. For some experiments, protein shape was visualized in samples measured at 12% contrast (12% D2O). For these studies, the scattering intensity of the protein was enhanced by preparing nHDL with deuterated apoA1. Both SANS at multiple levels of contrast and mass spectrometry studies independently confirmed that 70% of the nonexchangeable H atoms were substituted by D atoms during protein expression in D2O with deuterated nutrients, which corresponds to a level of contrast of 92% D2O. Using this information, calculations indicate that, under the conditions employed, the lipid phase (both head groups and acyl chains) contributed <7% of the intensity across all scattering angles examined in the 12% contrast experiment (supplementary Fig. I). We established this value by calculating contributions to intensity coming from the protein and lipid using a modification of the SASSIM program (
). The modification included splitting the total squared scattering amplitude into contributions assigned to various components of HDL (i.e., protein, PC head groups, and PC acyl chains). The contributions to intensity can be either positive or negative, a fact that reflects scattering interference between components. Similar calculations in the 42% contrast experiment, where HDLDMPC is formed with nondeuterated apoA1, indicate the protein component contributes <7% to the total intensity across all scattering angles examined.
Cross-linking mass spectrometry experiments on reconstituted nHDLDMPC and nHDLDMPC+Chol particles
Chemical cross-linking of apoA1 within nHDLDMPC and nHDLDMPC+Chol was performed by dissolving bis(sulfosuccinimidyl) suberate (BS3) cross-linker (Pierce, Rockford, IL) at 5 mg/ml in 20 mM sodium phosphate buffer, pH 7.4 and immediately adding to 1 mg/ml nHDL in PBS, pH 7.4 at a molar ratio of 200:1. The resulting solution was incubated at 4°C for 24 h with constant gentle mixing. The reaction was quenched by the addition of Tris at a final concentration of 50 mM for 15 min. The resulting cross-linked apoA1 solutions were split into aliquots, digested with sequencing grade chymotrypsin (Promega, Madison, WI) at a ratio of 1:100 (w/w) for 16 h at 25°C, and then stored at 4°C. Cross-linked and chymotrypsin-digested peptides were concentrated with strong-cation exchange (SCX) chromatography using HIL-SCX.25 UltraMicro Tip Columns (Nest Group, Southborough, MA). The samples were prepared for SCX chromatography by adding acetonitrile (5% v/v final) to the cross-linked peptide solution and adjusting to pH 3.0 with H3PO4 prior to being loaded onto the columns. The column was washed with PBS, pH 3.0 and eluted with 5% acetonitrile, 20 mM sodium phosphate buffer, pH 3.0 with steps of 300 and 500 mM NaCl. A FASTA database of potential cross-linked peptides of apoA1 was generated using the program xComb v1.3 (
). The Uniprot FASTA for apoA1 was used as the input file and the parameters for the database generation included two potential missed cleavages, both intra- and interprotein cross-links, amine to amine cross-linker, no missed cleavage necessary, and a minimum peptide length of four amino acids. Concentrated cross-linked peptides were loaded onto an IntegraFrit sample trap (ProteoPep C18, 300 Å, 150 μm × 2.5 cm; New Objective, Woburn, MA) with 5% acetonitrile in 0.1% formic acid in order to desalt the samples. The peptides were subsequently eluted through a PicroFrit column (75 μm × 15 cm; New Objective) packed in-house with XperTek 218TP, C18, 300 Å pore size, 150 μm particle size (Cobert Associates, St. Louis, MO) using a Proxeon Easy-nLC II system (Thermo Scientific, Waltham, MA) with a gradient of 5–65% acetonitrile, 0.1% formic acid over 120 min into an Orbitrap Velos mass spectrometer (Thermo Scientific). The initial MS screening experiment involved programming the Orbitrap MS to collect in a data-dependent mode and analyzing the data using the xComb FASTA database in Proteome Discoverer 1.1 (Thermo Scientific) to find precursor masses of potential cross-links. Observed monoisotopic masses within 5 ppm of the theoretical mass of cross-linked peptides of interest were collated into a “parent list” and the Orbitrap MS was programmed to specifically search for the peptides in this list and perform CID MS/MS on the specific peptides. The subsequent MS/MS analyses using the parent list utilized the same chromatographic method as detailed above in the screening experiment. The fragmentation patterns of the resulting data were inspected and assigned manually using Protein Prospector with a signal tolerance of 0.5 Da for the fragment ions (
Hydrogen-deuterium exchange mass spectrometry experiments on reconstituted nHDLDMPC particles
ApoA1 in nHDLDMPC was subjected to hydrogen-deuterium exchange (HDX) at 4°C and at a pH of 7.4 in PBS for various time intervals: 0.5, 1, 2, 75, and 167 min. The reaction was quenched by acidification and by reducing the temperature to 0°C in an ice/water bath. Proteins were subjected to rapid proteolytic digestion using pepsin and immediately injected into an LC-MS system consisting of binary LC20 UPLC pumps (Shimadzu, Kyoto, Japan) and an LTQ ion-trap mass spectrometer (Thermo Scientific). The gradient for elution of the peptides started at 200 μl/min of 0.5% formic acid for 1 min to flush the PBS from the system and then stepped down to 50 μl/min of 10% acetonitrile, 0.5% formic acid and continued to 50% acetonitrile, 0.5% formic acid over 10 min. The trap column consisted of an Everest 238 CV C18 guard column (2.1 mm diameter, Grace, Deerfield, IL) and the analytical column was a Vydac 238TP C18 guard column (2.1 mm diameter, Grace). The resulting spectra were manually interpreted.
) we matched computational models of HDL at multiple contrast levels (0, 12, 42, and 90% D2O) and established that two contrast levels (12 and 42% D2O) are sufficient to produce a consistent model for the entire particle. Thus, thirty-two low-resolution structures obtained from the 12% contrast SANS dataset were averaged, and the SANS shape closest in features to the average shape (reference shape) was further used for modeling. The all-atom model of nHDLDMPC was constructed by separately fitting an all α-helix apoA1 dimer into the reference low-resolution shape envelope, and by packing 160 DMPC molecules into the low-resolution structure of the lipid component (42% contrast SANS dataset, that was similarly averaged, and reference shape used for modeling). Additional constraints were imposed on the lipid packing in the sense that care was taken that lipids from the periphery of the lipid phase that face the apoA1 dimer were arranged such that their acyl chains faced the hydrophobic surface of the protein. Thus, most of lipids were arranged as a bilayer with the exception of about one quarter of them located in between the N/C-termini that form a half-micelle, which connects the two leaves of the bilayer.
The partitioning of the hydrophobic/hydrophilic surface of the apoA1 dimer in nHDLDMPC was determined using “patch” hydrophobicity, which identifies an average per residue hydrophobic index [range (0,1)] calculated by averaging the hydrophobic indices of neighboring amino acid residues located on the same side of the protein (i.e., either facing the solvent or the lipid) in a certain cutoff distance (here the cutoff distance used was 15 Å). The resulting model was further refined by gradually adjusting the conformation of apoA1 chains and lipids until the scattering intensities calculated with CRYSON (
) matched the experimental data. Atom clashes in the all-atom model, which resulted from manual adjustment of chain residues and individual lipid molecules, were removed by performing a short geometry optimization in vacuum. The CHARMM22 force field was used for protein (
), with walls 15 Å away from the particle. Sodium ions were added to compensate the negative charge of the two apoA1 chains, and then NaCl ion pairs were inserted to obtain a physiological salt concentration of 344 milliosmolar (standard PBS buffer). The solvated system was energy minimized [using the program GROMACS (
)]. The protocol for geometry optimization was as follows: the lipoprotein particle relaxed gradually (first the solvent, then the lipid, and finally the protein). In the last step of energy minimization all atoms relaxed. Potential energy calculations took into account periodic boundary conditions for the solvated particle. The evaluation of short-range (van der Waals) interactions used a 10 Å cutoff, while Coulomb interactions were calculated using the Ewald summation [Particle Mesh Ewald summation (
SANS decomposition analysis of the lipid core was carried out after splitting the lipid phase into lamellar and micellar domains by visual inspection. The scattering intensities for the whole range of scattering angles were calculated for each individual lipid subdomain. The individual scattering intensity curves were scaled such that the sum of the intensities at zero scattering angle, coming from the two lipid subdomains, equal the intensity at zero scattering angle given off by the entire lipid phase. The scaling procedure is justified by the fact that the intensity at zero scattering angle is proportional to the molecular mass of the lipid (i.e., with the number of lipids in the phase) (
). Still, the decomposition scheme is capable of giving insight into the scattering contributions of different subdomains of the lipid phase that are distinguished by different packing characteristics (lamellar vs. micellar).
Electron microscopy studies
Negative stained nHDL samples (dilution 1:300) were obtained by applying them to the clear side of the carbon-mica interface and staining with 2% (w/v) uranyl acetate. Images were recorded under low-dose conditions with a CM12 (Philips) electron microscope equipped with an LaB6 filament and working at 120 kV. Images have been recorded on an Orius SC1000 CCD camera (Gatan Inc.).
The experimental SANS intensities, the Protein Data Bank files of the SANS shapes for the protein and lipid components of nHDL examined, and the all-atom Turtle model of nHDLDMPC can be downloaded from the site http://www.lerner.ccf.org/cellbio/hazen/data/.
RESULTS AND DISCUSSION
Biochemical characterization of nHDL preparations formed using native human apoA1 and perdeuterated human apoA1
Reconstituted nHDL particles generated for SANS analyses in the present study were prepared with either isolated human apoA1 or recombinant human perdeuterated apoA1, as described under Materials and Methods. Before proceeding with SANS analyses of nHDLDMPC and nHDLDMPC+Chol preparations, particles were extensively characterized. SDS-PAGE and native gel analyses of nHDL forms generated using human apoA1 and deuterated apoA1 showed relatively homogeneous particle preparations (Fig. 1A, B). Chemical composition analyses confirmed that the final particle compositions were in close alignment with the initial 80:1, mol:mol (DMPC:apoA1), and 10:80:1, mol:mol:mol (cholesterol:DMPC:apoA1) composition of the DMPC:apoA1 used for particle formation (DMPC:apoA1 of 78.4 ± 2.0:1 and 81.7 ± 3.2:1, mol:mol, for nHDLDMPC generated using human apoA1 and deuterated apoA1, respectively; and cholesterol:DMPC:apoA1 of 8.7 ± 0.2:86 ± 1.7:1, mol:mol:mol, for nHDLDMPC+Chol). As expected, cross-linking studies were consistent with all nHDL particle preparations formed containing an apoA1 dimer (Fig. 1C). These nHDLDMPC and nHDLDMPC+Chol preparations were used for all SANS analyses.
As detailed below, all initial studies focused on nHDLDMPC particles. Negative staining of electron microscopy images of the nHDLDMPC formed from either human apoA1 or deuterated apoA1 are shown in Fig. 1D and Fig. 1E, respectively. The images show relatively homogeneous ellipsoidal particles of similar dimensions whether native or deuterated apoA1 was used in nHDLDMPC formation. Approximate overall dimensions of the nHDLDMPC are also shown (Fig. 1D, E; insets of the micrographs).
SANS studies reveal that the low-resolution structure of apolipoprotein A1 within nHDLDMPC is an open and slightly out of plane shape with three arms
To directly visualize the shape of apoA1 within nHDLDMPC, particles were analyzed by neutron scattering at 12% contrast (12% D2O). Under these conditions the average scattering length density of the lipid component matches the scattering properties of the solvent at zero scattering angle and is thus “invisible”. The selective deuteration of the protein further intensifies the scattering signal from the protein, making it possible to minimize (<7%, supplementary Fig. I) the contribution of the lipid to the intensity at higher scattering angles, thus allowing the use of the low-resolution shape obtained at 12% contrast to serve as a scaffold for constructing a model for the protein conformation. Initial SANS data collected during the 12% contrast variation study of the protein component of nHDLDMPC were used to construct 32 low-resolution structures for apoA1 in nHDLDMPC as described under Materials and Methods. All demonstrated similar essential structural features for apoA1, a shape with three distinct arms. Supplementary Fig. II includes a random sampling of 15 of the 32 low-resolution shapes, along with an average shape calculated from all of the low-resolution shapes. Because the scattering intensities predicted from the average shape do not agree as well with the experimental data (χ2= 5.42) as that calculated for each of the individual low-resolution shapes (χ2 < 0.85 for each), the low-resolution shape that most closely resembled the average shape was selected as the reference shape (supplementary Fig. II), and was then used to build a cartoon of the apoA1 double chain (vide infra), as described under Materials and Methods. The experimental scattering intensities obtained at various scattering angles (Fig. 2A), and the intensities collected at very small scattering angles (Fig. 2B) were used to determine the radius of gyration according to Guinier approximation (
). Deconvolution of the scattering intensities from the 12% contrast SANS experiments reveals a low-resolution structure (resolution 24 Å) of protein (also known as apoA1 dimer) within nHDLDMPC, as illustrated in Fig. 2D. The average thickness of the apoA1 double chain is 22–24 Å. So at the SANS resolution of the protein component we cannot distinguish the individual chains (as actually shown by the SANS shape), but we can resolve the shape of the double chain. To build a molecular model of apoA1 in nHDLDMPC we used information about the composition (number of apoA1 chains per nHDL particle) and alignment of the two chains with respect to each other, that is, the antiparallel orientation and the registry based on multiple published Lys-Lys proximity distances as probed by cross-linking mass spectrometry experiments. Notably, the protein shape is distinct from a classic “belt” configuration. Rather, the protein envelope visualized has three distinct arms, is open, and follows a contorted shape with two arms clearly out of plane. Measured parameters for the protein component of nHDLDMPC include a radius of gyration (Rg) of 52.4 Å, and overall outer dimensions of approximately 174 × 112 × 71 Å (Fig. 2D, Table 1). The scattering intensities calculated from the low-resolution protein envelope nicely overlay the experimental scattering data (Fig. 2A), and show an excellent statistical fit with experimental scattering intensities (root mean square deviation, χ2= 0.82; Table 1).
TABLE 1Radius of gyration, global dimensions, and goodness of fit of the Turtle model of nHDLDMPC with experimental SANS data
Radius of gyration (Rg) and overall dimensions of the low resolution structures (SANS shapes) and the Turtle, Belt, and Saddle models of nHDLDMPC. Calculated χ2 statistics (root mean square difference), that quantify differences between experimentally determined scattering intensities (measured by SANS with contrast variation) and theoretically calculated, show a better fit for the Turtle model compared with the Belt or various Saddle models of apoA1 dimer. EM, electron microscopy.
a Calculated with CRYSON.
b The root mean square difference between the calculated and experimental scattering intensities.
c Calculated with GNOM from experimental scattering intensity using the Guinier approximation.
d Calculated with DAMMIN.
e Particle dimensions of nHDLDMPC from negative staining EM images (see Fig. 1).
f A short energy minimization in vacuum was performed with GROMACS for each component (protein and lipid) individually (see Materials and Methods).
g Full energy minimization in solution was performed with GROMACS for the nHDLDMPC model (see Materials and Methods).
The low-resolution structure of the lipid component of nHDLDMPC has an oblate ellipsoidal shape
To visualize the low-resolution structure of the lipid phase of nHDLDMPC we performed contrast variation SANS studies in 42% D2O on nHDLDMPC particles reconstituted with nondeuterated apoA1. At this D2O concentration, the scattering length density of the solvent matches that of the nondeuterated protein at zero scattering angle; thus, the scattering signal measured arises mostly from the lipid phase of the lipoprotein and at higher scattering angle calculations indicates that the protein contributes <7% to the total intensity (see Materials and Methods). The scattering intensities, the Guinier range, and the distance distribution function P(R) of the lipid component of nHDLDMPC are shown in Fig. 3A–C. Deconvolution of the scattering data yields an oblate ellipsoid. Sixteen shapes were averaged (Materials and Methods, supplementary Fig. III), and a reference shape that best interlocked with the protein shape was selected as reference shape, which had dimensions 88 × 76 × 52 Å (Fig. 3D). The calculated scattering intensities from the low-resolution shape envelope of the nHDLDMPC lipid phase are in excellent agreement with the experimental data (Fig. 3A, Table 1; χ2= 2.29). On a cautionary note, we remark that the SANS studies by themselves lack the resolution to determine details of molecular packing of indivudual lipids within the particle (e.g., lamellar vs. micellar vs. other smectic mesophase), but further decomposition analysis of the intensities may actually provide further understanding of lipid packing in HDL (vide infra).
Overlays of the low-resolution structures of both protein (orange) and lipid (green) components of nHDLDMPC are shown in Fig. 3E. The lipid shape matches the protein shape like a key in a lock, with the overall low-resolution structure of the nHDLDMPC particle resembling a “turtle”, so from now on we will refer to the structural model of nHDLDMPC generated from 80:1, DMPC:apoA1 as the “Turtle model”.
ApoA1 within nHDLDMPC has an open conformation with a large hairpin and N/C-termini folded back
Having the low-resolution structure of the protein component of nHDLDMPC generates a puzzle in how to fit an apoA1 dimer within the observed protein shape. The known crystal structures of lipid-free apoA1 truncated mutants (
) argue that the two apoA1 chains are oriented in an antiparallel fashion. We therefore sought to develop a hypothetical three-dimensional model of the apoA1 antiparallel dimer that fit within the constraints of the low-resolution protein structure (Fig. 4). First, direct visualization of the overall shape envelope suggests only a few global configurations (in the crude sense) of how an apoA1 dimer can be arranged such that its full length fits within the protein shape envelope experimentally visualized. Closer inspection of the three “arms” (Fig. 4Aa–c) of the protein shape envelope reveals that the approximate cross-sectional area of two arms (Fig. 4Aa, c, each ∼120 Å2) are roughly the dimensions predicted by two classic α-helices (∼65 Å2 each, including residue side chains), whereas the third arm (Fig. 4Ab) appears to be substantially (∼2-fold) larger (∼234 Å2). We therefore chose to entertain an apoA1 configuration where the dimeric chains fold upon themselves forming a hairpin that fills arm b, leaving the other two arms to contain the N- and C termini (Fig. 4).
To help build a three-dimensional model of the apoA1 dimer, we traced the protein shape envelope with two tubular pathways, one for each monomer, each tube having the anticipated cross-section of a classic amphipathic α-helix (not including side chains). These “pipes” are gradient colored (dark → light, N terminus → C terminus) in red and blue (for the two apoA1 chains), and are superimposed in an antiparallel orientation onto the low-resolution protein structure of nHDLDMPC in Fig. 4A. Next, we built conformations of the two apoA1 monomers that matched the tube paths and maintained proper orientation of the hydrophobic/hydrophilic face of the amphipathic helical surfaces taking into account the presumed location of the lipid core of the particle. Following energy minimization, the crude proposed three-dimensional model obtained for the apoA1 chains within nHDLDMPC is shown in Fig. 4B (stereo views in supplementary Fig. IV), and its superposition onto the low-resolution structure of apoA1 is given in Fig. 4C. It should be noted that this hypothetical exercise was done not to claim that this was the precise structure of apoA1 within nHDLDMPC, but rather, to help build a model that displayed overall architecture that matched the experimental data collected by the SANS studies, and would then likely also match previously reported MS cross-linking, ESR, and FRET data on reconstituted nHDL particles (
To assess the goodness of fit of the protein component of the Turtle model with the experimentally determined scattering data (12% contrast), predicted scattering intensities were calculated. They showed excellent statistical fit with experimental scattering intensities (root mean square difference, χ2= 0.985; Table 1, Fig. 4D). To assess the orientation of the hydrophobic/hydrophilic residues in the amphipathic protein within the Turtle model, we calculated the patch hydrophobicity [15 Å (
)], since this would better reflect the predicted overall amphipathic partitioning of side chains on the apoA1 dimer. Visual inspection (Fig. 4E; hydrophilic residues orange, hydrophobic residues green) confirmed that the residues in the protein model had orientations in accord with a typical amphipathic peptide, with the green (hydrophobic) surface clearly oriented toward the cavity wherein the lipid core resides. Also notable is that the hydrophobic surfaces of the amphipathic helices abut one another within the hairpin in arm b (Fig. 4E), allowing for protection from the aqueous phase. Of note, many of the hydrophobic residues in the N- and C-terminal regions of apoA1 that are not associated with phospholipids are not predicted to be exposed to aqueous solvent because of being shielded by the protein fold (i.e., hydrophobic-hydrophobic interactions of amphipathic α-helical surfaces) in these regions (Fig. 4E). Further, the large hairpin houses residues (e.g., R149, R153, and R160) on one apoA1 chain previously identified as being important in LCAT activation (
Mass spectrometry cross-linking studies provide independent experimental data confirming the existence of a large hairpin domain of apoA1 in nHDLDMPC
We sought to obtain independent experimental data directly supporting the finding that a significant domain of apoA1 within nHDLDMPC particles is folded onto itself in a large hairpin. Using the Turtle model of nHDLDMPC and a 12 Å distance for the BS3 bifunctional cross-linking agent, we predicted in silico that up to 15 Lys-Lys cross-links (both intrachain and interchain) might exist in the predicted hairpin domain of ApoA1. Nevertheless, none of the postulated cross-links predicted have yet to be identified in other reported cross-linking studies of HDL preparations (N.B. no MS cross-linking studies are reported yet for a 1:80 (mol:mol), apoA1:DMPC HDL particle). Remarkably, using the exact mass of the proposed cross-links (see Materials and Methods for experimental approach), we were able to experimentally identify (using 5 ppm tolerance) all 15 cross-links by high-resolution mass spectrometry, and among these, eight were sequence confirmed by MS/MS (Fig. 5). For illustrative purposes, supplementary Fig. V shows the full scan MS/MS spectrum of cross-link Lys59-Lys133, which along with cross-link Lys133-Lys195, spans the base of the hairpin and affirms the large size of the hairpin (∼50 Å).
HDX tandem mass spectrometry studies provide additional independent confirmatory data in support of a distinct environment for apoA1 regions within the hairpin
Numerous studies support the antiparallel orientation of the two apoA1 chains within nHDL with a defined registry [similar to that observed for the crystal structure of lipid-free apoA1 (
) ]. Indeed, this registry for the two antiparallel apoA1 chains is needed to accommodate the many distance constraints obtained from multiple biophysical approaches including cross-linking mass spectrometry (
) studies. When incorporating the antiparallel apoA1 chains within the Turtle model, care was taken to retain this predicted registry. Closer inspection of the Turtle model suggests that specific regions of apoA1 within the hairpin on one chain would not be predicted to be within the hairpin on the antiparallel apoA1 chain (Fig. 6). We therefore performed HDX kinetic analyses, which are sensitive to the local environment, on three apoA1 peptides in nHDLDMPC particles (Y115-L126, L159-L170, and R171-L178) that are predicted to be located in the hairpin domain of one chain, but not of the other. Thus, we hypothesized that these peptides might demonstrate evidence of two distinct environments, such as the presence of bimodal kinetics when examining the rates of H/D exchange within amide protons in these sequences. The kinetic curves for the three peptides (percent of deuterium incorporation vs. HDX time) are shown in supplementary Fig. VI. Remarkably, each showed readily detectable bimodal kinetics (presence of both a slow and a fast exchanging peptide) for deuterium buildup rates, consistent with the presence of at least two distinct environments for these apoA1 sequences within the nHDLDMPC particles. The slow-exchanging peptides had rate constants approximately two orders of magnitude lower than that predicted for random-coil peptides having the same sequence. Conversely, we observed that the fast exchanging peptides incorporated deuterium either as fast as their random-coil counterpart in the case of peptide Y115-L126, or at least one order of magnitude faster than their slow-exchanging counterparts (e.g., peptides L159-L170 and R171-L178). Figure 6 shows the location and experimentally determined HDX protection factors of these three peptides (Y115-L126, L159-L170, and R171-L178) on the two apoA1 chains in the Turtle model. Peptides located in the hairpin domain do not interact with lipid, in contrast to their counterparts outside of the hairpin and located on apoA1 domains that interact with lipid. The distinct partitioning of the peptides and corresponding microenvironment (either interacting with the lipid or not) seems to correlate with their H/D exchange behavior, i.e., fast versus slow. Based on our HDX data on the Solar Flare region of apoA1 (L159-A180), which we found to exchange much faster when bound to lipid than in lipid-free apoA1 (
), we can hypothesize and extrapolate to assume that peptides that experience a protein-protein interaction environment only may demonstrate slower H/D exchange (i.e., higher protection factor), in general, because they may be more tightly packed and less susceptible to unfolding due to the diversity of intermolecular interactions they experience (e.g., charge-charge, charge-dipole, dipole-dipole, and H-bond). On the other hand, peptides interacting with lipids experience only weak van der Waals interactions, which make them more susceptible to unfolding, and thus for exposed amide H, to faster H/D exchange.
It is also noteworthy to mention that we found that the Solar Flare region (L159-A180), corresponding to peptides L159-L170 and R171-L178 on one chain of apoA1 in the nHDLDMPC particle, demonstrated fast H/D exchange (protection factor 10 and 11, respectively) even at 5°C. Thus, our HDX data on nHDLDMPC confirm our earlier result obtained for nHDLPOPC (nHDL reconstituted with POPC and cholesterol) (
) from their HDX data (5°C) and the overall predicted α-helical content of apoA1 based on circular dichroism data on nHDLPOPC.
The lipid domain of nHDLDMPC is predicted to be predominantly lamellar (bilayer) with a small micellar domain
We next sought to further investigate the possible organization of the phospholipids within the lipid core of the nHDLDMPC particles so that eventually, a hypothetical global all-atom model of the particle could be constructed. To build the three-dimensional model of the lipid phase, we used the reference low-resolution structure obtained from the 42% contrast SANS experiments (Fig. 3D, supplementary Fig. III) as the shape envelope within which the phospholipids were placed using the known composition of the particle as a guide for the number of phospholipids to include. Similar to the case of developing the protein model, the manner in which individual lipids might be arranged within the lipid core was not obvious from the SANS data alone. However, the fact that the apoA1 chains within nHDLDMPC have an open conformation, coupled with the known biophysical properties and packing characteristics of phospholipids, provide some logical suppositions to the packing. First, it is presupposed that the lipids within the particle will be organized predominantly as a bilayer domain, in keeping with this lipid organization being stable, a known packing structure for choline glycerophospholipids, and whose dimensions are compatible with the majority of the lipid SANS shape observed. However, the open arms of the apoA1 chains necessitate that a nonlamellar lipid organization occur in these regions, since it logically follows that a domain within the lipid core of the particle will have to accommodate the exposed region to the aqueous surface where the open apoA1 ends fail to overlap and enshrine the lipid (Fig. 4, and see below). What the packing organization of this region is cannot be unequivocally determined with the current data. Nevertheless, the protein shape argues that a smaller micellar domain that caps the space between the open N/C-termini of the apoA1 dimer might occur (Fig. 4A). A final hypothetical all-atom model of the lipid phase was thus obtained by combining a predominantly bilayer organization with a small micellar end-cap domain, followed by energy minimization. The combination of the two arrangements (bilayer and micellar) is evident from the different orientations shown in Fig. 7. Figure 7B displays the superposition of the calculated scattering intensities from the hypothetical lipid model and the experimental data, which demonstrated an excellent statistical match (χ2= 2.31, Table 1). Visual inspection of the superposition of the hypothetical lipid model developed and the low-resolution SANS structure (42% contrast reference shape envelope for lipid) is shown in Fig. 7C, and is notable for the excellent match. Overlay of the protein and lipid components, followed by energy minimization, yielded the three-dimensional all-atom model of nHDLDMPC (the Turtle model; Fig. 8, stereo view in supplementary Fig. VII), which will be discussed in detail below. Of note, the overall form is that of a discoidal particle with a protruding hairpin.
Discussion of the Turtle model in the historical context of prior discoidal models of nHDL
The first proposed discoidal model of nHDL, the bicycle tire micelle (a.k.a., Belt) model (
), was inferred from SANS (without contrast variation) and electron microscopy studies of nHDLDMPC with identical lipid composition to the particle studied herein, but using an N-terminal Δ43 truncated apoA1 mutant. Two decades later, an all-atom model was proposed for nHDLPOPC (
) have posited a closed apoA1 dimer in a belt configuration that completely enshrines a central lipid bilayer, regardless of lipid composition. We recently used SANS with contrast variation to directly visualize for the first time the protein component of nHDL reconstituted with apoA1:POPC:cholesterol (1:100:10, mol:mol:mol) (
)]. Further, when coupled with contrast variation SANS analyses of the lipid components of this nHDL preparation, H/D exchange tandem mass spectrometry results, and biochemical constraints (FRET, ESR, and mass spectrometry cross-linking), the lipid corresponding to this particle (composition 100:10:1, POPC:cholesterol:apoA1) appeared to adopt a micellar/pseudolamellar packing organization. Subsequent multinuclear NMR studies of this particle corroborated this assertion, showing multiple 31P NMR spectral features characteristic of a micellar packing organization, and distinct from a lamellar (bilayer) organization (
). It is thus remarkable that the present SANS studies of an nHDL particle generated from the classic starting composition of 80:1 (DMPC:apoA1), both confirmed prior assertions of a predominantly discoidal nHDL particle (generated at this lipid:protein composition), yet also revealed an open apoA1 configuration, and also posits the presence of a nonlamellar (bilayer) domain, presumably a small micellar lipid endcap. Contrast variation SANS allows for direct visualization of the time-averaged conformation of the protein within the particle, and clearly shows an open configuration for the apoA1 dimer (Figs. 2, 4). And while the SANS studies lack resolution to determine the molecular packing of the lipid within the particle, it logically follows that a domain within the lipid core of the particle will have to accommodate the exposed region to the aqueous surface where the open apoA1 ends fail to overlap and enshrine the lipid (Fig. 4, and see below).
Historically, nHDLDMPC has been used by many [e.g., (
)] to study the structure of nHDL because the shorter fatty acyl chains in DMPC enable large unilamellar vesicles (LUVs) of DMPC to spontaneously associate with apoA1, especially at the transition temperature (23°C) of DMPC vesicles. Despite the ease of formation of these HDL preparations, they also are relatively heterogeneous, with predominantly two sizes of nHDL particles [9.8 nm and 10.4 nm (
), making such preparations difficult to study and interpret by SANS, which requires monodispersed samples. Consequently, we used the classic cholate dialysis method for reconstituted nHDL formation with the DMPC in the present studies because we observed that superior (with respect to homogeneity) particle preparations of identical size to the major particles observed with DMPC/LUV:apoA1 mixtures were obtained, that could be purified more readily using size exclusion chromatography prior to SANS analyses (Fig. 1).
Further analyses of SANS data yield insights into potential lipid packing in nHDLDMPC
As noted above, the open protein shape observed from the 12% contrast SANS analyses of nHDLDMPC dictates that some phospholipid within the particle must possess a packing organization different from a simple bilayer. To this end, we sought to determine whether further examination of the 42% contrast SANS data could shed light on the different lipid packing that might exist in this region of the particle. In the Belt model of nHDLDMPC (
) all 160 DMPC lipids make a bilayer (supplementary Fig. VIIIA, B). One can calculate the theoretical scattering intensity given off by such a lipid bilayer (supplementary Fig. VIIIC, solid green line). Examination of this plot reveals that it does not agree well with the experimental data at higher scattering angles (supplementary Fig. VIIIC), while the predicted scattering curve from the lipid core comprised of both lamellar (bilayer) and micellar configurations, as proposed in the Turtle model (Figs. 7, 8), is in excellent agreement with the experimentally determined scattering intensities in the 42% contrast SANS study of nHDLDMPC (Fig. 7B, supplementary Fig. VIIIC).
To further investigate the impact of lipid packing on the predicted SANS scattering data and to better understand the discrepancy in predicted SANS scattering intensities from a purely lamellar (e.g., Belt) versus combined lamellar/micellar (Turtle) model, we examined the individual contributions of the two predicted domains within the Turtle model (Fig. 9). An illustration of the different domains within the lipid core of the particle is shown in Fig. 9A, wherein approximately 28% of the lipids form the micellar domain (see also Fig. 9B for different views of the micellar region). Of note, for SANS analyses, the scattering amplitudes, but not intensities, are additive and can be decomposed into components coming from various domains (
) (see Materials and Methods). Using this approach, we sought to determine the potential relative contribution of a bilayer versus micellar lipid packing organization to the scattering signal at different angles. Interestingly, the theoretical scattering intensities for neither the pure bilayer (blue line) nor the pure micellar (red line) lipid phases matched the experimental data (black open symbols). However, when computationally merged (yellow line), the composite of the two predicted domains demonstrated overall agreement with both the experimentally determined SANS scattering data (open symbols) and the theoretical SANS curve produced from the predicted lipid core of the Turtle model (green line, Fig. 9C). This purely computational exercise thus seems to provide some further insights into the SANS analyses, since the theoretical scattering intensities generated at very small angles (q < 0.1 Å−1) appear to predominantly arise from lipids within the bilayer domain, whereas at higher scattering angles (q > 0.1 Å−1) the scattering intensities are predicted to come more so from the micellar domain (Fig. 9C). The overall goodness of fit observed for the SANS decomposition analysis strongly suggests that lipids within nHDLDMPC are packed as a mixture of lamellar and micellar phases. It is also important to realize this finding serves as an independent validation of the low-resolution structure of the protein observed (i.e., that it has an open conformation as a critical component).
Expansion of the lipid cargo in nHDL is accommodated by loss of the apoA1 hairpin
We hypothesized that in an nHDL particle with a sufficiently large lipid cargo, the lipid can exert sufficient lateral pressure on the apoA1 domains that fold back into the hairpin to have the hairpin “unzip”. Consequently, we prepared and analyzed by SANS nHDLDMPC particles that contain cholesterol nHDLDMPC+Chol. Particle biochemical characterization confirmed a relatively homogeneous preparation with particles containing an apoA1 dimer, and an approximate 20% increase in lipid cargo (Fig. 1). The scattering intensities, the Guinier range, and the distance distribution function P(R) from SANS analyses of nHDLDMPC+Chol (prepared with deuterated apoA1) at 12% contrast (to directly visualize the protein component of the particle) are shown in Fig. 10. As before, multiple low-resolution shape envelopes for the protein were generated (supplementary Fig. IX) and the overall low resolution structure of apoA1 in nHDLDMPC+Chol obtained is depicted in various orientations in Fig. 10D. The protein structure observed is more clearly round (114 × 113 × 56 Å) and discoidal than that for nHDLDMPC, but importantly, is still both open and with ends of the apoA1 dimer that are slightly out of plane. SANS analyses of nHDLDMPC+Chol at 42% contrast to directly visualize the lipid phase of the particle are shown in Fig. 11 and supplementary Fig. X. The low-resolution structure of the lipid phase (Fig. 11) is overall ellipsoidal, though it seems more asymmetric than nHDLDMPC.. Overall the protein and lipid shapes fit remarkably well with each other, like a lock and key, with a clearly discoidal overall particle (Fig. 11E).
The protein structures observed by SANS analyses suggest a molecular mechanism accounting for the quantized sizes observed in nHDL particles
It is well recognized that the size of HDL particles is predominantly determined by the number of apoA1 incorporated per particle and its lipid cargo. Reconstituted HDL particles formed with apoA1 and varying amounts of DMPC (or other phospholipids) produce discrete well-defined sized particles predominantly containing two apoA1 chains [coined as R2 type (
)]. The molecular basis for the observed “quantized” size heterogeneity is ill defined, but has been suggested to arise from incremental incorporation of 11mer α-helical N-terminal domains into the closed apoA1 belt, a model referred to by Segrest and colleagues as the “hinge-domain” hypothesis (
). A cartoon of this model is shown in Fig. 12B (upper row), and is readily visualized by a belt that can be enlarged by 11mer increments (colored in green) as it expands from one to another “belt hole”. Indeed, in addition to nondenaturing gradient gel electrophoresis (
), the molecular basis for the quantized expansion in nHDL size following the interaction of apoA1 and phospholipids is unknown.
As illustrated by analyses of nHDLDMPC versus nHDLDMPC+Chol particle preparations herein, the present studies show that contrast variation SANS is quite sensitive to alterations in protein conformation induced by differences in lipid cargo content within the nHDL particle. Direct visualization of the structure within the protein component of nHDL provides an alternative molecular mechanism for the quantized size heterogeneity observed in nHDL. First, a key observation is that the global shape of the apoA1 dimer within nHDL is open, and not closed. Another key observation is that a large hairpin domain is clearly visible for the first time using contrast variation SANS in the nHDLDMPC particle with smaller lipid cargo. This direct experimental data reveals the hinge domain hypothesis, which depends upon the presence of a closed belt, is inaccurate. Rather, as a growing nHDL particle matures and expanding lipid content exerts lateral pressure on the apoA1 chains, there will be an incremental “unzipping” of the hairpin where the free energy of interaction between the hydrophobic protein interfaces on the hairpin is counterbalanced by the lateral pressure exerted by the expanding lipid core. This seems to happen each time when four 11mer increments (from both apoA1 chains), locked in mutual interaction, unzip and get exposed to lipids. One example is when cholesterol is added to the phospholipid (Fig. 12A) in reconstituted nHDL. In the present case the addition of cholesterol increased the lipid cargo by 20%, and this led to the unzipping of the hairpin. A cartoon of this hypothesis is illustrated in Fig. 12B (lower row). This quantized expansion seems to be facilitated by an α-helix breaking proline (yellow stripes in Fig. 12B) or other residue that caps an 11mer (black stripes), serving as a discrete “hinge”.
Previous studies have suggested the presence of a hairpin within apoA1 of nHDL based upon alternative approaches. For example, Davidson and colleagues used a series of elegant site-specific mutants in which single tryptophan reporters and nitroxide spin quenchers were engineered within each of the 22 amino acid amphipathic helical repeats of apoA1. Differences observed with these apoA1 forms in tryptophan fluorescence between large (96 Å) and small (78 Å) nHDL particles produced, suggested a movable hinge domain exists within regions historically referred to as helices 5, 6, and/or 7, presumably allowing apoA1 to accommodate more or less lipid content (
). In contrast to these prior approaches, which invariably require mutation of apoA1 and/or derivatization reactions, which can alter apoA1 conformation, the present studies provide a direct visualization of the apoA1 shape at low resolution, and are nonperturbing inasmuch as they require no mutations or derivatization.
Structural similarities in nHDL models derived from SANS low-resolution structures point to common features among apolipoproteins
)] have thus far been constructed from contrast variation SANS data (supplementary Fig. XI). Comparisons among these reveal similarities, suggesting that there are structural features shared by many apolipoproteins. First, apoA1 within both nHDL models and the spherical HDL model have an open conformation with sets off N/C-termini of apoA1 apart. This is in contrast to the discoidal model (and other twisted/bended simulation models), in which an apoA1 dimer forms a closed loop with the N/C-termini coming together (in truncated Δ43 apoA1). Second, in all contrast variation SANS-based models, the protein conformation is contorted and out of plane, a feature that is likely imparted by the presence of repeating proline residues which accentuate the bends in conformation. Of note, some discoidal models obtained by simulation bend out of plane, but they always remain closed loops (
). Another structural feature that sets apart the contrast variation SANS (direct experimental observation) versus simulation (hypothetical) models is the conformation of the N/C-termini. In all HDL structural models based upon direct visualization of protein shapes using contrast variation SANS thus far (
) first suggested the existence of this feature of apoA1 based on mass spectrometry cross-linking experiments. It is likely the two sets of termini in nHDL “stay apart”. This forces “end-capping” phospholipids in micellar configuration to connect the two leaves of the bilayer within the central lipid domain of nHDL. It may be that this behavior (N/C-termini folding back and staying apart) is imprinted within the primary sequence of apoA1, and it is tempting to speculate that the endcap containing micellar lipid configuration in HDL may even impart unique biological activity. For example, it is well-known that many lipases, and other proteins that act at a lipid interface, can function even more efficiently on a micellar surface (
Another remarkable similarity among SANS based models is that, regardless of lipid composition, they have the same radius of gyration (Rg=∼52Å), and very similar overall dimensions (Table 1 and Table 1A of Ref.
). The Belt model has Rg=∼49 Å, and the Rg for the bended simulation models is even lower (Table 1). Finally, in the contrast variation SANS-based models, the very nature of the open protein conformation allows individual lipids to diffuse freely throughout the entire lipid phase. This aspect may be essential for the ability of the particle to function as a cholesterol carrier because the lipid phase does not raise a free energy barrier to cholesterol diffusion through the entire lipid phase, and thus any cholesterol molecule within the lipid phase can access LCAT when bound to nHDL with the same easiness (i.e., there is no sidedness, like “heads” or “tails” of a coin, to the bilayer lipids as they can diffuse freely across the micellar end cap domain).
Geometrical analysis of the apoA1 three-dimensional model for nHDLDMPC shows that major bends in protein conformation correspond mostly to proline positions
As mentioned above, the locations of proline residues in apoA1 of nHDL are believed to be associated with hinge domains through which lipid-free apoA1 incrementally unfolds to accommodate particle enlargement (
). Interestingly, inspection of the Turtle model of nHDLDMPC reveals that most prolines are located at positions where major bends in apoA1 occur (Fig. 13B). For example, P99A and P165B (which are in close proximity) are located at the summit of the hairpin folds. Another pair of prolines, P66A and P121B, is close to the starting points of the hairpin. Another striking aspect of the conformation of apoA1 in the Turtle model is that the “Solar Flare” regions of the apoA1 chains (
), putative sites of LCAT interaction (L159-A180), are not equivalent. Virtually all “belt type” models have a significant degree of symmetry. In contrast, within the Turtle model, the solar flare positioned within the hairpin is not accessible to the lipid, which might suppress the LCAT activity at this location.
Also illustrated in Fig. 13B, C is the concept that while the individual HDLDMPC structure itself lacks symmetry, the nHDLDMPC will be comprised of a pair of indistinguishable symmetrically related structures. This is because the particles are generated from an antiparallel pair of identical apoA1 chains. To illustrate this point, the pair of structures shown in Fig. 13B are two views (front and back) of an nHDLDMPC formed with the hairpin located in the N terminus of chain A (between P66 and P143) and the C terminus of chain B (between P121 and A194). Note that while there is no proline between P165B and P209B, a residue with a relatively small and nonbulky side chain, A194B, is directly adjacent (opposite) to P66A of chain A, and readily accommodates the predicted bend in chain B opposite to P66 of chain A at the base of the hairpin. The structures in Fig. 13C are the same two views of an alternative nHDLDMPC where the hairpin is instead located in the C terminus of the A chain (between P121 and A194) and the N terminus of chain B (between P66 and P143). Note that in HDLDMPC structures depicted (Fig. 13B, C) the A chain is colored red and the B chain is colored blue (N terminus dark and C terminus light for each). Thus, there is a 50% chance that the hairpin is located in either the N or C terminus of apoA1, and the two forms of nHDLDMPC cannot be distinguished (in both structures the hairpin is between P66 and P143 of one chain and P121 and A194 of the other chain). While the nHDLDMPC structures shown in Fig. 13B and 13CFig. 13B and 13C themselves do not contain a C2 axis of symmetry [like the crystal structures of lipid-free N terminal or C terminal truncated mutant forms of apoA1 (
)], it is evident that the nHDLDMPC are related by a form of symmetry.
Lipoproteins are involved in many biological functions, one of which is sterol and lipid transport. Despite their essential role in many physiological processes, efforts thus far to solve their crystal structure in the presence of lipid has been unsuccessful, most likely due to their heterogeneous composition and dynamic nature when lipidated. The application of SANS with contrast variation to the structural interrogation of nHDLDMPC enabled us to directly visualize a time-averaged low-resolution structure for the protein and lipid components of nHDL individually, and has revealed an unexpected conformation of apoA1, a contorted open structure with a prominent hairpin that restricts the access of the lipid to only a portion of the hydrophobic surface of the protein. But, like the case of nHDLPOPC, the protein in nHDLDMPC offers mechanical strength to the HDL particle, while mediating lipid binding/transport and specific interactions with other proteins, like plasma enzymes and cell receptors. The observed open shape of apoA1 and the mixed packing (predominantly bilayer) of the lipid accounts for the experimental SANS intensities, while alternative shapes of the protein obtained by computer simulation do not (e.g., supplementary Figs. XI, XII).
In conclusion, we used contrast variation SANS to interrogate the structure of a reconstituted nHDL particle comprised of DMPC:apoA1 (80:1), a classic composition studied in numerous other investigations (
), as well as a particle with expanded lipid cargo. The results reveal overall shapes of both particles that are consistent with prior discoidal models of nHDL, in the sense that most of the lipid phase is predicted from SANS analysis to be lamellar packed, and the overall dimensions are of oblate ellipsoids. However, the protein shapes directly visualized within either nHDLDMPC or nHDLDMPC+Chol are clearly not classic belt configurations commonly envisaged; rather, the protein architecture of apoA1 within nHDLDMPC and nHDLDMPC+Chol both have an open configuration. The direct visualization for the first time of a presumptive hairpin loop by SANS within nHDLDMPC, combined with both the confirmatory cross-linking mass spectrometry studies and the absence of a hairpin within the expanded lipid cargo particle, nHDLDMPC+Chol, posit a molecular and structural mechanism for the well-known quantized size increases of nHDL during particle maturation. Moreover, the distinct environments experienced by the same sequences on distinct antiparallel apoA1 chains within nHDL explains the origin of bimodal exchange patterns observed during HDX studies.
This study was supported by National Institutes of Health Grants P01 HL-098055, P01 HL-076491, and by the Leduc Foundation. S.L.H. was partially supported by a gift from the Leonard Krieger Fund and V.G. acknowledges the Faculty Development Research Award from Cleveland State University. The electron microscopy facility used is part of the Grenoble Partnership for Structural Biology. Computational resources were provided by the Ohio Supercomputer Center and the National Center for Supercomputer Applications (NCSA, University of Illinois). Mass Spectrometry instrumentation used was housed within the Cleveland Clinic Mass Spectrometry Facility with partial support through a Center of Innovation by AB SCIEX. Some of the human apoA1 used in this study was purified from HDL provided as a generous gift from CSL Limited (Victoria, Australia).
double super helix
electron spin resonance
fluorescence resonance energy transfer
large unilamellar vesicle
nascent HDL/dimyristoyl phosphatidylcholine without cholesterol
nascent HDL/dimyristoyl phosphatidylcholine with cholesterol