Double Superhelix Model of High Density Lipoprotein

The Double Super Helix model of high density lipoprotein Zhiping Wu1,2 , Valentin Gogonea1,4, Xavier Lee1,2, Matthew A. Wagner1,2, Xin-Min Li1,2, Ying Huang1,2, Arundhati Undurti1,2, Roland P. May5, Michael Haertlein5, Martine Moulin5, Irina Gutsche6, Giuseppe Zaccai5, 7, Joseph A. DiDonato1,2, Stanley L. Hazen1,2,3

From Departments of 1 Cell Biology, 2Center for Cardiovascular Diagnostics and Prevention, and 3 Cardiovascular Medicine, Cleveland Clinic, Cleveland, OH 44195, 4 Department of Chemistry, Cleveland State University, Cleveland, OH 44115, 5 Institut Laue-Langevin, 6, rue Jules Horowitz, BP 156, 38042 Grenoble Cedex 9, France, 6 UVHCI, UMR 5233 UJF-EMBL-CNRS, 6, rue Jules Horowitz, BP 181 38042 Grenoble Cedex 9, France, 7 Institut de Biologie Structurale (CEA-CNRS-UJF), Grenoble, France

Running title: The structure of nascent HDL

Address correspondence to Stanley L. Hazen: Cleveland Clinic, 9500 Euclid Avenue, NE-10, Cleveland, OH 44195, Tel: 216-445-9763, Fax: 216-636-0392, E-mail: [email protected]

Keywords: Apolipoprotein A1, high density lipoprotein (HDL), small angle neutron scattering (SANS), hydrogen deuterium exchange mass spectrometry

ABSTRACT

High density lipoprotein (HDL), the carrier of so-called “good” cholesterol, serves as the major athero-protective lipoprotein and has emerged as a key therapeutic target for cardiovascular disease. We applied small angle neutron scattering (SANS) with contrast variation and selective isotopic deuteration to the study of nascent HDL to obtain the low resolution structure in solution of the overall time-averaged conformation of apolipoprotein A1 (apoA1) versus the lipid (acyl chain) core of the particle. Remarkably, apoA1 is observed to possess an open helical shape that wraps around a central ellipsoidal lipid phase. Using the low resolution SANS shapes of the protein and lipid core as scaffolding, an all-atom computational model for the protein and lipid components of nascent HDL was developed by integrating complementary structural data from hydrogen deuterium exchange mass spectrometry and previously published constraints from multiple biophysical techniques. Both SANS data and the new computational model, the Double Super Helix model, suggest an unexpected structural arrangement of protein and lipids of nascent HDL, an anti-parallel double super helix wrapped around an ellipsoidal lipid phase. The protein and lipid organization in nascent HDL envisages a potential generalized mechanism for lipoprotein biogenesis and remodeling, biological processes critical to sterol and lipid transport, organismal energy metabolism, and innate immunity.

INTRODUCTION

HDL functions in removal of cholesterol from peripheral tissues, such as within the artery wall, for delivery to the liver and ultimate excretion as biliary cholesterol within the intestinal lumen - a process called reverse cholesterol transport (1,2). Plasma levels of HDL cholesterol and apolipoprotein A1 (apoA1), the major protein component of HDL, are inversely related to the risk of developing coronary artery disease (3-5). Moreover, genetic alterations that induce changes in apoA1 levels in both animal and humans alter susceptibility for development of atherosclerotic heart disease (3-6). Thus, numerous interventions aimed at enhancing reverse cholesterol transport are being examined as potential novel therapeutic interventions for the prevention and treatment of cardiovascular disease (7,8). Examples include methods for generating new HDL particles through enhanced production or delivery of either intact apoA1 (9,10) or peptide mimetics of apoA1 (11), as well as modulating interactions between nascent HDL and proteins involved in HDL particle maturation and remodeling for potential therapeutic benefit (12-14). Structural elucidation often serves as the "Rosetta Stone" for enhanced understanding of function. It is thus remarkable that despite its importance to numerous biological and biomedical functions and its current prominent role as a target for therapeutic interventions, to date, the structures of neither the protein nor lipid components of nascent HDL have been directly visualized, and the high resolution structure of the particle remains unknown.

In the absence of high resolution structures, numerous models of nascent HDL have been

1

http://www.jbc.org/cgi/doi/10.1074/jbc.M109.039537The latest version is at JBC Papers in Press. Published on October 7, 2009 as Manuscript M109.039537

Copyright 2009 by The American Society for Biochemistry and Molecular Biology, Inc.

by guest on August 9, 2016

http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from


http://ww

w.jbc.org/

Dow

nloaded from

mailto:[email protected]

http://www.jbc.org/cgi/doi/10.1074/jbc.M109.039537

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

http://www.jbc.org/

proposed (recently reviewed in (15,16)). In general, these have relied on various biophysical and biochemical studies and are named after the proposed overall architecture of apoA1 within the particle (e.g. Picket Fence model (17), Double Belt model (18,19), various Hairpin Loop models (20-22), and most recently, the Solar Flares model (23)). Each of these models of nascent HDL assumes a structure comprised of a central phospholipid and cholesterol bilayer with amphipathic alpha-helical apoA1 arranged around the perimeter of the central lipid core. Refinement of the protein orientation has been dictated by biophysically determined constraints such as hydrophobic/hydrophilic character along the predicted alpha helices, overall alpha helical content as estimated by circular dichroism, measured distance constraints between inter- and intra-chain amino acids by multiple approaches, and more recently, quantitative indices of solvent accessibility and dynamics as determined by amide bond hydrogen / deuterium exchange throughout the apoA1 polypeptide backbone. While all current models share an anti-parallel orientation of two apoA1 predominantly alpha helical chains, the overall conformation of the apoA1 alpha helical chains within the particle is still debated. Of note, none of the models is based upon studies that provide direct imaging of the shape of apoA1 or the lipid core within nascent HDL in solution.

Structural data from individual low resolution platforms can have limited usefulness in resolving the molecular structure of biological compounds, but limitations in structural definition inherent in individual low resolution biophysical approaches can be partially overcome by combining them synergistically (24-26). Indeed, it is becoming increasingly recognized within the structural biology field that merging multiple complementary techniques can prove particularly valuable when only low resolution structures are available. For example, Fleishman et al. (27) have recently shown how combining multiple biophysical/biochemical constraints and computational analyses coupled with cryo-EM low-resolution data can result in enhanced resolution, as well as provide mechanistic understanding of membrane protein structure and function without crystallographic data.

Herein we develop and apply to the study of nascent HDL a broadly applicable, multidisciplinary methodological platform for investigating the solution structure of macromolecular complexes resistant to

traditional structural approaches. By uniting the diverse yet synergistic non-perturbing experimental techniques of small angle neutron scattering (SANS) with contrast variation, isotopic deuteration of selected macromolecule components and hydrogen / deuterium exchange tandem mass spectrometry (HD-MS/MS), we have been able to describe the time averaged conformation of protein and lipid components of nascent HDL separately at low resolution, and then to enhance the resolution of these shapes by incorporating multiple additional biophysical constraints to build a computational model for the particle.

EXPERIMENTAL PROCEDURES

Preparation of nascent HDL. The particle composition selected for our studies, the methods for particle generation, and the functional characterizations performed to demonstrate a biologically active particle closely align with the current HDL-based therapeutic interventions being evaluated in the clinic (28-30) and nascent HDL produced by macrophages (31).

Human apoA1 was isolated from plasma of healthy volunteers. Briefly, human HDL was isolated from a pool of fresh plasma by ultracentrifugation adjusted with KBr to a density range of 1.07 g/ml-1.21 g/ml as described. Lipid free human apoA1 was prepared by delipidation of isolated human HDL using methanol/ether/chloroform followed by ion exchange chromatography (32). The purity of isolated human apoA1 was verified by SDS-PAGE. Recombinant human apoA1 was generated in E. coli and isolated as previously described (33). To prepare deuterated recombinant human apoA1, the kanamycin resistant plasmid pET20b+ encoding 6xHis-tagged human apoA1 was transformed into E. coli BL21(DE3) cells. Kanamycin resistance allows efficient selection and maintenance of the expression construct under high cell density growth conditions required in deuterated minimal media (34-36). Cells were grown in 85% deuterated minimal medium (85% D2O, 15% H2O): 6.86 g/L (NH4)2SO4, 1.56 g/LKH2PO4, 6.48 g/L Na2HPO4

.2H2O, 0.49 g/L diammonium hydrogen citrate, 0.25 g/LMgSO4

.7H2O, 1.0 ml

2


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

L-1 (0.5 g/L CaCl2.2H2O, 16.7 g/L FeCl3

.6H2O, 0.18 g/L ZnSO4

.7H2O, 0.16 g/L CuSO4.5H2O,

0.15 g/L MnSO4.4H2O, 0.18 g/L CoCl2

.6H2O, 20.1 g/LEDTA), 40 mg/L kanamycin with hydrogenated glycerol as carbon source (5 g/L). Deuterated apoA1 was purified by nickel affinity chromatography using established methods (23,33).

Reconstituted nascent HDL was prepared using modified sodium cholate dialysis method (37) at an initial molar ratio of 100 : 10 : 1 of 1-palmitoyl, sn-2-oleoyl phosphatidyl choline (POPC) : cholesterol : isolated apoA1, which is a modification of the original seminal studies by Jonas and colleagues who first reported the cholate dialysis method (32) for reconstituted HDL preparation. HDL particles were further purified by gel filtration chromatography using Sephacryl S300 (GE healthcare) column. The size of reconstituted nascent HDL was routinely measured as previously described using dynamic light scattering, and non-denaturing PAGE (23). The stoichiometry of all nascent HDL preparations was determined as previously described (23). The amount of sodium cholate remaining in all reconstituted nascent HDL preparations was quantified by HPLC tandem mass spectrometry to confirm no significant residual cholate was present. The stoichiometry of apoA1 on nascent HDL made of human plasma isolated apoA1 (rHDL) and bacterial produced recombinant His-tagged apoA1 (rrHDL) were determined by chemical crosslinking studies as previously described (23).

HDL characterization by circular dichroism spectroscopy. Far-UV circular dichroism spectra were recorded on a JASCO 815 CD spectrophotometer (JASCO, UK). Reconstituted nascent HDL samples (rHDL and rrHDL) were analyzed at ambient temperature in continuous scan mode with a 1-nm bandwidth at a wavelength of 260 to 185 nm and with a path length of 1 mm. The spectra were normalized to mean residue ellipticity with the use of a mean residue molecular mass of 115.4 Da for apoA1.

Fractional α-helix contents were calculated using the neural network-based K2d program(38).

HDL characterized by Lecithin Cholesterol Acyltransferase (LCAT) activity assay and LCAT binding. The activation of LCAT by rHDL and rrHDL spiked with trace amount of 3H cholesterol was measured as described before with slight

modification (23). The reaction complex contained 0-35 μM cholesterol in a final concentration of 10 mM phosphate, pH 7.4, 1 mM EDTA, 150 mM NaCl, 2 mM β-mercaptoethanol, 0.6% fatty acid free bovine serum albumin and 20 ng of purified 6xHis-tagged human LCAT (rhLCAT). The reactions were carried out in triplicate at 370C for 35 minutes under argon. LCAT activity was determined by calculating the conversion efficiency of [3H] cholesterol to [3H] cholesteryl ester after lipid extraction of reaction mixture followed by thin-layer chromatography(23).

Measurements of apparent dissociation constants between LCAT and both rHDL and rrHDL were performed using a BIAcore 3000 SPR biosensor (BIAcore, AB) following the methods of Jin et al with modification (39). Briefly, approximately 8000 RU of polyclonal antibody against apoA1 (Biodesign, CA) was immobilized on a CM5 sensor chip through primary amino groups using reactive esters. Nascent HDLs were captured on the sensor chip through interaction with antibody against apoA1 by injecting 7 µM of HDL at a flow rate of 15µl/min in 10 mM PBS buffer (pH 7.4) into the flow cell. To determine the Kds between rhLCAT and nascent HDL, rhLCAT ranging from 500 nM to 2000 nM were prepared in binding buffers of 10 mM PBS (pH 7.4) and injected over the surface of the sensor chip at a flow rate of 20 µl/min. At the end of each cycle, surfaces of the

sensor chips were regenerated by injection of 15mM HCl at the same flow rate. The apparent dissociation constants were obtained by fitting background-subtracted SPR binding data to the 1:1 binding with drifting baseline model using the BIA evaluation software version 4.0.

HDL characterized by cholesterol efflux assay. Subconfluent J774A.1 murine macrophage cells in 48-well plates were loaded by 0.3 μCi/ml [3H] cholesterol overnight in 0.4 ml of DGGB (DMEM supplemented with 50 mM glucose, 2 mM glutamine, and 0.2% BSA). The day after cholesterol loaded, the cells were washed 3 times in PBS and treated with or without various 5 μg/ml HDLs for 6 hours in serum-free DMEM medium. The radioactivity in the chase media was determined after brief centrifugation to remove pellet debris. Radioactivity in the cells was determined by extraction in hexane:isopropanol (3:2, v/v) with the solvent evaporated in a scintillation vial prior to counting. The percent cholesterol efflux was calculated by radioactivity in the medium divided by the total radioactivity (medium radioactivity plus cell radioactivity) (33).

3


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

SR-B1 specific binding of HDL. rHDL was iodinated using Bolton-Hunter reagent to prevent oxidation of the particle (40). 293-T cells were transfected with vector or SR-B1 using lipofectamine-2000 according to manufacturer’s instructions. The next day, cells were plated in 24 well plates and after another 24 hours specific binding of the radiolabeled rHDL to SR-B1 was determined by incubating 125I-rHDL with either vector transfected or SR-B1 transfected cells for 1.5 hours at 40C. After 1.5 hours, cells were washed twice with 250 mM NaCl, 25 mM Tris (pH 7.4) and once with 250 mM NaCl, 25 mM Tris (pH 7.4) with 2 mg/mL BSA. Cells were solubilized in 1 mL 0.1 M NaOH at room temperature for 20 minutes and cell associated radioactivity subsequently determined with a Gamma4000 spectrometer (Beckman Coulter, Fullerton, California, USA). Specific binding was calculated as total binding minus binding in the presence of 30-fold excess of unlabeled rHDL. For the competition assay, a 30-fold excess of either unlabeled human HDL from plasma, unlabeled reconstituted HDL prepared with recombinant ApoA1 (rrHDL) or unlabeled POPC SUV was added along with 100 μg/ml iodinated human HDL to determine inhibition of iodinated human HDL binding to SR-B1 (41).

Measurement of anti-apoptototic activity of HDL. Human umbilical vein endothelial cells (HUVEC) were plated overnight in 60 mm dishes in MCDB media supplemented with 15% FBS. The next day, cells were washed with PBS and serum deprived for 6 h with simultaneous incubation with either 500 μg/ml pHDL, 500 μg/ml rHDL, 500 μg/ml apoA1 or 500 μg/ml small unilamellar vesicles (SUV) generated with 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC). After 6 hours apoptosis was measured using an Annexin V-FITC apoptosis detection kit (BD Pharmingem, Franklin Lakes, New Jersey). Briefly, cells were washed twice with PBS, harvested and suspended in 1x binding buffer (10 mM Hepes/NaOH pH 7.4, 140 mM NaCl, 2.5 mM CaCl2) at a concentration of 106 cells/ml. The solution was transferred to a 5 ml culture tube and incubated with annexin and propidium iodide (PI) for 15 min at room temperature in the dark according to the manufacturer’s instructions. Flow cytometry of labeled cells was performed on a FACScan (Becton-Dickinson).

Determination of surface VCAM-1 protein. HUVEC were plated overnight in 96 well plates in MCDB-105 media supplemented with 15% FBS. The

next day, cells were washed twice with PBS pH 7.4 and pre-incubated with either 500 μg/ml pHDL, 500 μg/ml rHDL, 500 μg/ml ApoA1 or 500 μg/ml POPC SUV for 2 h. After 2 h, 5 ng/ml of TNF-α was added and cells were incubated for an additional 6 h. Cells were then washed three times with PBS pH 7.4 and fixed in 4% paraformaldehyde for 30 min on ice. Cells were subsequently washed and blocked overnight with 5% BSA. The day after blocking, cells were incubated with anti-VCAM-1 primary antibody (sc-53778, Santa Cruz Biotechnology, California, USA) for 2 hours at room temperature. After three washes with PBS pH 7.4, cells were incubated with sheep anti-mouse HRP conjugated secondary antibody (GE Healthcare, Piscataway, New Jersey, USA), for 2 hours at room temperature. TMB substrate was subsequently added to each well and the reaction stopped after 20 min by addition of 1 M HCl. Absorbance was recorded at 450 nm on a 96 well plate reader (Spectramax 384 Plus, Molecular Devices, Sunnyvale, California).

SANS Experiment. Small angle neutron scattering experiments were carried out at the instrument D22 of the Institut Laue-Langevin, Grenoble, France. D22 (http://www.ill.eu/d22) is a classical pinhole camera that provides the highest neutron flux among all comparable SANS instruments in existence (42). Data collected from two positions of the detector (2 and 5.6 m with collimation lengths of 2.8 and 5.6 m) covered the momentum transfer (q) range from 0.008 to 0.35 Å-1. q is defined as 2π (sin θ)/λ, where 2θ is the scattering angle. The wavelength λ of the neutron beam used was 6 Å. Data beyond 0.25 Å-1 were too noisy to analyze.

To understand the organization of the protein and lipid in HDL contrast variation ([D2O] : [H2O + D2O] ratio) was required. The HDL samples were measured in 0, 12, 42 and 90% D2O solution to assure the required levels of contrast. From the corresponding scattering curves two important parameters were obtained for each contrast: the radius of gyration, Rg, and the intensity at 0 angle I(q=0). Rg were obtained based on the Guinier approximation (43), which is valid for q values (Rgq)2<1 (equation 1).

limq->0 Icorr (q) = I(0)exp(−Rg2q2⁄3) (1)

The logarithmic intensities varied linearly with q2 in the chosen Guinier range and the obtained Rg were

4


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

stable against slight change in q-range. The molecular mass (M) of the HDL particle was calculated as 200,000 Da based on the absolute scale measured I(q=0) (equation 2), in which C (mg/cm3) is the concentration of HDL, ∑b (cm) is the particle scattering length, V (cm3) is the particle volume, ρ0

(cm-2) is the solvent scattering length density, and NA is Avogadro’s number; the term in brackets is the particle excess scattering length normalized to unit molecular mass (44). The molecular mass calculated from the data corresponded to one HDL particle.

I(0) ⁄C=NA[(∑b −ρ0V)⁄M]2 M (2)

The program DAMMIN (45), a simulated annealing method, was used to calculate a low resolution model of the lipid core from the scattering curve of the sample in 42% D2O. Likewise the 12% scattering curve was used for the modeling of the protein. In the approach, the starting structure is a sphere of diameter equal to the maximum particle dimension, which was estimated from the scattering curve, via the distance distribution function P(r). The sphere was filled with dummy atoms with their size determined by the highest value of momentum transfer (q) of the scattering curve. To enhance the signal to noise ratio of the sample in 12% D2O, deuterated apoA1 was used for the reconstitution of the HDL particle. The χ value of the simulated annealing modeling for lipid is less than 1, and 5.2 for the protein. Both are in the excellent or reasonable range of statistics.

SAXS Experiment. Synchrotron X-ray scattering data of nascent HDL in solution were collected at the

X33 beamline (DESY, Hamburg) at particle concentrations ranging from 2 to 16.0 mg/ml. At a sample-detector distance of 2.7 m, the range of momentum transfer 0.01 < q < 0.5 Å–1 was covered (λ = 0.15 nm the X-ray wavelength). The data were processed with program PRIMUS (46) using standard procedures. The forward scattering intensity I(q=0) and the radius of gyration (Rg) were evaluated with program AUTORG (47) using the Guinier approximation. The effective molecular mass of the solute (MM) was estimated by comparison of the forward scattering intensity with that from reference solutions of bovine serum albumin (MM = 66 kDa).

Electron Microscopy Studies. Negatively stained HDL was obtained by applying a diluted solution of HDL particles (< 0.2 mg protein/ml) to the clear side of carbon on a carbon-mica interface and stained with

2 % (w/v) uranyl acetate. Images were recorded under low-dose conditions with a JEOL 1200 EX II microscope at 100 kV. Selected negatives were then digitized on a Zeiss scanner (Photoscan TD) at a step size of 14 μm giving a pixel size of 3.5 Å at the specimen level. Subsequent data processing was performed with the Imagic package. The data set, centered by translation, was subjected to multivariate statistical analysis and classification.

Computational modeling of nascent HDL. An all-atom computational model of nascent HDL was constructed by combining modeling techniques with experimental data including contrast variation SANS, HD-MS/MS data, and reported distance constraints from cross-linking, fluorescence resonance energy transfer and electron spin resonance experiments (20,21,48,49). The overall strategy in constructing a model of nascent HDL was to use the SANS low resolution structures obtained by deconvoluting the experimental SANS curves of 12% and 42% D2O as scaffolds to build molecular models for the protein and lipid components of nascent HDL, respectively. The overall model building involved over 250 iterative steps (i.e. over 250 models created for nascent HDL) with assessments of the goodness-of-fit with both SANS curves and experimentally determined H/D exchange data at every step. At each refinement step, precedence was always placed on SANS data for global conformation, while more fine-tuned refinements of local architecture primarily utilized H/D exchange data. It is worth noting that hydrogen / deuterium exchange, in conjunction with NMR and/or MS/MS analysis (50,51), has been used extensively in the past to determine the local environment, such as solvent accessibility and dynamics, of amino acid residues in proteins or in macromolecular complexes involving proteins. Determination of H/D exchange through either NMR or mass spectrometry methodology provides constraints that can help aide in structural modeling because amide hydrogen H/D exchange rates (and their degree of protection) are sensitive to the local environment of amino acid residues in various protein secondary structures (alpha helix, beta pleated sheets, etc.). The degree of H/D exchange in amide backbone hydrogen atoms depends upon both solvent accessibility and local interactions (e.g. participation in hydrogen-bonds in alpha-helical sites, hydrophobic contacts, protein dynamics).

An iterative co-refinement process was performed for each contrast variation study to first develop an

5


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

overall model of the protein component based upon SANS and HD-MS/MS data, then the lipid core, followed by a combined model with energy minimization at every step. A starting model for the lipoprotein was constructed using Modeller, Pymol, and Swiss-PDB-Viewer programs (52,53), by arranging the apoA1 chains into an anti-parallel super helical conformation to match the 12% D2O SANS low resolution structure. The crude protein model obtained in this way was further refined by iteratively adjusting the shape and the degree of protection of amide hydrogen for exchange. As described in detail within the Supplemental Information material available on-line, the program DEXANAL (23) was used to determine per residue deuterium incorporation factors (D0

i), residue unfolding constants (Kui) and

H/D exchange rate constants (kxci) from the

experimental H/D exchange data of overlapping peptic peptides. DEXANAL was also employed to calculate per residue H/D exchange probabilities (XPi) from three dimensional (3-D) molecular models. In this manner, an iterative co-refinement approach was used to improve the computational model so that predicted SANS and H/D exchange data matched experimental SANS and HD-MS/MS data. Use of DEXANAL for incorporating H/D exchange data into molecular models was validated using proteins with known crystallographic structure and published H/D exchange data as outlined in detail within Supplemental Information material available on-line (e.g. see Supplemental Information, Fig S1, S2, Table S1-S4).

The neutron scattering curves of the models of apoA1 within nascent HDL were computed using a modification of the SASSIM program (54). Instrument smearing was taken into account in calculating the scattering intensity as described in Merzel and Smith (54), and the relative wavelength spread of neutrons (Δλ/λ) used was 0.1 as suggested by Svergun et al (55). At each refinement step the model was energy minimized using the OPLS force field (56) (using the GROMACS program (57)).

The lipid phase (200 POPC molecules and 20

cholesterol molecules) was modeled to follow the helical orientation of protein hydrophobic surface filling the inside groove, and overall fits the experimentally visualized SANS prolate ellipsoid obtained by deconvoluting the 42% D2O scattering curve. The lipid model was refined further by performing energy minimization of the whole particle, and by matching its calculated scattering

intensity curve to the experimentally obtained one (42% D2O). Additional model refinement using experimental geometrical constraints such as chemical crosslinks, fluorescence resonance energy transfer and electron spin resonance distances are described in detail in the Supplemental Information. RESULTS

Small angle neutron scattering of nascent HDL.

Small angle scattering is a powerful approach for structural studies of macromolecules in solution. It provides a low resolution structure and is particularly useful in revealing details of the organization of a multi-component system. For example, in a contrast variation experiment of small angle neutron scattering (SANS), D2O/H2O levels are varied such that the scattering length density of the solvent is adjusted to match that of a component within a complex (e.g. protein, lipid, DNA or RNA), essentially rendering that component invisible (58). Thus, SANS can link structural and compositional information in solution in a way that is difficult to attain by other approaches. Indeed, SANS with contrast variation was the first method to correctly predict the structural orientation of protein and DNA within the fundamental subunit of chromatin, the nucleosome (59), as well as to triangulate the location of various proteins and RNA within the ribosome (60-62), the intracellular complex that translates the genetic code into proteins.

To probe the structure of apoA1 within nascent HDL particles we examined reconstituted nascent HDL particles produced using human apoA1 as described in Methods. Dynamic light scattering, biochemical and cross-linking/mass spectrometry analyses of HDL indicated monodispersed preparations with two apoA1 molecules per particle, and overall stoichiometry of apoA1 : phospholipid : cholesterol of approximately 1 : 100 : 10 (mol:mol:mol, Fig. 1). Prepared nascent HDL particles for structural studies were confirmed to be biologically functional with respect to a broad array of reported HDL activities including cholesterol efflux from cholesterol loaded macrophages, lecithin cholesterol acyl transferase (LCAT) binding, activity and catalytic efficiency, specific binding to the HDL receptor SR-B1, anti-apoptotic activity, and anti-inflammatory activity (Figs. 2 and 3). Absolute scale SANS analysis confirmed the predominance of a 200,000 Da lipoprotein particle in solution corresponding to the monomer of nascent HDL containing two apoA1 per particle. To obtain

6


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

information about the shape of apoA1 within HDL, scattering intensity data were collected in 12% D2O. At this concentration of D2O the scattering length density of solvent matches the average scattering length density of the lipid phase masking it at low angle. However, the initial low resolution structure of apoA1 obtained by analysis of the scattering curve at 12% D2O was inconclusive (ie not useful for determining the overall shape of the protein), presumably owing to the weak signal intensity from protein at this D2O level and the superposition of the residual signals produced by lipid polar head group scattering at higher angles. To overcome the issue of superimposed signals, we generated recombinant deuterated apoA1 (by growing E. coli in D2O with deuterated nutrients) for use in forming reconstituted nascent HDL. Because of the difference in the scattering length of hydrogen and deuterium (63), the deuterated apoA1 gives a much stronger signal in 12% D2O solvent.

Physical and biological properties of nascent HDL formed using recombinant deuterated His-tagged apoA1 again demonstrated a biologically active particle with the compositional and functional characteristics indistinguishable with reconstituted HDL particle preparations generated with apoA1 isolated from human plasma. The experimental scattering intensity data from deuterated apoA1 within nascent HDL in 12% D2O (Fig. 4A) allowed direct visualization of the overall time-averaged conformation of apoA1 following the simulated annealing method for ab initio structure determination (45). Unexpectedly, this low resolution SANS shape turned out to have an open helical conformation (Fig. 4B, 4C). Other measured parameters for the apoA1 component of HDL include a radius of gyration (Rg) of 51 Å, and the longest dimension (Dmax) of the protein to be approximately 170 Å (Table 1A).

SANS analyses of nascent HDL readily discriminates among HDL models. Because hydrogen and deuterium have markedly different neutron scattering properties, we have used a modified program to predict the scattering curve for all-atom structures of non-spherical macromolecular targets (Supplemental Information) (54). Visual inspection of the calculated theoretical scattering curves of more prominent models of HDL, the Belt (18,19) and Solar Flares (23) models, reveals significant deviations between the predicted and experimental SANS curves (Fig. 4A). Both the Belt and Solar Flares models deviate from the experimental SANS data at lower

scattering angles, and generate oscillatory scattering curves with minima and maxima characteristic of a ring shape at larger scattering angles. In additional investigations we evaluated the scattering behavior of alternative proposed shapes for apoA1 in nascent HDL, such as saddles of differing curvature (64) (Fig. S3), or protein shapes with hairpin loop conformation for what historically is termed Helix 5 (21) (Fig. S4). Again, owing to the closed symmetrical shape of apoA1 in these models, the scattering curve demonstrates marked variation from that observed experimentally, including oscillatory behavior similar to the Belt and Solar Flares models. SANS analyses indicate that closed symmetrical protein structures such as proposed in the Belt, Solar Flares and other ring models do not accurately represent the shape of apoA1 within recombinant nascent HDL, but rather reveal that an open helical conformation for apoA1 is fully supported by the SANS data.

The Double Super Helix model for apoA1 within nascent HDL. While SANS is a powerful approach for defining overall shape at low resolution, it is insensitive to local conformation at the level of amino acids and small peptides. In contrast, HD-MS/MS provides structural and dynamic information about the local environment of the amide protons throughout the polypeptide backbone at the amino acid or short peptide level of resolution, but is relatively insensitive to global particle shape. Hydrogen-deuterium exchange data can provide structural constraints for protein architecture and dynamics in solution, and is frequently used in NMR-based structural studies for the study of the local environment of amino acid residues in various protein secondary structures (helix, pleated sheets, etc.). H/D exchange of amide backbone hydrogen atoms is influenced by both solvent accessibility and protein dynamics, which reflect specific local interactions such as participation in hydrogen-bonds within alpha-helices, hydrophobic contacts, etc.(25).

We therefore sought to integrate for the first time structural information from both contrast variation SANS and HD-MS/MS to extend the low resolution SANS structures for HDL through development of computational models that accommodate both the overall global shape dictated by SANS, as well as the constraints imposed by measured hydrogen-deuterium exchange data for apoA1 within nascent HDL. Thus, a new model of apoA1 was built utilizing the low resolution SANS structure of HDL protein (Fig. 4B, left, Fig. S5) as scaffold for the construction of an

7


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

initial computational model of apoA1. The starting model was energy minimized and refined through iterative calculations of SANS intensity and H/D exchange incorporation factors (See Supplemental Information for detailed methods and Fig. S5, S6, Table S5). Subsequently, additional geometrical constraints (experimentally determined cross-linking/mass spectrometry, fluorescence resonance energy transfer and electron spin resonance coupling distance constraints (20,21,48)) were used to further refine the overall shape of the protein within the computational model. Clashes between atoms produced during modeling were removed by performing energy minimization at each step. Next, a model of the lipid phase of HDL was constructed based upon the experimental SANS structure of the lipid core (Fig. S7) and the model of apoA1 within HDL. The computational model developed for the solution structure for nascent HDL (called the Double Super Helix model - because of the super helical structure of the anti-parallel alpha helical apoA1 chains) was thus obtained by combining all experimental data (SANS, HD-MS/MS, cross-linking/mass spectrometry, fluorescence resonance energy transfer, electron spin resonance coupling) with integrative computational modeling tools (see Supplemental Information).

The predicted structure of apoA1 within the Double Super Helix model of nascent HDL is shown in Fig. 4B (right). Fig. 4A shows the calculated scattering intensity of the Double Super Helix model (blue line), which fits very well the experimental scattering intensity curve. As in the previous Belt and Solar Flares models, the protein in the Double Super Helix model has the two apoA1 chains in predominantly α-helix secondary structure and in anti-parallel orientation (Fig. 4B, right). The open helical conformation of the two apoA1 molecules is readily visualized in Fig. 4C, which provides a different vantage point of the low resolution SANS shape (left), the high resolution protein in the Double Super Helix model (middle) and their overlap (right). The relatively hydrophobic surface of apoA1 is oriented inwards, indicating the groove formed within the helical protein spiral abuts a central lipid core (Fig. 4D). In contrast, the majority of hydrophilic amino acids are oriented toward the outer solvent exposed surface (Fig. 4D). It is worth noting that to accommodate the HD-MS/MS data the N-termini of apoA1 are predicted to have predominantly α-helix secondary structure as previously suggested (65).

More quantitative analyses of the goodness of fit of experimental SANS and HD-MS/MS data and the Double Super Helix model are shown in both Table 1A, and Figs. 4E,F. Calculated χ2 statistics that quantifies differences between the experimentally determined and theoretical scattering intensities shows superior fit for the Double Super Helix model compared to Belt and Solar Flares models. Similarly, a close correlation is observed between experimental HD-MS/MS data and that predicted for the Double Super Helix model at both peptide (Fig. 4E) and amino acid level of resolution (Fig. 4F). The predicted conformation of apoA1 in the Double Super Helix model demonstrates substantially improved fit with experimental HD-MS/MS data compared to that predicted for apoA1 in the Belt model, and similar fit of experimental HD-MS/MS data in the Solar Flares model (Fig. 4F, Table 1A).

The lipid core of nascent HDL is ellipsoidal. To visualize directly the structure of the lipid core of nascent HDL we performed SANS at 42% D2O contrast variation (Fig. 5A, black dotted line). The HDL particles studied this time were generated with non-deuterated human apoA1 isolated from plasma. At 42% D2O, the solvent closely matches the scattering length density of the non-deuterated protein (match point 42%) and to a lesser extent, the lipid polar head groups (match point 30%), thus leaving mainly the acyl chains (match point ~3%) of the lipid visible (63). The experimental radius of gyration obtained for the lipid core is 39.7 Å, which is about 10 Å smaller than the radius of gyration of the protein (51.3 Å). These results confirm that the protein component of HDL is located predominantly toward the outside of the particle while the phospholipid acyl chains occupy a more central location within HDL. The low resolution structure of the lipid core was initially reconstructed from the experimental SANS data obtained from nascent HDL solvated with 42% D2O (Fig 5B, Fig. S7). The overall structure of the HDL particle lipid core visualized is clearly prolate ellipsoid. Next we built a hypothetical initial all atom model for the lipid of nascent HDL. As proposed in earlier HDL models, the hydrophobic portion of the lipids interacts with the hydrophobic surface of protein and the lipid polar head groups are oriented toward the aqueous phase. We thus used as scaffold for the initial model of the lipid both the helical grove partially bounded by the hydrophobic surface of the antiparallel apoA1 chains and the overall ellipsoidal structure of the lipid core visualized by SANS (Fig. S5, S7 and Table S5). Hydrophobic acyl chains of

8


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

phospholipids were oriented toward the hydrophobic protein surface (e.g. Fig. 4D) of the apoA1 helix. The initial model was refined through iterative energy minimizations and evaluations of the goodness of fit of SANS intensity for both protein and lipid core shapes, and amide proton solvent accessibility throughout the entire polypeptide chain. In this manner we optimized the HDL structure such that minimal differences between predicted vs. experimental data (for both SANS and HD-MS/MS) were achieved. Subsequently, reported cross-linking/mass spectrometry, fluorescence resonance energy transfer and electron spin resonance coupling distance constraints (20,21,48) were used as additional geometrical constraints for the overall shape of the protein within HDL, and then a final energy minimization of the intact solvated particle was performed (see Fig. S5, S6, S7, Table S5). Molecular dynamics simulation was also performed (Fig. S9) on the final computational model, the Double Super Helix model of nascent HDL, revealing a thermodynamically stable particle which retained overall helical conformation of apoA1 and preserved both the lipid organization and solvent exposed protruding apoA1 loops in the region previously identified as the LCAT binding site in HDL (aka, the Solar Flares (23)).

The structural model proposed for the lipid within the Double Super Helix model of nascent HDL is shown in Fig. 5C. The overall lipid mesophase is predicted to be predominantly micellar, while adopting a pseudolamellar arrangement in the vicinity of the protein (Fig. 5D). It should be noted that if one assumes that the packing arrangement of the lipid is essentially dictated by the shape of the super helical protein component of nascent HDL, orienting the acyl chains toward the hydrophobic amphipathic protein surface, an ellipsoidal lipid micelle is generated with overall shape compatible with the low resolution structure visualized for the lipid core of HDL (Fig. 5B, C). Comparison of the overall lipid shapes within various HDL models demonstrates the structure posited for the Double Super Helix model is more closely aligned with the experimental SANS scattering data collected in 42% D2O (Fig. 5A). Further, the calculated scattering intensity of lamellar mesophase (bilayer) lipids for both Belt and Solar Flares models are more skewed away from the experimental SANS curve at lower scattering angles, suggesting poorer fit with the experimental data (Fig. 5A). This is confirmed by quantitative analysis of the goodness of fit of the predicted scattering curves

derived from the shape of the lipid core of the Double Super Helix model vs. both Solar Flares and Belt models, where significantly greater deviation from the experimental SANS scattering data is observed with both the Solar Flares and Belt models, as indicated by the substantially larger χ2 statistics (Table 1A).

Global dimensions of the Double Super Helix model of nascent HDL. The overall dimensions of the experimentally determined low resolution (SANS) structures for nascent HDL and the Double Super Helix model are illustrated in Fig. 6A-D. Of note, the particle model is a prolate ellipsoid with global proportions in close agreement with that observed by both SANS and electron microscopy (EM) (Fig. 6A-D, Table 1A). The anti-parallel nature of the two apoA1 alpha helices are predicted to be stabilized by multiple inter-chain salt bridges (Fig. 6C and Table S7), providing structural support for the protein to serve as a super helical scaffolding upon which HDL lipids are transported. Individual particles in EM images correspond to projections of an ellipsoid of 55-65 Å in diameter and approximately 160 Å in length under different orientations. The internal structure of the particles was not clearly discernable on EM, even after an enhancement of the signal-to-noise ratio by classification, alignment and averaging. Nonetheless, some class-averages display striations reminiscent of projections of the proposed Double Super Helix model for HDL. Independent studies employing small angle X-ray scattering yielded similar particle dimensions (Table 1A, Fig. S8). The alpha-helical content of apoA1 within the Double Super Helix model as estimated by Pymol is 68%, similar to that observed by circular dichroism measurements on the recombinant HDL particles (67%), and in close agreement with prior reported values (66). Finally, better than earlier prominent HDL models reported in the literature, all sequence confirmed intra-chain and most inter-chain distance constraints identified for apoA1 based upon mass spectrometry analyses in cross-linking studies (20,48), fluorescence resonance energy transfer (21), or electron spin resonance studies (21), are accommodated by the super helical conformation of apoA1 in the Double Super Helix model (Table 1B, Fig. S4 and Table S6). An inter-chain cross-link previously reported (20,48) that would be structurally forbidden within a single HDL particle (K208-K208) could be readily accommodated by a cross link formed between a pair of particles such as with rouleaux formation.

9


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

DISCUSSION

Biological implications and context with prior studies. The model of nascent HDL presented here is a dramatic and surprising departure from the entrenched view. The open helical low resolution shape of apoA1 visualized by SANS should allow for a highly versatile particle that can accommodate alterations in shape and lipid composition during particle maturation and remodeling, properties that undoubtedly facilitate the ability of the particle to support its transport function. The super helical conformation of apoA1 helps to envision a path of HDL particle genesis from poorly lipidated apoA1 through association of phospholipid and sterol with a gradually unfolding protein. Further, the free energy barrier for assembling the lipid phase posited in the Double Super Helix model should be less than that required for constructing a circumferentially enshrined bilayer during HDL particle genesis such as required by discoidal (i.e. Belt, Solar Flares) models. Moreover, selective uptake, exchange or modification of lipids within HDL during interaction with both its cell based receptors or plasma enzymes will not generate an immediate asymmetric distribution of lipids within the particle (i.e. a “heads” vs. “tails” to the disk), a problem intrinsic to all discoidal/bilayer containing models of HDL that to date has not been adequately addressed. In contrast, lipids in the Double Super Helix model are free to diffuse throughout the entire particle.

Nearly three decades ago Atkinson and colleagues reported the first structural studies of HDL employing SANS with contrast variation (67). While the detailed topography of protein and lipid components remained undefined, the authors concluded that the protein of HDL is located circumferentially relative to lipid within the particle, and suggested a bilayer disk (oblate ellipsoid) model for HDL. Additional studies on reconstituted HDL performed by SAXS without contrast variation, and molecular dynamics simulations (68,69) either lent support to the idea of HDL being discoidal or did not define a low resolution model of nascent HDL with contrast variation. In preliminary SANS studies we similarly could not define the low resolution shape of apoA1 within HDL owing to both a weak protein signal and the overlap between the scattering signal from protein and phospholipid at higher angles. The use of deuterated apoA1 for nascent HDL generation proved key in permitting direct visualization of the open helical conformation of the protein within the

HDL particle. Moreover, the helical architecture of the protein within HDL suggests a predominantly micellar and pseudo-lamellar organization of the lipids within the prolate ellipsoidal shape visualized by SANS for the lipid core of nascent HDL. It is of interest that Weisgraber and colleagues used small-angle X-ray scattering to explore the structure of apolipoprotein E - DPPC particles in solution and observed a “quasi-spheroidal” particle with hair-pin containing apolipoprotein E and an ellipsoidal shaped lipid core with posited phospholipids packing similar to a micelle, which they termed the “twisted-bilayer” model. (70)

Prior computational modeling for nascent HDL has suggested a discoidal particle comprised of a double belt conformation of apoA1 encircling a lipid bilayer (18). This was based upon studies of particles with alternative composition (e.g. 80:1, mol:mol, PC:apoA1) and with an N-terminal (Δ43) truncated apoA1, the initial form of lipid free apoA1 crystallized (71). The HDL particle composition selected for this study, the methods for particle generation, and the detailed functional characterizations performed demonstrate a biologically active particle similar in composition to nascent HDL (96 Å form) generated by macrophages (31). Moreover, the reconstituted HDL preparations examined have similar composition to those used in many recent HDL-based human therapeutic intervention trials (9,28,30,72,73). While the present studies focus solely on a single reconstituted HDL form similar in composition and biological activities to nascent HDL, we cannot exclude the possibility that HDL particles comprised of different protein:lipid compositions and alternative apolipoprotein polypeptides chain length may assume different global architecture and lipid organization. The lipid structural studies performed do not directly visualize the detailed packing arrangement of the lipid of HDL, and thus do not exclude the possibility of an alternative organization.

The computational model of nascent HDL (the Double Super Helix model) was obtained by combining the low resolution structures obtained by SANS with contrast variation and selective deuteration of apoA1 and data from multiple experimental studies of HDL using novel integrative computational methods (e.g. HD-MS/MS, cross-linking/mass spectrometry, fluorescence resonance energy transfer, electron spin resonance coupling). Interestingly, while no crystal structures of lipid free

10


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

apoA1 were used (other than the incorporation of an anti-parallel orientation of apoA1 chains as originally suggested by the lipid free Δ43-N-terminal truncated apoA1 mutant (71)), it is remarkable to note that the conformation of apoA1 in the Double Super Helix model contains several turns found in the recently reported crystal structure of lipid-free full length apoA1 (74). As Fig. 7A shows, there are five turning loops in the reported crystal structure of lipid free apoA1 that start with leucine and most of them end with leucine (L38-L47, L82-A95, L137-L144, L189-A196, L214-L218). Three of these turns (L38-L47, L82-A95 and L214-L218) are also present as turns in the Double Super Helix model of apoA1 (Fig. 7B). Interestingly, a bend is also observed in both chains of apoA1 of the Double Super Helix model at the remaining 2 turns of lipid free apoA1 (Fig. 7B). Moreover, one of these turns (Turn 3, Fig. 7B) partially forms following molecular dynamics simulation of the Double Super Helix model (Fig. S9) and corresponds to a region on apoA1 suggested by Oda and colleagues to adopt a hairpin loop structure (21). These studies relied upon multiply site-specific mutated forms of apoA1 combined with derivatization with detector molecules (e.g. spin adducts or bulky aromatic species). While one of the inter-chain distance constraints observed (e.g. position 133 on each chain) is readily satisfied in the Double Super Helix model (Table 1B), another one (position 146 on each chain) is not (Table 1B). By changing the registry of the two apoA1 chains this inter-chain distance constraint can be accommodated; however, this also results in a structure that no longer satisfies some of the reported cross-links (Fig. S4). It should also be noted that introduction of a loop within the helix 5 region of apoA1 of the Double Super Helix model: (i) reduces the agreement of the predicted SANS scattering curve with the experimental SANS data; (ii) deteriorates the goodness of fit between experimental and theoretical HD-MS/MS deuterium incorporation factors for this region; and (iii) renders some of the reported cross-links forbidden regardless of whether the loop is introduced into the Double Super Helix or Belt models of apoA1 (Fig. S4). In contrast to studies with SANS and HD-MS/MS, which are non-perturbing, studies that employ mutations of apoA1, while informative, need to be interpreted within the context of recognizing that structural and functional alterations to the particle may be introduced. For example, HDL particles generated with recombinant mutant apoA1 that lack all four of the endogenous tryptophan residues, the mutant form of apoA1 often

used as a backbone for additional site-specific mutants studied by fluorescence resonance energy transfer and electron spin resonance, can have markedly impaired cholesterol efflux function depending upon the amino acid used to replace the tryptophan (75). Our present studies do not rule out the possible formation of transient loop structures anywhere throughout the apoA1 polypeptide since the overall shape of the proposed Double Super Helix model arises from the overall time-averaged conformation of apoA1 within the ensemble of various nascent HDL conformations in solution visualized by SANS. Indeed, the Double Super Helix model readily accommodates a highly dynamic conformation of apoA1 while retaining lipid binding properties within the particle. While the precise location of small structural elements along the apoA1 chain of nascent HDL may still be debated, the studies reported here suggest that the overall time-averaged global architecture of apoA1 within nascent HDL (at least for the studied lipid/protein composition) is an open spiral (Fig. 7C and D).

Another remarkable feature of the Double Super Helix model of nascent HDL is that numerous previously reported structural constraints (inter-chain distances based upon cross-linking/mass spectrometry, fluorescence resonance energy transfer and electron spin resonance studies) used to argue in favor of prior models can be readily accommodated within the observed open helical conformation of apoA1 (Table 1B, Table S6). Moreover, some regional structural features proposed in prior studies that were not easily accommodated by the closed belt structures can be more readily accommodated within the open and dynamic apoA1 helical conformation. For example, Sorci-Thomas and colleagues (20) reported that the N- terminus of apoA1 folds back upon the apoA1 chain allowing for detection of a cross-link between inter-chain residues Lys12 and Lys182. Another intra-chain cross-link between Lys94 and the apoA1 chain N-terminus was similarly reported by Davidson and colleagues (48). The Double Super Helix model accommodates the cross-link between Lys12 and Lys182 (Table 1B). Further, small changes in the conformation of the N-terminal region of apoA1 readily bring the N-terminus and Lys94 within permissible distance for cross-linking (Fig. S10).

A structural feature predicted in the Double Super Helix model is the presence of two highly dynamic and solvated apoA1 loops previously identified to support LCAT docking and activation within nascent

11


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

HDL – aka - the so called Solar Flares (23). Molecular simulation studies performed on the Double Super Helix model (Fig. S9) support the dynamic character and solvent accessibility of apoA1 in this region, and show that the “Solar Flares” remain highly solvent exposed over prolonged simulation time (> 4ns), with breaking and re-formation of salt bridges. This contrasts with a recently reported molecular dynamics simulation that claims that the Solar Flares “collapse” in the Solar Flares model due to disruption of one or more predicted salt bridge as visualized in a single time frame of a brief (1 ns) simulation (76). Because no His in their model were protonated, the model they used for simulation differs from that employed in our past and present studies, as well as in other computational studies (e.g. the Belt Model (18). As illustrated in Fig. 8, the experimentally measured deuterium incorporation factors for residues within the peptide corresponding to the Solar Flare loop (e.g. Arg160B-Arg177B) show a high degree of deuterium incorporation (average D0

i=0.8). This is because the residues within this region of apoA1 are very dynamic (derived average per residue unfolding equilibrium constant, Ku

i ~10-2) and their amide

hydrogens exchange rapidly (derived average per residue exchange rate constant, kxc

i ~0.23 s-1), regardless of whether they are involved in H-bonding or not (see Table S3). In contrast, the amide hydrogens of the residues in the adjacent α-helix from the anti-parallel apoA1 chain (Lys77A-Lys106A), by comparison, show reduced deuterium incorporation (average D0

i=0.47), are substantially less dynamic (Ku ~10-3), and their amide hydrogens exchange an order of magnitude slower (kxc ~2.67 x 10-2 s-1) (Fig. 8).

Lipoproteins participate in numerous biological functions including sterol and lipid transport, organismal energy metabolism, and innate immunity. Despite their central role in multiple patho-physiologically relevant processes, as a class, intact lipidated lipoproteins have defied high-resolution structural characterization because of their heterogeneous composition, dynamic/polymorphic nature and large size. This lack of structural definition hinders fundamental understanding of pathobiological processes and potential development of therapeutic interventions. By uniting the complementary and non-perturbing approaches of SANS and HD-MS/MS in combination with computational and bioinformatic methods, we have applied a multidisciplinary and broadly applicable methodology for the structural study of solution phase macromolecular complexes resistant to traditional high-resolution structural interrogation, such as lipoproteins. Application of this approach revealed a totally unexpected conformation of apoA1 in the nascent HDL particle – a double super helix. Overall, the protein seems to play the role of a backbone giving mechanical strength to the HDL particle, while acting as an interface for both lipid binding/transport and specific interactions with plasma enzymes and cell receptors. The observed open helical shape of apoA1, lipid arrangement and overall ellipsoidal particle shape accounts for geometric, biophysical and biochemical data reported to date. It also has not escaped our attention that a helical shape for a protein scaffolding for a lipoprotein envisages a credible generalized pathway for lipoprotein biogenesis, maturation and remodeling.

ACKNOWLEDGEMENTS

We thank Dimitri Svergun and Maxim Petouchov (EMBL Hamburg c/o DESY, Notkestrasse 85, 22603 Hamburg Germany) for assistance in performance of SAXS analyses. We thank Dr. Segrest for providing the PDB file of the Belt model of HDL. This study was supported by National Institutes of Health grants P01 HL076491-055328, P01 HL077107-050004, P01 HL087018-02001, 1R15 GM070469-01, and the Case Western Reserve University/Cleveland Clinic Clinical and Translational Science Award (grant 1KL2RR024990).

FOOTNOTE

The Double Super Helix model reported in this paper has been deposited in the Protein Model Database, http:/www.caspur.it/PMDB (accession no. PM0075984).

12


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

REFERENCES

1. Tall, A. R., Yvan-Charvet, L., Terasaka, N., Pagler, T., and Wang, N. (2008) Cell Metab 7(5), 365-375

2. Trigatti, B., Rigotti, A., and Krieger, M. (2000) Curr Opin Lipidol 11(2), 123-131 3. Barter, P. J., and Rye, K. A. (2006) Curr Opin Lipidol 17(4), 399-403 4. Rader, D. J. (2007) Nat Clin Pract Cardiovasc Med 4(2), 102-109 5. Assmann, G., and Gotto, A. M., Jr. (2004) Circulation 109(23 Suppl 1), III8-14 6. Plump, A. S., Scott, C. J., and Breslow, J. L. (1994) Proc Natl Acad Sci U S A 91(20), 9607-

9611 7. Chapman, M. J. (2006) Pharmacol Ther 111(3), 893-908 8. Linsel-Nitschke, P., and Tall, A. R. (2005) Nat Rev Drug Discov 4(3), 193-205 9. Nissen, S. E., Tsunoda, T., Tuzcu, E. M., Schoenhagen, P., Cooper, C. J., Yasin, M., Eaton, G.

M., Lauer, M. A., Sheldon, W. S., Grines, C. L., Halpern, S., Crowe, T., Blankenship, J. C., and Kerensky, R. (2003) Jama 290(17), 2292-2300

10. Shah, P. K., Nilsson, J., Kaul, S., Fishbein, M. C., Ageland, H., Hamsten, A., Johansson, J., Karpe, F., and Cercek, B. (1998) Circulation 97(8), 780-785

11. Navab, M., Anantharamaiah, G. M., Reddy, S. T., Hama, S., Hough, G., Grijalva, V. R., Wagner, A. C., Frank, J. S., Datta, G., Garber, D., and Fogelman, A. M. (2004) Circulation 109(25), 3215-3220

12. Tall, A. R. (2007) N Engl J Med 356(13), 1364-1366 13. Nissen, S. E., Tardif, J. C., Nicholls, S. J., Revkin, J. H., Shear, C. L., Duggan, W. T., Ruzyllo,

W., Bachinsky, W. B., Lasala, G. P., and Tuzcu, E. M. (2007) N Engl J Med 356(13), 1304-1316

14. Barter, P. J., Caulfield, M., Eriksson, M., Grundy, S. M., Kastelein, J. J., Komajda, M., Lopez-Sendon, J., Mosca, L., Tardif, J. C., Waters, D. D., Shear, C. L., Revkin, J. H., Buhr, K. A., Fisher, M. R., Tall, A. R., and Brewer, B. (2007) N Engl J Med 357(21), 2109-2122

15. Davidson, W. S., and Thompson, T. B. (2007) J Biol Chem 282(31), 22249-22253 16. Thomas, M. J., Bhat, S., and Sorci-Thomas, M. G. (2008) J Lipid Res 49(9), 1875-1883 17. Phillips, J. C., Wriggers, W., Li, Z., Jonas, A., and Schulten, K. (1997) Biophys J 73(5), 2337-

2346 18. Segrest, J. P., Jones, M. K., Klon, A. E., Sheldahl, C. J., Hellinger, M., De Loof, H., and

Harvey, S. C. (1999) J Biol Chem 274(45), 31755-31758 19. Koppaka, V., Silvestro, L., Engler, J. A., Brouillette, C. G., and Axelsen, P. H. (1999) J Biol

Chem 274(21), 14541-14544 20. Bhat, S., Sorci-Thomas, M. G., Alexander, E. T., Samuel, M. P., and Thomas, M. J. (2005) J

Biol Chem 280(38), 33015-33025 21. Martin, D. D., Budamagunta, M. S., Ryan, R. O., Voss, J. C., and Oda, M. N. (2006) J Biol

Chem 281(29), 20418-20426 22. Maiorano, J. N., Jandacek, R. J., Horace, E. M., and Davidson, W. S. (2004) Biochemistry

43(37), 11717-11726 23. Wu, Z., Wagner, M. A., Zheng, L., Parks, J. S., Shy, J. M., 3rd, Smith, J. D., Gogonea, V., and

Hazen, S. L. (2007) Nat Struct Mol Biol 14(9), 861-868 24. Gingras, A. R., Bate, N., Goult, B. T., Hazelwood, L., Canestrelli, I., Grossmann, J. G., Liu, H.,

Putz, N. S., Roberts, G. C., Volkmann, N., Hanein, D., Barsukov, I. L., and Critchley, D. R. (2008) Embo J 27(2), 458-469

25. Radford, S. E., Dobson, C. M., and Evans, P. A. (1992) Nature 358(6384), 302-307 26. Cowieson, N. P., Kobe, B., and Martin, J. L. (2008) Curr Opin Struct Biol 18(5), 617-622 27. Fleishman, S. J., Unger, V. M., and Ben-Tal, N. (2006) Trends Biochem Sci 31(2), 106-113

13


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

28. Shaw, J. A., Bobik, A., Murphy, A., Kanellakis, P., Blombery, P., Mukhamedova, N., Woollard, K., Lyon, S., Sviridov, D., and Dart, A. M. (2008) Circ Res 103(10), 1084-1091

29. Imaizumi, S., Miura, S., Nakamura, K., Kiya, Y., Uehara, Y., Zhang, B., Matsuo, Y., Urata, H., Ideishi, M., Rye, K. A., Sata, M., and Saku, K. (2008) J Am Coll Cardiol 51(16), 1604-1612

30. Ibanez, B., Vilahur, G., Cimmino, G., Speidl, W. S., Pinero, A., Choi, B. G., Zafar, M. U., Santos-Gallego, C. G., Krause, B., Badimon, L., Fuster, V., and Badimon, J. J. (2008) J Am Coll Cardiol 51(11), 1104-1109

31. Duong, P. T., Collins, H. L., Nickel, M., Lund-Katz, S., Rothblat, G. H., and Phillips, M. C. (2006) J Lipid Res 47(4), 832-843

32. Matz, C. E., and Jonas, A. (1982) J Biol Chem 257(8), 4535-4540 33. Peng, D. Q., Wu, Z., Brubaker, G., Zheng, L., Settle, M., Gross, E., Kinter, M., Hazen, S. L.,

and Smith, J. D. (2005) J Biol Chem 280(40), 33775-33784 34. Meilleur, F., Contzen, J., Myles, D. A., and Jung, C. (2004) Biochemistry 43(27), 8744-8753 35. Artero, J. B., Hartlein, M., McSweeney, S., and Timmins, P. (2005) Acta Crystallogr D Biol

Crystallogr 61(Pt 11), 1541-1549 36. Niemann, H. H., Petoukhov, M. V., Hartlein, M., Moulin, M., Gherardi, E., Timmins, P.,

Heinz, D. W., and Svergun, D. I. (2008) J Mol Biol 377(2), 489-500 37. Baker, P. W., Rye, K. A., Gamble, J. R., Vadas, M. A., and Barter, P. J. (2000) J Lipid Res

41(8), 1261-1267 38. Andrade, M. A., Chacon, P., Merelo, J. J., and Moran, F. (1993) Protein Eng 6(4), 383-390 39. Jin, L., Shieh, J. J., Grabbe, E., Adimoolam, S., Durbin, D., and Jonas, A. (1999) Biochemistry

38(47), 15659-15665 40. Bolton, A. E., and Hunter, W. M. (1973) Biochem J 133(3), 529-539 41. Gu, X., Trigatti, B., Xu, S., Acton, S., Babitt, J., and Krieger, M. (1998) J Biol Chem 273(41),

26338-26348 42. Johs, A., Hammel, M., Waldner, I., May, R. P., Laggner, P., and Prassl, R. (2006) J Biol Chem

281(28), 19732-19739 43. Guinier, A. (1939) Ann. Physi., 161-237 44. Zaccai, G., and Jacrot, B. (1983) Annu Rev Biophys Bioeng 12, 139-157 45. Svergun, D. I. (1999) Biophys J 76(6), 2879-2886 46. Konarev P. V., V. V. V., Sokolova A. V., Koch M. H. J. and Svergun D. I. (2003) J.Appl.Cryst.

(36), 1277-1282. 47. Petoukhov, M. K., PV; Kikhney, AG; Svergun, D. (2007) Journal of Applied Crystallography

(40), S223-S228 48. Silva, R. A., Hilliard, G. M., Li, L., Segrest, J. P., and Davidson, W. S. (2005) Biochemistry

44(24), 8600-8607 49. Li, H. H., Lyles, D. S., Pan, W., Alexander, E., Thomas, M. J., and Sorci-Thomas, M. G.

(2002) J Biol Chem 277(42), 39093-39101 50. Miranker, A., Robinson, C. V., Radford, S. E., Aplin, R. T., and Dobson, C. M. (1993) Science

262(5135), 896-900 51. Chung, E. W., Nettleton, E. J., Morgan, C. J., Gross, M., Miranker, A., Radford, S. E., Dobson,

C. M., and Robinson, C. V. (1997) Protein Sci 6(6), 1316-1324 52. Guex, N., and Peitsch, M. C. (1997) Electrophoresis 18(15), 2714-2723 53. Sali, A., and Blundell, T. L. (1993) J Mol Biol 234(3), 779-815 54. Merzel, F., and Smith, J. C. (2002) Acta Crystallogr D Biol Crystallogr 58(Pt 2), 242-249 55. Svergun, D. I., Richard, S., Koch, M. H., Sayers, Z., Kuprin, S., and Zaccai, G. (1998) Proc

Natl Acad Sci U S A 95(5), 2267-2272 56. Jorgensen, W. L., Tirado-Rives, J. (1988) Journal of the American Chemical Society 110(6),

1657-1666

14


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

57. Lindahl, E., Hess, B., and an der Spoel, D. (2001) J. Mol. Mod 7(8), 306 58. Serdyuk, I. N., Zaccai, N. R., and Zaccai, J. (2007) Methods in Molecular Biophysics, 776 59. Pardon, J. F., Worcester, D. L., Wooley, J. C., Cotter, R. I., Lilley, D. M., and Richards, R. M.

(1977) Nucleic Acids Res 4(9), 3199-3214 60. Ramakrishnan, V. (1986) Science 231(4745), 1562-1564 61. May, R. P., Nowotny, V., Nowotny, P., Voss, H., and Nierhaus, K. H. (1992) Embo J 11(1),

373-378 62. Willumeit, R., Diedrich, G., Forthmann, S., Beckmann, J., May, R. P., Stuhrmann, H. B., and

Nierhaus, K. H. (2001) Biochim Biophys Acta 1520(1), 7-20 63. Serdyuk, I. N., Zaccai, N. R., and Zaccai, J. (2007), 776 64. Catte, A., Patterson, J. C., Jones, M. K., Jerome, W. G., Bashtovyy, D., Su, Z., Gu, F., Chen, J.,

Aliste, M. P., Harvey, S. C., Li, L., Weinstein, G., and Segrest, J. P. (2006) Biophys J 90(12), 4345-4360

65. Zhu, H. L., and Atkinson, D. (2004) Biochemistry 43(41), 13156-13164 66. Rogers, D. P., Roberts, L. M., Lebowitz, J., Engler, J. A., and Brouillette, C. G. (1998)

Biochemistry 37(3), 945-955 67. Atkinson, D., Small, D. M., and Shipley, G. G. (1980) Ann N Y Acad Sci 348, 284-298 68. Shih, A. Y., Arkhipov, A., Freddolino, P. L., Sligar, S. G., and Schulten, K. (2007) J Phys

Chem B 111(38), 11095-11104 69. Denisov, I. G., Grinkova, Y. V., Lazarides, A. A., and Sligar, S. G. (2004) J Am Chem Soc

126(11), 3477-3487 70. Peters-Libeu, C. A., Newhouse, Y., Hall, S. C., Witkowska, H. E., and Weisgraber, K. H.

(2007) J Lipid Res 48(5), 1035-1044 71. Borhani, D. W., Rogers, D. P., Engler, J. A., and Brouillette, C. G. (1997) Proc Natl Acad Sci

U S A 94(23), 12291-12296 72. Tardif, J. C., Gregoire, J., L'Allier, P. L., Ibrahim, R., Lesperance, J., Heinonen, T. M., Kouz,

S., Berry, C., Basser, R., Lavoie, M. A., Guertin, M. C., and Rodes-Cabau, J. (2007) Jama 297(15), 1675-1682

73. Nicholls, S. J., Cutri, B., Worthley, S. G., Kee, P., Rye, K. A., Bao, S., and Barter, P. J. (2005) Arterioscler Thromb Vasc Biol 25(11), 2416-2421

74. Ajees, A. A., Anantharamaiah, G. M., Mishra, V. K., Hussain, M. M., and Murthy, H. M. (2006) Proc Natl Acad Sci U S A 103(7), 2126-2131

75. Peng, D. Q., Brubaker, G., Wu, Z., Zheng, L., Willard, B., Kinter, M., Hazen, S. L., and Smith, J. D. (2008) Arterioscler Thromb Vasc Biol 28(11), 2063-2070

76. Shih, A. Y., Sligar, S. G., and Schulten, K. (2008) Biophys J 94(12), L87-89 77. Rye, K. A., Garrety, K. H., and Barter, P. J. (1992) J Lipid Res 33(2), 215-224 78. Alexander, E. T., Bhat, S., Thomas, M. J., Weinberg, R. B., Cook, V. R., Bharadwaj, M. S., and

Sorci-Thomas, M. (2005) Biochemistry 44(14), 5409-5419 79. Koukos, G., Chroni, A., Duka, A., Kardassis, D., and Zannis, V. I. (2007) Biochem J 406(1),

167-174 80. Black, S. D., and Mould, D. R. (1991) Anal Biochem 193(1), 72-82 FIGURE. LEGENDS:

FIGURE 1 Characterization of reconstituted nascent HDL particles. Nascent HDL particles were prepared using a modified cholate dialysis method at a molar ratio of 100:10:1, POPC:cholesterol:apoA1(77). Before being used in structural analyses, HDL particles were characterized by a variety of biological, biochemical, and biophysical methods. A, Non-denaturing gel analyses of reconstituted nascent HDL containing purified human apoA1 from healthy donors (rHDL) (left panel) and reconstituted nascent HDL containing recombinant human

15


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

apoA1 purified from E. coli.(rrHDL) (right panel). Purified nascent HDL by gel filtration (10 μg protein) was loaded on a 4-20% gradient gel. The purity of the nascent HDL preparation was confirmed by ND-PAGE analysis. B, SDS PAGE gel analyses of crosslinked apoA1 within rHDL and rrHDL particles. After crosslinking, a broad band was shown at about 50-70 kDa position, indicating each HDL particle contains two apoA1 molecules. The broadened band pattern is consistent with prior reports for an apoA1 dimer in similarly cross-linked HDL.(20,37) C, Stoichiometry of reconstituted HDL. The chemical composition of HDL particles was determined using a variety of methods. D, The size of nascent HDL particles was determined both by comparison with the known diameter of globular standard proteins and by light scattering spectroscopy (23). The results represent an average ± SD of three measurements.

FIGURE 2 Cholesterol efflux and LCAT activity of nascent HDL. A, The cholesterol efflux activity of human plasma isolated HDL (pHDL), reconstituted nascent HDL containing purified human apoA1 from healthy donors (rHDL) and reconstituted nascent HDL containing recombinant human apoA1 purified from E. coli.(rrHDL) were measured by incubating different HDLs with subconfluent J774A.1 murine macrophage cells loaded with [3H] cholesterol. The cholesterol efflux was calculated by radioactivity in the medium divided by the total radioactivity (medium radioactivity plus cell radioactivity) as described under Methods. Note that all three different HDLs have similar cholesterol efflux activity. B, rrHDL activates LCAT similarly as rHDL. LCAT activity was measured as described in Methods. Note that rrHDL retains at least 85% of LCAT activation capacity compared with rHDL. C, Dissociation constants of nascent HDL/LCAT complex determined by surface plasmon resonance spectroscopy. The Kd of binding between the indicated HDL and recombinant human LCAT was determined as described under Methods. Note that rHDL and rrHDL demonstrate similar affinity with LCAT. D, Hanes&Woolf (S/V vs S) plot. The kinetic parameters of LCAT activity on rrHDL are similar to what is reported in the literature (Km, Vmax and Kcat)(78,79). All experimental results represent mean ± SD of three independent experiments.

FIGURE 3 Additional biological activities of nascent HDL. A, Demonstration of anti-inflammatory activity of HDL preparations. The capacity of the indicated HDL preparations (or their individual components) to inhibit TNF-α induced enhanced VCAM-1 protein expression in HUVEC cells was determined by cell based ELISA as described under Methods. Note that both human HDL (pHDL) and reconstituted nascent HDL containing purified human apoA1 from healthy donors (rHDL) prevent VCAM-1 expression following TNF stimulation. B, Demonstration of anti-apoptotic properties of HDL preparations. Apoptosis was induced in HUVEC by 6 h serum starvation. Cells were incubated simultaneously with pHDL, rHDL, apoA1 or POPC and the capacity of the indicated HDL particle (or its components) to inhibit apoptosis was determined as described under Methods. Note that both the pHDL and rHDL similarly protect HUVEC from apoptosis. C, Specific binding of HDL preparations to HDL receptor SR-BI. Binding was determined in SR-B1 and vector transfected 293-T cells by addition of iodinated rHDL. Specific binding was calculated as described in Methods. D, Competition binding assay of SR-B1. The capacity of the indicated HDL preparation to act as competitor and block SR-B1 specific binding of HDL isolated from plasma was determined as described under Methods. HDL isolated from plasma was iodinated and used as ligand, and competition binding studies were performed with 30-fold excess non-labeled plasma HDL, the indicated reconstituted HDL including reconstituted HDL made from recombinant human apoA1 (rrHDL), and small unilamellar vesicles (SUV) made of POPC. SR-B1-specific binding of iodinated human HDL is considered 100%.

FIGURE 4. Super helical structure of apoA1 within nascent HDL revealed by SANS and HD-MS/MS. A, Comparison of experimental SANS intensity from deuterated apoA1-nascent HDL in 12% D2O buffer (black dotted line) with calculated intensity of the Double Super Helix model (blue line), Solar Flares model (green line) and Belt model (red line). B, The SANS low resolution structure of apoA1 reconstructed from the experimentally obtained scattering curve of nascent HDL (left) and the all-atom Double Super Helix model of apoA1 within nascent HDL (right). C, 12% D2O SANS shape (black), apoA1 in the Double Super Helix model (red-blue), and their overlap from a viewing angle that emphasizes the helical conformation of the protein; N-termini are colored in dark red/blue and C-termini in light red/blue. D, Map of “patch” hydrophobicity of apoA1 within nascent HDL in the Double Super Helix model. Like the hydrophobic index, the patch hydrophobicity is

16


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

in the range [0,1], and is determined for each amino acid residue by averaging the hydrophobic indices of neighboring residues (e.g. within 15 Å) located on the same side of protein, i.e. either inside (facing lipid) or outside (facing solvent) (80). E, Correlation (R) between the experimentally measured deuterium incorporation factors (D0) for apoA1 peptic peptides from nascent HDL versus the predicted (calculated) D0 values for both the Belt (open circle) and Double Super Helix (open triangle) models of nascent HDL. F, The "goodness of fit" of the Double Super Helix, Solar Flares, and the Belt models at the individual residue level. The probability correction factor (PCF) plotted on the Y axis represents the multiplier needed to adjust the exchange probability of individual residues within the model to match that determined by hydrogen/deuterium exchange mass spectrometry studies. The dotted line at unity represents the PCF of a given amino acid within the model that possesses a conformation with the same H/D exchange as observed by experimental measurement.

FIGURE 5. Prolate ellipsoid structure of the lipid core within nascent HDL revealed by SANS. A, Comparison of experimental SANS intensity (black dotted line with error bars) at 42% D2O with the scattering intensity of the Double Super Helix (blue line), Solar Flares (green line) and Belt models (red line). B, The low resolution SANS structure of lipid core within nascent HDL reconstructed from the scattering intensity of nascent HDL in 42% D2O solution. C, The all atom model of lipids in the Double Super Helix model of nascent HDL illustrating both psuedo-lamellar and micellar organizational features. The phosphatidylcholine head groups are colored purple, the phospholipid acyl chains are green and cholesterol is orange (left). The prolate ellipsoid view shows the pseudo-lamellar arrangement of lipids where the helical protein overlies (note the acyl chain end on end orientation within the grove where the apoA1 has been removed) (right). The cross sectional view illustrates the micellar-like packing of lipids. The two apoA1 chains are shown as red and blue.

FIGURE 6. Geometrical dimensions and electron microscopy imaging of nascent HDL. A, Overlay of the SANS low resolution structures of apoA1 (light-blue) and lipid core (green) within nascent HDL. The lipid core structure shows clear invaginations that are approximately filled by protein. B, Overlay of apoA1 and lipid all atom Double Super Helix model of nascent HDL. The super helical shape of the protein is emphasized by the semi-transparent representation of the lipid core, in which phospholipid is depicted in green, and cholesterol in orange. C, Illustration of the antiparallel amino acid arrangement proposed in the Double Super Helix model of apoA1 in nascent HDL. The two apoA1 molecules are aligned in a head to tail anti-parallel arrangement using a helix 5/helix 5 registry (yellow). The N-termini are colored with dark red/blue and the C-termini are colored with light red/blue. Twenty eight inter-chain salt bridges that stabilize the anti-parallel double super helical conformation of apoA1 are also illustrated as sticks. D, Characteristic negative stain electron microscopic image of nascent HDL, along with two expanded views of the indicated insets. The scale bar corresponds to 15 nm.

FIGURE 7. The Double Super Helix model of nascent HDL. A and B Illustration that apoA1 within nascent HDL retains structural features of full length lipid free apoA1. A, The crystal structure of lipid free apoA1 (74) with residues in the turning loops colored in yellow: L38-L47 (Turn 1); L82-A95 (Turn 2); L137-L144 (Turn 3); L189-A196 (Turn 4); L214-L218(Turn 5). B, Turns in lipid free apoA1 mapped onto the Double Super Helix model of apoA1. Turns in lipid free apoA1 which no longer correspond to turns within the Double Super Helix model are colored in cyan. Turns 1, 2 and 5 from lipid free apoA1 are also present in the Double Super Helix model. C, Locations of the LCAT interaction site (residues 159-170, yellow) and MPO binding site (residues 190-203, light blue) in the Double Super Helix model of apoA1. Two residues, Tyr-166 and Tyr-192, which undergo site-specific oxidative modifications within human atherosclerotic lesions, are shown in black. D, The all atom Double Super Helix model of nascent HDL with known protein-protein interaction sites; POPC (green) and cholesterol (orange) are viewed in semitransparent mode. The dotted line shown represents the hypothetical overlay of a discoidal model of HDL, showing overall similar size.

FIGURE 8. The “Solar Flare” region of nascent HDL in the Double Super Helix model. A, The predicted solution structure of peptide Leu159-Arg177 (blue) in the chain B of apoA1 in the Double Super Helix model of nascent HDL, and the adjacent α-helix from chain A (Lys77-Lys106, red). B, Summary of average experimentally determined deuterium incorporation factors (D0), average unfolding constants (Ku) and average exchange rate constants (kxc) for the two peptides shown.

17


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

TABLE 1A Radius of gyration, global dimensions and goodness of fit of HDL models with experimental SANS, EM and HD-MS/MS data. Physical parameters including radius of gyration (Rg) and overall dimension of the Double Super Helix model of nascent HDL are more consistent with the values measured both by SANS and SAXS compared with the existing models of nascent HDL. Calculated χ2 statistics that quantifies differences between the experimentally determined and theoretical scattering intensities shows superior fit for the Double Super Helix model compared to Belt and Solar Flares models. More quantitative analysis of goodness of fit of HDL models with HD-MS/MS data including Root-mean square deviation (RSMD), probability correction factor (PCF) and correlation coefficient suggest that the Double Super Helix model fits better with the experimental data than the Belt and Solar Flares models.

TABLE 1B Amino acid distances between α and β chain of apoA1 within nascent HDL. The reported interchain structural constraints measured by Fluorescence Resonance Energy Transfer (FRET), Electron Spin Resonance (ESR) and sequence confirmed lysine cross links were examined in the Double Super Helix model.

18


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

rHDL

rrHDL

250kD

100kD

50kD

25kD

15kD

Before

A

1712.210.4

8.1

7.1

Diameter(nm)

17

12.210.4

8.1

7.1

Diameter(nm)rH

DLrrH

DLB

CStoichiometry of reconstituted nascent HDL

POPC Cholesterol ApoA1

Theoretical

Measured (rHDL)

100 10

89.3+4.8 9.7 + 0.5

1

1

Cholate

0

<0.01

Measured (rrHDL) 105 +1 7.4 + 0.4 1 <0.01

DSize of reconstituted nascent HDL

ND-PAGE

Light-scattering

Diameter

96 Ao

110 Ao+ 1

+ 1

Diameter

96 Ao

113 Ao+ 2

+ 1

rHDL rrHDL

150kD

75kD

37kD

20kD

10kD

apoA

1

After

crosslinkingBefo

reAfte

r

crosslinking

After

250kD

100kD

50kD

25kD

15kD

150kD

75kD

37kD

20kD

10kD

rHDL

rrHDL

Figure. 1

19


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

Kd (µM) 0.9

rHDL

Dissociation constants of human recombi-nant LCAT binding to rHDL and rrHDL bysurface plasmon resonance

rrHDL

1.4

C

-10 3020100

D

[S] (µM)

[S]/

V(µ

M/n

mol

/hr/

µgLC

AT)

2

4

6 Reaction Kinetics of HDL Particles with LCAT

1.0 + 0.4 9.5 + 1.0 11.2 + 4.5

1.7 + 0.3 20.5 + 0.9 12.1

rrHDL

rHDLb

µM nmol CE/hr/µg nmol CE/hr/µg/µM

1.0 + 0.2rrHDLa

ApparentKm

ApparentVmax

ApparentVmax/Km

a, Alexander ET et al., 2005 Biochemistryb, Chisholm JW, et al., 1999 Journal of Lipid Research

LCAT

activ

ity(%

)

B28

14

0

Contro

lpH

DLrH

DLrrH

DL

HD

L-m

edia

ted

mac

roph

age

chol

este

role

fflux

(%)

A100

50

0

rHDL

rrHDL

Figure. 2

20


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

16

8

0

p < 0.05 p < 0.05

p = 0.35

Ann

exin

posi

tive

(%)

Serum+ POPC

+ ApoA1

+ rHDL

+ pHDLNA

+ - - - - -

p = 0.33

P<0.001 P<0.05

P<0.05

Fold

incr

ease

VC

AM

-1pr

otei

n

0

5

1

3

+ rHDLNA

+ pHDL

+ ApoA1

+ POPC

TNF - + + + + +

P<0.69P<0.11

0 20 60 100 150

SR

-B1

spec

ific

bind

ing

(dpm

/µg

cell

prot

ein)

SR-B1

Vector

180

0

60

120

[I-125 rHDL] (µg/ml) NA

SR

-B1

spec

ific

bind

ing

(%)

100

50

0

+ pHDL

+rrHDL

+POPC

C D

BA

Figure. 3

21


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

Scattering Vector (1/A)O

log

I

0 0.1 0.2

-7

-3

-5

-6

-4

-2

-8

Experimental dataDouble Super HelixSolar FlareBelt

B

C

Nα

Nβ

Cβ

Cα

0.5-10-0.5

PatchHydrophobicity

z

yx

z y

x

Theo

retic

al

Hydrogens Exchanged

R=0.948

16

8

01

2401202401201

PC

F

Figure. 4

Actual

Chain A Chain B

A

D E

F

0 5 10 15

0

5

10

15

R=0.801

Double Super HelixSolar FlareBelt

22


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

B

Scattering Vector [1/A]O

-7

-3

1

-1

-5

0

-6

-4

-2

log

I

0 0.1 0.2

-8

Experimental dataDouble Super HelixSolar FlareBelt

A

C

Figure 5

23


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

Asp1

Gln243

Gly129Gly129

Asp1

Gln41

Gly81

Gln41 Ser201

Thr161

Ser201

Thr161

Gln243

Pro121

Gly81

Pro121

Nα

Nβ

Cβ

Cα

B

AO

171

73 AO

Fig. 3

171

AO

74 AO

A C

D

Figure 6

24


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

Nα

Nβ

Cβ

Cα

MPObinding

site

LCATbinding

site

Y192

Y166

D

A B

C

Turn 1

Turn 2

Turn 3

Turn 4

Turn 5

Turn 1

Turn 2

Turn 3

Turn 4Turn 5

Turn 1

Turn 2

Turn 3

Turn 4

Turn 5

N

C

Nα

Nβ

Cβ

Cα

MPObinding

site LCATbinding

site

Y192

Y166

Figure 7

25


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

Figure 8

A B

D0 Ku Kxc (S-1)

α-Helix (Lys77A - Lys106A)

Loop (Arg160B - Arg177B)

Fold difference

0.47

0.80

~ 2 ~ 10 ~ 10

2.04 x 10-3

2.72 x 10-2

2.67 x 10-2

0.229

26


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

Lipid

2,255

497

145

Models

Belt

Solar Flares

Double SuperHelix

12.7

6.9

1.1

Protein

Correlation

PCFd(α)

0.30

0.21

0.794

0.915

0.946

RMSDb

Peptide

2.39

1.65

0.82

1.50 1.72

0.29

0.24

PCF(β)

30.2

33.7

38.4

47.2

53.4

52.1

Experimental

51.3 39.7

LipidProtein

Radius ofo

(A)

SANS

Dimensionso

(A)

l x w x h

171 x 74 x 69

gyration coefficientb

Table 1A. Goodness of Fit of HDL models with experimental SANS, EM and HD-MS/MS data

171 x 73 x 67

98 x 98 x 45

140 x 115 x 50

EM 160 x 60

SAXS

HDL

52.0 167

a χ2 stands for the standard deviations between the experimental scattering data and calculated scattering dataof nascent HDL models obtained from our modified SASSIM program54.b Correlation coefficient was correlation between theoretical hydrogen-deuterium exchange probabilities of pepticpeptides in the model and the actual experimentally measured deuterium incorporation within apoA1 pepticpeptides.c RMSD stands for root-mean square deviation. It quantifies the difference between the experimental deuteriumexchange data and the calculated ones.d PCF stands for probability correction factor, which represents the multiplier needed to adjust theoretical deuteri-um incorporation of individual residues based upon the HDL models to match that which was actually experimen-tally determined

χ2, a

27


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

FRET (22.7 A)co

FRET (23.5 A)co

FRET (24.0 A)co

FRET (28.8 A)co

FRET (30-35 A)co

ESR (<15 A) c

ESR (<15 A)c

ESR (17-19 A)c

o

o

o

MS/MS (12 A) e,f,go

MS/MS (12 A) fo

MS/MS (12 A) fo

MS/MS (12 A) e,go

MS/MS (12 A) e,f,go

MS/MS (12 A) fo

MS/MS (12 A) fo

MS/MS (12 A) fo

Table 1B. Amino acid distance constraints between α and β chains of apo A1 withinnascent HDL obtained from fluorescence resonance energy transfer (FRET), HPLCtandem mass spectrometry (MS/MS) and electron spin resonance coupling (ESR).

Residues

W50-L230W72-A210

L90-A190W108-L170

K40-K239K118-K140K12-K182

Double Helixmodel (A)

o

22.8a

30.2

26.226.6

16.1

30.1b

17.3d

15.618.1

160.169.164.2

10.7

6.243.810.9

Methods

Q132-Q132

a The calculated distances between two residues in the Double Super Helix model of nascent HDL are from thefar end of the side chain of Trp to the far end of AEDANS acceptor assuming free rotation of the side chain ofamino acids.b The calculated distances in the Double Super Helix model of nascent HDL are between ε-amine groups oftwo lysines assuming free rotation of the side chain of lysines.c Martin et al, 2006 JBC;d Li et al., 2001, JBC;e Bhat et al, 2005, JBC;f Silva et al, 2005, Biochemistry;g Bhat et al., 2007, Biochemistryh not compatible with 5/5registry;i The calculated distances between two residues in the Double Super Helix model of nascent HDL are from theoxygen atom of methane thiosulfonate nitroxide spin label (MTSSL) on one residue to the oxygen atom ofMTSSL of the other residue.

K59-K208K77-K195

K208-K208h

K88-K118h

K96-K118h

K133-K133i

E146-E146i

L134-L134i

MS/MS (7.7 A) go

6.4K133-K140

28


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/

Supplemental Information

The Double Super Helix Model of High Density Lipoprotein

Zhiping Wu, Valentin Gogonea, Xavier Lee, Matthew A. Wagner, Xin-Min Li, Ying Huang, Arundhati Undurti, Roland P. May, Michael Haertlein, Martine Moulin, Irina Gutsche, Giuseppe Zaccai, Joseph DiDonato, Stanley L. Hazen

In the main text and Supplemental Information, we use the following abbreviations for various HDL preparations, pHDL (isolated human plasma HDL); rHDL (reconstituted nascent HDL containing purified human apoA1 from healthy donors); rrHDL (reconstituted nascent HDL containing recombinant human apoA1 purified from E. coli.).

EXPERIMENTAL PROCEDURES

SANS and SAXS.

Neutron scattering, similar to NMR, is a low energy structural technique that provides a non-perturbing structural analysis in solution. Unlike X-Ray scattering, the neutron beam used is “cold” (ie low energy) and interacts weakly with the nuclei without perturbing the electronic structure (1). Samples are thus not structurally altered (2-4).

SANS Experiment. Small angle neutron scattering experiments were carried out at the instrument D22 of the Institut Laue-Langevin, Grenoble, France. D22 (http://www.ill.eu/d22) is a classical pinhole camera that provides the highest neutron flux among all comparable SANS instruments in existence (5). Data collected from two positions of the detector (2 and 5.6 m with collimation lengths of 2.8 and 5.6 m) covered the momentum transfer (q) range from 0.008 to 0.35 Å-1. q is defined as 2π (sin θ)/λ, where 2θ is the scattering angle . The wavelength λ of the neutron beam used was 6 Å. Data beyond 0.25 Å-1 were too noisy to analyze.

To understand the organization of the protein and lipid in HDL contrast variation ([D2O] : [H2O + D2O] ratio) was required. The HDL samples were measured in 0, 12, 42 and 90% D2O solution to assure the required levels of contrast. From the corresponding scattering curves two important parameters were obtained for each contrast: the radius of gyration, Rg, and the intensity at 0 angle I(q=0). Rg were obtained based on the Guinier approximation

(6), which is valid for q values (Rgq)2<1 (Equation 1).

limq->0 Icorr (q) = I(0)exp(−Rg2q2⁄3) （Equation 1）

The logarithmic intensities varied linearly with q2 in the chosen Guinier range and the obtained Rg were stable against slight change in q-range. The molecular mass (M) of the HDL particle was calculated as 200,000 Da based on the absolute scale measured I(q=0) (equation 2), in which C (mg/cm3) is the concentration of HDL, ∑b (cm) is the particle scattering length, V (cm3) is the particle volume, ρ0

(cm-2) is the solvent scattering density, and NA is Avogadro’s number; the term in brackets is the particle excess scattering length normalized to unit molecular mass) (7). The molecular mass calculated from the data corresponded to one HDL particle.

I(0) ⁄C=NA[(∑b −ρ0V)⁄M]2 M (Equation 2)

The program DAMMIN(8), a simulated annealing method, was used to calculate a low resolution model of the lipid core from the scattering curve of the sample in 42% D2O. Likewise the 12% scattering curve was used for the modeling of the protein. In the approach, the starting structure is a sphere of diameter equal to the maximum particle dimension, which was estimated from the scattering curve, via the distance distribution function P(r). The sphere was filled with much smaller spherical dummy atoms with their size determined by the highest value of momentum transfer (q) of the scattering curve. To enhance the signal to noise ratio of the sample in 12% D2O, deuterated apoA1 was used for the reconstitution of the HDL complex. The χ value of the simulated annealing modeling for lipid is less than 1, and 5.2 for the protein. Both are in the excellent or reasonable range of statistics.

SAXS Experiment. Synchrotron X-ray scattering data of nascent HDL in solution were collected at the X33 beamline (DESY, Hamburg) at particle concentrations ranging from 2 to 16.0 mg/ml. At a sample-detector distance of 2.7 m, the range of

1

2

momentum transfer 0.01 < q < 0.5 Å–1 was covered (λ = 0.15 nm the X-ray wavelength). The data were processed with program PRIMUS (9) using standard procedures. The forward scattering intensity I(q=0) and the radius of gyration (Rg) were evaluated with program AUTORG (10) using the Guinier approximation. The effective molecular mass of the solute (MM) was estimated by comparison of the forward scattering intensity with that from reference solutions of bovine serum albumin (MM = 66 kDa).

All small angle scattering studies were performed under temperature controlled conditions and used reconstituted HDL generated from human apoA1 isolated from human plasma (rHDL) with the exception of deuterated recombinant apoA1 generated in E. coli (rrHDL), which was used in 12% contrast variation SANS studies with the purpose of enhancing the scattering signal of the protein without perturbing the overall structure of reconstituted nascent HDL.

SANS experiments were performed on 5 different occasions (multiple days each time) at the Institute Laue-Langevin, Grenoble, France, using D22, the best neutron beam-line. Experiments were performed on nascent HDL preparations generated from isolated human apoA1 (from plasma) (rHDL) as well as recombinant apoA1 from E. coli (His-tagged and deuterated) (rrHDL). Studies at both SANS facilities using nascent HDL from these two sources of apoA1 showed identical results (ie – rHDL and rrHDL showed similar SANS characteristics at all D2O levels examined; the rrHDL was used primarily for the 12% contrast variation studies; HDL from either apoA1 source showed indistinguishable shape for the lipid core in 42% contrast variation studies).

All SANS studies were performed in PBS buffer under temperature-controlled conditions. Initial measurements using higher concentrations of nascent HDL (e.g. >5 mg protein/ml) indicated a higher degree of aggregation (>5%) so all studies reported were limited to < 2 mg protein/ml. For the deuterated apoA1 containing nascent HDL (rrHDL) and high (>50% D2O) studies, protein concentration was kept < 1.3 mg/ml. Data acquisition for SANS was typically between 1-6 hours per sample under temperature control, and thus produced a time-averaged conformation.

SAXS studies only provided geometric parameters of the entire particle (Table 1, Fig. S8).

Data processing. Data reduction was done at the facilities where collected (ILL and Juelich Center for Neutron Sciences). Data reduction consists of the calibration of the neutron detector using a cadmium sample and minimizing the mechanical distortion of the detector using a sample of water. Subtraction of the background scattering followed these two processes. To ensure minimal error in the measurement of the background, the same buffer without protein was used for data acquisition under comparable conditions as the HDL-containing sample.

Final data were initially analyzed with the program PRIMUS (9) , followed by the program GNOM, (8) (that calculates the P(r) function using indirect Fourier transform). The goodness of fit of P(r) function ranged from good to excellent in all studies. In initial samples with high D2O content the distance distribution function, P(r), showed a small tail at high distances, suggesting the presence of low levels of aggregated particles (1). Use of lower concentrations of HDL (< 1.3 mg protein/ml) resulted in essentially a monodispersed system that produced the overall shape of the P(r) function on which the low resolution SANS models are based (8).

The program Dammin (8) was further used for generating the SANS shapes of the molecular components based on the output of GNOM. The χ statistics were close to 1 and 5 for the models of lipid core (42% D2O) and the deuterated protein component (12%) respectively. All programs used are part of the modeling suite at EMBL, Hamburg (8).

Electron Microscopy Studies. Negatively stained HDL was obtained by applying a solution of HDL particles to the clear side of carbon on a carbon-mica interface and stained with 2 % (w/v) uranyl acetate. HDL particle aggregation was minimized by limiting the concentration of the nascent HDL preparation for the EM studies (ie < 0.2 mg/ml). Images were recorded under low-dose conditions with a JEOL 1200 EX II microscope at 100 kV. Selected negatives were then digitized on a Zeiss scanner (Photoscan TD) at a step size of 14 micrometer giving a pixel size of 3.5 Å at the specimen level. 12456 sub-frames containing individual HDL particles were selected interactively from several micrographs using boxer, ctf corrected and low-path-filtered at 20 Å. Subsequent data processing was performed with the Imagic package. The data set centered by translation was subjected to multivariate statistical analysis and

3

classification that confirmed the ellipsoidal shape of the particles and enabled visualization of some internal striations. Dimensions of different class averages allowed interpretation of the images as projections under different orientations of an about 160 Å long ellipsoid approximately 55-65 Å in diameter. A small proportion (<3% combined by SANS) of aggregates and 200 Å long ellipses were also detected. Given the obvious similarity in general shape between the obtained class averages and re-projections of the SANS model of HDL, the latter were used for an additional round of multi-reference alignment in order to orient the views in the same way for a visual comparison and eventually to better reveal the internal structure of the images. However, no clear accentuation of the internal structural patterns was discernable. Thus, no attempts were made to further proceed to a three-dimensional reconstruction.

Theoretical issues and validation studies governing incorporation of H/D exchange data as a means of refining solution structure. We wished to develop a computational model of nascent HDL that better matched both SANS and H/D exchange experimental data. Before applying the co-refinement process of incorporating H/D exchange data to the low resolution SANS structure, we first validated the approach of incorporating H/D exchange data to several proteins for which both high resolution (crystallographic) structures and H/D exchange data (from either NMR or mass spectrometry) were available. Below we first describe the theoretical framework considered when incorporating H/D exchange data into a molecular model. We then validated the methodology and its implementation into computer code through application to multiple test proteins as described below.

Computational methodology for hydrogen deuterium exchange modeling. In prior studies we developed a computer program (DEXANAL) (11), which implements a theoretical method for determining the experimental per residue deuterium incorporation factors (D0

i) from measured D0 of peptic peptides, the residue unfolding constants (Ku

i) and the predicted rates of H/D exchange (kxc

i) for individual residues (i). As described below, we have validated (vide infra) our methodology on a set of globular proteins like lysozyme, cytochrome C, and thrombin by using either H/D exchange rates measured by NMR, for lysozyme (12), or D0 for peptic peptides measured by MS for cytochrome C (13) and thrombin (14). When

a three-dimensional (3D) structure (pdb from crystallographic, NMR or molecular modeling) is supplied together with the experimental data (either H/D exchange rates, D0 values, or a combination of both), the DEXANAL program has been designed to produce a set of H/D exchange probabilities (XPi) that are used to predict D0

i, Kui, and kxc

i, values for that particular model, and assesses the quality of the 3D input structure/model by gauging the difference between the unknown solution structure, as represented by the experimental D0

i values, and the model as described by the XPi.

The H/D exchange rate constant, kxci, for each residue

can be calculated from the experimental per reside deuterium incorporation factor, D0

i, and the exchange time as follows:

kxci =

− ln 1− D0i( )

t (Equation 3)

The residue unfolding constant, Kui, (the reciprocal of

the NMR protection factor) can also be obtained from kxc

i by using the intrinsic H/D exchange rate constant (kch

i) determined for individual residues as described by Englander and his coworkers (15):

Kui =

kxci

kchi

(Equation 4)

For an ensemble of molecules in solution with various conformations, XPi is defined as the ratio of the number of molecules in which the amide H of residue i exchanged to deuterium (ND

i) to the total number of molecules:

XPi =

NDi

NDi + NH

i (Equation 5)

When using a single molecular model (and not an ensemble), XPi for each residue i in the molecular model can be calculated as the product of two quantities that encode structural information about residue i (size, chemical composition) and its interactions with the rest of the system:

XPi = RAIi * BAIi (Equation 6)

This relationship is a basic equation describing the parameters which impact upon the exchange

4

probability XPi and incorporates information about the local solvent accessibility and protein dynamics at residue i. BAIi is the backbone accessibility index (0 ≤ BAIi ≤ 1), and RAIi is the residue accessibility index (0 ≤ RAIi ≤ 1) of residue i.

A residue conformation-adjustment factor can also be defined: this factor, called the probability correction factor (PCFi), is the ratio of the experimentally determined per residue deuterium incorporation factor D0

i (obtained by partitioning the experimentally measured D0 for overlapping peptic peptides) and the calculated probability XPi:

PCFi =

D0i

XPi (Equation 7)

PCFi for residue i is 1.0 for a residue whose amide H experiences the same degree of protection to exchange as in solution (i.e. when the PCFi =1, the predicted deuterium exchange based upon the molecular model is identical to that which was experimentally determined). Thus, PCF gauges the difference in protection to exchange of a backbone amide H atom in the molecular model with respect to the same H in the real structure, and from this point of view PCF can be used to further refine the conformation of individual residues (and adjust the protection of their amide H) and bring the molecular model in better agreement with the unknown solution structure.

For determination of the two indices, BAI and RAI, that together gauge the degree of protection of the amide H, we used a similar approach to the one developed by Freire and coworkers for the COREX method (16). For example, average molecular (van der Waals) and solvent accessible surface areas are determined for each amino acid residue in a random coil conformation from as set of tri-peptides (GXG, G=glycine, X=residue) with various ϕ and ψ angles for central residue. Similarly, the molecular and solvent accessible surface areas are calculated for each residue in the molecular model.

The BAI index is the ratio of the atomic (van der Waals) surface area of the backbone H atom in the model (AH

i)model and the random coil conformation (AH

i)rc, respectively.

BAIi =AH

i( )model

AHi( )rc

exp −c dHB − dHBo( )

j∑

⎛

⎝⎜⎞

⎠⎟ (Equation 8)

Because the protection of the amide H is offered by the molecular interactions of the residue with its surroundings (protein and solvent), the BAI and RAI indices quantify these interactions by themselves to a certain extent. The BAI index specifically quantifies the interactions of the amide H with its surrounding. For example, if the amide H is free of interactions (either when solvent accessible or buried in the protein core, but not involved in H-bonding) the molecular surface area of the amide H is the same in the model and the random coil conformation, thus the BAI index is 1. If the amide H is involved in H-bonding or hydrophobic contacts, then the BAI index is less than 1 and its actual value is proportional to the length of the H-bond (dHB, in the exponential factor in eq. 6). For example, the BAI index of an amide H involved in a tight H-bond (e.g. < 2.0 Å, such as within a classic α-helix) is practically 0.

The RAIi index of residue i is the ratio of the solvent accessible surface area of the residue in the model (Ai)model and in the random coil (Ai)rc conformation, respectively:

RAIi =Ai( )model

Ai( )rc

fiPD (Equation 9)

The factor fiPD takes into account to some degree the

dynamics of the protein (vide infra) that is captured by the experimental data but is not accounted for in a single static structure as used in this methodology for the prediction of D0 values. The RAI index thus encodes basically the interactions of the whole residue with its environment and the contribution of these interactions to the protection of amide H. For example, RAI becomes 0 when the residue is buried in the protein core, or is 1 when the residue is completely solvent accessible. When XPi is 1 the amide H has no protection, or the protection is total (i.e. no exchange) when XPi is 0.

Compared with the version of DEXANAL that we have used in our previous work (11), an improved version used in the present study includes the treatment of protein dynamics by an empirical relationship (fPD in Eq. 7) that relates the residue exchange probability with the hydrophobic index (HBIi)(17):

5

fiPD = 1+ 1− RAIi( )* a* 1− HBIi( )+ b⎡⎣ ⎤⎦ (Equation 10)

where a and b are adjustable scaling parameters. The relationship simulates the effect of protein dynamics by increasing the RAI index of residues to an extent that depends on their hydrophobicity and solvent accessibility. For example, polar residues that are not entirely exposed to solvent have an enhanced accessibility as compared to the accessibility calculated from a single protein conformation. This enhanced accessibility is driven by the favorable interaction of these residues with the solvent and cannot be captured with a single conformation of the protein. This extension of the program produces dynamic solvent accessibility for protein residues, which led to improvement in the predicted deuterium incorporation factors.

Validation of use of DEXANAL for refining solution structures based upon hydrogen-deuterium exchange. To validate the prediction of H/D exchange data and rate constants from molecular structures, we first examined the case of hen egg-white lysozyme, a protein with known crystallographic structure and H/D exchange data. This was then followed by application to alternative globular proteins with known crystal structure and H/D exchange data (e.g. cytochrome C), subsequent application to more dynamic proteins with known 3D structure and H/D exchange data (e.g. thrombin, IκBα), and finally application to the case of the highly dynamic lipoprotein nascent HDL. Below we begin with the case of hen egg-white lysozyme.

Hen Egg-White Lysozyme. The H/D exchange rate constants for many residues within lysozyme are reported by Radford et al.(12) We therefore first attempted to predict the H/D exchange rate constants, according to equations 3 and 6, from the crystal structure of hen egg-white lysozyme (pdb id: 193L) and compare these predicted values with the published measured rate constants.(12) To calculate the deuterium incorporation factors we defined peptic peptides for consecutive residue sequences for which measured rate constants were reported. Thus, for lysozyme we used a mixture of peptides and individual residues to predict the deuterium incorporation factors (D0), H/D exchange rate constants (kxc) and residue unfolding constants (Ku). Initial analysis of the crystal structure demonstrated predicted H/D rate constants that showed modest correlation (r=0.670) with the published experimentally determined values. We therefore used

DEXANAL to guide modifications to the starting crystal structure in order to produce a solution structure that yielded predicted D0 values in better agreement with those derived from the experimental rate constants, kxc.

In Fig. S1A we show the correlation between the D0 values derived from the experimental kxc and the predicted D0 values from the crystal and solution structures. The predicted solution structure obtained by altering the crystal structure produces D0 values that agree better with the experimental data (r=0.872). Fig. S1B shows a superposition of the crystal structure (orange) and the solution structure (green). Despite the marked improvement in agreement of predicted and experimental H/D exchange data in the solution structure, it is evident that the two structures superimpose well. In fact, the differences between the crystallographic and predicted solution structures are subtle. Differences involve H-bonds in loop regions, which appear more dynamic and on average slightly longer within the predicted solution structure, than in the crystal structure (Fig. S1).

To assess the performance of our method in gauging the degree of protection of amide H’s involved in H-bonding and part of an α-helix, we first looked at a specific peptide which is predominantly α-helical (e.g. Phe3-Leu17) and has many slow exchanging amide H’s. Table S1 lists the experimental kxc and D0 values, and the derived Ku (equation 4) for all residues in this peptide. Also shown are which residues are in α-helix, and which are in loop conformation, and whether they are involved in H-bonding or not. The data in Table S1 shows that for this particular peptide all amide H’s of the residues in α-helix are also involved in H-bonding, barely undergo unfolding (Ku~10-7), and therefore, their amide proton exchange is very slow (kxc~10-5). This behavior is in contrast with residues at the terminal points of the α-helix (Phe3, His15, and Leu17), which are much more dynamic (Ku~10-4) and have amide H’s that exchange a thousand times faster (kxc~10-2). The summary of Table S1 shows that on average, in this particular peptide, the amide H’s involved in H-bonding exchange 2-3 orders of magnitude slower than those that do not make a H-bond.

Fig. S1C shows the location of this α-helix in lysozyme, and Fig. 1D illustrates the superposition of the crystal and predicted solution structures of Phe3-Leu17 (left) and a table listing the length of the H-bonds and the D0, Ku, and kxc values (right). Note that

6

the superimposed structures are almost identical to the naked eye except at the termini of the helices where they deviate. Moreover, using the program DEXANAL, the predicted H/D exchange data concurs with the experimentally reported values, as well as the traditional view that amide H’s in α-helices in globular proteins exchange slowly because of a tight intermolecular environment. Globular proteins are relatively rigid and many amide H’s are protected from exchanging to deuterium on a time scale (e.g. 1 minute) used for mass spectrometry experiments. Similar analyses were performed on the model protein cytochrome C (data not shown). Based upon these analyses, and the recognition that compact globular proteins like lysozyme and cytochrome C are substantially less dynamic than apoA1 within HDL, we next progressed to examination of alternative proteins that are known to possess regions that are markedly more dynamic.

Thrombin. Thrombin is known to be a more dynamic globular protein and therefore might serve as a good test for further validation of the DEXANAL program. Koeppe and colleagues (18) used mass spectrometry to quantify H/D exchange on various regions of thrombin and inactive thrombin precursors (prethrombin2). Several regions of prethrombin 2 become much more dynamic and experienced H/D exchange upon cleavage at Arg320-Ile321. Among these regions with increased dynamics the authors identified peptides Tyr117-Ala132 and the α-helix Tyr228-Arg244 that showed enhanced H/D exchange in active thrombin. We used the MS measured deuterium incorporation factors (D0) for four peptides published by Thrular et al.(14), as well as the deposited crystal structure of thrombin (pdb id: 1ETS) as inputs for DEXANAL to produce a solution structure for thrombin. First we analyzed all four peptides: the correlation between the measured and calculated D0 are shown in Fig. S2A. The correlation is significantly improved for the solution structure that incorporates the experimentally determined D0. Of note, Koeppe et al.(18) reported that the same regions studied by Thrular et al.(14) have increased dynamics in activated thrombin. The peptic peptides corresponding to these regions are: Tyr117-Ala132 (Lys109-Leu129C) and Tyr228-Arg244 (Phe227-Lys235), with the peptides reported by Thrular shown in parentheses, and peptides from overlapping regions not in parentheses reported by Koeppe et al.(18) Peptide Lys109-Leu129C, which is a random buried coil in the crystal structure of thrombin, has 17 exchangeable amide H’s and all exchange. Table S2

shows that all residues in this peptide are very dynamic (Ku ~0.3) and exchange very fast (kxc ~ 0.3 s-

1). In peptide Phe227-Lys235, which is a combination of part α-helix and part β-pleated sheet within the crystal structure, 3 residues with amide H involved in H-bonding are more rigid and relatively resistant to unfolding (Ku ~10-3), and their amide H exchange slower (kxc ~ 10-3 s-1). Still these residues are much more dynamic and their amide H exchange ten times faster than those in lysozyme that have H-bonding (Ku ~10-6, kxc ~ 10−4 s-1 ). The residues with amide H not involved in H-bonding are more flexible (Ku ~0.1) and exchange a hundred times faster (kxc ~ 0.1 s-1).

Fig. S2B shows a superposition of the crystal (orange) and solution (green) structures for the entire protein thrombin. While most of the structures are superimposable, there are also regions with significant differences between the crystal and solution structure, particularly the above mentioned peptides where experimental H/D data shows significant exchange, consistent with enhanced dynamic motion. Shown in Fig S2C are the location of peptide Lys109-Leu129C in the solution structure of thrombin (left), a superposition of the crystal (orange) and solution (orange) structures of this peptide (center) and a summary of the average D0, Ku, and kxc, for the entire peptide (right). Similar information for peptide Phe227-Lys235 is displayed in Fig. S2D.We concluded from our validation tests on lysozyme, thrombin and other proteins not shown here (e.g. cytochrome C, IκBα) that our H/D exchange methodology reasonably predicts the H/D exchange rate constants for both slow and fast exchanging amide H’s regardless of their involvement in H-bonding, and whether they are/are not part of secondary structure motifs such as α-helices, β-sheets, or random coil loops.

ApoA1 in nascent HDL. We finally applied our validated H/D exchange methodology to the refinement of the solution structure of nascent HDL to produce a molecular model of nascent HDL that is based upon both the low resolution SANS structures and the experimental H/D data measured. Our H/D exchange values, D0, measured for 49 peptic peptides indicate that overall, apoA1 is much more dynamic than globular proteins such as lysozyme and cytochrome C, and similar to the more dynamic regions of proteins like thrombin and IκBα. For exemplary purposes, below we present experimental and predicted H/D exchange data for the LCAT binding loop we previously identified and named the

7

“Solar Flare” region (11) (Leu159B-Ala180B,). Also shown is an adjacent (from the anti-parallel apoA1 chain) predominantly α-helical region (Lys77A-Lys106A). Table S3 lists experimentally derived per residue D0, Ku, and kxc (s-1) values for each residue in these peptides. The summary of the table shows that for the Solar Flare loop Arg160B-Arg177B, all residues are very dynamic (Ku ~10-2) and their amide H exchange rapidly (kxc ~0.23 s-1) regardless of whether they are involved in H-bonding or not. On the other hand, residues in the adjacent α-helix, by comparison, are less dynamic (Ku ~10-3), and their amide H exchange an order of magnitude slower (kxc ~2.67 x 10-2 s-1). Even so, the degree of H/D exchange measured in the α-helical region of apoA1 is substantially higher than that observed in α-helices of more compact globular proteins like lysozyme and cytochrome C. Table S3 also shows that within the apoA1 α-helical region (Lys77A-Lys106A) there is only a marginal difference between the dynamics of the residues with amide H’s involved in H-bonding and those without. The H/D exchange data indicate that apoA1 is a highly dynamic protein throughout its length compared to globular proteins, but specific regions (e.g. the Solar Flare) show even more dynamic behavior, as indicated by measured H/D exchange. This suggests that the H-bonds within the α-helical regions of apoA1 are slightly longer (but only by fractions (<0.3) of angstroms) on average, and shorter lived than typical H-bonds of amide Hs in α-helices of globular proteins, but similar to more dynamic regions in other proteins.

Table S5 gives a summary of the results for the validation of the H/D exchange methodology performed on lysozyme and thrombin, as well as results from application to apoA1 in nascent HDL. The table lists the correlation coefficient for the experimental D0 for peptic peptides vs. those predicted by DEXANAL. Additional assessment of the goodness-of-fit is given by the root mean square deviation (RMSD) of the D0 values for peptic peptides, and of the probability correction factor (PCF). Formulas to calculate the RMSD values are given in the footnote of Table S5. Overall, the values show that the Double Super Helix model agrees well with the experimental H/D exchange data. The introduction of protein dynamics into the RAI index produces a modest improvement in the Double Super Helix model with respect to the Solar Flare model, and suggests that overall the Double Super Helix and Solar Flare models have the same quality as far as H/D exchange is concerned.

Computational modeling of nascent HDL. The time-averaged solution molecular model of nascent HDL was constructed by integrating multiple levels of contrast variation SANS with HD-MS/MS data as outlined above. In addition, reported cross-linking, fluorescence resonance energy transfer (FRET), electron spin resonance (ESR) distance constraints were also taken into account (19-21). Distance constraints reported in the literature (MS cross linking, ESR and FRET) were applied to make final refinements while development of the overall model building involved over 250 iterative steps (i.e. over 250 models created for nascent HDL) (See Fig. S4 for a few of the intermediate models of apoA1 within nascent HDL), with assessments of the goodness-of-fit with both SANS and HD-MS/MS data at every step (Table S1). At each refinement step, precedence was always placed on SANS shape data for global conformation, while more fine-tuned refinements of local architecture primarily utilized HD-MS/MS data. This iterative co-refinement process was performed for each contrast variation study (Fig. S4, S5, S6) to first develop an overall model of the protein component based upon SANS and HD-MS/MS data, then the lipid core, followed by a combined model with energy minimization at every step (vide infra). Lastly, the distance constraints reported in the literature (MS cross linking, ESR and FRET) were applied to make final refinements while again at every step giving precedence on SANS data for global conformation and more fine-tuned refinements of local architecture utilizing HD-MS/MS data. The final model of nascent HDL was solvated in a solution of physiological salt concentration and energy minimized as described below.

The overall strategy of extending the low resolution SANS structure to a higher resolution computational model was first to construct an initial model of nascent HDL (i.e. first iteration step in the refinement process) by using the low resolution structures obtained by deconvoluting the experimental SANS curves of 12% and 42% D2O as scaffolds to build molecular models for the respective protein and the lipid core components of nascent HDL (Fig. S5, S6, S7). Subsequently, this initial model was optimized iteratively by using the goodness-of-fit χ2 statistics (for SANS data) and the root mean square deviation (RMSD) of the MS/MS deuterium incorporation factors to guide the construction of new models of better quality.

8

Building the Protein Model. The starting model for the apolipoprotein was constructed using Modeller, Pymol, and Swiss-PDB-Viewer programs,(22,23) by arranging the apoA1 chains into an anti-parallel super helical conformation to match the 12% D2O low resolution model (Fig. 4, S10A). The scattering curves of the intermediate models of reconstituted human apoA1 within nascent HDL were computed using our modification of the SASSIM program(24). Instrument smearing was taken into account in calculating the scattering intensity as described in Merzel and Smith(24) and the relative wavelength spread of neutrons (Δλ/λ) used was 0.1 as suggested by Svergun et al (25). Our extensions to this program include the incorporation of H/D-MS/MS amide backbone exchange into SANS calculations as a function of the degree of deuteration of solvent, the use of calculated Voronoi volumes instead of scaled-tabulated atomic volumes to calculate the scattering length density of the atoms and the average scattering length density of the solvent. This allowed us to calculate fluctuations in scattering length densities of protein residues.

An algorithm was developed for calculating scattering intensities of non-spherical complexes with heterogeneous composition (e.g. both lipid and protein). The use of Voronoi volumes allowed for determination of the distribution of solvent scattering length density as a function of scattering angle for different concentrations of D2O. The distribution function was further used to determine the average scattering length density of the solvent. Modifications of the SASSIM program were validated by performing SAXS (X-ray) and SANS calculations on lysozyme and comparison of results with those published (26).

At each refinement step the all atom model was energy minimized using the OPLS force field(27) (as implemented in GROMACS program(28)). As noted above, to improve the accuracy of the theoretical scattering curve, the SANS calculations also took into account the H/D exchangeable hydrogen atoms in the following way: the scattering length assigned to all H/D exchangeable hydrogens (identified by the DEXANAL program from the experimental H/D exchange incorporation factors) was interpolated between the proton (-3.74 fermis) and deuteron (6.67 fermis) (1). By assigning such an “ensemble averaged” value for the scattering length of exchangeable protons we avoided the need to generate and calculate an ensemble of nascent HDL

particles, which would have been much more costly computationally, but leading to a similar result.

Building the Lipid Model. The lipid phase (200 POPC molecules and 20 cholesterol molecules) was modeled such that it follows the helical orientation of protein hydrophobic surface filling the inside groove, and overall fits the experimentally visualized SANS prolate ellipsoid obtained by deconvoluting the 42% D2O scattering curve. The lipid model was further refined by performing energy minimization of the whole particle, and by matching its calculated scattering intensity curve to the experimental one (42% D2O). Fig. S7 illustrates the overlay between the low resolution SANS shape of the lipid core of HDL obtained in the 42% D2O contrast variation experiment with the acyl chains and cholesterol in the penultimate model.

Incorporation of HD-MS/MS Data into the Model. Hydrogen-deuterium exchange data at each step were incorporated in the Double Super Helix model of nascent HDL (11). We used the H/D exchange methodology implemented in the program DEXANAL that was validated as described above. Data from at least three independent hydrogen deuterium exchange studies on each of three independent preparations of nascent HDL were averaged and used for determination of experimental deuterium incorporation factors.

Refinement of the HDL Model. The overall optimization of the 3-D structure of the molecular model of nascent HDL was a lengthy process involving over 250 iterative steps (256 models created for nascent HDL, see Fig. S6 for a few intermediate models of apoA1 within nascent HDL), with assessments of the goodness-of-fit with both SANS and HD-MS/MS data at every step (i.e. intermediate model built). At each refinement step, a new all-atom model of HDL was built by altering the structure of the model (both protein and lipid) built in the previous step. A new protein model was built by adjusting the shape of the double chain and the packing / orientation of the amino acid residues, and a new model of the lipid was obtained by adjusting the shape of the micelle. The quality of the new model was assessed performing SANS (at multiple D2O levels) and HD exchange calculations on it. Precedence was always placed on SANS data for global conformation, while more fine-tuned refinements of local architecture primarily utilized HD-MS/MS data. This iterative process was

9

performed for each contrast variation study (Fig. S5, S6, S7) to first develop an overall model of the protein component, then the lipid core, followed by a combined model with energy minimization at every step (vide infra). The goodness-of-fit χ2 statistics for SANS data and the RMSD for HD-MS/MS data, that gauge the quality of the all-atom computational model relative to the experimental data (Fig. S5B, S6, Table S5), were used at each step to assess the quality of the new model and guide the iterative development steps toward better models.

Incorporation of Geometrical Constraints. Experimental geometrical constraints such as reported chemical cross-links, fluorescence resonance energy transfer and electron spin resonance distance constraints have been used to further refine the model (Table 1B, Table S6, Fig. S10). The distance between two lysine residues involved in chemical cross-linking has been measured between the Nε atoms of the -NH3 groups of the two lysine residues after their conformations were adjusted such that the shortest possible distance was obtained without creating atom clashes and affecting the conformation of the protein backbone. To obtain hypothetical fluorescence resonance energy transfer distances in the Double Super Helix model, we have in silico mutated one of the amino acids in the fluorescence resonance energy transfer pair to a tryptophan and the other into a cysteine. Next we have attached the fluorescence resonance energy transfer acceptor (AEDANS) to the cysteine and adjusted the conformation of tryptophan and the fluorescence resonance energy transfer acceptor in such a way that the shortest distance was achieved. To obtain spin resonance distances we in silico mutated both amino acids in the spin resonance pair to cysteine. Next we attached the methane thiosulfonate nitroxide spin label (MTSSL) to both cysteine residues and adjusted the conformation of these chemically modified cysteine residues in such a way that the shortest distance was achieved. The cross-link, fluorescence resonance energy transfer and spin labeling distances measured after the model was modified as described above are those listed in Table 1b in the manuscript.

Solvation of the Final Model of Nascent HDL. Finally, the vacuum model of nascent HDL obtained in this way was solvated using methods similar to that previously described (11). Briefly, the HDL model was solvated in a box of TIP3P water molecules (27,29). The distance from the particle to solvent box walls was set to 15 Å. Next, eight sodium ions were

added to the solvated particle to neutralize the negative charge of the two chains of apoA1, and then twenty four NaCl ion pairs were inserted to produce a physiological salt concentration (300 milliosmollar). The solvated system was subjected to a series of energy minimizations (using Gromacs) in which the particle was allowed to relax gradually (first the solvent alone, then the lipid, and finally the protein). In the last step of energy minimization all atoms in the system were allowed to move. The calculation of the total potential energy of the system has taken into account the periodic boundary conditions for the solvated system, the cut-off for the van der Waals interactions was 10 Å, and for Coulomb interactions we used the Ewald summation as implemented in the PME method (30).

Molecular dynamics simulation on solvated nascent HDL. Molecular dynamics simulation (4.5 nanosecond) was performed on the Double Super Helix model of nascent HDL solvated in a box of SPC water molecules(31) (Fig. S9). The setup of the simulation box was similar with that used for the energy minimization of the Double Super Helix model in water described in the previous section. But, to shorten the amount of calculations performed per unit time we have used a united atom force field (28) to describe the interactions in the system. In this type of force field hydrophobic hydrogen atoms and the carbon atom they attach to are replaced by united atoms (28). To describe the POPC molecules we used a modified united atom force field with special parameters for the acyl chains as described by Sapay (32). Cholesterol was described by a united atom force field developed by Höltje and Brandt (33). The system has been prepared for simulation by performing first a sequence of energy minimizations in which the heavy atoms of the HDL particle where first restrained to move and then gradually relaxed in subsequent energy minimizations. In the final energy minimization all atoms in the system were allowed to move freely. Next, we performed 100 ps of simulation in which all heavy atoms of the particle were restrained to move in order to allow the solvent to rearrange and start equilibrating. The simulation was performed at 300 K in NVT ensemble using the Berendsen thermostat (34). Random initial velocities have been assigned for T=300K. The bonds between atoms in protein and lipid have been reset to their correct value by using the LINCS algorithm. Both the bond lengths and the bond angles in the water molecules were constrained by the SETTLE algorithm (35). The time integration step was 2

10

femtosecond and a coordinate frame was saved each 10 picosecond. Periodic boundary conditions have been applied and the Coulomb interaction energy has been calculated using the PME method as described in the previous section.

The change in the energy of the system as a function of the simulation time is displayed in Fig. S9, which shows that the system reaches thermodynamic equilibrium just before reaching 4 ns. Fig. S9D shows a snapshot of apoA-1 dimer after 4.5 ns. The shape of the protein is still clearly helical without much modification with respect to the starting structure, the lipid core is still predominantly micellar, and the apoA1 retains the highly solvent exposed “Solar Flares”, identified as the LCAT binding sites (11).

Theoretical models of poorly lipidated apoA1. Earlier molecular dynamics simulations and electron microscopy studies(36) suggested that apoA1 in HDL with lower lipid content than in nascent HDL, may adopt a saddle conformation and that the curvature of the saddle relates to particle composition. We therefore constructed four saddle type models that have increasing curvature (Fig. S3), in order to test the scattering properties of these theoretical shapes. We used as a starting point for building the saddle

models the Belt model of apoA1, which was bent into a saddle conformation. Models of higher curvature were obtained by more accentuated out of plane bending. Subsequently, we calculated the SANS curves for all saddle models and compared them with the experimental curve and the theoretical curve produced by the Double Super Helix model (Fig. S3).

Theoretical models of HDL that have helix 5 of apoA1 modeled as a loop. Another suggested change in the conformation of apoA1 within nascent HDL is the unfolding of helix 5 into a loop and the change of the registry from 129 to 139 (19). To test the scattering and solvent accessibility properties of models that possess these features, we built variants of the Double Super Helix, Belt and Solar Flares models in which the distance between residues 133 (chain A) and 133’ (chain B) is less than 15 Å and the distance between residues 146 (chain A) and 146’ (chain B) is less than 15 Å. Also the registry of apoA1 in the models was changed such that amino acid residues 139 from both apoA1 chains align as indicated by Oda and colleagues (19) (Fig. S4). Next, we fully investigated these derived models by performing SANS and hydrogen deuterium exchange calculations, and by inspecting the observed chemical cross-links.

Figure S1. Calculated versus experimental deuterium incorporation factors (D0) for a series of peptides of lysozyme. A. The calculated D0 values obtained from the crystal structure of lysozyme (pdb id: 193L) are shown with an open circle (r=0.670), while the D0 values obtained from a predicted solution structure of lysozyme derived through use of DEXANAL on the starting crystal structure are shown with an open triangle (r=0.872); B. Note the superposition of the crystal (pdb id: 193L, orange) and predicted solution (green) structures of lysozyme generated by incorporation of reported experimental H/D exchange data reveals similar overall structures. However, with the structural refinement, the predicted H/D exchange incorporation factor per residue (D0

i) throughout lysozyme is significantly better correlated with the experimentally determined H/D exchange data (r=0.670 vs. r=0.872). C. Shown is the location of a predominantly α-helix peptide Phe3-Leu17 of lysozyme in overlapping crystal (orange) and solution (green) structures, with the remainder of the protein colored grey. D. Expanded view of the superposition of the crystallographic (orange) and predicted solution (green) structures of the predominantly α-helix peptide Phe3-Leu17 of lysozyme. Shown are H-bond lengths for the amino acid residues of the α-helix peptide, the experimental per residue deuterium incorporation factor (D0) determined from the reported rate constant and 1 min exchange time, and the exchange probabilities (XP) calculated from the crystallographic and solution structures of lysozyme. The H-bonds between the backbone H and O atoms are shown as black dash lines. Note that the calculated XP from the solution and crystallographic structures agree well with the experimental D0 for residues in an α-helix, and that α-helix amide H’s involved in H-bonding show slow exchange (XP=0). Differences between experimental D0 and crystallographic-calculated XP are noticeable at the ends of the α-helix where the crystal structure predicts slower exchanging amide H’s than is observed experimentally. The predicted XP based upon the model of the solution structure for these terminal amide H’s agree reasonable well with the experimental D0 values.

11

Figure S2. Calculated versus experimental deuterium incorporation factors (D0) for a series of peptides of thrombin. A. The calculated D0 values obtained from the crystal structure of thrombin (pdb id: 1ETS) are shown with an open circle (r=0.924), while the D0 values obtained from the predicted solution structure of thrombin are shown with an open triangle (r=0.959); B. Superposition of the crystallographic (pdb id: 1ETS, orange) and predicted solution (green) structures of thrombin; C. Left: the location of peptide Lys109-Leu129C in the crystal structure (orange) and predicted solution structure (green) of thrombin, Center: superposition of the crystal (orange) and predicted solution (green) structures of this peptide, Right: the summary of average H/D exchange data for the entire peptide; D. Left: the location of peptide Phe227-Lys235 in the crystal structure (orange) and predicted solution structure (green) of thrombin, Center: superposition of the crystal (orange) and predicted solution (green) structures of this peptide, Right: the summary of average H/D exchange data for the entire peptide.

12

log

I

0 0.1 0.2Scattering Vector (1/A)

O

-7

-3

-5

-6

-4

-2

-8

Experimental dataModel 1Model 2Model 3Model 4Double Super Helix

E

Model 2Model 1

Model 3 Model 4

C

A B

D

FIGURE S3 Theoretical scattering curves of saddle shaped models of apoA1 within nascent HDL.Here we further investigate the scattering behavior of proposed saddle shaped models for apoA1,including saddles of differing curvature which were hypothesized computational models of HDLwith lower lipid content(38). A series of saddle models of apoA1 with different curvature within HDLwere generated and are shown as cartoon ribbon representations in (A-D) and the calculatedscatter-grams for each model are shown in E. Due to the closed symmetrical shape of apoA1within a saddle conformation, just like the Belt and the Solar Flares models (Fig. 4), the scatteringcurves of the saddle shaped models show characteristic oscillatory behavior.

Supplementary Figure S3

13

Residues

K40-K239a

K118-K140

K12-K182

Double SuperHelix model (A)

Helix 5 loop SolarFlares model (A)

16.1

30.1

a. The calculated distances in nascent HDL models are between ε-amine groups of two lysines assuming free rotation of the side chain oflysines. b. The calculated distances between two residues in the Double Super Helix model of nascent HDL are from the oxygen atom ofmethane thiosulfonate nitroxide spin label (MTSSL) on one residue to the oxygen atom of MTSSL of the other residue, assuming free rota-tion of the side chains of residues.

K59-K208 15.6

K77-K195 18.1

10.7

Helix 5 loop Beltmodel (A)

45.1

29.1

21.6

7.9

32.0

111.4

15.9

8.5

Helix 5 loop DoubleSuper Helix model (A)

n.a

n.a

Alo

gI


O

-7

-3

-5

-6

-4

-2

-8

Experimental dataDouble Super HelixDouble SuperHelix with Helix 5loop

Blo

gI


O

-7

-3

-5

-6

-4

-2

-8

C

log

I


O

-7

-3

-5

-6

-4

-2

-8

Experimental dataDouble Super HelixBelt with Helix 5loop

D

36.9

26.8

17.4

11.0

13.5

Models

BeltSolar Flares

DoubleSuper Helix

9.89.4

2.3

Protein

Correlation b

PCFd(α)

0.34

0.89

0.7790.897

0.784

RMSD c

Peptide

1.81.2

2.4

2.73 1.010.44

1.53

PCF(β)

48.154.2

52.3

Protein

Radius ofo

(A)gyration coefficient

a χ2 stands for the standard deviations between the experi-mental scattering data and calculated scattering data of nascentHDL models obtained from our modified SASSIM program25, bCorrelation coefficient for correlation between theoretical hydro-gen-deuterium exchange probabilities of peptic peptides in themodel and the actual experimentally measured deuterium incor-poration within apoA1 peptic peptides. c RMSD stands for root-mean square deviation. It quantifies the difference between theexperimental deuterium exchange data and the calculated ones.d PCF stands for probability correction factor, which representsthe multiplier needed to adjust theoretical deuterium incorpora-tion of individual residues based upon the HDL models to matchthat which was actually experimentally determined

E

o o o o

Experimental dataDouble Super HelixSolar Flares withHelix 5 loop

Goodness of Fit of Helix 5-loop HDL models withSANS, and H/D-MS/MS data

6.243.810.9

K133-K133b

E146-E146L134-L134

5.69.4

4.111.2

11.210.2

4.9 4.9 5.9

χ2a


14

FIGURE S4 SANS, HD-MS/MS and geometrical analyses of Helix5-loop HDL models.A, Comparison between the geometric constraints in the Double Super Helix modelwith Helix 5 loop models of HDL. Listed in the table are calculated distances betweenthe amino acids in the HDL models vs. reported distances determined experimentallyby FRET or ESR. K118-K140 crosslinking constraint satisfied in the Double Super Helixmodel is no longer satisfied in the same model with helix 5 as a loop. Both Belt andSolar Flares Helix 5 loop models do not accommodate as many geometric constraintsas the Double Super Helix model. B, Comparison of the experimental scattering curve(black dotted line) with theoretical scattering curve of the Double Super Helix model

(blue line) and the Double Super Helix with helix 5 loop HDL model (red line). C,Comparison of the experimental scattering curve (black dotted line) with the Solar Flaresmodel of HDL with helix 5 as a loop (red line). D, Comparison of the experimentalscattering curve (black dotted line) with the Belt model of HDL with helix 5 as a loop(red line). The scattering curve of the Belt and Solar Flares Helix 5 loop models exhibitthe same oscillatory behavior as observed for the original Belt and Solar Flares models,which indicates that the Helix 5 loop introduced in these models does not alter the generalfeature of the scattering curve. On the other hand, the curve produced by the Helix 5 loopDouble Super Helix model closely follows the curve of the original Double Super Helix modelsupporting our finding that the overall shape of the protein is open. E, The radius of gyration,χ2 statistics and HD-MS/MS theoretical data for the Helix 5 loop models of HDL. The χ2

statistics of all Helix 5 loop HDL models are significantly larger than that of the Double SuperHelix model. The difference between the experimental and calculated deuterium incorporationfactors is significant for all the Helix 5 loop HDL models, suggesting that introducing the loopat Helix 5 and shifting the registry to residue 139 significantly alters the solvent accessibilityaway from that derived experimentally.

15

A B


FIGURE S5 Overlay of SANS protein shape with initial and final apoA1 models.A, The overlay of Model 1 of apoA1 within HDL with the SANS shape (gray) obtained in 12% D2O.The figure shows that the initial model of apoA1 fits very well with the helical conformation of SANSstructure. B, The overlay of the final model of apoA1 within HDL with the SANS shape obtained in12% D2O. This figure shows that the initial helical conformation of apoA1 does not changesignificantly during modeling except the N- and C-terminals which fold back more in the final model.The original SANS structure is obtained by a deconvolution process that treats the protein as carbonspheres, wheras the predicted SANS scattergram for the models utilize all atom structures andexperimental HD- MS/MS data.

16

Model 1 Model 30 Model 50 Model 150

Model 175 Model 190 Model 200 Model 256FIGURE S6 Intermediate models of apoA1 within nascent HDL. The figure shows various stagesalong the process of generating the final experimental model. ApoA1 within nascent HDL startingfrom the first conformation (Model 1, top left) and ending with the Double Super Helix model is

shown. Model 1 was obtained by molding the protein chains arranged in anti-parallel orientation byusing the 12% D2O SANS shape as a scaffold. The iterative process of structure refinementgradually change the conformation of the N- and C-terminals by folding them back towards theprotein chain backbones to accommodate both SANS and HD-MS/MS data. This change ofconformation was required in order to increase the agreement between calculated and experimentalscattering intensity. This model refinement required building of more than 250 models, each ofwhich were subject to energy minimization, HD exchange incorporation and SANS calculation.Goodness of fit statistics at each step for both SANS and HD-MS/MS guide the iterative process(Table S5). The final model (Model 256) shown at the bottom right was solvated andrepresents the simultaneous optimization of SANS, HD exchange and geometric constraints insolution (chemical crosslinks, FRET and ESR).


17

A B

FIGURE S7 Overlay of SANS shape of the lipid with lipid core (acyl chains and sterol) model.A, The 42% D2O SANS shape (black) is revealed to be a prolate ellipsoid. B, Overlay of thefinal model of the lipid core (acyl chains colored green and sterols colored orange) within nascentHDL with the SANS shape (gray) obtained in 42% D2O. This figure shows that the lipid core fillswell the space occupied by the SANS shape.


18

0.015 0.065 0.115 0.1650

0.215 0.25

0.4

0.8

1.2

1.6

Scattering Vector (1/A)O

log

I

FIGURE S8 Small angle X-ray scattering curve of nascent HDL.


19

FIGURE S9 Molecular dynamics simulation results for the Double Super Helix model of nascent HDL.The simulation was done in physiological solution (300 milliosmollar salt concentration) for 4.5 ns atroom temperature. A, The fluctuation of total energy of the system (HDL+solution) during thesimulation time. B, The fluctuation of the temperature of the system during the simulation time. C,Root mean square deviation of the HDL structure from the initial structure, during the simulation.D, A snapshot of the Double Super Helix model of nascent HDL after 4.5 ns simulation. The structureof the particle is intact after 4.5 ns. The protein maintains its overall open helical shape and the lipidmaintains a predominantly micellar/pseudo-lamellar character. The Solar Flare loops previouslyreported (L159-A180) also persist during the simulation.


A

0 1500 3000 4500

Time (ps)

-2.725

-2.735

-2.745

-2.755

-2.765

Ene

rgy

(10-6

KJ

mol

-1)

B

0 1500 3000 4500

Time (ps)

310

300

290

Tem

pera

ture

(K)

DC

0 1500 3000 4500Time (ps)

RM

SD

1.4

0.7

0

20

13 Ao

Lys94

40 Ao

A B

FIGURE S10 A slightly modified Double Super Helix model of apoA1 in nascent HDL accounts forthe intra-chain Nter-K94 cross-link. Most of the previously reported structural constraints (inter-chaindistances based upon cross-linking/mass spectrometry, fluorescence resonance energy transfer andelectron spin resonance studies) are readily accommodated within the observed open super helicalstructure of apoA1. With modest alterations in the conformation of the Double Super Helix model ofapoA1, more structural constraints of apoA1 can be easily accommodated without changing theglobal super helical shape of apoA1 or shifting the registry. Shown here is a slightly modified DoubleSuper Helix model that accommodates the intra-chain N-terminal amine and Lys94 cross-linkconstraint reported for apoA1 of HDL. A, The ribbon representation of apoA1 from the modified (red)Double Super Helix model overlapping with the original Double Super Helix model (Blue). To get thecrosslink between N-terminal amine and ε-amine of lysine 94 in the same chain of apoA1 identified

by LC-MS/MS sequencing, only a slight conformational change of the N-terminal of one apoA1 isrequired. The N-terminal segment that needs minor conformational change to accommodate thecross-link is colored red. B, A magnified view of the region of apoA1 within HDL in which thecross-link is formed. A slight displacement of the N-terminus of apoA1 changes the distance betweenN-terminal amine and ε-amine of lysine 94 from 40 Å to 13 Å, within the distance for chemicalcrosslinking by Bis(Sulfosuccinimidyl) suberate . This adaptive feature of apoA1 is a directconsequence of its open conformation.


21

A B

C

FIGURE S11. Illustration of the three major HDL models studied. The Double Super Helix model (A)of nascent HDL. In this model, HDL particle is prolate ellipsoid in shape. The Solar Flare model ofnascent HDL (B) and the Double Belt model (C) of nascent HDL are both oblate ellipsoid shape.


22

Table S1. Experimental residue deuterium-incorporation factors (D0), the residue unfolding equilibrium constants (Ku) and the predicted H/D exchange rate constants (kxc) for residues in the predominantly α-helical peptide Phe3-Leu17 of Hen Egg-White Lysozyme (pdb id: 193L).

Residue H-bond α-Helix D0 Ku kxc(s-1)

Phe3 0.9 3.34 x 10-4 3.84 x 10-2

Leu8 0.01 2.47 x 10-6 1.27 x 10-4

Ala9 0 2.86 x 10-7 4.87 x 10-5

Ala10 0 1.65 x 10-8 4.55 x 10-6

Ala11 0 3.91 x 10-8 1.08 x 10-5

Met12 0 6.19 x 10-15 1.67 x 10-12

Lys13 0 2.08 x 10-7 6.74 x 10-5

His15 0.22 1.06 x 10-5 4.17 x 10-3

Leu17 0.4 8.03 x 10-5 8.61 x 10-3

Average for residues in an α-helix

0 5.03 x 10-7 4.31 x 10-5

Average for residues in a loop

0.51 1.42 x 10-4 1.71 x 10-2

Fold difference ~103 ~103

Average for residues with backbone H-bond

0.03 1.95 x 10-6 6.33 x 10-4

Average for residues without backbone H-bond

0.65 2.07 x 10-4 2.35 x 10-2


Residues with “ ” mark are either in an α-Helix or involved in a H-bond. Residues with “X” mark are not in either an α-Helix or a H-bond .

23

Table S2. Experimental residue deuterium incorporation factors (D0), the residue unfolding equilibrium constants (Ku) and the predicted H/D exchange rate constants (kxc) for residues in the two peptides: Lys109-Leu129C and Phe227-Lys235 of Thrombin (pdb id: 1ETS), identified to have enhanced dynamics after thrombin activation.

Residue H-bond α-helix/ β-sheet

D0 Ku kxc(s-1)

Lys109-Leu129CArg110 0.71 7.25 x 10-3 2.48 x 10-2

Ile112 1 1.99 0.461 Glu113 1 1.12 0.461 Leu114 0.69 5.74 x 10-2 2.37 x 10-2

Ser115 0.84 1.16 x 10-2 3.62 x 10-2

Asp116 1 0.207 0.461 Tyr117 1 0.59 0.461 Ile118 1 1.02 0.461 His119 1 9.17 x 10-2 0.461 Val121 0.64 8.18 x 10-2 2.03 x 10-2

Cys122 1 7.05 x 10-2 0.461 Asp123 1 0.228 0.461 Leu125 1 0.718 0.461 Lys126 1 0.348 0.461 Gln127 1 0.141 0.461 Thr128 0 2.20 x 10-6 6.43 x 10-6

Ala129 1 0.134 0.461 Ala129A 0.98 3.75 x 10-2 8.10 x 10-2

Lys129B 0.88 2.15 x 10-2 4.25 x 10-2

Leu129C 1 0.614 0.461 Average for residues in an α-helix

0.75 0.156 0.345


0.92 0.429 0.302

Fold difference ~1 ~3 ~1 Average for residues with backbone H-bond

0 2.20 x 10-6 6.43 x 10-6


0.93 0.394 0.372


Phe227-Lys235Tyr228 0 7.27 x 10-6 9.69 x 10-6

Thr229 0.46 5.99 x 10-3 1.24 x 10-2

His230 0 5.32 x 10-6 7.19 x 10-5

Val231 0 4.50 x 10-7 8.65 x 10-7

Phe232 0.86 4.31 x 10-2 3.88 x 10-2

Arg233 0.41 3.48 x 10-3 1.04 x 10-2

Leu234 0.74 2.83 x 10-2 2.67 x 10-2

Leu235 1 0.379 0.461 Average for residues in an β-sheet

0 7.27 x 10-6 9.69 x 10-6


0.5 6.57 x 10-2 7.84 x 10-2

Fold difference --- ~104 ~104


0.12 1.50 x 10-3 3.11 x 10-3


0.75 0.113 0.134

Fold difference ~6 ~102 ~102

Residues with “ ” mark are either in an α-Helix/β-sheet or involved in a H-bond. Residues with “X” mark are not in either an α-Helix or a H-bond

24

Table S3. Experimental residue deuterium incorporation factors (D0), the residue unfolding equilibrium constants (Ku) and the predicted H/D exchange rate constants (kxc) for the Solar Flare (Arg160B-Arg177B) and the adjacent predominantly α-helical peptide Lys77A-Lys106A of ApoA1 in nascent HDL.

Residue H-bond α-Helix D0 Ku kxc(s-1)

α-helical peptide of apoA1 chain A (Lys77A-Lys106A) Lys77A 0.92 5.59 x 10-3 6.21 x 10-2

Glu78A 0.18 7.25 x 10-4 5.10 x 10-3

Thr79A 0.24 6.60 x 10-4 6.85 x 10-3

Glu80A 0.61 2.78 x 10-3 2.35 x 10-2

Gly81A 0.71 1.35 x 10-3 3.07 x 10-2

Lys82A 0.30 1.32 x 10-3 8.81 x 10-3

Arg83A 0.82 3.37 x 10-3 4.29 x 10-2

Gln84A 0.99 3.37 x 10-3 1.10 x 10-1

Glu85A 0.71 3.69 x 10-3 3.12 x 10-2

Met86A 0.45 1.27 x 10-3 1.51 x 10-2

Ser87A 0.87 1.00 x 10-3 5.20 x 10-2

Lys88A 0.69 9.46 x 10-4 2.95 x 10-2

Asp89A 0 2.20 x 10-13 2.50 x 10-12

Leu90A 0.47 5.37 x 10-3 1.60 x 10-2

Glu91A 0.59 6.72 x 10-3 2.21 x 10-2

Glu92A 0 6.60 x 10-13 2.50 x 10-12

Val93A 0 1.03 x 10-12 2.50 x 10-12

Lys94A 0.53 1.65 x 10-3 1.87 x 10-2

Ala95A 0.82 1.90 x 10-3 4.29 x 10-2

Lys96A 0.09 1.47 x 10-4 2.30 x 10-2

Val97A 0.12 7.38 x 10-4 3.33 x 10-2

Gln98A 0.79 2.74 x 10-3 3.91 x 10-2

Tyr100A 0.13 6.41 x 10-4 3.40 x 10-3

Leu101A 0.17 9.42 x 10-4 4.77 x 10-3

Asp102A 0.87 9.41 x 10-3 5.01 x 10-2

Asp103A 0.08 3.44 x 10-4 1.96 x 10-2

Phe104A 0 3.82 x 10-13 2.50 x 10-2

Gln105A 0.57 9.29 x 10-4 2.10 x 10-2

Lys106A 0.79 1.56 x 10-3 3.87 x 10-2

Average for residues in an α-helix

0.47 2.04 x 10-3 2.67 x 10-2


0.41 1.84 x 10-3 2.42 x 10-2


0.57 2.43 x 10-3 3.15 x 10-2

Fold difference ~1.4 ~1.3 ~1.5

Random coil peptide (Solar Flare) of apoA1 chain B (Arg160B-Arg177B) Arg160B .30 7.09 x 10-4 9.02 x 10-3

Thr161B .97 3.48 x 10-3 8.45 x 10-2

His162B .92 1.49 x 10-3 6.43 x 10-2

Leu163B 1 5.92 x 10-2 5.76 x 10-1

Ala164B 1 5.44 x 10-2 5.76 x 10-1

Tyr166B 1 1.08 x 10-1 5.76 x 10-1

Ser167B 1 1.27 x 10-2 5.76 x 10-1

Asp168B 1 3.34 x 10-2 5.76 x 10-1

Glu169B 1 1.63 x 10-1 5.76 x 10-1

Leu170B 0.4 4.03 x 10-3 1.29 x 10-2

25

Arg171B 0.82 3.32 x 10-3 4.23 x 10-2

Gln172B 0.47 4.85 x 10-4 1.59 x 10-2

Arg173B 0.64 7.70 x 10-4 2.52 x 10-2

Leu174B 0.79 5.23 x 10-3 3.92 x 10-2

Ala175B 0.97 8.64 x 10-3 9.15 x 10-2

Ala176B 0.66 1.56 x 10-3 2.67 x 10-2

Arg177B 0.71 1.50 x 10-3 3.09 x 10-2


0.80 2.72 x 10-2 0.229


0.79 4.04 x 10-2 0.254


0.81 2.16 x 10-2 0.219

Fold difference ~1 ~0.5 ~1Average for residues in the α-helix Lys77A-Lys106A

0.47 2.04 x 10-3 2.67 x 10-2

Average for residues in the Solar Flare loop Arg160B-Arg177B

0.80 2.72 x 10-2 0.229

Fold difference ~2 ~10 ~10Residues with “ ” mark are either in an α-Helix or involved in a H-bond. Residues with “X” mark are not in either an α-Helix or a H-bond

26

Table S4. Goodness of Fit of experimental H/D data for lysozyme, thrombin and three HDL models (Belt, Solar Flares and Double Super Helix).

RMSD Protein/model Correlation coefficienta Peptideb PCFc

Lysozyme (pdb file: 193L) Predicted solution structure

0.670 0.872

0.656

0.419 20.557

3.703 Thrombin (pdb file: 1ETS) Predicted solution structure

0.924 0.959

6.983 2.280

23.616 0.499

HDL Models Double Super Helix: apoA1 α-chain apoA1 β-chain

0.946

0.819

0.212 0.237

Solar Flares: apoA1 α-chain apoA1 β-chain

0.915 1.648 0.297 0.293

Belt: apoA1 α-chain apoA1 β-chain

0.794 2.391 1.499 1.722

a For lysozyme and thrombin the first row shows data for the crystallographic structure and the second row shows data for the predicted solution structure.

b The root mean square deviation (RMSD) is calculated for peptide with the formula: RMSDpep =D0

n − XPii

n

∑⎛⎝⎜

⎞⎠⎟

2

n

Npep

∑Npep

,

where Npep is the number of peptic peptides in the set; D0n is the experimentally measured deuterium incorporation factor for

the peptic peptide n, and the sum over the exchange probabilities (XPi) is carried out over all residues in the peptic peptide n. c The RMSD for peptide with the the probability correction factor (PCF) is calculated with the following formula:

RMSDres =1− PCFi( )2

i

Nres

∑Nres

, where Nres is the number of residues in one peptide chain and PCFi is the probability

correction factor for residue i. Note that the correlation between predicted and experimentally measured H/D exchange for both Solar Flare and Double Super Helix models of HDL are similarly high, and the smaller RMSD values at the peptide and per-residue level for each indicates better agreement with experimental H/D data (as expected since H/D data is used to refine these models compared to the Belt model).

27

Models RgCorrelationCoefficient b

Protein Lipid Protein Lipid

RMSD c

150175190200

256

50

30

1 55.1 37.6 5.3 197 1.7 0.37 0.380.848

51.8 39.4 1.3 309 1.2 0.36 0.310.902

52.0 39.4 1.2 136 1.1 0.32 0.320.925

52.3 38.5 1.1 126 1.0 0.42 0.580.91450.9 37.0 0.9 150 1.0 0.29 0.350.911

51.1 37.8 1.0 147 0.8 0.27 0.240.945

52.1 38.4 1.1 145 0.8 0.21 0.240.946

51.1 37.5 1.0 145 0.7 0.40 0.280.940

Peptide PCFd(α) PCF(β)

Table S5. Goodness of fit for the intermediate models built during the extension to high resolu-tion model of nascent HDL

a. χ2 stands for the goodness of fit of the calculated scattering data versus experimental scattering data ofnascent HDL models obtained from our modified SASSIM program30.b. Correlation coefficient for the correlation between theoretical hydrogen-deuterium exchange probabilities ofpeptic peptides in the model and the actual experimentally measured deuterium incorporation within apoA1 pep-tic peptides.c. RMSD stands for root-mean square deviation. It quantifies the difference between the experimental deuteriumexchange data and the calculated ones.d. PCF stands for probability correction factor, which represents the multiplier needed to adjust theoretical deute-rium incorporation of individual residues based upon the HDL models to match that which was actually experi-mentally determined.

Vacuum

Hydrated

χ2, a

28

Table S6. Intra-chain amino acid distance constraints of apoA1 with-in nascent HDL reportedly identified by HPLC tandem mass spec-trometry (MS/MS) sequencing and Fluorescence Resonance EnergyTransfer (FRET).

Residues

K88-K94K96-K106K12-K23

Double SuperHelix model (A)a

o

18

1416

K133-K140 9

Nt-K94 40

K94-K96 13

K206-K208K107-K118

K88-K96

1417

16

Nt-K12 11

a, The calculated distances between ε-amine groups of crosslinked intra-chainresidues in the Double Super Helix model of nascent HDL, assuming free rota-tion of the side chains of residues (see Supplementary Methods).b, Silva et al, 2005, Biochemistryc, Bhat et al, 2005, JBC.d, Martin et al, 2006 JBC.e. Bhat et al., 2007 Biochemistry.

W50-L230 133

The reported intrachain structural constraints measured by FRET and lysine crosslinks(MS/MS) were examined in the Double Super Helix model. Nine of the eleven con-straints fit with the Double Super Helix model.

MS/MS (12 A) b,c,eo

FRET (22.7 A) d

MS/MS (12 A) bo

MS/MS (12 A) b,c,o

MS/MS (12 A) b,c,eo

MS/MS (12 A) b,c,eo

MS/MS (12 A) b,c,eo

MS/MS (12 A) bo

MS/MS (12 A) bo

MS/MS (12 A) b,eo

MS/MS (12 A) c,eo

Methods

K12-K94 38 MS/MS (7.7 A) eo

o

29

Table S7. Interchain saltbridges in apo A1 within the Double Super Helix model ofnascent HDL

Residuesa

H155-E111 1.49R151-E111 1.69

K140-E125 1.76

K118-E147 1.51

K239-E2 1.66K238-E2 3.07

R173-E78 1.51

Distance (A)o

K59-E198 1.55

K77-E183 1.55

K40-D213 1.65R27-E223 1.88K12-E234 1.67R10-E234 1.53

K96-E169 1.56

E147-K118 1.62

E136-R123 2.36E125-K140 1.59

E111-R151E111-R155

1.541.53

E136-H155 1.42

E92-R173 1.53D89-R173 1.65

E78-R177 1.42

E70-R188 1.53

D48-K206 1.69

E191-K59 1.47E92-R171 3.20

D62-K195 1.61

Residues

Twenty-eight predicted salt bridges between two apoA1 molecules in the Double SuperHelix model of nascent HDL are listed in the table. The distance between two aminoacids which forms salt bridge is less than 3.2 A.

o

a, Residues colored red are from apoA1 chain A of HDL and residues colored blackare from apoA1 chain B of HDL

Distance (A)o

30

FOOTNOTE

The Double Super Helix model reported in this paper has been deposited in the Protein Model Database, http:/www.caspur.it/PMDB (accession no. PM0075984).

REFERENCES

1. Serdyuk, I. N., Zaccai, N. R., and Zaccai, J. (2007) Methods in Molecular Biophysics, 776

2. Hjelm, R. P., Kneale, G. G., Sauau, P., Baldwin, J. P., Bradbury, E. M., and Ibel, K. (1977) Cell 10(1), 139-151

3. May, R. P., Nowotny, V., Nowotny, P., Voss, H., and Nierhaus, K. H. (1992) Embo J 11(1), 373-378

4. Ramakrishnan, V. (1986) Science 231(4745), 1562-1564 5. Johs, A., Hammel, M., Waldner, I., May, R. P., Laggner, P., and Prassl, R. (2006) J Biol

Chem 281(28), 19732-19739 6. Guinier, A. (1939) Ann. Physi., 161-237 7. Zaccai, G., and Jacrot, B. (1983) Annu Rev Biophys Bioeng 12, 139-157 8. Svergun, D. I. (1999) Biophys J 76(6), 2879-2886 9. Konarev P. V., S. A. V., Koch M. H. J. and Svergun D. I. (2003) J.Appl.Cryst., 1277-

1282 10. Petoukhov, M. K., PV; Kikhney, AG; Svergun, D. (2007) Journal of Applied

Crystallography (40), S223-S228 11. Wu, Z., Wagner, M. A., Zheng, L., Parks, J. S., Shy, J. M., 3rd, Smith, J. D., Gogonea,

V., and Hazen, S. L. (2007) Nat Struct Mol Biol 14(9), 861-868 12. Radford, S. E. B., M.; Topping, K. D.; Dobson, C. M.; Evans, P. A. (1992) Prot. Struct.

Func. Genet. 14, 237-248 13. Zhang, Z. S., D. L. (1993) Prot. Sci. 2, 522-531 14. Thrular, S. M. E. C., C. H.; Torpey, J. W.; Koeppe, J. R.; Komives, E. A. (2006) J. Am.

Soc. Mass Spectrom. 17, 1490-1497 15. Bai, Y. M., J. S.; Mayne. L.; Englander, S. W. (1993) Prot. Struct. Func. Genet. 17, 75-

86 16. Hilser, V. J., and Freire, E. (1996) J. Mol. Biol. 262, 756-772 17. Black, S. D., and Mould, D. R. (1991) Anal Biochem 193(1), 72-82 18. Koeppe, J. R. K., E. A. (2006) Biochemistry 45, 7724-7732 19. Martin, D. D., Budamagunta, M. S., Ryan, R. O., Voss, J. C., and Oda, M. N. (2006) J

Biol Chem 281(29), 20418-20426 20. Silva, R. A., Hilliard, G. M., Li, L., Segrest, J. P., and Davidson, W. S. (2005)

Biochemistry 44(24), 8600-8607 21. Bhat, S., Sorci-Thomas, M. G., Alexander, E. T., Samuel, M. P., and Thomas, M. J.

(2005) J Biol Chem 280(38), 33015-33025 22. Guex, N., and Peitsch, M. C. (1997) Electrophoresis 18(15), 2714-2723 23. Sali, A., and Blundell, T. L. (1993) J Mol Biol 234(3), 779-815 24. Merzel, F., and Smith, J. C. (2002) Acta Crystallogr D Biol Crystallogr 58(Pt 2), 242-249 25. Svergun, D. I., Richard, S., Koch, M. H., Sayers, Z., Kuprin, S., and Zaccai, G. (1998)

Proc Natl Acad Sci U S A 95(5), 2267-2272 26. Chen, L., Wildegger, G., Kiefhaber, T., Hodgson, K. O., and Doniach, S. (1998) J Mol

Biol 276(1), 225-237

31

27. Jorgensen, W. L., Tirado-Rives, J. (1988) Journal of the American Chemical Society 110(6), 1657-1666

28. Lindahl, E., Hess, B., and an der Spoel, D. (2001) J. Mol. Mod 7(8), 306 29. Jorgensen, W., Chandrasekhar J, Madura, JD., Impey, RW and Klein, ML. (1983) The

Journal of Chemical Physics 79, 926-935 30. Darden, T., York, D., Pedersen, L. (1993) J. Chem. Phys. (98), 10089–10089 31. Berendsen, H. J. C., Postma, J. P. M., van Gunsteren, W. F., Hermans, J. (1981), 331-342 32. Sapay, N. a. T., DP. (2008) Curr. Top. Memb (60), 111-130 33. Holtje, M., Forster, T., Brandt, B., Engels, T., von Rybinski, W., and Holtje, H. D. (2001)

Biochim Biophys Acta 1511(1), 156-167 34. Berendsen, H. J. C., Postma, J. P. M., DiNola, A., Haak, J. R. (1984) J. Chem. Phys (81),

3684–3690 35. Miyamoto, S., and Kollman, P. (1992) J. Comput. Chem. 13, 952-962 36. Catte, A., Patterson, J. C., Jones, M. K., Jerome, W. G., Bashtovyy, D., Su, Z., Gu, F.,

Chen, J., Aliste, M. P., Harvey, S. C., Li, L., Weinstein, G., and Segrest, J. P. (2006) Biophys J 90(12), 4345-4360

32

Giuseppe Zaccai, Joseph A. DiDonato and Stanley L. HazenArundhati Undurti, Roland P. May, Michael Haertlein, Martine Moulin, Irina Gutsche,

Zhiping Wu, Valentin Gogonea, Xavier Lee, Mathew A. Wagner, Xin-Min Li, Ying Huang,The Double Super Helix model of high density lipoprotein

published online October 7, 2009J. Biol. Chem.

10.1074/jbc.M109.039537Access the most updated version of this article at doi:

Alerts:

When a correction for this article is posted•

When this article is cited•

to choose from all of JBC's e-mail alertsClick here

Supplemental material:

http://www.jbc.org/content/suppl/2009/10/07/M109.039537.DC1.html

http://www.jbc.org/content/early/2009/10/07/jbc.M109.039537.full.html#ref-list-1

This article cites 0 references, 0 of which can be accessed free at


http://ww

w.jbc.org/

Dow

nloaded from

http://www.jbc.org/lookup/doi/10.1074/jbc.M109.039537

http://www.jbc.org/cgi/alerts?alertType=citedby&addAlert=cited_by&cited_by_criteria_resid=jbc;M109.039537v1&saveAlert=no&return-type=article&return_url=http://www.jbc.org/content/early/2009/10/07/jbc.M109.039537

http://www.jbc.org/cgi/alerts?alertType=correction&addAlert=correction&correction_criteria_value=early/2009/10/07/jbc&saveAlert=no&return-type=article&return_url=http://www.jbc.org/content/early/2009/10/07/jbc.M109.039537

http://www.jbc.org/cgi/alerts/etoc

http://www.jbc.org/content/suppl/2009/10/07/M109.039537.DC1.html

http://www.jbc.org/content/early/2009/10/07/jbc.M109.039537.full.html#ref-list-1

http://www.jbc.org/

Double Superhelix Model of High Density Lipoprotein

Documents

Transcript of Double Superhelix Model of High Density Lipoprotein