HDL surface lipids mediate CETP binding as revealed by electron microscopy and molecular dynamics simulation

Cholesteryl ester transfer protein (CETP) mediates the transfer of cholesterol esters (CE) from atheroprotective high-density lipoproteins (HDL) to atherogenic low-density lipoproteins (LDL). CETP inhibition has been regarded as a promising strategy for increasing HDL levels and subsequently reducing the risk of cardiovascular diseases (CVD). Although the crystal structure of CETP is known, little is known regarding how CETP binds to HDL. Here, we investigated how various HDL-like particles interact with CETP by electron microscopy and molecular dynamics simulations. Results showed that CETP binds to HDL via hydrophobic interactions rather than protein-protein interactions. The HDL surface lipid curvature generates a hydrophobic environment, leading to CETP hydrophobic distal end interaction. This interaction is independent of other HDL components, such as apolipoproteins, cholesteryl esters and triglycerides. Thus, disrupting these hydrophobic interactions could be a new therapeutic strategy for attenuating the interaction of CETP with HDL.

Although CETP has been intensively studied, a detailed understanding of how CETP senses and binds to HDL remains unknown due to the heterogeneity and dynamics of HDL. HDLs vary in size, shape, and composition 16,17 . Considering that most plasma CETPs are naturally bound to HDLs 18 , an investigation of how different components of HDL affect the interaction of spherical HDL with CETP is essential for a complete understanding of CETP function.
Here, we studied CETP interactions with various HDLs and liposome vesicles using optimized negative-staining (OpNS), cryo-electron tomography (Cryo-ET), and molecular dynamics simulations in order to understand how the different HDL components affect CETP binding.

Results
EM images of CETP bound to plasma HDL 2 . Based on previous reports, the spherical HDL 2 in plasma varies in diameter (,9 to ,15 nm) and in density (from 1.063 to 1.125 g/ml) 19,20 . Earlier studies 18,19,21 showed the average molecular weight of HDL 2 is 360 kDa and that the particles contain three major surface components: apolipoprotein A-I and A-II (apoA-I and apoA-II) (total ,40.2% of MW) and phospholipids (,31.3% of MW), as well as core lipids: cholesterol esters (,17.6% of MW) and triglycerides (,4.2% of MW). A small amount of free cholesterol (,5.8% of MW) is distributed between the particle surface and the core 21 .
In the present study, CETP was incubated with isolated HDL 2 at a molar ratio of 154 (HDL 2 : CETP), then prepared as described for optimized negative-staining (OpNS) EM method 22,23 (a method that minimizes rouleaux formation), and examined by electron microscopy (EM) [23][24][25] . A large field EM micrograph and representative particles showed that CETP-HDL 2 complexes had the appearance of rod shaped CETP penetrating spherical shaped HDL 2 (Fig. 1a). No CETP was found to bridge two HDL 2 particles or to adhere to the convex surface of HDL 2 via its concave surface as hypothesized by The samples of CETP incubated with, (a) plasma HDL 2 , (b) plasma HDL 3 , (c) reconstituted spherical HDL (rHDL), and (d) 1-Palmitoyl-2oleoylphosphatidylcholine (POPC) liposome vesicles were prepared by optimized negative-staining (OpNS). Survey view (left panel), representative conformations of particles (top right panel), and schematics of the particle structures and percentages of each conformation (bottom right panels). (e) Histogram of particle diameter (left panel, sampling step 0.5 nm, fitted with sixth-degree polynomial functions) and shape (right panel, sampling step 0.04, fitted with sixth-degree polynomial functions). The diameter is calculated based on the geometric mean of two measured diameters (in which one is the longest direction while the other is perpendicular to it); the shape is calculated based on the ratio of these two diameters. (f) The average diameters of CETP, lengths vs. widths. Particle window size is 37 nm in a, b and c. Particle window size is 59 nm in d. All scale bars, 90 nm. www.nature.com/scientificreports SCIENTIFIC REPORTS | 5 : 8741 | DOI: 10.1038/srep08741 crystallography 11 . The HDL 2 bound to CETP had an average diameter of 12.7 6 1.6 nm (Fig. 1e), similar to HDL 2 alone (12.8 6 1.3 nm) ( Supplementary Fig. 1e). The bound CETP particles were 8.7 6 1.7 nm in length and 3.2 6 0.5 nm in width (Fig. 1f). Although the width of bound CETP was similar to that of CETP itself (,3.5 nm), the length of the bound CETP was much shorter (,8.7 vs. ,12.5 nm) ( Supplementary Fig. 1e), suggesting that ,3-4 nm of the CETP molecule length had been inserted into the HDL 2 surface ( Supplementary Fig. 1e). The EM micrographs and statistical analyses showed that more than 50% of the HDL 2 bound to CETP, in which, ,30.3% 6 2.7% (mean 6 sd) of the HDL 2 particles bound to one CETP molecule, while 15.7% 6 4.9% (mean 6 sd) of the HDL 2 bound to two CETP molecules and 4.7% 6 4.3% (mean 6 sd) of the HDL 2 bound to more than two CETP molecules. Among these bound HDL 2 , ,40% bound to two or more CETP molecules. In contrast, no CETP molecule bound to more than one HDL 2 particle simultaneously.
TEM images of CETP bound to plasma HDL 3 . To investigate how three major HDL surface components: phospholipids, apoA-I and apoA-II affected the CETP binding, the above experiment was repeated with HDL 3 that contain a different surface protein and lipid ratio than HDL 2 . Earlier studies showed 21 the HDL 3 vary in size, but have an average molecular weight of 175 kDa, a density of ,1.125 to ,1.210 g/ml, and contain apoA-I and apoA-II (,55.5% of MW), phospholipids (,22.7% of MW), free cholesterol (,2.8% of MW), cholesterol esters (,14.7% of MW) and triglycerides (,3.4% of MW) 21 . Thus, HDL 3 has a lower surface percentage of lipids, but a higher surface percentage of proteins than HDL 2 .
In this study, the HDL 3 sample was incubated with CETP at a molar ratio of 451 (CETP5HDL 3 ), prepared for OpNS and examined by EM. The EM micrographs showed that the CETP-HDL 3 complexes were similar to CETP-HDL 2 complexes, with the rod-shaped CETP adhering to the spherical shaped surface of HDL 3 (Fig. 1b). As before, no CETP bridged two HDL 3 particles or adhered to the convex surface of HDL 3 via its concave surface. The diameter of HDL 3 that was bound to CETP ( Fig. 1e) was ,11.4 6 1.7 nm, similar to that of HDL 3 alone (,10.9 6 1.0 nm) ( Supplementary Fig. 1e). The rod shaped CETP in the complexes had a similar width to CETP itself ( Supplementary Fig. 1e), but were shorter in length (,8.6 6 1.2 nm, Fig. 1f). The micrographs and statistical analyses showed more than 40% of the HDL 3 particles were bound to CETP (compared to ,50% for HDL 2 ), in which 32.8% 6 7.0% (mean 6 sd) of the HDL 3 particles bound to one CETP molecule, while 8.2% 6 1.4% (mean 6 sd) of the HDL 3 particles bound to two and more CETP molecules. Among these bound HDL 3 , ,20.2% was also bound to two or more CETP molecules (Fig. 1b). Considering that HDL 3 and HDL 2 have the same surface components (apoA-I, apoA-II, and phospholipids), the lower percentages (,41.1% vs. ,50.7%) of CETP bound to HDL that have a lower percentage (22.7% vs. 31.3%) of surface phospholipids suggest that the surface phospholipids/protein ratio may have contributed to some variation in CETP binding.
TEM images of CETP bound to spherical recombined HDL (rHDL). To investigate which HDL surface component dominates CETP binding, one of three surface components, apoA-II, was excluded by repeating the above experiment using reconstituted HDL (rHDL) 26 . The rHDL contain apoA-I as the only protein (,43% in MW) and POPC as the only phospholipid (,32.9% in MW). Cholesteryl esters were the only core lipid components (,21.7% in MW) 26 and a small amount free cholesterol (,2.4%) was distributed between the surface and the core.
In this study, the rHDL sample was incubated with CETP (molar ratio of 451 for CETP5rHDL) prepared for OpNS 23 and examined by EM. The EM micrographs showed ,42.9% of rHDL particles were attached to CETP. No CETP was found to connect two rHDL particles or bind to the rHDL surface via its concave surface. The average diameter of rHDL particles in rHDL-CETP complexes was 10.2 6 1.2, similar to rHDL (10.3 6 0.7 nm) alone ( Supplementary Fig. 1e). The CETP protrusion length of ,9.2 6 1.3 nm was shorter than that of CETP alone, but slightly longer than the CETP protrusions on the surfaces of HDL 2 and HDL 3 (,9.2 nm vs. ,8.6 nm, Fig. 1f). The micrographs and statistical analyses showed ,34.6% 6 5.3% (mean 6 sd) of the rHDL particles bound to one CETP molecule, and ,7.6% 6 3.8% and ,0.7% 6 1.3% (mean 6 sd) of the rHDL particles bound to two and more CETP molecules. Approximately 19.4% of the CETP-rHDL complexes were bound to two or more protruding CETP molecules. Notably, most multi-CETP binding complexes (,75%) showed the protruding CETP sharing the same hemisphere of rHDL (Fig. 1c). Moreover, the total number of CETP molecules bound to rHDL can exceed the total number of apoA-I molecules in . Panel (b) shows a plot of the mean number of binding CETP molecules versus their binding liposome diameter (black line). As some CETP molecules that are located behind/in-front of the liposomes may not be counted, a geometric model (shown in supplementary Fig. 4) was used to adjust the binding number. The adjusted/calibrated number of bound CETP molecules versus the liposome diameter was also plotted (shown in blue line). The data was fitted with a sixth-degree polynomial function (histograms have a sampling step of 2.17 nm on liposome diameter) using a total of 3,282 liposome/liposome-CETP complexes. Each liposome diameter was calculated as the geometric mean (the square root of the product) of two perpendicular diameters.
www.nature.com/scientificreports SCIENTIFIC REPORTS | 5 : 8741 | DOI: 10.1038/srep08741 each rHDL particle ( Supplementary Fig. 2), which is consistent to the previous observation 13 . These observations showed that the apoA-II in HDL does not play a significant role in binding to CETP.
EM images of CETP bound to liposomes. To investigate which one of the remaining two HDL surface components, apoA-I or phospholipids, dominates the CETP binding, a lipid vesicle without apoA-I was further used to repeat the above experiments. The lipid vesicle (a POPC-only liposome) was incubated with CETP, and the sample was prepared by OpNS and examined by EM. The EM micrographs and representative liposomes showed the liposomes had a distribution of ,7.7 to ,116.2 nm in diameter (Fig. 1d), which is within the diameter range of all types of lipoproteins (from small HDL to large VLDL). Zoomed in views of a micrograph ( Supplementary Fig. 3) and representative particle images (Fig. 1d, top right panel) showed spherical liposomes with surface CETP protrusions with an average length of ,8.5 6 0.6 nm (Fig. 1f). The liposome particles retained their shape and structure after CETP insertion. Incubation with CETP did not cause particle disruption. Additionally, CETP did not form a bridge between two liposome particles, or bind to a liposome so that its concave surface was adjacent to the convex surface of the liposome. The micrographs and statistical analyses showed ,19.1% 6 3.2% of the liposome particles bound to 1-2 CETP molecules, ,6.0% 6 2.7% of the liposome particles bound to 3-4 CETP molecules, and ,5.1% 6 3.6% and 2.7% 6 1.7% bound to 5-6, or 7 and more CETP molecules respectively. In total, ,32% of the liposomes contained at least one surface CETP protrusion. Among these CETP bound liposomes, ,66% were bound to two or more CETP molecules (Fig. 1d). Notably, smaller liposomes had more CETP molecules bound (Fig. 2). For instance, 5-7 CETP molecules were bound to liposomes ,15 nm in diameter, while liposomes over ,40 nm were rarely bound to CETP ( Fig. 2b and Supplementary Fig.4). This relationship was analyzed based on a linear fitting the number of bound CETP against the liposome diameters (Fig. 2a). This result suggests that the smaller liposomes had a higher binding affinity to CETP. Moreover, it confirmed that apoA-I in HDL do not dominate the CETP binding.
Three dimensional structure of a complex of CETP bound to liposome by individual-particle cryo-electron tomography. To confirm that the conformation of liposome-CETP complex was not due to negative-stain artifact, the sample was also flash-frozen in vitreous buffer without any negative-staining. The frozen sample was then examined by cryo-electron microscopy (cryo-EM) at 2170uC and under a low-dose imaging condition (Fig. 3a). The micrographs and selected particles showed the CETP liposome surfaces adhering with rod-shape CETP (arrows in Fig. 3b), similar to that from the OpNS (Fig. 1d). To reveal the detailed structure in three dimensions (3D), the sample was also imaged by cryo-electron tomography (cryo-ET) through a series of tilting angles (angle range of 257u to 60u in steps of 1.5u). A 3D structure of a representative complex was reconstructed by individual-particle electron tomography (IPET) 27 . IPET was designed for 3D reconstruction of an asymmetric individual particle rather than averaging different particles like in conventional single-particle reconstruction. The naturally varying particle diameter of liposomes made it impossible to achieve any 3D reconstruction by conventional cryo-EM single-particle reconstruction methods.
A total of 79 tilting images of a representative liposome-CETP complex were windowed from CTF-corrected (by TOMOCTF 28 ) cryo-ET micrographs (Fig. 3c, left panel). Although the tilt images (but CTF corrected) were noisy, a protrusion of CETP was marginally visible (arrows in Fig. 3c, left panel). To confirm that the protrusion was from a bound CETP instead of noise, three images from consecutively tilted angles (28.5u, 30u and 31.5u) were averaged together to enhance the signal contrast and to reduce the noise level. The averaged image showed the protrusion was more clearly visible (arrow in Supplementary Fig. 5a) than in any individual tilt image, suggesting that the protrusion is indeed from CETP instead of noise. During IPET reconstruction, the tilt images were gradually aligned precisely to their ''global center'' via an iteration process 27 . The noise in the 3D reconstructions and projections was gradually eliminated as the signal was enhanced (Fig. 3c from second to last column). The protrusion remained visible in each intermediate 3D reconstruction and projection (arrows indicated in Fig. 3c from second to fifth column) as well as the final 3D reconstruction (Fig. 3d). To further confirm whether the protrusion in the final 3D reconstruction was from real signal instead of noise, five central slices (arrows indicated in Supplementary Fig. 5b) of the final 3D reconstruction were directly averaged together. This averaged image and its contour map showed the protrusion remained significant ( Supplementary  Figs. 5c and d). Fourier shell correlation (FSC) analysis showed the resolution is from ,35 Å to 72 Å based on Supplementary Fig. 6. By rigid-body docking the crystal structure of CETP (PDB entrance: 2OBD 11 ) into the protrusion, a near perfect match to the protrusion diameter in width, but ,5 nm shorter in length, suggests a CETP penetrating in the liposome surface ( Fig. 3e and f). This experiment confirmed the conformation where CETP penetrates the liposome surface is not due to a negative-staining artifact.
MD simulation study of the relationship between liposome size and surface hydrophobicity. To understand why CETP interacts with the lipid surface of HDL/liposome without apoA-I or apoA-II (Fig. 1) and why smaller liposomes had a higher binding affinity to CETP (Fig. 2), we hypothesized that a higher surface curvature in a smaller HDL/liposome 29 will generate a higher degree of surface hydrophobicity (Supplementary Fig. 7b) and increase the affinity of the interaction of the relatively hydrophobic distal end of the CETP N-terminal domain with the liposome surface 13 (Supplementary Fig. 7a).
To test this hypothesis, we employed MD simulation to study the liposome surface hydrophobicity against the liposome size by generating a series of liposome vesicles with diameters of ,12 nm, ,20 nm, ,27 nm, ,35 nm and ,42 nm (Fig. 4, and supporting Table 1). After energy minimization ( Supplementary Fig. 8), the analyses on the surface hydrophobicity of each lipid vesicle showed that the percentages of surface hydrophobic area (measured by the solvent accessible surface area, SASA 14 ) were ,12.9%, ,9.2%, ,8.0%, ,7.4% and ,7.3% respectively ( Fig. 4b and d), suggesting that smaller liposomes had a higher hydrophobicity than the larger liposomes (Fig. 4e).

Discussion
Our experiments showed that, i) the similar morphology in CETP binding between plasma HDL and rHDL implied that the absence of ApoA-II and TG in rHDL did not affect CETP binding to HDL (Fig. 1); ii) the total number of CETP molecules bound to rHDL can exceed the total number of apoA-I molecules in each rHDL particle; iii) non-apoA-I containing liposomes showed similar CETP protrusions to apoA-I containing HDL (Fig. 1d); iv) that there is a significant correlation between liposome size and number of bound CETP molecules; and v) a significant correlation between liposome size and the percentage of surface hydrophobic area. Those results suggest that apoA-I and apoA-II may not be involved in CETP binding; however, the surface phospholipids, and surface curvature likely dominate the interaction of CETP with the surface of HDL.
A larger surface lipid area provides a higher opportunity for CETP binding. We noticed that HDL 2 , which has has a larger diameter than HDL 3 (,13 vs. ,11), also has a higher percentage of bound CETP particles than HDL 3 (,50% vs. ,42%). This may due to the HDL 2 surface lipids occupying a larger area than the surface lipids of HDL 3 (,31.3% vs. ,22.7%). This is consistent with liposomes which are larger than HDL2 or HDL3 having ,6 bound CETP molecules ( Fig. 1d and 2b), compared to the average of 1-2 CETP molecules that are bound to HDL 2 and HDL 3 ( Fig. 1a and b). This suggests that the binding affinity of CETP is regulated by surface lipid area, as well as surface curvature.
Our results favor the tunnel mechanism rather than the shuttle mechanism for the transfer of CE from HDL to LDL via CETP. Both of these mechanisms were hypothesized to be involved in the CETPmediated transfer of CE between HDL and LDL two decades ago 30,31 . In the shuttle mechanism, CETP interacts with HDL and acquires CEs, after which the ''CE-enriched'' CETP is released from the HDL for subsequent deposition into LDL. The ''CE-poor'' CETP is then released from the LDL surface for subsequent interaction with HDL in another cycle of CE transfer 32 . In the tunnel mechanism, CETP binds to HDL and LDL simultaneously to form a ternary complex. CETP then undergoes a conformational change to form a hydrophobic tunnel whereby CEs transfer from HDL to LDL.
Our results showed that CETP bound to only one HDL particle at a time, and did not form a bridge between two HDL 2 HDL 3 or rHDL particles. This suggests that the binding of CETP to HDL is directional. When the fact that i) we were unable to demonstrate CETP bridging two HDL particles (Fig. 1), ii)small HDL/liposome particles with a higher surface curvature and hydrophobicity are more likely to bind to more CETP molecules ( Fig. 1 and 2), iii) the N-terminal b-barrel domain of CETP binds to HDL while the Cterminal b-barrel domain binds to LDL 13 , and iv) the distal end of the N-terminal b-barrel domain of CETP is relatively hydrophobic 11 , are considered in light of the current results which show that CETP binds to only one liposome particle at a time, and does not bridge two liposomes (Fig. 1, 2 and 3), it follows that the distal end of the CETP N-terminal b-barrel domain is most likely inserted into the liposome particle surface via a protein-lipid interaction. This interaction is different for LDL particles, where the distal end of the Cterminal b-barrel domain of CETP inserts to the LDL surface 13 , possibly via a protein-protein interaction. The conclusion that there is a reduced interaction of the distal end of the N-terminal domain of CETP with LDL surface lipids may be due to the fact that LDL particles have a flattened, ellipsoidal shape with planar opposing faces that, as shown in our earlier cryo-EM single-particle reconstruction 33 , are largely covered by apoB-100 (,500 kDa). The reduced lipid surface area combined with the reduced surface curvature of the particles is likely to have contributed to the low binding affinity of the hydrophobic distal end of the CETP N-terminal bbarrel domain to LDL.
CETP directionally bound to HDL is consistent with many earlier reports, including i) a CETP-mediated net transfer of CE into VLDL 34 , ii) CETP bridging HDL and LDL instead of bridging two HDL or LDL particles 13 , iii) the distal portion of the N-terminal bbarrel domain of CETP being more flexible in solution than is indicated by its crystal structure 14 ; and iv) MD simulations showing that the penetration of the N-terminal b-barrel domain of CETP into the HDL surface generates an opening that allows CE to access the CETP tunnel 15 . However, our results are inconsistent with those of other investigators who have shown that CETP binds to the edge of discoidal HDL 35 via an interaction with apoA-I 36 , and atomistic and coarse-grained simulations of CETP-HDL interaction which have shown that the N-terminus helix-X domain of CETP penetrates deeply into the HDL particle core 37 . Nonetheless, our results favor a tunnel, not a shuttle mechanism whereby CETP no longer binds to HDL when it has acquired an appropriate number of CEs from HDL.
While we have illuminated some of the mechanism behind CETP, several questions remain: i) how the CETP hydrophobic N-terminal b-barrel domain of CETP penetrates the HDL surface and with such high specificity (e.g. not using the C-term b-barrel domain? Normally, as hydrophobic interactions are not specific, it is unclear how much this affinity relate to the surface curvature interactions in concert with the N-term distal end to open up); ii) how CE molecules can be transferred through a ,10 nm channel; iii) how TGs can be transferred back to HDL from LDL; iv) how CE and TG exchange between LDL and VLDL; v) how CETP homo-exchanged the radiolabeled lipid transfer among HDL particles. Our results highlight the CETP N-terminal b-barrel domain hydrophobic distal end as a potential drug target, which may lead to a next-generation drug to treat CVDs.

Methods
Protein and liposome isolation. Recombinant human CETP (,53 kDa with no post-translational modifications) was expressed and purified from the Chinese hamster ovary cell line DG441, as previously described 11 . The CETP concentration was determined by absorbance at 280 nm. Native plasma HDL 2 and HDL 3 in phosphate buffered saline (PBS) were isolated from fresh, pooled samples of human plasma by ultracentrifugation as reported 38 . Pooled samples from multiple donors were used to make sure that the sample was representative and that the results could not be attributed to something that specifically related to a single individual. Spherical, reconstituted HDL (rHDL) in PBS were prepared 39,40 . Liposome vesicle samples were produced by Encapsula NanoSciences (Brentwood, TN). The sample contained 1 mg/ml 1-Palmitoyl-2-oleoylphosphatidylcholine (POPC, from Avanti Polar lipids) with a peak vesicle size of ,50 nm in a buffer containing 20 mM Tris-Cl, 154 mM NaCl, pH 7.4.
Specimen preparation for negative-staining EM. Specimens were prepared for EM using the optimized negative-staining (OpNS) protocol 23,25 , which minimizes the rouleaux artifact that is observed with lipoproteins [23][24][25] . In brief, CETP (0.28 mg/ml) was incubated with HDL 2 , HDL 3 , rHDL and liposomes, respectively at 37uC for 15 minutes at a molar ratio of ,451 (CETP: lipid macromolecular particle). CETP-lipid macromolecular complexes were diluted to 1.5 mg/ml with Dulbecco's phosphate buffered saline (DPBS). A 4 ml aliquot was placed on a thin-carbon-coated copper grid (300 mesh TEM grid, Cu-300CN, Pacific Grid-Tech, San Francisco, CA) that had been glow-discharged. After one minute, excess solution was blotted with filter paper, followed by washing and negative staining with 1% (w/v) uranyl formate (UF) 23,25 . After air drying, the grids were further dried for an hour at 40uC.
Electron microscopy data acquisition and image pre-processing. The OpNS micrographs were acquired under defocus between ,0.6 um to ,2.2 um on a Gatan UltraScan 4 K 3 4 K CCD equipped on a Zeiss Libra 120 Plus transmission electron microscope (Carl Zeiss NTS GmbH, Oberkochen, Germany). The TEM was operated under a high-tension of 120 kV, energy filtering of 20 eV and magnification range of 31.5 K to 80 K, in which each pixel of the micrographs corresponded to 3.68 to 1.48 Å respectively. A total of 15-72 micrographs were imaged from each sample. The contrast transfer function (CTF) of each micrograph was determined and then corrected by the phase-flip option using ctfit (EMAN software) 41 . ,300-,3000 particles from each sample were selected and windowed by the boxer software in the EMAN software package 41 and submitted for Gaussian low-pass filtering before size measurement.
Cryo-electron microscopy (cryo-EM) and cryo-electron tomography (cryo-ET) data acquisition. Cryo-EM specimens were prepared on lacey carbon film coated copper grid (Cu-200LC, Pacific Grid-Tech, San Francisco, CA). Cryo-EM data of liposome binding to CETP specimens were acquired under a less than ,2.3 mm defocus with a high-sensitivity 4 K 3 4 K pixel Gatan Ultrascan CCD camera at 50 K magnification by the same Zeiss Libra 120 TEM (each pixel of the micrograph corresponded to 2.4 Å in the specimens). Total dose for un-tilt 2D micrographs is about 10 to 30 e 2 /Å 2 . For cryo-ET data acquisition, the specimens mounted on a Gatan 915 high-tilt cryo-EM holder were tilted at angles ranging from 257u to 60u in steps of 1.5u. The total dose of electron illumination was up to ,120 e 2 /Å 2 or slightly higher. The tilt series of tomographic data was controlled and imaged by manual operation and by Gatan tomography software (Zeiss Libra 120 TEM) that was preinstalled in the microscope.
Correction of contrast transfer function (CTF) for cryo-ET data. Tilting series of micrographs were initially aligned together with the IMOD software package 42 . The defocus near the tilt-axis area of each tilt micrograph was examined by fitting CTF parameters with its power spectrum by ctffind3 in the FREALIGN software package 43 and then examined by ctfit (EMAN software package) 41 . The CTF was then corrected by TOMOCTF 28 . The tilt series of each CETP-liposome image in windows of 200 3 200 pixels was tracked and selected by IPET software.
Individual-particle electron tomography (IPET) 3D reconstruction. Ab-initio 3D reconstruction of an individual liposome-CETP complex was conducted by the IPET method 27 . In IPET, a small image containing only a targeted liposome-CETP complex was windowed from each tilted whole-micrograph (CTF corrected). An ab-initio initial model was generated by directly back-projecting these small-images into a 3D map. The map was then refined via three rounds of refinement loops (including more than a hundred iterations) by the focused electron tomography reconstruction (FETR) algorithm 27 . In FETR, an automatically-generated dynamic Gaussian low-www.nature.com/scientificreports SCIENTIFIC REPORTS | 5 : 8741 | DOI: 10.1038/srep08741 pass filter and an automatically generated soft-mask were applied to both the references and tilted images to achieve the final 3D reconstruction. Since the specimen holder has a limitation to tilt to 690u angle, a wedge shaped data was missing in final 3D reconstruction, which resulted in a certain level of artifact, especially along the Z direction. The missing wedge data was estimated via our newly developed interactive algorithm (related manuscript in preparation). As an implementation of this algorithm, we computed the missing wedge data and contributed to the final reconstruction. The crystal structure (PDB entry 2OBD 11 ) was fitted into the final IPET 3D reconstruction by using a rigid-body fitting option in UCSF Chimera 44 .
Fourier shell correlation (FSC) analysis. To analyze tomographic 3D reconstruction resolution, the center-refined raw ET images were split into two groups based on having an odd-or even-numbered index in the order of tilt angles 27 . Each group was used to generate a 3D reconstruction image; the two 3D reconstructions from both groups were then used to compute the FSC curve over their corresponding spatial frequency shells in Fourier space (using the ''RF 3'' command in SPIDER 45 ) ( Supplementary Fig. 5). The frequency at which the FSC curve declined to a value of 0.5 and 0.143 (golden standard 46 ) was used to represent the resolution of the final reconstruction.
Statistical analyses of CETP binding to HDL particles and liposomes. Particle size was determined by measuring the diameter in two orthogonal directions, as described before 25 . In brief, the geometric mean of the perpendicular diameters was used to represent the particle diameter. The aspect ratio of the long and perpendicular diameters was used to represent particle shape. Histograms of the particle diameters were generated with 2.17 nm sampling steps. Each histogram was fitted with a 9 th degree polynomial function in R for data analysis. Since the CETP bound to the liposome may have been blocked from certain viewing directions, we computed the probability based a geometric model ( Supplementary Fig. 3). The probability (P) is a function of the liposome diameter (d) and CETP protrusion length (l), in equation of ). The final histogram was adjusted by taking the measured histogram divided by this variable probability.
Liposome structures by molecular dynamics (MD) simulation and hydrophobicity analysis. Based on the size range of liposomes in our experiment, five initial liposome models with diameters of 12 nm, 20 nm, 27 nm, 35 nm and 42 nm were simulated. The initial models to simulate the liposomes were generated as below, the lipid bilayer vesicles (,65 Å 2 /lipids 47 and the average thickness ,36.8 Å 48 ) with an open pore (with 13% of surface area) 49 were surrounded by a solution containing water molecules and NaCl at 0.1 M. These pores on the vesicles would allow the lipid flipping and transferring between inner and outer membranes during energy minimization and MD simulations 50,51 .
As the liposomes contained a large number of molecules, we used the Residuebased Coarse Graining (RBCG) MD method to simplify our initial models in order to enable a long time-scale simulation 52 . In RBCG, about every 10 atoms were grouped together and assigned as a bead model based on their chemical functionality. One POPC molecule was assigned 13 beads (one bead for choline, one for phosphate, two beads for glycerol and nine for two fatty acid chains) based on Marrink's CG lipid model 53 . The interaction potentials between beads and the bead masses were used as previously reported 54 .
Energy minimization was conducted by Nanoscale Molecular Dynamics (NAMD) 55 . The criteria to determine whether the liposomes were stabilized were as follows: i) the surface pores were completely closed; ii) liposomes should have a spherical shape determined by Radial Distribution Function (RDF), which is computed with respect to the liposome's center of mass (supplementary Fig. 8); and iii) stabilized surface hydrophobicity. The area of the hydrophobic surface on each stabilized liposome was measured by the Solvent Accessible Surface Area (SASA) function in Visual Molecular Dynamics (VMD) 56 . The percentages of surface hydrophobic area on each liposome surface were computed by taking their hydrophobic surface area and dividing by the whole liposome surface area. The average of the last 300 frames (12 ns) was used to calculate the hydrophobicity.