Apolipoprotein A1 forms 5/5 and 5/4 antiparallel dimers in human high-density lipoprotein

Running title : Cross-linking and MS/MS reveals multiple registries in APOA1 Abbreviations ABSTRACT Apolipoprotein A1 (APOA1), the major protein of high-density lipoprotein (HDL), contains 10 helical repeats that play key roles in protein – protein and protein – lipid interactions. The current structural model for HDL proposes that APOA1 forms an antiparallel dimer in which helix 5 in monomer 1 associates with helix 5 in monomer 2 along a left – left (LL5/5) interface, forming a protein complex with a two-fold axis of symmetry centered on helix 5. However, computational studies suggest that other orientations are possible. To test this idea, we used a zero-length chemical cross-linking reagent that forms covalent bonds between closely apposed basic and acidic residues. Using proteolytic digestion and tandem mass spectrometry, we identified amino acids in the central region of the antiparallel APOA1 dimer of HDL that were in close contact. As predicted by the current model, we found six intermolecular cross-links that were consistent with the antiparallel LL5/5 registry. However, we also identified three intermolecular cross-links that were consistent with the antiparallel LL5/4 registry. Molecular dynamic simulations suggest that that LL5/5 and LL5/4 APOA1 dimers possess similar free energies of dimerization, with LL5/5 having the lowest free energy. Our observations indicate that phospholipidated APOA1 in HDL forms different antiparallel dimers that could play distinct roles in enzyme regulation, assembly of specific protein complexes, and the functional properties of HDL in humans.


INTRODUCTION
One important risk factor for atherosclerosis is a low level of HDL cholesterol (HDL-C), which has been proposed to protect against cardiovascular disease by removing excess cholesterol from artery wall macrophages (1,2). Thus, the risk of cardiovascular disease is inversely proportional to plasma levels of HDL-C and its major apolipoprotein, APOA1 (3)(4)(5). However, alterations in HDL's protein cargo change the lipoprotein's antioxidant and anti-inflammatory properties (6,7), creating particles that might be less able to protect against cardiovascular disease. Moreover, HDL carries protein families implicated in complement activation and the regulation of proteolysis, raising the possibility that they modulate the innate immune system and alter the cardioprotective properties of HDL (8).
The biogenesis of HDL begins with the secretion and phospholipidation of lipid-free APOA1 by the liver and intestine to form circulating discoidal particles containing two molecules of APOA1 along with free cholesterol and amphipathic phospholipids (9). These immature HDL particles then interact with lecithin: cholesterol acyltransferase (LCAT), which catalyzes the transfer of fatty acyl components from the sn-2 position of phosphatidylcholine to the hydroxyl moiety of free cholesterol (1,10). The resulting hydrophobic cholesteryl ester migrate to the core, forming mature spherical HDL particles (1,11). intermolecular salt bridges between the proteins (19). Importantly, the proposed docking faces of the two proteins have a left-left orientation (19), in which helix 5 in monomer 1 associates with helix 5 in monomer 2 along an antiparallel interface (LL5/5). The crystal structures of the lipidfree N-terminal and C-terminal truncated forms of APOA1 strongly support the proposed configuration (18,28), with both structures demonstrating extensive salt bridging between the APOA1 belts and an axis of symmetry in helix 5 in the central region of the dimer centered on the fifth repeated helical domain of APOA1 (18,28).
In the original description of APOA1 in discoidal HDL, it was noted that two other orientations of the LL ring pair, 5/4 and 5/6, also exhibited high calculated scores for potential intermolecular salt bridging (19). Subsequent work with cysteine mutants of APOA1 (29) supported the possibility that the LL5/4 orientation also exists in reconstituted HDL particles (rHDL). However, no direct biochemical evidence supports the existence of the LL5/4 orientation or other dimers in phospholipidated APOA1 of reconstituted discoidal HDL (rHDL) or human HDL.
To determine whether multiple structurally distinct dimers might form under physiological conditions, we used chemical cross-linking in concert with tandem mass spectrometry (MS/MS) (30)(31)(32) to search for intermolecular crosslinks in the central region (helices 4-6) of APOA1 in HDL. Our results strongly support the existence of both LL5/5 and LL5/4 APOA1 dimers in human HDL.
Human HDL. Plasma was prepared from EDTA-anticoagulated blood from four healthy adults who had fasted overnight. HDL (density 1.125-1.210 g/ml) was isolated from plasma by sequential ultracentrifugation (35) and stored at −80°C. HDL was further fractionated by highresolution size exclusion chromatography (Superdex 200, 350 μL flow/min) at 4°C using PBS.
Nine fractions (0.5 mL) with the apparent size of human HDL (7-12 nm diameter) were collected and used for analysis.
HDL particle concentration and size (HDL-PIMA). Analyses were performed on a scanning mobility particle sizer spectrometer (TSI Inc., Shoreview, MN, model 3080N) fitted with a nanodifferential mobility analyzer (TSI Inc., model 3085) and a charge-reducing electrospray ionization source (CR-ESI; TSI Inc., model 3480). Monodisperse particles exiting the differential mobility analyzer were detected by a condensation particle counter (TSI Inc., model 3788).
Samples were introduced into the electrospray chamber every 15 min by automated loop injections. Particle concentration was determined by external calibration (36). MS/MS data contained in the mzXML files were subsequently searched against the database for light labeled cross-links, using xQuest (version 2.1.1) (37,38). For rHDL, the database was built from the sequence of recombinant APOA1. For human HDL, the database contained the sequences of thirty-nine proteins (from Uniprot database) consistently detected in HDL in previous studies (6,8,39,40). xQuest searches for theoretical cross-links whose masses match measured precursor masses, and it subsequently assigns fragment masses from MS/ MS spectra, assuming that a cross-linked precursor will fragment at only one peptide bond. Spectra were searched using the default settings of xQuest (37) except for the cross-linked residues and the mass shift. The detailed parameters are the following: cross-linked residues: K, D, E; mass shift for cross-linked peptides, −18.010564686; MS ion mass tolerance, 10 ppm; MS2 ion mass tolerance for common-ions: 0.2 Da; MS2 ion mass tolerance for cross-linked ions: 0.3 Da; allow matching second isotopic peak; fixed modification, Cys carbamidomethylation; enzymatic digestion, trypsin; allowed number of missed cleavages, 2 (37). Matches from all searches were required to have less than 5% false-discovery rate (FDR) based on a target-decoy calculation strategy (a reverse decoy database) and more than 15% of the ion current in a given MS/MS spectrum assigned as b-and y-type ions. All matches were manually inspected to ensure that both peptides in the cross-linked product had been sequenced and that most of the abundant fragment ions could be assigned in the MS/MS spectra. The matched cross-links were also identified using two other search engines, Kojak (41) and SIM-XL (42).
Molecular dynamic simulations. Initial configuration of the planar double belt structures of LL5/5 (19) and LL5/4 were minimized for 10,000 steps using a conjugate gradient algorithm followed by 1 ns simulation in vacuo. All simulations were performed with using NAMD 2.12 (43) and the CHARMM 36 force field (44). Constant temperature and pressure was maintained at 310 K and 1 atm, respectively, using Langevin thermostat and the Nose-Hoover Langevin piston method (45). A time step of 2 fs was used, and all bond lengths involving H-atoms were subject to constraint using the SHAKE (46) and SETTLE (47) algorithms. Non-bonded interactions were cut off at 12 Å. The particle-mesh Ewald (48) was used to compute full electrostatic interaction.

RESULTS
We used EDC (1-ethyl-3-[3-dimethylaminopropyl]carbodiimide hydrochloride) (49) to generate intramolecular and intermolecular cross-links in APOA1 of rHDL and native HDL. Importantly, all cross-linking reactions were carried out at low concentrations of EDC in phosphate buffered normal saline at pH 6.5, which much more closely mimics physiological conditions than those used for crystallization of proteins. EDC reacts with the amino group of lysine resides in close proximity to the carboxylic acid group of aspartate and glutamate to form an amide bond. Crosslinks formed by EDC represent solvent-accessible salt bridges.
The maximum distance between the α-carbons in the amino acid residues of the cross-linked amino acids is the sum of the length of two side chains plus the length of the amide bond (10.5 Å for the K-D linkage and 12.1 Å for K-E). Considering a 3 Å motion-averaging factor (50, 51), we considered distances < 13.5 Å for K-D linkages and < 15.1 Å for K-E linkages as consistent with a proposed structure of APOA1.

APOA1 prepared using heavy and light isotopes distinguishes between intermolecular and intramolecular cross-links.
To distinguish between inter-and intra-molecular cross-links of APOA1 in rHDL, we used a mixture of human [ 14 N]APOA1 (light, L) and [ 15 N]APOA1 (heavy, H) (isotopic purity >99%) to generate rHDL particles (52). When rHDL containing two molecules of APOA1 is generated with

Chemical cross-linking in rHDL and MS/MS analysis reveal 5/5 and 5/4 APOA1 dimers.
Discoidal rHDL containing 2 molecules of APOA1 were prepared from phospholipid, cholesterol, and APOA1. Following cross-linking with EDC, the reaction mixture was fractionated by highresolution size exclusion chromatography to eliminate cross-linked HDL particles. The size of the discoidal rHDLs isolated by this procedure was 96 Å, as measured by calibrated ion mobility analysis (Fig. 1B) (36). We term these particles rHDL-2-100, indicating that they contain 2 moles of APOA1 and have a diameter of ~100 Å. These particles have been extensively characterized (34).
Chemical cross-linking and MS/MS analysis of rHDL-2-100 prepared with a 1:1 mol ratio of L and H APOA1 identified 11 inter-protein cross-links (Table I). Nine were between amino acids in the central region (helices 3-7) (Scheme 2) of the LL5/5 antiparallel APOA1 dimer in discoidal HDL proposed by Segrest and colleagues (19). The LL5/5 orientation is also present in two different crystal structures of non-lipidated N-and C-terminal truncated forms of APOA1 (18,28). Distances between the two α carbons in the linked residues (Table I) were determined by mapping the cross-linkers onto the crystal structure of the C-terminal truncated APOA1 dimer and simulated structure of lipid-bound APOA1 dimers (LL5/5 and 5/4) (28). The distances from the non-lipidated crystal structure are larger than the salt bridge distances in the lipid-bound forms of LL5/5 due to a small rotation in contacts between the antiparallel helical domains required for lipid-association (51). Two cross-linking distances were not available because they are involved in the C-terminal end of APOA1, which the crystal structure lacks. Six cross-links were between residues that were 5-15 Å apart in the crystal structure of APOA1, which is consistent with the theoretical maximum distance (15.1 Å) between EDC cross-linked amino acids. The positions of these cross-links are in excellent agreement with those predicted by the crystal structure of the truncated lipid-free APOA1 dimer (28).
We also identified three intermolecular cross-links that exceed the maximum plausible crosslinking distance in the crystal structures of truncated APOA1: K96-E147, K107-E125, and K133-E111. Molecular simulation of APOA1 dimers indicated that all three cross-links are consistent with the LL5/4 antiparallel dimer of APOA1 originally proposed by Segrest and colleagues (29). The distances of these three cross-links in LL5/4 were 8.7 Å, 12.9 Å and 9.8 Å, respectively. Based on the relative intensities of the ion current chromatograms for the intermolecular cross-links (Fig. 3A), the LL5/5 is likely to be the major conformation of the two dimers in rHDL-2-100.
The MS and MS/MS spectra of representative cross-links, K118-E147 from the 5/5 and K133-E111 from the 5/4 registry, are shown in Fig. 2A-B. In both MS1 spectra, there are three groups of cross-link peaks corresponding to the linkages of L-L, L-H and H-H, indicating that both are inter-molecular crosslinks. In the MS/MS spectra of cross-links from the L-L form of APOA1, fragments from both the α and β chains were observed and the majority of the abundant peaks can be assigned to predicted fragment ions. We also identified three cross-links (K12-E76, K23-D48 and K45-E34) in the N-terminus (Supplemental Table S1), that are consistent with the configuration in the crystal structure of a C-terminal truncated form of APOA1 (∆185-243, PDB ID 3R2P, Supplemental Fig. S1). For the C-terminus, we identified two cross-links: E198-K208 and K208-E223 (∆1-43, PDB ID 1AV1, Supplemental Fig. S2).
We also observed one intra-protein cross-link, K238-E34, which is strong evidence that the Nterminus of APOA1 is in contact with its own C-terminus in rHDL-2-100.
To determine if the size of HDL particles influenced the formation of 5/5 and 5/4 antiparallel dimers, we subjected rHDLs of 80 Å diameter (rHDL-2-80) and 120 Å diameter (rHDL-2-120) to cross-linking with EDC, proteolytic digestion, and MS/MS analysis. All of the intermolecular cross-links in the central region of APOA1 detected in rHDL-2-100 were also detected in the smaller and larger rHDL particles (Supplemental Table S1). The apparent abundance of the cross-links was similar to that observed in rHDL-2-100 as monitored by their total ion chromatograms, strongly suggesting that the size of the particles was not a major determinant of the formation of the 5/5 and 5/4 antiparallel APOA1 dimers.

APOA1 in human HDL contains both 5/5 and 5/4 homodimers
To determine whether APOA1 in human HDL might also form both LL5/5 and LL5/4 antiparallel dimers, we isolated spherical HDL from plasma by ultracentrifugation (d=1.019-1.063 g/mL), cross-linked the protein, digested it proteolytically, and analyzed it with MS/MS. The apparent size of the HDL used in these analyses was 89 Å as determined by calibrated ion mobility analysis (Fig. 1C) (36). This approach identified 10 intermolecular cross-links in the central region of the APOA1 dimer ( Table II). Four of the crosslinks were consistent with the LL5/5 antiparallel dimer and two were consistent with the LL5/4 antiparallel orientation. However, we did not detect two cross-linked peptides identified in rHDL-2-100, K107-D157 or K133-E125, in human HDL. We also observed one cross-link, K94-E125, that is consistent with 5/2 registry (19,53). Based on the relative intensities of the total ion current chromatograms of the crosslinks (Fig. 3B), it is likely that the LL5/5 isomer is the most abundant conformation in native HDL.
To further confirm the identity of cross-links, we compared the MS and MS/MS analyses of the cross-links from human HDL with those of two synthetic peptides, K96-E169 (synthetic peptide 1) and K96-E147 (synthetic peptide 2), that mimicked the proposed cross-link between the APOA1 dimers. Except for the anticipated mass shifts due to the incorporation of isotopic residues into the synthetic peptides, the MS and MS/MS spectra of the cross-links from human HDL and the synthetic peptides were indistinguishable (Fig. 4).  Fig. 4B). MS/MS analysis of synthetic peptide1 (Fig. 4B) also demonstrated the same fragmentation pattern as that of the native cross-link found in EDCtreated human HDL. Furthermore, the retention time of the synthetic peptide was virtually identical to that of native cross-link (Fig. 4B, Inset). For K96-E147 from APOA1 in the LL5/4 registry, the C-terminal residues Lys and Arg in the synthetic peptide 2 were replace by Lys-NH2 and [ 13 C6 15 N4]Arg-NH2 respectively, which would increase the mass of the peptide by 8 Da. As demonstrated in Fig. 4C and 4D 4+ . Importantly, the fragmentation pattern and retention times of the native and synthetic cross-links were indistinguishable. Taken together, these observations strongly support the identification of the cross-link K96-E169 from the LL5/5 registry and K96-E147 from the LL5/4 registry in human HDL.

Molecular modeling of the 5/5 and 5/4 antiparallel dimers in rHDL.
We created a closed planar circular model of APOA1 as described previously (54). We manually created the desired dimers and then, to maximize the formation of potential interchain salt bridges and other types of bonds, utilizing the low dielectric constant of vacuum. The idealized circular LL5/5 and LL5/4 structures were minimized for 10,000 steps using the conjugate gradient algorithm. The structures then were subjected to molecular dynamic simulations for 1 ns after fixing all backbones and sidechains, excepting the sidechains located in the wp (wheel position) 2, wp5 and wp9. We previously demonstrated that these were capable of forming interchain salt bridges (55). Using zero-length chemical cross-linking, proteolysis, and MS/MS, we identified nine peptide dimers that support the idea of intermolecular cross-linking in the central region (helices 4-6) of rHDL-2-100. Peptides derived from both heavy and light APOA1 provided unequivocal evidence that these cross-links were intermolecular. Six of the cross-links were consistent with the antiparallel LL5/5 APOA1 dimer observed in the crystal structures of the lipid-free C-terminal and N-terminal truncated forms of APOA1 (18,28). However, the distances between the remaining three intermolecular cross-linked peptides were not consistent with this orientation.
Instead, all three cross-links were entirely consistent with the distance constraints of the LL5/4 antiparallel registry first proposed by Segrest et al (19,29).
We also identified ten cross-linked peptides derived from helices 3-7 in spherical HDL isolated from human plasma by ultracentrifugation. Four of the cross-linked peptides were consistent with the 5/5 registry of HDL and two were consistent with the LL5/4 registry. However, there were some differences in the cross-linked peptides we detected in rHDL-2-100 and human HDL, likely reflecting differences in structure related to differences in the size and/or shape of discoidal and spherical HDL. Collectively, these observations strongly support the presence of two distinct antiparallel dimers in HDL -structural isomers that appear to be uncommon in proteins based on our search of the literature.
It is important to note that the crystallographic studies of APOA1 used non-lipidated and truncated forms of the protein and that the proteins were crystalized in high concentrations of salt, which can alter proteins' structures. We performed all of our cross-linking reactions in physiological buffer at pH 6.5, and the human APOA1 in the reconstituted discoidal and native spherical HDL particles was fully phospholipidated. These observations strongly suggest that the structures we observed are physiologically relevant and likely mimic those of HDL particles in vivo in humans.
Symmetry between subunits is the most common configuration in homodimeric proteins because it maximizes stability while avoiding protein aggregation (56, 57). The basis for the existence of two symmetrical homodimers in APOA1 is likely due to the similar energetics of salt bridging in the LL5/5 and LL5/4 orientations, suggesting that it is an intrinsic property of the protein. Moreover, we observed both orientations in rHDLs of different sizes, where APOA1 is the only protein, and in human HDL, which associates with more than 70 different proteins (8).
Thus, neither the size of rHDL, nor the interaction of APOA1 with other proteins is likely to be the driving force for the formation of different dimers.
Molecular dynamic simulations suggested that the LL5/5 and LL5/4 APOA1 antiparallel dimers should possess similar-but not identical-free energies of dimerization, with LL5/5 being favored with one-two additional salt bridges between the antiparallel chains. Consistent with this observation and the relative intensities of the ion currents for the antiparallel dimers, it is likely that LL5/5 is the major conformation of the two APOA1 homodimers present in both rHDL-2-100 particles and human HDL. The similarity of the summed ion currents of the cross-linked peptides detected for rHDL and human HDL suggests that the relative abundance of the isomers is similar in the two types of particles. A key question is whether the formation of both LL5/5 and LL5/4 dimers in APOA1 associates with functional advantages as well as with energetic expediency. For example, this behavior might permit the protein to attain multiple conformations with different functional and metabolic properties. It is noteworthy that our molecular modeling demonstrated marked differences in the surface plots of highly charged and aromatic amino acids on the solvent-accessible faces of the LL5/5 and LL5/4 homodimers. For example, aromatic residues appear positioned to form π-π bonds in LL5/5 (Y192-F171 and perhaps Y166-Y100), whereas no π-π bonds can be identified in LL5/4. These differences might affect APOA1's interactions with other proteins and lipids in ways that modulate HDL's cardioprotective, metabolic, and immune properties. Future studies need to investigate whether differences in APOA1's dimer structure lead to functional differences among HDLs in vivo and if disease states in humans associate with alterations in dimer structure.
In summary, our results strongly support the presence of two antiparallel dimers of APOA1 in HDL, with important implications for the lipoprotein's structure and functional properties.

Scheme and Figure Legends
aromatic residues, space filling magenta.  1 Relative to theoretical masses of cross-linked peptides. 2 Distance between α-carbons of cross-linked amino acids in the crystal structure of lipid-free Cterminal truncated APOA1 dimer (∆185-243, PDB ID 3R2P) (28)