Flipped C-Terminal Ends of APOA1 Promote ABCA1-Dependent Cholesterol Efflux by Small HDLs

BACKGROUND: Cholesterol efflux capacity (CEC) predicts cardiovascular disease independently of high-density lipoprotein (HDL) cholesterol levels. Isolated small HDL particles are potent promoters of macrophage CEC by the ABCA1 (ATP-binding cassette transporter A1) pathway, but the underlying mechanisms are unclear. METHODS: We used model system studies of reconstituted HDL and plasma from control and lecithin-cholesterol acyltransferase (LCAT)–deficient subjects to investigate the relationships among the sizes of HDL particles, the structure of APOA1 (apolipoprotein A1) in the different particles, and the CECs of plasma and isolated HDLs. RESULTS: We quantified macrophage and ABCA1 CEC of 4 distinct sizes of reconstituted HDL. CEC increased as particle size decreased. Tandem mass spectrometric analysis of chemically cross-linked peptides and molecular dynamics simulations of APOA1, the major protein of HDL, indicated that the mobility of C-terminus of that protein was markedly higher and flipped off the surface in the smallest particles. To explore the physiological relevance of the model system studies, we isolated HDL from LCAT-deficient subjects, whose small HDLs (like reconstituted HDLs) are discoidal and composed of APOA1, cholesterol, and phospholipid. Despite their very low plasma levels of HDL particles, these subjects had normal CEC. In both the LCAT-deficient subjects and control subjects, the CEC of isolated extra-small HDL (a mixture of extra-small and small HDL by calibrated ion mobility analysis) was 3- to 5-fold greater than that of the larger sizes of isolated HDL. Incubating LCAT-deficient plasma and control plasma with human LCAT converted extra-small and small HDL particles into larger particles, and it markedly inhibited CEC. CONCLUSIONS: We present a mechanism for the enhanced CEC of small HDLs. In smaller particles, the C-termini of the 2 antiparallel molecules of APOA1 are “flipped” off the lipid surface of HDL. This extended conformation allows them to engage with ABCA1. In contrast, the C-termini of larger HDLs are unable to interact productively with ABCA1 because they form a helical bundle that strongly adheres to the lipid on the particle. Enhanced CEC, as seen with the smaller particles, predicts decreased cardiovascular disease risk. Thus, extra-small and small HDLs may be key mediators and indicators of the cardioprotective effects of HDL.

HDL crosslinking results.xQuest is an algorithm that can search for theoretical crosslinks whose masses match measured precursor masses and to subsequently assign the fragment masses of MS/ MS spectra.In its interpretation of MS/MS spectra, xQuest assumes that a crosslink precursor will fragment at only one peptide bond.For our searches, xQuest was used with default settings, except that mass shifts for crosslinking products were manually set to −18.010564686 for the "xlink mass-shift." The modified residues were set to Lys, Asp, and Glu.Matches from all searches were required to have precursor mass errors ≤4.3 ppm and ≥5% of the ion current in a given MS/MS spectrum assigned as b-and y-type ions.xQuest matches for intra-protein and inter-protein crosslinks were required to meet different scoring thresholds.Intra-protein crosslink matches were required to have xQuest scores ≥25.Inter-protein crosslink matches were required to have xQuest scores ≥29.Inter-protein crosslink matches were disregarded unless three unique fragments were assigned on each peptide chain.Spectra matches were assessed for this characteristic after assigning neutral losses and second isotopic peaks.In addition, crosslink matches were required to have at least 8 of these sequence-specific fragments assigned.Finally, spectra that corresponded to crosslinks, either intra-protein or inter-protein, with multiple possible linkage patterns, were manually inspected.Exact linkages were proposed only when two crosslinked residues could be unequivocally defined by the observed fragmentation.
Incubation of control and LCAT-deficient plasma with LCAT.Plasma (150 µL) from 3 LCAT-/-subjects and 3 control subjects were incubated with or without 50 µg/mL recombinant human LCAT 32 at 37ºC for 1 h followed by addition of DTNB (2 mM final concentration) to inhibit the enzyme.Because DTNB interferes with Amplex red cholesterol assay, prior to addition of DTNB an aliquot (20 µL) was taken for cholesterol assays and immediately chilled on ice.ABCA1-specific CEC was measured in ABCA1expressing BHK cells as described in Methods, Figure 5 and Figure S3.
Molecular dynamics (MD) simulations of the APOA1 structures in rHDL.We used Anton 2 (a specialized supercomputer designed to accelerate molecular dynamics simulations) to study the dynamic structure of APOA1 molecules.Anton 2 rapidly generates all-atom trajectories for much longer time intervals (10's of microseconds) than possible on other computer systems which greatly increases confidence in the results because of increased sampling.
All-atom and CG simulations of r-HDL-90 (20 µs all-atom, 15 µs simulated tempering, 200 µs CG).8][29] The initial configuration of a 20 µs-long all-atom simulation of r-HDL-90 was developed by removing POPC and cholesterol from a disc of r-HDL-100. 27he disc was solvated by water and 0.15 mM NaCl, using CHARMM-GUI. 29The box was cubic ( =  =  = 137 Å) and the pressure was applied isotropically.The total number of particles was ~258,000, including 100 POPC and 10 cholesterol.The system was first equilibrated for ~2 ns on an in-house computer cluster (Biowulf) to generate an initial structure that was then further developed on the Anton-2 supercomputer.
To enhance sampling of the conformation space, a series of successive ST simulations totaling 15 µs on the Anton-2 supercomputer extended the 20 µs-long conventional MD simulation of r-HDL-90.One hundred and seventeen temperatures, ranging from 310 K to 450 K and distributed exponentially, 28 comprised the temperature ladder of ST simulations.Prior to ST simulations, the 20 µs frame of conventional MD was simulated for 5 ns at the 117 temperatures.The following equation determined the weights for the first ST simulation: , Equation 1where , , , and  ( are the weight, temperature, energy, and Boltzmann's constant.
The weights and maximum temperatures were further adjusted for subsequent ST simulations on an ad hoc basis to allow for more frequent transitions and to avoid long residence at certain temperatures.The exchange was allowed between adjacent temperature ladders with the following probability: ;.
Equation 2 Note that the probability is determined by the weight difference between two ladders rather than each weight.
A 200 µs-long CG simulation with the Martini force field on Gromacs (www.gromacs.org,version 2016.6) completed the study of r-HDL-90.The CG simulation frames were converted to all-atom for developing the contact maps.
All-atom and CG simulations of r-HDL-80 (1 µs all-atom, 200 µs CG).The initial configuration of a 1 µs-long all-atom simulation of r-HDL-80 was developed by removing POPC and cholesterol from the last configuration of the 20 µs-long conventional MD of r-HDL-90.The disc was solvated by water and 0.15 mM NaCl, using CHARMM-GUI. 29he number of phospholipid and cholesterol molecules in each size of r-HDL were determined using nondenaturing gradient gel electrophoresis. 30The box was cubic ( =  =  = 126 Å) and the pressure was applied isotropically.The total number of particles was ~ 200,000, including 50 POPC and 5 cholesterol (the number of phospholipid and cholesterol molecules are per particle, following the notation of our previous simulations. 27We choose the relatively large box size for r-HDL-80 (compared to that of r-HDL-90) because its APOA1 terminals exhibit enhanced mobility relative to r-HDL-90 particles.The use of a sufficiently large box size ensured that the protein does not interact with its periodic image during MD simulation.Consequently, reducing the number of lipids in rHDL-90 by half to obtain rHDL-80 does not necessarily result in a significantly smaller box size.
The system was first analyzed for ~ 2 ns on an in-house computer cluster (Biowulf).The main simulation was carried out on the Anton-2 supercomputer.A 200 µs-long CG simulation with the Martini force field on Gromacs completed the study of r-HDL-80.The CG simulation frames were converted to all-atom for developing the contact maps.
Additional details of all-atom simulations.
CHARMM36 was used for lipid and protein force field parameters.The TIP3P water model modified for CHARMM was used to describe water molecules.Lennard-Jones (LJ) parameters of Na + and Cl -as well as Na + and selected oxygens of lipids and proteins were taken from the CHARMM 36 ion force field parameters (NBFIX).
In-house all-atom trajectories were generated using CHARMM with a leapfrog Verlet algorithm and a time step of 2 fs.Temperature and pressure were kept at 310 K and 1 bar using Nose-Hoover thermostat and Langevin piston barostat ( = 0), respectively.Masses of the temperature and pressure pistons were 20% and 2% of the system masses, respectively.Lennard-Jones potentials were terminated at 12 Å, with a smoothing function operating between 8 Å and 12 Å.Electrostatics were evaluated using particle-mesh Ewald with approximately 1 grid point per Å, a sixth-order spline interpolation for the complementary error function, a real-space cutoff of 12 Å, and k = 0.32.All bonds to hydrogen atoms were constrained using SHAKE.
For the Anton2 simulation, a multigrator, which minimizes sources of error associated with limited-precision arithmetic and truncation errors, generated trajectories with a time step of 2 fs.Temperature and pressure were kept constant at 310 K and 1 bar, respectively, using a variant of Nosé-Hoover and the Martyna-Tobias-Klein. Electrostatic forces were calculated using the u-series method.Water molecules and all bond lengths to hydrogen atoms were constrained using M-SHAKE.

Figure S3 .
Figure S3.Scatter plots of the correlations of macrophage (J774 cells) and ABCA1 (BHK cells) CEC of serum HDL with HDL subspecies for all subjects in the LCAT study cohort.CEC was quantified as described in Fig. 5 of the paper.r, Pearson correlation.