An atomistic characterization of high-density lipoproteins and the conserved “LN” region of apoA-I

Abstract

The physicochemical characteristics of the various subpopulations of high-density lipoproteins (HDLs) and, in particular, their surface properties determine their ability to scavenge lipids and interact with specific receptors and peptides. Five representative spheroidal HDL subpopulation models were mapped from a previously reported equilibrated coarse-grained (CG) description to an atomistic representation for subsequent molecular dynamics simulation. For each HDL model a range of finer-level analyses was undertaken, including the component-wise characterization of HDL surfaces, the average size and composition of hydrophobic surface patches, dynamic protein secondary structure monitoring, and the proclivity for solvent exposure of the proposed β-amyloid (Aβ) binding region of apolipoprotein A-I (apoA-I), “LN.” This study reveals that previously characterized ellipsoidal HDL_3a and HDL_2a models revert to a more spherical geometry in an atomistic representation due to the enhanced conformational flexibility afforded to the apoA-I protein secondary structure, allowing for enhanced surface lipid packing and lower overall surface hydrophobicity. Indeed, the proportional surface hydrophobicity and apoA-I exposure reduced with increasing HDL size, consistent with previous characterizations. Furthermore, solvent exposure of the “LN” region of apoA-I was exclusively limited to the smallest HDL_3c model within the timescale of the simulations, and typically corresponded to a distinct loss in secondary structure across the “LN” region to form part of a significant contiguous hydrophobic patch on the HDL surface. Taken together, these findings provide preliminary evidence for a subpopulation-specific interaction between HDL_3c particles and circulating hydrophobic species such as Aβ via the exposed “LN” region of apoA-I.

Significance

This study advances our understanding of high-density lipoproteins (HDLs), key players in cardiovascular health and disease. Utilizing atomistic-scale molecular dynamics simulations, we explore the structural properties of the five major HDL subpopulations to better understand their specific roles in lipid transport and receptor interactions. This research reveals novel structural distinctions among HDL subpopulations, including differences in particle morphology, surface characteristics, and the exposure of specific protein regions of functional significance. This detailed understanding is crucial for developing precise therapeutic strategies targeting cardiovascular-related conditions and lays a foundational framework for further biophysical studies of lipoproteins and their role in human health.

Introduction

The compositional profile and physicochemical properties of high-density lipoprotein (HDL) subpopulations are inherently interconnected and act together to dictate biological functionality (1,2,3). As has been demonstrated in previous studies, a coarse-grained (CG) description of representative HDL subpopulation models with average molecular compositions can provide an experimentally consistent description of the long-scale, dynamic physicochemical properties of HDL particles including lipid packing densities, particle surface curvatures, the quaternary structures of surface proteins, and the availability of bioactive components and regions (4,5). These CG models allowed inference of generalized subpopulation-specific structure-function relationships for HDLs. For example, the higher relative apolipoprotein A-I (apoA-I) concentrations and specific conformations of apoA-I in smaller HDL particle models correlated with an increased surface hydrophobicity, which was attributed to the smaller particle surface area, higher surface curvature and, thus, a greater exposure of hydrophobic regions of apoA-I due to a lower lipidation and higher protein contortion. Taken together, the relatively enhanced solvent exposure of hydrophobic regions on the surface of small HDL particles can begin to explain their superior capacity to scavenge circulating species of lower polarity such as lipids and certain hydrophobic peptides (6,7), as well as an ability to interact with subpopulation-specific receptors including lecithin cholesterol acyltransferase (LCAT) via specific hydrophobic interactions (8).

Our previous work demonstrated that the CG description of HDL subpopulation models using the Martini force field v.2.2 (9,10,11) also provides a reliable description of the long-scale dynamic properties for representative particles including the ability to describe particle self-assembly, capture significant morphological and conformational rearrangements, and assess the influence of different compositional profiles across multiple HDL model subsets (4,5). However, in the current state of the art the Martini CG model is limited in its ability to explore and characterize finer atomic levels of detail, including the dynamic interchange of hydrogen bonds and protein secondary structure transitions. These finer-level aspects may not only be critical toward developing more accurate and realistic HDL models but also in revealing how HDLs undertake diverse and subpopulation-specific biological functions such as lipid scavenging, receptor binding, and possible β-amyloid (Aβ) sequestration, the latter functionality having implications for Alzheimer’s disease (AD) (12).

The Martini CG protein model is defined such that protein secondary structure elements remain static for the duration of a simulation due to amino acid backbone beads being allocated on the basis of an initial secondary structure mapping scheme. This assumption has proven to be both practical and effective toward exploring long-scale tertiary dynamics for a variety of proteins that exhibit a predominant and essentially stable secondary structure across long time periods (13). Indeed, a number of experimental studies have demonstrated that HDL-associated apoA-I chains remain remarkably stable in terms of their total secondary structure throughout the entire HDL life cycle (14,15,16,17) and are thus well suited to a static secondary structure regime (18,19,20,21). However, this assertion of suitability is predicated upon the assumption that the initial mapping scheme used to describe the CG apoA-I model accurately reflects that of its prevailing secondary structure in nature. It is thus interesting to consider that previous CG Martini studies of HDLs broadly assumed an entirely helical structure for apoA-I chains, whereas circular dichroism spectroscopy findings indicate that total protein helicity ranges from 61% to 76% for apoA-I in circulating HDLs (22). To improve upon a fully helical apoA-I model, the isolated multifoil models used in the development of CG HDLs in our previous work (4) initially underwent atomistic molecular dynamics (MD) simulations in neat chloroform to rapidly reduce total protein helicity to experimentally consistent quantities prior to CG mapping. Given that this approach yielded apoA-I models with representative helicity values, the non-specific and artificial nature of such a chloroform-induced protein destabilization does not guarantee apoA-I structures that accurately reflect the prevailing local secondary structure preferences of in situ apoA-I chains within the context of a full HDL particle. Thus, it is unclear as to the implications of using a potentially misrepresentative apoA-I secondary structure mapping scheme for the overall physicochemical properties of CG Martini HDL models.

It is well established that apoA-I is structurally and functionally important for periphery HDL (22,23,24,25,26); however, the extent to which local apoA-I conformations influence the structure, stability, and bioactivity of spheroidal HDLs has remained difficult to quantify experimentally (27). In nascent discoidal HDLs, the prominently repeated antiparallel α-helix (AαH) secondary structure motif throughout associated apoA-I chains plays a critical structural role in these particles (23,28). It is understood that AαHs are similarly important for spheroidal HDLs (24), particularly given the relatively high proportion of total apoA-I helicity found throughout periphery HDL subpopulations (22), as well as evidence that AαHs readily adsorb to phospholipid (PL) monolayers and can help to stabilize monolayer curvature (23). From a structural perspective, it was demonstrated in our previous study that the broadly helical mapped sections of apoA-I typically correlated with predominantly 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) and cholesteryl oleate (CO) interactions (4). Intriguingly, the prevailing non-helical helix-5/helix-5 (H5/H5) juncture of apoA-I chains across the three multifoil models represented a preferred site for free cholesterol (CHOL) migration, while the apoA-I termini regions (helix 1, H1, and helix 10, H10) tended to exhibit a lower relative helicity and corresponded with higher relative quantities of trioleate (TO) interactions. Taken together, the prevailing local conformations of apoA-I chains in those representative CG Martini HDL models seemed to inform the species-specific spatial arrangement of HDL lipids within the particle. Yet the biological relevance of these local annular lipid localization trends remains unclear, given that this could simply be a consequence of a static and misrepresentative apoA-I secondary structure that permits otherwise unrepresentative protein-lipid interactions. At worst, a fixed apoA-I secondary structure in CG Martini models of HDL has the potential to modify the local and perhaps global structures and dynamics of HDL models.

The interdependence of apolipoproteins and lipids within HDLs not only informs the structure and organization of each component, but this molecular interplay also almost certainly regulates HDL functionality in a reciprocal and dynamic fashion throughout the HDL life cycle (29). There is evidence that the spatial arrangement and composition of HDL molecular components indirectly informs HDL-subpopulation-specific bioactivity by modulating the solvent exposure and conformation of functional domains in apoA-I (6). For instance, in smaller HDL a lower relative quantity of lipids and, thus, smaller available particle surface area naturally imposes conformational restrictions on apoA-I relative to larger particles, which enhances protein solvent exposure and likely modulates the availability of specific binding regions. It has also been proposed that the lower relative fraction of CHOL in the lipidome of smaller, dense HDL particles could increase the surface lipid monolayer fluidity to further enhance protein exposure and conformational changes (30). In this respect, local exposure of apoA-I from a non-polar lipidic HDL environment to the solvent could intuitively beget significant protein conformational changes (such as the destabilization of AαHs), which may be critical to HDL bioactivity. Given that such conformational changes are otherwise inaccessible in a Martini CG protein model, a detailed assessment of the structure-function relationship of apoA-I would benefit from a finer resolution.

This study aims to characterize the subpopulation-specific interactions and characteristics within the HDL interactome, placing a particular emphasis on the seven-residue ⁴²LNLKLLD⁴⁸ sequence of apoA-I (herein referred to as “LN”). “LN” is a highly conserved and predominantly hydrophobic sequence preceding the lipid-binding domain of apoA-I (31) and has been identified as a potential binding site for the Aβ peptide (32). The prevailing consensus across experimental and computational studies of apoA-I is that the “LN” region exhibits a proclivity for solvent exposure despite being composed of mostly hydrophobic residues (33) and displays a transient secondary structure which is highly dependent on its local environment (33,34,35,36). Given that HDL subpopulations exhibit distinct compositional profiles that inform apoA-I structure and exposure, it stands to reason that the “LN” sequence and, potentially, Aβ binding may be structurally and functionally modulated depending on HDL subtype. Indeed, HDL size has been identified as a modulating factor for effective Aβ binding, such that larger HDL₂ particles have a relatively reduced capacity to bind Aβ compared with HDL₃ (37). Hence, determining how HDL composition and, thus, subpopulation impact the HDL interactome—and specifically the “LN” sequence in apoA-I—may have wider implications in terms of effective Aβ binding, modulation, and trafficking, and more broadly for AD pathophysiology (2,12,38). For these reasons, there is a need to further characterize representative HDL subpopulation models at an atomistic scale to effectively and comprehensively capture the finer-level structural and dynamic aspects that may inform HDL bioactivity.

There is also a limitation on the accessible length and time scales for atomistic simulations of large macromolecular systems such as HDL particles given their considerable system size and complexity. To this end, a multiscale MD simulation approach involving the sequential interchange from CG to atomistic resolution (or vice versa) for the development and refinement of HDL models has been previously demonstrated to be an efficient and effective approach toward overcoming the limitations of each method and allowing for analysis at both levels of accuracy (39,40,41,42). Yet the scope of previous multiscale simulation studies has typically focused on single, idealized HDL models and/or discoidal particles rather than realistic, representative spheroidal HDL subpopulation species. This study seeks to address this knowledge gap by mapping the five representative CG HDL subpopulation particles developed and characterized in our previous study (4) to an atomistic representation and simulating each model for a further 1 μs. In this way, the research described in this study aims to: 1) evaluate key morphological and structural properties for each HDL subpopulation model at an atomic resolution; 2) characterize the interactome for each HDL model in atomistic detail in terms of overall and local hydrophobicity trends; 3) assess apoA-I secondary structure transitions and tendencies within and between HDL subpopulation models; and 4) assess the solvent exposure and prevailing conformations for the proposed Aβ binding domain (“LN”) of apoA-I across each HDL subpopulation model.

Atomistic simulations of five representative spheroidal HDL subpopulation models revealed that the enhanced conformational flexibility afforded to apoA-I allows for enhanced surface lipid packing and lower overall surface hydrophobicity across all HDL models relative to previous CG models. Significantly, this capacity for apoA-I to undergo dynamic secondary structure transitions facilitated a morphological transition for HDL_3a and HDL_2a models from an ellipsoidal shape to a more spherical geometry, consistent with experimental characterizations (43,44,45,46). Remarkably, despite significant secondary structure rearrangement across apoA-I chains in all HDL models, the overall total apoA-I helicity proportions remained consistent with experimental measurements, and the quaternary multifoil apoA-I models were largely retained for the duration of each simulation. In agreement with previous CG characterizations, the proportional surface hydrophobicity and apoA-I exposure reduced with increasing HDL size. Intriguingly, solvent exposure of the “LN” region of apoA-I was exclusively limited to the smallest HDL_3c model for the given simulation times due to there being a local deficiency of HDL lipids to stabilize this segment on the HDL surface. This exposure of “LN” typically corresponded to a loss of secondary structure throughout the segment to represent a proportionally significant contiguous hydrophobic patch on the HDL surface. Taken together, these findings provide preliminary evidence for a subpopulation-specific interaction between HDL_3c particles and circulating hydrophobic species such as Aβ via the exposed “LN” region. More broadly, this work highlights the importance of undertaking a multiscale MD simulation approach to HDL modeling to overcome the limitations of each simulation technique (CG and atomistic) and provide a more comprehensive structural and dynamical description for HDL models.

Materials and methods

Simulation methods

Multiscale approach

The atomistic HDL subpopulation models presented in this study were derived from the five representative CG HDL models developed and characterized in our previous work (4). The relative molecular composition of each HDL model is detailed in Table S1. The overall model formulation process followed a multiscale approach broken into three stages: 1) atomistic protein preparation; 2) CG formation/equilibration; and 3) atomistic simulation and characterization (Fig. 1). The first two stages of this approach are discussed in depth in (4). In brief, idealized multifoil apoA-I arrangements composed of three, four, and five chains of apoA-I were constructed in a fully helical atomistic arrangement as described in (22), subjected to short (2-ns) MD simulations in chloroform to reduce total helicity reflective of experimental quantities, then mapped to a CG representation and superimposed above the surface of equilibrated lipid droplets, using the Martini force field v.2.2 (9,10,11). These CG lipid-protein complexes (HDLs) were then simulated for a further 25 μs to allow for protein incorporation across the particle surface as well as sufficient lipid diffusion and overall particle equilibration. For stage 3, final molecular structures were extracted from each CG HDL MD simulation trajectory and used as templates for atomistic mapping prior to subsequent MD simulations. This final stage allows for relatively finer-detail investigations and analyses of specific molecular behaviors including HDL interactome characterizations, as well as scrutinizing the potential for Aβ binding via the proposed apoA-I binding region, “LN.”

Figure 1.

Schematic of the multiscale approach undertaken for the simulation of HDL systems. Blue boxes represent atomistic representations, and green boxes represent CG representations. To see this figure in color, go online.

Atomistic simulations

Final equilibrated CG 5-lipid HDL systems were stripped of water and ion molecules prior to structure mapping to an atomistic representation. The backward protocol (47) was followed to map each HDL system from CG into atomistic representations consistent with the GROMOS 54A7 united-atom force field (48). In-built mapping schemes were used to convert POPC, CHOL, and protein residues; lysophosphatidylcholine (LysoPC), CO, and TO mapping schemes were generated according to the “backward” documentation. The “initram.sh” bash wrapper script included with the backward package was used to generate final atomistic structures of each system. In short, “initram.sh” performs an initial projection of atoms onto the CG topology, followed by two short (500 steps) energy minimization runs, then four position-restrained MD simulations in vacuum (300 K) with a step-wise increasing time step up to 2.0 fs. Final backmapped united-atom structures (with explicit atom counts ranging from approximately 10,500 atoms for HDL_3c to more than 35,000 atoms for HDL_2b, based upon the GROMOS 54A7 united-atom parameter set (48,49)) were then centered in a dodecahedron box and hydrated with 55,000–75,000 water molecules, representing a minimum distance of 24 Å between the surface of each HDL model and the simulation box edge. Na⁺ and Cl⁻ counter-ions were added to neutralize the charge of each HDL system.

Atomistic MD simulations were performed with the Gromacs 4.6.7 package (50). The standard GROMOS 54A7 united-atom parameter set (48) was used to describe protein molecules and counter-ions; POPC, LysoPC, and CHOL were modeled using the GROMOS 53A6L force field (which forms part of the GROMOS 54A7 parameter set (48,49)). United-atom models of CO and TO consistent with a modified version of the GROMOS 54A7 force field were generated by the Automated Topology Builder (ATB) v2.2 (51). The ATB-generated parameter sets for CO and TO were found to be in good agreement with previously parameterized atomistic models and experimental data (see Fig. S7). Water molecules were described by the SPC potential (52).

Each atomistic HDL and lipid droplet system was subjected to a short (fewer than 500 steps) energy minimization using the steepest descent algorithm prior to MD simulation. All systems were simulated for a total of 1.0 μs, with the first 500 ns monitored for system equilibration and the final 500 ns of the simulation used for analysis. Atomistic systems were considered equilibrated when a range of properties including radius of gyration, particle volume, and particle density had converged to stable values across rolling 50-ns time periods in the first 500 ns of MD simulation.

Lipids, protein, and solvent (water + counter-ions) were weakly coupled separately to a temperature bath at 310 K using the Nosé-Hoover extended ensemble thermostat (53,54) with a coupling time constant of 0.1 ps. Pressure was maintained at 1 bar isotropically via the Parrinello-Rahman barostat (55) with a time constant of 2.0 ps. Electrostatics were handled by the particle mesh Ewald scheme with a 0.8-nm real-space cutoff, a 0.12-nm reciprocal space gridding, and splines of order 4. The neighbor grid searching method was applied, and the neighbor list was updated at each time step. The LINCS algorithm (56) was used to constrain the bond length. Newton’s equations of motion were integrated using a leap-frog algorithm with a time step of 2.0 fs.

Analysis methods

A variety of the Gromacs 5.0.2 tools were used to measure the equilibrium properties of the atomistic HDL models across the final 500 ns of simulation. Radial density profiles were calculated on a component-wise basis from the center of the HDL particle using gmx rdf. Dynamic protein secondary structure was monitored using DSSP via the gmx dssp wrapper. Molecular contacts were measured with the gmx mindist protocol. Radii of gyration (R_g) and axial moments of inertia were measured with gmx gyrate following alignment and fitting of each HDL particle to its principal axes of inertia. Axial moments of inertia were used to determine eccentricity, η, for each HDL model according to Eq. 1 (57):

$$\eta =1-\frac{I_{\min }}{I_{\text{av}}}\text{.}$$ (1)

The effective physical radii, R_s, for HDL particles were predicted from R_g according to Eq. 2 (as described in (58,59)):

$$R_{\mathrm{s}}=\sqrt{\frac{5}{3}}\left(R_{\mathrm{g}}\right)\text{.}$$ (2)

Component-wise solvent-accessible surface area (SASA), hydrophobic and hydrophilic SASA, and approximate particle densities were calculated using gmx sasa with a probe radius of 0.14 nm. Hydrophobic surface patches were approximated every 2.5 ns using the QUILT program (https://github.com/plijnzaad/quilt), which utilizes the method for detecting hydrophobic patches via delineation of the SASA into contiguous segments composed solely of non-polar atom types, as first described by Lijnzaad et al. (60). The component-wise composition of hydrophobic patches was estimated as the proportion of contributing atoms per HDL component to the total number of exposed atoms within the contiguous patch.

Results and discussion

Structure and the interactome of HDL particles

Following 1 μs of MD simulation at an atomistic resolution, visual inspection of each representative HDL model reveals that all particles remained stable as a single macromolecular aggregate, with apoA-I chains organized across the surface in a contorted multifoil arrangement (Fig. 2; for a full axial presentation of each model, see Fig. S1). Individual apoA-I chains encircled discrete monolayer lunes composed primarily of POPC and LysoPC. As with previous CG models (4,5), the average HDL particle size progressively increased with increasing molecular composition, whereas average particle densities gradually decreased consistent with HDL subtype classifications (Fig. 3). Calculated R_s values for each subpopulation model fell within error of experimental ranges (61); however, for all intermediate subtypes (HDL_3b, HDL_3a, and HDL_2a) the average R_s was slightly overestimated relative to experimental quantities. This is not unexpected, given that R_s is calculated under the assumption that HDL particles can be treated as a solid sphere with a uniform density according to Eq. 2. In practice, however, embedded apoA-I chains impart a higher concentrated density at the particle surface leading to greater R_g values than that of a solid sphere of an equivalent average density, resulting in a slightly inflated R_s. Naturally, this effect was absent for the earlier CG Martini models of HDLs given that all regular CG beads are almost exclusively defined with an equivalent mass (small s-type beads being the single exception); therefore, CG HDL particles exhibited a largely uniform density. Intuitively, the real effective particle radius for the current atomistic HDL models should be expected to fall somewhere between calculated R_g and R_s values (Table 1), and indeed making this allowance results in reasonable HDL particle size estimates. Nonetheless, perhaps a less ambiguous measure of experimental consistency for these models is their average predicted particle densities, which are in excellent agreement with experimental ranges for each HDL subpopulation model (Fig. 3). Taken together, according to these metrics the five HDL models are consistent representations of their respective HDL subtype.

Figure 2.

Final snapshots of each mature HDL subpopulation model showing the apoA-I H5/H5 junction. Lipid molecules are drawn as Van der Waals spheres and colored according to the following convention: POPC headgroup atoms (gray) and tail atoms (cream); LysoPC headgroup atoms (brown) and tail atoms (tan); CHOL atoms (gold); CO atoms (dark orange); TO atoms (dark green). Individual apoA-I chains are drawn in a cartoon representation showing secondary structure, surrounded by a transparent surface, and colored individually by apoA-I chain (sky blue, red, bright green, purple, and orange). To see this figure in color, go online.

Figure 3.

Predicted HDL particle diameter versus particle density. Vertical error bars reflect the standard deviation in density predictions; horizontal error bars show the largest average size difference in radial semi-axes for each HDL model. Experimentally quantified ranges are shown as intersecting rectangles and colored according to HDL subpopulation. To see this figure in color, go online.

Table 1.

Summary of the average R_g, R_s, moments of inertia, and η for each atomistic HDL model

HDL system	Rg	Rs	Ixx	Iyy	Izz	Ixx/Iyy/Izz	η	HDL radius (exp) (nm) (278)
	(nm)	(nm)	(104 amu nm2)	(104 amu nm2)	(104 amu nm2)
3c	3.0	3.8	65.9	75.7	76.9	0.9:1:1	0.10	3.6–3.9
3b	3.2	4.1	94.7	105.7	109.0	0.9:1:1	0.08	3.9–4.1
3a	3.6	4.6	177.3	197.3	204.0	0.9:1:1	0.08	4.1–4.4
2a	3.8	5.0	214.8	248.8	253.3	0.9:1:1	0.10	4.4–4.9
2b	4.6	5.9	554.0	586.1	589.3	0.9:1:1	0.04	4.9–6.5

Evaluation of the average moments of inertia across simulation trajectories of HDL subpopulation models indicates that a spheroidal morphology is maintained for each particle, in agreement with experimental characterizations (Table 1). Interestingly, the resulting single-particle form factor for all HDL subpopulation models approaches a perfect sphere (η ≤ 0.10), in contrast to the prevailing prolate ellipsoidal morphology measured in CG HDL subpopulation models of HDL_3a and HDL_2a. In the case of these two HDL models, each backmapped particle was initially in an ellipsoidal form that underwent a rapid morphological transition toward a more spherical structure within the first 50 ns of the atomistic simulation, before stabilizing within the subsequent 200–300 ns (Fig. 4 a). This transition corresponded to a progressive total increase in the average apoA-I bend angle for all chains, indicative of an increase in protein contortion (Fig. 4 b). Remarkably, the enhanced apoA-I contortion following HDL morphological readjustment did not significantly affect the average proportion of helicity across protein chains (Fig. 4 c). Taken together, these findings are consistent with the view that a fixed apoA-I secondary structure in CG Martini models of HDL has the potential to constrain global particle structures to otherwise unfavorable morphologies.

Figure 4.

Changes to the morphology of the HDL3a model in an atomistic representation, reflecting a more spherical particle corresponding to (a) a reduction in the HDL3a radius of gyration (R_g) over the first 250 ns of simulation. This morphological change in HDL3a was accompanied by (b) an overall increase in apoA-I bend angles for associated chains to indicate protein warping across the surface. (c) Remarkably, enhanced protein bending in HDL3a did not greatly affect the average proportion of apoA-I helicity. To see this figure in color, go online.

The morphological transition from an ellipsoid toward a spherical HDL particle for HDL_3a and HDL_2a models following the transition from CG to atomistic resolution is interesting within the context of a previous multiscale MD simulation study of HDL_3b-like particles reported by Catte et al. (40). In that study, which began with a short 10-ns all-atom MD simulation followed by CG mapping and a subsequent 1-μs MD simulation, the initial atomistic component yielded an HDL particle with a predominantly spherical geometry that ultimately transitioned into a prolate ellipsoid during the subsequent CG simulation. In that study, the initial spherical HDL geometry was dismissed as an artifact stemming from the short atomistic simulation timescale (40), as opposed to being an artifact of the CG treatment of protein secondary structures. While the present atomistic MD simulations each extend over durations comparable to those of the CG component in that earlier work, it is imperative to acknowledge that direct temporal equivalences between atomistic and CG simulations are not straightforward. This complexity necessitates caution in directly attributing the observed spherical HDL geometries in this study solely to intrinsic geometric preferences of these particles. These variances in HDL geometry between CG and atomistic models underscore the inherent differences between and, thus, limitations of, both approaches. Atomistic representations of biomolecules allow for greater conformational freedom and flexibility relative to CG approaches at the expense of long-scale dynamics. Nonetheless, ellipsoidal and spherical particles have been experimentally characterized within the HDL subfraction (43,44,45,46), and thus the possibility of dynamic morphological transitions is conceivable throughout the HDL lifecycle.

Following atomistic mapping to final CG HDL structures, the total average apoA-I helicity content across all chains stabilized to quantities largely consistent with experimental measurements for each HDL subpopulation model (Fig. 5). Specifically, α-helicity tended to increase with an increasing HDL size, consistent with the notion that higher surface area and surface lipid content promotes better protein incorporation into the particle surface to more effectively stabilize AαHs. The time-wise secondary structure profiles (Figs. S2–S6) indicate that the conformations of apoA-I chains in each HDL remained relatively stable and were similar to the initial CG mapping schemes generated for multifoil models. This finding suggests that the atomistic protein model is relatively more conformationally flexible than the CG model, as opposed to a significant secondary structural rearrangement following backmapping. In terms of overall secondary structure variability, apoA-I chains associated with HDL_3c demonstrated the highest variation in secondary structure formation, consistent with these chains having the lowest overall α-helicity and, thus, reduced quantity of lipid-stabilized AαHs. More broadly, this finding is likely a reflection of the smaller relative particle size and lower available surface lipids, leading to a poorer incorporation of apoA-I into the particle surface and, thus, a higher level of protein contortion and exposure.

Figure 5.

Average total percentage of α-helical content for all apoA-I chains associated with HDL subpopulation models (blue) relative to experimental characterizations (pink) (22). Error bars represent the standard error in helicity predictions. To see this figure in color, go online.

HDL surface hydrophobicity is a key determinant of structure and function throughout the HDL life cycle (62). As with our previous CG models (4,5), the relative proportion of solvent-exposed hydrophobic surface area is enhanced in smaller HDLs, with almost 40% of the SASA of HDL_3c being hydrophobic in nature (Fig. 6 a). In contrast to the proportional solvent-exposed hydrophobic surface area for HDL_3c, the largest HDL_2b model has an average proportional hydrophobic surface area of 33%, although as Fig. 6 b demonstrates, this accounts for the highest average net hydrophobic surface area (540 nm²), which is three times larger than that for HDL_3c (180 nm²). Consistent with the view that exposure of hydrophobic apoA-I regions is a critical modulator for bioactivity in smaller HDL, the contribution of apoA-I to the average net hydrophobic surface area is considerably higher for smaller HDL, whereas the contribution by HDL lipids increases with increasing HDL size (Fig. 6 b). Remarkably, the average total apoA-I hydrophobic surface area remains relatively constant across all HDL subpopulation models (Fig. 6 b), which could reflect the functional importance of exposed hydrophobic regions of apoA-I to the entire HDL life cycle. Having said this, on a per-chain basis, the contribution of apoA-I regions to the total hydrophobic surface area naturally decreases with increasing HDL size, generally consistent with a view that the functional importance of apoA-I to HDL biology decreases with HDL maturation.

Figure 6.

Hydrophobic surface properties for HDL subpopulations. (a) Average hydrophobic surface area as a percentage of the total solvent-accessible surface area. (b) Component-wise breakdown of the total hydrophobic surface area. (c) Component-wise breakdown of the average largest conjoined hydrophobic patch. Corresponding values for component-wise contributions can be found as part of Table S1. All error bars represent the standard error in hydrophobic SASA predictions. To see this figure in color, go online.

There is an energetic cost to exposing hydrophobic surfaces which is proportional to the total surface area of HDL particles. The fact that the smallest HDL_3c model exhibits the highest proportional hydrophobic surface area is notable within the context of its overall structure as well as its functionality. Structurally, the higher relative hydrophobic surface area suggests that smaller HDLs exhibit relatively less stability than their larger counterparts, which is consistent with enhanced apoA-I flexibility (characterized for CG models in (4,5)) and greater secondary structure variability (Figs. S2–S6) as well as increased surface exposure of core lipids (4). From a functional perspective, this likely ensures that smaller HDLs are superior scavengers for circulating hydrophobic species such as lipids and are presumably more capable of interacting with subpopulation-specific receptors via hydrophobic interactions to undertake distinct biological functions. However, the relative proportion of hydrophobic surface does not necessarily confer surface-specific HDL activity. Rather, the size, stability, and compositional makeup of contiguous hydrophobic patches across a surface can provide a more complete measure for the identification and evaluation of functionally relevant recognition sites on the HDL interactome. Hence, the largest contiguous hydrophobic patch at the surface of each HDL particle was monitored across the final 500 ns of atomistic simulations to better gauge the functional significance of surface hydrophobicity in the overall HDL interactome.

In Fig. 6 c, the component-wise breakdown of the average largest conjoined hydrophobic patch is presented for each HDL subpopulation model. These plots demonstrate that the size of hydrophobic patches and the proportional lipid contribution to patches increases with increasing HDL particle size. On the other hand, the apoA-I contribution to the average largest conjoined hydrophobic surface patch for the HDL_3c particle is markedly larger as a proportion and as a net total than for the four other larger HDL subtypes. Indeed, as a function of simulation time the component-wise breakdown of the largest hydrophobic surface patch corresponding to HDL_3c (Fig. 7, top) demonstrates that the apoA-I contribution is not only significantly higher as a proportion of the surface patch across the simulation time but also that the contribution remains relatively stable for considerable time periods. At the other extreme, the apoA-I proportional contribution to the largest hydrophobic surface patch in the HDL_2b model is almost negligible in a relative sense, whereas large hydrophobic POPC patches predominate. Interestingly, the time-wise variability in hydrophobic surface patch size is highest for the HDL_2b model (Fig. 7), perhaps being indicative of the increased fluidity afforded to surface monolayer lipids in larger HDLs as well as a relative reduction in apoA-I flexibility due to greater protein incorporation (4).

Figure 7.

Component-wise breakdown of the largest hydrophobic surface patch versus time across the final 500 ns of simulation time for each HDL model. To see this figure in color, go online.

Within the context of HDL subtype functionality, the compositional differences in the largest hydrophobic surface patches across HDL subpopulation models share some consistencies with the known HDL bioactivities along the reverse cholesterol transport (RCT) pathway. For instance, smaller HDL₃ particles are high-affinity substrates for ABCA1, phospholipid transfer protein, and LCAT, which act to facilitate lipid cargo uptake and for which binding is mediated via direct interactions with HDL-associated apoA-I. On the other hand, cholesterol ester transfer protein binding increases with increasing HDL size and proceeds via non-specific interactions with the lipid monolayer surface of HDL to facilitate HDL lipid efflux. In terms of HDL maturation, the adsorption of additional apoA-I chains to HDLs is understood to proceed predominantly via non-specific lipid interactions following the uptake of additional lipid cargo, thus within the context of the transitional HDL subpopulation models (HDL_3b, HDL_3a, and HDL_2b), it follows that the composition of hydrophobic patches is skewed toward surface lipids. Nevertheless, protein-protein interactions are understood to stabilize apoA-I following initial adsorption, and thus the relatively constant contribution of hydrophobic apoA-I surface to the overall HDL hydrophobic surface for these transitional species may point toward a secondary stabilizing role for such hydrophobic regions of associated-apoA-I during additional apolipoprotein adsorption and, thus, particle maturation.

Structure and solvent exposure of “LN” in HDL subpopulation particles

In addition to the RCT pathway, HDL has been identified as a binding partner for the hydrophobic Aβ peptide, with broader implications for the pathophysiology of AD. Interestingly, the effectiveness of HDL to bind Aβ has been shown to decrease with increasing HDL size, despite larger particles having a greater available total surface area for interaction (37). Within the context of the present interactome characterizations for the five representative HDL subpopulation models, it thus stands to reason that HDL-Aβ interactions proceed via exposed hydrophobic regions of surface proteins in small HDLs as opposed to the relatively smaller available hydrophobic lipid surface. Such a mechanism of interaction is consistent with an earlier study demonstrating that the presence of lipid-free apoA-I effectively impedes Aβ aggregation in vitro (63), along with a subsequent study that identified the highly conserved and predominantly hydrophobic seven-residue “LN” region of apoA-I as a potential high-affinity Aβ binding site (32). In lipid-free apoA-I, the “LN” region exists as a solvent-exposed loop domain despite its considerable hydrophobicity (33,34,35,36). Intriguingly—and perhaps of more biological relevance—considerable secondary structure variability has been assigned to “LN” in lipidated apoA-I, such that the local level of “LN” lipidation appears to confer some influence on its prevailing structure and conceivably its solvent exposure (33,35,36,64,65,66,67,68,69,70). Altogether, there is strong evidence to indicate that Aβ binding via HDLs proceeds in a subpopulation-specific fashion that is dictated by particle size, composition, relative degree of apoA-I hydrophobicity, and more specifically the conformation and exposure of “LN” on the particle surface.

To this end, Fig. 8 a presents the average size and composition of contiguous hydrophobic surface patches containing a portion of at least one “LN” region of apoA-I for each HDL subpopulation model. Within the context of potential apoA-I-mediated Aβ binding, it is significant that on average the largest contiguous hydrophobic surface patch containing “LN” was observed for HDL_3c and is only marginally smaller than the global average largest hydrophobic patch for HDL_3c (Fig. 6 c: 6.33 nm² versus 7.48 nm²), accounting for roughly 4% of the total particle hydrophobic surface. On the other hand, the average largest “LN”-containing hydrophobic surface patch for HDL_2b is less than one-third the size of the global average largest hydrophobic patch (4.46 nm² versus 15.89 nm²; <1% of the particle hydrophobic surface area), demonstrating that the relative contribution of “LN” to the hydrophobic interactome in the largest HDL model is significantly diminished compared with HDL_3c. What is more, the average apoA-I contribution to the hydrophobic surface area progressively decreases with increasing HDL size to indicate that the hydrophobic contribution of “LN” to the HDL interactome gradually reduces with HDL maturation. Given that HDL maturation is characterized by a relative increase in the lipid-to-protein ratio and thus a progressive improvement in surface protein incorporation, it stands to reason that the solvent exposure of the mostly hydrophobic “LN” region of apoA-I naturally diminishes with increasing particle size. Under the assumption that the “LN” region of apoA-I is the primary mediator of HDL-Aβ binding, these findings provide some insight as to why effective Aβ binding is reduced for larger HDLs (37).

Figure 8.

(a) The largest hydrophobic surface patches containing the “LN” region of apoA-I were found in HDL3c and were composed of the highest proportion of protein surface. (b) Hydrophobic “LN” patches in HDL3c exhibited a majority protein composition throughout the final 500 ns of simulation. (c) Secondary structure of the “LN” region (background, primary axis) correlated with lipid interaction at the surface as measured by the minimum distance between the α-carbon of Leu42 and the HDL lipid surface (purple curve in the foreground, secondary axis). Error bars in (a) and (b) represent the standard error in hydrophobic surface patch predictions. To see this figure in color, go online.

Regarding the time-wise availability of the largest “LN”-containing hydrophobic patches, Fig. 8 b demonstrates that the relative protein contribution to the largest “LN”-containing patch remained stable throughout the final 500 ns of simulation. Visual inspection of identified patches indicated that the overwhelming majority was composed of the “LN” region from one particular apoA-I chain that exhibited a lower relative lipidation in this region. Moreover, this corresponding “LN” domain exhibited predominantly random coil conformation throughout the simulation (Fig. 8 c, background) and remained exposed away from the HDL surface for significant time periods (Fig. 8 c, foreground). On the other hand, helical structures predominated within the other “LN” regions across the five HDL models, wherein lipidation was notably higher and sustained throughout the simulations (Figs. S2–S6). On this note, it is interesting to consider that for the time periods where the identified “LN” region assumed some helical structure in Fig. 8 c, the minimum distance between the α-carbon of its Leu42 residue and the HDL_3c lipid surface was minimal, indicating lipid-protein interactions. Furthermore, these time periods tended to correspond with a relatively reduced hydrophobic patch size (Fig. 8 b), indicating that “LN” hydrophobicity is maximal when in a random coil conformation and when not in contact with the HDL lipid surface. Indeed, the largest recorded contiguous hydrophobic surface patch observed on the surface of HDL_3c (16.61 nm², corresponding to 8% of the total hydrophobic surface area at ∼819 ns in Fig. 8 b) was primarily composed of the non-polar atoms of “LN” in a fully extended coil arrangement (Fig. 9). While this configuration did not exhibit significant stability, its presence within the context of the overall hydrophobic and unstructured behavior of this “LN” region in the HDL_3c model provides evidence for the lipidation state of “LN”—and more broadly the compositional profile of HDL subpopulations—to play a crucial role in moderating “LN” solvent exposure and, potentially, HDL-Aβ binding.

Figure 9.

Snapshot showing the largest conjoined hydrophobic surface patch observed for the HDL3c model, composed primarily of non-polar apoA-I atoms and including a solvent-exposed segment comprising the carbon atoms of the “LN” region. The hydrophobic surface patch is drawn as a transparent magenta surface, with the “LN” side chains drawn in a thin licorice representation and colored according to residue type. To see this figure in color, go online.

Conclusions

This work explores the structure and interactome of representative HDL subpopulation models at an atomistic level following backmapping and subsequent microsecond-long simulations of each equilibrated CG Martini HDL structure characterized in our previous study (4). The limitations of the Martini CG model including a static protein secondary structure could be overcome to provide a more accurate description of the overall structure and interactome for HDL subpopulation particles, particularly within the context of HDL-mediated binding of the predominantly hydrophobic Aβ peptide. This work has allowed for a more detailed analysis of the solvent exposure (as a measure of bioactivity) of the proposed Aβ binding site of apoA-I, a largely hydrophobic seven-residue site referred to as the “LN” region, which has been shown to exhibit considerable structural heterogeneity depending on the lipidation state of apoA-I. Given that HDL particle size had also been identified as a modulating factor in HDL-Aβ binding, this work focused on understanding how HDL size and composition may moderate “LN” surface hydrophobicity, exposure and, thus, Aβ binding.

The interchange from a CG description of HDL particles to an atomistic resolution was structurally profound for HDL_3a and HDL_2a models, both of which underwent rapid morphological transitions from an ellipsoid to a more spherical shape. This transition was primarily attributed to the higher conformational flexibility afforded to apoA-I chains, such that a progressive increase in apoA-I bending was observed in concurrence with the increase in particle sphericity. Interestingly, the enhanced capacity for apoA-I to undergo dynamic secondary structural transitions did not correlate with a significant change to the overall helicity of apoA-I chains across all HDL models, which remained consistent with the average quantities measured experimentally for each respective HDL subpopulation model. Moreover, the dynamic protein secondary structure did not visibly disrupt the overall multifoil apoA-I arrangement on the surface of HDL particles, such that all protein chains remained primarily embedded in the HDL lipid surface, and the H5/H5 junction was retained for the simulation duration across all HDL subpopulation models.

On a broader level, the present study highlights the value of multiscale MD simulations to the development and refinement of representative HDL molecular models. Specifically, the ability for apoA-I to undergo dynamic secondary structure transitions appeared to facilitate large-scale morphological transitions in HDL_3a and HDL_2a. In this respect, the observed transition away from an ellipsoidal shape for HDL_3a and HDL_2a at an atomistic scale emphasizes the structural and proximal interdependence of apolipoproteins and lipids in HDLs as well as their overall significance to the particle shape and structure. Although ellipsoidal and spherical morphologies are each valid for HDL subtypes (43,44,45,46), the rapid transition away from an ellipsoid in atomistic simulations of these HDL models raises some queries regarding the structural bias that the initial apoA-I secondary structure mapping scheme imposed on the earlier CG Martini models presented in our previous studies (4,5). As such, future studies of HDLs may benefit from additional interchanges between CG and atomistic resolutions to better account for the limitations inherent to each approach.

Given the critical importance of hydrophobicity to HDL structure and function, the global and local size and composition of hydrophobic patches on the surface of each HDL model was evaluated as a measure of the differences in HDL interactomes between subpopulations. Consistent with the previously characterized CG models discussed in our earlier study (4), the proportional surface hydrophobicity decreased with increasing HDL size. Moreover, the proportional lipid contribution to HDL surface hydrophobicity progressively increased for larger HDL models, whereas the apoA-I contribution decreased in agreement with the notion that apoA-I embedment into the HDL lipid surface is improved in larger particles where the lipid/protein ratio is high. To this end, the hydrophobic contribution of the “LN” region of apoA-I to the average size of hydrophobic patches as well as to the overall HDL interactome was maximized in the smallest HDL_3c model. Specifically, solvent exposure of the hydrophobic “LN” region was significantly higher in the HDL_3c model and typically corresponded with an extended random coil conformation maximizing hydrophobic residue exposure to the solvent. On the other hand, interaction between the “LN” region and the HDL lipid surface tended to correspond with helical propensity to suggest that local lipidation of “LN” modulates structure and solvent exposure. Broadly, this provides preliminary evidence for a subpopulation-specific interaction between smaller, protein-rich HDL species and circulating Aβ where the proclivity for an exposed, unlipidated “LN” region of apoA-I is highest.

Perhaps of greater pertinence to the present findings for “LN” in apoA-I-associated HDLs is that the preceding 40 residues of apoA-I are absent in all protein models used in this study, given the lack of reliable structural information available within the context of spheroidal particles (as discussed in (4)). Granted, the N-terminal domain is not believed to play a major structural role in the maturation of HDL particles, nor is there strong evidence to suggest that its presence drastically modifies the proclivity for solvent exposure of “LN.” Nevertheless, the proximity of the N-terminal domain of apoA-I to “LN” must confer at least some influence on the overall tertiary structure and, thus, accessibility of “LN” to Aβ within the context of full HDL particles. For these reasons, some caution must be exercised when interpreting and extrapolating the present results for “LN” to a biological context.

It is also difficult to discount the possibility that the observed “LN” exposure in HDL_3c is a singular random event or an artifact of the initial starting configuration used in this study and may be less (or more) prevalent under different circumstances. To this end, future studies would benefit from additional simulations with varied initial conditions and increased replication, providing a more comprehensive understanding of the “LN” region’s behavior across different HDL subpopulations. Such an approach would enable a more robust assessment of the variability and reproducibility of “LN” exposure in HDL_3c, further elucidating its potential role in Aβ interaction. Moreover, incorporating a broader range of compositional diversity in the representative HDL models (such as the inclusion of lower-abundance lipid species, different apolipoproteins, and/or different surface protein arrangements) may uncover additional contributing factors toward “LN” exposure.

However, all things considered, the limitations of the present study do not negate the general premise that the lipidation state of “LN”—which can broadly be related to proportional protein-lipid composition and, thus, HDL size— can modify its secondary structure and modulate its bioavailability. Consequently, the findings presented in this work related to “LN” represent a promising avenue of enquiry toward better understanding the underlying molecular mechanisms that may influence HDL-mediated binding of circulating Aβ.

Data and code availability

We are committed to promoting open science and facilitating further research in this area. Simulation files including HDL model snapshots (final CG model structures, backmapped structures, starting structures, and final structures), backmapping schemes, parameter sets, and a selection of trajectories have been made available in the following Zenodo public repository: https://doi.org/10.5281/zenodo.10836226.

Additional data can be made available on request.

Author contributions

C.J.M. and R.L.M. conceived and designed the research project. R.L.M. supervised the work, secured computational resources, and provided ongoing guidance and support. C.J.M. carried out all simulations and data analyses, and drafted the manuscript in full. R.L.M. provided critical revisions and gave final approval to the manuscript.

Editor: Siewert Jan Marrink.