Disorders in lipid metabolism are related to a range of diseases, among them atherosclerosis and Alzheimer's disease (AD). In atherosclerosis, the leading cause of death in western society, elevated levels of high-density lipoproteins (HDL) are thought to have counter-atherosclerotic properties (Gordon et al., 1989). However, the presence of HDL has more recently been shown to play a neutral (Voight et al., 2012) or even negative (Madsen et al., 2017) role in aiding the prevention of this disease. Even though HDL has been used to help combat the onset of atherosclerosis (via HDL therapy or plaque remodelling therapy) (Gille et al., 2014; Van Capelleveen et al., 2014), some recent medical trials have failed to prove HDL's effectiveness against atherosclerosis (Angeloni et al., 2013; Toth et al., 2013).

In general, HDL consist of a core of triglycerides and cholesterol esters, encased in a monolayer of lipids and cholesterol, surrounded by protein. There are many different subclasses of HDL each with slightly varying size and composition (Jonas, 2002). In particular, the lipid-poor, nascent HDL (also known as Preβ-HDL) is thought to be discoidal in shape and transforms into a spherical, mature HDL particle upon esterification of cholesterol and transfer into its lipid core. While ApolipoproteinA1 (ApoA1) constitutes around 70% of total protein content in human HDL, there are various other proteins which are also key to the structure and function of HDL particles (Jonas, 2002). One of these is ApolipoproteinE (ApoE) (Utermann, 1975) which is found in most HDL subfractions but mainly in the largest, most buoyant mature subfraction (HDL2) that contains triglycerides and are lipid rich (Davidson et al., 2009). ApoE is also the most commonly found apolipoprotein in cerebrospinal fluid (Ladu et al., 2009). It plays a large role in cholesterol transport while maintaining local homeostasis of cholesterol within the brain (Mahley, 2016b).

In humans, ApoE allelic variants result in three isoforms E2, E3, and E4, with respective frequencies of 8.4, 77.9, and 13.7% and varying at only two amino acid residues, 112 and 158 (E2: 112Cys, 158Cys; E3: 112Cys, 158Arg; E4: 112Arg, 158Arg) (Weisgraber et al., 1981). These differences in the amino acid sequence largely contribute to the proteins' structure and function and determine their behaviour in the body (Chou et al., 2005; Mahley et al., 2009), and especially in the roles they play in certain diseases such as atherosclerosis and AD. The presence of ApoE2 is associated with very low risk of AD (Wu and Zhao, 2016; Reiman et al., 2020) and generally thought to be protective against atherosclerosis, apart from in the case of type III hyperlipoproteinemia which is associated with increased atherosclerotic risk. On the other hand, ApoE3 levels are neutral in both diseases while the presence of ApoE4 is indicative of both atherosclerotic advancement and the onset of AD (Mahley et al., 2009).

Lipids are important to these diseases as they coexist in both atherosclerotic and AD plaques (Kiskis et al., 2015; Sergin et al., 2016). Understanding the difference in the interaction of ApoE variants with model cellular membranes is key to deciphering the specific roles they play in the development of these diseases. The roles of lipoproteins in the development of atherosclerosis and, most likely in AD, directly involve lipid exchange. The type of lipid exchange occurring— namely deposition or removal—highly depends on the lipoprotein type present (Browning et al., 2017). Previously, HDL was shown to remove lipids from the cell membrane giving rise to the idea behind HDL therapy. Reconstituted HDL (rHDL), comparable to discoidal Preβ-HDL, exhibits similar lipid transfer properties to that of mature HDL (Davidson et al., 1995), and is used to mimic this, discoidal nanoparticles of phospholipids solubilised by encircling ApoA1 are used as artificial HDL or rHDL as a strategy in HDL therapy (Gille et al., 2014; Van Capelleveen et al., 2014). While less is known about the lipid exchange properties of ApoE, ApoE-containing rHDL (ApoE-rHDL) were shown to have potential for atherosclerosis treatment (Valanti et al., 2018), although further testing is needed.

In this paper, the initial interaction of ApoE3 and ApoE4 with saturated and unsaturated supported lipid bilayers (SLBs) is explored using neutron reflectometry (NR). Then, ApoE-rHDL are formed based on the vesicle solubilisation method often used to make ApoA1-rHDL. The ApoE-rHDL structure is further characterised using small angle neutron scattering (SANS). Finally, the lipid exchange capacity for ApoE-rHDL with model membranes is explored by NR. As ApoE is the main apolipoprotein in the brain, where there is a local homeostasis of cholesterol, the model membranes in question are in the absence or presence of cholesterol to determine the potential differences between these interactions. Comparisons are made between to ApoA1-based rHDL and mature HDL pooled from three healthy male volunteers [mature HDL is replotted from Waldie et al. (2020)].

In this work, the overall journey of ApoE is followed from its interactions with model membranes in the lipid-poor form, to the formation of a nascent-like HDL particle, and finally to the characterisation of the lipid exchange capacities for saturated lipids in the presence or absence of 20 mol% cholesterol. NR and SANS are ideal techniques for this task as they can distinguish between deuterated and non-deuterated components. Therefore, the use of deuteration in lipid-protein complexes allows highlighting of specific components within the system, together with the use of contrast variation (based on the different content of D₂O-based buffers, denoted as d-TBS) to visualise the differences between the components (Clifton et al., 2020). Here, the model cellular membranes are made up of tail-deuterated lipids and, in some membranes, the presence of non-deuterated or specifically matchout-deuterated cholesterol. The matchout deuterated cholesterol allows no net coherent scattering to be seen for this component in 100% D₂O-based buffer conditions. The protein and rHDL particles, instead, are unlabelled. The differences between the components in the membrane and the protein/rHDL allows NR to be used in determining: (1) the level of protein incorporation or lipid exchange, through the change in the scattering length density (SLD) of the tail region, (2) the amount of lipids that were removed and replaced by protein or solvent, through the change in the solvent quantity in the tail region, and (3) the amount of extra protein/rHDL bound on top of the model membrane. During the kinetics of interaction, the samples were measured in H₂O-based buffer (100% h-TBS) as this gives the best contrast against the deuterated membrane layer, while full characterisation in three contrasts (100% h-TBS, 100% d-TBS and 38% d-TBS) of the model membranes before and after kinetics were measured to increase the sensitivity and accuracy of the structural and compositional information obtained. The layer models describing the model membranes are determined from simultaneous fitting of the isotopic contrasts and allow for the analysis of the membrane composition and the decoupling of the information regarding protein incorporation/lipid exchange and lipid removal as described in the methods section and the Supporting Information. Figure 1 gives schematic representations of protein and nanodisc interactions with the bilayers. Representative structural parameters for the pristine SLBs are summarised in Table 1 and are in agreement with those reported in the literature for DMPC (Browning et al., 2018), POPC (Åkesson et al., 2012b; Waldie et al., 2020), 20 mol% cholesterol containing DMPC and POPC measured at 37°C (MMA increases with temperature and the reported values cannot be compared to MMA measured at 25°C (Åkesson et al., 2012a). POPC is fond to adopt a thicker, more expanded SLB structure than DMPC given its longer tail and the presence of double bonds that bend the acyl tails. Here, the thickening and compacting effect of cholesterol on fluid membranes is observed in agreement with previous reports (Gallová et al., 2004; Hung et al., 2007; Waldie et al., 2018, 2019). Moreover, the cholesterol molecules are expected to be fully miscible within these model membranes (Knoll et al., 1985; Barrett et al., 2013).

The samples measured by SANS (all unlabelled) were measured in three contrasts: 100% h-TBS, 100% d-TBS and 42% d-TBS. The different buffers used give varying levels of isotopic contrast; for example, the 42% d-TBS and 100% h-TBS highlight the lipids and the proteins respectively, whereas the 100% d-TBS buffer gives the largest contrast against the sample as a whole. Analysing all three contrasts simultaneously allows for higher sensitivity in the structural determination.

While the structural difference between the ApoE isoforms is subtle (Chou et al., 2005), clear differences can be seen in their behaviour upon interaction with lipid bilayers. In particular, ApoE4 binds to a larger extent to the membrane core than ApoE3 (Figure 2C), although about the same amount of protein sits on top of the membrane regardless of the lipid type, for each isoform, respectively (Figure 2D). These changes impact the ability of ApoE to bind lipids. In particular, ApoE4 has a greater ability to bind lipids due to structural differences in the helical segments of the C-terminal. This difference leads to a reduced ability to form tetramers, which are less capable than the monomer of binding lipids (Garai et al., 2011; Chetty et al., 2017). ApoE4's increased affinity for lipid binding is also suggested by the fact that ApoE4 preferentially binds to VLDL (for which its surface is about 60% lipids) rather than HDL, whereas ApoE3 binds preferentially to HDL (for which its surface is about 80% protein). Furthermore, ApoE4 lacks flexibility in its C-terminal domain leading to a preference for the lower curvature of the VLDL particles (Hatters et al., 2005; Nguyen et al., 2010). The increased affinity for VLDL leads to an increased clearance of VLDL from plasma via ApoE receptors in the liver. However, this in turn leads to LDL receptors being downregulated, raising overall levels of plasma LDL and leading to an increased risk for atherosclerosis (Dong and Weisgraber, 1996; Mahley, 2016a).

Here, we observed that most protein incorporation into bilayers (Figure 2A) occurred within the first hour for the unsaturated lipids but there was further increase with time for saturated ones. Earlier reports on the rate of lipid binding and vesicle solubilisation showed a biexponential decay when solubilising DMPC vesicles at their melting temperature of 24°C (Segall et al., 2002), with ApoE4 presenting a higher rate constant than ApoE3. Another study showed that ApoE4 displays an increased ability to disrupt DMPC vesicles compared to ApoE3 by measuring the release of fluorescent dye (Ji et al., 2002).

Recently, our group showed that native HDL, through its apolipoproteins, present a lower ability to exchange and replace unsaturated lipids as compared to saturated ones (Waldie et al., 2020). Both ApoE variants mirror this trend since these proteins remove more saturated than unsaturated lipids. Atherogenic and amyloid plaques are rich in saturated lipids and cholesterol (Touboul and Gaudin, 2014; Kiskis et al., 2015). In a recent study, a link between AD and the disruption of the metabolism of unsaturated fats such as omega 3 was found, which could provide an explanation for ApoE4's higher affinity for the unsaturated membrane (Snowden et al., 2017). Indeed, a previous report states that supplementing mono- and poly-unsaturated fats in the diet reduce the risk to AD, whereas the introduction of saturated fats increase this risk (Morris, 2004).

The fact that a larger fraction of ApoE4 binds the lipid core compared to ApoE3 and that it displays less sensitivity to the lipid type in the membrane (Figures 2C,D) could be linked to a decreased protein flexibility although NR cannot distinguish whether different ApoE molecules bind to the core and the head or if a single protein bends over across tails and heads. Indeed, ApoE4's greater rigidity forces it to maintain a more compact conformation around the lipid core which, in turn, could render it less able to bind to HDL. Earlier, ApoE4 was shown to disrupt the lipid membrane to a larger extent than ApoE3 (Mahley et al., 2009), in agreement with our results that show larger ApoE4 incorporation into the bilayers and greater structural changes in the SLBs required to fit the NR data after protein incubation. Indeed, ApoE4 could retain a more compact structure folded back on itself due to the Arg-61:Glu-255 salt bridge that could restrain the protein's structure even when associated with lipids (Raussens et al., 1998; Drury and Narayanaswami, 2005; Hatters et al., 2009).

Regarding the rHDL structure, the reports on ApoE-containing particles mainly involve electron microscopy and spectroscopic studies for structural determination (Narayanaswami et al., 2004; Newhouse et al., 2005); and ultracentrifugation (Raussens et al., 1998) and native gel electrophoresis (Raussens et al., 1998; Narayanaswami et al., 2004) for size determination. These reports are in agreement with the size of particles reported here: 10–15 nm in diameter (Figure 3 and Table 2). Contrary to ApoE-rHDL, ApoA1- or membrane scaffold protein (MSP)-based nanodiscs form particles that are about 8–12 nm in diameter (Denisov et al., 2004; Bayburt and Sligar, 2010; Skar-Gislinge et al., 2010; Del Giudice et al., 2017a). A variety of models have been presented to describe protein-lipid particles from the picket-fence model (Jonas et al., 1989), to the double superhelix (Wu et al., 2009) or double belt-like structure (Segrest et al., 1999; Midtgaard et al., 2015), though some of these have faced opposition within the nanodisc community (Jones et al., 2010; Gogonea, 2016). Ellipticity is the main structural feature which differed between the ApoE rHDL, with ApoE4 forming more elongated discs. Our data agree well with the double belt structure since the data suggest that there are two proteins present per nanodisc, also in agreement with ApoA1-rHDL. Previously reported data on ApoE-rHDL show the presence of two or more proteins per disc (Raussens et al., 1998; Narayanaswami et al., 2004; Yamamoto et al., 2008). There has been some debate as to whether ApoE3 and ApoE4 form nanodiscs in a different manner giving structural differences to the protein conformation (de Chaves and Narayanaswami, 2008) and, in turn, their lipid exchange ability. The hypotheses suggested include ApoE3 forming more of an extended belt conformation around the whole perimeter of the disc due to increased flexibility (Narayanaswami et al., 2004). The more compact structure of ApoE4 could lead to potentially more ApoE4 molecules per nanodisc compared to ApoE3 (Gong et al., 2002). However, the same number of molecules were found to be present in both nanodisc variants presented here. The NR lipid exchange data (Figure 4), on the other hand, suggest that the proteins adapt slightly different conformations giving rise to their differing exchange capabilities.

Where some reports have shown an incremental size increase of nanodiscs with increasing protein length (Denisov et al., 2004, 2005), it is of no surprise that the ApoE-rHDL particles reported here are of a slightly larger size than previously reported for ApoA1-rHDL, as the ApoE protein is longer. In turn, the number of lipid molecules per nanodisc is also higher for ApoE- than ApoA1-containing nanodiscs which does not seem to have an effect on the ability of these nanodiscs to exchange lipids: the similar size of ApoE3- and ApoE4-rHDL does not translate into similar functional capabilities, as clear differences were observed in their lipid exchange and removal affinity (Figures 4A,B). Moreover, the lipid exchange (Figure 4B) was similar between ApoE3- and ApoA1-rHDL while it was different between ApoE isoforms, despite them being of similar size but larger than ApoA1 (since the nanodisc concentration was constant across samples, smaller discs imply lower lipid concentration in the case of ApoA1). Therefore, it is clear that the rHDL size (area of the lipid nanodisc) does not determine the extent of lipid exchange; rather, specific apolipoprotein-lipid interactions must be behind the observed phenomena.

ApoA1 and ApoE have similar structures comprising amphipathic alpha-helices that enable them to form nascent HDL-like structures by solubilising phospholipids (Lund-Katz and Phillips, 2010). However, neither ApoE3 nor ApoE4 remove further lipids from the membrane than those exchanged, to a significant degree. This contrasts from what is observed for ApoA1-rHDL, which instead remove a notable quantity of lipids, although to a lesser extent than mature HDL (Figure 4A). The results obtained for ApoA1 are not surprising since it is the more abundant serum HDL protein, whose function is to clear lipids. Our results suggest that the function of ApoE cannot be mainly related to the lipid exchange, but rather to binding ApoE-receptors in the liver. Indeed, the fact that the structure of ApoE4 is more compact than ApoE3 might affect how it binds to ApoE- and other LDL receptors in the liver (Ruiz et al., 2005). However, ApoE-HDL serve as the primary brain lipoproteins, being produced there and lacking the ability to cross the blood-brain barrier (Pitas et al., 1987). The ApoE in the brain is the second most abundant place for ApoE production in the body after the liver (Mahley, 2016a). ApoE plays an important role in maintaining the homeostasis of cholesterol concentration in the brain, where this lipid comprises about 20–25% of the body's total cholesterol (Michikawa, 2006). Whilst HDL in the rest of the body participate in cholesterol efflux thereby removing cholesterol from artery walls, ApoE-HDL in the brain maintain constant cholesterol levels (Mahley, 2016b). Indeed, the current data clearly show that lipid exchange was not affected by the level of deuteration in cholesterol (Figure 4), suggesting that ApoE does not interact with cholesterol significantly. Instead, these results suggest that ApoE-enriched HDL rather transport saturated lipids over cholesterol and unsaturated lipids. By regulating the ratio of saturated lipids in cellular membranes, ApoE-HDL may help maintain membrane elasticity, which is key for the healthy function of the cell. Interestingly, cholesterol crystal nucleation starts sooner in unsaturated than saturated model biles (Halpern et al., 1993), which suggests a higher free cholesterol concentration is achieved in membranes comprised of saturated lipids. Thus, ApoE4 can potentially favour an increase in membrane saturation with cholesterol by exchanging and removing more saturated fats than ApoE3, which could potentially lead to cholesterol crystallisation in the brain.

Even though ApoE4 has a higher affinity to bind lipids than ApoE3 as demonstrated here and elsewhere (Saito et al., 2003), the ability of ApoE isoforms to catalyse the efflux of cholesterol has been disputed: some studies conclude the process is not ApoE isoform dependent (Krimbou et al., 2004) while others have found that ApoE4 has a lesser ability to perform cholesterol efflux compared to ApoE3, especially in relation to neurons in the brain (Huang et al., 1995; Gong et al., 2002; Michikawa et al., 2007). This lack of ability of ApoE4 to remove and deposit cholesterol in the brain efficiently has been proposed to be one of the main reasons for the onset of AD (Gong et al., 2002; de Chaves and Narayanaswami, 2008). Our simplified model shows that neither ApoE isoform has a special affinity for taking up cholesterol and it is therefore hypothesised that there is no significant difference in their capacity to efflux cholesterol.

Neutron reflection results suggest that ApoE4 adopts a different conformation to ApoE3 at model membranes, and that this conformation differs between saturated and unsaturated membranes for ApoE3 only. Moreover, neither ApoE isoform removes a significant amount of unsaturated lipids from the model membrane used but they were able to remove saturated lipids to similar extents. Small-angle neutron scattering was used to demonstrate that the structure of nascent-like rHDL particles made with DMPC and either ApoE3 or ApoE4 is similar, forming elliptical disc-like structures. Neutron reflection was then used to quantify the extent of lipid exchange and lipid removal by nascent-like rHDL particles and model membranes. The data show that either ApoE3- or ApoE4-rHDL particles have a low ability to remove saturated lipids as compared to ApoA1-rHDL or mature HDL. The extent of lipid exchange, on the other hand, is similar between the isoforms and is impaired by the presence of cholesterol. Finally, ApoE does not exchange or remove cholesterol molecules but rather saturated lipids. The results here mirror the physiological roles of ApoE-HDL and ApoA1-HDL particles in the brain and in serum respectively. This demonstrates that our models are suitable to study the function of these particles in a range of experimental conditions.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found at: DOIs: 10.5291/ILL-DATA.9-13-807 (FIGARO, June, September 2019), 10.5291/ILL-DATA.9-13-894 (FIGARO, February 2020) and 10.5291/ILL-DATA.8-03-979 (D11, September 2020).

SW produced the ApoE and ApoE-rHDL, participated in the planning of the experiments, performed all experiments, analysed all of the NR and SANS data and wrote the initial draft of this article. FS, NP, YG, SP, and MC participated in performing some of the experiments. FS, FR-R, and MC supported the data analysis. SW and MM produced the deuterated cholesterol and HP purified it. RDG produced the ApoA1-rHDL. TD and NY produced the POPC used. MM, SM, VF, MH, and MC participated in the planning of the experiments. MM, MH, and VF supervised activities in the ILL's Life Sciences Group. MC obtained the funding and supervised the work. All authors contributed to the discussions and editing of the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer DH declared a past co-authorship with one of the authors VF.