When NLPs, assembled using the cholate-free method, were purified by SEC only a singe NLP rich peak was observed (Figure 1A) [6,11,12]. NLP fractions were pooled from the NLP-rich peak and subjected to native gel electrophoresis (Figure 1B). As previously observed, there was a single NLP peak by SEC (Figure 1A) and a single band by native gel electrophoresis (600 kDa) (Figure 1B – lane 2), indicating assembly resulted in homogeneous NLPs that were presumably monodisperse in size. However, multiple sized species with four discrete diameters centered at ~14.5, 19, 23.5 and 28 nm were consistently observed by AFM (Figure 1C). In addition, in previous work we showed that these discrete sizes were not a result of an artifact of AFM imaging, such as tip-sample convolution, and similar discrete sizes were observed with other high resolution techniques such as electron microscopy and ion mobility spectroscopy [11]. It is also worth noting that the majority of NLPs were smaller than 25 nm and the SEC column used in this study (Superdex 200 HR 10/300 column) was expected to efficiently separate these different sized NLP constructs.

To examine the effects of cholate, NLPs assembled using the cholate-associated method were also subjected to SEC purification, native gel electrophoresis and AFM image analysis. In contrast to the results obtained using the cholate-free method, three NLP-rich peaks were observed by SEC (Figure 1D). In addition, when these three peaks (P1, P2, and P3) were subjected to native gel electrophoresis a single band, two bands and three bands were observed in P1, P2, and P3, respectively (Figure 1 E). Unexpectedly, the single band in P1 and the highest MW bands in P2 and P3 migrated to the same MW, (600 kDa), and the lowest molecular weight band in P2 and the middle band in P3 also all migrated to the same MW (500 kDa). In addition, there was a low molecular weight band in P3 (324 kDa) that was not observed in the other two peaks. These banding patterns strongly suggest that these particles represent discrete NLP sizes. When each peak was analyzed by AFM a very similar sizing pattern was observed (Figure 1F). P3 consisted of a single size NLP centered at 23.5 nm in diameter. However, two (19.0 and 23.5 nm) and three (14.5, 19.0, and 23.5 nm) different sized NLPs were observed in P2 and P3, respectively (Figure 1F). The diameter distributions of P1, P2 and P3 corresponded very closely with the banding patterns observed by native gel, where one, two and three different NLP sizes determined by AFM and native gel bands were observed in P1, P2 and P3, respectively (Figure 1E and 1F). Interestingly, these discrete sizes were centered at the same means observed for the NLPs assembled in the absence of cholate and there was no dramatic change in particle shape, which suggests that cholate does not appreciably affect the overall structure of the NLPs.

To assess whether the addition of cholate after cholate-free NLP assembly can promote separation by SEC and native gel, 5 mM cholate (below the CMC of cholate) was added to NLPs that were assembled in the absence of cholate. Figures 1G and 1H are the SEC and native gels, respectively, of the NLPs before and after cholate addition. As shown in Figure 1D, the single NLP peak became more broad displaying two overlapping peaks after the addition of 5 mM cholate. In addition, the single band in the native gel became four bands, which closely corresponded to the diameter distribution determined by AFM. To determine the effect of the post-addition of cholate on the NLP diameter distributions, the sample was imaged both before and after cholate addition (Figure 1I). From this analysis no changes were observed in either the mean or relative fraction of each discrete size (Figure 1I). It is also worth noting that the emergence of a small lipid-rich peak in the SEC trace after cholate addition (Figure 1D) was observed, suggesting that a small fraction of the particles have started to disassemble. This was expected since detergents will also promote the formation of micelle/lipoprotein complexes [24]. However, we did not observe any micelle-like structures by AFM, indicating that the constructs in the lipid-rich region most likely do not have favorable interactions with the mica surface under the buffering conditions used.

These combined results strongly suggest that even though the NLPs were dialyzed after cholate-NLP assembly a portion of the cholate may remain associated with the NLP construct. This is not surprising since it is known that the amphiphilic nature of cholate will cause them to intercalate into the NLP bilayer, and removal through dialysis alone may not be sufficient to remove all of cholate that is anchored into the NLP bilayer. The most intriguing aspect of these results is the effect cholate has on NLP separation by SEC and native gel. These results suggest that NLPs may be interacting with one another during electrophoresis and SEC purification such that different sized particles migrate to the same spot on native gel (~ 620 kDa) and elute with similar retention times during SEC purification. Quite possibly, these interactions may be disrupted when negatively charged cholate becomes associated with the NLPs, allowing for improved separation during SEC and native gel electrophoresis. It is worth noting that cholate appears to improve NLP separation with SEC, however both P2 and P3 still contained particles with a variety of diameters and thus are still polydisperse. In addition, the fact that no change was observed after cholate addition indicates cholate does not appreciably affect the absolute size of these particles.

The observation that NLPs of different sizes all eluted in a single SEC peak when NLPs were assembled using the cholate-free method strongly suggest that the particles are interacting dynamically, but it is still unclear whether the elution pattern of these particles is consistent across the entire peak. Therefore, the elution properties of the four different NLP sizes in the single NLP rich SEC peak were resolved by collecting 27 fractions across the entire peak and subjecting select fractions to AFM image analysis. In these experiments, two 20 mg NLP assemblies were pooled so there was sufficient material in each fraction for subsequent analysis.

Figure 2A is a truncated SEC trace of the NLP-rich peak after large scale assembly using the Superdex 200 26/60 column at a flow rate of 2.5 mL/min. Similar to the small scale assemblies, only a single NLP-rich peak was observed; however the NLP retention times in these experiments were different than the small scale assemblies (Figure 1A) due to the use of a different column. The vertical lines represent the fractions that were collected throughout the NLP-rich peak. NLP diameter distributions were determined by AFM and the relative fraction normalized to the SEC trace was calculated for each NLP size. Using this technique, the relationship between NLP size and elution time throughout the entire NLP-rich SEC peak was de-convoluted (Figure 2B). Interestingly, the elution of the four NLP sizes was not consistent across the entire peak. In fact, the maximal peak in the elution of larger particles occurred at earlier times where the peak elution times for the 28.0, 23.5, 19.0 and 14.5 nm NLPs were 49.9, 50.7, 51.5, 53.1 minutes, respectively (Figure 2B). Representative AFM images from three different fractions (highlighted in Figure 2A) are shown in Figure 2C. From these images it is clear that the size distributions vary throughout the NLP-rich peak.

Interestingly, the elution width of the 14.5 nm peak was narrower than that of the other sized particles and the majority of these particles eluted slightly to the right of the center of the entire NLP-rich peak (Figure 2B). Therefore, this region of the NLP-rich peak was highly enriched in the 14.5 nm size particle. In fact ~70% of the NLPs that eluted from 52.9 – 53.3 minutes were the ~ 14.5 nm sized NLP (referred to as SEC #1 –Center Cut), as opposed to ~ 40% for the pooled NLP-rich peak (referred to as SEC #1 – All). These results suggest that a second SEC injection of this elution fraction should result in a narrowing and shift in the main NLP peak. When this was done the peak in fact did narrow and shift to the right, indicating a more homogeneous sample of smaller particles (Figure 3B). To determine if there was an increase in NLP homogeneity in the center cut of the re-injected sample various elution fractions around this region of the NLP peak were imaged by AFM. Several fractions to the right of the center cut were found to be greater than 80% enriched in the 14.5 nm NLP but the fraction outlined by the black lines in Figure 3B (referred to as SEC #2 – Center Cut) was the fraction that contained the highest purity with 90% of the NLPs being ~ 14.5 nm in diameter (Figure 3C). The diameter distributions for the SEC #2 – Center Cut elution fraction and the SEC #1 – All elution fractions is given in Figure 3C. In addition, the exact fraction of the four different sized NLPs in the SEC #1 – All, SEC #1 – Center Cut, and SEC #2 – Center Cut elution fractions are given in Figure 3D.

These combined results indicate that the different sized particles are eluting as expected based on the principles of SEC but the presumed interactions severely tighten the difference in retention time; therefore, the different elution peaks of these particles appear as one single peak by SEC. Although the interactions severely hinder the ability of SEC to purify out a single NLP subpopulation it is possible to achieve homogeneity of the smallest 14.5 nm sized NLP through iterative SEC of the elution fraction that is to the right of the peak’s center.

One of the objectives of this study was to better understand the purification process of NLPs assembled using the cholate-free and cholate-associated methods in order to better produce homogeneous NLP sub-populations. The motivation for this approach is to utilize biophysical techniques that are sensitive to sample homogeneity. One such technique is the growth of high quality crystals for structural determination using X-ray crystallography. Obtaining discoidal lipoprotein crystals suitable for the determination of structure is a challenging task, although ellipsoidally-shaped full length apoE4 structures have been determined [30]. The benefits of these structures would significantly improve rational design of membrane protein solubilizing in NLPs. ApoE422k is of particular interest for solubilization of membrane proteins as it produces larger particles compared to other apolipoproteins [12] which may be relevant when accommodating large membrane proteins and protein complexes.

To demonstrate the utility of a homogeneous and monodisperse NLP subpopulation, samples of varying homogeneity were subjected to de novo crystallization screening [31] and the quality of the resulting crystals were compared. For these studies, we chose to use NLPs assembled using the cholate-free method and side-by-side crystallization was carried out with NLPs pooled from the entire single NLP peak from the first SEC run (40% homogeneity) and NLPs from the two most homogeneous fractions from the second SEC run (90% homogeneity). Both samples showed a propensity to crystallize under similar conditions but the more homogeneous sample crystallized with higher propensity. However, there were notable differences in the crystallization habits of the two samples (Figure 4A and 4B). The less homogeneous sample tended to give smaller crystals that formed in the presence of an amorphous precipitate (Figure 4A) whereas the more homogeneous sample gave well formed crystals with less precipitate (Figure 4B). The best crystals, determined by crystal volume and clean morphology, were obtained from the more homogeneous sample. These results underline the benefit and importance of homogeneity for crystal growth. The largest crystals obtained to date from the highly homogeneous sample were ~100 μm in maximum extent and were birefringent. Optimization efforts are currently underway to obtain larger crystals.

The phospholipid 1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC) was purchased from Avanti Polar Lipids, Inc. (Alabaster, AL).

The expression clone to produce apoE422K, the N-terminal 22 kDa fragment of apolipoprotein E4 (apoE4), as a 6His and thyrodoxin tagged construct was kindly provided by Dr. Karl Weisgraber. Production and purification of apoE422K has been described in detail elsewhere [12].

NLP formation and purification both in the presence and absence of cholate have been described in detail elsewhere [12]. Briefly, dried DMPC was re-suspended in either 10 mM Tris pH 7.4, 0.15 M sodium chloride, 0.25 mM EDTA, 0.005% sodium azide (TBS) buffer followed by probe sonication to clarity (cholate-free assembly) or TBS/cholate (20 mM) (cholate associated assembly). ApoE422k (200–250 μg or 20 mg for large scale assembly) was added to the TBS/DMPC solution +/− cholate at a mass ratio of 4:1. The particle formation process was started with three repeated sets of transition temperature incubations, above (10 minutes at 30 °C, 20 minutes for large scale assembly) and below the transition temperature of DMPC (10 minutes at 20 °C, 20 minutes for large scale assembly) followed by incubation at 23.8 °C overnight (43 hours for large scale). Samples containing cholate were then dialyzed against 1000x volume of TBS buffer using 3 changes in 24 hrs. The NLPs were purified by size-exclusion chromatography using a Superdex 200 HR 10/300 column (GE Healthcare), in TBS at a flow rate of 0.5 mL/min or on a Superdex 200 26/60 at a flow rate of 2.5 mL/min for large scale preparations. The NLP fractions were concentrated to approximately 0.1 mg/mL using molecular weight sieve filters (Vivascience) with molecular weight cutoffs of 50 kDa. Protein concentration was determined using the ADV01 protein concentration kit (Cytoskeleton, Inc.).

AFM images were acquired using a MFP-3D-CF AFM (Asylum Research, Santa Barbara, CA). Details of the imaging parameters have been described elsewhere [11,12].

Sets of 384 sitting-drop experiments were designed using our CrysTool crystallization design engine (Segelke, JCG 2001). Experiments were setup in IntelliPlate sitting drop trays using 100 μL of reservoir cocktail and 400:400 μL reservoir:NLP stock solution drops. NLP stock solutions were at a protein concentration of 2 mg/mL as determined by a Bradford assay. Both the highly homogeneous and less homogeneous samples showed a high propensity to crystallize in nearly the same set of conditions, where the more homogeneous sample crystallizing with slightly higher propensity. The lower homogeneity sample gives microcrystals or better in ~26% of CrysTool generated conditions and the higher homogeneity sample gives microcrystals or better in ~35% of the conditions tested. In addition, the lower homogeneity sample gave microcrystals or better in ~26% of CrysTool generated conditions and the higher homogeneity sample gives microcrystals or better in ~35% of the conditions tested. The large majority of conditions giving rise to NLP crystals are made up of mid molecular weight polyethylene glycol as the precipitating agent (2k-6k MW), a variety of buffer and range of pH (5.5–8.5).