HDL Isolated by Immunoaffinity, Ultracentrifugation, or Precipitation is Compositionally and Functionally Distinct

The HDL proteome has been widely recognized as an important mediator of HDL function. While a variety of HDL isolation methods exist, their impact on the HDL proteome and its associated function remain largely unknown. Here, we compared three of the most common methods for HDL isolation, namely immunoaffinity (IA), density gradient ultracentrifugation (UC), and dextran-sulfate precipitation (DS), in terms of their effects on the HDL proteome and associated functionalities. We used state-of-the-art mass spectrometry to identify 171 proteins across all three isolation methods. IA-HDL contained higher levels of paraoxonase 1, apoB, clusterin, vitronectin, and fibronectin, while UC-HDL had higher levels of apoA2, apoC3, and α-1-antytrypsin. DS-HDL was enriched with apoA4 and complement proteins, while the apoA2 content was very low. Importantly, size-exclusion chromatography analysis showed that IA-HDL isolates contained subspecies in the size range above 12 nm, which were entirely absent in UC-HDL and DS-HDL isolates. Analysis of these subspecies indicated that they primarily consisted of apoA1, IGκC, apoC1, and clusterin. Functional analysis revealed that paraoxonase 1 activity was almost completely lost in IA-HDL, despite high paraoxonase content. We observed that the elution conditions, using 3M thiocyanate, during IA resulted in an almost complete loss of paraoxonase 1 activity. Notably, the cholesterol efflux capacity of UC-HDL and DS-HDL was significantly higher compared to IA-HDL. Together, our data clearly demonstrate that the isolation procedure has a substantial impact on the composition, subclass distribution, and functionality of HDL. In summary, our data show that the isolation procedure has a significant impact on the composition, subclass distribution and functionality of HDL. Our data can be helpful in the comparison, replication and analysis of proteomic datasets of HDL.

Abstract The HDL proteome has been widely recognized as an important mediator of HDL function. While a variety of HDL isolation methods exist, their impact on the HDL proteome and its associated function remain largely unknown. Here, we compared three of the most common methods for HDL isolation, namely immunoaffinity (IA), density gradient ultracentrifugation (UC), and dextransulfate precipitation (DS), in terms of their effects on the HDL proteome and associated functionalities. We used state-of-the-art mass spectrometry to identify 171 proteins across all three isolation methods. IA-HDL contained higher levels of paraoxonase 1, apoB, clusterin, vitronectin, and fibronectin, while UC-HDL had higher levels of apoA2, apoC3, and α-1antytrypsin. DS-HDL was enriched with apoA4 and complement proteins, while the apoA2 content was very low. Importantly, size-exclusion chromatography analysis showed that IA-HDL isolates contained subspecies in the size range above 12 nm, which were entirely absent in UC-HDL and DS-HDL isolates. Analysis of these subspecies indicated that they primarily consisted of apoA1, IGκC, apoC1, and clusterin. Functional analysis revealed that paraoxonase 1 activity was almost completely lost in IA-HDL, despite high paraoxonase content. We observed that the elution conditions, using 3M thiocyanate, during IA resulted in an almost complete loss of paraoxonase 1 activity. Notably, the cholesterol efflux capacity of UC-HDL and DS-HDL was significantly higher compared to IA-HDL. Together, our data clearly demonstrate that the isolation procedure has a substantial impact on the composition, subclass distribution, and functionality of HDL. In summary, our data show that the isolation procedure has a significant impact on the composition, subclass distribution and functionality of HDL. Our data can be helpful in the comparison, replication and analysis of proteomic datasets of HDL.
Supplementary key words lipoproteins • proteomics • apolipoproteins • cholesterol/metabolism • cholesterol/efflux • density gradient ultracentrifugation • dextran-sulfate precipitation • mass spectrometry • paraoxonase 1 • HDL subclass distribution Since the discovery of HDL, a multitude of methods have been developed to isolate HDL (1). Ultracentrifugation (UC) is still the most commonly used method to isolate HDL since it was introduced by Havel et al. several decades ago (2). Many different versions of UC methods have been developed, and the differences are related to centrifugation time, centrifugation force, chemicals used to adjust plasma density, osmotic pressure present, and whether or not a density gradient was used. However, HDL isolation by UC is a lengthy procedure that is less suitable for clinical use. As a result, precipitation methods have been developed for the selective removal of lipoproteins from serum. These methods can also be used to isolate intact HDL particles and commonly use polyanions such as dextran sulfate (3). Precipitation of lipoproteins by dextran sulfate in the presence of divalent cations is depended on both the positive charge and negative charge of the lipoproteins protein moiety as well as the charged groups of the phospholipids present. (4) The interaction forms insoluble complexes that precipitate and can be solubilized again by the removal the reagents. In recent years, isolation of HDL by immunoaffinity (IA) chromatography using specific antibodies for apoA1 has gained importance and represents an alternative method that reflects Alaupovic's apolipoprotein-based definition of lipoprotein classes. (5,6) Recently, Furtado et al. used IA chromatography to define 16 unique HDL subspecies based on the presence or absence of specific proteins besides apoA1 (7). IA chromatography has several advantages over UC, as it does not require centrifugal forces and high osmotic pressure and can isolate apoA1-containing particles from the entire size/ density range of HDL. While these advantages are promising, the method also has drawbacks, such as rather low yield, high cost, and concerns about the specificity of the antibodies used. Over the past decade proteomics has expanded the list of HDL-associated proteins to over 200, while lipidomics has provided further new insights into the complexity of HDLassociated lipids (8)(9)(10)(11)(12).
A key factor in the quantification of the HDL proteome besides a precise quantification method is the isolation methodology (13). Although much effort has been invested into the development of these protocols, there is little understanding of the impact of different isolation techniques on HDL composition and function. The aim of our study was to systematically compare composition and function HDL particles isolated by common methods, such as IA, UC, and dextran-sulfate precipitation (DS). For that purpose, we used a combination of untargeted (data-dependent acquisition, DDA) and targeted (data-independent acquisition, DIA) mass spectrometry methods for precise proteomic analysis of isolated HDLs and assessed key metrics of HDL function, such as cholesterol efflux capacity, paraoxonase 1 activity, and endothelial barrier promoting activity.

Blood collection
Blood was sampled from 18 healthy control subjects (inclusion criteria: apparently healthy, free of chronic disease, and currently not on medication). Three independent serum pools of six participants each, matched for age and sex, were prepared from the sera of the 18 subjects (see Table 1). All subjects signed an informed consent form in agreement with the Institutional Review Board of the Medical University of Graz. All methods were carried out in accordance with the approved by the local ethics committee (Nr.: 21-523 ex 09/10) and the principles of the Declaration of Helsinki.

HDL isolation
UC. Serum density was adjusted with potassium bromide (Sigma, Vienna, Austria) to 1.24 g/ml, and a two-step density gradient was generated in centrifuge tubes (16 × 76 mm, Beckman, Nr. 342,413) by layering 3 ml density-adjusted plasma (1.24 g/ml) underneath a KBr-density solution (1.063 g/ml) as described (14). Tubes were sealed and centrifuged at 65.000 rpm (415.000 g) for 6 h in a 90Ti fixed angle rotor (Beckman Instruments, Krefeld, Germany). After centrifugation, the HDL-containing band was collected, desalted via PD10 columns (GE Healthcare, Vienna, Austria) and either immediately used for experiments or stored with 5% sucrose at −70 • C (15).
IA purification of HDL subspecies. HDL subspecies were isolated from human sera of healthy volunteers with modifications as described (7,16). Serum was incubated overnight with gentle turning at 4 • C with Sepharose 4B resin coupled to a polyclonal anti-apoA-I antibody (Catalog # S81-104, Fortis life science) at a ratio of 0.25 ml serum per 1 ml antibody resin. The unbound fraction was removed by washing three times with PBS. For elution, apoA1-resin was incubated for 5 min with 3 M sodium thiocyanate, and the eluate collected. Elution was repeated for a total of three times. Eluted samples were concentrated on Vivaspin Turbo 15 columns (VWR, Germany), followed by buffer exchanged to PBS on PD MiniTrap G-10 columns (Cytiva Life Science).

DS.
We used a commercial available kit from Cell Biolabs (Nr.: STA-607). The isolation was performed according to the manufacturer's instructions.

Gel electrophoresis and blotting
For native gel electrophoresis, isolated HDL (5-15 μg protein per lane) was separated by native gel electrophoresis on 4%-16% gels (BN1004BOX, ThermoFisher). Gels were run at constant voltage of 150 V for 120 min. As a high molecular weight marker (NativeMark, Nr.: LC0725, Life Technologies, Austria), containing bovine serum albumin (7.1 nm), lactate dehydrogenase (8.2 nm), B-phycoerythrin (10.5 nm, apoferritin band 1 (12.2 nm), and apoferritin band 2 (18.0 nm) was used to estimate the size of HDL. Afterwards, gels were either stained with a freshly prepared solution of Coomassie Brilliant Blue G-250 overnight (ThermoFisher) or used for blotting. Gels were transferred to polyvinylidene difluoride membranes with an iBlot 2 Dry Blotting System at 100V for 7 min at RT. Membranes were probed blocked with 5 % milk in PBS for 1 hour and incubated with the following primary antibodies diluted in 5 % milk in PBS overnight at 4 • C. The list of antibodies used can be found in the supporting information (supplemental Table S1).

MS proteomic analyses.
Digested HDL proteins (50 ng) were loaded onto a trap column (nanoViper C18, 3 μm, 75 μm × 2 cm, Thermo Scientific) and eluted onto a C18 column (nanoViper, 2 μm, 75 μm × 15 cm, Thermo Scientific). Peptides were analyzed using an Easy-nLC 1200 UHPLC system (Thermo Scientific) coupled to an Orbitrap Fusion Lumos (Thermo Scientific) equipped with a nanospray FlexNG ion source (Thermo Scientific) in a 44 min gradient and normalized collision energy of 30 for HCD fragmentation. For untargeted analysis (DDA), peptides were analyzed using MS1 resolution of 120,000 (at m/z 200) with AGC target set to 4 × MS data processing. MaxQuant software (version 2.0.1.0) was used to search raw shotgun MS spectra against the human proteome (Uniprot, 20,371 entries). The criteria for protein detection and quantification included at least two peptides (at least one of them unique), with methionine oxidation and protein N-terminal acetylation selected as variable modifications and carbamidomethylation of cysteine as fixed modification. DDA data were used to build a library for DIA analyses. DIA data were analyzed using Skyline software (version 21.1.0.146) as described (17). For DIA, we used only unique peptides (at least 4 transitions per peptide) and avoided choosing peptides susceptible to ex vivo modification (i.e., containing methionine), peptides with high interference signals and mass error higher than 10 ppm. All peaks used for quantification were manually inspected to select the best transitions and ensure correct peak detection and integration. A pooled quality control (QC) was made by combining unfractionated digested HDL samples isolated by IA, UC, and DS (n = 3 each). This QC was injected 7 times intercalating with samples to control for technical variability. Proteins and peptides that achieved coefficients of variation lower than 15% in the pooled HDL QC were considered for quantification. For each protein, quantification was performed by summing up the areas of 2-6 most intense peptides. To give an estimate of protein abundance within each HDL, the value obtained for each protein quantification was divided by the theoretical number of tryptic peptides.

Arylesterase activity of paraoxonase-1
Ca 2+ -dependent arylesterase activity was determined with a photometric assay using phenylacetate as previously described. HDL (0.5 μg protein) was added to 200 μl buffer containing 100 mmol/L Tris, 2 mmol/L CaCl2 (pH 8.0), and 1 mmol/L phenylacetate to a 96-well quartz glass plate (Hellma, Baden, Germany). The rate of hydrolysis of phenylacetate was monitored by the increase of absorbance at 270 nm, and readings were taken every 15 s at room temperature to generate a kinetic plot. The slope from the kinetic chart was used to determine the increase in fluorescence per minute. Enzymatic activity was calculated with the Beer-Lambert Law from the molar extinction coefficient of 1310 mol 1 *L −1 *cm −1 for phenylacetate.
Cholesterol efflux assay J774.2 cells were maintained in Dulbecco's Modified Eagle's Medium (DMEM) in the presence of 10% fetal bovine serum and 1% penicillin/streptomycin. Cells were plated on 48-well plates (300,000 cells/well), cultured for 24 h, and loaded with 0.5 μCi/ml radiolabeled [3H]-cholesterol in DMEM supplemented with 2% fetal bovine serum and 1% penicillin/streptomycin in the presence or absence of 0.3 mM 8-(4-chlorophenylthio)-cyclic adenosine monophosphate overnight to induce the expression of adenosine triphosphate-binding cassette subfamily A member 1 (ABCA1). After labeling, cells were rinsed with serum-free DMEM containing 1% penicillin/streptomycin and equilibrated with serum-free DMEM containing penicillin/streptomycin and 2 mg/ml bovine serum albumin for 2 h. Subsequently [ 3 H]-cholesterol efflux was determined by incubating cells for 3 h with 50 μg protein/ml HDL. Cholesterol efflux was expressed as radioactivity in the cell culture supernatant relative to total radioactivity (in the cell culture supernatant and cells) of three independent experiments respectively, measured in duplicates. All steps were performed in the presence of 2 μg/ml of an acyl coenzyme A cholesterol acyltransferase inhibitor (Sandoz 58-035).
Endothelial barrier promoting activity assay 96W20idf chips (Ibidi, Germany) were incubated with 10 mM L-cysteine for 10 min at RT, washed twice with PBS, followed by a coating with 1% gelatine for 30 min at 37 • C. The human umbilical vein cell line (Ea.hy926) was maintained in DMEM containing 10% FBS, 1% HAT and 1% Penstrept, seeded at 30.000 per well, and grown to full confluence for two days. Cells were serum-starved prior to experiments and baseline recorded until stable, which was routinely achieved after 1-2 h. Afterward, 100 μg/ml HDL was added, and impedance monitored over a period of 20 h. For quantification, we used the impedance recorded at 4000 Hz at the 10 h time point.

Statistical analysis
Differences between two groups (Fig. 4C, D; with and without thiocyanate treatment) were analyzed with the Student's t test. Differences between three groups (IA vs. UC vs. DS) were analyzed with one-way ANOVA followed by Bonferroni's multiple comparison test. Significance was accepted at *P < 0.05 and **P < 0.01. Statistical analyses were performed with GraphPad Prism, Version 6, and SPSS, Version 26. Principal component analysis was performed on Perseus (v.2.0.3.1) with imputation of missing values (less than 2% of the total).

RESULTS
We isolated HDL with three different methods, being IA, UC, and DS (Table 1) from three independent serum pools. While UC is based on the density difference of HDL to other serum compounds, IA directly targets the major HDL associated protein apoA1. Precipitation uses a specific interaction of polyanions with charged groups on proteins and lipids to precipitate HDL. The baseline characteristics for the serum pools used are given in the supplemental Table S2.
To investigate the differences in the proteome of isolated HDL, we used a combination of untargeted DDA and targeted DIA mass spectrometry methods. Global protein discovery by DDA allowed us to detect a total of 171 proteins across all three isolation methods, with 142 proteins being present in IA-HDL, 112 proteins in UC-HDL, and 123 proteins in DS-HDL (Table 1 and supplemental data). We used the DDA results to construct a library to analyze DIA samples. DIA chromatograms obtained in the Skyline software were carefully analyzed, and peptides of 51 HDL proteins that were consistently and confidently detected in a pool of all samples with low coefficient of variation (<15%) were selected for label-free relative quantitative analysis. Raw data and calculations of the mass spectrometry data are provided in the Supplementary Data section.
The results from the targeted DIA proteomics clearly showed that the composition of HDL from the different isolation methods is distinct (Fig. 1A). A principal component analysis of the results showed a distinct cluster for each individual isolation method indicating that the methods are indeed different (Fig. 1B). The principal component analysis clusters further suggested that the tested methods are reproducible, which was also confirmed by gel electrophoresis analysis of individual isolations from the same serum pool (supplemental Fig. S1). The results indicate that the major HDL-associated protein, apoA-I, accounts for 53.4%-59.8% of the total isolated protein and was lowest in UC-HDL (Table 2). Further changes in UC-HDL where higher levels of apoA2, apoC3, and α-1antytrypsin, while IA-HDL had higher levels of paraoxonase 1, PLTP, apoB, clusterin, vitronectin, fibronectin as well as a series of complement proteins ( Table 2). DS-HDL was markedly enriched in apoA4 and complement proteins, while the content of apoA2, apoD, and paraoxonase 1 was very low compared to the other methods (Table 2).
To investigate the size distribution of isolated HDL, we used size-exclusion chromatography (SEC). Interestingly, we found that HDL isolated by IA contained a fraction larger than 12 nm, which was almost entirely absent with the other isolation methods. The distribution between the HDL 2/3 subclasses was similar between IA-HDL and UC-HDL, while the HDL 3 subclass was largely missing in DS-HDL (Fig. 2). To further investigate the different subclasses, we collected three major fraction (above 12 nm, between 12 -7.5 nm and below 7.5 nm) from the SEC for proteomics analysis (Fig. 2).
Mass spectrometry data indicated that the subclass larger than 12 nm, which was only present in IA-HDL, contained primarily apoA1 (24%), IGκC (13%), clusterin (10%), apoC1 (9%), and apoC2 (4%) together with a small amount of apoB (0.3%) ( Table 3). The fraction larger than 12 nm from IA-HDL contained a large amount of various complement C1q proteins. However, recent reports have shown that this might be an artefact, at least in part, of protein isolation using sepharose as a matrix (19). The size fraction above 12 nm was barely present in UC-HDL and DS-HDL, where it was mainly composed of apoA1 and apoA2 (Tables 4 and 5). The main HDL subclasses HDL 2/3 is located in the size range between 7.5 and 12 nm (Fig. 2).
In comparison to total HDL, the proportion of apoA1 increased from 53.4%-59.8% to 53.7%-64.5% and for apoA2 increased from 18.0%-25.2% to 21.3%-28.1% in purified HDL 2/3 . The data from HDL 2/3 suggests an increase in apoA2 and lower levels of apoA1 in UC-HDL (Table 5). Such an increase could be due to either a loss of apoA-I and thus an accumulation of apoA2, which is known to be less exchangeable than apoA1 (20) or from differences in the HDL subclass distribution across the isolation methods, since apoA2 is known to be preferentially associated with HDL 3 . The latter is certainly the case for DS-HDL, which has a much lower content of the smaller HDL 3 subclass (Fig. 2). Looking at the apoA2/apoA1 ratio between the isolation methods, we found that the ratio was higher in UC-HDL compared to IA-HDL and DS-HDL (supplemental Table S3). Such an increase could be due to either a loss of apoA-I and thus an accumulation of apoA2, which is known to be less exchangeable than apoA1 (20) or from differences in the HDL subclass distribution across the isolation methods, since apoA2 is known to be preferentially associated with HDL 3 . The latter is certainly the case for DS-HDL, which has a much lower content of the smaller HDL 3 subclass (Fig. 2). Notable differences between the isolation methods for HDL 2/3 were a lower content of apoC's in IA-HDL, while clusterin and PLTP were higher (Table 3). UC-HDL contained high levels of apoC3 (Table 4), while DS-HDL contained more apoA4 and low levels of apoD and paraoxonase when compared to the other methods (Table 5).

Impact of different isolation methods on the functionality of HDL
Next, we investigated whether the compositional differences observed in isolated HDL are linked to altered functionality. Cholesterol efflux was measured with the widely used system using cyclic adenosine monophosphate-treated J774 macrophages (25). Within AMBP, adipocyte plasma membrane-associated protein; DS, dextran-sulfate precipitation; HGR, haptoglobin-related protein; IGHG1, immunoglobulin gamma-1 chain C region; IGκC, immunoglobulin kappa chain C region; ITIH4, Inter-alpha-trypsin inhibitor heavy chain H4; PLTP, Phospholipid transfer protein; PYCOX1, Prenylcysteine oxidase 1.
this system cholesterol efflux is primarily mediated by apoA1 through ABCA1 (∼40%) and aqueous diffusion (∼50%) and to lower extant by SR-BI (∼10%) (25,26). Importantly, we found that IA-HDL was significantly less effective at promoting cholesterol efflux from macrophages (Fig. 4A). Our data did not show a significant difference in apoA1 that would explain the reduction in cholesterol efflux from IA-HDL (Table 2). However, taking the SEC data showing the distribution of HDL subspecies into account (Fig. 2), we suggest that the reduced content of HDL 2/3 might be involved in the reduction of cholesterol efflux capability. Interestingly, DS-HDL was most potent in promoting cholesterol efflux. Our data suggest an increased content of apoA1 together with an increased content of larger HDL 2 particles as the prime causes. Paraoxonase 1 is an atheroprotective enzyme that is mainly bound to HDL in the circulation (27)(28)(29). We found that paraoxonase 1 activity was highest in UC-HDL, whereas activity was greatly reduced in DS-HDL and almost completely lost in IA-HDL (Fig. 4B). Importantly, the activity measurements were in striking contrast to the proteomics data, which indicated the highest level of paraoxonase 1 in IA-HDL (Table 2). We suspected that the elution conditions during IA purification may have caused the loss of function. To mimic the conditions during the elution process, we incubated UC-HDL with 3M thiocyanate and then removed it by gel filtration. This treatment resulted in an almost complete loss of paraoxonase-1 activity of UC-HDL (Fig. 4D). The reduced activity in DS-HDL was consistent with a lower paraoxonase-1 mass content (Table 2). We further tested whether thiocyanate also affected cholesterol efflux activity by treating UC-HDL with 3M thiocyanate. However, treatment of HDL with 3M-thiocyanate had no effect on cholesterol efflux capacity (Fig. 4C). We also analyzed the ability of HDL to promote endothelial integrity using an electrical impedance sensing system. We found that all three HDL isolates were able to improve endothelial barrier function by ∼10%. However, no significant differences were found between the three isolation methods, although UC-HDL tended to have a lower capacity (Fig. 3D).

DISCUSSION
While a variety of HDL isolation methods exist, their impact on the HDL proteome and its associated function remain largely unknown. In this study, we systematically compared the composition and function of HDL particles isolated by common methods such as IA, UC, and DS. Especially IA chromatography with apoA1specific antibodies has emerged as an alternative method for isolating HDL because UC alters the composition of lipoproteins (30,31). In contrast to UC and precipitation methods, IA purification isolates HDL across the full size spectrum of all apoA1-containing particles.
In the present study, we observed that IA-HDL contained subtypes that were completely absent in UC-HDL and DS-HDL and that differed in composition from the major forms of HDL. This is consistent with previous reports showing that apoA1 can form HDL subclasses with higher size ranges (32,33). Further experiments will be necessary to distinguish whether these subclasses An interesting observation of the present study was that IA-HDL was less potent in promoting cholesterol efflux and almost completely lost its paraoxonase activity. Remarkably, the prime mediator of cholesterol efflux, apoA1, was more abundant in IA-HDL than in UC-HDL. Therefore, other factors like changes in the subclass distribution and in the lipidation status might be responsible for the impaired cholesterol efflux capacity. For ABCA1-mediated efflux, the size and lipidation status of the cholesterol acceptor is of critical importance, and small and lipid poor forms of HDL are more effective acceptors (34). We did not observe enrichment of small lipid-poor forms of HDL, such as pre-β HDL, in IA-HDL, so other factors seem more important. However, the differences in HDL subclass distribution were evident from SEC analysis. IA-HDL contained a fraction above 12 nm that made up about 20% of the overall protein mass. If this subclass were less efficient in promoting cholesterol efflux, an overall decrease would thus be plausible.
A further important observation of our study was that DS precipitation yielded HDL distinct of UC-HDL and IA-HDL. This is line with a recent report demonstrating that precipitations reagents can have a significant impact on the compositions and size of isolated HDL (35). DS does not to affect the size distribution of HDL but alters the quantity of a subset of apolipoproteins (35). However, DS seems to be less damaging to HDL than the widely used polyethylene glycol treatment of serum, which leads to significant changes in the size and in the apolipoprotein distribution of HDL (25).
SEC has been used as an alternative approach for the isolation of HDL (36,37). While SEC is a method that allows purification of proteins in their native form, many plasma proteins have molecular weights in the same range as HDL, e.g., albumin dimer (∼135 kDa), IgG (150-180 kDa), and complement C3 (180 kDa) (13). In addition, many protein complexes overlap with the size distribution of HDL. Therefore, SEC alone is only able to enrich serum for HDL. To overcome this restrictions, SEC has to be combined with other methods to isolate Fig. 3. Detection of HDL-associated proteins after native-gel electrophoresis. Isolated HDL from immunoaffinity (IA), ultracentrifugation (UC), and dextran-sulfate (DS) precipitation isolation were loaded on 4%-16% native gels and separated by native gel electrophoresis. Gels were blotted and probed with specific antibodies: apoA1 (A1); apoA2 (A2); apoC1 (C1); apoC2 (C2); apoC3 (C3); apoE (E); apoL1 (L1); apoA4 (A4); clusterin (CLUS); paraoxonase 1 (PON1), α-1-antitrypsin (A1AT); retinol-binding protein 4 (RBP4); haptoglobin-related protein (HGR); complement C3 (CO3); transthyretin (TT); human serum albumin (HSA); serum amyloid A (SAA); plasminogen (PLAS). Dashed lines represent the usual size range of the main HDL 2/3 subclass. Fig. 4. Functional characterization of HDL isolated by immunoaffinity (IA), ultracentrifugation (UC), or dextran-sulfate precipitation (DS). A: cholesterol efflux was assessed by incubating cAMP-stimulated J774.2 macrophages with 50 μg protein/ml HDL for 3 h. Cholesterol efflux is expressed as the radioactivity in the medium relative to total radioactivity in medium and cells. B: Arylesterase activity of HDL-associated paraoxonase was measured by using phenylacetate as substrate. C, D: UC-HDL was preincubated with 3M thiocyanate for 30 min. Subsequently, cholesterol efflux and paraoxonase activity were measured as indicated above. E: Endothelial barrier promoting activity of HDL was measured using an electrical impedance sensing system (ECIS). Ea.Hy926 cells were grown to a confluent monolayer and incubated with 100 μg/ml HDL and the impedance of the endothelial monolayer monitored over time. After 10 h impedance values were taken for quantification. All experiments were performed at least three times. HDL with high purity, for example with a lipid binding resin (38) or UC (39)(40)(41). Adding SEC after UC has the advantage that proteins that overlap with HDL in density, but are not HDL associated, are removed by size separation. The results of previous studies by others and us suggested that many of the detected proteins within UC-HDL isolates are not present in the HDL 2/3 size range (39,40). These proteins are found exclusively in the smaller pre-β sized fractions (39,40) and crosslinking experiments revealed that no protein-proteins interaction with major HDL proteins are observed (39). When using a lipid binding resin to purify lipoproteins after SEC, it must be considered that other non-HDL plasma proteins with lipid-binding affinity are co-isolated. Moreover, re-solubilization of bound proteins requires enzymatic digestion. Therefore, the isolated lipoproteins cannot be used to study functional properties.
In conclusion, our data are the first to provide an in depth assessment of proteomic features of HDL isolated by UC, IA purification, and DS. We have demonstrated that the use of different isolation methods resulted in the isolation of HDL that was compositionally and functionally distinct. This is of particular importance as especially the use of IA purification gains widespread use for HDL isolation from clinical cohorts (42)(43)(44)(45). Special attention must be taken when IA isolated HDL will be used for functional assays as the elution conditions with 3M thiocyanate can significantly alter its functional properties. Our data showing that separation and purification of HDL subclasses have a profound effect on HDL structure and function may help in the selection of the most appropriate isolation method for experimental purposes.

Data Availability
The data supporting this study are available in the article, the supplemental data, or available from the corresponding author upon reasonable request.