Mapping Atheroprotective Functions and Related Proteins/Lipoproteins in Size Fractionated Human Plasma

Abstract

HDL has been shown to possess a variety of cardio-protective functions, including removal of excess cholesterol from the periphery, and inhibition of lipoprotein oxidation. It has been proposed that various HDL subparticles exist, each with distinct protein and lipid compositions, which may be responsible for HDL's many functions. We hypothesized that HDL functions will co-migrate with the operational lipoprotein subspecies when separated by gel filtration chromatography. Plasma from 10 healthy male donors was fractionated and the protein composition of the phospholipid containing fractions was analyzed by mass spectrometry (MS). Each fraction was evaluated for its proteomic content as well as its ability to promote cholesterol efflux and protect low density lipoprotein (LDL) from free radical oxidation. For each function, several peaks of activity were identified across the plasma size gradient. Neither cholesterol efflux or LDL antioxidation activity correlated strongly with any single protein across the fractions. However, we identified multiple proteins that had strong correlations (r values >0.7, p < 0.01) with individual peaks of activity. These proteins fell into diverse functional categories, including those traditionally associated with lipid metabolism, as well as alternative complement cascade, innate immunity and clotting cascades and immunoglobulins. Additionally, the phospholipid and cholesterol concentration of the fractions correlated strongly with cholesterol efflux (r = 0.95 and 0.82 respectively), whereas the total protein content of the fractions correlated best with antioxidant activity across all fractions (r = 0.746). Furthermore, two previously postulated subspecies (apoA-I, apoA-II and apoC-1; as well as apoA-I, apoC-I and apoJ) were found to have strong correlations with both cholesterol efflux and antioxidation activity. Up till now, very little has been known about how lipoprotein composition mediates functions like cholesterol efflux and antioxidation.

The risk of cardiovascular disease has been shown to be inversely related to high density lipoprotein cholesterol (HDL-C) ¹ levels in large human cohorts (1, 2). Although the exact mechanism that underlies this relationship has not been identified, numerous functions that are seen as atheroprotective have been attributed to HDL. For example, studies have shown that injecting ³H-cholesterol-labeled macrophages into mice that overexpress ApoA-I, the most abundant protein on HDL, results in a significant increase of ³H-cholesterol detected in the HDL and feces (3). This data supports the widely accepted 'reverse cholesterol transport' (RCT) hypothesis (4) which invokes HDL as the primary vehicle for movement of excess cholesterol out of the periphery, in which cells generally lack the ability to catabolize cholesterol, and back to the liver for excretion through the feces. Aside from RCT, HDL has been shown to have other potentially anti-atherosclerotic effects. It has well documented antioxidative properties and has been shown to prevent oxidative modification of low density lipoprotein (LDL) thus reducing macrophage foam cell generation in the vessel wall (5). It can also inhibit the expression of cell adhesion molecules on endothelial cells to prevent inappropriate capture of circulating monocytes (6–9), and reduce the activity of macrophage chemotactic factor 1 which signals the infiltration of surface-adhered monocytes into the vessel wall (10, 11). These varied functions may all contribute to HDL's well documented association with atheroprotection.

HDL's ability to carry out multiple atheroprotective functions may be explained by its compositional heterogeneity. Recent proteomic studies show HDL is highly compositionally heterogeneous, composed of particles, most all of which contain apoA-I, with related physico-chemical properties, but differ widely in their compositions and functions (12–17). Additionally, correlational network analysis identified distinct protein clusters, which may represent specific HDL subspecies (15). Many of the anti-atherosclerotic functions of HDL have been attributed to the class as a result of in vitro experimentation with ultracentrifugally isolated samples. Previous work has shown that in ultracentrifugally isolated HDL subclasses, dense HDL₃ was the most effective at inhibition of LDL oxidation, and that several proteins were highly correlated with this activity (15, 18). However, it is becoming clear that ultracentrifugation may selectively isolate certain subspecies and alter the proteomic content of others (19, 20). Indeed, studies using gel filtration separation of human plasma have suggested the presence of many additional proteins that are either not found in UC-isolated HDL or are found in much lower abundance (12).

We have hypothesized that HDL's role in CVD may be modulated by distinct HDL subspecies. To test this, we studied two of the most widely recognized functional roles of HDL, cholesterol efflux and ability to prevent LDL oxidation. Instead of using ultracentrifugation, we separated plasma from 10 healthy, normolipidemic males by gel filtration and compared the fractions in the functional assays at equal volumes. This allowed us to understand the contribution of different HDL size species in relation to other plasma components including LDL. We also tracked the proteome across all fractions in order to derive relationships between the protein contents of the lipoprotein species and their functions.

MATERIALS AND METHODS

Study Population and Plasma Collection

Ten healthy, normolipidemic males (fasting cholesterol ≤ 195 mg/dl; triglycerides ≤ 150 mg/dl) between the ages 18 and 40 (mean age 26.2 years) were recruited. Inclusion criteria included non-smokers, body mass ≤ 24.9, no history of taking lipid lowering medications, no diabetes, no history of heart disease and C-reactive protein less than 1.0 mg/dl. Fasted blood was collected for a lipid panel. A second vial of blood was collected for proteomic and functional analysis using citrate as an anticoagulant, and spun at ∼1590 × g for 15 min at room temperature to isolate plasma. Plasma was stored at 4 °C until subjected to gel filtration chromatography, within 2 h. Participants provided informed consent according to an approved protocol as overseen by the institutional review board at Cincinnati Children's Hospital Medical Center.

Plasma Fractionation

Plasma (354 μl) was fractionated on three Superdex 200 columns in series (10/300 GL; GE Healthcare Lifesciences, Pittsburgh, PA) as previously described (12). Fractions (1.5 ml) were collected and stored at 4 °C for up to 1 week for further analysis. Cholesterol and phosphatidylcholine containing phospholipid (PL) content of the fractions were analyzed within 24 h using enzymatic kits (Pointe Scientific, Canton, MI and Wako, Richmond, VA). Protein content was measured using the Markwell modified Lowry assay (21).

Mass Spectrometry Sample Preparation and Analysis

Plasma fractions (300 μl) were prepared for mass spectrometry (MS) as previously described (22). Briefly, samples were delipidated using chloroform: methanol, and then treated with dithiothreitol and iodoacetamide (Sigma Aldrich, St. Louis, MO) to reduce and carbamidomethylate the proteins prior to trypsinization. Trypsinized proteins were subjected to MS using a nanoLC-MS/MS (AB Sciex5600 +TripleTOF) mass spectrometer as previously described (22) with the exception that the trap column used was a Chrom XP C18-CL NanoLC Trap Column (350 μm × 0.5 mm with a 3 μm packed bed) from Eksigent (Dublin, CA). Peakview version 2.1 was used to convert the raw data files into the peaklist (.mgf) files. The resulting mass spectra were analyzed with Mascot (version 2.2.2, www.matrixscience.com) and X! Tandem (version 2001.01.01.1) search engines against the UniProtKB/Swiss-Prot Protein Knowledgebase (2011, containing 540,958 entries). Search criteria included: human taxonomy, and carbamidomethylation as a variable modification; peptide tolerance was set at ±20 ppm, MS/MS tolerance set to ± 0.6 Da, and up to 3 missed trypsin cleavages allowed. Peptide and protein identification from the MS/MS were validated using Scaffold software (version 3.3.1) and only peptides with >95% identification probability and proteins with >99% identification probability were included in the analysis. Additionally, at least 2 peptides from each protein were required to be considered in the analysis. False discovery rates were less than 0.6% for peptide identification (calculated as the percentage of the sum of exclusive spectrum counts of decoy proteins divided by the sum of exclusive spectrum counts of target proteins) and less than 0.1% for protein identification (calculated as number of decoy proteins divided by the number of target proteins). Using this method, we did not make any quantitative comparisons between different proteins in each fraction. However, because equal volumes of each fraction were used for analysis, the resulting spectral counts should be proportional to the relative abundance of a given protein across fractions. Previous studies have confirmed that the spectral counts across fractions correspond to their relative abundance across fractions based on immunological analyses (13, 15). When noted, the protein list was selected for proteins that had previously been identified as being either LDL- or HDL- associated, as found on the LDL and HDL watch list (http://homepages.uc.edu/~davidswm/HDLproteome.html http://homepages.uc.edu/~davidswm/LDLproteome.html). The complete list of proteins detected by MS can be found in Supplemental Figs. (supplemental Fig. S1). Originally, 206 proteins were identified, of which 78 proteins were considered LDL- and HDL- associated proteins. Fifty-seven of the proteins excluded belonged to the immunoglobulin family and 11 proteins were keratins. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (23) partner repository with the dataset identifier PXD005520 and 10.6019/PXD005520.

Cholesterol Efflux Assay

RAW 264.7 macrophage cells (ATCC, Manassas, VA) (4 × 10⁵ cells) were grown in 48-well plates. Cells were exchange labeled for 16 h with 0.5 μCi ³H-cholesterol (Perkin Elmer, Waltham, MA) per well, in the presence of 0.3 mm 8-Bromoadenosine 3′,5′-cyclic monophosphate sodium salt (8-Br-cAMP) (Sigma Aldrich). The next day, unincorporated ³H-cholesterol was washed off, and cells were treated for 6 h with or without cholesterol acceptors or fractions, in the presence of 8-Br-cAMP. Cholesterol acceptors used for controls included lipid-free apoA-I (10 μg/ml), or ultracentrifugally isolated HDL (10 μg/ml protein). To evaluate the efflux capacity of plasma fractions, 20 μl of each fraction was added to the cell medium. ³H-cholesterol detected in the medium was measured using liquid scintillation counting after filtering the medium to remove any floating cells. Total cellular efflux was determined by dividing the ³H-cholesterol in the medium, by the total ³H-cholesterol content extracted from a separate set of control wells that were measured at the end of the overnight labeling period. Background efflux was subtracted from each fraction. To compare between experiments, efflux was normalized to a standard UC-HDL pool that was included in each experiment. All fractions were evaluated in triplicate. Higher efflux indicates better HDL function. PeakFit software (Systat Software, San Jose, CA) was used to identify component peaks from the resulting cholesterol efflux curves using the residuals best fit method. The resulting component peaks correlated well with the original data and had an r² value of 0.98.

LDL Oxidation Assay

LDL oxidation was measured using the propagation rate (PR) of oxidation in a 96-well UV plate. LDL (20 μg cholesterol) was added to each well. To measure the antioxidative capacity of the plasma fractions, 50 μl of each fraction was pre-incubated with the LDL at 37 °C for 10 min. Oxidation products were measured by the production of conjugated dienes, initiated by the addition of 2,2,- Azobis (2-methylpropionamidine) dihydrochloride (AAPH, Acros Organics, Geel, Belgium) (5 mm final concentration). Accumulation of conjugated dienes was measured at A₂₃₄ for 8 h at 37 °C. The maximal slope of each curve represents the propagation rate (PR). Data are expressed as percent inhibition of LDL oxidation, as represented by PR, relative to LDL oxidation in the absence of any additions. Higher inhibition of LDL oxidation indicates “better” HDL function. Fractions from each volunteer were analyzed in duplicate. PeakFit software was used to identify individual component peaks from the resulting oxidation curves using the residuals best fit method. The resulting component peaks correlated well with the original data and had an r² value of 0.98.

Correlations of Plasma Fraction Proteomic Data with Functional Data

Pearson correlation coefficients (PCC), as mathematically shown below, were calculated to assess the relationship between protein total spectral counts and functional profiles across fractions of individual peaks. Assume that X_ij^k is the spectral count of protein Px in the plasma fraction j of peak k from subject i. Given one functional assay F, let Y_ij^k denote functional activity measurement in fraction j of peak k from the same subject i. Then, for each protein and function pair (Px, F), the PCC of peak k was calculated by:

$$R^k\left(P_x,F\right)=\frac{{\displaystyle {\sum }_i{\displaystyle {\sum }_j\left(X_{ij}^k-\overline{X}\right)\left(Y_{ij}^k-\overline{Y}\right)}}}{\sqrt{{\displaystyle {\sum }_i{\displaystyle {\sum }_j{\left(X_{ij}^k-\overline{X}\right)}^2}}\sqrt{{\displaystyle {\sum }_i{\displaystyle {\sum }_j{\left(Y_{ij}^k-\overline{Y}\right)}^2}}}}}$$

where X̄ and Ȳ are the mean of spectral count X_ij^k and functional activity Y_ij^k. Thus, PCC, as represented by R^k, simultaneously reflects both intra- and inter-subject correlation of plasma fractions proteomic data with functional data. The p value was calculated using the t test to test the null hypothesis that there is not a linear correlation between the abundance of proteins and functional activities against the alternative that there is a nonzero linear correlation. In this study, we defined R^k as the local correlation of individual peaks.

Correlations of HDL Subspecies with Functional Data

Linear correlation of HDL subspecies abundance and functional activity was also assessed using PCC. Putative HDL subspecies that are supported by multiple pieces of evidence were selected from our prior work (14). These putative subspecies were identified using a network based approach searching for common migration patterns of proteins separated using three orthogonal separation techniques. Network mining was performed using a maximal clique algorithm so that a total of 30 HDL subspecies were identified.

Existence of these subspecies was tested within individual plasma fractions by examining the non-zero MS spectral counts of all protein members. If all protein members of an HDL subspecies co-existed within that fraction, we calculated the mean of the normalized spectral counts of all members except apoA-I to estimate the elution patterns of the HDL subspecies. Using the normalized data effectively removed the bias of the more abundant proteins on the overall distribution pattern of the putative subspecies, and instead shifted the emphasis to the correlation of the complex with activity. ApoA-I was left out of the mean calculation because it is likely a member of most HDL subspecies, and the relative contribution of apoA-I to any given subspecies cannot be determined. Finally, PCC was obtained to represent the relationship of HDL subspecies with functional activities.

Experimental Design and Statistical Rationale

Proteomic content was analyzed on 18 fractions from 10 volunteers (total 180 samples). Triplicate measurements of cholesterol efflux capacity were obtained for each plasma fraction from all volunteers. Duplicate measurements of antioxidative capacity were made on 18 fractions from all 10 volunteers. All data are expressed as mean ± standard deviation unless otherwise specified. When necessary, Bonferroni multiple corrections were applied to account for multiple comparisons.

RESULTS

Study Populations

Table I defines the characteristics of the study population. The mean age of the subjects was 26.2 y. In addition to having normal values for total cholesterol, HDL-C, LDL-C, triglycerides, and glucose, their C-reactive protein levels were below 1.0 mg/dL.

Table I. Characteristics of study population.

n	10
Age (years)	26.2 ± 6.3
BMI (kg/m2)	22.9 ± 1.5
TC (mg/dL)	160.2 ± 18.4
TG (mg/dL)	81.4 ± 24.9
LDL-C (mg/dL)	92.8 ± 17.7
HDL-C (mg/dL)	51.1 ± 10.5
Glucose (mg/dL)	91.5 ± 8.0
Systolic blood pressure (mm Hg)	120.8 ± 10.5
Diastolic blood pressure (mm Hg)	72 ± 7.5

Lipoprotein Distribution of Fractionated Plasma

Plasma samples from each individual were fractionated using three tandem Superdex 200 columns, as described in Materials and Methods. Previous work (12, 13) has shown that the lipoproteins, as identified by the combination of PL, cholesterol and protein, elute from the columns in fractions 13–30, which correspond to a size range of roughly 30 to 927 kDa. Fig. 1 shows the distribution profiles of cholesterol, PL and protein across this range. Two distinct cholesterol and PL peaks were observed. Proteins typically found in VLDL/LDL were found in the first peak (fractions 15–20) (supplemental Fig. S1). Proteins typically associated with HDL were seen in the second peak (fractions 20–28) (supplemental Fig. S1). Because these separations depend on molecular size and not density, it is not accurate to refer to these using the density based designations of VLDL, LDL or HDL. However, to be consistent with convention, we refer to these ranges of fractions as the VLDL/LDL and HDL size ranges. Albumin and other smaller free plasma proteins eluted in fractions 28–30 (12).

Fig. 1.

Characterization of fractionated human plasma. Human plasma (354 μl) was separated on 3 Superdex 200 columns in series. Fractions (1.5 ml) were collected and analyzed for PL (open circles), cholesterol (closed circles) and protein content (closed diamonds). Data represent the average data from 10 volunteers, except for the protein data, which show data averaged from 3 volunteers.

Distribution of Proteins Detected by Mass Spectrometry Across Triple Superdex Fractions

Individual plasma fractions were delipidated and prepared for LC-MS/MS analysis. Proteins were identified in Scaffold using Mascot and X! Tandem, as described in Materials and Methods. The abundance (as represented by total spectrum counts) of individual proteins across the fractions were summed across all subjects and normalized. For normalization of each protein, we subtracted the mean from the summed spectral counts of individual fractions, and then divided the individual values by its standard deviation. The resulting values represent the relative distribution of an individual protein across fractions. The resulting 206 proteins were then further selected using the LDL or HDL watch list (Materials and Methods) to identify proteins that were known to associate with LDL or HDL, resulting in a list of 78 proteins. We then applied a hierarchical clustering approach in GENE-E software (24) to group the proteins based on their distribution patterns across fractions. The heat map shown in Fig. 2 illustrates the distribution of each protein and the clustering result. The first cluster is comprised of apoB-containing lipoproteins, and were found in the larger PL containing fractions (fraction 15–17), also called large VLDL/LDL fractions. Subunits of fibrinogen that are commonly observed in LDL were distributed across fraction 17–20 (cluster 2), which are defined as small VLDL/LDL fractions. ApoA-I and apoA-II were found in all PL containing fractions, as previously reported (12). However, there was significantly more apoA-I and apoA-II in the second PL peak (fraction 20–28, cluster 4) compared with the first. ApoA-I and apoA-II have been recognized as a scaffold for other HDL-associated proteins (13, 25–28). The overlap in the clusters 3–5 indicate that some proteins in cluster 4 (e.g. apoA-I or apoA-II) may carry other proteins in each of cluster 3, 4 or 5, comprising separate subspecies. Thus, we defined fractions 20–24 as large HDL size fractions and fractions 25–28 as small HDL size fractions. The proteins in cluster 6 were mostly distributed in fractions 28–30, which we called the minimally lipidated/free proteins range. Each of these newly defined size-based fractions may contain one or more PL-associated subspecies. Because apoA-I and apoA-II were detected in every PL containing fraction (13–30), we analyzed the HDL functional activities in all fractions.

Fig. 2.

Proteomic heat map with hierarchical clustering of MS identified proteins across the PL containing plasma fractions. Individual proteins were identified by MS and abundance of proteins across fractions was then normalized as described in Materials and Methods. The list has been filtered to contain only those proteins found on the human HDL or LDL watch list.

Cholesterol Efflux Capacity of Plasma Fractions

The ability of each PL containing fraction (fractions 13–30) to promote cholesterol efflux was measured for all ten participants. The mouse macrophage cell line, RAW 264.7 cells, were exchange labeled with ³H-cholesterol in the presence of 8-Br-cAMP to upregulate ABCA1. Fig. 3A shows the cholesterol efflux capacity (mean and standard deviation) of each fraction. PeakFit software (29) was used to mathematically model individual peaks of efflux from the entire range of phospholipid containing fractions. This helped us to better identify the different components contributing to total cholesterol efflux capacity. Using the residuals best fit method, three peaks of activity were identified, as shown in Fig. 3B. Peak 1 corresponds to the LDL range, peak 2 to the HDL range, and peak 3 to the minimally lipidated/free protein range of fractions. The two PL-rich peaks clearly demonstrate a stronger cholesterol efflux capacity than the third peak in the minimally lipidated/free proteins range, again when compared on an equal volume basis.

Fig. 3.

Cholesterol efflux across all triple superdex fractions. A, Total cholesterol efflux capacity of each fraction, loaded at equal volume, is shown. Efflux measurements were performed on each fraction in technical triplicates. Data are background subtracted (fraction 13 was set as background) and normalized to a standard UC-HDL sample that was analyzed on each plate to compare across experiments. Data shown represent the average and standard deviation of 10 volunteers. B, PeakFit software v.4.12 was used to identify individual component peaks derived from the cholesterol efflux data using the residuals method. Individual peaks (peak 1, closed circles; peak 2, open circles; peak 3, closed triangles) identified using the residuals methods, as well as the sum of the peaks (open triangles) are shown.

Correlation of Phospholipid with Cholesterol Efflux of Plasma Fractions

Previous reports have suggested a strong link between PL content and cholesterol efflux. In order to determine whether or not PL was a strong predictor of the cholesterol efflux activity of the plasma fractions, we looked at the correlation of PL, total cholesterol and protein content of the fractions and their ability to stimulate cholesterol efflux. Fig. 4A shows a very strong linear correlation between the averaged PL content of each fraction from all 10 volunteers and the corresponding averaged efflux capacity of each fraction (r = 0.953, p < 0.001). Previous studies have shown that the PL content of HDL is directly related to its ability to efflux cholesterol, likely because the PL provides a large sink for cholesterol solubilization (30–35). Total cholesterol also showed a high linear correlation (r = 0.82, p < 0.001) with efflux capacity, although not as strong as the PL correlation (Fig. 4B). Interestingly, it appears that as the cholesterol content of the fractions increases, there appears to be a leveling off of the cholesterol efflux activity. Indeed, the data can also fit a 2-site saturation with nonspecific binding curve model with a higher correlation (r = 0.89, p < 0.001) (supplemental Fig. S2). However, cholesterol efflux is a complex, multicomponent process. It is thus, difficult to interpret the significance of this type of correlation. In contrast, the protein content in individual fractions does not exhibit any discernable correlation with efflux activity (Fig. 4C).

Fig. 4.

Correlation between cholesterol efflux and PL, cholesterol or protein levels of plasma fractions. Scatterplots showing the correlation of PL (A), cholesterol (B) and protein (C) concentrations to the efflux capacity of individual fractions. Data represent the average PL, cholesterol and protein content of each fraction plotted against the average normalized efflux of that same fraction. Pearson correlation coefficient for PL is 0.950 and the p < 0.0001; for cholesterol is 0.82, p < 0.0001 and for protein is 0.06 (n.s.).

Correlation of MS Identified Proteins to Cholesterol Efflux

Next, we sought to determine if any individual proteins, in contrast to total protein content of each fraction, are correlated with cholesterol efflux. Because we have shown that total phospholipid is the strongest predictor of efflux activity, we limited our search to proteins that have already been shown to be associated with LDL or HDL by mass spectrometer studies. Proteins in each fraction, identified by MS, were correlated with the individual cholesterol efflux peaks using PCC. Table II shows the top ranked proteins based on their correlation with cholesterol efflux activity of each peak. In peak 1 (LDL/VLDL range), the top efflux correlated protein is apoB, the major component of LDL. In peak 2, apoA-I and apoA-II are strongly correlated with efflux activity, as well as immunoglobulins and serum amyloid P. They are distributed across both large and small HDL fractions. This is consistent with data showing apoA-I and apoA-II's ability to promote cholesterol efflux (36–40). In peak 3, we identified several minimally lipidated/free proteins, including anti-thrombin 3, α-1-antitrypsin, transthyretin and albumin. Fig. 5 shows the distribution of the proteins most highly correlated with each peak's cholesterol efflux activity.

Table II. Correlation of cholesterol efflux activity with MS spectral counts.

Efflux Peak 1			Efflux Peak 2			Efflux Peak 3
	r	p		r	p		r	p
apolipoprotein B	0.944	4.29E-09	apolipoprotein A-II	0.993	2.36E-16	antithrombin-III	0.967	6.88E-11
IGHM	0.912	1.31E-07	Ig kappa chain C region	0.978	3.02E-12	α-1-antitrypsin	0.965	1E-10
C4b-binding protein α chain	0.749	0.0003	Ig gamma-1 chain C region	0.974	1.02E-11	transthyretin	0.958	4.1E-10
fibrinogen α chain	0.720	0.0008	apolipoprotein A-I	0.951	1.46E-09	albumin	0.935	1.35E-08
fibrinogen β chain	0.710	0.0010	serum amyloid P	0.944	4.16E-09	α-1-acid glycoprotein 1	0.849	8.39E-06
fibrinogen γ chain	0.709	0.0010	complement C3	0.894	5.83E-07	serotransferrin	0.836	1.55E-05
apolipoprotein(a)	0.564	0.0147	inter α trypsin inhibitor 4	0.838	1.41E-05	α-1-acid glycoprotein 2	0.830	2.04E-05
			apolipoprotein C-I	0.832	1.87E-05	apolipoprotein H	0.826	2.37E-05
			apolipoprotein J	0.764	0.0002	angiotensinogen	0.765	0.0002
			paraoxonase 1	0.760	0.0003	vitamin D binding protein	0.754	0.0003
			apolipoprotein D	0.658	0.0030	apolipoprotein A-IV	0.697	0.0013
			PtdIns-glycan-specific phospholipase D	0.646	0.0038	kallistatin	0.690	0.0015
			ceruloplasmin	0.614	0.0067	retinol binding protein 4	0.684	0.0017
			lecithin:cholesterol acyltransferase	0.564	0.0149	gelsolin	0.655	0.0032
			complement C2	0.560	0.0157	zinc-α-2-glycoprotein	0.650	0.0035
			lumican	0.530	0.0238	α-2-HS-glycoprotein	0.622	0.0059
			plasma kallikrein	0.505	0.0327	plasminogen	0.617	0.0064
						hemopexin	0.563	0.0149

Fig. 5.

Overlay of proteins that correlate most highly with individual peaks of cholesterol efflux activity. Distribution of MS spectral counts across fractions are shown overlaid with the cholesterol efflux activity for the proteins most highly correlated with efflux of the individual peaks for A, the first peak, [closed circles, apoB; open circles, IGHM; closed triangles, C4BPA; open triangles fibrinogen α chain; gray circle, cholesterol efflux] B, the second peak [closed circle, apoA-II; open circle, IGKC; closed triangle, IGHG1; open triangle, ApoA-I; and gray circle, cholesterol efflux] and C, the third peak [closed circle, Antithrombin 3; open circle, α-1-antitrypsin; closed triangle, transthyretin; open triangle, albumin and gray triangle, cholesterol efflux.

Antioxidative Capacity of Plasma Fractions

We also assessed the plasma fractions from our 10 participants for their antioxidative capacity. The capacity of each fraction to inhibit LDL lipid peroxidation was measured using AAPH to initiate LDL oxidation (Materials and Methods). Inhibition of oxidation is calculated as the percent decrease in PR relative to LDL only. When antioxidant activity is plotted across fractions, Fig. 6A shows that there are 3 peaks of activity where fractions inhibit LDL oxidation. Individual peaks of antioxidative activity were deconvoluted using PeakFit software. Fig. 6B shows the identified component curves, as well as the sum of the individual curves, as identified using the residuals best fit method. Peak 1 (fractions 16–21) is found in the LDL size range fractions, peak 2 (fractions 22–27) is found in the HDL range of fractions, whereas peak 3 (fractions 26–20) represents fractions containing minimally lipidated or free protein. Although the PeakFit software identified three component peaks of oxidation, similar to the efflux curves, we noted that these peaks are not identical to the component efflux peaks.

Fig. 6.

Inhibition of LDL oxidation by gel filtration fractionated plasma. A, Plasma fractions (50 μl) were incubated with LDL and AAPH to initiate lipid peroxidation. Percent inhibition of LDL oxidation was calculated using the propagation rate (PR) in the presence of each plasma fraction relative to PR of LDL only. B, Individual components of the total oxidation curve were identified using PeakFit. Three peaks were identified that likely contribute to the entire range of antioxidation activity (peak 1, closed circles; peak 2, open circles; peak 3, closed triangles, and the sum of all peaks, open triangle). The r² value for the peak fit was 0.98.

Protein Content Correlates with Antioxidant Capacity

Similar to cholesterol efflux activity, we sought to determine if the total PL, cholesterol or protein content of the plasma fractions are correlated to the corresponding antioxidative capability. Fig. 7 shows that, unlike cholesterol efflux which correlated with PL and cholesterol content, antioxidative activity of plasma was most closely correlated with total plasma protein level (Fig. 7C; r = 0.746, p < 0.0005). This is consistent with previous work determined that total plasma proteins accounted for a major portion of the total plasma antioxidant capacity (41). Interestingly, the data in Fig. 7C can also fit a model of saturation with higher statistical accuracy (r = 0.87, p < 0.0001) (supplemental Fig. S3), however, the interpretation of this is hindered by the complexity of the system. Because the antioxidant activity of the fractions are likely due to multiple components in each fraction, we are unable to determine the meaning of this curve fit. On the other hand, neither PL nor CH demonstrated a clear correlation to antioxidative capability (Fig. 7A, 7B).

Fig. 7.

Correlation of antioxidative activity with total PL, cholesterol or protein content of fractions. Scatterplots showing a relatively strong correlation with total protein content of fractions (r = 0.746, p < 0.0005) (C), but the lack of correlation of PL (A) or CH content (B) of fractions with antioxidative capacity of plasma fractions, (PCC of r = −0.271 (PL) and −0.0243 (CH), and p = n.s.).

Correlation of MS Identified Proteins to Inhibition of LDL Oxidation

We assessed the correlation of individual proteins with antioxidant activity using the Pearson's correlation coefficient. As with cholesterol efflux, we selected proteins that are lipoprotein associated. No specific proteins correlated with the entire set of fractions and their ability to inhibit LDL oxidation. However, significant correlations were identified between individual proteins and individual peaks of activity (Table III). It is interesting to note that in the small VLDL/LDL range fractions, the most highly correlated proteins were three subunits of fibrinogen, followed by α-2 macroglubulin. In the HDL range, in addition to some immunoglobulins, the most highly correlated proteins included apoA-II, serum amyloid P, inter-α-trypsin inhibitor heavy chain H1, apoA-I and a few different complement components. In the minimally lipidated/free protein fractions, the antioxidation correlated proteins include α-1 antitrypsin, albumin and serotransferrin. These highly correlated proteins are overlaid with the three anti-oxidation peaks in Fig. 8A (peak 1), 8B (peak 2) and 8C (peak 3).

Table III. Correlation of antioxidant activity with MS spectral counts.

Oxidation peak 1			Oxidation peak 2			Oxidation peak 3
	r	p		r	p		r	p
fibrinogen β chain	0.986	8.32E-14	Ig gamma-1 chain C region	0.975	7.71E-12	α-1-antitrypsin	0.934	1.43E-08
fibrinogen α chain	0.976	5.26E-12	Ig kappa chain C region	0.973	1.37E-11	albumin	0.934	1.47E-08
fibrinogen γ chain	0.972	1.91E-11	apolipoprotein A-II	0.958	3.93E-10	serotransferrin	0.918	7.98E-08
α2-macroglobulin	0.905	2.57E-07	serum amyloid P	0.949	2.01E-09	angiotensinogen	0.890	7.4E-07
fibronectin	0.726	0.0006	inter α trypsin inhibitor 4	0.916	9.41E-08	antithrombin-III	0.889	7.98E-07
apolipoprotein B	0.605	0.0078	complement C3	0.876	1.91E-06	transthyretin	0.883	1.21E-06
IGHM	0.557	0.0164	apolipoprotein A-I	0.832	1.89E-05	apolipoprotein H	0.873	2.23E-06
			apolipoprotein C-I	0.804	5.86E-05	gelsolin	0.835	1.62E-05
			apolipoprotein J	0.701	0.0012	α-2-HS-glycoprotein	0.813	4.07E-05
			ceruloplasmin	0.604	0.0080	plasminogen	0.791	9.26E-05
			complement C2	0.560	0.0157	α-1-acid glycoprotein 1	0.766	0.0002
			paraoxonase 1	0.539	0.0210	α-1-acid glycoprotein 2	0.740	0.0004
			Ptdlns-glycan-specific phospholipase D	0.501	0.0342	hemopexin	0.734	0.0005
						α-1-antichymotrypsin	0.650	0.0035
						complement C9	0.599	0.0086
						α-1B-glycoprotein	0.576	0.0123
						vitamin D binding protein	0.567	0.0140
						apolipoprotein A-IV	0.564	0.0147
						α-2-antiplasmin	0.556	0.0167
						vitronectin	0.548	0.0184

Fig. 8.

Overlay of top identified proteins that correlate with inhibition of LDL oxidation. Distribution of MS spectral counts across fractions are shown overlaid with the antioxidant activity of the individual peaks for A, the first peak, [closed circle, fibrinogen α chain; open circle, fibrinogen β chain; closed triangle, fibrinogen γ chain; open triangle, α-2-macroglobulin; gray circle, anti-oxidation activity] B, the second peak [closed circle, IgHG1; open circle, IgKC; closed triangle, apoA-II; open triangle, serum amyloid P; gray circle, antioxidation activity] and C, the third peak of activity [closed circle, α-1-antitrypsin; open circle, albumin; closed triangle, serotransferrin; open triangle, angiotensinogen; gray circle, anti-oxidation activity].

Correlation of Previously Identified Putative Subspecies with HDL Functions

We further wanted to determine if there was any correlation of cholesterol efflux or antioxidant activity with lipoprotein subspecies that we had previously postulated using a systematic computational approach (14). As described in the Methods section, PCC coefficient was calculated to determine how well each putative subspecies correlated with either cholesterol efflux or antioxidant activity. Using the list of putative subspecies previously identified, we identified 2 subspecies that are highly correlated with cholesterol efflux activity and 3 subspecies that associate significantly with antioxidant activity. As shown in Fig. 9A, cholesterol efflux activity in peak 2 was highly correlated with complex 1: apoA-I, apoA-II, and apoC-I, with a PCC of 0.961 as well as complex 2: apoA-I, apoC-I, and apoJ, with a PCC of 0.897. Fig. 9B shows the subspecies highly associated with antioxidant activity. Two subspecies were identified that correlated highly with peak 2. ApoA-I, apoA-II and apoC-I comprised one correlated complex, whereas the other was apoA-I, apoC-I and apoJ. Both subspecies that correlated with peak 2 of antioxidant activity had a PCC value of greater than 0.8, indicating a strong correlation. Additionally, the fibrinogen complex (FibA, FibB, and FibG) is highly correlated with antioxidant activity in peak 1, with a PCC of 0.98 indicating a nearly perfect correlation with antioxidant activity. The correlation of the fibrinogen complex with antioxidant function is in agreement with Olinescu et al. and Abudu et al. (42, 43), who have also shown that fibrinogen has antioxidant activity.

Fig. 9.

Correlation of previously identified putative subspecies with cholesterol efflux and antioxidant activities in individual peaks. Pearson correlation coefficients were calculated for the identified subspecies as described in Methods. A, PCC of each subspecies compared with cholesterol efflux activity. Black bars, apoA-I, apoA-II, apoC-I; gray bars, apoA-I, apoC-I, apoJ. B, PCC of each subspecies compared with antioxidation activity. Black bars, fibrinogen α chain, fibrinogen β chain, and fibrinogen γ chain; light gray bars, apoA-I, apoA-II, and apoC-I; dark gray bars, apoA-I, apoC-I and apoJ.

DISCUSSION

HDL is a heterogeneous population of particles with distinct size, charge and compositional characteristics. Given the wide variety of functions associated with HDL proteins, it stands to reason that different subpopulations may exhibit different functionalities. Here, we investigated cholesterol efflux and protection of LDL from oxidation, arguably two of HDL's most attributed functions. Rather than study the HDL particles, purified by potentially perturbing ultracentrifugation (19), we investigated these functions in the context of size-fractionated human plasma. This allowed the analysis of relatively native lipoproteins and the correlation of these functional activities with the proteomic composition of the fractions. Based on our proteomic analysis of size based fractionation of human plasma, we have indeed shown that a variety of HDL subspecies exist, as defined by their heterogeneous protein compositions. Furthermore, we demonstrated that the two HDL atheroprotective functions investigated were differentially influenced by the various plasma components. For example, cholesterol efflux appears to be most highly associated with phospholipid content of the fraction, whereas inhibition of LDL oxidation was more closely associated with total protein content. Each atheroprotective function was correlated with different sets of proteins. Finally, we identified 2 separate putative subspecies that were highly correlated with both cholesterol efflux and inhibition of LDL oxidation (apoA-I, apoA-II, apoC-I; and apoA-I, apoC-I and apoJ), suggesting that together these proteins may have a role in HDL function.

Our analysis of the protein distribution of size separated plasma further delineates the heterogeneity of lipoproteins. In Fig. 2, we showed hierarchical clustering of proteins based on their distribution patterns. Similar co-migration patterns of two proteins indicate they may reside on the same particles. Thus, we defined five fraction ranges each containing proteins that migrated with similar patterns: large VLDL/LDL, small VLDL/LDL, large HDL, small HDL and minimally lipidated/free protein fractions. These hierarchical grouping of fractions imply at least five types of subspecies. In fact, it is likely that the actual number of PL-associated subspecies is far more than five, because the number of proteins residing on the same particles is likely small because of their biophysical limitations. In any given PL-containing fraction, we detected between 13 and 54 known HDL- or LDL-associated proteins. However, we do not expect that all proteins reside on a single particle because of size constraints. Therefore, we expect that there are multiple PL-containing subspecies in each fraction. As verification of our method, we found one known HDL subspecies, trypanosome lytic factor (TLF) (27, 44, 45), which contains three main proteins, apoL-I, haptoglobin-related protein (HPR), and apoA-I, in fractions 19–23 (supplemental Fig. S1), demonstrating that our approach to separation does not disrupt the lipoprotein particles. Thus, according to our hierarchical clustering analysis, ApoL-I and HPR are grouped in the cluster 3 across large HDL fractions, whereas apoA-I, likely working as a scaffold for proteins in cluster 3–5, is grouped with cluster 4, which contains proteins found in both large and small HDL fractions (cluster 3 and 5) (Fig. 2). Considering the number of proteins detected in each fraction and their possible combinations, dozens of subspecies may exist. In a recent study, a series of HDL subspecies candidates were inferred using a network-based computational method (14). Experimental validation is still necessary to further our understanding of these plasma subspecies.

ApoA-I and apoA-II (HDL's main structural proteins) were detected in all phospholipid containing fractions, even in the VLDL/LDL size range. Thus, all phospholipid containing fractions were examined for both cholesterol efflux and antioxidation activity. Correlational analysis examining the proteomic content of each fraction with the HDL functions revealed that similar, but different subsets of proteins were linked to each functional activity. For example, cholesterol efflux activity was detected in multiple peaks across the fractions, indicating that cholesterol efflux is related to more than one subspecies, and those multiple PL species may coordinate with each other to promote cholesterol efflux. In an attempt to identify subspecies that may be related to cholesterol efflux, we analyzed the top proteins whose distribution across each efflux peak were most highly correlated with cholesterol efflux activity. In terms of individual proteins, we found the top correlated proteins consist of diverse functional categories including lipoproteins traditionally involved with lipid metabolism, immunoglobulins, proteins associated with the alternative complement cascade and innate immunity as well as many proteins associated with the clotting cascade. The lipoproteins most commonly associated with cholesterol efflux were identified in peak 2 (HDL range) and include the most abundant and central structural apolipoproteins in HDL, apoA-I and apoA-II. Other lipoproteins with high correlations in peak 2 include apoC-1 and apoJ (clusterin). Each of these apolipoproteins have previously been shown to have the ability to efflux cholesterol from cells (46–49). Besides these apolipoproteins well known for their efflux capacity, multiple immunoglobulins and various components of the alternative complement pathway, including complement C3, C5, C6 and C7 were found to be highly correlated with efflux activity in peak 2, along with proteins of the innate immune response, serum amyloid P and inter-α-trypsin inhibitor H4. In the LDL size range fractions, it appears that an apoB containing particle is most highly correlated with efflux activity. These studies utilized radio-labeled free cholesterol to specifically measure cholesterol efflux in one direction, i.e. from the cell to the lipoproteins. It has been shown that the net mass transfer of cholesterol between LDL and cells results in a net influx of cholesterol mass to the cell (50). Nevertheless, our work clearly shows that cellular cholesterol can end up in LDL populations, suggesting that it participates in cellular cholesterol homeostasis in more ways than simple cholesterol loading. Other proteins found to be highly correlated with efflux activity in peaks 1 and 3 include multiple chains of fibrinogen in peak 1 and anti-thrombin 3, α-1 antitrypsin and albumin in peak 3. Many of these proteins have well established functions in regulating the clotting cascade (51). However, it is unclear whether they contribute directly to cholesterol efflux, or if they are cargo proteins on lipoprotein particles that have efflux capacity because of the presence of apoA-I, because it is well established that minimally lipidated apoA-I is the primary mediator of ABCA1 dependent cholesterol efflux (52, 53). Interestingly, we also found albumin to be highly correlated with efflux activity in peak 3. Consistent with our findings, multiple studies have previously shown that albumin has the capacity to accept cholesterol from cells (54, 55). Each of these highly correlated proteins were also detected in our previous studies (12, 22), which analyzed only phospholipid bound proteins, suggesting that our highly correlated proteins are lipid bound.

Inhibition of LDL oxidation is another potentially important atheroprotective function of HDL. Similar to cholesterol efflux activity, antioxidation function is clearly associated with multiple subspecies in multiple peaks of activity across PL containing fractions. Indeed, there have been many studies showing a variety of antioxidants present in plasma. These include both small molecules, such as ascorbate, urate and vitamin E, as well as proteins, such as paraoxonase, transferrin, and albumin (41, 56–64) Although our study does not address the small molecules association with our antioxidation activity, we did, however, find some interesting associations with specific proteins. Proteomic analysis showed that the proteins most highly correlated with antioxidative activity in peak 1 are fibrinogen α, β and γ chains, the three subunits of fibrinogen that come together to form the fibrinogen complex. Several previous studies have detected fibrinogen associated with lipoprotein particles (12, 15, 16, 65, 66), but it is unclear whether the association is specific, or is simply because of the abundance of the fibrinogen complex non-specifically sticking to the lipoprotein particles. The fibrinogen complex has a mass of ∼342 kDa, and its peak elution fraction is fraction 18 (Fig. 2). According to our calibration standards, proteins and protein complexes with a MW of 340–417 kDa will elute in fraction 18. Thus, using our sizing information, we are unable to determine whether this fibrinogen complex is PL-associated or not. Nevertheless, fibrinogen has previously been shown to have antioxidant activity (42, 43), which lends support to our findings that fibrinogen is highly correlated with antioxidant activity in peak 1. Furthermore, correlation analysis demonstrated that the fibrinogen complex (FibA, FibB and FibG) was also highly correlated to antioxidant activity in peak 1 (Fig. 9B, r = 0.98). In peak 2, apoA-I and apoA-II are both highly correlated with antioxidant function. This effect may be because of these proteins specifically, as these proteins have both been shown to have antioxidative properties (67–71). Additionally, apoA-I and apoA-II may be the structural proteins of particularly effective subspecies of HDL containing other antioxidant proteins. For instance, ceruloplasmin and paraoxonase, both proteins with documented antioxidant properties (60, 72, 73), were also found to be highly correlated with antioxidation activity in peak 2. Other proteins in peak 2 that were highly correlated with antioxidation include complement components, as well as other inflammatory and immune response proteins (immunoglobulins, serum amyloid P, inter-α-trypsin inhibitor), indicating the potential overlap and interaction of immune response and antioxidation processes. Antioxidative activity in peak 3 is highly correlated with albumin and serotransferrin, both proteins known to have antioxidant properties (62, 74, 75). Other interesting proteins found in peak 3 that had significant correlations with antioxidant activity include both hemopexin and apoA-IV, both of which have previously demonstrated antioxidant activity (76, 77).

In our previous study (14), network analysis identified 30 potential subspecies that were supported by multiple lines of evidence. Further analysis of the proteomic compositions of our plasma fractions yielded support for the existence of some previously identified potential HDL subspecies. For instance, apoA-I, apoC-I and apoJ (clusterin) were found together in peak 2, and were individually found to be significantly correlated with both cholesterol efflux and antioxidation activity (p < 0.05 after Bonferroni multiple correction). We noted that these three proteins appear to comprise an HDL subspecies as indicated in our previous study (14). The existence of this HDL subspecies is supported by multiple lines of evidence. First, multiple literature reports have documented an association between pairs of these proteins (41, 78–80). Second, recent data from our group has shown that when apoA-I is knocked out in a mouse model, apoC-I was also significantly decreased (28). Finally, using co-migrational analysis (14), we were able to detect a shift in migration of apoC-I as well as apoJ in a subject who was deficient in apoA-I, implying that in the absence of apoA-I, apoC-I and apoJ migrate with smaller particles than they would in the presence of apoA-I. These data, taken together, support the notion that apoA-I, apoC-I and apoJ may constitute a specific subspecies of HDL. Similarly, a second previously identified putative subspecies was found to be highly correlated with cholesterol efflux activity, as well as antioxidation activity in peak 2: apoA-I, apoA-II and apoC-I. Indeed, the apolipoproteins in each of these complexes have demonstrated efflux potential; it will be interesting to determine whether a lipoprotein complex or complexes, containing these efflux-capable proteins, will act synergistically to enhance cholesterol efflux compared with each protein on its own. Because each of these subspecies is highly correlated with both cholesterol efflux and antioxidant activity, and because many of these exchangeable lipoproteins are capable of performing similar functions, it is difficult to ascertain the importance of one subspecies compared with another in carrying out each specific function. The presence of additional proteins on these subspecies may account for differences in functional activities. Regardless, this data supports our hypothesis that distinct HDL subspecies may be responsible for various biological functions related to CVD.

Although our results suggest multiple proteins/subspecies are associated with cholesterol efflux and antioxidation functions, we caution that both efflux and antioxidation associated proteins were derived from numerical correlation analysis. It is difficult to know if the correlated proteins are causative or simply associations. However, correlations between the proteins and the two functions are very strong and these candidates are actively being studied through interventional experimental approaches in our laboratory. This work is among the first to go beyond individual proteins and link putative HDL subspecies to specific functions. Further work will be needed to confirm the existence of specific subspecies and their direct role in a specific HDL function. However, our data suggest that certain plasma proteins may serve as better biomarkers than HDL-C for CVD risk assessment, especially when it comes to precision medicine; specific functional evaluation (e.g. antioxidation and cholesterol efflux activities) may provide richer information than general disease risk assessment.