Proteomic analysis of HDL from inbred mouse strains implicates APOE associated with HDL in reduced cholesterol efflux capacity via the ABCA1 pathway[S]

Cholesterol efflux capacity associates strongly and negatively with the incidence and prevalence of human CVD. We investigated the relationships of HDL’s size and protein cargo with its cholesterol efflux capacity using APOB-depleted serum and HDLs isolated from five inbred mouse strains with different susceptibilities to atherosclerosis. Like humans, mouse HDL carried >70 proteins linked to lipid metabolism, the acute-phase response, proteinase inhibition, and the immune system. HDL’s content of specific proteins strongly correlated with its size and cholesterol efflux capacity, suggesting that its protein cargo regulates its function. Cholesterol efflux capacity with macrophages strongly and positively correlated with retinol binding protein 4 (RBP4) and PLTP, but not APOA1. In contrast, ABCA1-specific cholesterol efflux correlated strongly with HDL’s content of APOA1, APOC3, and APOD, but not RBP4 and PLTP. Unexpectedly, APOE had a strong negative correlation with ABCA1-specific cholesterol efflux capacity. Moreover, the ABCA1-specific cholesterol efflux capacity of HDL isolated from APOE-deficient mice was significantly greater than that of HDL from wild-type mice. Our observations demonstrate that the HDL-associated APOE regulates HDL’s ABCA1-specific cholesterol efflux capacity. These findings may be clinically relevant because HDL’s APOE content associates with CVD risk and ABCA1 deficiency promotes unregulated cholesterol accumulation in human macrophages.


Plasma lipid and APOA1 measurements
Triglycerides and phospholipids (Wako Diagnostics) and cholesterol levels (Invitrogen) were determined biochemically. APOA1 levels were determined by immunoblot analysis with a goat anti-mouse APOA1 antibody (US Biological).

Cholesterol effl ux assays
Macrophage cholesterol effl ux capacity was assessed with J774 macrophages labeled with [ 3 H]cholesterol and stimulated with a cAMP analog, as described by Rothblat and colleagues ( 10 ). Effl ux via the ABCA1 or ABCG1 pathways was measured with BHK cells expressing mifepristone-inducible human ABCA1 or ABCG1 that were radiolabeled with [ 3 H]cholesterol ( 20 ). Effl ux of [ 3 H] cholesterol was measured after a 4 h incubation in medium with APOB-depleted serum HDL (2.8% v/v) or isolated HDL (30 g protein per milliliter). ABCA1-specifi c cholesterol effl ux capacity was calculated as the percentage of total [ 3 H]cholesterol (medium plus cell) released into the medium of BHK cells stimulated with mifepristone after the value obtained with cells stimulated with medium alone was subtracted.

HDL isolation
Serum HDL was prepared by adding calcium (2 mM fi nal concentration) to plasma and using polyethylene glycol (8 kDa; Sigma) to precipitate lipoproteins containing APOB (VLDL, IDL, LDL). After centrifugation at 10,000 g for 30 min at 4°C, serum HDL was harvested from the supernatant. HDL was isolated from serum or EDTA-anticoagulated plasma using sequential ultracentrifugation (d = 1.063-1.21 mg/ml) ( 15,21 ). HDL was stored on ice in the dark and used within 1 week of preparation.
It has been diffi cult to assess the clinical relevance of cholesterol effl ux. However, recent studies of cAMPstimulated J774 macrophages radiolabeled with cholesterol, which measures cellular effl ux by multiple pathways (ABCA1, ABCG1, SR-B1, and diffusion) ( 10 ), demonstrate strong inverse correlations between cholesterol effl ux capacity of serum HDL (serum depleted of APOB-containing lipoproteins) and prevalent coronary artery disease ( 11 ). Moreover, macrophage effl ux capacity remained a strong predictor of prevalent coronary artery disease after adjustment for HDL-C levels ( 11 ).
The cholesterol effl ux capacity of serum HDL with cAMP-stimulated J774 macrophages can also be assessed with fl uorescently labeled cholesterol, which primarily measures ABCA1-mediated effl ux from these cells ( 12 ). A recent study of a population-based cohort that was free of CVD demonstrated that ABCA1-specifi c cholesterol effl ux capacity assessed by this method associates strongly and negatively with the risk of future cardiac events ( 13 ). This association persisted after multivariate adjustment, suggesting that altered HDL function affects cardiovascular risk by processes distinct from those involving HDL-C or traditional lipid risk factors. Impaired cholesterol effl ux capacity with J774 macrophages radiolabeled with cholesterol also predicted incident cardiac events in a healthy cohort ( 14 ). Taken together, these observations provide strong evidence that cholesterol effl ux capacity is a better measure of HDL's cardioprotective effects than HDL-C.
The molecular factors that control HDL's serum effl ux capacity are poorly understood ( 10,12 ). For example, HDL-C predicts only 34% of serum HDL's macrophage cholesterol effl ux capacity ( 11 ); HDL-C fails to predict HDL's ABCA1-specifi c cholesterol effl ux capacity with cells labeled with fl uorescent cholesterol ( 12,13 ). Macrophage cholesterol effl ux capacity of serum HDL also correlates poorly with plasma APOA1 levels ( 11 ), even though APOA1 is HDL's major structural protein, accounting for ‫ف‬ 70% of protein mass. In human HDL, APOA2 accounts for ‫ف‬ 20% of HDL's protein, while >80 proteins account for the other 10% of HDL's proteome, as assessed by MS ( 15,16 ). Moreover, various HDL subspecies have very different proteomes ( 17,18 ). It is therefore unclear which, if any, of the proteins contribute to HDL's proposed cardioprotective effects.
In the current studies, we used MS and HDL isolated by ultracentrifugation from fi ve inbred mouse strains that differed in HDL-C levels and atherosclerotic susceptibility ( 19 ) to investigate the relationships among HDL's size, protein composition, and cholesterol effl ux capacity, using both J774 macrophages and BHK cells expressing ABCA1. Our unbiased approaches identifi ed candidate proteins that correlate with HDL's size and with its cholesterol effl ux capacity with macrophages and the ABCA1 pathway. Importantly, we used mice defi cient in APOE and APOA2 to confi rm that these proteins infl uence HDL's effl ux capacity and size. counter. HDL peak areas were converted into aqueous particle concentrations using glucose oxidase calibration curves.
Because electrophoretic mobility depends chiefl y on size, IMA data are expressed in terms of particle diameter (nanometers), which corresponds to the calculated diameter of a singly charged spherical particle with the same electrophoretic mobility ( 29 ). The method yields a stoichiometry of APOA1 and the sizes and relative abundance of HDL subspecies in excellent agreement with those determined by nondenaturing gradient gel electrophoresis and analytical ultracentrifugation ( 29,30 ). Because calibrated IMA accurately quantifi es the size and concentration of gold nanoparticles and reconstituted HDL ( 29 ), we term the concentration and size of HDL determined by this method, HDL-P ima .

Statistical analyses
Data are represented as mean ± SEM. Differences among the HDLs of the fi ve groups were assessed with ANOVA followed by Fisher's exact test to correct for multiple comparisons. The results of ANOVA are presented as F(df 1 , df 2 ), where F is the test statistic of the F distribution, and df 1 and df 2 are the degrees of freedom of the analysis. The Kruskal-Wallis one-way ANOVA by ranks followed by Dunn's test (to correct for multiple comparisons) was used for proteomic analyses. Linear correlations were assessed with Pearson's product-moment coeffi cient. P < 0.05 was considered signifi cant. Data were analyzed with Prism and R software.

HDL's protein cargo is genetically controlled
Structural and quantitative variations in APOA1 and APOA2 of HDL isolated from inbred strains of mice have been previously reported ( 31,32 ). To test the hypothesis that the HDLs in different strains of mice have different protein cargos that affect size and function, we used shotgun proteomics to analyze HDL isolated from the plasma of fi ve inbred mouse strains that differed in atherosclerosis susceptibility (least to most susceptible: NZW/LacJ < SWR/J < DBA/2J < C57BL/6J < C57BLKS/J). The HDLs were isolated by ultracentrifugation and digested with trypsin, and the peptide digest was analyzed by LC-ESI-MS/MS. Judged by stringent statistical criteria (Methods), shotgun proteomics identifi ed 72 proteins with high confidence in one or more mouse strain (supplementary Table 1).
Hierarchical clustering, as assessed by Euclidean distance analysis, of the HDL proteome recapitulated the genealogy of the strains, as determined by high density single nucleotide polymorphism genotyping ( 33 ).Thus, C57BL/6J and C57BLKS/J are most closely related to each other and these are then most related to DBA/2J, SWR/J, and NZW/LacJ, in that order ( Fig. 1A ). These observations support the proposal that a wide range of proteins in HDL are under genetic control.
Sixty of the 72 proteins were detected in all fi ve strains, and 45 of the 60 were differentially expressed (supplementary Table 1, Kruskal-Wallis, P < 0.05; e.g., APOA2, APOE, APOC3, APOD, SAA1, SAA2, C3, PLTP). Adjusted spectral counts of representative proteins for fi ve strains are presented in Fig. 1B-G . DBA/2J had the most diverse proteome ( Fig. 1A ). For example, GPLD1, AQP4, SERPINA1b, and CTSD were expressed at high levels in HDL from the Inc.) at a fl ow rate of 1 l/min over 180 min, using a linear gradient of 5-35% buffer B (90% acetonitrile, 0.1% formic acid) in buffer A (5% acetonitrile, 0.1% formic acid). ESI was performed using a CaptiveSpray source (Michrom BioResources, Inc.) at 10 ml/min fl ow rate and 1.4 kV setting. HDL digests were introduced into the gas phase by ESI, positive ion mass spectra were acquired with a linear ion trap mass spectrometer (LTQ; Thermo Electron Corp.) using data-dependent acquisition (one MS survey scan followed by MS/MS scans of the eight most abundant ions in the survey scan) with a m/z 400-2,000 scan. An exclusion window of 45 s was used after two acquisitions of the same precursor ion ( 15,18 ).

Protein identifi cation
MS/MS spectra were matched against the mouse International Protein Index database (mouse v.3.54), using the SEQUEST (version 2.7) search engine with fi xed Cys carbamidomethylation and variable Met oxidation modifi cations. The mass tolerance for precursor ions was 2.5 ppm; SEQUEST default tolerance was 2.5 Da for precursor ion mass and 1 Da for fragment ion mass. SEQUEST results were further validated with PeptideProphet and ProteinProphet ( 22,23 ), using an adjusted probability of у 0.90 for peptides and у 0.95 for proteins. Each charge state of a peptide was considered a unique identifi cation.
We used the gene and protein names in the Entrez databases [National Center for Biotechnology Information; based on the nomenclature guidelines of the Human Gene Nomenclature Committee (http://www.gene.ucl.ca.uk/nomenclature) for human guidelines ( 24 ), and Mouse Genome Informatics (http:// www.infromatics.jax.org.nomen/) for mouse guidelines ( 25 )] to identify HDL proteins and to eliminate the redundant identifi cations of isoforms and protein fragments frequently found in databases used in proteomic analysis ( 26 ).This approach also permits cross-referencing of proteins from different species. When MS/ MS spectra could not differentiate between protein isoforms, the isoform with the most unique peptides was used for further analysis.

Protein quantifi cation
Proteins were quantifi ed using spectral counts, the total number of MS/MS spectra detected for a protein ( 15,27,28 ). Proteins considered for analysis had to be detected in three or more analyses with two or more unique peptides. When MS/MS spectra could not differentiate between protein isoforms, the isoform with the most unique peptides was used for further analysis. Spectral counts for each protein, normalized to total spectral counts for peptides from each sample, were used to calculate a spectral index to compare the relative protein composition of mouse strains' HDLs ( 15 ). We included in our analyses only proteins that were detected in 75% of samples in at least one mouse strain. Supplementary Table 1 provides the total calculated spectral counts for each protein, the individual peptides that identifi ed each protein, the total number peptide spectra counts, and relative quantifi cation.

HDL particle concentration and size
HDL particle concentration and size (HDL-P ima ) were quantifi ed by calibrated ion mobility analysis (IMA) ( 29 ). Briefl y, HDL isolated by ultracentrifugation from EDTA plasma was introduced into the gas-phase ions by ESI. The resulting highly charged ions were largely neutralized by ␣ particles, yielding a small proportion of singly charged cations, which were introduced into the mobility analyzer. As the particles moved through a strong electromagnetic fi eld, they were separated according to their electrophoretic mobility and then enumerated by a particle Fig. 1. The HDL proteome of inbred mouse strains. LC-ESI-MS/MS analysis of proteins in HDL isolated by ultracentrifugation from fi ve different strains of mice. Proteins were quantifi ed by spectral counting (total number of peptides identifi ed for a given protein normalized to total spectral counts). A: Heat map of differentially expressed proteins. Relative protein abundance was calculated as z-scores that were generated from adjusted spectral counts. Red, upregulated; green, downregulated. Mouse strains were clustered by Euclidian hierarchical cluster analysis. B-G: Quantifi cation of representative mouse HDL proteins. DBA/2J mice, but not in the other strains of mice, while three proteins that were expressed by all the other strains were undetectable in the HDL of DBA/2J mice [retinol binding protein 4 (RBP4), TCEAL5, IGKAPPA].
Cluster analysis suggested that certain proteins are coexpressed on HDL (supplementary Fig. 1), perhaps indicating coordinate regulation or functional complexes. For example, several closely related proteins clustered together. Examples include SAA1 and SAA2 acute-phase response proteins that share a high degree of sequence identity. In contrast, SAA4, which is constitutively expressed, did not cluster with the other SAA isoforms. Histocompatibility 2 Q region locus 10 (H2-Q10), B2-M, HBA-A2, and HBB-B1 also clustered together. These proteins are not members of a single gene family, but each is implicated in immune regulation, raising the possibility that their expression levels in HDL might be coordinately regulated as part of the immune response. orthologs of certain human HDL proteins linked to the acute response, such as HPX, ORM2, and TTR, in mouse HDL.
More than 80 proteins have been identifi ed with high confi dence in human HDL by at least two independent laboratories ( 15,16,36 ), However, only 42 of the 72 proteins we detected in mouse HDL have been identifi ed in human HDL. Many of those mouse proteins lack human orthologs. Examples include H2-Q10, APON, H2-L, SER-PINA-3K, PSAP, AHSG, and NAPSA. However, Gene Ontology analysis ( Fig. 2 ) revealed that both human ( 15 ) and mouse HDLs contain the same major functional categories of proteins. These data demonstrate that the HDL proteomes of mice and humans share similar functions and have many of the same proteins, but are also distinct.

The size of HDL varies signifi cantly in different strains of mice
HDL particles in the different strains of mice differed signifi cantly in size [ Fig. 3A ; ANOVA, F(4, 31) = 6.1, P < 0.001, and F(4, 31) = 6.2, P < 0.001 for size and concentration, respectively]. HDL particles isolated from the NZW/ LacJ mice had the largest diameter (10.14 ± 0.13 nm), while HDL from the C57BLKS/J mice had the smallest (9.9 ± 0.05 nm, P < 0.0001). Based on calibration with proteins of known molecular mass ( 29 ), we estimate the difference in molecular mass between the largest and smallest particles to be ‫ف‬ 10 kDa. It is likely that alterations in HDL particle size refl ect changes in particle lipid and/or protein composition. The fi ve strains of mice had signifi cantly different levels of HDL-C, APOA1, and phospholipids in serum HDL (APOB depleted serum; supplementary Fig.  2B-E).
To determine whether any strain-specifi c compositional differences were present, we measured the HDL phospholipid, free cholesterol, cholesteryl ester, and protein content of particles isolated by ultracentrifugation. We normalized the values to HDL particle concentration (HDL-P ima ) to estimate composition of HDL particles for each strain (supplementary Fig. 2F). We observed less than 5% differences among proteins, cholesteryl esters, and phospholipids. Free cholesterol varied ‫ف‬ 40% among strains (6.4% vs. 10.1% between DBA2/J and NZW/LacJ, respectively). The lack of major differences in HDL's core lipids and phospholipids between strains suggests that size differences are due to HDL's protein composition.

Candidate HDL proteins that affect HDL size
We next determined whether differential expression of HDL proteins in the fi ve strains of mice correlated with HDL particle size. All fi ve strains of mice had both small ( ‫ف‬ 10.1 nm) and large ( ‫ف‬ 12.6 nm) HDLs; the small HDL subspecies was ‫ف‬ 10-fold more abundant than the large subspecies. Particle size correlated strongly with HDL's content of APOA2 and H2-Q10 ( Fig. 4A, B ; r = 0.67 and 0.64, both P < 0.0001), but not with that of APOA1 or HDL-C (data not shown). APOA2 is HDL's second most abundant protein, and H2-Q10 is a component of the murine histocompatibility complex that participates in antigen processing.
Mouse HDL proteins are involved in lipid metabolism, immunity, and inhibition of proteolysis Gene Ontology analysis ( Fig. 2 ) demonstrated that mouse HDL is enriched in proteins associated with lipid metabolism (18 proteins, P = 10 Ϫ 15 ), the immune response (25 proteins, P = 10 Ϫ 6 ), the acute-phase response (3 proteins, P = 10 Ϫ 5 ), and antigen processing (5 proteins, P = 10 Ϫ 3 ). Four proteins associated with proteinase inhibition ( P = 10 Ϫ 7 ). Importantly, the existing annotation databases used for Gene Ontology analysis are incomplete, and only a subset of genes is functionally annotated. Indeed, we have used MS/MS analysis to demonstrate that over half the proteins detected in HDL are acute-phase response proteins in mice injected with silver nitrite ( 34 ), a widely used model of infl ammation. Collectively, these observations indicate that mouse HDL, like human HDL ( 15 ), carries proteins that participate in lipid metabolism, acute infl ammation, immunity, and inhibition of proteolysis.
Like human HDL ( 15,35 ), mouse HDL contains major histocompatibility complex proteins. However, mouse HDL appeared to be more enriched in proteins linked to the adaptive immune system, including ␤ -2-microglobulin (the ␤ -chain of the heterodimer that presents major histocompatibility complex class I molecules), H2-Q10, and H2-L, which the human genome lacks . In contrast, we did not detect transgenic mice (all in the C57BL/6J background). Then we determined the sizes and concentrations of the HDL particles by calibrated ion mobility ( Fig. 5A ).
The mean size of medium HDL from the wild-type mice was 9.8 nm ( Fig. 5B ). In contrast, it was 9.4 nm for the Apoa2 Ϫ / Ϫ mice and 10.1 nm for the transgenic mice expressing human APOA2. The mean sizes of the large HDLs exhibited the same trends, but the absolute differences were even more marked ( Fig. 5C ). These observations confi rm that APOA2 has a key infl uence on the size of HDL. Plasma levels of both large and medium HDL were markedly reduced in the Apoa2 Ϫ / Ϫ mice ( Fig. 5 ), as previously reported for HDL-C levels in these animals ( 37 ).

The cholesterol effl ux capacity of HDL varies signifi cantly in different strains of mice
We used cAMP-stimulated J774 macrophages to quantify macrophage cholesterol effl ux capacity of serum HDLs of the fi ve strains of mice because this assay predicts both prevalent and incident human coronary artery disease ( 11,14,38 ). The differences were highly signifi cant ( Fig. 3B ; ANOVA, F(4, 31) = 21.6, P < 0.0001). Serum HDL of C57BL/6J displayed the highest macrophage effl ux capacity (6.2 ± 0.8%), while C57BLKS/J showed the lowest (4.7 ± 0.4%). The difference in effl ux capacity between the two strains is similar to that observed between control subjects and subjects with incident or prevalent CVD ( 11,13 ). Dunn's post hoc analysis indicated signifi cant differences in macrophage cholesterol effl ux capacity between C57BL/6J and the other four strains ( P < 0.0001), as well as between C57BLKS/J and SWR/J ( P = 0.01).
Because the cholesterol effl ux capacity of serum HDL with J774 macrophages refl ects the activity of multiple cellular pathways and because human ABCA1 defi ciency causes cholesterol-loaded macrophages to accumulate in many different tissues, we used BHK cells with mifepristone-inducible expression of human ABCA1 to quantify that pathway's effl ux capacity ( Fig. 3C ). In this system, ABCA1-specifi c cholesterol effl ux capacity is calculated as the percentage of total [ 3 H]cholesterol (medium plus cell) released into the medium of BHK cells stimulated with mifepristone minus the value obtained with cells incubated with medium. The differences in ABCA1-specifi c cholesterol effl ux capacity between the strains of mice were highly signifi cant [ANOVA, F(4, 31) = 7.9, P = 0.0002]. Serum HDL from DBA/2J mice had the largest effl ux capacity (5.7 ± 0.5%), while serum HDL from NZW/ LacJ mice had the lowest (3.7 ± 0.6%). Importantly, the rank orders of effl ux capacity of serum HDLs with macrophages and ABCA1-expressing cells were also clearly different, strongly suggesting that different factors control HDL's cholesterol effl ux capacity of ABCA1-expressing BHK cells and J774 macrophages loaded with radiolabeled cholesterol.
We next examined the relationships between macrophage cholesterol effl ux capacity, ABCA1-specifi c cholesterol effl ux capacity, and macrophage ABCA1-specifi c cholesterol effl ux capacity using serum HDL from all fi ve strains of mice (14 animals). Macrophage cholesterol Fig. 3. Macrophage cholesterol effl ux capacity, ABCA1 effl ux capacity, and HDL size in the different strains of mice. A: HDL size was measured by calibrated IMA in HDL isolated from serum by ultracentrifugation. Serum HDL was obtained by polyethylene glycol precipitation of APOB-containing lipoproteins from plasmaderived serum. Macrophage cholesterol effl ux capacity (cAMPstimulated J774 macrophages) (B) and ABCA1-specifi c cholesterol effl ux (mifepristone-stimulated BHK cells minus BHK cells incubated in medium alone) (C) were measured after a 4 h incubation with serum HDL (2.8% v/v) as described in the Methods.

APOA2 regulates HDL particle size
To determine whether APOA2 controls the size of HDL particles, we isolated HDL by ultracentrifugation of plasma from wild-type mice, Apoa2 Ϫ / Ϫ mice, and human APOA2 Fig. 4. Correlations of HDL protein levels with HDL size, macrophage effl ux capacity, and ABCA1 effl ux capacity. A, B: HDL size was measured by calibrated IMA in HDL isolated from serum by ultracentrifugation. Macrophage cholesterol effl ux capacity of serum HDL (C, D) or ABCA1-specifi c cholesterol effl ux capacity (E-H) were measured with serum HDL as described in the legend to Fig. 3 . The relationship of relative HDL protein abundance to HDL size and effl ux capacity was quantifi ed by Pearson's correlation. All proteins with r у |0.5| and P < 0.001 were included in this analysis . HDL (supplementary Fig. 2), we next investigated the relationship between these metrics, macrophage cholesterol effl ux capacity, and ABCA1-specifi c cholesterol effl ux capacity (supplementary Fig. 3). Macrophage cholesterol effl ux with serum HDL correlated strongly with plasma HDL-C (Pearson's correlation coeffi cient, r = 0.77, P < 0.0001) and phospholipids ( r = 0.73, P < 0.0001), but not with plasma APOA1 ( r = 0.41, P = 0.01) (supplementary Fig. 3A, C, E). In contrast, serum HDL's ABCA1-specifi c cholesterol effl ux capacity correlated strongly with APOA1 ( r = 0.7, P < 0.001), but not with HDL-C ( r = 0.2, P = 0.2) or phospholipids (supplementary Fig. 3B, D, F).
These observations support the proposal that ABCA1 accounts for only ‫ف‬ 30% of cholesterol effl ux from J774 macrophages, as assessed with radiolabeled cholesterol ( 10 ). Importantly, 40-50% of the variance in serum HDL's effl ux capacity with macrophages or cells expressing ABCA1 is not predicted by HDL-C or APOA1 levels, strongly suggesting that other factors mediate cholesterol effl ux in these systems.

Identifi cation of candidate HDL proteins that affect cholesterol effl ux capacity
Because plasma APOA1, HDL's major protein, failed to explain most of the variation in serum HDL's macrophage effl ux capacity, we next determined whether other HDL proteins might help modulate this HDL metric. Pearson's analysis revealed that six HDL proteins detected by LC-ESI-MS/MS on the HDL isolated from the different strains of mice ( Fig. 1 ) had signifi cant ( P р 0.0001) and strong correlation coeffi cients ( r у |0.5|) with serum HDL's effl ux capacity with either macrophages or the ABCA1 pathway .

APOE modulates the effl ux capacity of HDL with the ABCA1 pathway
Because APOE correlated negatively and strongly with ABCA1-specifi c effl ux capacity of serum HDL ( Fig. 4G ), we quantifi ed the effl ux capacity of serum HDLs prepared effl ux capacity (effl ux from J774 cells stimulated with cAMP) did not correlate ( r = Ϫ 0.09, P = 0.73) with ABCA1specifi c cholesterol effl ux capacity (effl ux in mifepristonestimulated BHK cells minus nonstimulated effl ux). In contrast, ABCA1-specific cholesterol efflux of macrophages (effl ux in cAMP-stimulated J774 macrophages minus nonstimulated effl ux) correlated strongly ( r = 0.86, P < 0.0001) with that of BHK cells. These observations confi rm that cholesterol effl ux from cAMP-stimulated J774 macrophages measures effl ux by pathways distinct from ABCA1 ( 10 ). They also are in good agreement with our data indicating that different factors control macrophage and ABCA1-specifi c cholesterol effl ux capacity ( Fig. 3 ).

HDL-C and APOA1 play different roles when they promote cholesterol effl ux from macrophages or cells expressing high levels of ABCA1
Because the fi ve strains of mice had signifi cantly different levels of HDL-C, APOA1, and phospholipids in serum Ϫ / Ϫ mice, and human APOA2 transgenic mice. Relative HDL particle abundance (A) and size (B, C) were determined by calibrated IMA in HDL isolated by ultracentrifugation of the plasma of Apoa2 tg mice (n = 15), Apoa2 +/+ mice (n = 8), and Apoa2 Ϫ / Ϫ mice (n = 4). arb, arbitrary units .
To determine the impact of APOE defi ciency on the effl ux capacity of HDL itself, we isolated HDL by ultracentrifugation from the plasma of Apoe +/+ and Apoe Ϫ / Ϫ mice and measured macrophage effl ux, ABCA1-specifi c effl ux, and ABCG1-specifi c effl ux, using cells exposed to identical protein concentrations of the lipoprotein (30 g/ml). HDL isolated from the Apoe Ϫ / Ϫ mice was 30% ( P = 0.004) more effi cient at promoting macrophage cholesterol effl ux capacity than wild-type HDL ( Fig. 6D ). HDL isolated from the Apoe Ϫ / Ϫ mice promoted ABCA1-specifi c effl ux even more effectively ( Fig. 6E ). In contrast, HDL isolated from Apoe Ϫ / Ϫ and Apoe +/+ mice exhibited similar activities with cells expressing ABCG1 ( Fig. 6F ). Taken together, these observations demonstrate that in mice, HDL associated APOE impairs the effl ux capacity of HDL with macrophages by a mechanism selectively involving the ABCA1 pathway.

DISCUSSION
We used fi ve inbred mouse strains to investigate the relationships between HDL's size, protein cargo, and cholesterol effl ux capacity. Proteomic analysis of isolated HDL from wild-type ( Apoe +/+ ) and APOE-defi cient ( Apoe Ϫ / Ϫ ) mice in the C57BL/6J background. The macrophage cholesterol effl ux of serum HDL from the Apoe Ϫ / Ϫ mice was markedly lower than that of wild-type mice ( Fig. 6A ). This difference likely refl ects a decrease in the number of HDL particles, because HDL-C levels are markedly lower in Apoe Ϫ / Ϫ than Apoe +/+ mice ( 39 ). When we normalized effl ux capacity to HDL particle concentration (HDL-P ima as determined by calibrated IMA), the effl ux capacity of the HDL from the Apoe Ϫ / Ϫ mice was higher than that of the HDL from the wild-type mice (9.68 ± 2.57% vs. 2.12 ± 1.02%, respectively, n = 3, P = 0.052), strongly suggesting that APOE expression impairs HDL's effl ux capacity with macrophages.
In striking contrast to effl ux capacity with macrophages, the ABCA1-specifi c cholesterol effl ux capacity of serum HDLs from Apoe Ϫ / Ϫ and Apoe +/+ mice were similar ( Fig.  6B ), despite the markedly lower HDL-C and HDL particle concentration in these mice. When we used calibrated IMA to quantify HDL particle concentration in the Apoe +/+ and Apoe Ϫ / Ϫ mice, ABCA1-specifi c cholesterol effl ux per particle correlated negatively and signifi cantly ( r = Ϫ 0.58, P = 0.0002) with the APOE content of the HDL particles ( Fig. 6C ). When we normalized effl ux capacity to HDL particle number in serum, ABCA1-specifi c cholesterol effl ux was markedly higher in the Apoe Ϫ / Ϫ mice than in the associate with increased CVD risk ( 45 ). We also found that the cholesterol effl ux capacity of serum HDL from Apoe Ϫ / Ϫ mice was comparable to that of wild-type mice with the ABCA1 pathway when we used the same concentration of serum HDL in the assays. Because of the marked differences in HDL-C and HDL particle concentration in the two strains (>50% reduction of both metrics in Apoe Ϫ / Ϫ mice), these observations are consistent with the idea that APOE associates with HDL directly or indirectly inhibits HDL's effl ux capacity by the ABCA1 pathway. The J774 macrophages we used to quantify cholesterol effl ux capacity do not secrete APOE. Thus, cell-derived APOE would not affect the effl ux capacity of serum HDL in our studies. This is important because mouse and human macrophages secrete APOE ( 46,47 ) and lipid-free APOE ( 48,49 ), like other lipid-free apolipoproteins, promotes cholesterol effl ux by ABCA1 due to its ␣ helical structure ( 6 ). Cell autonomous production of APOE promotes cholesterol effl ux from hematopoietic stem cells ( 50 ), demonstrating the potential impact of macrophagederived APOE on effl ux capacity.
We used wild-type and Apoe Ϫ / Ϫ mice to determine whether HDL-associated APOE helps regulate the cholesterol effl ux capacity of serum HDL. HDL isolated by ultracentrifugation from Apoe Ϫ / Ϫ mice had signifi cantly better effl ux capacity with macrophages (based on HDL's protein content or HDL particle concentration) than did wild-type HDL. Moreover, its effl ux capacity increased markedly with cells expressing ABCA1, but not with cells expressing ABCG1. These observations provide strong evidence that APOE associated with APOE impairs HDL's cholesterol effl ux capacity from macrophages by a pathway involving ABCA1. Further experiments are required to understand how APOE affects the cholesterol effl ux capacity of HDL subspecies.
We previously demonstrated that HDL's APOE content was elevated in HDL isolated from humans with established CVD ( 15,21 ). Moreover, in a prospective study of CVD risk in subjects with established atherosclerosis, the APOE content of APOA1-containing HDL particles was a much stronger predictor of future cardiac events than HDL-C or LDL-cholesterol levels ( 51 ). These observations raise the possibility that APOE impairs HDL's cardioprotective effects, perhaps in part by affecting cholesterol effl ux capacity.
Signifi cantly, HDL-associated APOE interacts very differently with the ABCG1 pathway ( 52,53 ), where it acts in concert with LCAT to drive net cholesterol effl ux by promoting cholesteryl esterifi cation and an increase in HDL size. It remains to be determined whether the cholesterol effl ux capacity of serum HDL with the ABCG1 pathway associates with cardiovascular risk.
HDL's content of PLTP correlated strongly with its ability to promote cholesterol effl ux from macrophages. This observation is consistent with the fi nding that PLTP facilitates ABCA1-dependent cellular cholesterol effl ux in concert with HDL ( 54 ). In vitro, PLTP remodels mediumsized HDL particles, generating larger HDL particles and pre ␤ HDL, a ligand for ABCA1 ( 55 ). If this reaction takes identifi ed more than 70 proteins linked to lipid metabolism, the acute-phase response, proteinase inhibition, and the immune system, as originally described for human HDL ( 15 ). Using size-exclusion and phospholipid affi nity chromatography, Gordon et al. ( 36 ) obtained similar fi ndings for HDL isolated from C57BL/6J mice . These observations indicate that the HDLs of the two distantly related species carry the same functional categories of proteins. However, only 42 of the 72 proteins we detected in mouse HDL have been identifi ed in human HDL ( 36 ), suggesting both qualitative and quantitative differences.
Hierarchical cluster analysis of the HDL proteome recapitulated the genealogy of the fi ve mouse strains. We found signifi cant variability in HDL size and the cholesterol effl ux capacity of serum HDL, supporting the idea that these proposed metrics of HDL's cardioprotective effects are under genetic control. Moreover, the lipoprotein's content of specifi c proteins strongly correlated with each metric, raising the possibility that HDL's protein cargo helps regulate its size and function. For example, the APOA2 content of HDL strongly correlated with HDL size, as previously observed ( 37,(40)(41)(42)(43).
Using mice that were defi cient in APOA2 or expressed human APOA2, we confi rmed that the protein controlled the size of both the medium and large HDL subspecies detected in mice. Because mouse and human APOA2 are only ‫ف‬ 60% identical in sequence and APOA2 is dimerized in human, but not mouse, HDL, the impact of APOA2 on size appears to be independent of its precise sequence. The fact that HDL's lipid composition varied little between the mouse strains strengthens the proposal that APOA2 is an important structural protein that regulates HDL's size.
The major protein of HDL is APOA1, but its content did not correlate with serum HDL's ability to remove cholesterol from macrophages. However, we noted a strong correlation between cholesterol effl ux capacity and isolated HDL's content of RBP4 and PLTP. In contrast, serum HDL's ability to remove cholesterol from BHK cells that expressed ABCA1 correlated strongly with isolated HDL's content of APOA1, APOC3, and APOD. These observations imply that the proteins that control HDL's effl ux capacity with macrophages differ from those that infl uence cholesterol removal via the ABCA1 pathway. This is consistent with the fi nding that the ABCA1 pathway accounts for only about one-third of serum HDL's effl ux capacity with J774 macrophages ( 10 ). Although it is generally believed that lipid-free or lipid-poor apolipoproteins initially interact with ABCA1, a recent report shows that small HDLs (HDL3b and HDL3c) are effective in removing cholesterol from ABCA1-expressing cells ( 7,44 ).Thus, HDL size is also likely to be an important determinant of ABCA1dependent cholesterol effl ux.
Two proteins in isolated HDL, APOE, and SAA1 strongly and negatively correlated with cholesterol effl ux capacity when we used serum HDL to accept cholesterol from cells selectively expressing ABCA1. There was no such correlation with SAA2. It is noteworthy that elevated blood levels of total SAA (proteins carried exclusively by HDL in humans)