Inflammatory remodeling of the HDL proteome impairs cholesterol efflux capacity.

Recent studies demonstrate that HDL's ability to promote cholesterol efflux from macrophages associates strongly with cardioprotection in humans independently of HDL-cholesterol (HDL-C) and apoA-I, HDL's major protein. However, the mechanisms that impair cholesterol efflux capacity during vascular disease are unclear. Inflammation, a well-established risk factor for cardiovascular disease, has been shown to impair HDL's cholesterol efflux capacity. We therefore tested the hypothesis that HDL's impaired efflux capacity is mediated by specific changes of its protein cargo. Humans with acute inflammation induced by low-level endotoxin had unchanged HDL-C levels, but their HDL-C efflux capacity was significantly impaired. Proteomic analyses demonstrated that HDL's cholesterol efflux capacity correlated inversely with HDL content of serum amyloid A (SAA)1 and SAA2. In mice, acute inflammation caused a marked impairment of HDL-C efflux capacity that correlated with a large increase in HDL SAA. In striking contrast, the efflux capacity of mouse inflammatory HDL was preserved with genetic ablation of SAA1 and SAA2. Our observations indicate that the inflammatory impairment of HDL-C efflux capacity is due in part to SAA-mediated remodeling of HDL's protein cargo.

Kentucky. Saa1 / 2 Ϫ / Ϫ mice and littermate WT mice were in a 129SvEv/C57BL/6 background ( 29 ). Experiments with C57BL/6 mice were performed at the University of Washington. Animal experiments were approved by the Institutional Animal Care and Use Committees of each institution. All mice were 8-12 weeks of age, fed a low-fat diet, and maintained in a pathogen-free facility with 12 h light-dark cycles and free access to food and water. Acute infl ammation was induced in female mice by subcutaneous injection of silver nitrate (0.5 ml, 2% w/w) ( 30,31 ). Control mice were injected with 0.5 ml of sterile normal saline. Blood anticoagulated with EDTA was collected 24 h after the injection, and plasma was prepared by centrifugation. Plasma samples from two mice were combined for HDL isolation.

HDL
HDL was isolated using sequential ultracentrifugation (d = 1.063-1.21 g/ml), as described for mouse ( 29,32 ) and human HDL ( 33,34 ). HDL was stored on ice in the dark and used within 1 week of preparation.

Cholesterol effl ux assays
J774 macrophages were loaded with cholesterol for 24 h at 37°C in DMEM containing acetylated-LDL (50 g protein/ml) and [ 3 H]cholesterol (1 Ci/ml). The cells were then washed with DMEM and incubated for an additional 24 h in DMEM containing cAMP (0.5 mM) and the liver X receptor (LXR) agonist, T0901317 (5 M) ( 37 ). Effl ux of [ 3 H]cholesterol was measured after a 6 h incubation with DMEM supplemented with 0.1% BSA without or with 30 g/ml of HDL protein. Cholesterol effl ux mediated by HDL was calculated as the percentage of total [ 3 H]cholesterol (medium plus cell) released into the medium after the value obtained with DMEM/BSA alone was subtracted.

LC-MS/MS analysis
Tryptic digests of mouse HDL (2 g protein) were injected onto a C18 trap column (Paradigm Platinum Peptide Nanotrap, 0.15 × 50 mm; Michrom Bioresources, Inc.), desalted (50 l/min) for 5 min with 1% acetonitrile/0.1% formic acid, eluted onto an analytical reverse-phase column (0.15 × 150 mm, Magic C18AQ, 5 m, 200 Å ; Michrom Bioresources, Inc.), and separated 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) on a Paradigm M4B HPLC (Michrom Bioresources, Inc.). Positive ion mass spectra were acquired with ESI in a linear ion trap mass spectrometer (LTQ; Thermo Electron Corp., San Jose, CA) with data-dependent acquisition (one MS survey scan followed by MS/MS scans of the eight most abundant ions in the survey scan). An exclusion window of 45 s was used after two acquisitions of the same precursor ion. Alternatively, the HDL from WT mice, with and without infl ammation, were analyzed on Thermo QE+ mass spectrometer using data-independent acquisition ( m/z 20 window with m/z 10 overlap, 17,500 resolution, normalized collision energy 25) ( 38 ) and data for apoA-I native and oxidized methionine-and tryptophan-containing peptides were analyzed and quantifi ed using Skyline ( 39 ).
Tryptic digests of human HDL (2 g protein) were injected onto a C18 trap column (Magic AQ C18 200A, 5 m, 0.1 × 20 mm; burden that is independent of HDL-C and apoA-I levels. However, the molecular factors controlling the sterol effl ux capacity of serum HDL are poorly understood ( 8,13 ). For example, a recent study found that the majority of radiolabeled cholesterol released from macrophages did not reside in HDL, suggesting that impaired sterol effl ux capacity does not necessarily refl ect alterations in HDL itself ( 11 ).
Infl ammation and metabolic disorders have been proposed to convert HDL to a dysfunctional form lacking anti-atherogenic properties (14)(15)(16)(17)(18)(19). For example, HDLs from mice and humans with acute infl ammation are less able to promote sterol effl ux from macrophages (20)(21)(22)(23). Detailed metabolic studies demonstrate that multiple steps in reverse cholesterol transport from macrophages are inhibited in infl amed mice ( 21,22 ). Furthermore, proteins cotransported with HDL in plasma, such as paraoxonase and clusterin, have been proposed to have antioxidant and anti-infl ammatory properties, and their levels change in response to infl ammation ( 15,(24)(25)(26)(27). Loss of sterol effl ux capacity and of anti-infl ammatory and/or antioxidant proteins, perhaps in concert with gain of pro-infl ammatory proteins, may thus be key factors in generating dysfunctional HDL ( 7 ).
In the current study, we investigated the protein cargo and function of HDL isolated from humans and mice with infl ammation. Our observations indicate that infl ammatory remodeling mediated by serum amyloid A (SAA) is one mechanism for generating HDL whose capacity to promote sterol effl ux is impaired.

Human studies
The study was approved by the institutional review board of the Clinical Research Center, New Orleans, and reviewed by the institutional review board of the National Institute of Allergy and Infectious Diseases ( 28 ). Twelve healthy male volunteers aged 22-49 years underwent a complete history and physical examination prior to entering the study. All subjects had normal physical exams and routine blood and urine chemistries, and none were taking medications or had known medical conditions. The 12 subjects were randomly assigned to the endotoxin injection groups. The subjects were injected iv with 1 ng/kg or 2 ng/kg National Institutes of Health (NIH) "equivalent" endotoxin ( ( 28 ). Consistent with previous studies with these endotoxin preparations, we found that highsensitivity C-reactive protein levels in the group that received the 4 ng/kg dose of FDA endotoxin were comparable to those in the group that received the 1 ng/kg dose of NIH endotoxin. Because SAA and CRP levels were similar in the 1 ng/kg NIH endotoxin and 4 ng/kg FDA endotoxin groups, we combined the subjects into a 1 ng/kg endotoxin group in subsequent analyses.

Mouse studies
Experiments with mice lacking SAA1.1 and SAA2.1 (termed here as Saa1 / 2 Ϫ / Ϫ mice) were performed at the University of hallmarks of acute infl ammation, as indicated by: i ) symptoms (chills, headache) and signs (elevated body temperature); ii ) a rapid transient elevation of infl ammatory cytokines; and iii ) a dose-dependent increase in blood levels of CRP and SAA after the injection ( 28 ).

Acute infl ammation induced by low doses of endotoxin remodel the human HDL proteome and impair cholesterol effl ux from macrophages
To investigate the ability of HDL and infl ammatory HDL to promote sterol effl ux from macrophages, we used ultracentrifugation (d = 1.063-1.21 g/ml) to isolate HDL from blood collected from each subject 30 min prior to (control HDL) and 24 h after (infl ammatory HDL) the injection of saline or low doses of endotoxin. At these doses, endotoxin induced only a mild infl ammation. The effl ux capacity of infl ammatory HDL was reduced ( Fig.  1A ) by ‫ف‬ 10% at 1 ng/kg of endotoxin and by ‫ف‬ 20% at 2 ng/kg ( P = 0.009, ANOVA with Tukey's HSD). These observations are consistent with previous reports that the infl ammatory response impairs HDL's ability to remove sterol from macrophages in mice and rabbits, as well as in humans (20)(21)(22)(23).

Acute infl ammation induced by low doses of endotoxin selectively increases SAA levels in HDL
To assess how changing the relative abundance of proteins in HDL might affect the lipoprotein's function, we analyzed control and infl ammatory HDL with MS. LC-ESI-MS/MS of tryptic digests of the HDL identifi ed with high confi dence (estimated FDR 2%) 82 proteins associated with HDL (supplementary Table 1). This approach identifi ed all of the proteins found in our earlier studies of HDL ( 33,34 ). We also identifi ed a number of proteins that were not known to reside in HDL ( 45 ). The fact that we identifi ed more proteins in this analysis than in our previous studies likely refl ects improvements in LC and mass spectrometers.
To determine which HDL proteins changed in relative abundance when humans were challenged with endotoxin, we used the G -test and t-test to fi nd signifi cant differences in spectral counts, a measure of relative protein abundance ( 42 ). We estimated the FDR by using the same statistical tests with all possible permutations of the data ( 42 ). Permutation analysis revealed that G > 3.5 ( G -test) and P < 0.001 ( t-test) yielded the most true-positive changes in protein abundance after endotoxin challenge, with an estimated FDR of 4% (data not shown).
Using these stringent statistical criteria ( Fig. 1D , supplementary Table 1), the only proteins that changed in relative abundance in infl ammatory HDL were the acute-phase proteins, SAA1 and SAA2. The relative concentration of SAA1/2 (SAA1 and SAA2 are quantifi ed together because the proteins share 95% sequence homology) increased 6-fold in the subjects who received the 1 ng/kg dose of endotoxin and 9-fold in those who received the 2 ng/kg dose.
We used two independent approaches to confi rm that levels of SAA1 and SAA2 were elevated. First, we used extracted ion chromatograms to quantify the relative Michrom Bioresources, Inc.), desalted for 15 min with water/0.1% formic acid (4 l/min), eluted onto an analytical column (Magic AQ C18 90A, 5 m, 0.1 × 200 mm; Michrom Bioresources, Inc.), and separated at a fl ow rate of 0.4 l/min over 180 min, using a linear gradient of 5-35% buffer D (acetonitrile/0.1% formic acid) in buffer C (0.1% formic acid) on a NanoAquity HPLC (Waters, Milford, MA). Positive ion mass spectra were acquired with ESI in a hybrid linear ion trap-Orbitrap mass spectrometer (LTQ Orbitrap XL; Thermo Fisher, San Jose, CA) with datadependent acquisition of MS/MS scans (linear ion trap) on the eight most abundant ions in the survey scan (Orbitrap, resolution 60,000). An exclusion window of 45 s was used after two repeated acquisitions of the same precursor ion.

Protein identifi cation and quantifi cation
HDL protein was digested and analyzed essentially as previously described ( 40 ). MS/MS spectra were matched against the human International Protein Index (IPI) database (mouse v.3.54, January 2009; human v.3.72, April 2010), using the SEQUEST (v2.7) search engine with fi xed Cys carbamidomethylation and variable Met oxidation modifi cations. SEQUEST results were further validated with Trans-Proteomic Pipeline tools, using an adjusted probability of у 0.90 for peptides and у 0.95 for proteins. Proteins considered for analysis had to be detected in у 4 analyses with у 2 unique peptides. Relative protein quantifi cation was performed using spectral counting ( 41 ). Signifi cant differences in spectral counts were identifi ed using the combination of G -test and t-test together with permutation analysis to estimate false discovery rate (FDR) ( 42 ). The abundance of SAA relative to apoA-I was estimated using extracted ion chromatograms and sum of peak areas for the three most abundant peptides from each protein according to the approach of Silva et al. ( 43 ).

Biochemical and immunochemical assays
HDL protein levels were quantifi ed by Bradford assay and corrected to set of standards quantifi ed by Lowry assay. HDL and plasma lipids were quantifi ed biochemically [cholesterol and cholesteryl ester (Amplex Red, Invitrogen), triglycerides (Cayman Chemical, Ann Arbor, MI), phospholipids (Wako Chemical, Richmond, VA)]. CRP (Invitrogen, Carlsbad, CA) and SAA were measured by ELISA (Invitrogen, Camarillo, CA). HDL particle concentration was measured by calibrated ion mobility analysis ( 44 ).

Statistical analyses
Analyses were performed with the Statistical Package for the Social Sciences (SPSS v. 19) and R statistical package (v.2.14). Results represent means and SEMs. The statistical signifi cance of differences between groups was evaluated by the two-tailed Student's t -test or ANOVA with Tukey's honest signifi cant difference (HSD) post hoc test.

RESULTS
The human study involved 12 male subjects (40 ± 7 years old; fasting plasma levels of lipids and infl ammatory proteins: HDL-C, 39.9 ± 3.8 mg/dl; LDL-C, 107 ± 4.9 mg/dl; CRP, 2.0 ± 0.6 mg/l; SAA, 14.3 ± 5.3 g/ml). All subjects were apparently healthy, normolipidemic, not using medications, and had no symptoms or signs of infl ammation. To induce infl ammation, subjects were injected with the equivalent of 1 ng/kg (n = 8) or 2 ng/kg (n = 4) of reference endotoxin ( 28 ). All of the subjects developed the with an ELISA (which detects all the SAA isoforms) similarly indicated 15-and 17-fold increases of SAA in HDL with 1 and 2 ng/kg of endotoxin. Based on extracted ion chromatograms (see Methods), we calculated that SAA represented approximately 13 and 20% of apoA-I with 1 and 2 ng/kg doses of endotoxin, but less than 1% of apoA-I in HDL from the same subjects prior to endotoxin challenge.
abundance of peptides unique to SAA1 and SAA2, because this approach quantifi es the individual isoforms and estimates protein ratios more accurately than spectral counting. This approach demonstrated 19-and 30-fold increases in ion current for peptides specifi c for SAA1 (FFGHGAEDSLADQAANEWGR) and SAA2 (GPGGA-WAAEVISNAR) in the subjects treated with 1 ng/kg and 2 ng/kg of endotoxin, respectively. Quantifi cation of SAA We therefore fi rst characterized the HDL proteome of C57BL/6J mice isolated by ultracentrifugation (d = 1.063-1.21 g/ml) and compared it to human HDL. Of the 75 and 82 proteins identifi ed in mouse and human HDL, respectively, 17 were identifi ed only in human HDL, while 8 proteins were detected only in mouse HDL (supplementary Fig. 1). Nine of the proteins identifi ed in only mice or human HDL lacked orthologs in the other species. However, gene ontology analysis of the identifi ed proteins revealed that the HDLs of both species contain the same functional categories of proteins. These data demonstrate that the HDL proteomes of mice and humans share many proteins, but also are distinct.
To examine the impact of infl ammation on the protein cargo of mouse HDL, we injected saline or silver nitrate subcutaneously into WT C57BL/6 mice, and collected blood 24 h later. We used silver nitrate because this model of infl ammation, while unsuitable for human studies, has been widely studied in animal models ( 30,31 ). It also avoids the possibility that any endotoxin adsorbed by HDL would affect cell-based assays. With acute infl ammation, HDL isolated from the WT animals exhibited ‫ف‬ 20% less effl ux capacity in J774 macrophages ( Fig. 3A ; P = 0.0002) and exhibited 40% less ability to decrease cholesterol mass with macrophage foam cells isolated from the peritoneum of cholesterol-fed Ldlr Ϫ / Ϫ mice (supplementary Fig. 2).
Proteomic analysis identifi ed 81 proteins with high confi dence in HDL isolated from infl amed mice. As expected, based on higher level of infl ammation in mice in contrast to HDL isolated from endotoxin-treated humans, onethird of the proteins in infl amed mouse HDL had changed their relative abundance ( Fig. 3E , supplementary Table  2). Peptides for SAA1 and SAA2 were virtually undetectable in HDL of mice injected with saline ( Fig. 3B-D ). In infl amed mice, SAA1 and SAA2 were among the most abundant HDL proteins ( Fig. 3E ). Extracted ion chromatograms of peptides specifi c to each SAA isoform demonstrated a marked increase in levels of both SAA1 and SAA2 in infl ammatory HDL ( Fig. 3B-D ). Based on extracted ion chromatograms (see Methods), we calculated that SAA1/2 increased from less than 1% of apoA-I in control HDL to approximately 70% of apoA-I in infl ammatory HDL.
With acute infl ammation, 18 proteins in HDL increased and 11 decreased in relative abundance (supplementary Table 2). Many of those proteins have previously been shown to change in abundance in acute-phase HDL ( 17,49 ), including apoJ, apoE, apoA-V, phospholipid transfer protein, lipopolysaccharide binding protein, multiple apoCs, apoM, LCAT, and PON1. Haptoglobin was detected in infl ammatory HDL but not control HDL. Infl ammation did not signifi cantly affect levels of the ␣ and ␤ chains of hemoglobin, which have been proposed as markers of dysfunctional HDL ( 50 ). While infl ammation induced profound changes in the HDL proteome, it only modestly altered plasma phospholipid and total cholesterol levels ( Fig. 3F, G ). In contrast, infl ammation failed to alter HDL-C or HDL particle concentration ( Fig. 3H, I ).
To assess how changing the relative abundance of SAA in HDL might affect the lipoprotein's function, we measured the abilities of control HDL and infl ammatory HDL to promote cholesterol effl ux from macrophages ( Fig. 1B, C ). There was a strong inverse linear correlation between HDL sterol effl ux ability and levels of SAA in HDL, as assessed by MS ( r = Ϫ 0.56, P = 0.003) and biochemically ( r = Ϫ 0.64, P = 0.001). These observations are consistent with the hypothesis that SAA impairs HDL's ability to promote sterol effl ux from macrophages. Because regression analysis indicates that SAA levels only explain approximately 36% of the variance in effl ux capacity, other factors also likely contribute to altered effl ux capacity, including interindividual differences in sterol effl ux capacity without infl ammation.

Infl ammation with low doses of endotoxin fails to affect plasma lipids in humans
At the time point and low endotoxin doses used in our study, we did not observe signifi cant changes in HDL-C or other plasma lipids ( Fig. 1E-G ), the lipid composition of isolated HDL ( Fig. 1H-J ), or plasma LCAT activity (data not shown). High-resolution size-exclusion chromatography also failed to reveal alterations in the distribution of plasma cholesterol in the endotoxin-treated subjects (data not shown). These observations, in concert with our demonstration of increased levels of SAA1/2 in infl ammatory HDL of humans treated with endotoxin, suggest that alterations in the lipoprotein's protein cargo impair its effl ux capacity.

SAA1 generates dysfunctional HDL in a model system
To determine whether SAA1 alters sterol effl ux from macrophages in vitro, human HDL was fi rst incubated with recombinant SAA1 at various ratios for 3 h at 25°C, and the lipoproteins were then re-isolated by ultracentrifugation (d = 1.21 g/ml). SDS-PAGE and densitometric quantifi cation of Imperial Blue-stained proteins ( Fig. 2A, B ) indicated that SAA1 accounted for ‫ف‬ 10-30% of the protein in re-isolated HDL, levels markedly higher than those observed with mild infl ammation in humans (see above) and similar to that observed in infl ammatory HDL from mice (see below). The molar fraction of SAA1 incorporated into HDL increased in concert with the loss of the ability of the lipoprotein to promote sterol effl ux from macrophages ( Fig. 2C ). ESI-MS/MS of tryptic digests demonstrated that apoA-I and SAA1 were the major proteins that changed in relative abundance when HDL was enriched with SAA1 in vitro ( Fig. 2D ). These observations support the proposal that SAA enrichment of HDL in vitro impairs its sterol effl ux capacity with macrophages.

Acute infl ammation alters plasma lipids and remodels the HDL proteome in mice
The human HDL proteome has been extensively investigated ( 26,33,46,47 ), but much less is known about the protein composition of mouse HDL. Previous studies have shown differences in the apoA-I and apoA-II content and apparent sizes of mouse and human HDL ( 48 ).  Table 3). SDS-PAGE of Coomassiestained proteins indicated that SAA content of HDL of infl amed Saa1 / 2 Ϫ / Ϫ mice was markedly reduced ( Fig. 4C ); MS/MS analysis indicated that SAA1/2 was undetectable in these HDLs (see below).

Infl ammation-induced alterations in plasma and HDL lipid composition are similar in WT and Saa1 / 2 ؊ / ؊ mice
Previous studies demonstrate that infl ammation markedly alters the lipid composition of plasma and HDL, and that lipids can contribute to the sterol effl ux capacity of HDL by certain pathways (e.g., ABCG1) ( 49,55,56 ). We therefore quantifi ed levels of specifi c lipid classes in plasma and in HDL of WT and Saa1 / 2 Ϫ / Ϫ mice, with and without infl ammation.
Plasma levels of HDL-C, cholesterol, and phospholipids were altered signifi cantly 24 h after injection of silver nitrate ( Fig. 4E-G ). However, the patterns of change in plasma lipids and HDL-C during infl ammation were similar in WT and Saa1 / 2 Ϫ / Ϫ mice ( Fig. 4E-G ). Moreover, there were no signifi cant differences in the infl ammationinduced pattern of changes in phospholipid or cholesteryl ester of HDL isolated from WT and Saa1 / 2 Ϫ / Ϫ mice ( Fig.  4H, I ). The free cholesterol content of HDL increased signifi cantly with infl ammation in both strains. The enrichment of HDL from infl amed Saa1 / 2 Ϫ / Ϫ mice was signifi cantly greater than that of infl amed WT mice ( Fig. 4J ). Together with the data on humans and on HDL enriched in vitro with SAA1, these observations suggest that changes induced by infl ammation in the proportions of lipid classes

Acute infl ammation does not alter oxidation state of HDL
In addition to altering HDL, proteome infl ammation may also increase HDL and apoA-I oxidation, a modifi cation known to impair HDL sterol effl ux capacity ( 10,(51)(52)(53). We therefore interrogated the proteomics data of WT mice with and without infl ammation to determine whether acute infl ammation increased oxidation of apoA-I methionine or tryptophan residues ( 10,54 ). Extracted ion chromatograms of tryptic peptides showed no difference in oxidized peptides containing Met91, Met180, or Trp77 between HDL isolated from control and infl amed mice (data not shown), demonstrating that oxidation is not contributing to impaired sterol effl ux of infl ammatory HDL.

HDL of infl amed SAA-defi cient mice is protected from loss of cholesterol effl ux capacity
We next compared the abilities of HDLs isolated from WT and Saa1 / 2 Ϫ / Ϫ mice on the same genetic background, without and with infl ammation, to promote sterol effl ux from macrophages. In the absence of infl ammation, there was no difference between the two strains ( Fig. 4A ). The effl ux capacity of HDL from infl amed WT animals exhibited ‫ف‬ 40% loss of effl ux capacity compared with HDL of WT animals without infl ammation ( P < 0.001, n = 6). In striking contrast, HDL from infl amed Saa1 / 2 Ϫ / Ϫ mice exhibited no impairment in the ability to promote sterol effl ux ( Fig. 4A ). The effl ux capacity of HDL from infl amed WT mice and infl amed Saa1 / 2 Ϫ / Ϫ mice were signifi cantly different ( P = 0.008, n = 6). HDL isolated from infl amed WT mice was signifi cantly enriched in protein content

Loss from HDL of proteins with proposed cardioprotective effects is mediated by SAA
We next compared levels of proposed anti-atherogenic proteins in WT and Saa1 / 2 Ϫ / Ϫ mice. The molar fractions of apoA-I, apoE, phospholipid transfer protein, and apoA-II were all increased in infl ammatory HDL isolated from Saa1 / 2 Ϫ / Ϫ mice relative to HDL from WT mice ( Fig. 4D ,  inset). Moreover, levels of PON1 ( r = 0.65, P = 0.007) and apoA-II ( r = 0.78, P = 0.0004), but not apoA-I ( r = 0.37, P > 0.05) or apoE ( r = Ϫ 0.34, P > 0.05), correlated with the sterol effl ux capacity of HDL (supplementary Fig. 4). These observations support the proposal that SAA modulates the abundance in HDL of proteins that affect atherosclerosis in mice, and that this in turn may alter the in HDL are unlikely to explain the normal effl ux capacity of HDL isolated from infl amed Saa1 / 2 Ϫ / Ϫ mice. However, it is possible that infl ammation-induced alterations in a specifi c species of lipids [such as phosphatidylserine ( 57 )] contribute to impairment of sterol effl ux capacity.

SAA defi ciency reduces HDL protein remodeling during infl ammation
These observations suggest that increased levels of SAA1/2 and/or lower levels of other HDL proteins might contribute to the loss of effl ux capacity of HDL of infl amed WT mice. We therefore compared the protein composition of HDL isolated from WT and Saa1 / 2 Ϫ / Ϫ mice, with and without infl ammation (supplementary Table 4). In the absence of infl ammation, only two proteins (SAA4 and APOA2) met our dual statistical criteria for differential protein abundance between HDL isolated from WT and Saa1 / 2 Ϫ / Ϫ mice (supplementary Fig. 3). In contrast, markedly more proteins differed in relative abundance between control and infl amed WT mice (n = 8) than in infl amed endotoxin [ ‫ف‬ 1/2 dose previously published ( 21 )], stringent statistical analysis demonstrated that only two proteins, SAA1 and SAA2, were differentially abundant in infl ammatory HDL. Moreover, the ability of HDL isolated from the infl amed subjects to accept cholesterol from macrophages correlated inversely with the HDL's SAA1/2 content. Importantly, infl ammation did not signifi cantly change HDL's lipid composition or plasma levels of HDL-C, LDL-C, and triglycerides, raising the possibility that the increased SAA1/2 content was the main factor impairing cholesterol effl ux from macrophages.
Using WT mice and mice defi cient in SAA1/2, we directly determined whether those proteins alter HDL's cholesterol effl ux capacity. In the absence of infl ammation, the protein composition of HDL isolated from WT mice was essentially identical to that of HDL from the Saa1 / 2 Ϫ / Ϫ mice. However, acute infl ammation induced with silver nitrate markedly remodeled the HDL proteome of the WT mice. Indeed, the relative abundance of cardioprotective effects of HDL by mechanisms independent of HDL-C.

DISCUSSION
HDL's cardioprotective effect is attributed, in part, to its ability to mobilize excess cholesterol from artery wall macrophages ( 2-4, 6, 7 ). Consistent with this proposal, recent studies demonstrate that impaired sterol effl ux capacity of serum HDL from J774 macrophages strongly associates with prevalent and incident CAD status, but is independent of HDL-C and apoA-I levels ( 9,11,12 ). It is therefore critical to uncover the molecular mechanisms that modulate effl ux capacity. Because acute infl ammation markedly impairs the sterol effl ux capacity of HDL in mice and in humans (20)(21)(22)(23), we used MS to investigate the role of HDL proteins in modulating cholesterol effl ux from macrophages. In humans challenged with a low dose of SAA content during the time points used for the study. A recent study of type 2 diabetics found negative association of SAA levels with SRBI-mediated cholesterol effl ux capacity and no association with ABCG1-mediated cholesterol effl ux capacity in 500 diabetic and nondiabetic subjects, although clinical relevance of these assays of cholesterol effl ux capacity is unknown ( 67 ). In contrast, another study suggested that infl ammation-induced increase in the phospholipid content of HDL modestly improved the ability of HDL to promote sterol effl ux by the ABCG1 pathway ( 55 ). However, effl ux by this pathway only contributes about one-fourth of the effl ux from macrophages to HDL ( 68 ). Moreover, while the major determinant of sterol effl ux by the ABCG1 pathway to lipoproteins is their phospholipid content ( 69 ), we observed similar levels of the phospholipid in HDL isolated from infl amed WT and Saa1 / 2 Ϫ / Ϫ mice, and the phospholipid as well as the cholesteryl ester content of HDL isolated from control and infl amed Saa1 / 2 Ϫ / Ϫ mice were similar to those of control and infl amed WT mice. In contrast to lipids, protein content signifi cantly increased only in the HDL of infl amed WT mice, with this increase largely accounted for by the marked increase in SAA1/2 content of the lipoprotein by addition to, rather than by displacement of, apoA-I and other HDL proteins ( 70 ). Our model system studies further strengthen the hypothesis that SAA is important for generating HDL with impaired sterol effl ux capacity from J774 macrophages. Taken together, our observations provide strong evidence that SAA1 and SAA2 play critical roles in rendering HDL dysfunctional during infl ammation ( 71 ). Recent studies of hypercholesterolemic mice have reached different conclusions regarding the role of SAA in atherosclerosis. apoE-defi cient mice defi cient in SAA1/2 fed a Western-type diet were not protected from atherosclerosis ( 72 ). In contrast, a study using lentiviral-induced expression of SAA in apoE-defi cient mice fed a Westerntype diet ( 73 ), as well as a study using adenoviral transfection of SAA in apoE-defi cient mice on chow diet, found signifi cantly increased atherosclerosis when SAA was expressed ( 74 ).
A key issue then becomes whether the modestly elevated level of SAA found in HDL isolated from hypercholesterolemic animal models of atherosclerosis ( 75 ) is likely to affect the effl ux capacity of HDL. Those levels appear to be lower than the levels of SAA1/2 in HDL we observed in our human and mouse studies. The levels of SAA1/2 in humans injected with low doses of endotoxin were significantly lower than those observed in mice injected with silver nitrate (approximately 13 and 20% of apoA-I in HDL in humans compared with 70% in mice), and we observed a more modest impairment of the sterol effl ux capacity of infl ammatory HDL isolated from humans than from mice.
Our model system studies also demonstrated that impaired effl ux capacity of HDL was quantitatively linked to HDL's content of SAA. Collectively, these studies indicate that signifi cant levels of SAA in HDL are necessary to cause signifi cant impairment of HDL sterol effl ux capacity. Such levels may be relevant in chronic infl ammatory diseases one-third of the proteins we detected in WT mouse HDL changed signifi cantly. These observations suggest that the level of infl ammation induced by silver nitrate injection in mice was much greater than the level we observed in humans treated with low doses of endotoxin. Indeed, we estimate that the relative level of SAA was approximately 70% that of apoA-I in mouse infl ammatory HDL compared with ‫ف‬ 20% that of apoA-I in infl ammatory HDL in humans exposed to the highest dose of endotoxin. Consistent with previous studies, levels of SAA1/2 were markedly higher in infl ammatory HDL of the WT mice, representing ‫ف‬ 30% of the protein mass. Moreover, the ability of infl ammatory HDL isolated from WT mice to promote sterol effl ux from J774 macrophages was markedly impaired, as previously reported by other investigators ( 21,22 ). In contrast, HDL isolated from infl amed Saa1 / 2 Ϫ / Ϫ mice was completely protected from the loss of cholesterol effl ux capacity. Plasma lipids, HDL-C levels, and HDL lipid composition changed to the same extent in infl amed WT and Saa1 / 2 Ϫ / Ϫ mice. Collectively, our observations suggest that SAA enrichment impairs the sterol effl ux capacity of HDL in both humans and mice by mechanisms independent of altered lipid composition. However, changes in specifi c lipid subclasses, such as phosphatidylserine, might also contribute to impaired sterol effl ux capacity during infl ammation. In vivo, it is likely that factors other than SAA1/2 associated with infl ammation also reduce HDL's ability to inhibit atherogenesis ( 49,58,59 ). For example, infl ammation downregulates expression of apoA-I in the liver, lowering its circulating levels. It can also alter HDL's lipid composition by changing levels of plasma triglycerides, CETP, secretory phospholipase A2, and other lipidmetabolizing enzymes ( 36,49,58,60,61 ). Such enzymes are important for regulating HDL's ability to accept cholesterol by both the ABCA1 and ABCG1 pathways. Altered levels of lipids and anti-infl ammatory and antioxidant proteins, in concert with the gain of SAA and perhaps other infl ammatory proteins, may thus be key factors that deprive HDL of its cardioprotective functions. In addition to altered lipid composition, other factors such as HDL protein and lipid oxidation may also contribute to impaired sterol effl ux capacity during infl ammation ( 10,62 ).
The impact of SAA on sterol effl ux capacity is controversial. Early studies showed that lipid-free SAA2 promotes sterol effl ux by the ABCA1 pathway ( 63 ). However, virtually all circulating SAA1/2 is associated with HDL in humans and mice ( 32,64,65 ), and there is no convincing evidence that free SAA1 or SAA2 exists in plasma or blood. A recent study of WT and Saa1 / 2 Ϫ / Ϫ mice suggested that SAA does not contribute to impaired reverse cholesterol transport during infl ammation in vivo ( 66 ). This study used macrophages containing radiolabeled cholesterol to assess the impact of SAA on HDL function. Because the radiolabeled macrophages were injected into the animals' peritoneum only 4 h after endotoxin treatment, while circulating SAA levels in HDL generally peak at approximately 24 h after treatment, it is possible that the ability of HDL to promote effl ux from macrophages, the initial step of reverse cholesterol transport, was not affected by elevated associated with elevated risk of cardiovascular disease (e.g., systemic lupus erythematosus, rheumatoid arthritis) ( 76 ) or acute coronary syndrome ( 77 ).
Infl ammatory HDL that is enriched in SAA1 and SAA2 is also depleted in specifi c proteins, including several that may be cardioprotective including apoA-I ( 78 ). Moreover, we found that levels of PON1 and apoA-II associated with HDL's cholesterol effl ux capacity. Collectively, our observations provide strong support for the proposal that SAA alters HDL's biological effects by replacing cardioprotective proteins.
In summary, our observations have identifi ed a specifi c mechanism, enrichment of the HDL proteome with SAA1/2, which alters HDL's functionality. It is noteworthy that both infl ammation and elevated levels of SAA strongly associate with an increased risk of cardiovascular disease in humans ( 79,80 ). These fi ndings may have important implications for understanding how HDL helps prevent CAD and for developing HDL therapeutics that increase its cardioprotective effects.