Diabetes Impairs Cellular Cholesterol Efflux from ABCA1 to Small HDL Particles

Abstract

Rationale:

High-density lipoprotein (HDL) may be cardioprotective because it accepts cholesterol from macrophages via the cholesterol transport proteins ABCA1 and ABCG1. The ABCA1-specific cellular cholesterol efflux capacity (ABCA1 CEC) of HDL strongly and negatively associates with cardiovascular disease (CVD) risk, but how diabetes impacts that step is unclear.

Objective:

To test the hypothesis that HDL’s cholesterol efflux capacity is impaired in subjects with type 2 diabetes.

Methods and Results:

We performed a case-control study with 19 subjects with type 2 diabetes and 20 control subjects. Three sizes of HDL particles, small-HDL, medium-HDL and large-HDL, were isolated by high-resolution size exclusion chromatography from study subjects. Then we assessed the ABCA1 CEC of equimolar concentrations of particles. Small-HDL accounted for almost all of ABCA1 CEC activity of HDL. ABCA1 CEC—but not ABCG1 CEC—of small-HDL was lower in the subjects with type 2 diabetes than the control subjects. Isotope dilution tandem mass spectrometry demonstrated that the concentration of serpin family A member 1 (SERPINA1) in small-HDL was also lower in subjects with diabetes. Enriching small-HDL with SERPINA1 enhanced ABCA1 CEC. Structural analysis of SERPINA1 identified 3 amphipathic α-helices clustered in the N-terminal domain of the protein; biochemical analyses demonstrated that SERPINA1 binds phospholipid vesicles.

Conclusions:

The ABCA1 CEC of small-HDL is selectively impaired in type 2 diabetes, likely because of lower levels of SERPINA1. SERPINA1 contains a cluster of amphipathic α-helices that enable apolipoproteins to bind phospholipid and promote ABCA1 activity. Thus, impaired ABCA1 activity of small HDL particles deficient in SERPINA1 could increase CVD risk in subjects with diabetes.

Graphical Abstract

INTRODUCTION

The prevalence of type 2 diabetes mellitus is increasing worldwide, and atherosclerotic cardiovascular disease (CVD) remains the leading cause of death in that population ^1–4. Previous studies have shown that subjects with type 1 or type 2 diabetes mellitus have a two-to-four-fold higher risk of CVD than people without diabetes ^{2, 5} and, even with statin therapy, their risk remains high ^6–7. Also, patients with type 2 diabetes often suffer from diabetic dyslipidemia—elevated levels of plasma triglycerides (TG) and low levels of high-density lipoprotein cholesterol (HDL-C) ^8–9—which strongly associates with the risk of CVD ¹⁰.

A low level of HDL-C strongly and inversely correlates with CVD risk ^11–13. However, the failure of HDL-C elevating interventions to lower cardiac events in statin-treated subjects suggests that the association between HDL-C levels and CVD risk may be indirect and that HDL may not play a causal role in cardiovascular disease ^14–21. It is therefore critical to identify new HDL metrics that better predict CVD risk and lead to potential therapeutic targets.

Many lines of evidence indicate that one of HDL’s cardioprotective functions is to mediate reverse cholesterol transport from macrophages, a key cellular component of the atherosclerotic lesion ^22–26. Two steps in this pathway involve ATP-binding cassette transporters in the cell membrane—ABCA1 and ABCG1—that are highly induced when macrophages accumulate excess cholesterol ²⁷. First, ABCA1 mediates cholesterol efflux from macrophages to lipid-poor apolipoproteins ²⁸ and small dense HDL ²⁹. ABCG1 then transports cellular cholesterol to the more mature lipidated HDL particles generated by ABCA1 ^30–31. Humans deficient in ABCA1 accumulate cholesterol-loaded macrophages in many different tissues ³²; and cholesterol efflux from macrophages is also severely impaired when ABCA1 is deleted ³³. Moreover, recent studies demonstrated that the ABCA1-specific cholesterol efflux capacity (ABCA1 CEC) of serum HDL (serum depleted of APOB-containing lipoproteins) inversely associates with both incident and prevalent CVD ^34–36. Importantly, that association was independent of HDL-C levels. These results strongly suggest that HDL’s cholesterol efflux capacity is a critical contributor to its anti-atherogenic effects in humans.

The ABCA1 CEC of serum HDL and isolated HDL correlate highly, indicating that altered HDL function is likely a key factor in impaired ABCA1 CEC ³⁷. Biochemical and genetic studies of mice strongly implicate apolipoprotein A1 (APOA1), the main structural protein of HDL, and APOE as key mediators of ABCA1 CEC ^38–39. However, the molecular factors that control HDL’s CEC are still not fully understood. Moreover, HDL is a heterogeneous population of particles, with sizes ranging from 7 nm to 14 nm, that collectively contain >80 different proteins ^40–41. Du et al. recently used reconstructed HDL and human HDL isolated by ultracentrifugation to demonstrate that small, dense HDLs are the most efficient lipidated mediators of cholesterol efflux by the ABCA1 pathway ²⁹. However, it is still not clear how diabetes affects the function of small HDL.

In the current studies, we tested the hypothesis that HDL’s function is impaired in subjects with type 2 diabetes relative to control subjects without diabetes. We found that ABCA1 CEC was significantly lower in small HDL (S-HDL) of subjects with type 2 diabetes than in that of control subjects. Furthermore, levels of SERPINA1 (serpin family A member 1, also termed α1-antitrypsin [ATT]) was also significantly lower in S-HDL of subjects with type 2 diabetes, and enriching S-HDL with SERPINA1 increased its ABCA1 CEC. Our observations suggest that ABCA1 CEC, the first key step in cholesterol export from macrophages, is impaired in diabetes when S-HDL contains subnormal levels of SERPINA1.

METHODS

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Experimental design.

Subjects were selected from the Carotid Lesion Epidemiology and Risk (CLEAR) study ^42–43 that was approved by institutional review boards at the University of Washington, Virginia Mason Medical Center, and Veterans Affairs Puget Sound. The CLEAR study is a Seattle-based prevalent carotid artery disease case-control study composed primarily of veterans. All subjects were collected through an Epidemiology Research and Information Center project at the Puget Sound Veterans Affairs Health Care System from 9/20/96 to 10/17/11. Censored age matching was based on the age at the time of the blood draw for controls and the age at the diagnosis of vascular disease for cases. Subjects gave written informed consent. All plasma samples were deidentified, randomized, analyzed in a blinded manner, and contained no information that might allow for the identification of individuals. Exclusion criteria were: total cholesterol >400 mg/dl, known coagulopathy, or unable/willing to consent. All subjects underwent carotid ultrasound and were accepted only if they had <10% carotid stenosis bilaterally. Subjects with atherosclerosis-related diagnoses were excluded. All subjects were on statin therapy. None was on triglyceride-lowering therapy.

Diabetic status was defined as HbA1c level >6.5% or use of insulin or antidiabetic medications. We studied 19 subjects with type 2 diabetes and 20 controls. The two groups were similar in age, gender, and HDL-C, LDL-C, and triglyceride levels (Table 1).

Table 1.

Clinical characteristics of study subjects

Characteristic	Control (n=20)	Diabetic (n=19)	P value
Age (years)	61±6	63±7	0.35
Female/Male (n)	2/18	4/15	0.34†
HDL-C (mg/dL)	50±10	46±7	0.14
LDL-C (mg/dL)	110±34	99±37	0.36
Triglycerides (mg/dL)	118±49	130±39	0.40
HbA1c (%)	5.5±0.4	7.2±1.2	4.0×10−6
Total-HDL (µM)	14.9±0.52	13.8±0.4	0.13
S-HDL (µM)	5.2±0.4	5.5±0.4	0.64
M-HDL (µM)	7.1±0.5	6.3±0.4	0.18
L-HDL (µM)	2.6±0.3	2.1±0.2	0.17

^†Chi-square test

HDL isolation.

Plasma was quickly thawed at 37°C, and subjected to sequential ultracentrifugation (UC) to isolate HDL (density 1.063–1.210 g/mL) ⁴⁴. The protein concentration in HDL was measured using the Bradford assay, with albumin as the standard.

Ultracentrifuge-isolated HDL (total-HDL) was separated (0.1 mL/min) by sequential fractionation on two tandem Superdex 200 Increase 5/150 GL columns at 4°C in 150 mM ammonium acetate buffer with an ÄKTA 10 fast protein liquid chromatography system (GE Healthcare Bio-Sciences, Pittsburgh, PA). Fractions containing APOA1 (200 μL) were collected and combined to obtain different sizes of HDL (Fig. I): fractions 12–13 for L-HDL (11.9±0.5 nm, mean ± standard deviation); fractions 14–15 for M-HDL (9.8±0.4 nm); fractions 16–18 for S-HDL (8.1±0.1 nm).

HDL Particle Concentration (HDL-P_ima).

The sizes and concentrations of HDL particles were quantified as total HDL (Total-HDL), small HDL (S-HDL), medium HDL (M-HDL), and large HDL (L-HDL) by calibrated ion mobility analysis ⁴⁵. HDL particles were separated by size in the gas phase and quantified by particle counting. HDL particle counts were converted into aqueous particle concentrations with glucose oxidase calibration curves. The method yields a stoichiometry of APOA1 and sizes and relative abundances of HDL subspecies that are in excellent agreement with those determined by non-denaturing gradient gel electrophoresis and analytical ultracentrifugation ^45–46 .

Cholesterol Efflux Capacity (CEC).

ABCA1 CEC and ABCG1 CEC of isolated HDLs were quantified using BHK (baby hamster kidney) cells with mifepristone-inducible human ABCA1 or ABCG1 ²⁸. BHK cells were cultured in DMEM/high glucose media with 4.00 mM L-glutamine (Cytiva, Marlborough, MA) plus 10% fetal bovine serum (FBS), 100 units/mL of penicillin/streptomycin (Cytiva, Marlborough, MA). Cells were radiolabeled by incubation with [H³]cholesterol for 24 h, washed twice, and then incubated for 4 h with 30 pmol of S-HDL, M-HDL, L-HDL, or Total-HDL in 250 μL of medium ⁴⁷. ABCA1-specific and ABCG1-specific CEC were calculated as the percentage of total [H³]cholesterol (medium plus cell-associated) released into the medium of BHK cells stimulated with mifepristone minus that of cells stimulated with medium alone.

Targeted Protein Quantification by Selected Reaction Monitoring (SRM).

Details are provided in the Supplemental Materials.

Incubation of SERPINA1 with S-HDL.

S-HDL was isolated from pooled HDL (five healthy adult subjects), using two tandem Superdex 200 Increase 5/150 GL columns as described above. S-HDL (1.2 mg/mL protein) was incubated with human SERPINA1 at molar ratios of 0.25:1, 0.5:1, and 1:1 in phosphate-buffered saline (PBS) at 37°C for 1 h and then re-isolated by ultracentrifugation (density 1.063–1.210 g/mL). SERPINA1 concentrations used during incubation were 0.25-1 mg/mL, which are lower than or close to SERPINA1 levels in human plasma (1-3 mg/mL). About 50% of APOA1 was recovered by ultracentrifugation of the reaction mixture. MS/MS analysis of re-isolated S-HDL and SERPINA1-enriched HDL demonstrated no difference in their relative amounts of APOA1. Furthermore, SERPINA1 levels in SERPINA1-enriched HDL were in the lower range of those of HDL isolated by ultracentrifugation from human plasma. There was no impact of SERPINA1 supplementation on the size of the re-isolated S-HDL as determined by calibration ion mobility analysis (Fig. II).

Identification of Potential Amphipathic α-Helical Sequences with Lipid-Binding Affinity.

The LOCATE algorithm uses termination rules to identify potential amphipathic α-helices in proteins ^48–51. WHEEL is a complementary algorithm ^48–50 that creates Schiffer-Edmundson helical wheel diagrams ⁵². WHEEL uses the helical wheel diagram to calculate a parameter of lipid affinity for each sequence, termed Λα ⁵⁰, which increases with increasing lipid affinity. Both algorithms are based on a minimum of 16 sequential amino acid residues. LOCATE followed by WHEEL was used to identify putative amphipathic α-helical regions in the sequences of SERPINA1 and APOA1.

Phospholipid Binding of SERPINA1.

Phospholipid binding was measured with the biotin capture lipid affinity assay we previously described ⁵³. Briefly, small unilamellar vesicles (SUVs) were made with 99% POPC (1-palmitoyl,2-oleoyl phosphatidylcholine), 0.5% BT-PE ((biotinyl)-1,2-dipalmitoyl phosphatidylethanolamine), and 0.5% Rhod-PE ((N-lissamine rhodamine B sulfonyl)-1,2 dioleoyl phosphatidylethanolamine) (w/w) ⁵³. SUVs (43 µg of phospholipid) were incubated with SERPINA1 in a series of concentrations (0, 10, 25, 50, 100, 150 and 225 µg/mL) in PBS (100 µL) for 30 min at room temperature. Samples were then incubated for 30 min on immobilized streptavidin columns in a 96-well filter plate. The columns were washed 3 times with 100 µL PBS. Bound protein and PL were eluted from the columns, using PBS with 1% sodium cholate (w/v). The eluted PL was quantified by fluorescence (excitation of rhodamine B fluorophor at 550 nm with monitoring emission at 590 nm). Protein was quantified by a 3-(4-carboxybenzoyl)quinoline-2-carboxaldehyde (CBQCA) fluorescence protein assay kit (Invitrogen, Carlsbad, CA).

Statistical Analyses.

All statistical analyses were performed with STATA software version 12 (Stata Corp, College Park, TX). Power calculations were based on data from previous analyses and indicated that 15 subjects per group would give 80% power to detect a 15% difference in CEC and 90% power to detect a 30% difference in protein abundance (two-sided α=0.05). To compare groups, we used the chi square test for categorical values and the two-sample t-test for normally distributed continuous variables. One-way ANOVA was used to compare the means of three or more groups. Multivariate linear regression models with an interaction term (diabetic status x lipid component) were used to investigate the association of efflux with each lipid component after controlling for diabetic status. A P value of 0.05 was the threshold for significance. Unless otherwise stated, values represent means ± standard deviations.

For proteomic analyses, differences in HDL-associated protein expression between control subjects and subjects with type 2 diabetes were first calculated by the two-sample t-test. To account for multiple comparisons, P values obtained from the t-test were compared with its Benjamini-Hochberg critical value q, (i/m)Q, where I is the rank, m is the total number of tests, and Q is the false discovery rate (15%) ^54–55. Protein levels were considered significantly different when the P value from the t-test was lower than or equal to its q value ^54–55.

RESULTS

Study design.

To test the hypothesis that CEC from ABCA1 to HDL particles is impaired in subjects with type 2 diabetes, we performed a case-control study with 39 subjects from the CLEAR study ^42–43. Subjects meeting our study criteria (Methods) were randomly selected. The diabetic (n=19) and control (n=20) groups had similar ages and percentages of females (Table 1). Mean hemoglobin A1c levels were 7.2% for the cases and 5.5% for the controls (P=4.0×10⁻⁶). The two groups showed no significant differences in HDL-C, LDL-C, or plasma triglyceride levels, or in total-HDL, S-HDL, M-HDL, or L-HDL (Table 1). All the subjects were on statin therapy.

An overview of the study’s workflow is shown in Figure 1. HDL was isolated by ultracentrifugation from freshly thawed plasma (stored at −80 °C after collection), and further separated by high-resolution size exclusion chromatography into three fractions: S-HDL (8.1 nm±0.1nm), M-HDL (9.8±0.4 nm), and L-HDL (11.9±0.5 nm). For each isolated fraction as well as for total-HDL, we used calibrated ion mobility analysis to quantify HDL particle concentration and size (HDL-P_ima) ⁴⁵. Examples of size distributions of S-HDL, M-HDL, and L-HDL are shown in Figure I (Supplemental Material).

Figure 1.

Study design.

All subsequent analyses were performed using the same concentration of HDL particles in each assay. Therefore, the composition (triglycerides, phospholipids, cholesterol, cholesteryl ester and total protein) of the HDL particles as well as CEC and quantitative proteomics were based on equal molar particle concentrations. In the proteomic analyses, we determined relative quantities of 30 proteins that had been consistently detected in HDL in previous studies ^{40, 56–57} using SRM and isotope dilution with ¹⁵N-labeled APOA1 as the internal standard.

Small HDL particles are the major lipidated acceptors of cholesterol from the ABCA1 pathway in both diabetic and control subjects.

We first tested the capacity of HDL particles of different sizes to promote ABCA1- and ABCG1-specific CEC independently of diabetic status. ABCA1 CEC and ABCG1 CEC were measured using BHK cells with or without expression of mifepristone-inducible human ABCA1 or ABCG1. The cells were exposed to equal concentrations of S-HDL, M-HDL, and L-HDL particles based on HDL-P_ima. Regardless of diabetic status, the S-HDL fraction was the major contributor to ABCA1 CEC (Fig. 2A), which was much less efficient with M-HDL and, especially, L-HDL particles. The differences among the ABCA1 CECs of the three sizes of HDL particles were significant after adjustment for diabetic status (P=6×10⁻¹⁰, P=6×10⁻¹⁰ and P=2.1×10⁻⁴ from one-way ANOVA, Tukey-Kramer post-test). In contrast, L-HDL and M-HDL particles were the major acceptors of cholesterol from the ABCG1 pathway (Fig. 2B). After adjustment for diabetic status, both M-HDL and L-HDL promoted significantly more efflux through the ABCG1 pathway than did S-HDL (P=2.1×10⁻⁴ and P=1.2×10⁻⁶, respectively). There was no significant difference between the ability of M-HDL and L-HDL to promote ABCG1 efflux (P=0.082).

Figure 2. ABCA1-mediated and ABCG1-mediated cholesterol efflux capacity (CEC) of different sizes of HDL particles.

Box plots of ABCA1 CEC (A) and ABCG1 CEC (B) for different sizes of HDL particles in the entire study population (n=39 for S-HDL, M-HDL and Total-HDL; n=18 for L-HDL). HDL was isolated by sequential ultracentrifugation and high-resolution size exclusion chromatography. ABCA1 CEC and ABCG1 CEC were measured after a 4-h incubation with HDL (120 nM), using mifepristone-stimulated BHK cells and BHK cells stimulated with medium alone. Efflux was calculated as the % of radiolabeled cholesterol in the medium relative to the total radiolabeled cholesterol content of the medium and cells. CEC is based on equal particle concentrations. The central rectangle of the box plot is the first quartile to the third quartile (interquartile range) range of the data. A segment inside the rectangle shows the median, and “whiskers” above and below the box are the minimum and maximum values. Outliers (●) are values >1.5 times the interquartile range; they were included in the statistical analysis. P value: one-way ANOVA with Tukey-Kramer post-test, controlling for diabetic status.

Small HDL particles from subjects with type 2 diabetes exhibit defective ABCA1 CEC.

S-HDL particles from subjects with type 2 diabetes were less able to promote efflux through the ABCA1-specific pathway than those of control subjects (Fig. 3A; 7.6 % efflux vs 9.9 % efflux from controls, P=0.028). In contrast, the ABCA1 CEC of the M-HDL and L-HDL particles of the subjects with type 2 diabetes did not differ significantly from that of the controls (Fig. 3A; P=0.52 for M-HDL and 0.86 for L-HDL). Moreover, all three sizes of HDL particles from the subjects with type 2 diabetes had similar ABCG1 CECs when compared with particles from the controls (Fig. 3B). Importantly, there were no significant differences in total HDL particle concentration or S-HDL, M-HDL, or L-HDL particle concentration between control and subjects with type 2 diabetes (Table 1).

Figure 3. Box plots of ABCA1 CEC (A) and ABCG1 CEC (B) for different sizes of HDL particles in subjects with and without diabetes.

All analyses were based on equal particle concentrations (n=20 for control and n=19 for subjects with type 2 diabetes in S-HDL, M-HDL and Total-HDL; n=12 for control and n=6 for subjects with type 2 diabetes in L-HDL). ABCA1 CEC and ABCG1 CEC (mifepristone-stimulated BHK cells minus BHK cells incubated in medium alone) were measured after a 4-h incubation with HDL as indicated in the legend to Fig. 1. P value, two-tailed Student’s t-test.

Diabetic status significantly affects triglyceride levels in S-HDL.

To investigate potential determinants of ABCA1 CEC, we measured the lipid composition of S-HDL in diabetic and control subjects (Fig. 4). All of the lipid levels were normally distributed (P>0.05; Shapiro-Wilk test). Triglyceride levels in S-HDL particles were significantly lower in the subjects with type 2 diabetes than in the controls (P=0.004) (Fig. 4A). However, no significant differences were found for the other lipids: free cholesterol content (Fig. 4B, P=0.34), phospholipids (Fig. 4C, P=0.35), and cholesteryl ester (Fig. 4D, P= 0.22). S-HDLs from the control and subjects with type 2 diabetes contained similar amounts of protein (73.3±12.8 mg/µmol particle in control subjects and 69.3±10.9 mg/µmol particle in subjects with diabetes).

Figure 4. Triglycerides (A), free cholesterol (B), phospholipids (C) and cholesteryl ester (D) of S-HDL particles in subjects with and without diabetes.

The box plots show the distribution of the data (median, interquartile ranges), while the dots represent outliers (n=20 for control and n=19 for subjects with type 2 diabetes). Outliers were included in the statistical analysis.

ERPINA1 and APOC2 levels are altered in diabetic S-HDL, and levels of both proteins correlate with ABCA1 CEC.

To further investigate the mechanism whereby diabetes might impair S-HDL function, we quantified HDL proteins using MS/MS analysis with selected reaction monitoring, a sensitive and precise method for quantitative proteomics analysis in complex biological materials ^58–60. We monitored the 30 most abundant HDL proteins. All proteins had been identified previously in at least two different human populations ^{40, 56–57} (Table I).

After controlling for multiple comparisons, we found that levels of two proteins—SERPINA1 and APOC2—differed significantly in S-HDL from subjects with type 2 diabetes (Fig. 5). The mean level of SERPINA1 was 35% lower in the subjects with type 2 diabetes than in the control subjects (P=0.0034, Fig. 5A). This deficit was unlikely to be explained by reduced plasma levels of SERPINA1, which did not differ between control and subjects with type 2 diabetes (Fig. IIIA). Moreover, SERPINA1 levels in S-HDL did not correlate with those in plasma (Fig. IIIB), indicating that differences in plasma levels of SERPINA1 did not account for the differences in the SERPINA1 levels in S-HDL of control and subjects with type 2 diabetes. The SERPINA1 level in S-HDL showed a significant positive correlation with ABCA1 CEC (Fig. 5C, r=0.61, P=4×10⁻⁵).

Figure 5. Box plots of SERPINA1 (A) and APOC2 (B) in subjects with and without diabetes.

Peptides were quantified by isotope dilution MS/MS as the integrated peak area relative to that of [¹⁵N]APOA1 peptide (n=20 for control and n=19 for subjects with type 2 diabetes). P values were calculated from log transformed data. Scatter plots show the correlation of MS/MS analysis and ABCA1-specific efflux for SERPINA1 (C) and APOC2 (D). r represents the Pearson correlation. AU, arbitrary units. ● Control, ○ Diabetic.

In contrast, the mean level of APOC2 was higher in the subjects with type 2 diabetes than in the control subjects (P=0.011, Fig. 5B), while its level in S-HDL correlated negatively with ABCA1 CEC (r=−0.33, P=0.043, Fig. 5D). SERPINA1 levels, but not APOC2 levels, correlated significantly and positively with the triglyceride level in S-HDL (r= 0.40, P=0.011, Fig. IV).

SERPINA1 promotes ABCA1 CEC of small HDL particles.

To determine whether SERPINA1 and APOC2 might regulate the ABCA1 CEC of S-HDL particles, we incubated S-HDL isolated by high-resolution size exclusion chromatography with purified SERPINA1 or APOC2 and then re-isolated the lipoproteins by ultracentrifugation. The impact of enriching S-HDL with APOC2 on ABCA1 CEC was variable, perhaps because the apolipoprotein dissociated from S-HDL during ultracentrifugation or displaced APOA1 from S-HDL. In contrast, MS/MS analysis demonstrated that levels of SERPINA1 in re-isolated HDL consistently increased when the lipoprotein was incubated with increasing amounts of the protein (Fig. 6A) without changing the size of the S-HDL particles (Fig. II).

Figure 6. Impact of S-HDL-associated SERPINA1 and free SERPINA1 on ABCA1 CEC.

S-HDL particles were incubated with the indicated concentrations of SERPINA1 (mol/mol particle) for 1 h. HDL was re-isolated by ultracentrifugation, and the relative concentrations of SERPINA1 (A) and ABCA1 CEC (B) were determined by MS/MS analysis and with BHK cells expressing ABCA1 (mifepristone-stimulated BHK cells minus BHK cells incubated in medium alone), respectively. CEC is based on equal particle concentrations. SERPINA1 in panel B represents ABCA1 CEC of lipid-free human SERPINA1 (4 µg/mL). Results are means with standard deviations (n=4, technical replicates) and are representative of those found in 3 independent experiments. AU, arbitrary units. To compare means, 1-way ANOVA with Tukey-Kramer post-test was used in B: #P=0.020, ##P=0.011.

Moreover, enrichment of S-HDL particles with physiological levels of SERPINA1 promoted ABCA1-specific cholesterol efflux from BHK cells. Stimulation was half-maximal at 0.25 mol SERPINA1/mol S-HDL during the incubation; it plateaued at approximately 0.50 mol SERPINA1/mol S-HDL (Fig. 6B). No free APOA1 was detected by immunoblot analysis in the medium of BHK cells incubated with S-HDL or SERPINA1-enriched S-HDL (Fig. V), and the APOA1 contents of the two lipoproteins were indistinguishable by MS/MS analysis, strongly suggesting that SERPINA1 was not displacing APOA1 from HDL.

Biotin labeling followed by MS/MS analysis demonstrated similar levels of ABCA1 protein in plasma membranes isolated from BHK cells exposed to S-HDL or SERPINA1 enriched S-HDL (Fig. VI). S-HDL and SERPINA1-enriched HDL also inhibited TNF-α-stimulated mRNA levels of inflammatory genes ICAM1 and VCAM1 in human endothelial cells to a similar degree (Fig. VII). Collectively, these observations strongly suggest that the increased ABCA1 CEC we found with SERPINA1 enrichment of HDL was not due to altered levels of free or HDL-associated APOA1, increased cell-surface expression of ABCA1 protein, or altered anti-inflammatory properties of S-HDL.

SERPINA1 contains amphipathic α-helices.

Amphipathic α-helices are key mediators of lipid binding in apolipoproteins and membrane proteins ⁴⁸, and they play a critical role in modulating the ABCA1 activity of lipid-free APOA1 ²⁸. We used the LOCATE and WHEEL algorithms ⁵⁰ to identify potential amphipathic α-helical lipid-binding domains in SERPINA1. Lipid affinity, termed Λα ⁵⁰, was calculated for each sequential sequence containing a minimum of 16 amino acids. Based on the analysis of known lipid-binding amphipathic α-helices in apolipoproteins, Λα lipid affinity >6.0 was defined as a potential lipid-binding domain. In SERPINA1, we identified 7 sequences with Λα values >6.0 (Fig. 7A). Analysis of APOA1 (Fig. 7B), which binds phospholipids via α-helical domains in the C- and N-terminal domains ^{49, 61}, identified 4 potential amphipathic α-helical sequences in the C- and N-terminal domains with Λα values >6.0. All of the identified amphipathic α-helical sequences identified in APOA1 are α-helical sequences in the crystal structure of APOA1 ⁴⁹ and are known to mediate phospholipid binding ⁶¹.

Figure 7. Potential lipid-binding amphipathic α-helices in SERPINA1 (A) and APOA1 (B) identified with the LOCATE and WHEEL algorithms.

LOCATE uses termination rules to identify potential amphipathic α-helices in proteins ^48–50. WHEEL calculates a parameter of lipid affinity for each amino acid sequence (termed Λα) ⁵⁰. Based on the analysis of known α-helices with lipid-binding affinity ⁵⁰, sequences with Λα lipid affinity >6.0 (dotted line) were considered potential lipid binding domains. The locations of each putative amphipathic α-helix in the two sequences are denoted by dark lines. (A) Location of α-helical lipid-binding domains in the amino acid sequence of SERPINA1. Seven potential binding domains were identified, each of which is mapped as a single sequence. (B) Location of α-helical lipid-binding domains in the amino acid sequence of APOA1. (C) Crystal structure of SERPINA1. The locations of the seven putative amphipathic helical domains with lipid-binding affinity identified with WHEEL and LOCATE in (A) are mapped in blue in the cartoon. The remainder of the SERPINA1 structure is mapped in CPK colors, using the ribbon format. Four of the putative amphipathic α-helices mapped onto α-helical regions in the crystal structure (1, 2, 3, 6) and 3 of these (1, 2, 3) were clustered at the N-terminus of the protein.

We mapped the locations of the 7 potential lipid-binding domains identified by LOCATE and WHEEL on the crystal structure of SERPINA1 (Fig. 7C). Four were in α-helical domains, and 3 of those were clustered at the N-terminal end of the protein’s globular structure. Taken together, these observations raise the possibility that amphipathic α-helical domains of SERPINA1 mediate lipid binding to HDL.

SERPINA1 binds phospholipid vesicles.

We used phospholipid SUVs⁵³ to investigate the lipid-binding properties of lipid-free SERPINA1 (Fig. 8). This approach demonstrated that SERPINA1 binds SUVs. Its apparent binding affinity was ~22% that of APOA1, but it had a much higher apparent maximal binding capacity. In contrast to SERPINA1 and APOA1, Rnase A (which lacks amphipathic α-helical sequences) failed to bind phospholipid SUVs (Fig. 8). These observations demonstrate that SERPINA1 is a phospholipid-binding protein.

Figure 8. Phospholipid-binding profiles of SERPINA1, APOA1, and Rnase.

Lipid-binding of proteins (43 µg) were assessed in 100 μL of PBS for 30 min at 25°C, using SUVs (99% POPC/0.5% BT-PE/0.5% Rhod PE [w/w]). Values are the mean of three independent replicates. Results are representative of two independent experiments. Error bars are ±1 SD.

DISCUSSION

Because pharmacological interventions that raise HDL-C have failed to reduce CVD risk in clinical trials, it is critical to identify new metrics that reflect HDL’s biological roles in humans ^{11–13, 62–65}. One proposed functional metric is ABCA1 CEC, which predicts incident and prevalent CVD in humans independently of HDL-C ^34–36. We tested the hypothesis that HDL particles are functionally impaired in subjects with type 2 diabetes, which might contribute to increased CVD risk in these subjects. Except for diabetic status, the clinical characteristics of the control and case subjects we studied were very similar. Importantly, we always compared HDL function on the basis of equimolar concentrations of particles_. When HDL concentration is based on other metrics (e.g. protein content), the absolute concentration of the particles can vary widely for the different size classes of HDL.

We found that the ABCA1 CEC of S-HDL was selectively impaired in subjects with type 2 diabetes. The level of impairment (~23%) was much greater than that seen in subjects with prevalent and incident CVD (~9%) ^{35–36, 66}. Because S-HDL particle concentrations were similar in patients with diabetes and controls, these finding suggest a functional impairment of S-HDL in subjects with type 2 diabetes that reduces ABCA1 CEC. In contrast, there was no difference in the ABCG1 CEC of S-HDL, M-HDL, or L-HDL between subjects with and without diabetes.

We also found that S-HDL particles are the major contributors to ABCA1 CEC. These observations raise the possibility that S-HDL particles in patients with diabetes are unable to effectively promote efflux by the ABCA1 pathway, the first key step in reverse cholesterol efflux from macrophages in the artery wall.

These observations confirm and extend the work of Du et al., who demonstrated that small, dense HDLs were the most efficient mediators of cholesterol efflux by the ABCA1 pathway ²⁹. However, that group based its analyses on equal protein concentrations, using reconstituted HDL particles and HDL isolated by ultracentrifugation of plasma from healthy humans. Because the stoichiometry of APOA1 ranges from 2–6 in the different sizes of HDL particles ⁶⁷, this approach did not allow the investigators to determine the specific activity of each particle size (i.e., ABCA1 CEC based on the molar concentration of the particles) or the contributions of the different sizes of HDL to the CEC of total-HDL.

To determine why small HDL particles are dysfunctional in humans with diabetes, we quantified levels of proteins and lipids in S-HDL. MS/MS analysis showed that levels of SERPINA1 were significantly lower in the S-HDL of the subjects with diabetes. Diabetes associates with lower levels of SERPINA1 in plasma ^68–72. In contrast, plasma levels of SERPINA1 were similar in our study of subjects with and without diabetes. Moreover, levels of SERPINA in S-HDL did not correlate with plasma levels of SERPINA1, strongly suggesting that altered levels of plasma SERPINA1 did not account for the differences in SERPINA1 levels we found in S-HDL. Importantly, when we enriched S-HDL with physiological levels of SERPINA1, the ABCA1 CEC of the re-isolated HDL increased, suggesting that SERPINA1 enhances S-HDL’s capacity to promote cholesterol efflux from cells by this pathway and that low levels of SERPINA1 in S-HDL contribute to impaired ABCA1 CEC.

We also found differences in the lipid composition of HDL from subjects with and without diabetes. Triglyceride levels of S-HDL of subjects with diabetes were significantly lower than in control subjects. Levels of SERPINA1 in S-HDL correlated linearly and strongly with the triglyceride content of the particles, suggesting that low triglyceride levels impair the association of SERPINA1 with S-HDL. These observations raise the possibility that lipid composition is another important factor influencing the ABCA1 CEC of S-HDL and that lowering the concentration of triglycerides in HDL particles also might lower SERPINA1 levels.

Because amphipathic α-helices are key mediators of lipid binding in apolipoproteins and membrane proteins, and because they are critical modulators of ABCA1 activity of lipid-free APOA1 ²⁸, we searched for evidence that SERPINA1 contains amphipathic α-helices. The LOCATE and WHEEL algorithms identified 7 sequences in SERPINA1 that could be amphipathic α-helices. Four were located in α-helical domains in the crystal structure of SERPINA1, and three of those were clustered at the N terminal end of the globular protein structure, raising the possibility they could be a lipid-binding domain. Biochemical studies with SUVs demonstrated that SERPINA1 is indeed a phospholipid-binding protein. Its apparent binding affinity was ~22% that of APOA1, but it had a much higher apparent maximal binding capacity.

Because apolipoproteins are known to use amphipathic α-helical sequences to associate with lipid and activate ABCA1 ^{28, 48–50}, our observations strongly suggest that the amphipathic α-helical regions we identified in SERPINA1 enable to the protein to both bind to lipid in S-HDL and modulate ABCA1 activity. The domain may promote ABCA1 CEC by directly interacting with ABCA1. Alternatively, SERPINA1’s amphipathic α-helical helices could alter APOA1’s ability to bind to lipid, in turn enhancing its ability to interact with ABCA1.

SERPINA1 may play other roles in CVD. Gordon et al. have suggested that both its anti-elastase activity and anti-inflammatory functions are enhanced when it binds to HDL ⁷³. As neutrophil infiltration of culprit coronary lesions is implicated in plaque rupture and thrombus formation in acute myocardial infarction ⁷⁴, the lower level of SERPINA1 in S-HDL of subjects with diabetes might contribute to the increased risk of morbidity and mortality from myocardial infarction ⁷⁵ by impairing HDL’s ability to protect tissue from promiscuous proteolysis by serine proteases.

Our study has several limitations. First, the number of subjects was relatively small and the majority were male. It will be important to confirm our results in larger cohorts with more female participants. Second, the size distributions of the isolated HDL subpopulations overlapped to some degree, reflecting the limited resolution of size-exclusion chromatography. Importantly, the mean sizes of S-HDL, M-HDL and L-HDL were well separated. Moreover, there were marked differences in the ABCA1 CEC of the different sizes of HDL particles. These observations strongly support the proposal that the size of HDL has a major impact on both its biological functions and composition ⁴⁶. Finally, it will be important to carry out additional biochemical and cellular studies to identify the functional domains and mechanisms that enable SERPINA1 to enhance ABCA1 CEC in HDL.

In conclusion, we found that the ABCA1 CEC and SERPINA1 of S-HDL were selectively impaired in subjects with diabetes. In contrast, enrichment of S-HDL particles with SERPINA1 increased ABCA1 CEC. Structural and biochemical studies supported the proposal that SERPINA1 contains amphipathic α-helical regions and is a phospholipid-binding protein, suggesting that SEPINA1 binds to HDL phospholipids, altering the lipoprotein’s ABCA1 activity. Thus, the S-HDL produced during diabetes could be unable to remove adequate amounts of cholesterol from the artery wall due to its decreased SERPINA1 content, placing patients at higher risk for CVD.

NOVELTY AND SIGNIFICANCE.

What Is Known?

• Patients with type 2 diabetes often have low levels of high density lipoprotein cholesterol (HDL-C) and a higher risk of cardiovascular disease (CVD) than non-diabetic patients.

• One of HDL’s important cardioprotective functions is likely its cholesterol efflux capacity—the ability to remove cholesterol from macrophage foam cells.

• HDL is a heterogeneous population of particles that vary widely in size and molecular composition.

What New Information Does This Article Contribute?

• Small HDL particles are the primary mediators of cholesterol efflux that involves the ATP-binding cassette transporter A1 (ABCA1).

• Small HDL particles in patients with type 2 diabetes have impaired cholesterol efflux capacity and contain subnormal amounts of serpin family A member 1 (SERPINA1).

• Reconstituting small HDL with SERPINA1 improves cholesterol efflux capacity.

• SERPINA1 binds phospholipids. Three putative amphipathic α-helical domains in the N-terminus of SERPINA1 likely mediate this binding.

• Selective impairment of small HDL’s cholesterol efflux capacity may contribute to increased CVD risk in patients with type 2 diabetes.

HDL may help protect against CVD because it accepts the cholesterol that is removed from arterial wall macrophages by ATP-binding cassette transporter (ABCA1). Type 2 diabetes greatly increases the risk of CVD, but its impact on cholesterol efflux by ABCA1 is poorly understood. We isolated HDL from control subjects and subjects with type 2 diabetes and fractionated the particles into three size classes. We found that small HDL was the major contributor to ABCA1-dependent cholesterol efflux from cells. Moreover, the efflux capacity of small HDL particles was selectively impaired in the patients with type 2 diabetes. Using quantitative proteomics, we showed that those patients had lower SERPINA1 levels in their small HDL particles. Enrichment of small HDL with SERPINA1 markedly improved cholesterol efflux capacity. Structural studies demonstrated that SERPINA1 contains multiple putative amphipathic α-helical domains in its N-terminus that likely mediate its ability to bind to phospholipids and promote cholesterol efflux. Our findings suggest that type 2 diabetes might increase CVD risk by depleting small HDL of SERPINA1, thereby impairing its ability to remove artery wall cholesterol.

Nonstandard Abbreviations and Acronyms:

ABCA1

ATP-binding cassette transporter A1

ABCG1

ATP-binding cassette transporter G1

BHK

baby hamster kidney

CEC

cholesterol efflux capacity

CVD

cardiovascular disease

FC

free cholesterol

HDL-C

high-density lipoprotein cholesterol

HDL-P

HDL particle concentration

HDL-P_ima

HDL-P concentration determined by calibrated ion mobility analysis

LDL-C

low-density lipoprotein cholesterol

L-HDL

large HDL

M-HDL

medium HDL

MS

mass spectrometry

PL

phospholipid

rHDL

reconstructed HDL particles

S-HDL

small HDL

SRM

selected reaction monitoring