HDL-apolipoprotein A-I exchange is independently associated with cholesterol efﬂ ux capacity patient-oriented and epidemiological research

various and HDL Abstract HDL is the primary mediator of cholesterol mobilization from the periphery to the liver via reverse cholesterol transport (RCT). A critical ﬁ rst step in this process is the uptake of cholesterol from lipid-loaded macrophages by HDL, a function of HDL inversely associated with prevalent and incident cardiovascular disease. We hypothesized that the dynamic ability of HDL to undergo remodeling and exchange of apoA-I is an important and potentially rate-limiting aspect of RCT. In this study, we investigated the relationship between HDL-apoA-I exchange (HAE) and serum HDL cholesterol (HDL-C) efﬂ ux capacity. We compared HAE to the total and ABCA1-speciﬁ c cholesterol efﬂ ux capacity of 77 subjects. We found that HAE was highly correlated with both total ( r = 0.69, P < 0.0001) and ABCA1-speciﬁ c ( r = 0.47, P < 0.0001) efﬂ ux, and this relationship remained signiﬁ cant after adjustment for HDL-C or apoA-I. Multivariate models of sterol efﬂ ux capacity indicated that HAE accounted for approximately 25% of the model variance for both total and ABCA1-speciﬁ c efﬂ ux. We conclude that the ability of HDL to exchange apoA-I and remodel, as measured by HAE, is a signiﬁ cant contributor to serum HDL efﬂ ux capacity, independent of HDL-C and apoA-I, indicating that HDL dynamics are an important factor in cholesterol efﬂ ux capacity and likely RCT.

room with chest pain, but did not have angiographically diagnosed coronary artery disease, subjects with diagnosed coronary artery disease (medical history of myocardial infarction (MI), percutaneous coronary intervention (PCI), or coronary artery bypass graft surgery (CABG)) but not acute coronary syndrome, and subjects suffering from acute coronary syndrome [confi rmed MI, both ST-elevation myocardial infarction (STEMI) and nonSTEMI ]. All subjects gave written informed consent that was approved by the University of Washington Institutional Review Board .

Plasma lipid analysis
The lipid panel (total cholesterol, HDL-C, and triglycerides) was measured in fasted EDTA-plasma by standard validated clinical assays on a Beckman DXC, and LDL cholesterol (LDL-C) was calculated using the Friedewald formula. Plasma apoA-I levels were determined by validated nephelometric method.

Cholesterol effl ux assays
Cholesterol effl ux assays were performed essentially as described by Khera et al. ( 13 ). The apoB-containing particles were precipitated from serum using polyethylene glycol and the resulting LDL/VLDL-free serum, representing serum HDL (at 2.8%), was used in the assays. J774 macrophages were labeled with [ ABCA1-specifi c sterol effl ux was performed in BHK cells overexpressing human ABCA1 under mifepristone control. The ABCA1-BHK cells were labeled for 24 h at 37°C in DMEM containing [ 3 H]cholesterol (1 m Ci/ml). Cells were then incubated with or without mifepristone (10 nM) for 24 h, followed by 4 h incubation with serum HDL (2.8%) in DMEM supplemented with 0.1% BSA. Cholesterol effl ux was calculated in mifepristonetreated and untreated cells as described for J774 cells, and the ABCA1-specifi c effl ux was determined as the difference between effl ux in cells treated with and without mifepristone. Each sample was analyzed in duplicate and an average is reported. Both assays were performed on the same day and a control sample was included on each plate. The coeffi cient of variability across all plates was 3.7% for J774 and 4.5% for ABCA1 effl ux.

HAE assay
HAE assay methods were performed with apoB-depleted plasma as described in Borja,et al. ( 31 ), except that the EPR spectrum collection at 6°C was omitted, as analysis of previous data revealed the assay gave the same results when this step was not performed. Spin labeling of apoA-I is described in Oda et al. ( 32 ). Plasma was thawed and mixed 1:4 with PBS [20 mM phosphate, 150 mM NaCl (pH 7.4)] and 24% w/v PEG 6000 (Sigma) was added to a fi nal concentration of 4%. Samples were centrifuged at 13,000 rpm for 10 min at 4°C to remove apoB-containing lipoproteins. The clarifi ed plasma was mixed with 3 mg/ml spin-labeled apoA-I in a 3:1 ratio. Samples were incubated for 15 min at 37°C. EPR measurements were performed with a Bruker eScan spectrometer outfi tted with a temperature controller (Noxygen). The peak amplitude of the nitroxide signal was compared with the peak amplitude of a proprietary internal standard provided by Bruker. HAE was determined by subtracting the sample/internal standard ratio of EPR intensities obtained from spin-labeled apoA-I in PBS from the ratio obtained from the plasma sample enzymes ( 21 ). apoA-I is a highly dynamic protein ( 22 ), and this property is important for both the remodeling of HDL particles (23)(24)(25)(26) and the promotion of cholesterol effl ux ( 27 ). apoA-I structure and function are critical to its anti-atherogenic molecular processes ( 28 ), largely due to its ability to promote cholesterol effl ux from ABCA1 ( 29 ). In vitro studies suggest that the latter may be impaired by oxidative modifi cation of apoA-I, which we hypothesized to act by inhibiting HDL remodeling/exchange of apoA-I ( 24,30 ).
We recently developed a rapid and precise assay employing electron paramagnetic resonance (EPR) spectroscopy that measures the relative rate of HDL-apoA-I exchange (HAE). HAE provides a measure of the ability of HDL to remodel and release lipid-poor apoA-I ( 24,31 ). The assay capitalizes on the high specifi city of apoA-I for HDL and the exchangeable nature of HDL-associated apoA-I, which can be released upon the addition of exogenous lipid-free apoA-I ( 31 ). As spin-labeled apoA-I associates with HDL, the EPR spectra's peak amplitude increases due to structural changes in apoA-I from a lipid-free to a lipid-bound conformation ( 31,32 ). The ratio of lipid-free to lipidbound apoA-I provides a measure of the relative exchangeability of endogenous apoA-I and the dynamic nature of HDL particles. Our early investigation into the clinical relevance of HDL dynamics suggests that HAE is impaired in people with prevalent CVD, and as such it may be a CVDrelevant measure of HDL function ( 31 ).
Despite the widely held conjecture that HDL dynamics are important in RCT, there are limited data on the role of HDL remodeling and dynamics in cholesterol effl ux. In macrophages, cholesterol effl ux is thought to be facilitated by ABCA1, with contributions from ABCG1, scavenger receptor class B type I (SR-BI), and passive diffusion ( 33 ). In turn, the prevailing understanding is that ABCA1mediated cholesterol effl ux is most effi ciently driven by small dense HDL and/or lipid-poor apoA-I ( 29,34 ). In this study, we investigated the relationship between HDL dynamics and cholesterol effl ux capacity in human subjects. We quantifi ed effl ux in both J774 macrophages and ABCA1-transfected baby hamster kidney (BHK) cells (ABCA1-BHK) to assess the relative contribution of HAE to total effl ux and ABCA1-mediated effl ux, respectively. We found that HAE was signifi cantly correlated ( P < 0.0001) with both measures of HDL effl ux capacity, independent of HDL-C and apoA-I, supporting the conclusion that HDL dynamics/remodeling is an important factor in HDL's sterol effl ux capacity.

Study subjects
Plasma samples (n = 77) were collected from subjects undergoing coronary angiography at the University of Washington Medical Center. Coronary angiography was clinically indicated either for ischemic chest pain or as a part of preoperation evaluation for other surgeries in subjects with known coronary heart disease. The cohort included subjects who reported to the emergency indicating that apoB precipitation does not affect this measure of HDL function ( Fig. 1 ). Additionally, we examined the relationship between the two cell-based effl ux assays (J774 versus ABCA1-BHK). They exhibited a high positive correlation with each other ( Fig. 2A ; r = 0.71, P < 0.0001), confi rming ABCA1 as the major contributor to mobilization of cholesterol from J774 cells. HAE had a similarly high positive correlation with effl ux from J774 cells ( Fig. 2B ; r = 0.69, P < 0.0001) and was also correlated to effl ux from ABCA1-BHK cells ( Fig. 2C ; r = 0.47, P < 0.0001). Further, HAE, J774 effl ux, and ABCA1-BHK effl ux all demonstrated signifi cant positive correlations with apoA-I and HDL-C levels ( Fig. 3 ). The signifi cant correlations between the three assays suggest that they represent interdependent aspects of HDL function. To ascertain the degree to which apoA-I and HDL-C determine HAE and cholesterol effl ux, we adjusted the correlations between HAE, J774 effl ux, and ABCA1-BHK effl ux for apoA-I or HDL-C levels. Adjusting for both HDL-C and apoA-I attenuated the relationship between HAE and J774 cholesterol effl ux, but the correlation remained signifi cant ( Table 2 ). The relationship between HAE and ABCA1-BHK effl ux was similarly attenuated, but remained signifi cant following adjustment for HDL-C, and nearly reached signifi cance after adjustment for apoA-I ( Table 2 ; P = 0.068). Taken together, these results suggest that total HDL effl ux capacity, as measured by J774 effl ux, is in part associated with HAE, independent of apoA-I and HDL-C levels; whereas, the relationship between ABCA1-specifi c effl ux capacity and HDL dynamics is independent of HDL-C, but dependent on apoA-I.
To further investigate the factors driving the relationship between sterol effl ux capacity and HAE, we employed multivariate linear regression modeling. We built models predicting sterol effl ux capacity using HAE, controlling for the traditional CVD risk factors of gender, age, HDL-C, apoA-I, LDL-C, triglycerides, diabetes, hypertension, statins, and smoking. We applied step-wise model optimization that systematically eliminated variables that did not contribute to the prediction. We estimated the contribution of individual variables to the model by the lmg algorithm ( 36,37 ). The optimized model for J774 sterol and dividing by the maximum sample/internal standard ratio obtained from spin-labeled apoA-I in a solution of SDS. The inter-assay coeffi cient of variability was 5.3%.

Statistical analysis
All HAE and cholesterol effl ux assays were performed in duplicate and the mean was computed and used for analysis. Associations between different parameters were established by using linear regression and Pearson's correlation coeffi cient. Multivariate regression was used to investigate the predictors of cholesterol effl ux and HAE. Initial multivariate models were developed including the following variables: age, gender, plasma HDL-C, plasma apoA-I, plasma LDL-C, plasma triglycerides (log transformed), and presence of diabetes, hypertension, statins, and smoking. For the model building, all variables except gender, age, and binary variables (diabetes, hypertension, statins, and smoking) were standardized as a z-score. Models were optimized using a backward and forward stepwise approach and selecting the model with a minimum Akaike information criterion (AIC) . Models for both J774 and ABCA1-BHK effl ux were assessed for relative importance of individual variables [R, lmg algorithm in relaimpo package ( 35 )]. The relative importance of each variable is presented as percent of the dependent variable (effl ux) variance explained by the given model. Statistical signifi cance was determined for P < 0.05 for all tests. Statistical analyses were performed using SAS (version 9.3) and R software (version. 3.1).

RESULTS
The clinical characteristics of 77 subjects, along with cholesterol effl ux capacity and HAE, are summarized in Table 1 . Subjects exhibited a spectrum of cardiovascular risk factors (see Materials and Methods), HAE, and effl ux capacities. Due to the small size of the study population, our aim was to use the whole population with the assumption that the disease state would introduce physiologically relevant variance in the measured HDL properties, facilitating modeling of the relationship between HAE and sterol effl ux capacity.
We investigated the effect of apoB depletion on HAE. Plasma from healthy volunteers (n = 10) was tested directly or after PEG-induced precipitation of apoB particles. There was no difference between the two sample preparations,   1. Plasma from healthy individuals (n = 10) was used to confi rm that HAE values are similar whether obtained in apoBdepleted or whole plasma. ApoB depletion was performed as described in the Materials and Methods. Samples with whole plasma were diluted 1:5 in PBS prior to EPR. Each apoB-depleted and non-apoB-depleted pair was measured by EPR, and the data was compared using a two-tailed t -test.
lower HAE ( P = 0.008), this association disappeared after adjustment for HDL-C ( P = 0.31), indicating that HAE is lower in diabetics due to reduced HDL-C. Adjustment for age and sex did not alter this relationship . In summary, multivariate models reveal that HDL dynamics, as measured by HAE, are a signifi cant predictor of cholesterol effl ux capacity from both J774 and ABCA1-BHK cells.

DISCUSSION
Cell-based cholesterol effl ux studies have been instrumental in demonstrating that the ability of HDL to accept cholesterol from lipid-loaded macrophages is linked to prevalent and incident CVD, and that HDL-C and apoA-I levels do not fully refl ect HDL's anti-atherogenic properties ( 13,14,20 ). However, factors controlling HDL sterol effl ux capacity are largely unknown. Our previous work indicated that the ability of HDL to remodel and/or exchange apoA-I is attenuated in CVD subjects with similar apoA-I and HDL-C levels, suggesting that direct measurement of HDL dynamics measures a HDL property distinct from apoA-I or cholesterol concentration ( 31 ). To establish whether HAE and cholesterol effl ux capacity assays measure independent HDL properties, we investigated the relationship between HDL dynamics, as measured by the HAE assay, and cholesterol effl ux capacity, as measured by the cell-based assays. We found that HDL dynamics (HAE) and cholesterol effl ux capacity are strongly associated and are, in part, independent of circulating levels of HDL-C and apoA-I. Multivariate models predicting serum HDL-C effl ux capacity in J774 and ABCA1-BHK cell systems revealed that while HDL-C (and apoA-I) contributes to a large proportion of the sterol effl ux capacity of serum HDL, HAE accounts for as much as 26% of model variance, indicating that HDL dynamics are a major factor in HDL sterol effl ux capacity.
The capacity of HDL to remodel and exchange apoA-I is a critical aspect of HDL quality ( 24,31 ), which is linked to HDL's ability to interact with proteins involved in RCT, including ABC transporters, HDL remodeling proteins [including LCAT, cholesteryl ester transfer protein (CETP), phospholipid transfer protein (PLTP), hepatic lipase, and endothelial lipase] ( 38 ), and potentially SR-BI ( 33 ). In vitro oxidation studies have demonstrated that proteinprotein interactions and dynamics of apoA-I are critical elements in HDL's interaction with these factors (39)(40)(41). HDL remodeling and exchange of apoA-I can be induced when the equilibrium between lipid-associated and lipidfree apoA-I is altered, as when exogenous lipid-free apoA-I is added during the HAE assay ( 24,31 ). The HAE assay is not appreciably affected by apoB depletion of plasma ( Fig. 1 ), supporting the notion that lipid transfer with apoBcontaining lipoproteins does not signifi cantly infl uence HDL to apoA-I exchange under the conditions of the HAE assay. We have previously shown that, within the time frame of the HAE assay, lipid-free apoA-I predominantly exchanges onto smaller HDL particles ( 31 ). Furthermore, in vitro studies of HDL remodeling in the absence of apoB particles catalyzed by LCAT ( 42,43 ), PLTP ( 44,45 ), or effl ux capacity included HAE, gender, age, HDL-C, triglycerides, and statin use, and explained over 80% of the J774 effl ux variance, with HDL-C ( b = 0.884, P < 0.0001, 59% model variance explained) and HAE ( b = 0.202, P = 0.031, 26% model variance explained) as the most signifi cant contributors to the model ( Table 3 ). The model for ABCA1-specifi c effl ux capacity explained over 50% of the variance, with HAE as the strongest predictor ( b = 0.342, P = 0.015, 24% explained) followed by HDL-C ( b = 0.390, P = 0.01, 23% explained). In both models, male gender was a signifi cant negative predictor of sterol effl ux capacity (ABCA1-BHK sterol effl ux, b = 2 0.754, P = 0.006; J774 sterol effl ux b = 2 0.493, P = 0.007). As expected in both J774 and ABCA1-BHK models, apoA-I and HDL-C were highly correlated and, therefore, only one of them was included in the fi nal model. As confi rmation, models containing either apoA-I or HDL-C were nearly identical. Because statin use and diabetes were signifi cant contributors to the sterol effl ux prediction model, we also tested whether they might affect HAE in particular. Logistic regression was used to evaluate the effect of smoking , statin use, and diabetic status on HAE. Neither statin use nor smoking was associated with HAE with or without adjustment for HDL-C, age, and sex. While diabetic status was associated with signifi cantly HDL particle subspecies ( 21 ). Structural transitions in apoA-I and HDL are therefore important to the process of cholesterol effl ux. In ABCA1-specifi c cholesterol effl ux, for instance, lipid-poor pre b -1 HDL is associated with enhanced cholesterol effl ux from ABCA1 ( 20 ), but high circulating levels of pre b -1 HDL are also associated with CVD ( 48 ), which may be indicative of dysfunctional HDL maturation or remodeling ( 49 ). A potential mechanism responsible for the latter is oxidative modifi cation of HDL by myeloperoxidase, which inhibits desorption of lipidpoor apoA-I from HDL and its interaction with ABCA1, resulting in reduced HAE and cholesterol effl ux capacity ( 24,31,50 ). CETP ( 46 ) suggest that HDL remodeling by these enzymes occurs on the time-scale of hours rather than minutes and, thus, is unlikely to affect HAE. While we cannot rule out the possibility that these factors may infl uence HDL's propensity to exchange apoA-I in vivo, within the context of the HAE analyses reported here, any relationship between HAE and these remodeling enzymes is unlikely. Future studies will be necessary to examine the likely subtle interplay between HDL remodeling factors and HAE.
During RCT, apoA-I transitions through multiple lipidation states, during which it assumes an array of conformations ( 47 ). RCT is thus mediated by a spectrum of diverse to why cholesterol effl ux capacity is a superior predictor of CVD risk compared with HDL-C levels. The J774 macrophage effl ux assay has been used to demonstrate the inverse association between cholesterol effl ux capacity and prevalent CVD, as well as incident CVD ( 13,14,55 ). Our fi nding that HAE strongly associates with sterol effl ux capacity suggests that it may have similar prognostic value, although this will need to be validated by cross-sectional and longitudinal studies with large clinical cohorts. Given the highly heterogeneous distribution of HDL with respect to particle size, charge, protein, and lipid composition, HAE offers a facile means of investigating the effects of these properties on HDL dynamics and their relationship to cholesterol effl ux capacity and CVD risk.
In conclusion, HDL dynamics, as measured by HAE, are positively correlated with serum HDL effl ux capacity, measured in J774 macrophages and ABCA1-BHK cells. This association is independent of both HDL-C and apoA-I levels in the total effl ux (J774) assay and HDL-C in the ABCA1-specifi c effl ux assay. Our fi ndings demonstrate that HDL dynamics are an important factor in HDL's ability to effl ux cholesterol from cells and ultimately its facilitation of RCT.
We observed that while HAE is correlated with ABCA1specifi c effl ux capacity independent of circulating levels of HDL-C, the relationship appears to be dependent on apoA-I ( Table 2 ). This likely refl ects the dependence of ABCA1specifi c sterol effl ux on the release of lipid-poor pre b -1 particles from HDL ( 29 ). Total effl ux capacity measured from J774 macrophages was associated with HAE independent of both HDL-C and apoA-I, suggesting that HAE is more refl ective of total effl ux capacity and represents more than the mere release of endogenous apoA-I from HDL particles. In addition to the ABCA1-mediated pathway, cholesterol effl ux from J774 cells is also driven by ABCG1, SR-BI, and passive diffusion ( 33 ). Cholesterol mobilization via these processes is predominantly through larger HDL particles ( 33 ). The ABCG1 and SR-BI components of effl ux are subtracted from the ABCA1-BHK assay to provide the ABCA1-specifi c component of cholesterol effl ux (see Materials and Methods). Correlation between the ABCA1-BHK and J774 macrophage assays ( Fig. 1A ; r = 0.71) suggests that about 50% of the cholesterol effl ux from J774 cells is mediated by the ABCA1 pathway, consistent with previous observations ( 51 ). The binding of exogenous apoA-I to HDL involves both the release of lipid-poor apoA-I and remodeling of HDL to both larger and smaller particles ( 24 ). Thus, the full repertoire of cholesterol-accepting HDL particles is better described by the J774 effl ux assay, as total serum HDL or purifi ed HDL 3 yields a greater cholesterol effl ux capacity versus lipid-free apoA-I alone ( 52 ).
Multivariate regression modeling of cholesterol effl ux capacity describes HDL dynamics, as measured by HAE, as a signifi cant independent predictor of cholesterol effl ux capacity from both J774 and ABCA1-BHK cells. The optimized J774 model explained more than 80% of the observed variance in effl ux capacity, with HDL-C and HAE having the greatest contribution to the model (over 80% combined). The ABCA1-BHK model explained more than 50% of the variance in effl ux, with HAE, HDL-C, and gender as the strongest predictors ( ‫ف‬ 75% combined). In both models, gender contributed signifi cantly to the effl ux capacity prediction, in agreement with previous data suggesting that effl ux capacity has a gender-specifi c component ( 53,54 ). The fi nding that HAE is a signifi cant predictor of both J774 and ABCA1-BHK effl ux provides a mechanistic rationale as a Standardized b coeffi cient; change of 1 SD in independent variable corresponding to b standard deviations of change in dependent variable.
b Relative importance of individual independent variables as a fraction of the explained variance of dependent variable ( lmg algorithm, see Materials and Methods).