Impact of Self-association on Function of Apolipoprotein A-I*

Background: Self-association is an intrinsic property of exchangeable apolipoproteins but an under-explored feature of the major protein of good cholesterol, apolipoprotein A-I. Result: Different degrees of apolipoprotein A-I self-association exhibit distinct in vitro lipid remodeling and cellular lipid release efficiencies. Conclusion: Self-association of apolipoprotein A-I modulates the biogenesis of high density lipoprotein. Significance: This is the first study to demonstrate that self-association of apolipoprotein A-I attunes key steps in reverse cholesterol transport. Self-association is an inherent property of the lipid-free forms of several exchangeable apolipoproteins, including apolipoprotein A-I (apoA-I), the main protein component of high density lipoproteins (HDL) and an established antiatherogenic factor. Monomeric lipid-free apoA-I is believed to be the biologically active species, but abnormal conditions, such as specific natural mutations or oxidation, produce an altered state of self-association that may contribute to apoA-I dysfunction. Replacement of the tryptophans of apoA-I with phenylalanines (ΔW-apoA-I) leads to unusually large and stable self-associated species. We took advantage of this unique solution property of ΔW-apoA-I to analyze the role of self-association in determining the structure and lipid-binding properties of apoA-I as well as ATP-binding cassette A1 (ABCA1)-mediated cellular lipid release, a relevant pathway in atherosclerosis. Monomeric ΔW-apoA-I and wild-type apoA-I activated ABCA1-mediated cellular lipid release with similar efficiencies, whereas the efficiency of high order self-associated species was reduced to less than 50%. Analysis of specific self-associated subclasses revealed that different factors influence the rate of HDL formation in vitro and ABCA1-mediated lipid release efficiency. The α-helix-forming ability of apoA-I is the main determinant of in vitro lipid solubilization rates, whereas loss of cellular lipid release efficiency is mainly caused by reduced structural flexibility by formation of stable quaternary interactions. Thus, stabilization of self-associated species impairs apoA-I biological activity through an ABCA1-mediated mechanism. These results afford mechanistic insights into the ABCA1 reaction and suggest self-association as a functional feature of apoA-I. Physiologic mechanisms may alter the native self-association state and contribute to apoA-I dysfunction.

Exchangeable apolipoproteins play a central role in lipoprotein assembly, lipid transport, and lipid metabolism by mediating interactions with receptors, enzymes, and lipidtransport proteins. At the secondary structure level, amphipathic ␣-helices are the common lipid-binding element of apolipoproteins (1,2), but a structural landscape more complex than a simple amphipathic ␣-helix is required to regulate apolipoprotein lipid binding (3,4). Structural elements implicated in apolipoprotein function may involve tertiary and quaternary interactions, such as the four-helix bundle, a structural motif that is essential for function in apolipoprotein (apo) E (5-7) and apoA-I (8 -10). At the quaternary structure level, propensity for self-association is a widespread feature of apolipoproteins that may also affect function (apoE (11)(12)(13), apoA-I (14,15), apoA-II (14)), apoA-IV (16), and apoCs (17)). In the case of apoE, for instance, wherein the apoE4 allele associates with higher incidence of Alzheimer disease (18), the isoform-specific distribution of self-associated species (tetramers, dimers, and monomers) has been shown to contribute to the different lipid-binding efficiencies of the two major isoforms (apoE3 and apoE4) and has been proposed to participate in the mechanism determining the dysfunctional phenotype associated with apoE4 (19 -21).
Although the effect of self-association on apoE structurefunction is under scrutiny, much less is known about the influence of self-association on apoA-I function. Increasing our knowledge of this fundamental property of apoA-I has clinical significance because apoA-I plasma levels uniquely predict risk of cardiovascular disease (22)(23)(24). One pathway through which apoA-I exerts its beneficial lipid trafficking function involves the plasma membrane transporter ATP-* This work was supported by a new investigator award from the Tobacco-Related Disease Research Program of California (Grant 18KT-0021) and an International Atherosclerosis Society Visiting Fellowship Award. Partial support was also provided by a Grant-in-Aid for Science Research in Japan from the Ministry of Education, Culture, Sports, Science and Technology (Grant 21591164) and by National Institutes of Health Grants GM067260 and HL026355. □ S The on-line version of this article (available at http://www.jbc.org) contains supplemental Table S1 and Figs. S1-S4. 1  binding cassette A1 (ABCA1) 3 (25)(26)(27). In particular, cholesterol-overloaded macrophages, one of the foremost players in atherosclerosis progression (28,29), primarily efflux excess cholesterol through an ABCA1-dependent mechanism (30,31). The main extracellular recipient of ABCA1mediated cholesterol release is lipid-free apoA-I (32,33), and minimal lipidation of apoA-I greatly reduces its ABCA1-mediated cholesterol release efficiency (34 -37).
Early studies indicated that lipid-free apoA-I self-associates to form dimers, tetramers, and octamers in a concentration-dependent manner and that simple dilution to Յ0.1 mg/ml disrupts the self-associated species to monomeric apoA-I (15,38,39). However, the actual distribution of apoA-I self-associated species in solution is not known, and their presence has been reported even at 0.1 mg/ml (40). In humans, the plasma concentration of apoA-I is ϳ1 mg/ml, where ϳ90 -95% is HDLassociated and ϳ5-10% lipid-free/lipid-poor apoA-I (41,42). Based on this rough estimation, the general notion is that plasma concentration of lipid-free apoA-I is lower than 0.1 mg/ml. Furthermore, binding of as few as two lipid molecules is sufficient to disrupt apoA-I self-association in favor of monomeric lipid-poor species (39,43). Thus, based on a probable lipid-free apoA-I concentration below 0.1 mg/ml and the abundance of lipids, monomers have been hypothesized to be the predominant form of lipid-free apoA-I in plasma. However, the true physiological self-association state of lipid-free apoA-I in the subendothelial artery space, wherein apoA-I exerts its main antiatherogenic function, remains elusive.
Importantly, increasing evidence suggest that dysfunctional apoA-Is, either by natural mutations (44,45) or by oxidative modifications (46), correspond to an altered self-association state that may contribute to the mechanism underlying protein malfunctioning. Recently, the heterozygous apoA-I S36A mutation has been linked to severe reduction in plasma pre-␤-HDL levels and a modified self-association state of the lipid-free protein, which has been proposed to contribute to the inability of the protein variant to produce a normal HDL phenotype (44). ApoA-I Nichinan, a C-terminal single amino acid deletion variant (⌬E235) that associates with low plasma HDL levels and abnormal distribution of HDL subspecies, is another case of potentially self-association-dependent alteration of apoA-I functionality. In fact, the higher than wild-type self-association levels of apoA-I Nichinan may play a role in the lower lipidbinding and ABCA1-mediated cholesterol release efficiencies, leading to the dysfunctional HDL phenotype of Nichinan variant carriers (45).
Furthermore, altering the apoA-I self-association state may impair lipid-free apoA-I stability in solution and lead to functional consequences. Recently, a mechanistic link between apoA-I self-association and amyloidosis has been identified (46). In the atherosclerotic plaque tissue, deposition of apoA-I amyloids contributes to the mechanism of atherosclerosis pro-gression (47,48). ApoA-I methionine oxidation (49) has been implicated in the formation of full-length apoA-I amyloid fibrils through a mechanism involving disruption of apoA-I self-association, reduction of conformational stability in solution, and higher propensity to protein misfolding and aggregation (46).
The above cases suggest that increased or stabilized self-association can be a negative determinant of apoA-I lipid binding efficiency, whereas reduced native self-association may affect the stability of apoA-I structure in solution. The two opposite alterations both have dysfunctional consequences, and it can be hypothesized that maintenance of the native self-association state of lipid-free apoA-I is essential to preserving different aspects of apoA-I functionality.
Here we compared the structural and functional properties of wild-type apoA-I (WT-apoA-I) and a Trp to Phe variant (⌬W-apoA-I) with enhanced propensity to self-association. Characterization of ⌬W-apoA-I structure-function showed that modifications of specific apoA-I residues can cause global structural rearrangements that propagate to the quaternary structure level with striking functional consequences. Interestingly, ⌬W-apoA-I retains wild-type-like lipid-binding and ABCA1-mediated cellular lipid release efficiencies when monomeric or in its low self-association state but displays severely reduced activity when in a high self-association state. These results demonstrate for the first time the effect of selfassociation on the efficiency of apoA-I as a cellular lipid recipient. Furthermore, structure-function data for specific ⌬W-apoA-I self-associated subclasses permit novel conclusions about the structural requirements underlying ABCA1mediated cellular lipid release.

EXPERIMENTAL PROCEDURES
Generation, Expression, and Purification of Protein Variants-Primer-directed PCR mutagenesis was used to introduce single amino acid mutations within human apoA-I cDNA as described previously (50). Multiple Trp to Phe mutations were generated one by one using the mutated DNA as template for the following primer-directed PCRs. DNA sequences were confirmed by automated sequencing and subcloned into the pET-20b bacterial expression vector (Novagen). ApoA-I proteins were expressed in BL21 E. coli (Invitrogen) and purified by nickel affinity chromatography as described previously (51). Protein masses were verified by SDS-PAGE and MALDI-TOF mass spectrometry (29, (52). The denatured samples were refolded at ϳ0.1 mg/ml by dialysis versus PB with three buffer changes. All stock solutions were stored at 4°C. The concentration of each preparation of "High" and "Low" stock solutions was determined by amino acid analysis with an accuracy of Ϯ2% (Molecular Structure Facility at Genome Center, University of California, Davis, CA) and confirmed by the BCA assay (Pierce). After storage at 4°C and prior to each experiment, concentration of the stock solutions and protein integrity were checked by BCA and SDS-PAGE, respectively. Protein stock solutions were prepared in alternative buffer systems: Tris-buffered saline (TBS; 8.2 mM Tris, 150 mM NaCl, 1 mM EDTA, pH 8.0) and sodium phosphate-buffered saline (PBS; 20 mM phosphate, 500 mM NaCl, pH 7.4). No differences in protein stabilities and self-association properties were detected in these buffers.
Size Exclusion Chromatography (SEC)-Protein analysis and isolation of self-associated species were performed on a Superdex 200 prep grade XK 16/100 column and/or a Superose 6 10/300 GL column (GE Healthcare) controlled by an AKTA UPC 10 FPLC system (GE Healthcare). Elution by PB was performed at flow rates of 1.5 and 0.5 ml/min for Superdex 200 and Superose 6 columns, respectively.
Identity of Lipid-free ⌬W-apoA-I Self-association Subclasses-Initial estimation of the identity of individual self-associated species was performed by chemical cross-linking by bis(sulfosuccinimidyl) suberate (BS 3 ) and SDS-PAGE analysis, as described previously (40,54). An increase in the molecular weight of cross-linked species was observed upon cross-linking of higher order self-association species (data not shown). However, interpretation of chemical cross-linking results, specifically for lipid-free apolipoproteins, is not straightforward, because the amount and nature of cross-linked species depend on reagent concentrations and reaction conditions (stoichiometry, temperature, and incubation time). At high cross-linker/ protein ratios, inter-oligomer cross-linking may produce apparent high molecular weight species that are not present in the native mixture (55). At low cross-linker/protein ratios, nonexhaustive cross-linking may not bind all of the protein units present in the higher molecular weight species. These two possibilities cannot be completely ruled out in any experimental conditions; thus, chemical cross-linking is a weak technique for secure identification of self-associated species of lipid-free apolipoproteins. To identify the self-associated species, we rather relied on the following three independent techniques: SEC, using both Superose 6 and Superdex 200 columns; NDGGE with native protein standards; and dynamic light scattering. In SEC experiments, the apparent mass of protein samples was determined by comparing their distribution coefficients (K av ) with those of globular proteins of known mass as described in the supplemental material. Wherever applicable, dynamic light scattering (see supplemental material) was used to confirm the molecular weight of the self-associated species.
Far UV Circular Dichroism (CD) Spectroscopy-CD data were recorded using AVIV 400 or AVIV 62DS spectropolarim-eters with thermoelectric temperature control. Unless mentioned otherwise, far-UV CD spectra (185-250 nm), and melting data were recorded from 20 g/ml protein solutions in PB. The CD data were normalized to protein concentrations and reported as molar residue ellipticity, [⍜]. Protein ␣-helical content was estimated from the molar residue ellipticity at 222 nm ([⍜] 222 ) using the equation, % ␣-helix ϭ ((Ϫ[⍜] 222 ϩ 3000)/ 36,000) ϫ 100 (56). In melting experiments, CD data [⍜] 222 (T) were recorded upon sample heating from 5 to 98°C at a constant rate of 80 K/h. The reversibility of thermal unfolding was checked by cooling the protein to 5°C and repeating the temperature scan for the same sample at the same scan rate.
Differential Scanning Calorimetry (DSC)-Excess heat capacity, C p (T), was measured using a VP-DSC Microcal differential scanning microcalorimeter (Northampton, MA). WT [High] and ⌬W[High] stock solutions where diluted to 0.5 mg/ml in PB and rapidly degassed just before DSC analysis. Data were recorded during sample heating from 5 to 98°C at a rate of 90°C/h, corrected for buffer base line, and normalized to protein concentration. No protein precipitation was visually observed after temperature scanning. However, to test for potential protein loss by precipitation from solution, post-DSC samples were centrifuged at 10,000 ϫ g, and protein concentration was measured by BCA analysis. Sample protein concentrations before and after DSC were within the experimental error, suggesting that exposure of proteins to high temperature did not lead to any significant protein loss from solution.
Phospholipid Clearance Assay-ApoA-I-induced clearance of multilamellar vesicles (MLV) of 1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC) was monitored at 25°C by turbidity using a Varian Cary-300 UV-visible spectrophotometer equipped with thermoelectric temperature control. DMPC-MLV were produced as described (57). The DMPC-MLV suspension (1 ml) was placed in a 1-cm path length quartz cell and equilibrated at 25°C for 5-10 min, during which the turbidity was recorded at 325 nm with an averaging time of 5 s as described (45). No significant changes in turbidity were detected during equilibration. The clearance experiment was initiated by briefly stopping the data recording and adding a small volume of the stock protein solution to final lipid and protein concentrations of 80 and 20 g/ml, respectively. The sample was mixed gently, and the turbidity measurements were immediately resumed and continued for 1 h.
Reconstitution of DMPC-HDL Particles-DMPC-HDL particles were reconstituted by incubating ⌬W[High], ⌬W[High]60, and ⌬W[High]90C with DMPC-MLV (1:4, w/w) overnight at 25°C. After incubation, the samples were span down to pellet any unbound lipid (not soluble at the incubation condition used), and total lipid and protein contents were quantified in the supernatant by Bartlett and BCA assays, respectively. The final protein:lipid content was similar (ϳ1:100 molar ratio) for all reconstituted HDL samples, irrespective of the self-associated state of apoA-I. Furthermore, lipid-free monomeric apoA-I was not detected in the samples by NDGGE and SEC analysis after 24 h of incubation (data not shown). Taken together, these observations suggest that the amount of lipidated protein was similar in the three samples.
Electron Microscopy (EM)-DMPC-particles were analyzed by phosphotungstate negative staining EM under low dose conditions in a CM12 transmission electron microscope (Philips Electron Optics) as described (58). 200 -250 particles/micrograph were assigned to five different diameter classes (7.2, 9.6, 12.0, 14.4, and Ͼ14.4 nm) by determining their diameters with an accuracy of Ϯ1.2 nm.
Preparation of Lipid-free ⌬W-apoA-I Self-association Subclasses-Distinct self-association ⌬W-apoA-I subclasses were prepared either by thermal treatment of ⌬W[High] (tetramers and monomers) or isolation from ⌬W[High] by SEC (hexadecamers and octamers) as described in the supplemental material.
ABCA1-mediated Cellular Lipid Release-HEK293 cells stably expressing human ABCA1-EGFP fusion protein (293/2c) (59) were maintained in a 1:1 (v/v) mixture of Dulbecco's modified Eagle's medium and Ham's F-12 medium (DF medium) supplemented with 10% (v/v) fetal calf serum (FCS). Cells were subcultured in a 6-well tray in 10% FCS-DF medium at 1.2 ϫ 10 6 cells/well. After 48-h incubation, the cells were washed with buffer H (Hanks' balanced salt solution, containing 20 mM HEPES-KOH (pH 7.5) and 14 mM glucose), and incubated with 1 ml/well DF medium containing 0.02% bovine serum albumin (BSA) and apoA-I as indicated, with or without 300 M dibutyryl cyclic AMP. Because 300 M dibutyryl cyclic AMP significantly enhances apoA-I-mediated lipid release without affecting dose dependence (data not shown) (59), this condition was employed to evaluate the ABCA1-dependent lipid release efficiency, specifically. The medium and cells were harvested after a 24-h incubation for lipid analysis. Procedures for lipid extraction and enzymatic assays for choline-phospholipids (PL), total cholesterol, and free cholesterol (FC) have been described (60). Lipid release obtained in the absence of apoA-I was used as background. After lipid extraction, cellular debris in each well was solubilized in 0.1 N NaOH, and protein levels were determined by the BCA assay. Lipid release results were normalized per mg of cellular proteins per well.
Data Processing and Statistical Analysis-Origin software (OriginLab) was used to process CD, DSC, and SEC data. Statistical analysis was performed using the Excel program. Com-parisons between groups were performed by t test methods. Each result reported is a representative of at least three independent experiments.

RESULTS
Unique Self-association Properties of ⌬W-apoA-I-Self-association of WT-apoA-I and over 300 full-length apoA-I variants incorporating single or multiple Trp 3 Phe, Xaa 3 Trp, Xaa 3 Cys, Met 3 Leu, and Tyr 3 Phe were compared by NDGGE and SEC analyses. In particular, altered states of self-association, such as a different size or more resolved self-associated species, were screened. Among this large sample of point mutations, only substitution of the four native tryptophans (Trp-8, Trp-50, Trp-72, and Trp-108) with slightly more hydrophobic phenylalanines (⌬W-apoA-I) significantly altered the self-association behavior of the protein variant (Table 1 and Fig. 1). In agreement with previous studies (43), at high concentration, WT-apoA-I (WT[High] ϳ3.0 mg/ml) ran as a diffused band encompassing native globular markers with sizes of 8.2 and 7.1 nm in diameter (Fig. 1A, lane 1). Similarly, a majority of WT[High] eluted by SEC as a broad peak with three main unresolved components at K av ϭ 0.32, 0.35, and 0.45 (Fig. 1B, black line). By data regression analysis using globular protein markers as calibration standards, we identified these species as tetramers, trimers, and monomers, respectively, of WT-apoA-I. In contrast, ⌬W-apoA-I self-associated species focused by NDGGE in two main bands running at about 12.2-and 9.5-nm apparent globular diameters (Fig. 1A, lane 4) and eluted by SEC as two sharp and well resolved peaks at K av ϭ 0.12 and 0.21 (Fig.  1C, black line), corresponding to hexadecamers and octamers. To our knowledge, such a high order of self-association has not been reported for any other full-length apoA-I variants.
Effect of Concentration on Self-association of ApoA-I Variants-In WT-apoA-I, self-association was readily disrupted by dilution to ϳ0.1 mg/ml (Fig. 1B), in accordance with the largely accepted notion that self-association of apoA-I is concentration-dependent (15,38). In fact, the single narrow peak at K av ϭ 0.47 ml in the SEC chromatogram of diluted WT[High] (Fig. 1B, gray line, and Table 1) represents WT-apoA-I monomers. Strikingly, the NDGGE and SEC pro- a ␣-Helical content was calculated from the molar ellipticity at 222 nm with 5% accuracy, which includes variability from different samples (at least three independent sample preparations) and instrument variability. b The ratio ͓⍜͔ 222 /͓⍜͔ 208 was estimated within an error of Ϯ0.02 from 10 different spectra per experimental condition. files of ⌬W[High] were virtually unchanged by dilution to ϳ0.1 mg/ml in the same conditions (Fig. 1, A (lanes 4 and 5) and C (black versus gray lines) and Table 1). Upon denaturation by GdnHCl and refolding at 0.1 mg/ml, WT-apoA-I (WT[Low]) exists predominantly as monomer (Fig. 1, A (lane 3) and B (light gray line) and Table 1). Interestingly, in ⌬W[Low], high molecular weight oligomers reassembled rapidly. By NDGGE and SEC analysis, five species were clearly visible in ⌬W[Low] stored at 4°C for 48 h (Fig. 1, A (lane 6) and C (light gray line) and Table 1), with predominance of octamers. Taken together, SEC and NDGGE analysis revealed the propensity of ⌬W-apoA-I to self-associate in large oligomeric species even at low protein concentration conditions. Secondary Structure Analysis of ApoA-I Variants-To determine the effect of self-association on the secondary structure of apoA-I, we performed a systematic far-UV CD analysis. In the CD spectra of both WT-and ⌬W-apoA-I, a double negative minimum at 208 and 222 nm and a positive maximum at 190 nm indicate the presence of an ␣-helical conformation ( Fig.  2A). The estimated ␣-helical content of WT[High] (ϳ60%) was similar to previously reported values under the same conditions (15,61), whereas the ϳ82% ␣-helical content of ⌬W[High] was significantly higher and comparable with the ␣-helicity of apoA-I on reconstituted discoidal HDL (15,61,62).
Upon dilution in the 1.00 -0.01 mg/ml range, a small but significant loss of ␣-helical content was measured in WT[High] (Fig. 2B), consistent with a previous report of Massey et al. (63), wherein a reduction in ␣-helical content was observed consequent to disruption of plasma apoA-I self-association. Loss of ␣-helical content (Fig. 2B) alongside self-association (Fig. 1, A  and B) upon dilution of WT[High] suggests a relationship between these two properties of apoA-I. Interestingly, dilution did not change either self-association (Fig. 1, A and C) or ␣-helical content (Fig. 2B) of ⌬W[High], confirming an interplay between self-association and ␣-helicity.
Thermal Stability of ApoA-I Variants-To assess the combined effects of high self-association and ␣-helicity on protein thermal stability, we analyzed ⌬W-apoA-I and WT-apoA-I by two independent methods: CD spectroscopy (Fig. 3) and calorimetry ( Fig. 4 and supplemental Fig. S2A).
Heat-induced changes in protein secondary structure were measured by CD spectroscopy at 222 nm. Before starting the melting experiments, protein stock solutions were diluted to 20 g/ml, wherein WT[High] is monomeric (Fig. 1). Thermal denaturation of WT-apoA-I monomers was reversible and with apparent melting temperature of ϳ63°C (Fig. 3, A and B, black  lines), which is similar to previously published data (56). The melting curve and the apparent melting temperature of a second temperature scan were not significantly different from the first melting results (Fig. 3, A and B, light gray 1 and 4 and black lines). Lanes  3D, black line) was similar to that of WT[High], nearly 30% of the ␣-helical content was lost upon cooling to 25°C. The irreversible nature of the heat-induced transition is probably due to disruption of high order self-associated species. Consistent with this interpretation, the melting curve of a second temperature scan was completely reversible (Fig. 3C, light gray line), confirming that the self-associated species were disrupted by the high temperatures at the end of the first heating cycle and did not reform in the time scale of the consecutive melting. It is notable that the apparent melting temperature in the second scan of ⌬W[High] (ϳ58°C, Fig. 3D, light gray line) was ϳ5°C lower than the first melting temperature (Fig. 3D, black line). Similar to the increased reversibility, the lower apparent melting temperature in the second scan suggests that the self-associated species are disrupted upon completion of the first heating cycle, and the second melting curve represents the reversible unfolding of less stable ⌬W-apoA-I monomers.
To compare the thermodynamic stability of ⌬W-apoA-I and WT-apoA-I self-associated species, we analyzed ⌬W[High] and WT[High] by DSC at higher protein concentrations (0.5 mg/ml). In DSC of WT[High], endothermic transitions occurred at ϳ63°C (main peak) and ϳ51°C (shoulder) (Fig. 4A,  black line). To investigate the nature of these transitions, we performed a second temperature scan on the same sample, wherein self-association had probably been disrupted by the high temperature at the end of the first heating cycle. In the second DSC scan, the absence of a shoulder at ϳ51°C demonstrates that this transition is related to dissociation of self-associated species and that the DSC peak at ϳ63°C represents melting of the ␣-helices of WT-apoA-I monomers (Fig. 4A, light  gray line).
In contrast to WT[High], only one peak with a maximum at ϳ63°C was observed in the DSC of ⌬W[High] (Fig. 4B, black  line). The absence of transitions at lower temperatures indicates that self-associated species in ⌬W-apoA-I are more stable than in WT-apoA-I. In a second DSC scan of ⌬W[High], the peak was broader and shifted to lower temperature with a maximum around ϳ56°C (Fig. 4B, light gray line). This temperature change suggests that during the first DSC scan, both ␣-he-lix melting and dissociation of self-associated species contribute to the observed endothermic transition at ϳ63°C, whereas in the second DSC scan, only melting of the ␣-helices of monomers is observed. Interestingly, the different melting temperatures of the second CD and DSC scans of ⌬W[High] and WT[High] (56 -58°C and ϳ63°C, respectively; Figs. 3 (B and D) and 4) suggest that the ␣-helical structure in ⌬W-apoA-I monomers is less stable than in WT-apoA-I monomers.
When measured by pressure perturbation calorimetry, the volume changes accompanying protein unfolding were remarkably higher for ⌬W[High] than for WT[High] (⌬V/V ϭ Ϫ0.95 and Ϫ0.3%, respectively) (supplemental Fig. S3), and other globular proteins (64). In agreement with the DSC results, these unusually large volume changes in ⌬W[High] may be due to the combined effect of secondary structure unfolding and self-association disruption, which contribute to increasing the solvent exposure of charged groups otherwise restricted in a solvent-excluded environment within self-associated species. Consistent with this view, a significant reduction in the magnitude of volume change was observed in the second consecutive pressure perturbation calorimetry scan of ⌬W[High] (data not shown).
Taken together, the thermal stability measurements indicate that the Trp to Phe mutations partially destabilize the ␣-helical structure of apoA-I monomers and prompt formation of selfassociated species that are more stable than those in WT-apoA-I.
Isolation and Analysis of Low Order ⌬W-apoA-I Self-associated Species-The thermal stability studies (Figs. 3 and 4) suggested that heating of ⌬W[High] above the transition temperature disrupts high order self-associated species. To systematically investigate the effect of heating on self-associated species, we treated ⌬W[High] and WT[High] at different temperatures for 1 h, followed by NDGGE and SEC analysis (Fig. 5). Upon thermal treatment, no protein degradation was detectable by SDS-PAGE analysis (data not shown).

JOURNAL OF BIOLOGICAL CHEMISTRY 35615
trast, heating WT[High] at 60°C, which is above the WT-apoA-I self-associated species transition temperature, led to a significant increase in low order self-association species (in the monomer to tetramer range), as shown by NDGGE and SEC analyses (Fig. 5, A and B, gray line). Unlike WT[High], 60°C treatment of ⌬W[High] (⌬W[High]60C) disrupted octamers and hexadecamers toward only one species with an apparent globular protein size of 7.5 nm (Fig. 5A) and SEC K av of 0.32 (Fig. 5C, gray line), corresponding to a tetramer of ⌬W-apoA-I. The presence of tetramers in ⌬W[High]60C was further confirmed by dynamic light scattering measures (supplemental Table S1). Although we observed no significant amount of monomeric species in ⌬W[High]60C, heating at 90°C resulted in largely monomeric forms of both WT-and ⌬W-apoA-I (K av ϭ 0.42 and 0.43, respectively; Table 1 and Fig. 5, B and C, light gray lines). These results suggest thermal treatment as a simple tool to produce individual low order self-associated sub-classes from otherwise stable high order ⌬W-apoA-I self-associated species.
Analysis of the ␣-helical content of heated samples illustrates the correlation between protein self-association and secondary structure in individual self-association subclasses. Whereas the ␣-helical content of WT[High] converged to 52-53% after heating at 60 or 90°C, the ␣-helicity of ⌬W[High] was remarkably proportional to the degree of self-association: hexadecamers/octamers (⌬W[High]), 82%; tetramers (⌬W[High]60), 72%; monomers (⌬W[High]90), 54% (Fig. 5A). Thus, upon disruption of self-association, the secondary structures of monomeric ⌬W-apoA-I and WT-apoA-I were similar. Furthermore, a large structure rearrangement occurs upon selfassociation in tetramers because these were 18% more ␣-helical than monomers, whereas only an extra 10% ␣-helicity was acquired upon formation of octamers/hexadecamers. The increased stability of higher order self-associated spe-

. Effect of consecutive heating and cooling on thermal denaturation of WT[High] (A and B) and ⌬W[High] (C and D).
Protein stocks were diluted to 20 g/ml in PB just before starting the melting experiments. Intact samples were denatured by heating (first scan, black lines), followed by cooling to 5°C and immediate heating (second scan, light gray lines) at a scan rate of 80 K/h. A and C, protein unfolding during thermal denaturation was monitored by CD at 222 nm. Heating and cooling directions are marked by arrows. B and D, first derivatives of curves in A and C. Peak maxima are marked by arrows. Melting temperatures (see "Results") were estimated from peak maxima with an accuracy of Ϯ2°C, which includes variability from different samples (at least three independent sample preparations) and instrument variability.
cies must therefore arise from tertiary and quaternary structure contributions.
Evaluation of the ratio of molar residue ellipticity at 222 and 208 nm ([⍜] 222 /[⍜] 208 ) yields useful information on the nature of these tertiary and quaternary interactions. The [⍜] 222 /[⍜] 208 ratio provides an estimate of the level of coiled-coil inter-␣helical contacts because the ratio is Ͻ1.00 for non-interacting ␣-helices and Ͼ1.00 in the presence of two-stranded coiledcoils (65,66). The [⍜] 222 /[⍜] 208 ratio was within the 0.81-0.88 range for all WT-apoA-I samples at any protein concentration tested (Table 1 and Fig. 2B), suggesting that in WT-apoA-I, no coiled-coil structure is formed at any degree of self-association. Likewise, the ratio was 0.83-0.85 for monomeric or self-associated ⌬W-apoA-I up to tetramers. Strikingly, the [⍜] 222 / [⍜] 208 ratio was 1.03 for ⌬W [High], which is composed of octamers and hexadecamers and 1.01 for ⌬W [Low], which mainly contains octamers and hexadecamers together with a small amount of lower order self-associated species. Thus, in contrast with the mechanism of self-association of apoE, where coiled-coil interactions contribute to the formation of low order self-associated species, such as dimers (13), in ⌬W-apoA-I, coiled-coil interactions participate in the stabilization of octamers/hexadecamers but are absent in lower order self-associated species.
Effect of Self-association on Lipid Binding-Massey et al. observed increased rates of DMPC clearance when plasma apoA-I self-association was disrupted by performing the DMPC clearance experiment in the presence of 0.5 M GdnHCl (63). Although no changes in secondary structure were reported at these denaturant concentrations, effects at the tertiary structure level with potential consequences on protein's lipid-binding efficiency cannot be ruled out. In fact, small but significant differences in the near-UV CD spectra of WT-ApoA-I incubated for 1 h with and without 0.5 M GdnHCl are indicative of tertiary structure changes (data not shown). In the current study, we took advantage of the unique self-association properties of ⌬W-apoA-I to test the dependence of lipid-binding kinetics on different degrees of apoA-I self-association in completely native conditions. Remarkably, DMPC clearance rates increased proportionally to the reduction in self-association levels (Fig. 6, A and B). Furthermore, lipid-binding kinetics of monomeric ⌬W-apoA-I (⌬W[High]90C) and WT-apoA-I (WT[Low]) were comparable (Fig. 6, B and A, respectively). This result demonstrates that the introduction of four amino acid mutations in ⌬W-apoA-I did not significantly change the lipid-binding efficiency of the protein in its monomeric form, making ⌬W-apoA-I a reliable model for studying the effect of self-association on apoA-I function as a recipient of cellular lipid release.
Size and Morphology of DMPC-HDL Particles-By negative staining EM (Fig. 7), high order self-associated species in ⌬W[High] (hexadecamers and octamers) produced particles heterogeneous in size but significantly larger (the most represented diameter class was 12.0 nm) than those generated by ⌬W[High]60C and ⌬W[High]90C. Surprisingly, monomeric ⌬W[High]90C formed particles with the broadest distribution of diameters (7.2-14.4 nm) (Fig. 7C). This observation suggests that a variable number of monomers can associate on the MLV surface and produce particles of diverse protein-lipid stoichiometry.
By analysis of the amount of rouleau formation, higher order self-associated species (Fig. 7A) produced relatively fewer discoidal particles than tetrameric and monomeric apoA-I (Fig. 7, B and C, respectively). It is notable that tetramers (⌬W[High]60C) yielded a narrow particle size distribution centered at the 9.6-nm diameter class. Based on size and morphology, these particles probably contain two apoA-I molecules per particle (53).
Because the amount of lipidated protein was similar in the three samples (see "Experimental Procedures"), the observed differences in particle morphology and size are probably due to the different self-association states of the initial protein component.  High] (B). Intact samples were denatured by heating at 90°C/h (first scan, black lines), followed by cooling below 25°C and immediate heating (second scan, light gray lines) at the same scan rate. Endothermic transition temperatures are marked by arrows. The numeric values (see "Results") were estimated from peak maxima with an accuracy of Ϯ3°C, which includes variability from different samples (at least three independent sample preparations) and instrument variability. Impact of Self-association on ApoA-I-mediated Cellular Lipid Release-Lipid-free apoA-I is known to activate cellular lipid release through an ABCA1-dependent pathway (30,34). To study the effect of self-association on the efficiency of apoA-I as the recipient of ABCA1-dependent cellular lipid release, four distinct self-association subclasses of ⌬W-apoA-I (namely hexadecamers, octamers, tetramers, and monomers) were prepared as described under "Experimental Procedures" (Fig. 8 and supplemental Fig. S1). We measured the cellular lipid release efficiencies of the four apoA-I self-association subclasses by incubating 293/2c cells expressing high levels of ABCA1 in the presence of increasing concentrations of apolipoproteins.
As a control, WT[High] was heated at 90°C (WT[High]90C) at the same conditions as for ⌬W[High]90C, and its cellular lipid release activity was measured at 2, 5, and 10 g/ml. At most of the conditions tested, WT[High]90C was similar to WT[High], ⌬W[High]60C, and ⌬W[High]90C in activating cellular lipid release. We also evaluated non-ABCA1-mediated lipid release by HEK293 cells, which do not express ABCA1. In this cellular system, none of the apolipoproteins at any of the concentrations tested was able to activate PL or FC release at levels significantly different from control samples incubated in the absence of apolipoproteins (data not shown). Cellular lipid release results indicate that the ABCA1-mediated lipid release efficiency of apoA-I is severely impaired by formation of stable quaternary interactions in high order self-associated species, but disruption of self-association at or below the tetrameric level restores efficiency to levels similar to those of monomeric WT-apoA-I.

DISCUSSION
Tryptophans appeared in the apoA-I sequence relatively recently in evolution (67); one Trp (corresponding to human Trp-108) is conserved in the apoA-Is of terrestrial vertebrates, but all known fish apoA-Is have no tryptophans. Interestingly, the other three tryptophans of human apoA-I are specific to mammals and are extremely well conserved in this class (67,68). Thus, we speculated that the acquisition of the four tryp-  tophans in the apoA-I of the mammalian ancestor led to an important functional property that drove fixation of this new form in mammals. The current work suggests that this property may involve the modulation of self-association.
Edmundson helical wheel projections of the three tryptophans predicted to be in ␣-helical regions (Trp-50, Trp-72, and Trp-108) (57,73) locate the tryptophans within the non-polar face of amphipathic ␣-helices (Trp-50 and Trp-108) or close to the interface between the polar and non-polar faces (Trp-72) (supplemental Fig. S4) (67,74). Consistent with this observation, solvent accessibility and Trp fluorescence measurements of monomeric lipid-free apoA-I demonstrate that the tertiary  structure folding provides a hydrophobic environment for all four tryptophans (62,74,75). This suggests that tryptophans are directly involved in interhelical interactions within the fourhelix bundle. Although they share aromatic characteristics, Trp is considerably less hydrophobic than Phe (76). Thus, replacing tryptophans with phenylalanines may affect the interhelical arrangement of apoA-I. In fact, Trp to Phe substitutions reduced the stability of apoA-I monomers (62). In the work described here, mutation of all four native tryptophans to phenylalanines had a dramatic effect on the overall protein structure. In fact, lipid-free ⌬W-apoA-I was composed exclusively of hexadecamers and octamers and about 82% ␣-helical, a marked 22% increase compared with WT-apoA-I at the same concentration and similar to the ␣-helicity of apoA-I in nascent HDL.
The N-terminal domain of monomeric lipid-free WT-apoA-I is predominantly ␣-helical (52,70,71), and thus the high ␣-helicity of self-associated ⌬W-apoA-I (ϳ200 of the 243 residues) must involve a propagation of the ␣-helical character to the largely unfolded C terminus (52,72). This result is consistent with the notion that self-association of apoA-I is the structural response of an otherwise unstructured C-terminal domain, wherein hydrophobic residues seek protection from solvent exposure by establishing new intermolecular contacts (15,77). Furthermore, the only two self-associated species present in native ⌬W-apoA-I are significantly larger and more stable than the self-associated species of WT-apoA-I. The increased stability of ⌬W-apoA-I hexadecamers/octamers is illustrated by a dissociation temperature higher than the melting temperature of the residual ␣-helices in monomeric ⌬W-apoA-I (Figs. 4 and 5), which suggests that their structure is stabilized through strong intermolecular and, potentially, new intramolecular (tertiary) contacts.
We speculate that the Trp to Phe substitutions destabilize the N-terminal four-helix bundle, exposing hydrophobic resi-dues to solvent and potentially disrupting N terminus-C terminus interactions. This destabilization is compensated for by an overall structural rearrangement where C-terminal random coil regions fold in amphipathic ␣-helices, which participate in intermolecular hydrophobic interactions. Further supporting evidence is provided by our finding that formation of coiledcoil structures occurs exclusively in octamers/hexadecamers, indicating that interhelical interactions contribute to the stabilization mechanism in high order self-associated species.
Although we cannot exclude the possibility that specific structural features of the self-associated species generated by the Trp to Phe substitutions are unique to this variant, ⌬W-apoA-I serves as a model for studying the gross effect of self-association on apoA-I biological function. In fact, the large self-association-dependent structural rearrangement had striking consequences for the functional activity of ⌬W-apoA-I. The DMPC clearance and cellular ABCA1-mediated lipid release efficiencies of monomeric ⌬W-apoA-I were not significantly different from the activity of monomeric WT-apoA-I, indicating that modification of the apoA-I primary sequence by Trp to Phe substitutions did not exert any direct influence on the main function of apoA-I. In contrast, DMPC clearance rates and ABCA1-dependent lipid release efficiencies of self-associated ⌬W-apoA-I species were significantly reduced.
It has been proposed that the membrane insertion of the loosely folded C terminus of monomeric apoA-I is a requirement for maintaining the HDL-forming activity of apoA-I through a structural rearrangement encompassing the whole molecule (52,78). Within the process of HDL formation from PL vesicles in vitro, binding of lipid-free apoA-I to the vesicle surface is an essential step that is driven by a large enthalpic contribution provided by the transition of non-helical residues to an ␣-helical conformation (70, 79 -82). The large number of ␣-helical residues in high order ⌬W-apoA-I self-associated species (ϳ200 of 243) indicates that their ability to form new ␣-helices is greatly reduced. Remarkably, the DMPC clearance rates of ⌬W-apoA-I were proportional to the number of residues in non-␣-helical conformation from hexadecamers/octamers (ϳ18% non-␣-helical) to tetramers (ϳ28%) and monomers (ϳ46%). However, in a cellular model where lipid release is ABCA1-dependent, apoA-I tetramers and monomers exhibited similar lipid release efficiencies, despite an 18% difference in ␣-helicity between the two species. In contrast, the cellular lipid release efficiency of octamers and hexadecamers was reduced to Յ50% of WT-apoA-I levels, despite only a 10% difference in ␣-helicity from tetramers. Thus, our observations suggest that the enthalpic gain provided by the formation of new ␣-helices determines the efficiency of HDL production from DMPC vesicles in vitro. However, in ABCA1-mediated conditions, where the cell membrane curvature is actively increased by the membrane transporter, lipid release efficiency is less dependent on the protein ␣-helix-forming ability and strongly affected by the presence of stable intermolecular interactions. The formation of stable quaternary structures reduces the cellular lipid release efficiency of apoA-I; we speculate that it does so by limiting protein flexibility at sites where flexibility is needed to promote the structural changes that follow initial lipid-binding and determine the efficiency of HDL formation in vivo.
Nascent HDL produced by lipidation of lipid-free apoA-I at the cell surface contains two or three molecules of apoA-I. Thus, in a scenario wherein lipid-free apoA-I is self-associated, two pathways are possible: (i) dissociation of self-associated apoA-I to monomers and reassembling of monomers at the cell membrane to generate nascent HDL and (ii) disruption of high order self-associated species to dimeric or trimeric apoA-I, which can act as minimal functional units directly interacting with lipids. Our results demonstrate that monomers and tetramers have similar cellular lipid release efficiencies, raising the interesting possibility that dimers generated from tetramers rather than monomers are the minimal functional unit of lipid-free apoA-I. Analysis of DMPC particles generated by different ⌬W-apoA-I self-associated species revealed that tetramers preferentially form 9.6-nm discoidal particles, whereas the particles produced by monomeric apoA-I are largely heterogeneous in size and, potentially, number of apoA-I molecules per particle (Fig. 7). This evidence suggests that dimeric apoA-I dissociating from tetrameric species could be the species that most efficiently generates two-apoA-I-containing lipoproteins. However, cell culture conditions are very different from in vitro DMPC binding, and, although our in vitro studies ( Fig. 8 and supplemental Fig. S2) did not detect monomer formation from high order self-associated ⌬W-apoA-I during a 24-h incubation at 37°C, we cannot exclude the possibility that  OCTOBER 14, 2011 • VOLUME 286 • NUMBER 41 dissociation of tetramers to monomers occurs in cell culture. Thus, further experiments will be required to unequivocally establish which of the two reaction pathways happens in vivo.

Self-association-dependent Function of Apolipoprotein A-I
According to the existing paradigm of structure-function relationship in apoA-I, the integrity of specific residues (i.e. primary structure) is necessary for preservation of the lipid acceptor function of apoA-I. This notion is derived from the dysfunctional consequences of naturally occurring point mutations and of in vivo oxidation of specific residues. For instance, in the atherosclerotic plaque, the macrophage-produced myeloperoxidase (MPO) catalyzes oxidation of several apoA-I residues and reduces the cellular cholesterol release efficiency of apoA-I (83)(84)(85). However, work from the Heinecke, Smith, and Hazen groups (86 -88) did not unequivocally identify the MPO-targeted residues (i.e. Tyr, Lys, and Met) that are directly implicated in the loss of the ABCA-1-mediated cholesterol release efficiency of apoA-I. To date, resistance to MPO-dependent loss of cholesterol release efficiency has been shown to occur only when tryptophans, which are only marginally modified by MPO (89), have been mutated to phenylalanines (90). We propose that this protection from MPO oxidation is mediated by the high self-association state of ⌬W-apoA-I, which reduces widespread oxidative damage by limiting access of MPO-generated oxidants to the solvent-excluded residues within selfassociated species.
The Trp to Phe case prompts a more general hypothesis, that loss of function by modifications of specific residues does not necessarily imply direct involvement of those residues in binding of apoA-I to ABCA1, cell membrane, or lipids. Alternatively, modification of certain residues could affect the overall structure of apoA-I, such as the stability of high order selfassociated species, and potentially create two opposite effects. In some conditions, such as general atherosclerosis progression, reduced cellular lipid release efficiency by limited local protein flexibility is the predominant effect. In more specific cases, such as MPO-mediated oxidation, more stable quaternary structure may provide extra protection from deleterious protein modification reactions. Finally, similarly to the case of full-length apoA-I amyloid fibril formation, oxidation may destabilize the native self-association state of apoA-I in solution and promote protein aggregation.
In conclusion, we have demonstrated that self-association is one property of apoA-I that strongly influences its structure, the size and morphology of the HDL particles produced upon phospholipid binding, and the ability of apoA-I to release lipid from cells. The large self-association-mediated structural changes may explain why at high concentrations (similar to the conditions used for protein crystallization) specific stable structural subspecies are selected, although these are not necessarily predominant at physiological concentrations. We hypothesize that in vivo the lipid acceptor function of apoA-I does not directly depend on the activity of specific residues but on a different and more general structural feature of the protein. Thus, protein self-association should be taken into consideration as one of the functional structure-determinant proper-ties that can be targeted by in vivo protein modifications that render apoA-I dysfunctional.