Thermal transitions in serum amyloid A in solution and on the lipid: implications for structure and stability of acute-phase HDL.

Serum amyloid A (SAA) is an acute-phase protein that circulates mainly on plasma HDL. SAA interactions with its functional ligands and its pathogenic deposition in reactive amyloidosis depend, in part, on the structural disorder of this protein and its propensity to oligomerize. In vivo, SAA can displace a substantial fraction of the major HDL protein, apoA-I, and thereby influence the structural remodeling and functions of acute-phase HDL in ways that are incompletely understood. We use murine SAA1.1 to report the first structural stability study of human plasma HDL that has been enriched with SAA. Calorimetric and spectroscopic analyses of these and other SAA-lipid systems reveal two surprising findings. First, progressive displacement of the exchangeable fraction of apoA-I by SAA has little effect on the structural stability of HDL and its fusion and release of core lipids. Consequently, the major determinant for HDL stability is the nonexchangeable apoA-I. A structural model explaining this observation is proposed, which is consistent with functional studies in acute-phase HDL. Second, we report an α-helix folding/unfolding transition in SAA in the presence of lipid at near-physiological temperatures. This new transition may have potentially important implications for normal functions of SAA and its pathogenic misfolding.

lipid-poor apoA-I, are expected to infl uence HDL functions, such as the effl ux of cell cholesterol at the critical steps of reverse cholesterol transport. In fact, some studies report that HDL enrichment with SAA increased HDL binding to macrophages, decreased cholesterol effl ux from macrophages, and increased uptake of cholesteryl ester ( 34,35 ). Moreover, SAA-HDLs are ineffective mediators of cholesterol esterifi cation by LCAT ( 36 ) and have reduced antioxidant potential ( 37 ). These reports suggest that SAA diminishes the antiatherogenic properties of HDL. However, the effects of SAA on HDL functions are not necessarily deleterious ( 26,38 ). For instance, similar to apoA-I, lipid-free SAA promotes effl ux of cholesterol and phospholipids from cells and, hence, is proposed to play a protective role in cholesterol removal from macrophages at the sites of infl ammation ( 14,26,39 ). Interestingly, certain key functions of HDL in reverse cholesterol transport are relatively unaffected by the biochemical changes in acute-phase HDL; this includes the stimulation of cholesteryl ester transfer protein (CETP) or the net effect on cell cholesterol effl ux ( 15 ). Thus, the complex effects of SAA on the protective properties of HDL are far from clear. Moreover, the effects of SAA on HDL structure, stability, and functional remodeling during reverse cholesterol transport remain unclear and are addressed in the current study.
Metabolism of normal HDL critically depends on HDL remodeling by plasma factors such as LCAT and other lipases, CETP and phospholipid transfer protein (PLTP), and HDL scavenger receptor class B type I (SR-BI). This remodeling alters the biochemical composition of HDL and increases its lipid load, thereby shifting the balance between the polar surface of HDL and its apolar core. This balance can be restored upon apolipoprotein dissociation and HDL fusion, which can be induced by LCAT, hepatic lipase, PLA 2 , CETP, or PLTP (40)(41)(42)(43)(44). Similar to apoA-I dissociation in these reactions, SAA can also be released from acute-phase HDL upon the action of CETP, PLA 2 , and other plasma factors ( 45 ). At the last step of reverse cholesterol transport, HDL disintegrates (or ruptures) and releases its core lipids that are taken up by the SR-BI receptor; this disintegration is accompanied by apolipoprotein dissociation ( 46 ). Notably, important aspects of metabolic HDL remodeling by plasma factors, including apoA-I dissociation and HDL fusion and rupture, are mimicked in chemical or thermal denaturation ( 47,48 ). Therefore, we use thermal denaturation as a tractable experimental model to determine the effects of various HDL modifi cations on the functional remodeling of HDL ( 49 ).
Here, we combine several biochemical and biophysical methods to determine, for the fi rst time, the effects of SAA enrichment on the HDL stability to heat-induced fusion and rupture. We report the fi rst study of SAA in solution and on HDL using differential scanning calorimetry (DSC). To glean further insights into the molecular basis of the observed effects, we investigate the stability and conformational changes of SAA in the presence of lipids. A structural model of SAA-HDL is proposed that explains our fi ndings. These fi ndings may have important implications for understanding functional remodeling of acute-phase HDL as well as the normal functions of SAA and its pathological misfolding in amyloid.
Although the functional role of SAA is subject of debate, various biological functions have been attributed to this protein. In addition to its putative role in immune response, SAA was proposed to modulate blood coagulation ( 11 ), monocyte and neturophil chemotaxis ( 12 ), and HDL functions in cholesterol transport and metabolism (13)(14)(15)(16). SAA is implicated in retinol transport ( 10 ), and elevated plasma SAA is a cardiovascular risk factor ( 17 ). Many aspects of normal functions of SAA and its pathological deposition in amyloid remain unclear and are intimately linked to HDL metabolism.
Plasma HDLs are heterogeneous nanoparticles (8-12 nm) composed of lipids and specifi c amphipathic proteins, termed apos. The proteins and polar lipids (mainly phospholipids and cholesterol) form the HDL surface, while apolar lipids (cholesteryl esters and triacylglycerols) are sequestered in the core. Plasma levels of HDL cholesterol and the major HDL protein, apoA-I, have long been known to correlate inversely with the risk of cardiovascular disease ( 18 ), although recent studies show that HDL quality is also very important ( 19,20 ). The cardioprotective effect of HDL is attributed, in part, to its antioxidant, antithrombolytic, and anti-infl ammatory properties, but mainly to its central role in reverse cholesterol transport, which is the sole pathway for removal of excess cholesterol from peripheral cells such as arterial macrophages ( 21,22 ). ApoA-I (28 kDa, 243 amino acids) plays a key functional role in this complex pathway and also forms an important structural scaffold on the HDL surface ( 23,24 ). In normal HDL (nHDL), apoA-I constitutes ‫ف‬ 70% of protein mass, apoA-II (77 amino acids) constitutes ‫ف‬ 20%, and the remainder is composed of minor apos ( 25 ). The biochemical composition of HDL can drastically change in infl ammation upon upregulation of two major acute-phase proteins, SAA and secretory phospholipase A 2 (sPLA 2 ), which remodel HDL ( 26 ).
Studies in mice and humans show that, in the acute phase, SAA binds to HDL, displaces a large fraction of apoA-I and other minor proteins, and becomes the major HDL protein that can constitute up to 80% of the total protein content in HDL ( 27,28 ). Although SAA-enriched HDLs (SAA-HDLs) have larger size, they are associated with denser HDL 3 fraction, which is consistent with higher protein content in acutephase human HDL compared with nHDL ( 29 ). This situation differs from acute infl ammation in SAA-defi cient mice where the increase in HDL size results from increased lipid content ( 30,31 ). Because apoA-I is the major functional protein on HDL that modulates several lipid-processing reactions, the displacement of a large fraction of apoA-I with SAA is expected to alter HDL lipid composition. In fact, acute-phase HDL is enriched in free cholesterol and depleted in cholesteryl esters, which is likely due to the depletion in apoA-I, the preferred cofactor for LCAT. Acute-phase HDL is also enriched in triglycerides and free fatty acids that are elevated in infl ammation ( 31 ). Importantly, HDL plasma levels can decline up to 2-fold in acute infl ammation ( 32 ). This decline may result either from potential destabilization of acutephase HDLs, which would lead to their enhanced catabolism ( 33 ), or, more likely, from HDL processing by sPLA 2 ( 26 ).
Changes in the biochemical composition and plasma levels of acute-phase HDL, along with the displacement of range of relative protein concentrations found in plasma. In fact, in acute-phase plasma, SAA can reach up to 1 mg/ml, while apoA-I can decline 2-fold from its normal levels of 0.8-1.4 mg/ml ( 52,53 ). Therefore, concentrations of 0.85 mg/ml SAA and 0.5 mg/ml apoA-I, which exemplify acute-phase plasma, correspond to 4:1 mol/mol SAA:apoA-I used in our studies.
To assess HDL remodeling upon incubation with SAA, total samples of nHDL and SAA-HDL were characterized by size-exclusion chromatography (SEC) controlled by an AKTA UPC 10 FPLC system (GE Healthcare), and by nondenaturing gel electrophoresis (NDGE) followed by immunoblotting for apoA-I. The results clearly show that a large fraction of apoA-I progressively dissociates from HDL upon incubation with SAA in a dosedependent manner ( Fig. 1 ). Thus, SEC data show that dissociated apoA-I is absent from nHDL, is present in SAA-HDL (2:1), and increases in SAA-HDL (4:1) where more than 50% of total apoA-I is dissociated ( Fig. 1A ). This result agrees with densitometry analysis of the immunoblot for apoA-I ( Fig. 1C, D ), which was carried out by using ImageJ software ( 54 ). The fraction of dissociated apoA-I increases from ‫ف‬ 3% at 0.5:1 SAA:apoA-I molar ratio to more than 60% at 4:1 ratio ( Fig. 1D ). These and other data, such as SDS-PAGE quantifi cation of SEC fractions of SAA-HDL (not shown), consistently show that at 4:1 SAA:apoA-I mol ratio, approximately two-thirds of all apoA-I is dissociated from HDL. To maintain the particle density in the HDL 3 range, a comparable or slightly larger amount of SAA must be associated with these particles. These results agree with previous in vitro studies in SAA-HDL obtained by incubation, as well as with protein analysis in acute-phase plasma HDL reporting that up to three-fourths of the total apoA-I can be displaced from these particles ( 52,53 ).

Preparation and characterization of normal human plasma HDL and SAA-HDL
Plasma of healthy volunteer donors was purchased from Research Blood Components LLC, according to their rules and regulations and upon approval of the Institutional Review Board. Single-donor HDLs of healthy humans, termed nHDL, were isolated from fresh EDTA-treated plasma by KBr density gradient ultracentrifugation in the density range 1.08-1.21 g/l as previously described ( 48 ). HDL stock solution containing 10 mM Na phosphate buffer at pH 7.5 (which is a standard buffer used throughout this study) and 0.25 mM Na EDTA was stored in the dark at 4°C and was used within 4 weeks. To obtain SAA-HDL, nHDLs were incubated with SAA at 37°C for 3 h, during which the exchange reaction reached equilibrium (data not shown). No changes in the particle size, composition, or stability were detected in control HDL upon incubation. The SAA to apoA-I ratios ranged from 0:1 (control) to 4:1 mol/mol, or 1.7:1 mg/mg, which approximates the Fig. 1. Characterization of SAA-HDL by SEC and immunoblotting. A: SEC profi les of SAA-HDL using Superdex-200 preparative grade XK 16/100 column. Elution by PBS (10 mM Na phosphate, 150 mM NaCl, pH 7.5) was carried out at a fl ow rate of 1 ml/min. The data were recorded from nHDL and from SAA-HDL that have been incubated at 2:1 or 4:1 SAA:apoA-I molar ratio. The peaks corresponding to HDL-bound and dissociated proteins are indicated. Free apoA-I and SAA are shown for comparison. The data were shifted along the y -axis to avoid overlap. The relative peak intensities, which were obtained by integration using Unicorn software of AKTA, are proportional to weight % of total protein: 100% HDL-bound protein in nHDL; 83% HDL-bound protein and 17% dissociated apoA-I in SAA-HDL (2:1); 52% HDL-bound protein and 48% dissociated apoA-I in SAA-HDL (4:1). (B-D) Quantifi cation of HDL-bound and dissociated apoA-I in SAA-HDL by immunoblotting. SAA-HDL (0.5:1 to 4:1 mol/mol of SAA:apoA-I as indicated) were analyzed by NDGE (B) followed by immunoblotting for apoA-I (C) and band quantifi cation (D). NDGE shows progressive increase in HDL particle size with concomitant protein dissociation (C). Western blot shows progressive dissociation of apoA-I from HDL. The apoA-I content was quantifi ed by ImageJ analysis of the Western blot ( 54 ). The area under the peak was calculated in the calibrated mode. The % fraction of protein in each band is indicated. The results suggest preferential displacement of apoA-I from small HDL.

Temperature-dependent structural transitions in lipid-free SAA
First, we characterized the temperature dependence of the solution conformation of lipid-free SAA. Secondary structure and aromatic packing were monitored by far-and near-UV CD using 0.1 and 0.5 mg/ml protein, respectively. Far-UV CD spectra showed that SAA was ‫ف‬ 33% ␣ -helical at 5°C but largely unfolded at 25°C ( Fig. 2A , insert) and fully unfolded at 37°C, as evident from the melting data ( Fig.  2A ) and the CD spectrum at 37°C (not shown). To monitor ␣ -helical structure as a function of temperature, CD melting data were recorded at 222 nm. The results showed a sigmoidal unfolding transition centered at 18°C, which was largely reversible upon heating to 60°C and cooling to 5°C ( Fig.  2A ), in agreement with other SAA studies ( 57 ).
Near-UV CD spectra of free SAA at 5°C showed a negative peak at ‫ف‬ 283 nm indicating Trp and/or Tyr packing resulting from tertiary and/or quaternary structure, which disappeared at 25°C ( Fig. 2B , insert). Comparison of the melting data recorded at 222 nm and at 280 nm showed concomitant changes ( Fig. 2A, B ) indicating that the ␣ -helical unfolding was accompanied by loss of aromatic packing. In addition to the protein unfolding below 30°C, irregular "bumps" in near-UV CD melting data were detected at 30°C-45°C, which were not accompanied by large far-UV CD changes ( Fig. 2A, B ) and probably refl ected changes in protein oligomerization (described subsequently).
Structural transitions in lipid-free SAA were further explored by DSC. Fig. 2C shows consecutive heating scans from 10°C to 60°C recorded from the same sample containing 0.3 mg/ml SAA. An endothermic cooperative largely reversible transition centered at 36°C was observed between 30°C and 40°C. Because SAA is unfolded at these temperatures ( Fig. 2A ), this transition must refl ect changes in protein self-association rather than secondary structure. DSC data recorded at various SAA concentrations support this conclusion. Although the calorimetric peak showed little change upon increasing SAA concentration from 0.18 to 0.3 mg/ml, further increase to 1.0 mg/ml SAA led to increased amplitude and a high-temperature shift of this peak (supplementary Fig. 1), supporting the idea that it involves changes in protein oligomerization. Further support comes from the previous studies reporting changes in SAA1.1 selfassociation at near-physiological temperatures ( 57,58 ).
In sum, SAA heating to 30°C leads to nearly complete unfolding detected by far-and near-UV CD, followed by changes in protein self-association between 30°C and 40°C detected by DSC. These structural transitions are largely reversible.

Temperature-dependent structural transitions in SAA-HDL
To explore structural and stability properties of HDL that have been enriched with SAA, such SAA-HDLs were prepared by using various SAA:apoA-I ratios as described in Materials and Methods and were analyzed for particle size, protein composition, conformation, and structural stability. As expected, NDGE and negative-stain EM showed Next, HDL and SAA-HDL that were intact or were subjected to controlled heating and cooling were characterized by SDS-PAGE, NDGE, and immunoblotting (described below), and by transmission electron microscopy (EM). For EM analysis, lipoproteins were negatively stained with Na phosphotungstate and were visualized under low-dose conditions in a CM12 transmission electron microscope (Philips Electron Optics) as previously described ( 47 ).

Protein complexes with DMPC and POPC vesicles
Model lipoproteins were reconstituted by incubating SAA or apoA-I with multilamellar vesicles of DMPC (1:4 w/w) overnight at 24°C following established protocols ( 51 ). The unbound lipids were pelleted by centrifugation. The lipid and protein content in the supernatant was quantifi ed by Bartlett and BCA assays, respectively.
Homogenous small unilamellar vesicles (SUVs) of POPC, which were ‫ف‬ 22 nm in size, were prepared by sonication following published protocols ( 55 ). SAA or apoA-I was mixed with SUVs at a protein:lipid molar ratio of 1:100 and was incubated at 37°C for 6-12 h. The protein-lipid complexes were analyzed by NDGE and negative-stain EM as previously described ( 51 ).

Gel electrophoresis and immunoblotting
For NDGE, Novex™ 4-20% Tris-glycine gels (Invitrogen) were loaded with 6 g protein per lane and run to termination at 1,500 V-h under nondenaturing conditions in Tris-glycine buffer. For SDS-PAGE, Novex™ 4-20% Tris-glycine gels were loaded with 5 g protein per lane and run at 200 V for 1 h under denaturing conditions in SDS-Tris-glycine buffer. The gels were stained with Denville blue protein stain. For Western blotting, the proteins were separated either on native or on SDS gels and were transferred to a polyvinylidene difl uoride membrane for 1 h at 100 V, 4°C. The membrane was blocked for 1 h in Tris-buffered saline/casein blocking buffer. The blots were probed with antibodies for apoA-I or SAA in blocking buffer for 1 h. The blots were washed three times for 10 min each and were visualized using the ECL system (NEN Life Science Products).

Thermal stability studies
Calorimetric data were recorded from protein or lipoprotein solutions using the VP-DSC microcalorimeter (MicroCal) as previously described ( 56 ). Sample concentrations ranged from 0.18 to 1.0 mg/ml protein in standard buffer. Briefl y, the heat capacity data, C p ( T ), were recorded during sample heating from 5°C up to 120°C at a rate of 90°C/h. DSC data of SAA alone were normalized to protein concentration and are reported in units kcal/mol(SAA)×°C. DSC data of HDL, which contained a mixture of various proteins and lipids, are reported in units of cal/°C. Circular dichroism (CD) data were obtained using an AVIV415 spectropolarimeter to monitor protein secondary structure, aromatic packing, and thermal stability. Far-UV CD spectra (185-250 nm) were recorded from protein or lipoprotein solutions in standard buffer. Far-UV CD data from samples containing only one protein (SAA or apoA-I) were normalized to protein concentration and expressed as molar residue ellipticity, [ ⌰ ]. The CD data recorded from mixed protein systems, such as SAA-HDL, were not normalized and are reported in millidegrees. Protein All experiments in this study were repeated three to fi ve times to ensure reproducibility. progressive increase in the particle size upon increasing SAA:apoA-I molar ratio from 0.5:1 to 2:1 to 4:1 ( Fig. 3A , C ). At 0.5:1 ratio, all protein comigrated with HDL on NDGE, while at 2:1 and 4:1 ratios, both HDL-bound and dissociated protein was detected ( Fig. 3A, B ).
Next, excess protein was removed by SEC from SAA-HDL that was prepared at 2:1 protein ratio as described in supplementary Fig. 2A, and the isolated lipoprotein-only fraction, termed SEC-HDL (2:1), was analyzed ( Fig. 3A , right lane). Western blotting showed that SAA-HDL and SEC-HDL contained both SAA and apoA-I, and that the dissociated protein also contained both SAA and apoA-I that was displaced from HDL ( Figs. 1, 3B ). SDS-PAGE confi rmed these results and showed that SAA displaced a large fraction of apoA-I but not apoA-II from HDL (supplementary Fig. 2B). As a result, SAA and apoA-I became the major proteins in HDL-bound and in dissociated fractions. Protein quantifi cation by SEC and immunoblotting shows that apoA-I is progressively displaced from HDL upon incubation with SAA in a dose-dependent manner ( Fig. 1 ); at 4:1 mol/mol SAA:apoA-I, up to two-thirds of total apoA-I is displaced ( Fig. 1D , bottom). These results agree with previous  to fusion and rupture of nHDL and of HDL that have been incubated with SAA were very similar, indicating similar structural stability of these particles ( Fig. 4A ).
Apart from the high-temperature transitions, DSC data of SAA-HDL showed an additional peak near 39°C ( Fig. 4A ). This peak did not involve irreversible changes in lipoprotein morphology or substantial protein dissociation, as indicated by NDGE of SAA-HDLs that have been heated to 50°C ( Fig. 4B ), although some protein dissociation was detected by immunoblotting. Notably, this peak was detected only in SAA-HDL, and its amplitude progressively increased upon increasing SAA concentration ( Fig. 4A ), suggesting that a structural transition in SAA was involved. An alternative explanation (i.e., that the peak refl ected a transition in apoA-I that was progressively displaced from HDL) was dismissed because the lipid-free or lipid-poor apoA-I undergoes structural transitions at much higher temperatures approaching 60°C ( 59 ) . Consequently, the 39°C peak refl ected a structural transition in SAA that interacted with other HDL constituents. This SAA was probably bound to HDL, as suggested by the very similar peak observed in SEC-HDLs that contained only HDL-bound protein (SEC-HDL (2:1) in Fig. 4A ; intact SEC-HDL (2:1) in Fig. 4B ). The peak temperature and its shape did not change much upon increasing SAA concentration ( Fig. 4A , vertical line), suggesting that this transition may involve either monomeric or oligomeric SAA.
In sum, we show for the fi rst time that HDLs that have been incubated with SAA undergo fusion, rupture, and apolipoprotein dissociation that are similar to those in nHDLs and occur at similar temperatures. Consequently, despite the displacement of up to two-thirds of apoA-I with SAA, the structural stability and remodeling of the resulting SAA-HDL and nHDL are very similar. Another novel result is the calorimetric transition in SAA-HDL at nearphysiological temperatures, which probably refl ects structural transitions in SAA that interacts with HDL.
The structural basis for these transitions was further probed by CD spectroscopy ( Fig. 5 ). Far-UV CD spectra in in vitro studies in SAA-HDL, as well as with protein analysis in acute-phase plasma HDL reporting that up to threefourths of the total apoA-I can be displaced ( 52,53 ). Thus, the protein composition in our in vitro studies in SAA-HDL is representative of the acute-phase plasma HDL.
Next, thermal stability and remodeling of SAA-HDL and nHDL were analyzed by DSC and NDGE ( Fig. 4 ). Previously, we showed that HDL heating above 80°C leads to two irreversible kinetically controlled transitions: i ) fusion into larger HDL-like particles accompanied by dissociation of a fraction of apoA-I; and ii ) rupture of these particles, that is, lipoprotein disintegration, release of core lipids, and dissociation of lipid-poor apoA-I and apoA-II. Biochemical analysis of the products of this HDL transition was reported previously ( 48,59 ). Consistent with these studies, DSC data in Fig. 4A show two high-temperature peaks in nHDL, which correspond to lipoprotein fusion and rupture. SAA-HDL showed similar DSC transitions ( Fig. 4A ) whose nature was confi rmed by NDGE, immunoblotting, and EM. For example, SEC-HDLs have been heated to 50°C (well below the fi rst high-temperature DSC peak), 95°C (above the fi rst but below the second peak), or 110°C (above the second peak). NDGE showed that heating to 50°C did not cause HDL fusion, although some dissociation of SAA and apoA-I was detected by immunoblotting (not shown). Heating to 95°C led to increase in the particle size due to fusion accompanied by dissociation of a larger protein fraction (detected by NDGE; Fig. 4B ) that comprised SAA and apoA-I (Western blot not shown). Further heating to 110°C led to lipoprotein disintegration and lipid coalescence into droplets, as indicated by NDGE ( Fig. 4B ) and EM (not shown). In sum, similar to nHDL, SAA-HDL also undergoes irreversible fusion, rupture, and apolipoprotein dissociation at high temperatures.
Compared with nHDL, SAA-HDL showed reduced amplitudes of these high-temperature DSC transitions, which refl ected reduced amounts of apoA-I in these particles ( Fig. 2 ) and perhaps less extensive apoA-I-lipid contacts, leading to reduced enthalpic contribution from apoA-I dissociation and unfolding. Surprisingly, the peak positions corresponding Fig. 4. Thermal denaturation of SAA-HDL analyzed by DSC and NDGE. A: DSC data of SAA-HDL were recorded during heating from 5°C to 120°C at a rate of 90°C/h. The samples contained 0.8 mg/ml apoA-I and various amounts of SAA at indicated SAA:apoA-I molar ratios. SEC-HDL is the lipoproteinonly fraction isolated by SEC from the total SAA-HDL (2:1 SAA:apoA-I). The data are shifted along the yaxis to avoid overlap. B: NDGE of SEC-HDLs that were intact or heated to 50, 95, or 110°C. Heating to 50°C does not cause HDL remodeling. Heating to 95°C causes partial protein dissociation and HDL fusion; Western blotting detects SAA and apoA-I both in fused HDL and in the dissociated protein (data not shown). Heating to 110°C causes lipoprotein disruption and coalescence into protein-containing lipid droplets that do not enter the gel. For more detailed biochemical analysis of the products of HDL thermal denaturation, see ( 59 ). only HDL-bound protein, such as SAA-HDL (0.5:1) ( Fig. 5B ) or HDL-only fraction isolated by SEC from SAA-HDL (2:1) (supplementary Fig. 3). This suggests that excess protein is involved in this transition and explains why it was missed in the previous studies that used isolated SAA-HDL.
To test the involvement of SAA in this novel transition, we used human apoA-II that, similar to SAA, can displace apoA-I from HDL [( 60 ) and references therein]. HDLs have been incubated with apoA-II at 37°C for 3 h using 2:1 mol/mol apoA-II:apoA-I, and CD melting data were recorded (supplementary Fig. 4). Although significant amounts of apoA-I were displaced with apoA-II from these HDLs, the helical transition at near-physiological temperatures was lacking (supplementary Fig. 4), indicating that SAA is necessary for this transition to be observed.

Thermal denaturation of SAA:DMPC complexes
To further test the effect of lipid binding on SAA structure and stability, we reconstituted binary complexes of SAA with DMPC, termed SAA:DMPC. NDGE showed that all protein was incorporated into discrete complexes that were 9-19 nm in size, and negative-stain EM detected discoidal lipoproteins and small vesicles in this size range ( Fig. 6A ). Far-UV CD spectra of SAA:DMPC complexes indicated ‫ف‬ 50% ␣ -helix content at 5°C and at 25°C. Therefore, binding to DMPC stabilizes ␣ -helical structure in SAA ( Fig. 6B ) as compared with free protein ( Fig. 2A , insert). To test the stability of this structure, CD melting data were recorded at 222 nm from SAA:DMPC complexes during heating and cooling from 5°C to 98°C; similar data of apoA-I:DMPC complexes are shown for comparison ( Fig. 6C ). The heating curves showed cooperative ␣ -helical unfolding in SAA:DMPC between 45°C and 55°C versus 80°C-90°C in Fig. 5A were recorded at 25°C from nHDL and from SAA-HDL by using a constant HDL concentration and increasing SAA to apoA-I ratios. The results showed formation of additional ␣ -helical structure in SAA in the presence of HDL at 2:1 and 4:1 mol/mol SAA:apoA-I. Thus, in contrast to lipid-free SAA that is largely unfolded at 25°C ( Fig. 2A ), SAA acquires ␣ -helical structure in the presence of HDL. Because ␣ -helical content in human apoA-I is nearly 80% on HDL and 60% in solution, our CD data suggest that, at 4:1 molar (or 1.7:1 weight) ratio of SAA to apoA-I in SAA-HDL, the ␣ -helical content in SAA reaches ‫ف‬ 35%. To probe the stability of this helical structure, heating and cooling data were recorded from SAA-HDL by CD at 222 nm ( Fig. 5B ). All samples showed irreversible thermal unfolding above 80°C, which corresponds to HDL fusion and dissociation and unfolding of an apoA-I fraction; this result is in good agreement with previous studies in nHDL ( 48 ). Furthermore, CD melting data indicate that nHDL and SAA-HDL undergo fusion at similar temperatures ( Fig. 5B ), in agreement with the DSC results ( Fig. 4A ).
Unexpectedly, SAA-HDL at 2:1 and 4:1 ratios showed an additional ␣ -helical unfolding transition between 30°C and 40°C ( Fig. 5B , marked by an asterisk). In contrast to irreversible HDL fusion and rupture at higher temperatures, this structural transition was largely reversible upon heating and cooling to 60°C ( Fig. 5C ). The latter is consistent with the absence of irreversible changes in HDL size at 50°C observed by NDGE ( Fig. 4B ) and EM (not shown). This novel ␣ -helical transition at near-physiological temperatures increased in amplitude but not in temperature upon increasing the SAA:apoA-I ratio from 2:1 to 4:1. Interestingly, this transition was observed only in samples containing both HDL-bound and excess SAA. It was not detected in samples containing

Thermal transition in SAA bound to POPC vesicles
Another model system used to test the effects of lipids on SAA conformation were SUVs of POPC. This zwitterionic phospholipid is an established model for the unsaturated phosphatidylcholines (PCs) found in HDL and in the plasma membrane. POPC SUVs were incubated with either SAA or apoA-I at 1:100 protein:lipid molar ratio at which the vesicle surface is expected to be saturated with the bound protein ( 54 ). NDGE of the incubation mixtures showed that most protein was bound to vesicles that were more than 20 nm in size ( Fig. 7A , SAA-SUV). In addition, a small fraction of SAA remodeled SUVs into HDL-size particles (8-12 nm) that were similar in size to SAA:DMPC complexes. This fraction was not seen in apoA-I that does not spontaneously remodel POPC SUVs at a high enough rate; instead, excess apoA-I migrated as free protein ( Fig. 7A , apoA-I-SUV). Far-UV CD spectra of SAA-SUVs suggested 38% ␣ -helix at 5°C and at 25°C ( Fig. 7B ), and the CD melting data showed that this helical structure unfolded upon heating with T m,app ‫ف‬ 50°C and gradually refolded upon cooling to near-physiological temperatures ( Fig. 7C ). Similar experiments using apoA-I-SUV incubation mixture showed two unfolding transitions, one at 60°C and another at 85°C, which refl ect the respective unfolding of free and SUV-bound apoA-I ( Fig. 7C ). This comparison illustrates once again that SAA:lipid complexes are less stable than their apoA-I-containing counterparts. Further, similar to other protein-lipid complexes, SAA-SUVs show hysteresis and scan rate dependence in their heating and cooling data, indicating a thermodynamically irreversible transition ( Fig. 7D ).

DISCUSSION
This is the fi rst calorimetric analysis complemented by farand near-UV CD analysis of SAA in lipid-free and HDL-bound apoA-I:DMPC, indicating much greater structural stability of the latter complexes. These results agree with a recent study showing that complex formation with DMPC increases the resistance of SAA to proteolysis and induces ␣ -helical structure that unfolds upon heating near 50°C ( 61 ).
Notably, CD melting data of SAA:DMPC and apoA-I:DMPC showed hysteresis and incomplete signal recovery upon heating and cooling ( Fig. 6C ), indicating that the unfolding transition was thermodynamically irreversible. In contrast to the helical unfolding upon heating, which was centered at 49°C, helical refolding upon cooling occurred at near-physiological temperatures in SAA:DMPC ( Fig. 6C ). NDGE shows that heating and cooling of SAA:DMPC complexes beyond 90°C leads to their remodeling and dissociation of some SAA ( Fig. 6A ). Similar studies in other lipoproteins also showed a hysteresis and thermodynamic irreversibility, which was rooted in the partial unfolding and dissociation of lipid-poor apos accompanied by lipoprotein fusion ( 51,59,62 ). The results reported here suggest that similar effects occur in SAA:DMPC complexes.
Furthermore, the unfolding transition in SAA:DMPC complexes shifted to lower temperatures at slower heating rates: the apparent melting temperature, T m,app , decreased from 49°C to 47°C upon slowing down the heating from 70°C to 6°C/h ( Fig. 6D ). This small but signifi cant shift suggests low Arrhenius activation energy, E a , of lipoprotein denaturation ( 62 ). Larger heating rate effects were observed in other similar complexes, indicating high activation energy of their denaturation ( E a = 50 kcal/mol for apoA-I: DMPC disks, with lower values observed for the complexes containing smaller proteins) ( 51 ). In sum, compared with apoA-I:DMPC, SAA:DMPC complexes show lower structural stability and lower apparent activation energy of denaturation, suggesting that SAA forms less extensive interactions with the lipid surface than apoA-I. HDL remodeling by plasma factors involves dissociation of lipid-poor apoA-I and HDL fusion (mediated by LCAT, hepatic lipase, PLA 2 , PLTP, or CETP), followed by HDL rupture and release of core lipids (mediated by SR-BI receptor) and dissociation of the remaining apos (40)(41)(42)(43)(44). Therefore, our fi nding of similar structural stability in nHDL and in HDL that have been enriched with SAA is consistent with the fi nding that such enrichment per se does not reduce HDL levels in plasma ( 64 ). Together, these fi ndings support the idea that decreased HDL levels in acute-phase response result from HDL hydrolysis by sPLA 2 rather than from HDL enrichment with SAA ( 26 ).
In contrast to plasma HDL, binary complexes of DMPC with apoA-I, which are established mimics of nascent HDL, are much more stable than those with SAA. Compared with apoA-I:DMPC, SAA:DMPC complexes undergo lipoprotein fusion and protein dissociation at much lower temperatures ( T m,app ‫ف‬ 85°C for apoA-I:DMPC vs. 50°C for SAA:DMPC; Fig. 6C ) and have lower apparent activation energy evident from the small scan rate effects on their melting data ( Fig. 6D ). These results suggest less extensive protein-lipid interactions in SAA:DMPC than in apoA-I:DMPC complexes, which can perhaps be explained by a relatively small SAA-lipid contact area involving the N-terminal ␣ -helix that forms the primary HDL binding site in SAA ( 9 ) as compared with the much larger lipid binding surface formed by the apolar faces in the apoA-I ␣ -helices that encircle HDL ( 21,23,65,66 ).
Why do HDLs that have been incubated with SAA and nHDL have similar stability, whereas SAA:DMPC complexes are much less stable than their apoA-I-containing counterparts? The probable explanation is that, in contrast to SAA:DMPC complexes, SAA-HDLs contain not only SAA but also other proteins, mainly apoA-I and apoA-II. To explain why HDL stability does not signifi cantly change upon forms and the fi rst report on structural stability of SAA-containing human HDL. For free SAA1.1, we report a DSC transition near 39°C ( Fig. 2C ), which involves changes in the oligomerization degree of SAA that is devoid of ordered secondary structure at these temperatures ( Fig. 2A ). For SAA-HDL, we report two surprising fi ndings. First, irreversible fusion and rupture of HDL, which occur above 80°C, are very similar in nHDL and in HDL that have been incubated with SAA in the wide range of SAA:apoA-I ratios explored, from 0:1 to 4:1 mol/mol ( Fig. 3 ), which encompass the range found in vivo in normal and in acute-phase human plasma ( 28,29,31 ). Second, SAA-HDL shows a largely reversible cooperative helical folding/unfolding at near-physiological temperatures ( Fig. 5B, C ). This novel structural transition is observed in samples containing both excess protein and HDL and probably involves SAA that interacts with lipid. These two new fi ndings are discussed in detail below.

Effects of apoA-I displacement on HDL stability
Surprisingly, the high-temperature transitions in human plasma HDL do not signifi cantly change upon displacement of a large fraction of apoA-I ( Figs. 4,5 ). Even at 4:1 SAA:apoA-I molar ratio when roughly two-thirds of the total apoA-I is displaced ( Fig. 1 ) ( 29, 63 ), the HDL transition temperatures remain nearly invariant ( Fig. 4A ). Therefore, the displacement of the exchangeable fraction of apoA-I has little effect on HDL stability to fusion and rupture. Consequently, the structural stability of human plasma HDL is determined primarily by the nonexchangeable fraction of apoA-I.
Previous studies proposed that structural remodeling of HDL induced by various perturbations (heating, denaturants, detergents, serum factors, etc.) mimics important aspects of HDL remodeling during reverse cholesterol transport ( 47,48,59 ). In fact, similar to biophysical perturbations, Fig. 7. Characterization of SAA complexes formed upon incubation with SUVs of POPC. SAA was incubated with POPC SUVs at 1:100 protein:lipid molar ratio to form SAA-SUVs. A: NDGE shows that all SAA was bound to lipid, either to SUV ( ‫ف‬ 22 nm) or to SAA:POPC complexes in the HDL size range (9-11 nm). A similar sample containing apoA-I and POPC SUV, along with lipid-free apoA-I and SAA are shown for comparison. B: Far-UV CD spectra of SAA-SUV samples suggest similar helix content at 5°C and 25°C. C: CD heating and cooling data of POPC SUVs that have been incubated with SAA (black) or with apoA-I (gray) were recorded by CD at 222 nm. In intact samples, all SAA is lipid-bound, while apoA-I is distributed between lipid-bound and dissociated fractions whose thermal unfolding transitions are indicated. The data are in units of molar residue ellipticity, [ ⌰ ]. D: Scan rate effects on the melting data of SAA-SUVs. The CD data at 222 nm were recorded from two identical samples at a fast (70°C/h) and a slow (6°C/h) rate. dimer ( Fig. 8 ). Regardless of the exact structural model, the DSC and CD results reported here suggest that this nonexchangeable apoA-I is the key determinant of HDL stability to fusion and rupture ( Figs. 4A, 5B ).
Our stability studies are consistent with functional studies reporting that acute-phase HDLs have essentially intact functions in cell cholesterol effl ux and CETP-mediated lipid transfer ( 15 ). Together, these studies suggest that the nonexchangeable fraction of apoA-I is the major determinant of the structural stability and functional remodeling of HDL in reverse cholesterol transport. We speculate that this important fraction is a double-belt apoA-I dimer.

SAA folding/unfolding at near-physiological temperatures
Another surprising fi nding of this study is the reversible ␣ -helical folding/unfolding in SAA detected by CD at nearphysiological temperatures in the presence of HDL ( Fig. 5 ,  asterisk). This helical transition is distinct from the calorimetric transition observed in lipid-free SAA at ‫ف‬ 36°C ( Fig.  2C ), which involves changes in the oligomerization degree of SAA that is fully unfolded at near-physiological temperatures in the absence of the lipid ( Fig. 2A ). In contrast, SAA in the presence of HDL shows substantial ␣ -helical structure ( ‫ف‬ 34%) that unfolds with the midpoint at ‫ف‬ 38°C (asterisk in Fig. 2B, C ). Our results suggest that this helical transition involves SAA interacting with the lipid surface and occurs only in samples containing excess protein. In fact, this helical transition is not detected in lipid-free SAA ( Fig. 2A ) or in the lipid-containing samples that lack excess displacement of a large fraction of apoA-I with SAA, we propose a hypothetical structural model of SAA-HDL. This model is based on the double-belt conformation of apoA-I on HDL, which is widely accepted in the fi eld and is supported by extensive experimental evidence, including the two available X-ray crystal structures of truncated forms of apoA-I ( 21,23,65,66 ). We also rely on the modifi ed trefoil/ tetrafoil model fi rst proposed by Davidson's group to explain how multiple copies of apoA-I can adapt their conformation to the surface of spherical HDL (67)(68)(69).
One possible model of intact HDL is illustrated in Fig. 8 (left). It shows four copies of apoA-I in a putative tetrafoillike arrangement. SAA initially binds to the available HDL surface without displacing apoA-I (0.5:1 molar ratio). As SAA:apoA-I ratio increases, a progressive amount of apoA-I is displaced with SAA, until most exchangeable apoA-I is displaced at about 4:1 ratio ( Fig. 8 , left to right).
The presence of two populations of apoA-I on HDL, only one of which is readily exchangeable, was previously proposed to refl ect the presence of two distinct apoA-I conformations [ ( 60,70,71 ) and references therein]. Further, previous studies in model lipoproteins showed that smaller apos are generally more exchangeable than their larger counterparts; for example, under otherwise identical conditions human apoA-II monomer dissociates from the lipid much faster with only half the activation energy as compared with apoA-II dimer ( 51 ). Therefore, it is reasonable to assume that the exchangeable population of apoA-I on plasma HDL corresponds to protein monomer, while the nonexchangeable population corresponds to the double-belt Fig. 8. Hypothetical structural model of mature plasma HDL containing increasing amounts of SAA. The model that exemplifi es nHDL shows four copies of apoA-I, including one molecular dimer in a double-belt conformation and two monomers in a putative double-hairpin conformation. For details of these conformations and alternative models, see ( 69 ) and references therein. The SAA:apoA-I molar ratios are indicated. Addition of SAA initially leads to its binding to the available HDL surface (0.5:1 SAA:apoA-I mol/mol), followed by progressive displacement of the exchangeable fraction of apoA-I. We propose that this readily exchangeable apoA-I is probably monomeric, while the less exchangeable apoA-I that determines HDL stability is dimeric. HDL incubation at 4:1 molar (or 1.7:1 weight) ratio of SAA:apoA-I, which displaces approximately two-thirds of apoA-I from HDL ( Fig. 1 ), must be compensated by binding of at least an equivalent amount of SAA in order to maintain the particle density in the HDL 3 range. Thus, the HDL-bound proteins comprise roughly two-thirds SAA and one-third apoA-I by weight, which resembles their ratio in the incubation mixture. Such a ratio of bound SAA to apoA-I can be achieved if HDL contains 2 apoA-I molecules and 8 to 10 SAA molecules (right panel). protein, such as 0.5:1 SAA:apoA-I (when all protein is bound to HDL; Fig. 5B ) or SEC-HDL from which the excess protein has been removed by SEC (supplementary Fig.  3). It is also not observed in the apoA-II-HDL sample despite the presence of excess apoA-I and HDL (supplementary Fig. 3), clearly indicating that SAA is involved.
These observations suggest that interactions of SAA with HDL and perhaps with the dissociated lipid-poor apoA-I can induce formation of a marginally stable helical structure in this intrinsically disordered protein. We hypothesize that reversible folding/unfolding of this helical structure at near-physiological temperatures ( Fig. 5C ) may modulate functional interactions of SAA with other proteins at the lipid surface and/or infl uence amyloid deposition. Studies in other proteins reported that, in contrast to the stable ␣ -helical structure that inhibits amyloid formation, marginally stable ␣ -helices (such as those formed in amyloid-␤ peptide and other intrinsically disordered proteins at the lipid surface) tend to promote intermolecular interactions leading to ␣ -helix to ␤ -sheet conversion in amyloid (72)(73)(74).
The results of the current study prompt us to speculate that the precarious ␣ -helical structure formed in SAA1.1 at the lipid surface may modulate local functional interactions of SAA at the sites of infl ammation as well as its pathological deposition in amyloid.