The Orientation of Helix 4 in Apolipoprotein A-I-containing Reconstituted High Density Lipoproteins*

The three-dimensional structure of the high density lipoprotein (HDL) component apolipoprotein (apo) A-I and the molecular basis for its protection against coronary artery disease are unknown. In terms of discoidal HDL particles, there has been a debate as to the orientation of the apoA-I α-helices around the disc edge. The “picket fence” model states that the α-helical repeats, separated by turns, are arranged parallel to the phospholipid acyl chains of the enclosed lipid bilayer. On the other hand, the “belt” model states that the helical segments run perpendicular to the acyl chains. To distinguish between these models, we used nitroxide spin labels present at various depths in the bilayer of reconstituted HDL (rHDL) to measure the position of Trp residues in single Trp mutants of human proapoA-I. Two mutants were studied; the first contained a Trp at position 108, which was located near the center of helix 4. The second contained a Trp at position 115, two turns along the same helix. The picket fence model predicts that these Trp residues should be at different depths in the bilayer, whereas the belt model predicts that they should be at similar depths. Different sized rHDL particles were produced that contained 2, 3, and >4 molecules of proapoA-I per complex. In each case, parallax analysis indicated that Trp-108 and Trp-115 were present at similar depths of about 6 Å from the center of the bilayer, consistent with helix 4 being oriented perpendicular to the acyl chains (in agreement with the belt model). Similar experiments showed that control transmembrane peptides were oriented parallel to the acyl chains in vesicles, demonstrating that the method was capable of distinguishing between the two models. This study provides one of the first experimental measurements of the location of an apoA-I helix with respect to the bilayer edge.

mediator of high density lipoprotein (HDL) function over 25 years ago (1), little progress has been made in defining the detailed mechanisms underlying its well known protection against coronary artery disease. HDL and apoA-I are thought to be critical for the uptake of cellular cholesterol in the periphery and, after interactions with numerous plasma factors, for its delivery to the liver and adrenals for bile acid and steroid hormone synthesis, respectively. Arguably, the most prominent obstacle to a molecular understanding of these events is the lack of information on how the structure of apoA-I modulates HDL function. Most of our knowledge of apoA-I conformation was generated within the first few years of the publication of its primary structure (2), largely from theoretical analysis of the amphipathic ␣-helical repeats in the sequence (3,4). Circular dichroism and infrared spectroscopy (IR) have shed light on the rough secondary structural contents of various forms of apoA-I (5, 6), but hard experimental evidence concerning the threedimensional arrangement of these elements in the native protein is lacking.
Recently, an x-ray crystal structure of a lipid-free fragment of apoA-I was reported (7). The structure was a homotetramer of highly ␣-helical apoA-I molecules arranged in a ring with the monomers associating via the hydrophobic faces of their amphipathic ␣-helices. While the applicability of this structure to the native lipid-free forms of apoA-I is undetermined, this advance has rekindled an old debate on the structure of nascent, discoidal HDL particles. It is widely accepted that discoidal HDL particles reconstituted in vitro (rHDL) are composed of a patch of phospholipid/cholesterol bilayer stabilized at its edges by the amphipathic ␣-helices of apoA-I. Segrest et al. and others (8,9) proposed early on that the ␣-helices were arranged around the circumference of the disc with the long axis of the helices perpendicular to the phospholipid (PL) acyl chains. This became known as the "belt" or "bicycle wheel" model. However, other investigators theorized that the 22 aa helical repeats were an ideal length to traverse the bilayer edge with the helices parallel to the acyl chains (10). This "picket fence" model has been favored in recent years because of supporting IR spectroscopy studies (6,11) and the model's superb ability to account for the experimentally observed size classes of apoA-I containing rHDL (12). Borhani et al. (7) proposed that the ring shaped structure observed in their crystal was new evidence that the belt model may be correct. This view has been supported in recent months by methodologically updated IR experiments (13). A detailed computer analysis of the belt model has also been recently published (14).
To provide information on the three-dimensional structure of apoA-I, we have taken the approach of combining fluorescence spectroscopy with single Trp mutants of human proapoA-I. Our initial studies focused on lipid-free apoA-I (15). However, in light of the picket fence versus belt model debate and the conflicting IR data, our current study utilized a simple strategy to distinguish between the two models in discoidal rHDL par-ticles. We employed lipid-based quenching studies using rHDL reconstituted with the single Trp mutants and synthetic phospholipids containing nitroxide spin labels at various positions down the length of the fatty acyl chain. These labels can quench Trp residues in lipid interacting areas of proteins and are used in conjunction with a technique called the parallax analysis. This method has been used to accurately measure the depth of Trp penetration in transmembrane proteins (16 -19). We have extended this analysis to the situation of apoA-I wrapping around a patch of lipid bilayer in rHDL particles. Our hypothesis was straight forward. If the picket fence model is correct, Trp residues at the center of a helical domain should be measured near the center of the bilayer in rHDL particles, whereas those near the ends of the domain should be considerably more shallow (see Figs. 1 and 2). If the belt model is correct, Trp residues should be measured at similar bilayer depths regardless of their location on the helix.
The results of this study suggest that at least helix 4 of apoA-I exists in the belt conformation in rHDL of various sizes. We believe this study provides one of the first measurements of the location of an apoA-I helix with respect to the bilayer edge.

Mutagenesis, Protein Expression, and Purification-Human
proapoA-I (containing the 6-aa prosegment) was used in this study because of its high expression level in bacteria and the lack of structural or functional differences from mature plasma apoA-I (20). To generate single Trp mutants, each of the five naturally occurring Trp residues (positions Ϫ3, 8, 50, 72, and 108) in proapoA-I were converted to Phe. This yielded a mutant that contained no Trp residues (W@). W@ was then used as a template to re-insert a Trp residue at position 108 (W@108) and, in a separate mutant, to convert Tyr-115 to a Trp (W@115). The site-directed mutagenesis was accomplished using either standard polymerase chain reaction-based techniques or the Quick-Change TM mutagenesis kit (Stratagene). The sequence of each construct was verified on an Applied Biotechnology System DNA sequencer. The resulting constructs were ligated into the PET-30 vector (Novagen) directly downstream of a histidine tag (His tag) sequence and transfected into BL-21 (DE3) Escherichia coli cells. The overexpression of the mutant proapoA-I was performed using freshly transfected cells as described previously (15). After harvesting, the cells were resuspended in 10 mM Tris buffer, pH 8.0, containing 0.15 M NaCl (standard Tris buffer, STB). The cells were lysed for 1 h at room temperature using B-PER lysis detergent (Pierce) at 4 ml of detergent solution for every 100 ml of bacterial cell culture. The soluble cell contents were applied to a His bind column (Novagen) and eluted according to manufacturer's instructions. The His-tagged proteins were dialyzed into 10 mM ammonium bicarbonate buffer, pH 8.0, and lyophilized. Lyophilized proteins were solubilized in 3 M guanidine HCl, dialyzed into STB to make rHDL particles (see below). The His tag was then removed by cleavage with 0.5 unit of enterokinase per 250 g of proapoA-I at 25°C for 16 h. The intact proapoA-I containing rHDL particles were purified away from the enterokinase and the cleaved His tag by gel filtration chromatography on a Superdex 200 (Amersham Pharmacia Biotech) column. Proteins prepared by this method were Ͼ95% pure as visualized by SDS electrophoresis stained with Coomassie Blue and the yields were about 5 mg of protein/100 ml of original culture.
Preparation and Characterization of the Lipid-containing Particles-rHDL particles were prepared as described previously (21) using POPC, mutant proapoA-I and spin-labeled phospholipids. All phospholipid stock solutions were assayed before the reconstitution by the phosphorus method of Sokoloff and Rothblat (22). Initial lipid to protein molar ratios were 85:15:1 (POPC:DOXYL-PSPC:proapoA-I) to isolate the 98and 160-Å particles and 34:6:1 (POPC:DOXYL-PSPC:proapoA-I) to isolate the 78-and 90-Å particles. For each mutant, a set of particles was reconstituted without the spin-labeled phospholipids. The different particle sizes were isolated by gel filtration chromatography on a Superdex 200 gel filtration column (23) to remove any unreacted protein and vesicular structures. To ensure that the gel filtration step did not change the concentration of quencher within the particles, we performed control experiments using rHDL preparations that contained 12-NBD-PC. The various DOXYL quenchers were able to quench this label identically whether in complexes studied prior to or after gel filtration chromatography. The phosphorus assay and the Markwell modification of the Lowry protein assay (24) determined the final lipid and protein concentrations. Particle hydrodynamic diameters were measured by gradient native polyacrylamide electrophoresis (23). The number of apolipoprotein molecules per rHDL complex was determined by BS 3 cross-linking (25). The secondary structure content of the proapoA-I in the rHDL particles was estimated by circular dichroism at 222 nm (23).
Small unilamellar vesicles (SUV) containing the transmembrane peptides were prepared as by Ren et al. (26). For all peptide experiments, the lipid to protein molar ratios were 68:12:1 (POPC:DOXYL-DSPC:peptide). The appropriate lipids in chloroform were dried under nitrogen for at least 2 h. The appropriate amount of peptide was applied to the lipids in a volume of 10 l absolute ethanol. The lipid/peptide mixture was vortexed, incubated on ice for 10 min, and then diluted to 1 ml with filtered STB. The small amount of ethanol in the mixture has been shown not to interfere with the fluorescence characteristics of the sample (18). The final concentrations of lipid in the samples were 200 M. Identical samples were made containing POPC SUV only (blank for fluorescence runs) and POPC SUV with peptide in the absence of spin label (for determination of F o ).
Fluorescence Spectroscopy-All fluorescence measurements were performed on a Photon Technology International Quantamaster spectrometer in photon counting mode. The emission and excitation band passes were 3.0 nm. The excitation wavelength for all Trp studies was 295 nm to minimize the contribution of tyrosine fluorescence in proapoA-I. The samples (in STB for all studies) were measured at 25°C in a semimicro quartz cuvette. The emission spectra from 303 to 375 nm were corrected for the background fluorescence of either buffer alone (rHDL samples) or SUV alone in buffer (peptide samples). The spectra were not corrected for the spectral characteristics of the emission and excitation monochromators.
Depth Calculations, Theory-The parallax method for determining the depth of penetration of a fluorophore into a lipid bilayer was derived by Chattopadhyay and London (27) from classical relationships of static quenchers to randomly distributed fluorophores. The method depends on the differences in fluorescence intensities of the fluorophore in the presence of known concentrations of nitroxide quencher groups at known locations in a phospholipid acyl chain. Since quenching is a distance-dependent phenomena, the quenching ratio from two quenchers, one shallow and one deep in the membrane, are used calculate the relative depth of the fluorophore relative to the quenchers. The method was initially applied to lipid-based fluorescent groups that are exposed to quencher on all sides and both leaflets of the bilayer. The theory was later modified to allow for the quenching of Trp residues on transmembrane peptides (18). The distance of the Trp from the center of the bilayer (Z cf ) is given by, where F s is the fluorescence intensity in the presence of the shallow quencher, F d is the same for the deep quencher, L cs is the distance from the center of the bilayer to the shallow quencher, L ds is the distance between the shallow and deep quenchers, and C is the concentration of the quencher molecules per Å 2 . Equation 1 holds when the Trp residue is quenched by nitroxide groups in the same leaflet of the membrane as the Trp. However, in the case where the Trp is deeply buried within the membrane, it is subjected to quenching from groups in the opposing leaflet of the membrane as well. For this situation, a second relationship is required to account for trans-bilayer quenching, where F o is the fluorescence intensity in the absence of quencher, L cd is the distance from the center of the bilayer to the deep quencher, and L ds is the distance between the shallow and deep quenchers. For a detailed discussion and the derivation of these equations, see Refs. 18 and 33. Equation 2 was used in this study whenever a Trp residue was deter-mined to be Ͻ5 Å from the center of the membrane by Equation 1. These equations have been demonstrated to accurately describe the behavior of single Trp transmembrane helices in model membrane systems (18). Although, in theory, any pair of quenchers can be used for the analysis, the quenching pair of C-5 and C-12 was chosen for our calculations for two important reasons. First, in the case of our rHDL particles, these two quenchers gave the largest difference in quenching efficiencies (Fig. 3). Where the data was available, we also calculated depth using the C-5/C-10 and the C-5/C-7 quencher pairs. While these gave similar mean values, the standard deviations among the values between samples were consistently 2-3-fold higher than with the C-5/ C-12 quencher pairs. Second, Abrams and London (17) have published detailed information on the quenching of 12-NBD phospholipids using the C-5/C-12 pair. The NBD quenching data allowed the calibration of our particular batches of spin-labeled lipid to estimate Trp depths as accurately as possible (see below). For the quenchers at positions 5 and 12, the relevant depth values were: L cs ϭ 12.15 Å, L ds ϭ 6.3 Å, and L cd ϭ 5.85 Å.
The C term is the mole fraction of spin-labeled lipids divided by the average area occupied by a lipid molecule, 70 Å 2 (17). For a fluorophore that is surrounded on all sides by quencher (as is the case for the transmembrane peptides used in this study), C was calculated using the mole fraction value of 0.15 (the fraction of label for all particles in this study). However, in the case of the rHDL particle, the Trp residue is on the edge of the bilayer and is only exposed to the lipid on one side. Since the diameter of a phospholipid molecule is small compared with the diameter of the rHDL disc, the Trp in a rHDL particle is expected to have a lipid exposure that is about 50% that of a free floating peptide. Thus, a mole fraction of 0.075 was used to calculate C for all of the rHDL particles in this study.
Spin Label Assay-The spin label content of the purchased spinlabeled phospholipids was measured as suggested by Abrams and London (17). These authors have reported F/F o values for the quenching of 12-NBD-PC in PBS at pH 7 in SUV in the presence of the DOXYL-5 and DOXYL-12 phospholipids present at 10 mol %. The spin label/phosphorus ratio for the particular batch of spin labels had been previously calibrated against known standards by electron spin resonance (ESR) spectroscopy. As an ESR machine was not available for calibration in our work, we performed the same quenching experiments. We determined that our batch of lipids contained an average of 0.95 spin labels/ mol of phosphorus. This factor was taken into account during the reconstitutions of the rHDL and SUV used in this study.

RESULTS
Experimental Design-To determine the orientation of apoA-I helices with respect to the lipid bilayer, we produced two single Trp mutants of proapoA-I, one containing a Trp at position 108 and the other containing a Trp at position 115. Both of these sites occur in the putative helical repeat 4 (as numbered by Roberts et al. (28)). This region was chosen because (a) it already contained a naturally occurring Trp at position 108 and (b) repeat 4 is punctuated on both ends with a Pro residue and is universally predicted to exist entirely as a ␣-helix in rHDL particles (29 -31). Both Trp residues at pos. 108 and 115 are predicted to be in the center of the hydrophobic face of the helix as one looks down the long axis of the repeat (Fig. 1) and likely interact with lipid. However, when one looks perpendicular to the long axis of the helix (from a side view, as in Fig. 2), Trp-108 is near the center of the helix, roughly equidistant from the N-and C-terminal Pro residues that punctuate the repeat. Trp-115, however, is two helical turns Cterminal to Trp-108, about 12 Å closer to Pro-121. The hypothesized predictions for lipid quenching studies using these two mutants in nitroxide label-containing rHDL are shown in Fig.  2. If the helix is in the picket fence orientation, Trp-108 should be situated near the center of the membrane and Trp-115 should be substantially more shallow. If the helix is in the belt orientation, both Trp residues should exhibit similar depths and will be situated some distance up each leaflet in the bilayer.
In our initial experiments, each of the two single Trp mutants were reconstituted with phospholipids containing quenchers at each of the five commercially available positions along the phospholipid acyl chain (C-5, -7, -10, -12, and -16). A sixth preparation contained POPC only. The first set of reconstitutions was performed at 100:1 (mole of PL:mole of proapoA-I). The reaction mixture produced two populations of rHDL particles of 160 and 98 Å in diameter that were easily separated by gel filtration. To obtain information about particles that contained only two molecules of proapoA-I per particle, a second set of reconstitutions was performed at a PL:proapoA-I molar ratio of 40:1. In this series, the particles were only made with quenchers at positions C-5 and C-12 for the parallax analysis (for cost reasons) resulting in two populations of 93 and 76Å in diameter. These were also isolated by gel filtration.
Characterization of the Various Size rHDL-The use of the Trp mutants in fluorescence studies depends on the assumption that the amino acid replacements do not perturb the structure and function of the WT proapoA-I. In previous work, we have characterized W@108 in detail and determined that it was similar to WT using several indices of protein structure and function (15). Table I lists the physical characteristics of the four particle size classes made with W@108 and W@115 in this study. Both mutants formed rHDL particles of similar size classes as measured by native polyacrylamide gradient gel electrophoresis. A comparison between the two mutants within a given size class indicates that they were of similar phospholipid to protein molar composition and similar numbers of proapoA-I molecules per particle. Also, within each size class, their fractional ␣-helical contents were similar as measured by far UV circular dichroism. These data are consistent with previous studies in which rHDL particles of these sizes were generated with human plasma apoA-I (23,25), indicating that the mutations in both mutants do not affect the structure and composition of the rHDL particles. It was also important to verify that the presence of the nitroxide spin labels did not adversely affect the rHDL particles. Each value in Table I is  Notice that Trp-108 is shown to be in the center of the hydrophobic face of the helix. Tyr-115, which was converted to a Trp in this study, is two ␣-helical turns C-terminal to Trp-108 (3.6 aa per turn) and is also near the center of the hydrophobic face.
98-and 160-Å particles) for a given mutant in a given size class. A student's t test performed within each set of data failed to identify a significant structural or compositional difference among the particles with or without the DOXYL quenchers, regardless of position. Table I also lists the wavelength of maximum fluorescence ( max ) measured for each complex. All rHDL particles exhibited blue-shifted -max values, which are consistent with the Trp residues being involved in lipid contacts (32). We also determined that W@108 and W@115 in rHDL particles were not significantly quenched by the soluble quencher acrylamide (data not shown). This indicates that the Trp residues were not exposed to the aqueous buffer and, therefore, were appropriate for lipid-based quenching studies.
Fluorescence Quenching of the Lipid-bound Proteins-Once the particles were characterized, the fluorescence intensity of the Trp residue in the absence of quencher was compared with those in the presence of the nitroxide-labeled phospholipids. The results are shown in Fig. 3. In the case of the 98-Å particles (Fig. 3A), the shallowest quencher at carbon 5 (C-5) in the acyl chain minimally quenched the fluorescence of Trp-108. As the quencher depth was increased, Trp-108 fluorescence was increasingly quenched to a maximum near C-12 and then decreased at C-16. The Trp at position 115 followed a nearly identical pattern. The data were remarkably consistent between experiments; similar results were obtained in three experiments performed on three independently prepared sets of  a All data presented in this table are average values for each size class of particles derived from all particles regardless of DOXYL content. For example, the PL:proapoA-I ratio data derived for the W@108 162-Å particle was averaged from the particles containing no spin label, DOXYL-5, -7, -10, -12, and - 16. b The Stokes diameter as determined by native polyacrylamide gel electrophoresis. The error of 0.3 nm is based on maximal S.D. values obtained for multiple determinations of similarly sized particles in previous studies (23). All complexes were judged to be Ͼ95% pure by coomassie densitometry with the exception of the 77-78-Å particles which had about 10 -15% contamination with the 90 -93-Å particle.
c The ␣-helical contents were determined from the molar ellipticity at 222 nm at 25°C as calculated according to (38). Each value represents the average of at least three observations. rHDL particles. Analogous quenching patterns were found using the largest 160-Å particles containing the same array of quenchers (Fig. 3B).
We performed a positive control experiment to convince ourselves that the quenching protocol was capable of determining the orientation of helices that are known to be parallel to the acyl chains in a bilayer. We used two hydrophobic ␣-helical peptides that each contained a single Trp residue at different points along the length of the helix (gifts of Dr. Erwin London). Ren et al. (18) have demonstrated that these synthetic peptides traverse the membrane of SUV and are, necessarily, oriented parallel to the acyl chains. Peptide K 2 WL 9 AL 9 K 2 A contains a 22-aa transmembrane domain composed largely of repeating leucines flanked with pairs of lysine residues (Fig. 4). The lysines stabilize the ends of the helix in the polar region of the bilayer and allow the intervening hydrophobic region to stretch across the acyl chains. The Trp in this peptide is situated high in the membrane, just below the lysine residues. Peptide K 2 GL 9 WL 9 K 2 A is similar in design, but the Trp has been moved halfway down the length of the hydrophobic region of the helix. We reconstituted these peptides into SUV and performed lipid-quenching studies similar to those performed with the rHDL particles (Fig. 4). As expected, the Trp in K 2 WL 9 AL 9 K 2 A was maximally quenched when the quencher was at C-5. The quenching decreased as the spin label was moved down the acyl chains. K 2 GL 9 WL 9 K 2 A exhibited the opposite quenching pattern; the deeply situated Trp was minimally quenched near the surface of the bilayer and was maximally quenched near C-16, close to the bilayer center.
To quantitate the location of the Trp residues with respect to the bilayer, the distance of each Trp from the center of the bilayer (Z cf ) was calculated using the parallax analysis (27). Table II lists the data for each of the rHDL size classes. In the case of rHDL particles containing Trp-108, the Trp depth was found to range from 5.6 to 6.6 Å with an average of 6.2 Å for all four size classes. The Trp depth was not statistically different from the average for any size class. Similar results were obtained for the W@115 particles, which ranged from 5.0 to 6.0 Å with an average of 5.5 Å. It appears that our extension of the parallax method to the case of an rHDL particle is satisfactory because the calculated Trp depths in Table II agree well with qualitative judgments that one would make as to the depth of the most effective quencher in Fig. 3. Therefore, these data indicate that the Trp residues at positions 108 and 115 were located at similar depths in the membrane and this did not vary significantly with the size or number of proapoA-I molecules in the rHDL particle. In the case of the control peptides (Table III), the average distance of the Trp residue from the bilayer center for K 2 WL 9 AL 9 K 2 A was determined to be about 14 Å, while that for K 2 GL 9 WL 9 K 2 A was about 3 Å. These values compare favorably with the previously published values (18) for these peptides. The max values for the peptides were increasingly blue shifted with the depth of penetration of the Trp residue into the membrane (18). DISCUSSION Lipid-based quenching experiments and the parallax analysis have been useful for defining the positions of Trp residues in transmembrane proteins (33). Here, we have extended this method to the related situation of apoA-I wrapping around the outside of a phospholipid bilayer (see "Methods") to determine the orientation of helix 4 with respect to the bilayer. Our results are discussed below in terms of the picket fence and belt models introduced in Fig. 2.
If one assumes that helix 4 is in the picket fence orientation, Trp-108 should be situated about 3 Å from the center of the membrane (assuming that the geometric center of the helix (99 -121) is residue 110 and a 1.5-Å rise per ␣-helical residue). Trp-115 should then be about 8 Å from the bilayer center on the opposing leaflet. However, Table II shows that both Trp residues were consistently measured at similar depths of about 6 Å in the various particles. In contrast, the Trp residue in peptide K 2 WL 9 AL 9 K 2 A was measured at 14 Å from the center of the bilayer, in excellent agreement with the picket fence theoretical value of 13.5 Å. The Trp residue in peptide K 2 GL 9 WL 9 K 2 A was measured at 3 Å from the center, reasonably close to the theoretical value of 1.5 Å. Thus, the data for the transmembrane peptides are most consistent with the picket fence model   Fig. 3, and the data were plotted the same way. The peptide K 2 WL 9 AL 9 K 2 A data are shown with the filled squares and that for K 2 GL 9 WL 9 K 2 A is shown with the open squares. The schematic illustrations to the right of the figure show the sequence of each peptide, highlighting the location of the Trp residue in each.  prediction (as expected), whereas the data for helix 4 of apoA-I are most consistent with the prediction for the belt model. It may be argued, however, that the data for helix 4 can be consistent with the picket fence model if the helix were moved 3-4 Å across the bilayer in the direction of the N terminus. This would have the effect of pulling Trp-108 away from the bilayer center and Trp-115 toward the center. While our current data cannot preclude this possibility, we believe that this scenario is less likely because such an arrangement would require turn sequences that are no longer centered on Pro residues. The helix breaking propensity of Pro has been proposed to be an important requirement for the rather sharp helical turns required by the picket fence model.
If we assume that helix 4 is in a belt orientation, it is possible to construct the model in Fig. 5. We assumed that the total bilayer width for a liquid crystalline POPC bilayer is 50 Å as measured by x-ray reflection techniques (15 mol % H 2 O at 23°C) (34). Of that, 8 Å is attributed to the polar head group regions of each leaflet (16 Å total), leaving 34 Å for the hydrophobic portion of the membrane. This gives two leaflets of about 17 Å that must be in contact with apoA-I to prevent unfavorable solvation of the acyl chains. We further assumed that one helix interacts with each leaflet as proposed for the belt model (7,14), although we make no suggestion of the identity of the second helix. Our quenching measurements placed the Trp residues of helix 4 of apoA-I about 5.5-6 Å from the center of the bilayer (Table II). Since the Trp residues were located near the centerline of the hydrophobic region of the helix (Fig. 1), this predicts that there were about 12 Å between the centers of the two helices. Our analysis of the x-ray crystal structure for the lipid-free variant of apoA-I (7) indicates that there is an average of 10 -11 Å between centers of helix 4 and helix 6 of a second apoA-I molecule in the structure. Therefore, assuming a typical apoA-I helical radius of about 5.5 Å, our fluorescence measurements are consistent with the edges of the ␣-helices in rHDL particles being clustered together at the center of the bilayer. Furthermore, they appear to be close enough to each other for possible salt bridge interactions such as those proposed for the belt model (14).
It is interesting to note that if the analysis of the helical radii and the placement of the helices by the fluorescence measurements are correct, then two helices together can only account for 22-24 Å of the total 34 Å of acyl chain region. The reasons for this apparent shortfall are not entirely clear although physical measurements on rHDL particles made in the past have suggested that protein to phospholipid interactions at the disc edge can account for a reduced bilayer thickness at the perimeter versus that in the center of the particle (35). It is also possible that the phospholipid head groups can bend slightly around the sides of the helices to interact with negative charges along the polar surface of the helix, shortening the effective distance that the protein needs to cover. Such phospholipid to amphipathic helix interactions have been proposed to occur on the surface of spherical HDL particles or vesicles (36,37).
Another informative observation in this work was that the Trp depths were similar in rHDL that contained 2, 3, and Ͼ4 molecules of apoA-I per particle. Segrest et al. (14) have suggested that when two apoA-I molecules are present in a complex, they are arranged on top of each other in a staggered, anti-parallel fashion with each C-shaped molecule wrapping entirely around a leaflet of the bilayer. In this model, the helix A depicted in the cross-section in Fig. 5 would come from one molecule of apoA-I, while helix B would come from the other. Four molecules of apoA-I can be envisioned in this scenario by simply opening up the original two molecules and slipping in two more to surround a larger lipid patch (7,14). However, three molecules of apoA-I are more difficult to envision because the two leaflets are already occupied. Therefore, if the helices are perpendicular to the acyl chains and to fit three molecules on two leaflets, one is forced into postulating that the molecules of apoA-I can double back on themselves. In this scenario, half of each molecule is on leaflet A, there is an intervening turn, and the other half of the molecule proceeds back toward the N terminus on leaflet B. In this model, helices A and B in Fig. 5 would be from the same molecule of apoA-I. This allows the presence of both even and odd numbers of apoA-I on a particle. Our experimental approach in this study was incapable of distinguishing between these two variants of the belt model, but we feel that this "double back" belt model is the simplest explanation for our observation of three molecules of apoA-I in a rHDL particle with helices consistent with the belt configuration. Finally, it is important to note that these results apply only to helix 4, and they do not exclude the possibility that some areas of apoA-I may exist in a belt conformation, while others exist in a picket fence or other orientation. More Trp residues will have to be introduced in other helical areas of the molecule to investigate this possibility. We believe that the strategy of using carefully designed single Trp mutants with lipid-based quenching studies is a promising approach for understanding the structure of apoA-I and HDL. In addition, we are excited about the potential of this method for structural studies of other apolipoproteins, in both discoidal and spherical HDL particles.