Membrane-anchoring and charge effects in the interaction of myelin basic protein with lipid bilayers studied by site-directed spin labeling.

Myelin basic protein (MBP) maintains the compaction of the myelin sheath in the central nervous system by anchoring the cytoplasmic face of the two apposing bilayers and may also play a role in signal transduction. Site-directed spin labeling was done at eight matching sites in each of two recombinant murine MBPs, qC1 (charge +19) and qC8 charge (+13), which, respectively, emulate the native form of the protein (C1) and a post-translationally modified form (C8) that is increased in multiple sclerosis. When interacting with large unilamellar vesicles, most spin-labeled sites in qC8 were more mobile than those in qC1. Depth measurement via continuous wave power saturation indicated that the N-terminal and C-terminal sites in qC1 were located below the plane of the phospholipid headgroups. In qC8, the C-terminal domain dissociated from the membrane, suggesting a means by which the exposure of natural C8 to cytosolic enzymes and ligands might increase in vivo in multiple sclerosis. The importance of two Phe-Phe pairs in MBP to its interactions with lipids was investigated by separately mutating each pair to Ala-Ala. The mobility at F42A/F43A and especially F86A/F87A increased significantly. Depth measurements and helical wheel analysis indicated that the Phe-86/Phe-87 region could form a surface-seeking amphipathic alpha-helix.

The 18.5-kDa isoform of myelin basic protein (MBP) 1 is a stabilizing factor in the myelin sheath. A major function of MBP is to bind to the apposing cytoplasmic faces of the myelin membrane and maintain compaction for efficient nerve transmission (1,2), but it may also be involved in signal transduction (3). Because of a diversity of post-translational modifica-tions, MBP exists as a number of charge isomers denoted C1-C8 with a net positive charge decreasing from ϩ19 to ϩ13 at pH 7.0 (4 -6). The C8 component is characterized by the enzymatic deimination of arginine to citrulline. Each conversion results in the loss of one positive charge, and C8 is thus the least basic form of the protein and has a diminished ability to cause adhesion of lipid bilayers (7)(8)(9). Component C8 occurs in greater amounts in patients with the demyelinating disease, multiple sclerosis (9,10).
We have previously produced and characterized a recombinant murine 18.5-kDa MBP (11). Here, we will denote this protein quasi-C1 (qC1), because it is unmodified post-translationally (with the exception of an LEH 6 tag) and emulates the least-modified, most basic charge isomer C1. We have also generated by site-directed mutagenesis a quasi-deiminated form of recombinant murine 18.5-kDa MBP that we call qC8, since it was designed to mimic the less cationic natural form C8. The recombinant qC8 consists of Arg/Lys 3 Gln substitutions at the same deimination sites in human MBP that predominate in chronic multiple sclerosis and has properties similar to those of natural C8 (7). The net charge of qC1 is ϩ19 at neutral pH, whereas that of qC8 is ϩ13 as for their natural counterparts.
In this study, we investigated the electrostatic and hydrophobic components of MBP-lipid interactions by site-directed spin labeling (SDSL) of MBP and electron paramagnetic resonance (EPR) spectroscopy (12,13). The technique of SDSL involves replacement of residues at selected sites by cysteines, which are then labeled with a methanethiosulfonate spin label that can be probed by EPR spectroscopy. This approach enabled us to monitor the electrostatic lipid interaction profiles of qC1 and qC8 at numerous specific sites. The importance of hydrophobic interactions was evaluated by spin-labeling sites adjacent to each of the two Phe-Phe pairs and determining the effects of Phe-Phe 3 Ala-Ala substitution on spin label mobility and accessibility to lipid-soluble O 2 and water-soluble nickel ethylenediaminediacetic acid (NiEDDA) as applied by Victor et al. (14) to the myristoylated alanine-rich C kinase substrate effector region. The technique of SDSL is particularly well suited to MBP, because there are no native cysteinyl residues to be removed prior to mutagenesis and because the EPR spectrum is not affected by light diffraction and immobilization associated with MBP-induced lipid vesicle aggregation. Lipid vesicle aggregation is a powerful mimic of the in vivo function of MBP in the myelin sheath (15,16), and SDSL/EPR offers the advantage of studying residue-specific interactions within this natural environment.

EXPERIMENTAL PROCEDURES
Materials-The sulfhydryl reactive spin label [1-oxyl-2,2,5,5-tetramethyl-D-pyrroline-3-methyl]methanethiosulfonate (MTS-SL) was purchased from Toronto Research Chemicals (Toronto, Ontario, Canada). This spin label is referred to as R1 in previously published literature (17,18). The paramagnetic reagent NiEDDA was synthesized as previously described (19). All of the other chemicals were of reagent grade and were acquired from either Fisher Scientific or Sigma. The Ni 2ϩ -NTA-agarose beads were obtained from Qiagen (Mississauga, Ontario, Canada).
Labeling and Purification of Cysteine-containing Mutants of qC1 and qC8 -The purification and spin labeling of recombinant murine 18.5-kDa MBPs were done using a modification of a published purification protocol (11). Two types of lysis buffer were used, one containing 20 mM Tris-HCl, pH 8.0, 6 M urea, 5 mM imidazole, and 500 mM NaCl, and the other being identical with the exception of the addition of 1 mM dithiothreitol to prevent the formation of intermolecular disulfide bonds. The cell pellet was lysed using the dithiothreitol-containing buffer, and subsequent loading and washing of the Ni 2ϩ -NTA column were done using this buffer. Finally, the column was washed with normal lysis buffer using 25 column volumes to remove the excess dithiothreitol. An additional 25 volumes of spin labeling buffer with urea (20 mM HEPES-NaOH, pH 7.4, 6 M urea, 10 mM NaCl) were then run through the column. The agarose beads were removed from the column and resus-pended in 8 ml of spin labeling buffer, and a 10-fold molar excess of MTS-SL was dissolved in 100 l of Me 2 SO was added to the bead slurry. The slurry consisting of the protein bound to the Ni 2ϩ -NTA beads and MTS-SL was incubated on a nutator overnight in a 15-ml Falcon tube at room temperature. The column was repoured the next day and washed with normal lysis buffer to remove excess MTS-SL, and the bound protein was eluted. The purity of each labeled protein was verified using SDS-PAGE, and the protein-containing fractions were dialyzed against 20 mM HEPES-NaOH, pH 7.4, 10 mM NaCl, and 1 mM EDTA. The labeling efficiency was assessed for one sample using electrospray ionization mass spectroscopy (11,20) and was found to be complete (data not shown). The protein concentration was estimated by measuring the absorbance at 280 nm using the published extinction coefficients (7). The protein samples were adjusted to 1 mg/ml in the same buffer for EPR spectroscopy.
Preparation of Large Unilamellar Vesicles (LUVs)-Aliquots of the chloroform solutions of the lipids were combined in the following molar ratios to form LUVs with a lipid composition similar to that estimated for the cytoplasmic face of the myelin membrane (Cyt-LUVs): 44 mol % cholesterol; 27 mol % phosphatidylethanolamine; 13 mol % PS; 11 mol % PC; 3 mol % sphingomyelin; and 2 mol % phosphatidylinositol (15,21). The solvent was evaporated under a stream of nitrogen, and the lipid mixture was dried further in a vacuum dessicator overnight. It was hydrated in 20 mM HEPES-NaOH, pH 7.4, 10 mM NaCl, and 1 mM EDTA. The lipid suspension was then passed through a 100-nm polycarbonate membrane 17 times using a syringe extruder (Avanti Polar Lipids). The final lipid concentration after extrusion was verified using a phosphorous assay (22).
EPR Spectroscopy-For signal-averaged EPR spectroscopy, spin-labeled qC1 and qC8 solutions were added to LUVs at a concentration of 100 g of protein to 2 mg of LUVs, yielding a molar protein:lipid ratio of 1:600. This value is a little less than the MBP:lipid ratio in myelin. Borosilicate glass 50-l capillary tubes (Fisher Scientific) were used to hold the samples. Aggregation of the vesicles by MBP occurred immediately, and after 10 min, the preparations were spun at 1000 ϫ g to loosely pellet the MBP-LUVs. Most of the supernatant was removed, and the pellets were taken up in capillary tubes. The opposite end of the capillary tube was sealed using a Bunsen burner, and the tubes were centrifuged at 4000 ϫ g for 15 min to give a compact pellet.
The protein-lipid pellet was positioned in the center of the EPR cavity of a Bruker ECS 106 spectrometer (Bruker BioSpin, Milton, Ontario, Canada). The EPR spectra were recorded at room temperature using a microwave power of 10.0 milliwatts and a modulation amplitude of 1.0 G. To estimate the amount of free unbound MBP, the supernatant of the final centrifugation step (see above) was also probed by EPR spectroscopy and no detectable signal was recorded. We conclude that all of the MBP reacted with the lipid vesicles and that there was no unbound protein.
An empirical motion parameter 0 was determined from the first derivatives of the absorption spectra using Equation 1, where K ϭ 6.5 ϫ 10 Ϫ10 s, ⌬H is the width of the center-line and h 0 and h Ϫ1 are the heights of the center and high field lines, respectively ( Fig.  2a) (23). For the membrane-bound samples, changes in the mobility of the probe were quantified using the width of the center-line (⌬H) (a greater ⌬H value represented a decreased mobility of the spin label) (17,24).

Depth Measurements via Continuous Wave (CW) Power Saturation-
The depths of penetration of spin-labeled sites of all of the proteins into Cyt-LUVs were determined from CW power saturation. Experiments and data analysis were performed essentially as described previously (25,26) using a Varian E102 Century Series spectrometer (Varian Associates, Palo Alto, CA) equipped with a loop-gap resonator (Medical Advances, Milwaukee, WI). Because of the increased sensitivity of this resonator, the amount of sample used for CW saturation was halved (1 mg of LUVs, 50 g of protein). After spinning down the lipid-protein aggregate, 5 l of it was loaded into a gas-permeable TPX capillary (27). The field modulation was 1.25 G, and the incident microwave power was varied from 0.1 to 81 milliwatts. The P1 ⁄2 value was determined by fitting the amplitude A of the m I ϭ 0 peak to Equation 2, where I is a scaling factor, ⑀ is a measure of the homogeneity of FIG. 1. a, amino acid sequences of qC1 and qC8 with residues that were mutated to Cys and spin-labeled indicated by shaded squares. The basic residues in qC1 that were converted to Gln to yield qC8 are indicated with an arrow (1), and the two Phe-Phe sites that were mutated to Ala-Ala are indicated by broken open squares. The LEH 6 tag was omitted for clarity. b, scheme for the spin labeling reaction of cysteinyl residues in qC1 and qC8. The Cys-containing mutants were reacted with the MTS-SL while bound to the Ni 2ϩ -NTA resin.
saturation of the resonance line, P is the microwave power, and P1 ⁄2 is the power necessary for reducing amplitude (A) to half of its unsaturated value. Power saturation was done on one full set of samples when equilibrated first with nitrogen and then with air and on another set of samples when equilibrated with nitrogen in the presence of 20 mM NiEDDA. The ⌬P1 ⁄2 was obtained by the difference in the P1 ⁄2 values in the presence and absence of NiEDDA and O 2 . The normalized accessibility parameter (⌸) was calculated from ⌬P1 ⁄2 as described previously (28) using diphenylpicrylhydrazine to standardize resonator performance. The ⌬P1 ⁄2 values were used to calculate ⌽ according to Equation 3, This parameter describes the relative depth of the spin label in the bilayer due to the gradient of NiEDDA and O 2 along the bilayer normal.
The dependence of ⌽ on distances from the membrane surface was determined using head group-labeled TEMPO-PC and various PCs with doxyl nitroxides along the acyl chain at positions 5, 7, 10, and 12. The nitroxide-labeled PCs were mixed with the lipids used for Cyt-LUVs at a molar ratio of 1:500 in chloroform, and LUVs were prepared as described above. The calibration was done using Cyt-LUVs without protein as well as two concentrations of unlabeled qC1. The first concentration was a molar ratio of lipid to protein of 600:1 (identical to that used for the depth measurements), and the second concentration was a saturating amount of unlabeled qC1 (150:1 lipid to protein). The resultant ⌽ values were used in conjunction with the known depth (distance of the nitroxide nitrogen from the phosphate) of spin-labeled PC doxyl nitroxides (29) to generate a standard curve for determining the absolute depth of the MTS-SL labels in the protein when bound to the bilayer. The distance of the nitrogen to the phosphate in TEMPO-PC was estimated to be 5 Å based on molecular models by Farahbakhsh et al. (28).
Previous results have demonstrated that the linear dependence of the depth parameter breaks down with increasing distance into the aqueous phase and behaves according to Equation 4 (30), where x is the distance from the nitrogen of the nitroxide label to the lipid phosphate, A and D define the bulk values of ⌽ in water and hydrocarbon, and B and C describe the slope of the curve and the inflection point, respectively. This equation was then used to solve for the distance (x) using the experimentally derived values of ⌽.

RESULTS
Labeled MBP Sites Are Not Sequestered in Aqueous Solution-In aqueous solution, all the spin-labeled residues of both qC1 and qC8 gave a spectrum similar to that in Fig. 2a. These spectra were consistent with unstructured nitroxide-labeled proteins or peptides with no sites that were predominantly sequestered in a hydrophobic pocket (31). The spectra of similar shape were obtained in 30% sucrose used to restrict tumbling motion of the protein molecule (17). The type of isotropic motion represented indicated that there was very little immobilization at any of the sites probed with the nitroxide radicals. The motional parameters ( 0 ) of each residue were also calculated and averaged to yield 0.69 ns (Ϯ0.07 ns S.D.). Since the error associated with each 0 is roughly 0.13 ns as estimated from several replicate experiments, we conclude that there is no significant difference in mobility for any of the spin-labeled sites in solution.
The qC8 Spin Labels Are Less Immobilized than the qC1 Spin Labels in the Bilayer-The interactions with Cyt-LUVs of qC1 and qC8 spin-labeled at sites S17C, S44C, S67C, H85C, S99C, S129C, and S159C were monitored from the EPR spectra. Upon interaction of these proteins with Cyt-LUVs, the motion became more anisotropic and the sharp hyperfine lines were broadened, indicating a decrease in mobility (Figs. 2b and  3, a and b). In some cases, especially H85R1, S129R1, and S159R1 in qC1, a second more immobilized component was observed as indicated by an arrow on the low field side in Fig.  3a. There was no exchange broadening indicative of MBP oligomerization in any of the spectra.
The spectra in Fig. 3, a and b, showed several differences between qC1 and qC8. The mobility of most labeled residues in qC8 was greater as can be seen especially from the sharper high field peak at m I ϭ Ϫ1. There was also a decrease in the more immobilized component seen for qC1 at H85R1 and S129R1. To quantify the degree of mobility, the widths of the center-lines (⌬H) were measured and compared as 1/⌬H (Fig.  4). Generally, a higher value of 1/⌬H means that the relative mobility is greater. The spin labels at sites S17C, S67C, S129C, and S159C in qC8 were more mobile than at the same sites in qC1. Overall, the difference between the spin-label environments in qC1 and qC8 was more pronounced the closer the label was to an Arg/Lys 3 Gln substitution. However, H85R1 was more mobile in qC8 than in qC1 (seen especially from the loss of immobilized component from the spectrum in Fig. 3b, rather than from the change in ⌬H in Fig. 4) even though there was no Arg/Lys 3 Gln substitution in the vicinity. This result indicates that this site might be more sensitive to its environment or that quasi-deimination and reduced lipid binding effected a local structural perturbation.
Role of the Phe-Phe Sites in Anchoring qC1 and qC8 in the Bilayer-The spin labels at sites S44C and H85C next to two Phe-Phe pairs at positions Phe-42/Phe-43 and Phe-86/Phe-87 were the most immobilized in qC1 and qC8 (Figs. 3b and 4), suggesting that they anchor qC1 and qC8 in the bilayer. The   FIG. 2. a, first derivative EPR spectrum of 1 mg/ml spin-labeled qC1-S17R1 in aqueous solution (20 mm HEPES-NaOH, pH 7.4, 10 mm NaCl, 1 mm EDTA). The width of the center-line is ⌬H, and h 0 , h 1 , and h Ϫ1 are the heights of the center-, low field, and high field first derivative lines, respectively. b, first derivative EPR spectrum of the complex of qC1-S44R1 and Cyt-LUVs.
two Phe-Phe pairs were expected to provide a hydrophobic component to the free energy of binding of MBP to lipid bilayers.
To estimate the relative importance of these Phe-Phe sites, both Phe-42/Phe-43 and Phe-86/Phe-87 were substituted by Ala-42/Ala-43 and Ala-86/Ala-87 in qC1-S44C and qC1-H85C, respectively. The Phe-Phe 3 Ala-Ala substitutions caused significant differences in the environments of adjacent probes and hence mobilities and spectral line shapes, especially for qC1-H85R1 (Fig. 3c, arrows). In fact, qC1-F86A/F87A/H85R1 was the most mobile of all of the sites probed as can be seen from the ⌬H values in Fig. 4. In contrast, qC1-F42A/F43A/S44R1 was motionally restricted even with the Phe-Phe 3 Ala-Ala substitutions.
Depth Measurements via CW Power Saturation-The EPR spectral line shapes (and hence values of ⌬H and 0 ) could depend not only on the depth of penetration of the spin label into the lipid bilayer but also on secondary structural changes that can be induced in MBP by its interaction with lipids. For this reason, direct measurements of depth of penetration of each spin label into the membrane were performed using power saturation approaches (25,26). Table I summarizes the ⌸ and ⌽ values calculated for all of the proteins when bound to Cyt-LUVs.
Due to the unique combination of lipids present in the Cyt-LUVs, proper calibration of the ⌽ value using the spin-labeled lipids was essential. The NiEDDA accessibility of the lipid nitroxide was significantly higher in the cholesterol-containing Cyt-LUVs than in the PC or PC/PS lipid mixtures usually used for these measurements (18,30,32). Subczynski et al. (32,33) have shown that cholesterol increases the water permeability and reduces oxygen permeability of the bilayer up to C7 or C9 of the acyl chain of saturated and unsaturated lipids, respectively. Consequently, NiEDDA would actually penetrate Cyt-LUVs more easily up to a depth of 10 Å, which means that all of the nitroxide spin labels in MBP would be more available for collisional quenching than they would be in a bilayer lacking cholesterol. The experimentally determined depth of the spin labels (29) was plotted versus the ⌽ values (Fig. 5a), and a curve was then fit to the data using the hyperbolic tangent function described in Equation 4. This function was first reported by Frazier et al. (30) and provides a description of the limiting behavior of ⌽ that can be applied to sites on the aqueous side of the bilayer interface.
To determine the lipid spin label accessibility in the presence of MBP bound to the surface of the membrane, the calibration curve was also determined with the addition of 50 g of unlabeled qC1 (as used for spin-labeled MBP) and with 180 g of unlabeled qC1 to saturate the acidic phospholipid head groups. The presence of MBP increased the depth parameter (Fig. 5a), and the curve was fit to data for the samples containing 180 g of qC1. Despite the differences in lipid composition from that used by Frazier et al. (30), Equation 4 resulted in an extremely good fit of the data (r 2 ϭ 0.9998) (Fig. 5a).
The distance data for the various spin-labeled MBPs were obtained by solving for the distance using experimentally determined ⌽ values (Fig. 5b). The calibration curve changes very little below a ⌽ value of Ϫ2, making it impossible to solve for the relative depths of some of the more exposed spin label sites. To calculate a distance for ⌽ values below Ϫ2, we used linear regression to fit data for the head group PC and 5-and 7-doxyl-PCs (these values served for comparative purposes only).
The depth values for qC1 showed that with the exception of S17R1, spin-labeled residues in the N-terminal half were more deeply embedded in the bilayer than those in the C-terminal half, whereas those in the midsection were located near the level of the nitrogen of the PC head groups (Fig. 5b). The spin label S44R1 next to Phe-42/Phe-43 was the most deeply pene-  3. a, first derivative EPR spectra of the qC1 (red lines) and qC8 (blue lines) spin label mutants bound to Cyt-LUVs. The spectra were normalized to the amplitude of the center peak. An arrow on the low field side indicates a second more immobilized spectral component in the spectra of qC1-S129R1 and qC1-S159R1. This component is much less pronounced in the corresponding sites in qC8. b, first derivative EPR spectra of qC1 (red lines) and qC8 (blue lines) labeled at S44C or H85C near the Phe-Phe sites at positions Phe-42/Phe-43 and Phe-86/ Phe-87 (qC1-44-SL, qC8 -44-SL, qC1-85-SL, and qC8 -85-SL) bound to Cyt-LUVs. These protein sites are more immobilized than those in a as can be seen from the broadening of the peaks. c, first derivative EPR spectra of qC1 spin-labeled at S44C or H85C in mutants in which Phe-Phe is changed to Ala-Ala (qC1-F42A-F43A, and qC1-F86A-F87A) (blue lines) bound to Cyt-LUVs compared with the spectra of the normal Phe-Phe-containing proteins (red lines).
trating one, but H85R1 next to Phe-86/Phe-87 penetrated only a little below the nitrogen of the PC head group, even though it was the most immobilized residue.
Comparison of the depth values of qC1 and qC8 revealed dramatic differences at sites in the C terminus, which relocated to the aqueous phase in qC8. In contrast to qC8, S129R1 and S159R1 of qC1 were found at the interface and at 3.5 Å below the phosphates, respectively. The relocation of residues in the C-terminal domain of qC8 to the aqueous phase accounts for the greater mobility of these residues.
At the N terminus, S17R1 was slightly more exposed in qC1 as opposed to qC8, despite the fact that the mobility was significantly higher in qC8. Accessibility of S44R1 and S99R1 was similar in qC8 and qC1, consistent with the lack of effect of Arg/Lys 3 Gln substitution on mobility of these residues. In addition, S99R1 was found in the aqueous region consistent with its high mobility, whereas the location of S44R1 in the bilayer was consistent with its low mobility.
The Arg/Lys 3 Gln substitution in qC8 caused H85R1 to move further out of the bilayer, consistent with an increase in mobility. However, the location of this residue in the head group region in qC1 suggests that the low mobility of this residue may be due to secondary structure in this region rather than the result of deep bilayer penetration. There was a significant increase in the aqueous accessibility of H85R1 after the F86A/F87A substitution, but no change for S44R1 with the F42A/F43A mutation. This result was somewhat surprising considering the depth of S44R1; however, the increase in mobility observed at this site was not nearly as extreme as it was for H85R1 upon substitution of Ala-Ala for Phe-Phe near these residues. DISCUSSION MBP has a significant and specific role in the compaction of the myelin sheath (1, 2) and may also be involved in signal transduction. Extracellular signals imparted by anti-glyco- The lipid spin label standards were incorporated into Cyt-LUVs with the addition of 0.2 mol % spin-labeled lipid with bound unlabeled qC1. Measurements were performed in the absence of protein with 50 g of unlabeled qC1 and with 180 g of unlabeled qC1 (data shown in table). There was no significant difference between the two proteincontaining samples.
b The average standard error associated with ⌸ was estimated from replicate experiments.
c The distances were determined using the fit to Equation 4 using the spin-labeled standards (Fig. 5a). The uncertainty in the distance calculation increases substantially near the aqueous phase.
d Since values of ⌽ Ͻ Ϫ2 could not be solved using the fit to Equation 4, a linear regression of the head group PC, 5-doxyl-PC, and 7-doxyl-PC spin labels were used for comparative purposes only.
FIG. 5. Depth measurement of spin labels into Cyt-LUVs via power saturation. a, calibration curve for doxyl spin-labeled lipids using Cyt-LUVs (44% cholesterol (CHL), 27% phosphatidylethanolamine (PE), 13% PS, 11% PC, 3% sphingomyelin (SPM), and 2% phosphatidylinositol (PI), molar ratios). Open circles represent the ⌽ values in the absence of unlabeled qC1, whereas the closed circles represent the ⌽ values in the presence of 180 g of unlabeled qC1 with 1 mg of Cyt-LUVs. The curve shows the fit to Equation 4 using the latter protein-containing samples. The curve was constrained by the aqueous bulk limit (⌽ ϭ Ϫ2.6) and the upper ⌽ limit of the membrane interior (⌽ ϭ 4.5) as described previously (30). The parameters from the best fit using the various spin-labeled PCs with the Cyt-LUV lipid mixture with bound unlabeled qC1 were: A ϭ 3.2, B ϭ 0.15, C ϭ 12.0, and D ϭ 1.3. The ⌽ values for spin-labeled MBP ϽϪ2 could not be determined from this fit due to the limiting nature of the curve as it approaches the bulk solvent; therefore, a linear regression of the data for TEMPO-PC, 5-doxyl-PC, and 7-doxyl-PC was used to calculate these distances. The closed squares represent the data for spin-labeled MBPs that were fit to the hyperbolic tangent function, whereas open squares represent the data for spin-labeled MBPs that were fit to the linear regression. Error bars were calculated by propagating the uncertainties in the ⌬P1 ⁄2 measurements. b, depths of penetration of all of qC1 and qC8 spin labels into the lipid bilayer of the Cyt-LUVs. The gray shading indicates values below the lipid phosphates (in the bilayer), whereas the crosshatching specifies the tentative region of distance determination that employed the linear regression for ⌽ ϽϪ2. The horizontal line at Ϫ5 Å indicates the location of the nitrogen atom of the TEMPO-PC nitroxide (28). sphingolipid antibodies are transmitted across the membrane to the cytoskeleton by a mechanism involving MBP (3). Therefore, it is important to delineate the specific sites of interaction of MBP with lipid bilayers and to determine how this interaction is influenced by lipid composition and post-translational modifications of the protein. MBP is known to be a protein that is "intrinsically unstructured" (or "natively unfolded") (34) and is primarily disordered in aqueous solution, but gains ordered secondary structure upon binding to lipids (11,(35)(36)(37)(38). Several segments of MBP, including both the N and C termini, are predicted to have a high degree of intrinsic disorder (36,39). In myelin, MBP is thought to lie on the surface of the lipid membrane at the level of the head group region with some hydrophobic side chains dipping part of the way into the bilayer (40 -42). In this study, we have used site-directed spin labeling to probe the topology of MBP when bound to membranes of composition akin to that of the myelin sheath.
Electrostatic Interactions of MBP with Lipid Bilayers-We used a fully charged version of recombinant murine MBP (qC1, net charge ϩ19 at pH 7.0) as well as a less cationic form (qC8, net charge ϩ13 at pH 7.0) to study how the overall charge affected lipid-protein interactions at various sites. Most spinlabeled sites were found to be more mobile in qC8 than in qC1, although the accessibility measurements indicated that the C terminus in qC8 was the most accessible to the aqueous phase, indicating a potentially decreased interaction of these qC8 sites with the lipid bilayer.
Other studies have shown that MBP/C1 can bind to actin filaments while bound to Cyt-LUVs and that the protein dissociates from the membrane when calmodulin is added (43). This dissociation would be predicted to be faster with qC8. Thus, not only does the modified protein aggregate lipid to a lesser extent, but the greater mobility of spin-labeled sites and the increased accessibility at the C terminus indicate that it is less embedded in the bilayer, which would expose it to potential ligands. In summary, long range Coulombic forces appear to play an important role in attracting MBP to the surface of a membrane and in determining how deeply it is embedded in the bilayer.
The high mobility and accessibility of S99R1 in qC1 and qC8 indicated that this region was exposed to the aqueous environment. This segment of 18.5-kDa MBP is of interest because of its proximity to the PRTPPPS motif (the triproline repeat region characteristic of all of the known mammalian MBPs) (Fig.  1a). The sequence of PRTP represents a PXXP motif that can bind to an SH3 domain (Src homology 3) containing protein such as non-receptor tyrosine kinases (44). Moreover, the threonyl residue within this motif is a mitogen-activated protein kinase target (45). Our results suggest that this segment of MBP is naturally exposed and available for modification and recognition.
The Two Phe-Phe Pairs in MBP Are Not Equivalent-The importance of aromatic residues in the two dual Phe-Phe sites was studied by substituting these residues for Ala-Ala. The spin-labeled residues immediately adjacent to the Phe-Phe pairs, S44R1 and H85R1 in both qC1 and qC8, demonstrated significant motional restriction compared with other sites. After the Phe-Phe 3 Ala-Ala substitutions, the mobility of each of these sites increased, especially H85R1.
When the regions surrounding Phe-42/Phe-43 and Phe-86/ Phe-87 are portrayed as helical wheels (Fig. 6), there are distinct disparities. Segment mMBP (38 -49) has neither a hydrophobic nor a hydrophilic face indicative of an amphipathic ␣-helix. In contrast, segment mMBP-(82-93) is strongly amphipathic with very hydrophobic amino acids (Val, Phe, and Ile) on one side and polar amino acids (Lys, Thr, His, and Asn) on the other side of the ␣-helix (a similar prediction has been noted previously (46,47)). The segment mMBP-(82-93) has a hydrophobic-hydrophilic residue ratio of 13:5, which in peptides has been shown to result in immersion of their hydrophobic regions into lipid bilayers (48). In fact, this type of ␣-helix had the largest hydrophobic face in that study and had the strongest binding to bilayers. The mean helical hydrophobic moment ( H ) of the putative ␣-helix mMBP-(82-93) was calculated using the sum of the hydrophobicities of the side chains (49) and found to be 0.332, indicating that the ␣-helix is amphiphilic perpendicular to its axis. The hydrophobic moment of this section of MBP places it directly in the domain of known surface-seeking ␣-helices (48). Moreover, this ␣-helix may even be stable in solution as indicated by digestion of MBP with cathepsin D, which cleaves at Phe-Phe linkages. The Phe-42/ Phe-43 pair is cleaved far more quickly than Phe-86/Phe-87 (7,50). The CW power saturation accessibility of H85R1 indicates that this site is above the plane of the lipid phosphate groups. Since H85R1 is located on the polar side of the putative amphipathic ␣-helix (Fig. 6), the conclusion that it is positioned at the interface is sound.
The CW power saturation data positions H85R1 in the Ala-Ala-containing mutant in the bulk aqueous phase several angstroms from the membrane location of H85R1 in the Phe-Phecontaining protein. It might have been expected that this site would have been even more exposed based on the considerable change in mobility that was observed. Another factor to consider is the fact that the spectrum of the spin-labeled residue is sensitive to ␣-helix motion (51,52). The membrane-anchoring effect of the Phe-Phe pair might prevent rocking motions of the ␣-helix. Rocking motion would become much more pronounced when the Phe-Phe was converted to Ala-Ala.
Another interesting feature of segment mMBP-(82-93) is the terminal prolyl residues at positions Pro-82 and Pro-93. Prolines are commonly found in kink or bend points within proteins and disrupt ␣-helices, especially in membrane proteins (53). These two residues could represent natural extents of the putative amphipathic ␣-helix, and the points at which this section of MBP would dip into the membrane. This idea is supported by the observation that both prolyl residues line up in an orientation where they could connect with the rest of the protein (Fig. 6), whereas the amphipathic ␣-helix penetrates the membrane.
The indicated that the depth of this residue was unchanged by this substitution, even though it was the most deeply penetrating site. From this result, it does not appear that the presence of the Phe-42/Phe-43 pair alone is vital for membrane anchoring of this region of the protein. Thus, Phe-42/Phe-43 and Phe-86/ Phe-87 may play different roles in terms of the association of MBP with the membrane. The reason for the Phe-Phe-independent deep penetration of S44R1 and for the deep penetration of the N-terminal half of MBP is not known. However, it is consistent with greater labeling of the N-terminal half of MBP by the hydrophobic photolabel relative to the C-terminal half (40). In myristoylated alanine-rich C kinase substrate, a highly basic 25-residue lipid-effector region interacts with the cell membrane by a combination of electrostatic interactions and partial insertion of hydrophobic phenylalanines into the lipid bilayer (14,54). Our results show that this highly effective design for strong membrane association is only partially shared by MBP with significant differences in behavior (and thus role) of one of its Phe-Phe pairs.
Another amphipathic ␣-helix was predicted earlier for the C terminus of MBP (55), suggesting that this region might bind acidic lipids with high affinity and/or contain a calmodulinbinding site. The more immobilized component in the spectrum of S159R1 of qC1 (not present in the spectra of S17R1, S67R1, and S99R1) and its deeper penetration at a depth of 3.5 Å below the lipid phosphates are consistent with the localization of S159R1 on the hydrophobic side of this putative amphipathic ␣-helix. CONCLUSIONS We have used SDSL and EPR to examine the local electrostatic and hydrophobic contributions to MBP interactions with lipid bilayers. The diminished electrostatic charge in quasideiminated MBP increased the mobility of most of the spin labels spanning the protein, especially at the C terminus, which dissociated from the membrane. The implication is that the natural C8 isomer will be more exposed than the C1 isomer to proteolytic degradation, further post-translational modifications, or ligand-or protein binding. These effects may be operative in multiple sclerosis. Spin-labeled residues near the Phe-42/Phe-43 and Phe-86/Phe-87 pairs were motionally restricted in the unmodified protein, and their mobilities increased when Phe-Phe was mutated to Ala-Ala. The greater effect was seen adjacent to Phe-86/Phe-87 embedded in a segment found to be a superb candidate for an amphipathic ␣-helix that would bind tightly to a membrane surface. The spin-labeled residue was on the polar side of this putative amphipathic ␣-helix, consistent with its high aqueous accessibility. Future studies involving Cys scanning of the entire region surrounding Phe-86/Phe-87 will provide further insight into its association with lipid bilayers.