R. M. Epand, Department of Biochemistry, 1200 Main Street West, McMaster University Health Sciences Centre, Hamilton, ON, L8N 3Z5 Canada. Fax: 905 521 1397, Tel.: 905 525 9140, E-mail: firstname.lastname@example.org
We have studied the interaction of β-17, a potent synthetic antimicrobial β-peptide, with phospholipids. We find that unlike other antimicrobial peptides such as magainin II, β-17 facilitates the formation of nonbilayer phases, indicating that the peptide promotes negative curvature. Studies of liposomal leakage also indicate a different mode of membrane interaction relative to magainin II, but both leakage and membrane binding show that β-17, like magainin II, has strong affinity for membranes containing anionic lipids. This is likely to be an important factor contributing to the antimicrobial specificity of the β-peptide.
There is currently extensive interest in developing new antimicrobial agents to counter the growing clinical problem of bacterial resistance to traditional antibiotics . A large variety of natural peptides and synthetic derivatives display antimicrobial activity , and these peptides have been viewed as potential sources of new therapeutic agents. Nearly all of the synthetic derivatives have been constructed from α-amino acids, as are the natural host-defense peptides. Recently, antimicrobial activity has been reported for a number of β-amino acid oligomers (β-peptides) [3–8]. β-Peptides differ from conventional peptides (α-amino acid residues) in two important ways. First, the unnatural backbone of β-peptides confers resistance to degradation by proteolytic enzymes [8–10]. Second, β-peptides constructed from appropriately rigidified residues display higher conformational stability, on a per-residue basis, than do conventional peptides . It is therefore of interest to compare the mechanism(s) of β-peptide antibacterial action with the antibacterial mechanisms of analogous conventional peptides.
Here we characterize the interactions of one antimicrobial β-peptide, β-17 [7,8], with lipid vesicles, which are simplified models of cell membranes. β-17 was designed to mimic natural host-defense peptides like the magainins, which are thought to exert their antimicrobial effects by disrupting bacterial membranes. The structure of β-17 is shown in Fig. 1. Magainins are cationic peptides that contain 23 α-amino acid residues and adopt an amphiphilic α-helical conformation in the presence of membranes . β-17 contains 17 β-amino acid residues and adopts a 12-helical conformation, which is defined by a network of 12-membered ring C=O(i) → NH(i + 3) hydrogen bonds [12–15]. (The α-helix contains 13-membered ring C=O(i) → NH(i + 4) hydrogen bonds.) The β-peptide 12-helix has approximately 2.5 residues per turn and a rise of 5.5 Å per turn; therefore, a 17-residue 12-helix should be similar in length to a 23-residue α-helix formed by a conventional peptide. β-17 contains only two types of β-amino acid residue, hydrophobic trans-2-aminocyclopentanecarboxylic acid (ACPC) and cationic trans-3-aminopyrrolidine-4-carboxylic acid (APC). The repeating APC-ACPC-APC-ACPC-ACPC pentad gives rise to a 12-helix with distinct cationic and hydrophobic sides, which mimics the amphiphilic α-helical conformation adopted by magainin peptides .
We have characterized interactions between β-17 and vesicles by differential scanning calorimetry, to gain insight on lipid phase transitions, and by circular dichroism (CD), to gain insight on β-peptide conformation in the presence and absence of membranes. We have also examined the ability of β-17 to lyse model membranes as a function of lipid composition, in an effort to explain the specificity of action and characterize the energetics of binding to the different liposomal systems used. Like magainins and other host-defense peptides, β-17 is more effective at killing bacterial cells than at inducing lysis of human red blood cells (which are taken to represent eukaryotic cells in general) [7,8]. The basis for this cell specificity among natural peptides remains a subject of debate [1,16–19], although electrostatic factors are thought to be important, as bacterial cell surfaces generally have a greater negative charge density than do eukaryotic cell surfaces . Vesicle model studies have been conducted with many conventional antimicrobial peptides ; therefore, our results with β-17 have an ample basis for comparison. The lipids used in this study as components of lipid mixtures have the structures shown in Fig. 2.
Materials and methods
Phospholipids were purchased from Avanti Polar Lipids (Alabaster, AL).
The labeled lipids were made by chemical modification of phosphatidylethanolamine that was prepared from egg phosphatidylcholine by transphosphatidylation. [Ala8,13,18]-magainin II amide was purchased from the Sigma Chemical Co (St Louis, MO, USA).
Synthesis of NBD-β-17
Fmoc-ACPC-OH and Fmoc-APC(Boc)-OH were synthesized as previously reported [20,21]. β-17 was prepared with a conventional automated solid-phase peptide synthesizer (Synergy 432 A; Applied Biosystems, Foster City, CA) using extended coupling and deprotection steps. After removal of the N-terminal Fmoc protecting group, the resin-bound peptide was treated with three equivalents of 4-fluoro-7-nitrobenzofurazan(4-fluoro-7-nitrobenz-2-oxa-1,3-diazole) (Molecular Probes, Eugene, OR) in dry dimethylformamide containing 5% v/v diisopropylethylamine. The reaction vessel was covered with foil and shaken at room temperature. After 48 h, the resin was washed with dimethylformamide and then CH2Cl2. NBD-labeled β-17 was cleaved from the resin and purified by RP-HPLC using a Vydac C4214TP510 column. Trifluoroacetic acid counterions were replaced with chloride ions through three dissolutions of the β-peptide in 50 mm HCl followed by lyophilization. The replacement of trifluoroacetic acid was confirmed by 19F-NMR. A small amount of decomposition was seen during this step; the final purity of NBD-β-17 was ≈ 95%, and the purity of β-17 itself was >98%.
Differential scanning calorimetry
Lipid films were made from dipalmitoleoylphosphatidylethanolamine (ΔPam2PtdEtn) dissolved in chloroform/methanol (2 : 1, v/v). Increasing mole fractions of peptides dissolved in methanol were added and the solvent removed by evaporation with nitrogen. Final traces of organic solvent were removed in a vacuum chamber attached to a liquid nitrogen trap for 2–3 h. The lipid films were hydrated at room temperature by vortexing with 20 mm Pipes buffer containing 0.14 m NaCl, 1 mm EDTA and 0.002% sodium azide pH 7.4. The final lipid concentration was 5 mg·mL−1. Lipid suspensions were degassed under vacuum before being loaded into a NanoCal high sensitivity calorimeter (CSC, American Forks, UT). A heating scan rate of 0.75 K·min−1 was employed. The observed phase transition was fitted with parameters describing an equilibrium with a single van't Hoff enthalpy and the transition temperature reported as that for the fitted curve. Data were analyzed with the program Origin 5.0. For some curves the endothermic peak was not symmetrical and it appeared that including more than one component would provide a better fit to the data. However, there was no progressive trend showing a systematic increase in one component relative to another as a function of β-peptide/lipid ratio. Fitting the transition to one component provides a measure of the average effect of the β-peptide on the curvature properties of the lipid.
Circular dichroism (CD)
The CD spectra were recorded using an AVIV model 61 DS CD instrument (AVIV Associates, Lakewood, NJ). The sample was contained in a 1-mm pathlength quartz cell that was maintained at 25 °C in a thermostated cell holder. The CD data are expressed as the mean residue ellipticity. All CD runs were made with β-peptide dissolved in 10 mm sodium phosphate buffer containing 0.14 m NaF and 1 mm EDTA at pH 7.4. For samples containing lipid, the lipid was first made into a dry film from a solution of chloroform/methanol, and then hydrated by vortexing with buffer. The lipid suspension was then sonicated to clarity to make small or sonicated unilamellar vesicles (SUVs).
Large unilamellar vesicles
Lipid films were made by dissolving appropriate amounts of lipid in a mixture of chloroform/methanol (2 : 1, v/v) and dried in a test tube under nitrogen. Final traces of solvent were removed in a vacuum chamber attached to a liquid nitrogen trap for 2–3 h. Dried films were kept under argon gas at −30 °C if not used immediately. Films were hydrated with buffer, vortexed extensively at room temperature and then subjected to five cycles of freezing and thawing. The homogeneous lipid suspensions were then further processed by 10 passes through two stacked 0.1 µm polycarbonate filters (Nucleopore Filtration Products, Pleasanton, CA) in a barrel extruder (Lipex Biomembranes, Vancouver, BC), at room temperature. Large unilamellar vesicles (LUVs) were kept on ice and used within a few hours of preparation. Lipid phosphorus was determined by the method of Ames .
Aqueous content leakage from liposomes was determined using the ANTS-DPX assay . Lipid films were hydrated with 12.5 mm 8-aminonaphthalene-1,3,6-trisulfonic acid (ANTS), 45 mmp-xylene-bis-pyridinium bromide (DPX), 68 mm NaCl, 10 mm Hepes at pH 7.4. The osmolarity of this solution was adjusted to be equal to that of the buffer as measured with a cryoosmometer (Advanced Model 3MOplus Micro-Osmometer, Advanced Instruments Inc., Norwood, MA). LUVs of 0.1 µm diameter were prepared by extrusion as described above. After passage through a 2.5 × 20 cm column of Sephadex G-75, the void volume fractions were collected and the phospholipid concentration was determined by phosphate analysis. The fluorescence measurements were performed in 2 mL of buffer composed of 10 mm Hepes, 0.14 m NaCl, 1 mm EDTA, pH 7.4, in a quartz cuvette equilibrated at 37 °C with stirring. Aliquots of LUVs were added to the cuvette to a final lipid concentration of 25 µm. The fluorescence was recorded as a function of time using an excitation wavelength of 360 nm and an emission wavelength of 530 nm with 8 nm bandwidth slits. A 490-nm cutoff filter was placed in the emission path. Leakage was started by addition of the peptide in buffer solution. The value for 100% leakage was obtained adding 20 µL of a 10% Triton X-100 solution to the cuvette. Runs were performed in duplicate.
NBD-labeled compounds have a low quantum yield in aqueous solution, but the quantum yield becomes higher when the probe enters a hydrophobic environment such as in a membrane . To assess the degree of β-17 association with the membrane systems tested in leakage, a series of LUVs were made as previously described and aliquots of a suspension of LUVs were added sequentially to NBD-labeled α-peptide in the cuvette and the fluorescence intensity changes followed until a plateau was reached.
Fluorescence intensity was measured with an Aminco-Bowman Series II spectrofluorimeter, with magnetic stirring, at 25 °C; excitation was set at 467 nm and emission at 539 nm, using 4 nm bandpass slits in excitation and emission and a 490-nm cut off filter in the emission path. Polarizers were set at 90° for the excitation and 0° for the emission to minimize scattering effects. Siliconized glass cuvettes were used, containing 2 mL of 10 mm Hepes buffer with 0.14 m NaCl and 1 mm EDTA at pH 7.4. β-Peptide solutions were made in the same buffer and added at a concentration of 0.4 µm to the cuvette, before addition of LUVs. After every addition of vesicles, the system was allowed to incubate for a few minutes. Controls with addition of LUVs to buffer only were also made and subtracted. The intensity changes were used to calculate binding affinity [25–27].
Isothermal titration calorimetry (ITC)
The experiments to measure the heat of reaction at 30 °C were carried out on a VP-ITC isothermal titration calorimeter manufactured by MicroCal, Northampton, MA. The solutions were degassed under vacuum prior to being loaded in the calorimeter. The β-peptide solution (25–50 µm) was placed in the reaction cell and was titrated with 3–10 µL of 5–10 mm lipid mixtures in the form of LUVs delivered from a motor-driven syringe. The reference cell contained the same buffer as the solution placed in the cell. A buffer composed of 10 mm Hepes, 0.14 m NaCl, 1 mm EDTA at pH 7.4 was used in all titrations. The binding curves obtained were fitted with software provided by Microcal.
Differential scanning calorimetry
We measured how increasing mole fractions of β-17 changed the bilayer to hexagonal phase transition temperature (TH) of a model unsaturated phospholipid, ΔPam2PtdEtn, to obtain an indication of the relationship between insertion of the peptide and the stability of the bilayer phase. ΔPam2PtdEtn is a zwitterionic lipid with a bilayer to hexagonal phase transition at 43 °C. Small mole fractions of additives that interact with this lipid have been shown to either lower this transition temperature, thereby destabilizing the lamellar phase and facilitating the conversion to a more highly curved nonlamellar phase, or vice versa. The change in transition temperature in the presence of small amounts of additives has been therefore used as a measure of propensity to induce curvature in the membrane . Changes in membrane curvature are required for several of the proposed mechanisms by which antimicrobial peptides induce membrane leakage. Thus, the formation of a large pore or a carpet-like disruption on the surface of the bilayer requires increased positive membrane curvature. However, not all lytic peptides promote positive membrane curvature. For example, the peptide mastoparan, isolated from wasp venom, has the opposite effect on membrane curvature. It promotes negative curvature. How this effect of promoting negative curvature is related to the lytic action of mastoparan is not well understood, except for the fact that it would tend to destabilize the bilayer phase toward the formation of inverted phase morphology. This is not meant to imply that a peptide that promotes negative curvature will cause the formation of nonlamellar structures, but rather that it will facilitate the formation of transient structures with increased negative curvature. β-17 also promotes negative curvature and lowers the transition temperature of ΔPam2PtdEtn by −708 ± 104 °C·mol−1 fraction of peptide added. This is opposite to the direction of the shift of TH of +1800 ± 300 °C·mol−1 found with the addition of magainin II  and does not correspond to the type of curvature required for a large toroidal pore. We have also compared our previous differential scanning calorimetry results using magainin II with those of the more potent magainin analog, [Ala8,13,18]-magainin II amide . We find that the peptide-induced changes in the transition of ΔPam2PtdEtn are similar for both magainin II and for [Ala8,13,18]-magainin II amide (data not shown). In a toroidal pore there are two types of curvatures to be considered, i.e. one in the plane of the membrane and the other along the bilayer normal; for large pores, the positive curvature required along the bilayer normal predominates. If β-17 promoted the formation of a similar pore arrangement, but the diameter of the pore were smaller than that of magainin, then the negative curvature in the dimension around the rim of the pore would predominate. However, the pore formed by β-17 does not appear to be very small as it allows the release of β-galactosidase from bacterial cells at a similar rate to magainin . It is thus possible that β-17 lyses membranes by a pore-like mechanism, but the properties of the pore would have to be different from that formed with magainin in order to be consistent with the different curvature tendencies of these peptides.
Solutions of β-17 in water were diluted into phosphate buffer, and the far UV CD spectra were measured in the presence and absence of SUVs (Fig. 3). When visually transparent SUVs of lipid mixtures are added to a solution of the β-peptide, there is a red shift in the CD spectra with an increase in signal intensity. Similar changes in 12-helical β-peptides are observed in going from water to methanol or trifluoroethanol [31,32]. Trifluoroethanol and methanol are known to stabilize helical conformations of both β-peptides  and β-peptides [3,32]. Thus the CD spectra in Fig. 3 suggest that the 12-helical conformation of β-17 is modestly stabilized in the presence of vesicles relative to aqueous solution. DeGrado et al.  have observed strong helix stabilization by vesicles for a different class of β-peptides that are more conformationally flexible than is β-17.
Leakage experiments were designed to obtain information regarding the role of anionic headgroup, fatty acyl chain unsaturation and vesicle curvature in the action of β-17 on model membranes. For this purpose, different anionic lipids dioleoylphosphatidylglycerol (Ole2PtdGro) or dioleoylphosphatidylserine (Ole2PtdSer) were mixed with dioleoylphosphatidylethanolamine (Ole2PtdEtn), dielaidoylphosphatidylethanolamine or phosphatidylcholine extracted from egg yolk (eggPtdCho), as the zwitterionic lipid. Phosphatidylethanolamine has a smaller headgroup and therefore a larger tendency to promote negative curvature compared to phosphatidylcholine. We included Ole2PtdEtn and dielaidoylphosphatidylethanolamine, two species of phosphatidylethanolamine, as components of these mixtures, because the trans double bond at position 9–10 in dielaidoylphosphatidylethanolamine results in less deviation of the acyl chain from the bilayer normal than does the cis double bond of Ole2PtdEtn. Thus dielaidoylphosphatidylethanolamine has less intrinsic negative curvature than Ole2PtdEtn, which is illustrated by the fact that pure dielaidoylphosphatidylethanolamine has a TH of 65 °C compared with 10 °C for pure Ole2PtdEtn, despite their similarity in chemical structure. The presence of the anionic lipids in the membrane allows for the formation of stable liposomes at room temperature in the presence of Ole2PtdEtn; pure PtdEtn does not easily form liposomes, presumably due to limited hydration of the headgroup and the higher energy cost of bending the membrane. In addition, the presence of the anionic lipid PtdGro in these mixtures is important as this is a major component of prokaryotic membranes, while the use of phosphatidylserine mimics the anionic lipid that becomes exposed to the extracellular side of the bilayer in cancer cells or in apoptotic mammalian cells. The zwitterionic lipid was mixed with the negatively charged lipid Ole2PtdSer or Ole2PtdGro at a 1 : 2 molar ratio.
The rate and final extents of leakage is shown in Fig. 4A,D. The leakage of aqueous contents from liposomes of different lipid compositions is summarized by showing the dependence of the percentage leakage at 300 s on the β-peptide to lipid ratio (Fig. 5). In Fig. 6, leakage from Ole2PtdEtn LUVs containing the anionic lipids Ole2PtdGro, Ole2PtdSer, 1-palmitoyl-2-oleoylphosphatidylserine (PamOlePtdSer) or phosphatidylserine extracted from bovine brain is shown. Substituting either bovine brain phosphatidylserine; dielaidoylphosphatidylethanolamine, or PamOlePtdSer for Ole2PtdSer had little effect, highlighting again the predominance of headgroup charge over chain length or degree of unsaturation of the fatty acid in controlling α-peptide induced leakage. EggPtdCho alone had a lower rate of leakage than did the lipid mixtures, implying the requirement for an anionic lipid. This result is expected for a peptide with antimicrobial activity, as microbial membranes have a net negative surface charge.
All vesicle systems tested that contained Ole2PtdGro were found to bind strongly to NBD-β-17 peptide, as would be expected of a cationic molecule (Fig. 7). Binding to LUVs of Ole2PtdEtn/Ole2PtdGro and dielaidoylphosphatidylethanolamine/Ole2PtdGro was virtually indistinguishable, while the lowest binding affinity was found for Ole2PtdGro/Ole2PtdSer.
Isothermal titration calorimetry (ITC)
Isothermal titration calorimetry was performed with β-17 in the cell and incremental addition of vesicles. The results of titration of LUVs of dielaidoylphosphatidylethanolamine/Ole2PtdGro (1 : 2) into a solution of β-17 is given in Fig. 8A. Similar titrations were performed for mixtures of Ole2PtdEtn/Ole2PtdGro (2 : 1) as well as Ole2PtdEtn/Ole2PtdSer (2 : 1) (not shown). The titration for pure eggPtdCho LUVs did not evolve significant heat with either β-17 or NBD-β-17 (Fig. 8B). However, addition of Ole2PtdGro to eggPtdCho produced a titration curve similar to that of the other LUV containing anionic lipids. The two modes of binding observed could be explained by aggregation at the high concentrations of peptide required for the ITC experiment and in the presence of high local concentrations of LUVs after each addition.
For this reason, thermodynamic data was not extracted to evaluate binding. The titrations however, allowed discrimination between the behaviour of the peptide in lipids containing Ole2PtdGro and with eggPtdCho alone.
In this study we have characterized the interactions of β-17 with model membranes, in order to rationalize the observations that the antimicrobial activities and hemolytic activities of β-17 and those of [Ala8,13,18]-magainin II amide , are similar [7,8]. This finding indicates that these peptides must both cross the bacterial cell wall at similar rates. It has been observed that although a cyclic derivative of magainin has similar activity against liposomes when bound to the membrane, its activity against biological membranes can be altered by its ability to access the membrane surface . Both β-17 and [Ala8,13,18]-magainin II amide have in common their affinity for anionic lipids, but we found significant differences between β-17 and magainin II or [Ala8,13,18]-magainin II amide in the way these antimicrobial peptides interact with vesicle systems. While both magainin II  and [Ala8,13,18]-magainin II amide raise the TH of ΔPam2PtdEtn, β-17 lowers it. The fact that magainin raises the TH is in accord with the suggested pore mechanism  in which the peptide has to promote a large positive curvature of the lipid monolayer lining the pore. It is less clear how the promotion of positive curvature relates to the action of the β-17 peptide. The fact that homologous antimicrobial peptides can have different effects on liposomal membranes is not unique to the present paper. It has recently been shown that a group of model Leu-containing diastereomeric peptides micellized vesicles while the Ile-containing diastereomeric homologs fused model membranes .
The lipid dependence of leakage is also very different for magainin II and for β-17. In part, this reflects their different curvature tendencies. For example, replacing Ole2PtdEtn with dielaidoylphosphatidylethanolamine (Fig. 4A,B), which has less negative intrinsic curvature, reduces the lysis caused by β-17 as can be seen by the lower rate of contents release caused by 5 µmβ-17 with dielaidoylphosphatidylethanolamine:Ole2PtdGro (1 : 2) compared to 5 µmβ-17 with Ole2PtdEtn:Ole2PtdGro (1 : 2). Similarly 10 µmβ-17 is required for complete leakage to be achieved with dielaidoylphosphatidylethanolamine/Ole2PtdGro (1 : 2) as opposed to 6.5 µmβ-17 with Ole2PtdEtn/Ole2PtdGro (1 : 2). In contrast, magainin II promotes greater lysis of vesicles when Ole2PtdEtn is substituted with dielaidoylphosphatidylethanolamine . The effects of changes in lipid composition (to modulate membrane curvature) on liposomal leakage rates have not been studied for [Ala8,13,18]-magainin II amide. Magainin II also exhibits much more lysis with liposomes containing PtdGro than with liposomes containing PtdSer . We found that β-17 (Fig. 6) did not discriminate, in terms of inducing leakage, between vesicles containing Ole2PtdGro or Ole2PtdSer, even though the binding affinity of β-17 to PtdSer-containing vesicles is lower than for PtdGro-containing vesicles. Magainin II has been shown in several studies to have antitumor activity [37–45]. With the greater interaction of β-17 with PtdSer-containing liposomes, as compared with magainin, one would anticipate that β-17 would be a more potent anticancer agent.
Both magainin II and β-17 lyse anionic liposomes with greater potency than they lyse zwitterionic lipsosomes (Fig. 4C,D), and they both bind more strongly to anionic membranes than to the zwitterionic membrane, as seen by ITC. Therefore, their microbial specificity is likely a consequence of improved electrostatic interactions with microbial vs. eukaryotic membranes. For several antimicrobial peptides it is believed that the basis of the selective toxicity towards bacteria is that the peptide is cationic and can therefore partition to a greater extent into microbial membranes, which have exposed anionic lipids, than into eukaryotic cell membranes. Alternatively, the antimicrobial activity could result from interaction of the peptide with a specific component of microbial membranes. Recently it has been shown that the antimicrobial peptide nicin has greater specificity for microbial membranes because of a specific interaction with lipid II, a component of these membranes [46–49].
Like other cationic antimicrobial peptides, β-17 displays specificity for anionic lipids. However, β-17 differs from several other such antimicrobial peptides, such as magainin II and its analog [Ala8,13,18]-magainin II amide, in that β-17 promotes negative rather than positive membrane curvature. Therefore it is likely that there are differences in the manner in which β-17 affects membrane properties, compared with magainin. One manifestation of these differences is the lack of specificity of interaction of the peptide with PtdSer vs. PtdGro. Despite these mechanistic differences, however, β-17 behaves as a potent antimicrobial peptide having a minimal inhibitory concentration against several bacteria at least as good as or better than that of [Ala8,13,18]-magainin II amide . [Ala8,13,18]-magainin II amide is an analog of the natural peptide magainin II that has a greater antimicrobial activity . This analog is more hydrophobic than magainin II, having three Ala substituting for a Ser and two Gly. The antimicrobial activity of this analog for several bacteria is of a similar potency to its hemolytic activity  in contrast to magainin II where there is approximately 100-fold greater antimicrobial activity compared with hemolytic activity . Being more hydrophobic, [Ala8,13,18]-magainin II amide will likely penetrate more deeply into the membrane than the parent peptide, magainin II. Deeper membrane penetration also facilitates the promotion of negative curvature, as the regions of the bilayer below the pivotal plane are expanded more than regions close to the interface. Thus, although [Ala8,13,18]-magainin II amide still induces positive curvature effects on membranes, as measured by shifts in TH, its effects on membranes would tend to be somewhat different from that of magainin II. When the [Ala8,13,18]-magainin II amide analog is compared with β-17 by HPLC analysis, both peptides have similar overall hydrophobicities, with similar retention times. However, the effects of these peptides on the TH of ΔPam2PtdEtn suggest that the β-17 has a greater depth of penetration. Nevertheless the two peptides exhibit similar microbial specificity.
This work was supported by the Canadian Institutes of Health Research, Grant 7654 and by the US National Institutes of Health (GM56414). N.U. was supported in part by a Research Fellowship for Young Scientists from the Japan Society for the Promotion of Science, and E.A.P. was supported in part by a Biotechnology Training Grant from NIGMS.