Reconstruction of the Conserved β-Bulge in Mammalian Defensins Using d-Amino Acids*

Defensins are cationic antimicrobial mini-proteins that play important roles in the innate immune defense against microbial infection. Six invariant Cys residues in each defensin form three structurally indispensable intramolecular disulfide bridges. The only other residue invariant in all known mammalian defensins is a Gly. Structural studies indicate that the invariant Gly residue is located in an atypical, classic-type β-bulge with the backbone torsion angles (ϕ, Ψ) disallowed for l-amino acids but permissible for d-enantiomers. We replaced the invariant Gly17 residue in human neutrophil α-defensin 2 (HNP2) by l-Ala or one of the d-amino acids Ala, Glu, Phe, Arg, Thr, Val, or Tyr. Although l-Ala17-HNP2 could not be folded, resulting in massive aggregation, all of the d-amino acid-substituted analogs folded with high efficiency. The high resolution x-ray crystal structures of dimeric d-Ala17-HNP2 were determined in three different crystal forms, showing a well preserved β-bulge identical to those found in other defensins. The seven d-analogs of HNP2 exhibited highly variable bactericidal activity against Gram-positive and Gram-negative test strains, consistent with the premise that interplay between charge and hydrophobicity dictates how amphiphilic defensins kill. Further, the bactericidal activity of these d-amino acid analogs of HNP2 correlated well with their ability to induce leakage from large unilamellar vesicles, supporting membrane permeabilization as the lethal event in microbial killing by HNP2. Our findings identify a conformational prerequisite in the β-bulge of defensins essential for correct folding and native structure, thereby explaining the molecular basis of the Gly-Xaa-Cys motif conserved in all mammalian defensins.

A ␤-bulge is defined as a region between two consecutive ␤-type hydrogen bonds, that includes two or more residues on one strand opposite a single residue on the other strand (1,2). Found primarily in anti-parallel ␤-sheets, ␤-bulges occur commonly, on average twice per protein (2). The extra residue(s) on the bulged strand not only disrupts the normal alternation of side chain direction but also impacts the directionality of ␤-strands and accentuates the typical right-handed twist of ␤-sheets (1,2). For these reasons, ␤-bulges are often well conserved in proteins. However, the underlying molecular basis for the formation of ␤-bulges in proteins remains poorly understood. Carefully selected model protein systems provide valuable tools in elucidating the functional implications of the occurrence of these irregular secondary structural elements, thus likely shedding light on protein de novo design. In this respect, mammalian defensins are a suitable example.
Mammalian defensins are a family of cationic and Cys-rich antimicrobial mini-proteins expressed predominantly in leukocytes and epithelial cells (3,4). They form part of the first line of defense in the innate immune arsenal against pathogens, killing a broad range of microbes primarily through membrane disruption. In addition, defensins function as effective immune modulators in adaptive immunity by activating certain types of immune cells via receptor-mediated signaling pathways (5)(6)(7)(8). Based on the connectivity of their six cysteine residues, mammalian defensins are classified into ␣, ␤, and families (9). Despite the divergent sequences and different disulfide pairings, the tertiary structures of defensins from ␣ and ␤ families are quite similar (10 -15). Characteristic of the overall fold is a three-stranded anti-parallel ␤-sheet constrained by three disulfide bonds. A sequence alignment of known ␣-defensins from human, mouse, rhesus macaque, rabbit, guinea pig, and rat ( Fig. 1) reveals seven conserved residues, the six Cys residues and the glycyl residue located within the second ␤-strand of the molecule. Separated by a variable residue (Xaa), the invariant Gly and the fourth Cys constitute the sequence motif Gly-Xaa-Cys, which is also conserved in all known mammalian ␤-defensins, as well as in other broader classes of antimicrobial peptides (16). Collectively, the structural studies of several ␣and ␤-defensins show that the Gly residue is part of the conserved ␤-bulge (17).
We have recently solved high resolution crystal structures of human neutrophil ␣-defensin 4 and human epithelial ␣-defensins 5 and 6, also known as HNP4, HD5, and HD6, respectively. 4 Structural analysis of HNP4, HD5, and HD6 indicates that the invariant glycyl residue (Gly 17 in HNP1) adopts the backbone dihedral angles (, ) centered around ϭ 170°and ϭ Ϫ150°in the Ramachandran diagram ( Fig. 2) (18). These torsion angles are rarely observed for L-amino acid residues but are energetically preferable for D-enantiomers (19). Thus, we hypothesized that the glycyl residue is conserved in mammalian ␣-defensins because of the stringent structural requirement for ␤-bulges that cannot be met by any L-amino acids with restricted backbone conformational space. Furthermore, substitutions of D-amino acids for the invariant glycyl residue in mammalian ␣-defensins can be structurally tolerated. Because antimicrobial peptides kill microorganisms by mechanisms independent of stereospecific molecular recognition (20,21), we predicted that D-amino acids in place of the conserved Gly in defensins will be functionally allowed. The current report integrated biochemical, structural, and functional approaches to investigate whether or not the conserved ␤-bulge in mammalian defensins can be reconstructed through D-amino acid substitutions without structural and functional consequences.

MATERIALS AND METHODS
Synthesis of HNP2 Analogs-Peptide synthesis was carried out on an Applied Biosystems 433A synthesizer using an in-house Boc chemistry (22,23) tailored from the previously published 2-(1H-benzotriazolyl)-1,1,3,3-tetramethyluroniumhexafluorophosphate activation/N,N-diisopropylethylamine in situ neutralization protocol developed by Kent and coworkers (24,25). Crude peptides, after chain assembly and hydrogen fluoride cleavage/deprotection, were purified on a Waters Delta Prep 600 system equipped with a Vydac reversed phase (RP) 5 preparative C18 column (15-20 m, 50 ϫ 250 mm). Analytical RP-HPLC was performed on an Agilent 1100 series system equipped with a Waters Symmetry 300 TM analytical C18 column (5 m, 4.6 ϫ 150 mm). Solvents used for HPLC purification were water containing 0.1% trifluoroacetic acid (solvent A) and acetonitrile containing 0.1% trifluoroacetic acid (solvent B). Electrospray ionization mass spectrometry (ESI-MS) analyses were conducted on a Micromass ZQ-4000 single quadruple mass spectrometer. Samples were suspended in methanol:water:acetic acid (49:49:2) and infused by a syringe pump at 10 l/min. Peptides were quantified spectroscopically at 280 nm using molar extinction coefficients calculated according to a published algorithm (26).
Oxidative Folding-The protocol of oxidative folding for HNP2 analogs was essentially as described previously (23). Reduced purified peptides were first dissolved at 1 mg/ml in 8 M urea containing 12 mM reduced and 1.2 mM oxidized glutathione, followed by a 4-fold dilution into a dimethylformamide solution (33% v/v) buffered to pH 8.3. The air-tight folding reaction proceeded at room temperature with gentle stirring. 40 l was withdrawn periodically for analysis on HPLC, and peak integration was performed to quantify time-dependent folding yields. Folded HNP2 analogs were purified to homogeneity by RP-HPLC and their molecular masses ascertained by ESI-MS.
Determination of the disulfide topology in D-Ala 17 -HNP2 was performed according to the published procedures (27,28 [3][4][5] . It is worth noting that, in contrast to what was found with wild-type HNPs (28), the Tyr 16 -D-Ala 17 peptide bond was not cleaved by chymotrypsin. This finding is consistent with the fact that the scissile peptide bond flanked by D-amino acids is proteolytically resistant due to steric incompatibility in the enzyme-substrate complex.
Crystallization, Data Collection, and Processing-The crystallization experiments were carried out at room temperature using the hangingdrop vapor diffusion method. Droplets were prepared by mixing equal FIGURE 1. Amino acid sequence alignment of mammalian ␣-defensins from human, mouse, rhesus macaque, rabbit, guinea pig, and rat (us.expasy.org/sprot). The invariant residues, including six Cys and Gly 17 (HNP1 numbering), are shaded in black. The boxed residues, Arg 5 and Glu 13 , form a salt bridge in the structure and are highly conserved.  (18). Most residues are located in conformationally allowed regions, such as A (␣-helix (red)) and B (␤-strand (red)). Gly 17 from a total of 12 defensin chains (filled triangles circled in red) clusters around ϭ ϩ157 Ϯ 20°and ϭ Ϫ155 Ϯ 9°, with 10 centered at ϭ 170°and ϭ Ϫ150°. The blue circle depicts the most probable region preferred by D-amino acids (19). This figure was created using the program Procheck (63).
volumes of protein (15 mg/ml in 10 mM Tris buffer, pH 8.5) and reservoir solution. Three different crystal forms of D-Ala 17 -HNP2 were obtained, which belong to the monoclinic, tetragonal, and trigonal systems, respectively. Monoclinic crystals were grown from the solution containing 0.5 M lithium sulfate and 15% (w/v) polyethylene glycol 8000. Tetragonal crystals were obtained from the solution consisting of 0.2 M trisodium citrate dihydrate, 0.1 M Tris/HCl, pH 8.5, and 30% (w/v) polyethylene glycol 8000. Trigonal crystals grew from the solution containing 0.5 M lithium sulfate and 2% (w/v) polyethylene glycol 8000.
The x-ray diffraction data were collected using the synchrotron radiation source (line 22BM at the Southeast Regional Collaborative Access Team station in the Advanced Photon Source, Argonne, Chicago, IL), with the wavelength tuned to either 1.0 or 0.92 Å. The intensities were recorded using a MarCCD detector (x-ray Research, Germany). For data reduction and scaling, we used the program HKL2000 (29), and the identity of the space group was unambiguously deduced from the systematic extinctions observed for the scaled intensities. The x-ray data collection and processing statistics are shown in TABLE ONE.
Antibacterial Activity-Antimicrobial assays against Escherichia coli American Type Culture Collection (ATCC) 25922 and Staphylococcus aureus ATCC 29213 were conducted using a previously detailed 96-well turbidimetric method dubbed virtual colony counting (30), except tenpoint (rather than nine-point) 2-fold dilution series of HNP2 and analogs in 10 mM sodium phosphate, pH 7.4, were made, resulting in concentrations of 0.5-256 g/ml. In addition to "negative" controls containing no defensin, a negative control termed "input" consisted of a portion of the same cell suspension kept on ice during the initial 2 h of defensin exposure and was added to the microplate just before the addi-tion of 100 l of twice-concentrated Mueller-Hinton broth. Data analysis proceeded as described in Ref. 30, except a Visual Basic algorithm was written to automate the determination of the time necessary for each growth curve to reach a threshold change in optical density at 650 nm (⌬OD 650 ) of 0.02, mimicking the previous visual inspection method with equal accuracy and greater precision. Bactericidal activity in the absence of added tryptic soy broth was calculated using slope and y-intercept values for a threshold ⌬OD 650 of 0.02 from TABLE ONE of Ref. 30.
Large Unilamellar Vesicles (LUVs) Leakage Assays-LUVs encapsulating the low molecular weight fluorophore/quencher pair (ANTS/ DPX) were prepared using the standard extrusion method. Specifically, phospholipids DPPC and POPG were dissolved in chloroform at an equal molar ratio and dried as a film by solvent evaporation. After removal of residual solvent, the lipid film was hydrated in the fluorescent solution containing 5 mM HEPES, 12.5 mM ANTS, 45 mM DPX, and 20 mM NaCl, pH 7.0, freeze-thawed for 10 cycles, and extruded 10 times through 0.4-m polycarbonate membranes. LUVs were separated from unencapsulated materials by gel filtration chromatography using a Sepharose CL-4B column eluted with 5 mM HEPES and 100 mM NaCl, pH 7.4 Leakage of ANTS from LUVs, monitored on an LS-55 PerkinElmer luminescence spectrometer, was characterized by an increase in fluorescence, which was quenched by DPX when encapsulated together inside liposomes (31). 270 l of ANTS/DPX-encapsulated LUVs were added to each well of a 96-well plate to a final lipid concentration of 600 M. 30 l of H 2 O was added to the first well of each row as a blank and 30 l of 2.5% (v/v) Triton X-100 to the last (twelfth) well as the control

Engineering Mammalian Defensins Using D-Amino Acids
for 100% leakage. Upon the addition of 30 l of a 2-fold dilution series of defensin and incubation for 1 h, the fluorescence signal was recorded at 515 nm with an excitation wavelength of 353 nm, 10-nm bandwidths, and a 390-nm cutoff filter in the emission path. Percent leakage is expressed as, where F t is the fluorescence determined at different time points after the addition of defensin, F 0 is the background fluorescence of the "blank" cells, and F 100 is the fluorescence of the control cells containing 0.25% Triton X-100.

RESULTS
D-Xaa 17 -HNP2 but Not L-Ala 17 -HNP2 Folds Productively-Substituting any of the seven D-amino acids in place of Gly 17 and applying the previously published oxidative folding protocol for HNPs resulted in productive folding of the defensin with a yield of Ͼ65%. As illustrated in Fig. 3A, folded D-Ala 17 -HNP2 emerged predominately as a narrow peak on RP-HPLC after 8 h, shifting ϳ4 min earlier in retention time compared with the peptide before oxidative folding. By contrast, L-Ala 17 -HNP2 under the same folding conditions massively precipitated (Ͼ60%). An analysis of the supernatant on RP-HPLC gave rise to a broad trace of much lower UV light absorbance with no distinct peak (Ͻ0.5%) at the expected retention time for the folded species over a period of 24 h (Fig. 3B). These findings suggest that, although D-amino acids at the Gly 17 position are fully compatible structurally, substituting L-amino acids at that position causes defensins to misfold.
After folding and purification, all seven D-amino acid-substituted HNP2 analogs were analyzed by RP-HPLC and ESI-MS, yielding chromatographic traces and mass spectra highly similar to those of D-Ala 17 -HNP2 (Fig. 3C). Oxidative folding of D-Xaa 17 -HNP2 resulted in a loss of six mass units, consistent with the formation of three disulfide bridges in the protein. Peptide mass mapping coupled with complete tryptic/chymotryptic digestion and Edman degradation unequivocally established the correct disulfide connectivity in D-Ala 17 -HNP2, of which crystal structures were subsequently solved at high resolution.
Structural Characterization of D-Ala 17 -HNP2-The crystals of all three forms of D-Ala 17 -HNP2 showed excellent diffraction properties. Consequently, the electron density was well defined for all the protein residues, including the D-Ala 17 site (Fig. 4A). Although the asymmetric units of monoclinic and trigonal crystals contain four protein molecules arranged in two dimers, in the tetragonal crystals, only two monomers are structurally independent. The tertiary structures of D-Ala 17 -HNP2 are quite conserved for all 10 independent monomers of this protein, although some systematic differences could be observed. The root mean square deviations calculated for 29 C-␣ atoms varied between 0.2 and 0.8 Å, depending on the choice of aligned monomers. Despite the different crystal environment of each of 10 independent monomers, the conformations of most side chains appeared quite similar, as shown in Fig. 4B.
A significantly different outcome arises from comparison of five quaternary structures of D-Ala 17 -HNP2. These differences are particularly visible after aligning dimers based on the equivalent 29 C-␣ atoms from one monomer only (Fig. 4C). It is clear that the variability shown in the quaternary structures is a result of the different relative locations of monomers participating in dimerization. A better understanding can be obtained from the analysis of dimer-stabilizing interactions. The primary contribution to dimer formation and stabilization comes from four (N-H⅐⅐⅐O) main-chain hydrogen bonds between Thr 18 and Ile 20 from the equivalent ␤-strands of both monomers. Further stabilization of the dimer is provided by hydrophobic contacts between the phenyl rings of Phe 28 from one monomer and the sulfur atoms forming the disulfide bridge between Cys 4 and Cys 19 of the second monomer. Finally, in some dimers, we observed short hydrogen bonds between the main-chain atoms of the two equivalent Cys 2 residues mediated by water molecules or Cl Ϫ ions.

-HNP2 Is Indistinguishable from HNP2 in Its Antibacterial
Activity-Reinforcing our interpretation of structural data, antibacterial functional results indicated that D-Ala-substituted HNP2 killed E. coli in a manner indistinguishable from wild-type HNP2 (Fig. 5).
D-Ala 17 -HNP2 was very slightly less effective than HNP2. This differ-

Engineering Mammalian Defensins Using D-Amino Acids
ence in killing was proportional to defensin concentration, up to a full log at 32 and 64 g/ml, but at no concentration tested was D-Ala 17 -HNP2 clearly outside standard error compared with HNP2. Both peptides achieved complete killing at 128 and 256 g/ml in both replicates of both experimental conditions. Survival at each concentration was essentially identical to data previously published (TABLE TWO in  D-Val 17 -HNP2 was essentially identical to HNP2 and D-Ala 17 -HNP2 against both strains. The other neutral side chains produced similar results to HNP2 and each other, except the bulky hydrophobic D-Phe 17 noticeably improved potency against S. aureus at all concentrations. By contrast, a charged D-amino acid at position 17 consistently altered antimicrobial activity at all concentrations against both strains. Against E. coli, survival was at least two orders of magnitude lower for D-Arg 17 -HNP2 than for D-Glu 17 -HNP2 at 8 g/ml and above. D-Arg 17 -HNP2 was the only peptide tested that produced zero E. coli survival at 256 g/ml during the 12 h of the assay. This difference was even more pronounced against S. aureus. D-Glu 17 -HNP2 killed only one log at the  Ϫ F c , red). Several residues are labeled for clarity. Individual red balls represent the oxygen atoms of ordered solvent molecules. The figure was prepared using Bobscript (64) and POV-ray (www. povray.org). B, the stereo representation of ten structurally independent (representing the asymmetric unit contents of three crystal forms) monomers of D-Ala 17 -HNP2 with equivalent C-␣ atoms superimposed. Selected residues have been labeled for ease of interpretation. High structural conservation, clearly evident for the backbones of protein chains, extends also over most of the side chains of D-Ala 17 -HNP2. The figure was prepared with programs Ribbons (65) and POV-ray. C, flexibility of the D-Ala 17 -HNP2 dimers. The green and yellow C-␣ traces represent dimers present in the monoclinic crystals. The trace shown in red describes the tetragonal structure, whereas violet and blue reflect two dimeric assemblies from the trigonal crystals. Five dimers of D-Ala 17 -HNP2 were superimposed based on the C-␣ atoms of one monomer (shown on the left) only; therefore, discrepancies in the relative positions of the intimate monomers represent an extent of the flexibility of the dimer. A good estimate of the latter is gained by comparison to the 5 Å bar. It is clear that the most significant "deformation" is observed for the dimer of the monoclinic crystal form, i.e. assembly interacting in crystal lattice with the polyethylene glycol molecule. N, N terminus; C, C terminus.
highest concentration, whereas D-Arg 17 -HNP2 killed completely at the highest four (32-256 g/ml). Our results are consistent with previous perturbations of antimicrobial peptides that showed an increase in efficacy with increasing cationic charge, and to a lesser extent, with increasing hydrophobicity. We found no evidence that any substitutions at position 17 change the nature of the dose versus survival relationship. As we previously observed, survival curves against S. aureus were all concave-down on a log-log scale and essentially linear on a log-normal plot, consistent with simple exponential killing and suggesting the absence of any resistance mechanism. By contrast, all E. coli curves deviated from simple exponential killing, reproducing the shape of the survival curve previously reported for HNP2 against this strain (30).
The Bactericidal Activity Correlated with the Ability of D-Xaa 17 -HNP2 to Induce Leakage from LUVs-Cationic antimicrobial peptides selectively target prokaryotic cells, because the bacterial cell wall and membrane is negatively charged (32,33). It is believed that defensins kill microorganisms by permeabilizing the cytoplasmic membrane and inducing leakage of cellular contents (34 -38). Liposomes encapsulating fluorophores are commonly used as a model system for studies of the interaction between phospholipid membranes and cationic antimicrobial peptides, such as defensins (39 -42). We found that HNP2 and its D-amino acid analogs were fast-acting in inducing leakage from LUVs in a dose-dependent manner (Fig. 6). D-Arg 17 -HNP2 caused a 50% leakage of ANTS from DPPC/POPG LUVs at a concentration of ϳ0.5 g/ml, whereas the required concentration of D-Glu 17 -HNP2 was as high as 75 g/ml. For HNP2 and other D-amino acid analogs, the concentrations needed to induce a 50% leakage ranged from 0.8 to 22 g/ml. Significantly, the relative potency of the membrane activity of these

DISCUSSION
Amino acid residues in proteins are conserved for structural and/or functional reasons. Changing a structurally conserved residue may result in loss of the native three-dimensional fold of a protein, whereas substitutions of functionally conserved residues interfere with various biological processes, often attributable to the difference in side chain chemistry. Because the glycyl residue lacks a side chain, its role is likely structural. The lack of a side chain affords Gly greater access to backbone conformational space than L-amino acid residues (18). Consequently, Gly frequently occurs in the loops and ␤-turns of a protein, where it is often structurally required for the backbone to adopt dihedral angles (, ) not accessible to L-amino acids. Its large conformational freedom, however, renders Gly entropically deleterious within regular secondary structural elements, such as ␣-helices and ␤-sheets (43). Despite this entropic disadvantage, Gly can be conserved in ␤-sheets of irregularity in proteins. The Gly 17 residue in the ␤-bulge of mammalian defensins is a classic example of such conservation.
Gly and the ␤-Bulge in Defensins-On the basis of local structure, the hydrogen bonding pattern, and the number of residues involved, ␤-bulges are grouped into five different subclasses: classic, G1, wide, bent, and special (1,2). The most common ␤-bulge is the classic type in anti-parallel ␤-sheets, where two residues (numbered 1 and 2) on the bulged strand participate in main-chain H-bonding with the residue (lettered X) on the opposite strand. In a classic ␤-bulge, residue X donates one H-bond to the CϭO group of residue 2, or NH(X)3 CO(2), and accepts two from the NH groups of residues 1 and 2, or CO(X)4NH(1) and CO(X)4NH (2). The CO(X)-NH (2) bond is always longer than the other two H-bonds, and in some cases, too long to be considered effective (2). For classic ␤-bulges, the (, ) angles of residue 1 must fall into an ␣ R conformation (Ϫ60°, Ϫ30°), whereas residues 2, preferably small amino acids, such as Gly, Ala, and Ser, and residue Xaa, typically occupy the ␤-sheet region in the Ramachandran diagram (2). We analyzed the H-bonding pattern of the ␤-bulge in the recently determined structures of HNP4, HD5, HD6, 4 and D-Ala 17 -HNP2 (4 chains in each defensin) and derived the following characteristics for the constituting hydrogen bonds: CO(X)-NH(1) ϭ 2.8 Ϯ 0.05 Å, CO(X)-HN(2) ϭ 3.8 Ϯ 0.4 Å, NH(X)-CO(2) ϭ 2.9 Ϯ 0.1 Å. Apparently, the H-bonding pattern of the ␤-bulge in defensins is the same as that of a classic ␤-bulge. Further, although the torsion angles of residue 1 (Tyr 16 in HNP2) fall into the ␣ R region, residue Xaa (Phe 28 in HNP2) has a preference in the ␤-sheet region. However, we classify the ␤-bulge in mammalian defensins as an atypical classic type, because the conserved Gly (residue 2) adopts the (, ) angles centered at ϭ ϩ170°a nd ϭ Ϫ155°rather than a ␤-conformation.
Mitchell and Smith (19) categorize the D-amino acid residues present in Protein Data Bank entries and find that the torsion angles preferred by the vast majority of D-residues are loosely centered around ϭ ϩ120°and ϭ Ϫ135°, with a small portion around the ␣ L region (ϩ60°, ϩ45°). It is therefore reasonable to assume that substitutions of D-amino acids for Gly 17 in defensins will introduce little perturbation on the protein structure because of local conformational compatibility. Our findings clearly support this assumption (Fig. 7), demonstrating the importance of (, ) angles permissible only to D-amino acids for defensin folding. Interestingly, the (, ) angles of D-Ala 17 in the crystal structure of HNP2 shifted to the ␣ L region, which is consistent with the secondary dihedral preference of D-amino acids in the Ramachandran diagram.
Structural and Functional Effects of ␤-Bulging-One important structural effect of a ␤-bulge is to accentuate the twist of the ␤-strand (1,2). Structural perturbations of ␤-bulges often translate into profound functional consequences in proteins. In studying a ␤-bulge in E. coli dihydrofolate reductase, Dion-Schultz and Howell (44) have shown that deletion mutations in the ␤-bulge region invariably destabilize the protein and decrease catalytic efficiency. Similar results are reported by Axe et al. (45) for barnase, a bacterial ribonuclease, where deletion or substitution of the Gly residues in a conserved ␤-bulge severely impairs the stability and catalysis of the enzyme. Chen et al. (46) have observed that deletion of Gly in the N-terminal ␤-bulge of ubiquitin slows down the refolding and unfolding of the protein, suggesting that the ␤-bulge functions as a nucleation center in the folding process. Interestingly, the effect of inserting a ␤-bulge into a protein can be functionally neutral. Keefe et al. (47) have studied a double Gly insertion mutation in staphylococcal nuclease and find that, when two Gly residues are inserted on the protein surface, a three-residue ␤-bulge forms to accommodate the extra residues without pronounced functional and structural consequences. Recently, Di Nardo et al. (48) have reported that mutations at the conserved Gly 48 in a ␤-sheet of Src homology 3 (SH3) domains, although highly destabilizing, dramatically accelerate protein folding. Although it adopts the unusual (, ) angles in the same region occupied by Gly 17 of defensins, the Gly 48 residue of the SH3 domains is located in a regular ␤-sheet rather than a ␤-bulge.

Engineering Mammalian Defensins Using D-Amino Acids
So why did substituting an L-Ala residue for Gly 17 of HNP2 result in a misfolded aggregate? All mammalian ␣-defensins contain six conserved Cys residues that, by forming disulfide bonding, play a crucial role in maintaining the three-stranded ␤-sheet structure. It is well documented that peptide fragments from ␤-sheeted proteins are prone to aggregation, making de novo designs of ␤-proteins extremely challenging (49). It is plausible that accentuated twisting introduced by the ␤-bulge in defensins necessitates the proximal cysteinyl residue (Cys 19 ) to correctly position itself for native disulfide bonding with Cys 4 . Productive folding intermediates will likely be disrupted structurally by the substitution of L-Ala for Gly 17 due to incompatible backbone torsion angles. Alternatively, loss of the ␤-bulge as a result of the Gly 3 L-Ala substitution in HNP2 may also impact the side-chain alternation necessary for the formation of correct folding intermediates leading to protein aggregation. Our findings with D-amino acid substitutions at position 17 appear to underscore the importance of the unusual (, ) angles in the ␤-bulge of defensins for the stabilization of folding intermediates.
D-Amino Acids in Proteins-D-amino acids occur naturally, even in the absence of codons for their inclusion among the 19 L-amino acids and the achiral glycine. A stereospecific post-translational epimerization event can produce diasteromeric peptides found in a variety of lower animal species, such as amphibians (50,51). Much more commonly, D-amino acids are specifically synthesized and incorporated into antibiotic peptides by prokaryotes. Nevertheless, the L-amino acids remain by far the predominant enantiomers found in nature (52).
D-amino acids have been chemically engineered into peptides and proteins for various reasons. They can be used as structural probes because of their known properties as ␣-helix or ␤-sheet breakers (53,54). Because D-amino acids stabilize ␤-turns (55, 56), they are highly valuable in protein de novo designs (57,58). A more drastic use of D-amino acids is the construction of mirror image proteins composed entirely of D-amino acid residues to screen novel ligands (59,60). Notably, D-amino acids have been extensively used to enhance the therapeutic index and in vivo stability of small antimicrobial peptides (61). It is well established that antimicrobial peptides do not engage in highly regiospecific interactions with target molecules, in sharp contrast with most macromolecular interaction systems in biology. In fact, antimicrobial peptides consisting of all D-amino acids are equally active as their natural L-form counterparts (20,62), suggesting that peptide-mediated cell lysis involves no protein receptors on the surface of microbial membranes. As a result of the structural leniency seen with antimicrobial peptides, incorporation of D-amino acids into defensins was expected to be functionally tolerated.
Electrostatic interactions between cationic peptides and negatively charged membranes are among the first steps in microbial killing. Interplay between charge-initiated peptide-membrane association and hydrophobicity-mediated burial of amphiphilic defensin molecules into the lipid phase of microbial membranes leads to eventual membrane disruption and cell lysis. Results obtained from the substitutions of Damino acids for Gly 17 in HNP2 are consistent with this premise. Notably, the effect of charge and hydrophobicity at position 17 on antimicrobial activity is strain-dependent, providing a unique opportunity for modulating the antimicrobial activity and specificity of defensins using D-amino acids. Because Gly residues with unusual torsion angles commonly occur in proteins, engineering secondary structural elements using D-amino acids in place of Gly could also have important implications for protein de novo designs.