Functional Determinants of Human Enteric α-Defensin HD5

Background: Human α-defensin HD5 is a multifunctional antimicrobial peptide whose functional determinants have yet to be elucidated. Results: Alanine scanning mutagenesis aided by x-ray crystallography identified Leu29 at the dimer interface as crucial; N-methylation of Glu21 to debilitate HD5 dimerization also affected activity. Conclusion: Dimerization and hydrophobicity are important for HD5 function. Significance: The molecular basis of α-defensin function is better understood. Human α-defensins are cationic peptides that self-associate into dimers and higher-order oligomers. They bind protein toxins, such as anthrax lethal factor (LF), and kill bacteria, including Escherichia coli and Staphylococcus aureus, among other functions. There are six members of the human α-defensin family: four human neutrophil peptides, including HNP1, and two enteric human defensins, including HD5. We subjected HD5 to comprehensive alanine scanning mutagenesis. We then assayed LF binding by surface plasmon resonance, LF activity by enzyme kinetic inhibition, and antibacterial activity by the virtual colony count assay. Most mutations could be tolerated, resulting in activity comparable with that of wild type HD5. However, the L29A mutation decimated LF binding and bactericidal activity against Escherichia coli and Staphylococcus aureus. A series of unnatural aliphatic and aromatic substitutions at position 29, including aminobutyric acid (Abu) and norleucine (Nle) correlated hydrophobicity with HD5 function. The crystal structure of L29Abu-HD5 depicted decreased hydrophobic contacts at the dimer interface, whereas the Nle-29-HD5 crystal structure depicted a novel mode of dimerization with parallel β strands. The effect of mutating Leu29 is similar to that of a C-terminal hydrophobic residue of HNP1, Trp26. In addition, in order to further clarify the role of dimerization in HD5 function, an obligate monomer was generated by N-methylation of the Glu21 residue, decreasing LF binding and antibacterial activity against S. aureus. These results further characterize the dimer interface of the α-defensins, revealing a crucial role of hydrophobicity-mediated dimerization.

The exact mechanisms of defensin function are not fully understood. Human ␣-defensins kill Gram-positive bacteria preferentially to Gram-negative bacteria, suggesting a mode of action that depends upon bacterial physiology (14). The traditional view holds that electrostatic interactions and disruption of the bacterial membrane cause lethality, which appears to explain the sensitivity of the Gram-negative Escherichia coli (32). However, a recent study failed to find a correlation between ␣-defensin membrane activity and killing of the Gram-positive Staphylococcus aureus, instead attributing cell death to lipid II binding and sequestration (11), a mode of action reminiscent of the lantibiotic peptide nisin (33,34) and other defensins (35)(36)(37). Defensins also neutralize secreted bacterial toxins and virulence factors (1, 38 -42), including anthrax lethal toxin (43)(44)(45), which is a binary complex of two proteins secreted by B. anthracis: lethal factor (LF) 4 and protective antigen (46). LF is a Zn 2ϩ -dependent protease that induces cell death in macrophages through disruption of signaling pathways, and its inhibition has relevance for the development of therapies to treat anthrax exposure (47).
␣-Defensins adopt a three-stranded ␤-sheet core structure stabilized by three intramolecular disulfides and associate into a dimer via their second ␤-strands (48,49). The molecular basis of the myriad ␣-defensin functions is gradually being elucidated (1). Many of the prior studies have been focused on two model systems: HNP1 and ␣-defensins from mouse crypts, also known as cryptdins (50). Collectively, these studies have provided important insights into the structural and functional roles of disulfide bonding (51,52), cationicity (53,54), and conserved elements, such as the Arg-Glu salt bridge (55)(56)(57)(58) and invariant Gly residue (59,60) in the action of ␣-defensins. Recently, a comprehensive Ala scanning mutagenesis of HNP1 has discovered that a hydrophobic residue near the C terminus, Trp 26 , governs the ability of this ␣-defensin to kill Staphylococcus aureus, inhibit anthrax lethal factor, and bind HIV-1 envelope glycoprotein gp120 (61); methylation of a peptide bond at the putative dimer interface of HNP1 debilitates its dimerization and is functionally detrimental (62). Despite structural conservation, mammalian ␣-defensins share rather limited sequence identity. Whether or not these findings seen with HNP1 can be generalized to other members of the ␣-defensin family remains to be tested.
The two enteric human ␣-defensins HD5 and HD6 play important roles in intestinal host defense and homeostasis (63). Although several structure-function relationship studies have been reported on HD5, focusing on disulfide bonding, the conserved salt bridge, and selective cationic and hydrophobic residues (8, 52, 56, 64 -71), the molecular determinants of HD5 function remain largely unexplored. In the present study, an Ala scanning approach, aided by x-ray crystallography, enabled the thorough examination of the bactericidal properties and anthrax lethal factor inhibition of HD5 as a function of its sequence. Somewhat analogous to HNP1 Trp 26 , HD5 Leu 29 at the C terminus was identified as the most important residue. For functional interrogation, a series of hydrophobic amino acid substitutions were synthesized at position 29 using side chains of increasing aliphatic or aromatic chain length, including aminobutyric acid (Abu), norvaline (Nva), norleucine (Nle), Phe, and biphenylalanine (Bpa) (Fig. 1). To further study the effect of dimerization on HD5, Glu 21 at the dimer interface was N-methylated to produce an obligate HD5 monomer. Together, these mutant forms of HD5 portrayed the influence of Leu 29 and dimerization in the context of a richer understanding of the significance of hydrophobicity in the C-terminal region of the human ␣-defensins.
Surface Plasmon Resonance (SPR)-based LF Binding-SPR experiments were performed on a BIAcore T100 system (BIAcore, Inc., Piscataway, NJ), unless stated otherwise, at 25°C in 10 mM HEPES, 150 mM NaCl, 0.05% surfactant P20, pH 7.4 (with or without 3 mM EDTA). LF was immobilized on a CM5 sensor chip at a level of 2500 response units by the amine coupling protocol. Analytes were introduced into the flow cells at 30 l/min in the running buffer. Association and dissociation were assessed for 5 and 10 min, respectively. Resonance signals were corrected for nonspecific binding by subtracting the background of the control flow cell. After each analysis, the sensor chip surfaces were regenerated with 10 mM glycine solution (pH 2.0) and 50 mM NaOH and equilibrated with the buffer before the next injection. Binding isotherms were analyzed with BIAevaluation software and/or GraphPad Prism.
LF Inhibition Kinetics-The inhibition of LF by various defensins was quantified using an enzymatic kinetic assay (52). Briefly, freshly prepared LF at a final concentration of 1 g/ml (10 nM) was incubated at 37°C for 30 min with a 2-fold dilution series of defensin in 20 mm HEPES buffer containing 1 mM CaCl 2 and 0.5% Nonidet P-40, pH 7.2. 20 l of LF substrate (1 mM) was added to each well to a final concentration of 100 M in a total volume of 200 l. The enzyme activity, characterized as a time-dependent absorbance increase at 405 nm due to the release of p-nitroaniline, was monitored at 37°C over a period of 5 min on a 96-well V max microplate reader (Molecular Dynamics, Inc.). Data are presented in a plot showing percentage inhibition versus defensin concentration, from which IC 50 values (the concentration of defensin that reduced the enzymatic activity of LF by 50%) were derived by a non-linear regression analysis.
Virtual Colony Count-Antimicrobial assays against E. coli ATCC 25922 and S. aureus ATCC 29213 were conducted using a previously detailed 96-well turbidimetric method called "virtual colony count" (14). A 2-fold dilution series of defensin, ranging from 0.195 to 50 M in 10 mM sodium phosphate, pH 7.4, plus 1% tryptic soy broth, was incubated at 37°C for 2 h with E. coli or S. aureus (1 ϫ 10 6 cfu/ml), followed by the addition of twice concentrated Mueller-Hinton broth. Kinetic measurements of bacterial growth at 650 nm over 12 h were taken using a V max plate reader kept in a 37°C room. Data analysis utilized a Visual Basic script to calculate the time necessary for each growth curve to reach a threshold change in optical density at 650 nm of 0.02. To increase the sensitivity of bacterial killing, the phosphate buffer used in the 2-h incubation period contained 1% tryptic soy broth, as was used previously for defensins (30,52).
Crystallization, Data Collection, and Structure Determination-Crystallization screenings were conducted at room temperature using the hanging drop, vapor diffusion method and the commercially available crystallization Sparse Matrix Screens (Hampton Research). The drops were generated by mixing 0.5 l of defensin solution (prepared at 20 mg/ml in water) with 0.5 l of reservoir solution and placed over 0.8 ml of reservoir solution. The final compositions of the reservoir and cryosolutions are listed in supplemental Table S1. In all cases, the crystals appeared after 1 day and grew to their final sizes within a week. X-ray diffraction data were collected from flashfrozen crystals mounted on a rotating anode x-ray generator Rigaku-MSC Micromax 7 equipped with a Raxis-4ϩϩ image plate detector (at the X-ray Crystallography Core Facility, University of Maryland, Baltimore, MD) and at the Stanford Synchrotron Radiation Lightsource BL7-1 beamline on an ADSC QUANTUM 315 area detector. Crystal diffraction images were indexed, integrated, scaled, and merged using the HKL2000 package (74). The data collection statistics are shown in supplemental Table S2.
The structures of L29Abu-HD5, L29Nle-HD5, and MeGlu 21 -HD5 were solved by molecular replacement using the program Phaser from the CCP4 suite (75) with the HD5 monomer as a search model (Protein Data Bank entry 1ZMP (48)). The structures were refined with the program Refmac (76), and the models were corrected by manually refitting into the electron density and rebuilt using the program COOT (77). The refinement results are summarized in supplemental Table S2. The coordinates and structure factors have been deposited in the Protein Data Bank with accession codes of 4E86, 4E83, and 4E82 for L29Abu-HD5, L29Nle-HD5, and MeGlu 21 -HD5, respectively. Molecular graphics were generated using PyMOL (Schrodinger LLC, New York).

All Mutants Could Be Folded without Pro-domain Except for
L29A-HD5-For this study, we synthesized and folded wild type HD5 and its 28 analogs, including 21 alanine scanning mutants, five L29X-HD5 mutants, one N-methylated peptide, MeGlu 21 -HD5, and E21I-HD5. We excluded mutations of the six conserved cysteine residues and the invariant Gly 18 as well as the salt bridge-forming residues Arg 6 and Glu 14 because these residues ensure the structural integrity of HD5 and other ␣-defensins. We have previously established robust, facile synthesis and folding protocols for human ␣-defensins, which we employed to synthesize HD5 and all of its mutants except L29A. L29A could not be productively folded alone using the standard procedure, resulting in precipitation and a highly heterogeneous HPLC trace. However, ␣-defensin pro-domains have a chaperone-like effect on the oxidative folding of the mature defensin (30). Native chemical ligation (78,79) between the defensin and its pro-domain followed by oxidation resulted in the correctly folded full-length pro-defensin (56). The pro-domain was subsequently removed via cyanogen bromide cleavage, leaving the folded defensin intact. All peptides were purified to homogeneity by preparatory HPLC, and their experimentally determined molecular masses were in agreement with the calculated values based on their average isotopic compositions (data not shown).
Alanine Scanning Mutations Point to Leu 29 as Most Crucial Residue-We utilized SPR to assay the binding of the analogues to anthrax lethal factor. As shown in Fig. 2, wild type HD5 bound to LF with high affinity, as observed previously (52). Of all of the alanine scanning HD5 analogues tested, only S15A-HD5 bound to LF more efficiently. Interestingly, mutations of predominantly hydrophobic residues in the C-terminal region of the peptide, such as L26A, Y27A, and especially L29A, showed the weakest binding to LF of the panel of peptides. Also, mutations that reduced the net positive charge of the molecule (Arg-to-Ala substitutions) showed weaker binding compared with the wild type peptide only when residues were located toward the C terminus. For example, the R28A and R25A analogues showed weaker binding than R9A or R13A, with the final C-terminal R32A analog being the exception. Notably, the increase of cationicity due to the replacement of the single anionic residue in the sequence (E21A) also significantly reduced binding affinity for LF.
We have shown previously that wild type HD5 binds to and neutralizes the enzymatic activity of anthrax lethal factor or LF, a major virulence factor from Bacillus anthracis (52). We determined the effects of alanine mutations on the ability of HD5 to inhibit the enzymatic activity of LF. Fig. 2 shows the percentage residual LF activity as a function of the defensin concentration. Non-linear regression analysis was used to calculate the IC 50 values, listed in Table 1. Wild type HD5 showed a higher IC 50 value than the one reported previously (332 versus 192 nM) (52). As was the case for LF binding described above, most alanine scanning analogues inhibited LF activity comparably to the wild type peptide, as illustrated by only marginal differences in IC 50 values. Also consistent with the binding results, LF activity inhibition was weakest for analogues mutated in the C-terminal region. Notably, of all peptides tested, the L29A analog showed the weakest ability to inhibit LF activity (IC 50 ϭ 1672 Ϯ 500 nM). Unexpectedly, the G8A analog showed the second weakest LF inhibition (IC 50 ϭ 1418 Ϯ 183 nM), suggesting that this glycine may be structurally significant. Previous studies highlighted the importance of a glycine corresponding to Gly 18 that is conserved in all of the human ␣-defensins (59,60). With the exception of G8, these results suggest that the single alanine mutations that most severely affect LF interactions are near the C terminus. Furthermore, they suggest that a decrease in hydrophobicity is functionally deleterious for LF binding and inhibition of LF activity, more so than a decrease in cationicity.
We tested the bactericidal capacity of selected alanine scanning HD5 mutants, R9A, S17A, E21A, I22A, Y27A, and L29A, and also the E21I mutant, in which the negatively charged HD5 residue was replaced with the corresponding hydrophobic residue from the HNP1 sequence. Fig. 3 shows the survival of E. coli or S. aureus against HD5 and analogues ranging in concentration from 0.195 to 50 M. Virtual lethal dose values are given in Table 2. As observed previously, S. aureus was more susceptible to HD5 than E. coli (14,65). Strikingly, L29A-HD5 showed essentially no bactericidal activity against S. aureus, even at the highest concentration tested (50 M), and showed reduced killing of E. coli. The decrease in the positive charge of R9A-HD5 attenuated killing against both strains, whereas the increase in cationicity of E21A-HD5 and E21I-HD5 enhanced killing. Y27A-HD5 had strongly reduced killing of S. aureus but not E. coli.
HD5 L29X Mutations Alter Mode of Dimerization and Emphasize Role of Hydrophobicity-The results of bacterial killing and the LF binding and inhibition experiments showed the importance of Leu 29 for HD5 function. To further examine the nature of side chain hydrophobicity for the functional activity of HD5 at this position, we introduced the unnatural amino acids ␣-aminobutyric acid (L29Abu-HD5), norvaline (L29Nva-HD5), and norleucine (L29Nle-HD5). Hydrophobicity of these analogues is progressively restored compared with the alanine substitution by increasing the length of the aliphatic side chain. Two additional analogues were prepared, L29F-HD5 and L29-Bpa-HD5 (biphenylalanine), to determine the effects of aromatic hydrophobicity. Interestingly, although it was necessary to synthesize the L29A analog as a pro-peptide, the other Leu 29 analogues, L29F, L29Bpa, L29Abu, L29Nva, and L29Nle, folded correctly without the need for ligation to the pro-peptide. All peptides were purified to homogeneity, and their respective molecular masses were verified by mass spectrometry.
In addition, to validate the structural changes introduced by Abu and Nle substitutions, we solved the x-ray structures of L29Abu-HD5 and L29Nle-HD5 by a molecular replacement approach at resolution 2.75 and 1.9 Å, respectively   Table S2). L29A-HD5 also crystallized but diffracted poorly, and we were unable to solve the structure by molecular replacement. In all cases, a continuous and well defined electron density clearly confirmed the identity of each mutation as well as the connectivity of three disulfide bonds, which are identical with the wild type protein (Cys 3 -Cys 31 , Cys 5 -Cys 20 , and Cys 10 -Cys 30 ) (Fig. 4). The tertiary structure is very well preserved among all structurally independent monomers of HD5 analogs, which are essentially identical to each other and to the wild type HD5 (Fig. 5).
The root mean square deviations between 32 equivalent C␣ atoms within monomers of HD5 analogs and wild type HD5 are in the range of 0.36 -1.1 Å. Analysis of intermolecular contacts within L29Abu-HD5 and L29Nle-HD5 crystals indicate that these analogs form stable but distinct dimers in the crystals (Fig. 6). L29Abu-HD5 monomers associate into dimers that are very similar to the   6A and supplemental Fig. S1). The basket-shaped dimer is formed by the interaction of the long ␤2 strand of one monomer with the equivalent strand of the other monomer to form a four-stranded antiparallel ␤-sheet and the interaction of the short ␤1 strand of one monomer with the equivalent strand of the other monomer to form a two-stranded antiparallel ␤-sheet (Fig. 6A). The fourstranded sheet is stabilized by four backbone-backbone H-bonds donated reciprocally by Val 19 nitrogen and Glu 21 nitrogen of one monomer to Glu 21 oxygen and Val 19 oxygen of another, whereas the two-stranded sheet is maintained by two reciprocal backbone-backbone H-bonds formed by Cys 3 nitrogen and Cys 5 oxygen. The L29Abu-HD5 dimer is additionally stabilized through extensive hydrophobic interactions involving the side chain of Abu 27 and the Cys 3 -Cys 31 disulfide of one monomer and the side chains of Ile 22 and Tyr 27 and the Cys 5 -Cys 20 disulfide of the opposite monomer (Fig. 6A, middle). Formation of this dimer results in the average burial of more than 431 Å 2 of molecular surface per monomer, which compares with an average of 399 Å 2 for the wild type HD5 dimer (Protein Data Bank entry 1ZMP (48)). The slight increase in the average value of the molecular surface buried within intermonomer interactions of L29Abu-HD5 as compared with the wild type dimer could be attributed to the presence of two, instead of one, N-terminal H-bonds formed between Cys 3 and Cys 5 . Pairwise superposition of crystallographically independent dimers of L29Abu-HD5 and wild type HD5 indicates extensive structural conservation of all residues and results in an average distance between equivalent atoms in a range of 1.7-2.0 Å. Two structurally independent monomers of the asymmetric unit of the L29Nle-HD5 crystal form a dimer that is topologi-cally distinct from the "canonical" defensin dimer (Fig. 6B). L29Nle-HD5 monomers are arranged in a parallel fashion and form a four-stranded ␤-sheet by interaction of the ␤2 strand of one monomer with the ␤1 strand of another monomer. The dimerization is mediated through four hydrogen bonds formed between the backbone atoms of Val 19 and Glu 21 of the ␤2 strand and Ala 1 oxygen, Cys 3 nitrogen, Cys 3 oxygen, and Cys 5 nitrogen of the ␤1 strand (Fig. 6B, left) and hydrophobic contacts involving the side chains of Ile 22 , Tyr 27 , and Nle 29 and Cys n1 -Cys n2 disulfides of each monomer (Fig. 6B,  middle and right). The value of the molecular surface buried within intermonomer interactions was found to be 427 Å 2 /monomer.
We tested the bactericidal capacity of the L29Abu, L29Nva, L29Nle, L29F, and L29Bpa mutants ( Fig. 3 and Table 2). L29F-HD5 was more active than HD5 against E. coli in the region of the virtual lethal doses and had activity against S. aureus similar to that of HD5. L29Bpa-HD5 was notably less active against both strains, with a vLD 99.9 of 9.9 M versus E. coli and 12 M versus S. aureus compared with HD5 with vLD 99.9 values of 3.2 and 3.6 M, respectively. L29Abu-HD5 had little activity against S. aureus and less activity against E. coli than HD5. L27Nva-HD5 was less active against both strains. L29Nle-HD5 was more active against E. coli in the region of the virtual lethal dose values but less active against S. aureus in that region. In summary, aliphatic chain length is directly proportional to activity, and the single aromatic ring of phenylalanine gave better activity than the very bulky biphenylalanine side chain.
MeGlu 21 -HD5 Crystallized as a Monomer and Had Decreased LF Binding Activity, LF Inhibition, and Activity against S. aureus but Not E. coli-MeGlu 21 -HD5 was tested for its LF binding, LF-inhibiting, and antibacterial activity (Figs. 3 and 7 and Tables 1 and 2). MeGlu 21 -HD5 was less effective at inhibiting LF compared with wild type, as evidenced by a more than 2-fold increase in LF inhibition assay IC 50 , to 750 nM. LF binding was reduced 2.5-fold as measured by SPR. Against E. coli, MeGlu 21 -HD5 activity was similar to HD5. However, against S. aureus, virtual lethal dose values differed by as much as 6-fold, with a vLD 99.9 of 20 M for MeGlu 21 -HD5 compared with 3.6 M for HD5.
We determined the crystal structure of MeGlu 21 -HD5 at 1.7 Å resolution and found the overall structure of the monomer not altered, presenting all of the previously identified secondary structure elements and resembling very closely the wild type HD5 (Fig. 5). Analysis of intermolecular contacts within the  MeGlu 21 -HD5 crystal unambiguously rules out the formation of any quaternary structure. In this regard, monomers of MeGlu 21 -HD5 closely resemble dimerization-impaired MeIle 20 -HNP1 monomers (62). The MeGlu 21 -HD5 crystal structure once again validates the utility of the employed N-methylation method of the defensins' design, forcing the defensins to be represented in the crystal as monomers.

DISCUSSION
There are at least two distinct roles of hydrophobicity in the mechanism of ␣-defensins. One is that an amphiphilic structure is induced upon membrane binding with hydrophobic surfaces exposed for membrane penetration. Another is that hydrophobicity maintains the dimeric inter-   Table 1. face between defensin subunits. Our results emphasize the latter role.
Two biochemical underpinnings for defensin binding and activity are electrostatic attraction and the hydrophobic effect. The importance of positive charge in bacterial killing by defensins and their murine orthologues the cryptdins, viewed as essential for electrostatic interaction with the anionic bacterial membrane, is experimentally well supported (53,54,64). In further support, anionic residues in the pro-domain of cryptdins prevent their antibacterial activity (80 -82), although more recent studies point to hydrophobicity of the pro-domain as the major factor governing the intramolecular functional inhibition of proHNPs (29). Although the interplay of cationicity and hydrophobicity clearly dictates defensin function, growing evidence suggests that the latter plays a dominant role in the action of human neutrophil ␣-defensins HNP1 to -3 (61), which are significantly less cationic than cryptdins or typical ␤-defensins. The work described in this report on HD5 lends additional support to the findings seen with human neutrophil peptides and underscores the importance of hydrophobicity for human ␣-defensin function.
We find that C-terminal hydrophobic residues of HD5, and in particular Leu 29 , are important for its bactericidal activity. Substitution of Leu by Ala abolished S. aureus killing and severely decreased E. coli killing. The lost bactericidal activity of L29A-HD5 against both strains progressively improved as the hydrophobicity of residue 29 in L29X-HD5 (where X represents Abu, Nva, Nle, or Phe) increased, and both Nle and Phe were improvements over Leu against E. coli. Similarly, L29A was the most deleterious mutation for LF binding and inhibition by HD5, and the L29F mutation fully rescued the lost function of L29A-HD5 against LF. These results are in accordance with those previously reported for W26X-HNP1 analogs (61). Because both Trp 26 (HNP1) and Leu 29 (HD5) are bulky hydrophobic residues near the C termini of the respective defensins, they point to hydrophobicity, rather than cationicity, as the keystone of human ␣-defensin activity.
Members of the ␣-defensin family from several species (human, mouse, rhesus macaque, rabbit, guinea pig, and rat) commonly contain Leu or Phe at the position just before the fifth and sixth cysteines (55), which corresponds to Leu 29 of HD5 or Phe 28 of HNP1. Not surprisingly, in the previously published Ala scanning mutagenesis of HNP1 (61), the F28A mutation was the second most deleterious for bactericidal activity against S. aureus, next to W26A. However, because Tyr 27 of HD5 corresponds to Trp 26 of HNP1, the question remains; why is the Y27A mutation substantially less detrimental than the L29A mutation in HD5 and the W26A or F28A mutation in HNP1? The answer probably lies in differences in the mode of dimerization between the two ␣-defensins HNP1 and HD5 (Fig.  8) as well as different hydrophobicities of Leu, Phe, Tyr, and Trp residues.
Structural analysis of the HNP1 dimer indicates that Trp 26 in each monomer stacks against Tyr 21 , which, in turn, makes hydrophobic interactions with Phe 28 of the opposing monomer, among others (62). The dimer is also stabilized by four reciprocal backbone H-bonds formed by Thr 18 and Ile 20 . According to the Wimley-White whole-residue hydrophobicity scale (water-octanol) (83,84), Trp is more hydrophobic than Phe, with Tyr being the least hydrophobic among the three aromatic residues. Thus, Trp 26 and, to a lesser extent, Phe 28 prominently mediate HNP1 dimerization, which strongly correlates with defensin activity (61,62). Compared with HNP1, however, the C terminus of HD5 is significantly more polar, and the only aromatic residue in the region, Tyr 27 (corresponding to Trp 26 in HNP1), does not directly participate in dimer formation as shown in the crystal structure of HD5 (48). In fact, the dimer interface of HD5 is formed mainly by Ser 17 , Leu 29 , and the Cys 3 -Cys 31 disulfide bond of one monomer, which pack against the Cys 5 -Cys 20 disulfide bond and Ile 22 of the opposing monomer. On the Wimley-White whole-residue hydrophobicity scale, Leu is still more hydrophobic than Tyr, albeit less than Phe and Trp (83,84). It is therefore not surprising that Leu 29 was found to be more important than Tyr 27 for HD5 function despite the fact that the latter and Trp 26 of HNP1 occupy equivalent positions in the amino acid sequences. Of note, the difference in hydrophobicity between Leu and Trp may provide a ready explanation for the finding that W26A-HNP1 showed a greater decrease in LF binding and inhibition than L29A-HD5 compared with the respective wild type peptides. W26A-HNP1 decreased the inhibition of LF enzymatic activity 19-fold compared with wild type HNP1, whereas L29A-HD5 decreased the inhibition of LF activity only 5-fold compared with HD5.
As is the case with Trp 26 in HNP1, the functional importance of Leu 29 probably stems from its prominent role in mediating HD5 dimerization, contributed additionally by four reciprocal intermolecular backbone-backbone H-bonds involving Val 19 and Glu 21 . This has been confirmed at both the structural and functional levels by methylation of the Cys 20 -Glu 21 peptide bond to debilitate HD5 dimerization. In fact, MeGlu 21 -HD5 crystallized as an obligate monomer, and its bactericidal activity against S. aureus and inhibition of LF were attenuated in a similar way to the corresponding mutant of HNP1, MeIle 20 -HNP1, as was recently reported (62). However, the deleterious effects of HD5 methylation at Glu 21 are much smaller than those seen with MeIle 20 -HNP1. The decrease in activity of MeGlu 21 -HD5 compared with HD5 was 6-fold against S. aureus, as measured by vLD 99. 9 . By contrast, MeIle 20 -HNP1 did not achieve 99 or 99.9% killing of S. aureus, resulting in vLD 99 and vLD 99.9 values of Ͼ50 M. Similarly, although MeIle 20 -HNP1 had an 11-fold greater IC 50 than HNP1 with respect to LF inhibition, MeGlu 21 -HD5 had only a 2-fold increase in IC 50 (750 nM) compared with HD5 (332 nM). Thus, dimerization appears to be functionally more important for HNP1 than HD5.
Compared with HNP1, the attenuated functional effects of the L29A mutation and Glu 21 methylation in HD5 probably underscore differences in the mode of dimerization between HNP1 and HD5. In contrast to the HNP1 dimer, which is stabilized by only four reciprocal intermolecular backbone-backbone H-bonds, the HD5 dimer is stabilized additionally by one H-bond formed by the N-terminal backbone atoms of Cys 3 . This additional H-bond contributes significantly to the solvation energy gain of the interface. These differences of dimerization features are well revealed by structural studies of L29Abu-HD5 and the previously published W26Abu-HNP1 (61). The W26Abu mutation caused little change to HNP1 tertiary and quaternary structures except for an increased local structural mobility of Tyr 21 at the dimer interface (61). By contrast, the L29Abu mutation slightly increased the surface area buried at the dimer interface of HD5 due to enhanced intermolecular interactions of N termini accompanied by a conformational adjustment. This subtle but favorable structural change presumably offsets, at least partially, the deleterious mutational effects of L29Abu at the functional level. These structural findings also suggest that molecular defects at the dimer interface of HD5 may be functionally tolerable. Indeed, the L29Nle mutation caused a dramatic switch of the HD5 dimer from the (canonical) anti-parallel mode to a (non-canonical) parallel mode to avoid steric clashes at the dimer interface involving the elongated side chains of Nle-29. Despite this marked change in the mode of dimerization, L29Nle-HD5 had S. aureus-killing and LF-inhibiting/binding activities comparable with the wild type defensin HD5. Of note, in contrast to HNP1, structural plasticity seen in HD5 dimerization presumably makes it diffi-cult to accurately predict its activity on the basis of its tendency to dimerize, oligomerize, or multimerize.
Leu-Leu interactions appear in protein motifs that stabilize dimeric structures, such as Leu-rich repeats (85) and Leu zippers (86), to achieve specific binding to ligands such as DNA. This general hydrophobic property of Leu is exploited in HD5 on a much smaller scale, where a pair of Leu 29 side chains interact indirectly through the hydrophobic patch of the dimer interface. Changing Leu 29 to Ala might have thermodynamic consequences similar to those of Leu-to-Ala substitutions that cause cavities within the hydrophobic core of T4 lysozyme (87,88) or Leu zippers (89). The "cavities" in the case of L29A-HD5 would lie at the dimer interface, where the absence of the van der Waals contacts of the bulky Leu side chain might greatly decrease its hydrophobic surface area or eliminate dimerization altogether. Quite possibly, the Abu side chain might be the shortest permissible that provides sufficient hydrophobicity. Indeed, mutating Leu 29 to the smaller and less hydrophobic Ala caused aggregation and misfolding, which was only alleviated through the chaperone-like pro-peptide. Replacing Leu 29 with larger hydrophobic side chains, however, resulted in productively folded peptides. The series of L29X mutations of increasing hydrophobicity had self-association and activity proportional to side chain size with the exception of the extremely bulky Bpa. The results with Bpa indicate that there is a limit to the beneficial effect of hydrophobicity; a side chain much bulkier than that of Leu apparently cannot be accommodated by the defensin dimer without steric clashes.
Finally, unlike S. aureus, E. coli killing by HD5 or HNP1 was more sensitive to the change in cationicity than hydrophobicity. Moreover, MeGlu 21 -HD5 activity was comparable with HD5 activity against E. coli, indicating that E. coli killing is independent of defensin dimerization. We and others have shown that a number of Gram-negative bacterial strains can be killed effectively independent of ␣-defensin structure and chirality (1). The present study further underscores the mechanistic difference in the bactericidal activity of ␣-defensins against these two bacterial strains, one mediated by achiral electrostatic attractions, such as to membrane phosphate groups, and the other mediated partially by hydrophobic interactions that result in specific binding to molecules, such as lipid II. Thus, strain selectivity could be explained by differences in defensin binding as it relates to the physiology of the bacterial cell wall and inner or outer membranes.
In summary, the results obtained from the comprehensive alanine scanning mutagenesis and the series of L29X mutations have identified hydrophobicity as the dominant functional trait, more important than cationicity (as shown by the selection of L29A over arginine mutations) and even more important than canonical dimerization (as shown by the relatively mild effects of monomerization in the MeGlu 21 -HD5 mutant and the reorientation of the quite functional L29Nle-HD5 dimer). Disrupting dimerization appears to be less deleterious for HD5 than for HNP1, which also forms multimers that have been modeled to be governed by Ile 20 and Leu 25 (62). Despite the variation in the importance of quaternary structure among the human ␣-defensins, hydrophobicity-mediated ␣-defensin dimerization, oligomerization, and multimerization appear to be an overriding molecular event that governs the great variety of biological activities of these multifunctional peptides.