Initial Insights into Structure-Activity Relationships of Avian Defensins*

Background: Avian defensins are antimicrobial peptides of a bird's immunity. Results: The target of chicken AvBD2 defensin is not chiral. Its structure is not amphipathic. The reduced and AvBD2-K31A forms dramatically decrease antibacterial activity. Conclusion: AvBD2 may disrupt the bacterial membrane through a nonchiral, nonspecific interaction. Significance: Knowledge of the structure-function relationships of avian defensins is a prerequisite for their use as alternatives to antibiotics. Numerous β-defensins have been identified in birds, and the potential use of these peptides as alternatives to antibiotics has been proposed, in particular to fight antibiotic-resistant and zoonotic bacterial species. Little is known about the mechanism of antibacterial activity of avian β-defensins, and this study was carried out to obtain initial insights into the involvement of structural features or specific residues in the antimicrobial activity of chicken AvBD2. Chicken AvBD2 and its enantiomeric counterpart were chemically synthesized. Peptide elongation and oxidative folding were both optimized. The similar antimicrobial activity measured for both l- and d-proteins clearly indicates that there is no chiral partner. Therefore, the bacterial membrane is in all likelihood the primary target. Moreover, this work indicates that the three-dimensional fold is required for an optimal antimicrobial activity, in particular for Gram-positive bacterial strains. The three-dimensional NMR structure of chicken AvBD2 defensin displays the structural three-stranded antiparallel β-sheet characteristic of β-defensins. The surface of the molecule does not display any amphipathic character. In light of this new structure and of the king penguin AvBD103b defensin structure, the consensus sequence of the avian β-defensin family was analyzed. Well conserved residues were highlighted, and the potential strategic role of the lysine 31 residue of AvBD2 was emphasized. The synthetic AvBD2-K31A variant displayed substantial N-terminal structural modifications and a dramatic decrease in activity. Taken together, these results demonstrate the structural as well as the functional role of the critical lysine 31 residue in antimicrobial activity.

Numerous ␤-defensins have been identified in birds, and the potential use of these peptides as alternatives to antibiotics has been proposed, in particular to fight antibiotic-resistant and zoonotic bacterial species. Little is known about the mechanism of antibacterial activity of avian ␤-defensins, and this study was carried out to obtain initial insights into the involvement of structural features or specific residues in the antimicrobial activity of chicken AvBD2. Chicken AvBD2 and its enantiomeric counterpart were chemically synthesized. Peptide elongation and oxidative folding were both optimized. The similar antimicrobial activity measured for both L-and D-proteins clearly indicates that there is no chiral partner. Therefore, the bacterial membrane is in all likelihood the primary target. Moreover, this work indicates that the three-dimensional fold is required for an optimal antimicrobial activity, in particular for Gram-positive bacterial strains. The three-dimensional NMR structure of chicken AvBD2 defensin displays the structural three-stranded antiparallel ␤-sheet characteristic of ␤-defensins. The surface of the molecule does not display any amphipathic character. In light of this new structure and of the king penguin AvBD103b defensin structure, the consensus sequence of the avian ␤-defensin family was analyzed. Well conserved residues were highlighted, and the potential strategic role of the lysine 31 residue of AvBD2 was emphasized. The synthetic AvBD2-K31A variant displayed substantial N-terminal structural modifications and a dramatic decrease in activity. Taken together, these results demonstrate the structural as well as the functional role of the critical lysine 31 residue in antimicrobial activity.
Defensins belong to a family of antimicrobial peptides characterized by cationicity, small size, ␤-sheet structure, and the presence of three disulfide bonds (1). Three subclasses (␣, ␤, and ) have been defined depending on the disulfide arrangement and the positioning of the six conserved cysteines. The ␣and -defensin families have been considered to evolve by duplication and divergence from ␤-defensin ancestor genes because the former are not reported in evolutionarily old vertebrates such as fish and bird classes. Defensins play a major role in both innate and adaptive immunity (2). They have been found to be constitutively or inducibly expressed by neutrophils and epithelial cells from many mammals and birds, including chicken (1,3,4). They display a wide range of microbicidal or microbistatic activities against Gram-negative and Gram-positive bacteria, fungi, and viruses (4). A substantial body of evidence indicates that the mechanism of action of defensins mainly relies on several structural features such as cationicity and amphipathy, which drive the antimicrobial peptide to interact with bacterial membranes and tend to divide peptides into two mechanistic classes, membrane-disruptive and nonmembrane-disruptive (5,6). In the latter case, there is growing evidence that defensins induce killing by acting on chiral anionic intracellular targets (for review see Refs. 7,8).
Interest in defensins as therapeutic drugs is growing because defensins may constitute an alternative to the controversial use of antibiotics. In birds, a potential use of these peptides has been proposed in particular to fight antibiotic-resistant bacteria, including Salmonella, a major zoonotic agent that causes food poisoning (9). Numerous ␤-defensins were identified in birds from isolated peptides or gene sequences (for review see Ref. 4). In a previous study, it was shown that chicken ␤-defensin genes (avBD1 and -2) were highly expressed in the intestinal tissue of birds that are resistant to Salmonella colonization (10). Three defensins (AvBD1, AvBD2, and AvBD7) were therefore purified from chicken bone marrow, and their antimicrobial activity was tested on a series of Gram-positive and Gram-negative bacteria (11). Only chicken AvBD2 was shown to be more active against Gram-positive than Gram-negative strains, as reported for the king penguin spheniscin (AvBD103b) (12), the only other avian ␤-defensin whose three-dimensional structure has been determined to date.
Concerning the molecular patterns involved in the activity of avian defensins, the sole data currently available refer to ostrich AvBD1 and AvBD2 defensins, which share 39 and 78%, respectively, of identity with chicken AvBD2. Ostrich defensins were shown to create a slow and partial depolarization of the Escherichia coli membrane but were unable to provoke bacterium death by membrane disruption (13). This indicated that the ostrich defensins could cross the bacterial membrane to target a cytoplasmic molecule. Considering that the ostrich defensins were efficient in shifting the mobility of bacterial DNA in a gel electrophoresis assay, it has been proposed that DNA could be the target of defensin (13). In the context of the long term objective of improving knowledge of immunity in birds, this work was carried out to gain information on structure-activity relationships of the chicken AvBD2 defensin, at the atomic level, which is an essential first step to understanding how avian ␤-defensins function.

Reversed Phase High Performance Liquid Chromatography
HPLC analyses were carried out on either an Elite LaChrom system composed of a Hitachi L-2130 pump, a Hitachi L-2455 diode array detector, and a Hitachi L-2200 autosampler or a LaChrom 7000 system composed of a Merck-Hitachi L-7100 pump, a Merck-Hitachi L-7455 diode array detector, and a Merck-Hitachi D-7000 interface, which was also used for semipreparative purification. The machines were equipped with C18 reversed phase columns (Nucleosil), 300 Å, 5 m, 250 ϫ 4.6 mm for the analytical separations, or 250 ϫ 10.5 mm for purification. Solvents A and B containing 0.1% trifluoroacetic acid (TFA) were H 2 O and MeCN, respectively.

Synthesis of the Oxidized Defensins
In a syringe fitted with a frit, the S-Acm-protected peptide resin (15 mol) was swollen in NMP (two times, 5 ml for 1 min). Silver tetrafluoroborate (58.4 mg per Acm group, 20 eq.) in NMP/H 2 O, 9:1 mixture (4 ml), was transferred to the resin by suction, and the resulting suspension was stirred by rotation for 5 min at RT, in the absence of light, followed by washes with NMP/H 2 O, 9:1, and then N,N-dimethylformamide. This treatment was repeated once (60 min of stirring), and the resin was further washed with pyridine (five times, 6 ml), then treated alternatively with sodium diethyldithiocarbamate (25 mM in NMP) and pyridine hydrochloride (1 M in CH 2 Cl 2 /MeOH, 95:5) (three and two times, 5 ml), followed by extensive washes with N,N-dimethylformamide. The peptide resin was then treated for 3 h at room temperature with TFA/H 2 O/i-Pr 3 SiH/PhOH, 87.5:5:2.5:5, and the linear peptide was precipitated by dilution with ice-cold diethyl ether. The crude reduced form of AvBD2 was dissolved in 20% acetic acid (AcOH) and purified by semipreparative C18 reversed phase HPLC.
The oxidative folding was performed at a peptide/GSH/ GSSG molar ratio of 1:100:10 in deoxygenated MeCN, 200 mM Tris-HCl buffer, pH 8.5 (50:50, v/v), containing 1 mM EDTA. The peptide concentration (50 g/ml) was measured using UV spectrophotometry at 280 nm (⑀ Trp , 5579 M Ϫ1 cm Ϫ1 ). The kinetics of the oxidative folding was monitored by analytic C18 reversed phase HPLC. Aliquots (100 l) of the folding reaction mixture were taken at regular intervals, and the reaction was stopped by adding 2 l of TFA before HPLC analysis. The oxidative folding was quantitative over 30 min. The peptide was purified on to a Resource S column (GE Healthcare) using a linear gradient of 0 -0.5 M NaCl in 50 mM Tris, pH 7.5. The fractions corresponding to the pure peptide were loaded on a Sep-Pak C18 column (6-ml column, Waters) followed by washing with 5% aqueous AcOH, eluted by MeCN/H 2 O/ AcOH, 5:4:1, and lyophilized.

Mapping of Disulfide Bridges by Proteolytic Cleavage and Mass Spectrometry
Proteolytic Cleavage-Protein cleavages were performed in a total volume of 20 l. To avoid the scrambling of disulfide bridges known to occur at basic pH, cleavages were performed in 30 mM ammonium acetate buffer adjusted to pH 6.5. Trypsin (Roche Diagnostics) cleavage of AvBD2 was performed at an enzyme/substrate ratio of 1:20 (w/w) for 4 h at 37°C. Papain (Roche Diagnostics) was incubated with AvBD2 for 4 h at 25°C using an enzyme/substrate ratio of 1:5 (w/w). For papain cleavage, the following amino acids were considered for proteinase specificity: Arg, Ala, Asn, Asp, Glu, Gln, Gly, His, Lys, Phe, Leu, and Tyr.
Mass Spectrometry-Intact and proteolyzed synthetic L-AvBD2 was analyzed by matrix-assisted laser desorption ionization-time of flight (MALDI-TOF) using an Autoflex instrument (BrukerDaltonics, Bremen, Germany) equipped with a 337-nm nitrogen laser and a gridless delayed extraction ion source. Sample deposition on the MALDI plate was performed using the ultrathin layer method as described previously (14,15). Samples were diluted at a ratio of 1:20 with a matrix solution consisting of 4-hydroxy-␣-cyanocinnamic acid (Bruker) saturated in a solution of 66.5% water, 33.3% MeCN, and 0.1% TFA. A 0.5-l aliquot of this analyte/matrix solution was spotted onto the ultrathin layer plate. The MALDI spot was irradiated using a 4 Hz laser pulse to produce ions. At least 200 laser shots were accumulated for each spectrum. Ions were analyzed in positive ion reflector mode with a 150-ns delay and an accelerating voltage of 19 kV. The measured m/z values correspond to the a0 peak as determined by the SNAP algorithm on the isotopic ion distribution. The spectra were calibrated externally and internally using the Pepmix calibrant mixture (Bruker) consisting of bradykinin, angiotensin, substance P, bombesin, renin substrate, adrenocorticotropic hormone 19 -38, and somatostatin. Instrument parameters were adjusted using Flex-Control (Bruker). Data analysis and internal calibration were performed using FlexAnalysis (Bruker). The disulfide-bridged cleavage peptides were mapped to the known AvBD2 sequence using the PeptideMap software tool from the PROWL website (The Rockefeller University, New York). Given the amino acid sequence of the protein and the proteinase cleavage specificity, PeptideMap automatically computes all theoretically possible combinations of bridged peptides and matches the observed masses to the corresponding theoretical masses.

Antimicrobial Activity Test
The antibacterial activities of the peptides were measured by radial diffusion assay (16) as described in Derache et al. (11) in gel containing either one of the following Gram-positive bacterial strains: Bacillus cereus ATCC 14579, Staphylococcus aureus ATCC 29740, and Listeria monocytogenes strain EGD, or one of the following Gram-negative bacterial strains: E. coli ATCC 25922, Salmonella enterica serovar Enteritidis ATCC 13076, and S. enterica serovar Typhimurium ATCC 14028. For each bacterial strain, three identical independent measurements of the antibacterial activity were performed. The minimal inhibitory concentration (MIC) of each peptide was determined from a graph constructed by plotting the log peptide concentration against the diameter of the clear zone on the plate minus the diameter of the well. The best fit straight line was determined using linear regression with GraphPad Prism 5 software (GraphPad Software). The MIC was calculated by finding the x-intercept of the line, indicating the peptide concentration where no clear zone was obtained. For each bacterial strain, the statistical difference between native and variant peptide MICs was assessed by comparing the slope and intercepts of both regression lines with GraphPad Prism 5 software (GraphPad Software). The level of significance was set at p Ͻ 0.05.

Circular Dichroism Experiments
The CD experiments were carried out on a Jasco J-810 spectropolarimeter. Solutions of 30 M (10 mM phosphate buffer, pH 7.2) of both L-AvBD2 and D-AvBD2 enantiomers were compared.

Three-dimensional NMR Structure
A standard set of two-dimensional 1 H NMR experiments (COSY, 80-ms TOCSY and 160-ms NOESY) was performed on a 0.1 mM aqueous solution of the synthetic L-AvBD2 (H 2 O/D 2 O 90:10 and 100% D 2 O) at pH 4.1 and at 293 K. An additional set of data, recorded at 303 K, was used to resolve assignment ambiguities due to spin system overlaps. All spectra were recorded on a Bruker 800 MHz spectrometer (NMR Facilities, Gif-sur-Yvette, France). The NMR data sets were processed using the NMRPipe/NMRDraw software package (17,18). 1 H chemical shifts were assigned according to classical procedures (19). NOE cross-peaks were integrated and assigned within the NMRView software (17). Covalent bonds were built between the sulfur atoms of the paired cysteines. Structure calculations were performed with the ARIA 1.1 software (20). The calculations were initiated using the default parameters of ARIA and a first set of easily assigned NOEs. At the end of each run, the new assignments proposed by ARIA were checked manually and introduced (or not) in the following calculation. This iterative process was repeated until complete assignment of the NOESY map. A last run of 1000 structures was then performed with the final list of NOE-derived distance restraints, and 200 structures were submitted to the last step on ARIA. The 10 structures without residual NOE violation and with the lowest residual NOE energy were selected and considered as characteristic of the peptide structures. Representation and quantitative analysis of the calculated structures were performed using MOL-MOL (21) and in-house programs.
The same sets of experiments were recorded on a VARIAN 600 MHz spectrometer for the variant AvBD2-K31A (2.4 mM of the synthetic peptide in aqueous solution at pH 4.2 and at 293 K). The same protocol was followed except that ambiguous constraints were introduced between cysteine residues, using the "ambiguous disulfide bridges" protocol of the ARIA 1.1 software (20).

Chemical Synthesis of AvBD2, Chemical and Functional Characterization Versus Extracted AvBD2
The peptide elongation of AvBD2 was carried out by solidphase peptide synthesis following the Fmoc/t-butyl strategy. Besides repeating most of the coupling steps twice, optimization of the elongation yield required the combined use of pseudoproline dipeptide derivatives and a polar resin (22). Our synthetic strategy also involved the use of the acetamidomethyl group as a TFA-stable protection of cysteinyl residues to obtain the linear S-Acm-alkylated AvBD2 (AvBD2-Acm). To obtain the linear nonalkylated AvBD2 from the same batch of peptide resin, we developed conditions for the removal of the Acm groups on the peptide resin before the final TFA treatment. After HPLC purification of the reduced form of AvBD2, the oxidative folding was carried out using a procedure based on a thermodynamically controlled disulfide shuffling, in the presence of reduced and oxidized glutathione at pH 8.5. The folding kinetics was followed by quantitative analytical HPLC, and the reaction was shown to be complete in 30 min (supplemental Fig. S1). MALDI-TOF MS analysis of oxidized AvBD2 showed a 6-Da difference in mass compared with the reduced form, consistent with the fully oxidized form of this peptide (data not shown). The oxidized AvBD2 was then purified to homogeneity by cation exchange chromatography. Reversed phase HPLC analysis showed that synthetic AvBD2 co-eluted with the natural product extracted from chicken bone marrow (supplemental Fig. S2). As further evidence of the identity of the synthetic and natural peptides, their activities measured in MIC assays were in the same range for every bacterial strain tested (supplemental Table S1). Altogether, our data validated an efficient optimized protocol for the production of highly pure and biologically active synthetic AvBD2. It was successfully applied to the synthesis of the all-D-enantiomeric homologues of AvBD2 (D-AvBD2) and the AvBD2-K31A variant (supplemental Fig.  S3). The all-D-form was checked by circular dichroïsm where CD spectra of the two enatiomers show equal and opposite spectra (supplemental Fig. S4). In the case of the AvBD2-K31A variant, the increase in hydrophobicity led to a poor folding yield. Organic solvents were then screened as folding additives (supplemental Table S2) and MeCN, which greatly enhanced the yield, was selected for preparative scale oxidative foldings. Our optimized protocol, including an efficient peptide elongation and the use of a co-solvent for the folding step, enabled an enhanced production yield up to 30 -40% for all the AvBD2 peptides.

Antibacterial Activity
The antimicrobial activities of D-AvBD2 and L-AvBD2 were tested on a selection of three Gram-positive (B. cereus, L. monocytogenes, and S. aureus) and three Gram-negative (E. coli, S. enterica serovar Enteritidis, and S. enterica serovar Typhimurium) bacterial strains. As shown in Table 1, the MICs measured for the two enantiomers are identical for every tested strain.
To investigate the role of the well conserved three-dimensional frame of ␤-defensin in AvBD2 functionality, the antibacterial activities of the linear S-alkylated AvBD2 (AvBD2-Acm) were compared with the activities of its oxidatively folded counterpart ( Table 1). The linear AvBD2-Acm is less active than the folded AvBD2 for every bacterial strain tested except for E. coli (p ϭ 0.06), as shown by the dramatic increase in the MIC of AvBD2 when linear. In particular, the linear form of AvBD2 is 10 and 16 times less efficient than the folded peptide against the Gram-positive strains B. cereus (p ϭ 0.0002) and L. monocytogenes (p ϭ 0.0002), respectively. The linear form is even ineffective in our conditions toward the Gram-positive strain S. aureus, showing the strict requirement of the three-dimensional fold for an optimal antimicrobial activity. The effect of the three-dimensional structure on the activity is more limited for the Gram-negative strains. Indeed, for S. enterica serovar Enteritidis and S. enterica serovar Typhimurium, the linear form of AvBD2 displays an activity one and a half (p Ͻ 0.0001) to three times (p ϭ 0.0002) lower than that of the folded AvBD2 peptide.

AvBD2 Solution Structure
Partial Determination of AvBD2 Disulfide Bridge Arrays-The determination of the correct disulfide pairing is generally achieved using enzymatic proteolysis of proteins and mass spectrometry analysis of the obtained cleavage products. These data can thus be introduced as additional constraints in structure calculations, allowing the three-dimensional models to converge more efficiently. For the chicken AvBD2 defensin, the However, these adjacent cysteines could not be differentiated. It is a known limitation of this method that, because of the impossibility to cleave between adjacent half-cystinyl residues, connected peptides containing a single disulfide bond cannot be obtained in such cases (22). intra-residues, sequential and easily determined long range peaks, was assigned. Additional assignments were progressively proposed during the ARIA runs (20) and manually validated. The use of ambiguous intersulfur distances, an option assuming that a given half-cystine is part of a bridge without supposing a particular partner, could not be successfully applied for AvBD2. This method is considered as a reliable and robust method for disulfide-rich proteins (23), but the calculations did not converge satisfactorily enough, even based on a very convenient set of NOEs (around 15 NOEs, including 3 long range restraints each). Therefore, it was more convenient to add the disulfide bridges as constraints. The two remaining possibilities were compared: 3-29, 8 -23, 13-30 or 3-30, 8 -23, 13-29. Convergence to a well formed three-stranded ␤-sheet was only obtained in the first case, with a convenient residual number of NOE violations and satisfactory energies. The last iterations to refine the structure were then performed with the 3-29, 8 -23, 13-30 disulfide bridges array. The final numbers of distance restraints used in the last run of ARIA calculations are detailed in Table 2. The solution structures of AvBD2 were represented by 10 conformers refined in a shell of water ( Fig. 2A) and were deposited in the Protein Data Bank with the 2gl5 entry code. The three-dimensional structure of AvBD2 displays the structural characteristics of ␤-defensins as follows: a three-stranded antiparallel ␤-sheet (Ser 7 -His 9 , Ile 18 -Ser 22 , and Ser 28 -Lys 31 ) stabilized by a conserved array of three disulfide bridges. The sequential Cys 29 and Cys 30 belong to the middle strand of the ␤-sheet; therefore, their side chains point in opposite directions. The measured distances greater than 8 Å between the sulfur atoms of Cys 13 and Cys 29 , or between Cys 3 and Cys 30 , definitively preclude the possibility of C1-C6; C2-C4, and C3-C5 pairing that could match NOE NMR data. The structures were in very good agreement with the experimental data; there was no violation of distance restraints larger than 0.3 Å. Most of the residues (92.7%) were found in the most favorable regions of the Ramachandran plot. On the whole, the secondary structure elements were well defined, and the root mean square deviation value calculated for secondary structures was 0.62 Å ( Table 2). The analysis of the surface properties clearly indicated that the positive and hydrophobic residues were well distributed on the three-dimensional structure of the molecules (Fig. 2, B and C). Contrary to many antibacterial molecules, AvBD2 did not display any amphipathic character, neither along the primary structure ( Fig. 1) nor on the three-dimensional structure of the molecules (Fig. 2).

Three-dimensional NMR Structure and Antibacterial Activity of AvBD2-K31 Variant
The protocol applied for AvBD2 was followed for AVBD2-K31A except that ambiguous constraints were introduced between cysteine residues, using the ambiguous disulfide bridges option. In the first calculations, each half-cystine was allowed to be linked to one of the five others, leading to 15 possibilities of pairing. During the calculations, each disulfide bridge is then allowed to float freely, and the protocol is driven to the most compatible disulfide bridges array, under the influence of the other NMR restraints. Once aberrant conformations (bridging more than two sulfur atoms) were discarded, a majority (72%) of structures correspond to the "3-29, 8 -23, 13-30" disulfide bridges array, the residual 28% corresponding to the "3-30, 8 -23, 13-29" pairing. At this stage, comparing two parallel calculations differing only by the disulfide bridges array imposed, 3-29, 8 -23, 13-30 or 3-30, 8 -23, 13-29, only the first calculations converged to a well formed three-stranded ␤-sheet. In the second calculations, the strands could not form properly. There was 25% more NOE violations, and the total energy was multiplied by 2. The refinement of the structure in the last iterations was then performed with the 3-29, 8 -23, 13-30 disulfide bridges array.
A very accurate model of AvBD2-K31A variant was determined by NMR (the root mean square deviation value calculated for secondary structures is 0.19 Å; Table 2). The structures were in very good agreement with the experimental data, and most of the residues (96.2%) were found in the most favorable regions of the Ramachandran plot. AvBD2-K31A displays the typical three-stranded anti-parallel ␤-sheet of ␤-defensins (6 -10, 18 -22, 27-31) stabilized by the conserved array of three disulfide bridges C1-C5, C2-C4, and C3-C6. An additional short anti-parallel ␤-strand was observed in the N-terminal part (Fig. 3). Finally, the C-terminal extremity systematically formed a final 3 10 helix turn at the end of the molecule, whereas this turn was observed on only two of the 10 AvBD2 solution structures. The assignment of all proton resonances has been deposited in the BioMagResBank (entry code 17798). Ten conformers representative of the AvBD2-K31A variant in solution have been deposited in the Protein Data Bank (2gl6 entry code).
The effects of this point mutation on the antimicrobial activity were measured. By comparison with the wild type AvBD2, the AvBD2-K31A variant exhibited a dramatic decrease in antimicrobial activity against the six bacterial strains tested, as shown by a significant increase of the MICs (Table 1). This result demonstrates the essential role played by this lysine residue at position 31 in antimicrobial activity. Furthermore, when comparing the effect of AvBD2-K31A on the Gram-positive and Gram-negative bacteria, the folded AvBD2-K31A variant was globally more damaging than the linear AvBD2 for the Gram-positive bacteria, whereas the opposite was observed for the Gram-negative ones. The linear form of the variant AvBD2-K31A was almost completely inactive in our conditions against all the bacterial strains (Table 1).

DISCUSSION
The aim of this work was to gain initial insights into the structure-activity relationships of avian ␤-defensins. We thus focused on the following three crucial questions. 1) Does AvBD2 require a chiral partner for its antimicrobial activity? 2) Is the three-dimensional structure of AvBD2 essential for the antimicrobial activity? 3) Can specific residues or features be pointed out as playing a role in the antimicrobial activity? To address these questions, chicken native AvBD2 and various peptides derived from its sequence were successfully synthesized in their linear or fully oxidized forms.
Involvement of a Chiral Partner in AvBD2 Antimicrobial Activity-To address the requirement of a chirality-dependent target in the antimicrobial activity, we synthesized and tested the all-D enantiomer of AvBD2. It is well established that the D-enantiomer of a native protein does not recognize the protein partners of the L-enantiomer or vice versa because of steric incompatibility (24,25). In the field of antimicrobial peptides, it was shown earlier that the all-D-enantiomer homologues of magainins and cecropins exert antimicrobial potency comparable with the naturally occurring all-L-peptides, which indicates the absence of a specific receptor-mediated mechanism and the achiral lipid chains of the cell membrane as the main target (26 -29). By contrast, during the discovery process of the outer membrane protein LptD as the chiral target of the peptidomimetic L27-11, it was shown that the all-D-enantiomer was essentially inactive (30). The measured MICs for both Dand L-AvBD2 enantiomers are identical toward various bacterial strains, either Gram-positives or Gram-negatives (Table 1). This clearly indicates that there is no chiral requirement for the antimicrobial activity. However, AvBD2 interacts with DNA in a gel shift assay (see supplemental Fig. S6) as reported for ostrich AvBD2 defensins (13), which share 78% of identity with chicken AvBD2 (Fig. 1). A similar behavior of AvBD2 and of its linear AvBD2-Acm form was observed in gel shift assays. These data suggested an unspecific interaction because of the cationic nature of the molecules, as already noticed for most antimicrobial peptides tested in vitro for their binding to nucleic acids (7). Therefore, the bacterial membrane appears to be the AvBD2 target.
Importance of the Three-dimensional Structure in AvBD2 Antimicrobial Activity-Although the structural organization of most defensins stabilized by a network of disulfide bonds is crucial to maintain the antimicrobial activity, some linearized defensins retain their antimicrobial activity (31). On the series of bacterial strains used in our studies, the linear AvBD2-Acm peptide always proved to be less active than the fully oxidized form, indicating the requirement of the three-dimensional fold for optimal antimicrobial activity. This requirement is particu-  larly critical for Gram-positive strains. To draw the first structure-activity relationships for bird defensins, we determined the three-dimensional NMR structure of chicken AvBD2. It displays the structural characteristics of ␤-defensins, i.e. a three-stranded antiparallel ␤-sheet, stabilized by the conserved array of three disulfide bridges (C1-C5, C2-C4, and C3-C6). Most mammalian ␤-defensins display an additional N-terminal helix. The king penguin AvBD103b, the only avian defensin three-dimensional structure that is currently available, displays a high propensity of the N-terminal part to form a helix in aqueous solution (32). By contrast, chicken AvBD2 lacks the possibility to form an N-terminal ␣-helix because of its shorter sequence (Figs. 1 and 2A). Hence, this helix appears to be nonessential for antibacterial activity. Although other studies have shown that helix conformation is essential for the action on zwitterionic lipid membranes, this structural feature appears less significant for the permeabilization of negatively charged bilayers (33)(34)(35). This helix may be involved in activity against fungi or host-cell membrane and indeed involved in selectivity, as suggested by the fungicidal activity of AvBD103b (12) compared with the lack of AvBD2 potency against Candida albicans (36).
Structural Features-The ability of antimicrobial peptides to cross bacterial membranes and/or disrupt them is often governed by amphipathy (37)(38)(39)(40). The analysis of the surface properties of chicken AvBD2 clearly showed that, contrary to many antibacterial molecules, AvBD2 did not display any amphipathic character (Fig. 2, B and C), even though it contains positively charged and hydrophobic residues. This organization of positive and hydrophobic residues, which are well distributed on the three-dimensional structure of the molecule (Fig.  2), certainly provides an appropriate equilibrium to interact with bacterial membranes.
To determine whether some of these positive and/or hydrophobic residues could play a role in this charge/hydrophobic equilibrium, the consensus sequence of the 32 avian ␤-defensins currently known was analyzed. As conserved residues often display a structural and/or a functional role in a given protein family, the consensus sequence was analyzed (Fig. 1) in the light of chicken AvBD2 and king penguin AvBD103b threedimensional structures (32), which share 33% of sequence identity with AvBD2. Globally, the amino acid composition of avian ␤-defensins is highly variable, and only the six cysteines were strictly conserved (Fig. 1). These six cysteine residues, involved in a conserved array of three disulfide bridges, ensure the high stability of the molecule and the high resistance to enzymatic degradation and therefore undoubtedly have a structural role. For AvBD2, three half-cystines, one for each bridge, were totally embedded in the core of the protein. Their accessibility to the solvent calculated with NACCESS software (41) was 8.4, 0.0, and 0.2% for Cys 8 , Cys 29 , and Cys 30 , respectively. (For AvBD103b, the corresponding Cys 5 , Cys 33 , and Cys 34 residues were totally embedded, with a solvent accessibility of 4.8, 0.1, and 0.3%, respectively). Subsequently, the consensus sequence highlighted two very well, but not strictly conserved, glycine residues (Gly 6 and Gly 21 ), belonging to Gly-Xaa-Cys motifs. Their role is most likely not only structural, but their presence could impact the neighboring residues. 1) Because of their small side chain, glycines are known to be highly flexible and to have a small steric size. At position 6, the short side chain of Gly 6 prevented steric "clashes," in particular with the bulky well conserved Lys 31 side chain of the "Cys-Cys-positive" motif (similarly, the totally embedded Gly 10 of AvBD103b prevented steric clashes with the bulky well conserved Arg 35 ). For two of the three exceptions not containing Gly at position 6 (mallard duck AvBD10 and chicken AvBD10, see Fig. 1), the Gly-Xaa-Cys and Cys-Cys-positive motifs are replaced by Gly-Xaa-Xaa-Cys and "Cys-Cys-Xaa-positive," respectively ( Fig. 1), which could ensure the same steric function. 2) The flexible and short side chain of Gly 21 (Gly 25 for AvBD103b), conserved in all 32 avian defensins except turkey AvBD3, chicken AvBD12, and chicken Gallins, was involved in a bulge where Val 20 -Gly 21 in the second strand of the ␤-sheet are facing Cys 29 in the third one (Ile 24 -Gly 25 facing Cys 33 for AvBD103b). This bulge could assist in placing the neighboring Val 20 residue (or Ile 24 in AvBD103b) in a favorable position, and/or it could ensure the proper folding of the protein (42), and/or it could give flexibility to this part of the protein (43). It is noticeable that this bulge is present in all the mammal ␤-defensin three-dimensional structures presently known: human hBD1-6, bovine BD12, and mouse mBD7-8 (Protein Data Bank codes 1kj5, 1fd3, 1kj6, 1zmm, 1zmp, 1zmq, 1bnb, 1e4t, and 1e4r, respectively). Moreover, the consensus sequence depicted in Fig. 1 highlighted well conserved positive residues at positions 4 and 31, and well conserved hydrophobic residues at positions 7, 10, 18, 20, and 26 (AvBD2 numbering), which did not seem to be involved in the fold itself, and consequently could have a functional role. The role of the well conserved hydrophobic residues at position 7 (but replaced by Ser in AvBD2) or at position 10 (but replaced by Arg in AvBD103b) is tricky to extrapolate with the only two three-dimensional structures available. They probably participate in the global hydrophobic/positive properties at the surface of the protein, as do the exposed Lys 4 and Phe 26 (Fig. 2D). Lysine 31 was pointed out (Arg 35 in AvBD103b). This positive residue keeps only its charged extremity accessible to the solvent, whereas its hydrophobic side chain is surrounded by the hydrophobic N-terminal Leu 1 and the well conserved Ile 18 and Val 20 residues, pointing toward the solvent (Fig. 2D). A similar feature is observed for AvBD103b, where the hydrophobic part of Arg 35 lies in a hydrophobic environment provided by Ile 22 , Ile 24 , and Val 37 pointing toward the solvent. In the case of AvBD103b, the positively charged extremity of Arg 35 , accessible to the solvent, was reinforced in the three-dimensional structure by two close additional positive charges, Arg 8 and Arg 9 .
Role of Lys 31 in the Antibacterial Activity and the Structure of AvBD2-To assess the structural role and to confirm, or reject, the functional role of the positively charged Lys 31 in the mechanism of bacterial killing by AvBD2 and/or in its specificity toward different bacterial strains, the AvBD2-K31A variant was synthesized and studied. The point mutation of lysine 31 by an alanine residue (K31A) caused a dramatic decrease in activity (Table 1), showing the critical functional role of Lys 31 . However, this point mutation also causes a large structural modification in the N-terminal part of the molecule (Fig. 3), where an additional N-terminal ␤-strand is formed. A fine analysis of the three-dimensional models showed that the side chain interactions between the hydrophobic parts of Leu 1 and Lys 31 , holding these residues in contact in AvBD2, are lost in the AvBD2-K31A variant. At this juncture it is not possible to precisely evaluate the contribution of these structural modifications to the decrease in activity. However, the critical functional and structural role of Lys 31 has been evidenced.
Global Cationicity Versus Structural Distribution of Charges-A common feature of most antimicrobial peptides/proteins is their net positive charge, which is essential for the initial association with bacterial membranes, through electrostatic interactions with the anionic surface of bacteria. However, the specificity of each defensin is certainly linked to its own distribution of charged and hydrophobic residues on the one hand and to the differences in the membrane composition of bacteria cell membranes on the other hand (Gram-positive versus Gramnegative, or between species). From our results, the global cationicity of the molecule, which is reduced in the variant form of AvBD2-K31A, appears to be more critical for Gram-negative strains. This could be explained by the higher exposure of negative charges on the Gram-negative bacterial surface because of the lipopolysaccharide. Moreover, the discrepancy we have observed in this study between the Gram-positive and Gramnegative susceptibility to linear AvBD2 might thus come from their difference in bacterial membrane accessibility and composition (44,45). However, the positive net charge of AvBD2 is one of the lowest among the avian ␤-defensins, and the AvBD2-K31A variant is charged only with three positive residues without losing all of its activity. In that variant, the loss of the threedimensional structure has a dramatic effect on activity, as shown by the MIC of AvBD2-K31A-Acm, which was almost above the concentration range (Table 1). Thus, even if cationicity seems to be more important in the mechanism of action of AvBD2 against the Gram-negative bacteria than against the Gram-positive ones, the role of the three-dimensional structure, and the associated distribution of positive and hydrophobic residues at the surface, predominates in the activity of this avian ␤-defensin. In the absence of any amphipathic character, the interaction of AvBD2 with the bacterial membrane may be governed by an adequate distribution of positive and hydrophobic residues at the surface, which could be described as an appropriate partition constant (46). Recently, it has been proposed that synthetic ␣-helical amphiphilic antimicrobial peptides (47), and the amphiphilic human hBD3 (48), may act like "sand-in-a-gearbox." This mechanism of action may be based on the ability of antimicrobial peptides to disrupt over space and/or time the highly dynamic membrane-bound protein complexes involved in essential processes of bacterial life. Even if not amphiphilic, AvBD2 could show an adequate partition constant to insert into the membrane, through nonchiral nonspecific interaction, and could disrupt the membrane equilibrium like sand in a gearbox.
Conclusion-The similar antimicrobial activity measured for both L-and D-enantiomeric chicken AvBD2 proteins clearly indicates that there is no chiral partner for the antimicrobial activity. Although the membrane emerges as the target, the resolution of the three-dimensional structure and the analysis of the AvBD2 surface revealed no amphiphilic distribution of its positively charged and hydrophobic residues. Thus, we propose that chicken AvBD2 antimicrobial activity may be based on a disorganization of the membrane through nonchiral nonspecific interaction.
Moreover, we highlighted a series of well conserved but not strictly conserved residues that could be involved in the antimicrobial properties and/or in the bacterial strain specificity of bird defensins. In particular, we pointed out lysine 31 of chicken AvBD2, lying in the hydrophobic environment provided by well conserved, accessible, hydrophobic residues. This study demonstrates the critical functional as well as structural role of Lys 31 in antimicrobial activity.