Solution structure of the recombinant penaeidin-3, a shrimp antimicrobial peptide.

Penaeidins are a family of antimicrobial peptides of 47-63 residues isolated from several species of shrimp. These peptides display a proline-rich domain (N-terminal part) and a cysteine-rich domain (C-terminal part) stabilized by three conserved disulfide bonds whose arrangement has not yet been characterized. The recombinant penaeidin-3a of Litopenaeus vannamei (63 residues) and its [T8A]-Pen-3a analogue were produced in Saccharomyces cerevisiae and showed similar antimicrobial activity. The solution structure of the [T8A]-Pen-3a analogue was determined by using two-dimensional 1H NMR and simulated annealing calculations. The proline-rich domain, spanning residues 1-28 was found to be unconstrained. In contrast, the cysteine-rich domain, spanning residues 29-58, displays a well defined structure, which consists of an amphipathic helix (41-50) linked to the upstream and the downstream coils by two disulfide bonds (Cys32-Cys47 and Cys48-Cys55). These two coils are in turn linked together by the third disulfide bond (Cys36-Cys54). Such a disulfide bond packing, which is in agreement with the analysis of trypsin digests by ESI-MS, contributes to the highly hydrophobic core. Side chains of Arg45 and Arg50, which belong to the helix, and side chains of Arg37 and Arg53, which belong to the upstream and the downstream coils, are located in two opposite parts of this globular and compact structure. The environment of these positively charged residues, either by hydrophobic clusters at the surface of the cysteine-rich domain or by sequential hydrophobic residues in the unconstrained proline-rich domain, gives rise to the amphipathic character required for antimicrobial peptides. We hypothesize that the antimicrobial activity of penaeidins can be explained by a cooperative effect between the proline-rich and cysteine-rich features simultaneously present in their sequences.

Only recently, such effectors of innate immunity were isolated from crustaceans, whereas numerous peptides have been characterized from other arthropods, both insects and chelicerates. Three peptides, named penaeidins (Pen), 1 were initially purified in their active form (5.48Ϫ6.62 kDa) from the hemocytes of the shrimp Litopenaeus (Penaeus) vannamei, and they were fully characterized at the amino acid level (Pen-1, Pen-2, Pen-3a) and by cDNA cloning from a hemocyte library (pen-2, -3a, -3b, and-3c) (11). Penaeidins are an original peptide family composed of an N-terminal proline-rich sequence and a Cterminal region containing six cysteines engaged in three intramolecular disulfide bridges. Recently, using the molecular EST approach (Expression Sequence Tag) and sequence homology search, members of the penaeidin family have been identified in several shrimp species, including Litopenaeus setiferus (12,13), Penaeus monodon, 2 Fenneropenaeus chinensis, 3 Litopenaeus stylirostris, 4 and Penaeus japonicus (14). These data confirm that the penaeidin family is characterized by the presence of two distinct domains and by the conserved position of the six cysteines in the C-terminal domain. Interestingly, these studies have also revealed a new class of penaeidins, Pen-4 ( Fig. 1), and as evidenced for the Pen-3 class, an abundance and diversity of isoforms can be produced in a single species (13). Penaeidins are synthesized as precursor molecules consisting of a peptide signal (19 -21 residues) preceding the mature bioactive peptide. Pen-1, -2, and Pen-4 have between 48 * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The atomic coordinates and structure factors ( and 50 amino acids, whereas Pen-3 class peptides are the longest peptides (63 amino acids), characterized by the insertion of a medial sequence (3-5 amino acids), linking together the proline-rich and the cysteine-rich domains, and by additional residues at the C-terminal sequence. Pen-2 and Pen-3a were produced in the recombinant system Saccharomyces cerevisiae to study their biological properties and their spectra of antimicrobial activities (15). These peptides exhibit broad activity at rather low concentration (below 10 M) against mainly Gram-positive bacteria and filamentous fungi, and comparatively weak activity against the Gram-negative bacteria (ranging from 10 to 50 M). In addition, for Pen-3a, a chitin-binding property has been evidenced, which appeared to be displayed by the C-terminal region (16).
Penaeidins are original peptides, which possess six cysteine residues organized in two doublets in their C-terminal domain. This is unusual in antimicrobial peptides since only vertebrate defensins, namely ␤-defensins, and hepcidin are known to present one and three cysteine doublets, respectively. Hepcidin (initially called liver-expressed antimicrobial peptide, LEAP-1), which displays antimicrobial activity, is presently known as an iron-regulatory hormone (17,18). Neither the disulfide bond arrangement nor the three-dimensional structure of the penaeidins has yet been reported. For this reason, the Pen-3 family, which appears to be most abundantly produced in shrimp, was chosen for such a structural study. Given the fact that native peptides are difficult to purify from natural samples, the Pen-3a isoform was expressed in a heterologous expression system (15). In this study, we report the solution structure of the recombinant Pen-3a, as determined by 1 H NMR derived-constraints, and its disulfide arrangement. Finally, from the determined structure, the distribution of positively charged and hydrophobic side chains, both in the unconstrained proline-rich and in the highly constrained cysteinerich domains is discussed in terms of the amphipathic character required for antimicrobial activity of peptides.

EXPERIMENTAL PROCEDURES
Expression and Purification of Pen-3a and of Its [T8A] Analogue-Pen-3a used in this study was obtained as previously described (15) by recombinant expression in the S. cerevisiae strain TGY 48-1 (20). Briefly, the Pen-3a coding sequences fused to the yeast ␤-glucanase/ MF␣1 preprosequence, were subcloned into a shuttle vector pTG4812. This construct was used to transform the yeast. An analogue of Pen-3a having the Thr to Ala substitution at residue 8, and termed [T8A]-Pen-3a analogue, was also expressed in the same yeast strain. The substitution was performed using the QuickChange site-directed mutagenesis kit (Stratagene) by changing only one nucleotide from the Pen-3a coding sequence. 5 S. cerevisiae TGY-transformed cells, expressing respectively Pen-3a and its [T8A]-Pen-3a analogue, were inoculated into selective medium YNBG (yeast nitrogen base 0.67%, glucose 1%) supplemented with 0.5% casamino acids lacking uracil. After 72 h of incubation at 30°C under vigorous shaking, the cultures were centrifuged (1000 ϫ g, 20 min, 4°C) and the supernatants acidified to pH 2.9 by addition of trifluoroacetic acid. Large scale purifications of recombinant penaeidins were essentially performed as previously described (15). To increase peptide purity, the molecules were finally purified on a RP18 reversedphase column (Symetry Shield TM RP18; 4.6 ϫ 250; 5 mm; Waters) by a linear biphasic gradient of 2-18% acetonitrile in acidified water over 10 min, and of 18 -23% over 40 min at a flow rate of 0.8 ml/min. Finally, the fractions containing the penaeidins were pooled, lyophilized, and kept as dry powder.
Antimicrobial Assays-Antimicrobial activity of the [T8A]-Pen-3a analogue was assayed against different micro-organisms already used in a previous study relative to the activity spectrum study of the recombinant Pen-3a (15). The test microorganisms were the Grampositive bacteria Aerococcus viridans, Micrococcus luteus, and Bacillus megaterium; the Gram-negative strains, Escherichia coli 363, Salmonella thyphimurium, Klebsiella pneumoniae, and marine bacteria Vibrio harveyi pathogenic for shrimp. Finally, the filamentous fungus Fusarium oxysporum isolated from shrimp was also used.
MICs were determined in triplicate by the liquid growth inhibition assay as described (11). Poor broth nutrient medium (PB: 1% bactotryptone, 0.5% NaCl, w/v, pH 7.5) or saline peptone water (1.5% peptone, 1.5% NaCl w/v, pH 7.2) was used for the standard bacterial strains and the marine bacteria, respectively. Briefly, in a sterile microtitration plate, 10 l of peptide, or deionized water as a control was added to 90 l of a mid-logarithmic growth phase culture of bacteria diluted in culture medium to OD 600 ϭ 0.001. Plates were incubated for 24 h at 30°C with vigorous shaking and monitored spectrophotometrically at 620 nm. An identical liquid growth inhibition assay as that used to measure anti-bacterial activity was used to measure anti-fungal activity on Potato Dextrose Broth (Difco) medium. After incubation for 48 h at 25°C in the dark without shaking and in a moist chamber, growth inhibition of the fungus was measured at 600 nm.
Trypsin Cleavage-Pen-3a and its [T8A]-Pen-3a analogue (100 g) were subjected to bovine trypsin (Promega Corporation) treatment with a trypsin/substrate ratio of 1/20 (w/w) in 500 l of a 200 mM Tris-HCl buffer (pH 8.0) at 37°C. After 2, 5, 10, and 17 h of incubation, 100-l aliquots of the solution were frozen to stop the reaction and freezedried. ESI-MS spectra were recorded and analyzed as described above.
NMR Spectroscopy-D 2 O (99.95%) was purchased from the CEA (Saclay, France). The pH values were measured at room temperature with a 3-mm electrode and are given uncorrected for the deuterium isotopic effect. 1 H chemical shifts were referenced with respect to sodium 4,4-dimethyl-4-silapentane-1-sulfonate (DSS) according to the IUPAC recommendations (22).
Proton NMR experiments were performed both on a Bruker Avance 600 and Avance 500 spectrometers equipped with a triple resonance probe and with a cryoprobe, respectively. In all experiments, the carrier frequency was set at the center of the spectrum at the water frequency. To identify several overlapping spin systems, three sets of spectra were recorded at 12, 20, and 27°C. DQF-COSY (23), z-TOCSY (24), and NOESY (25) spectra were acquired in the phase-sensitive mode using the States-TPPI method (26). For spectra recorded in H 2 O, and except for the DQF-COSY spectra (where low-power irradiation was used), the water resonance was suppressed by the WATERGATE method (27). z-TOCSY spectra were obtained with a mixing time of 50 and 70 ms and NOESY spectra with a mixing time of 100, 150, and 200 ms.
Slowly exchanging amide protons were identified from successive TOCSY and NOESY spectra recorded at 27°C following the solubilization of Pen-3a and of [T8A]-Pen-3a in D 2 O from 15 min to 24 h. Data were processed by using either the XWINNMR or the GIFA (28) software. The data were zero-filled before processing and shifted sinebell functions were used for apodization. The processed data were baseline corrected using a five-order polynomial function. The full sequential assignment was achieved using the general strategy described by Wü thrich (29).
Structure Calculation-The NOESY cross-peaks were measured from the NOESY spectrum acquired at 27°C with a mixing time of 150 ms and were divided into five classes, according to their intensities. Very strong, strong, medium, weak, and very weak NOEs were then converted into 1.8 -2.4, 1.8 -2.8, 1.8 -3.6, 1.8 -4.4, and 1.8 -5.0 Å distance constraints, respectively. For equivalent protons or non-stereospecific assigned protons, pseudo-atoms were introduced. The angle restraints were derived from the 3 J HN-C␣H coupling constants, and the 1 angle restraints were derived from the combined analysis of the 3 J H␣-H␤ coupling constants and intra-residues NOEs, respectively.
To calculate three-dimensional structures, these distance and dihe-5 C. Zatylny, unpublished results. dral angle restraints were used as input in the standard distance geometry (DG)/simulated annealing (SA) refinement and energy-minimization protocol using X-PLOR 3.8 (30). In the first stage of the calculation, an initial ensemble of 60 structures was generated from a template structure with randomized , dihedral angles and extended side chains, using a DG protocol followed by restrained SA and refinement (31). No hydrogen bond restraint was used. Since the arrangement of the disulfide bonds had to be determined, structures were generated without using any disulfide bond constraints. Analyzing the obtained structures and comparing them with the NMR data allowed us to identify more additional NOE restraints, which were introduced into the subsequent calculation. After a number of these processes, 552 NOE-derived distance restraints (73 medium range and 93 long range) and 28 dihedral angles (15 and 13 1) were used as final input data. Analysis of the average S␥-S␥ distances in the lowest energy structures allowed us to unambiguously determine the arrangement of the three disulfide bonds. Finally, a calculation of 60 conformers including the disulfide bonds was carried out, and the resulting 20 structures with a minimum of restrained violations were submitted to 5000 cycles of restrained Powell energy minimization.
Structure Analysis-The visual display and the calculation of rmsd were performed with INSIGHT 97 (Molecular Simulation Inc., San Diego). Hydrogen bonds were considered as present if the distance between heavy atoms was less than 3.5 Å and the donor hydrogen acceptor angle was greater than 120°. The Ramachandran analysis was performed with PROCHECK (32), and the limits of the secondary structure elements and the van der Waals surfaces were determined with the STRIDE program (33). The chemical shifts and coordinates of the energy-minimized conformers of [T8A]-Pen-3a are deposited in the BMRB (accession number 5806) and in the Protein Data Bank (PDB entry: 1ueo), respectively.

Production of Pen-3a and of Its [T8A]
Analogue-S. cerevisiae-expressed Pen-3a was purified by reversed-phase HPLC. As already shown in a previous study, in addition to molecules similar to the native one, several forms of the recombinant penaeidins were found to be glycosylated (15). Despite additional successive purification steps performed to obtain highly pure Pen-3a suitable for the structural study, residual contaminating glycosylated molecules were observed (see below). Nevertheless, a small amount of the non-glycosylated peptide was isolated and characterized by ESI-MS (6686.62 Da). It displayed a glutamine residue in its N terminus and therefore corresponded to the non-post-translationally modified Pen-3a [the N terminus of native Pen-3a was shown to be a pyroglutamic amino acid (11)]. To obtain enough non-glycosylated material to carry out the structural study, a [T8A]-Pen-3a analogue was further produced. Since in a previous study (15), the threonine in position 8 was identified as a site of O-glycosylation, this residue was substituted for alanine by directed mutagenesis to avoid glycosylation events in the yeast. The [T8A]-Pen-3a analogue was expressed in S. cerevisiae, purified by RP-HPLC, and characterized by ESI-MS and by NMR. Its monoisotopic mass (6656.30 Da) was in agreement with the theoretical one of 6656.26 Da (Table I). Its 1 H NMR spectrum was similar to that of the parent peptide suggesting that they share the similar fold.
Antimicrobial Activity-The antimicrobial activity of the recombinant Pen-3a and of its [T8A]-Pen-3a analogue was measured using a panel of microorganisms including Gram-positive and Gram-negative bacteria, and the filamentous fungus F. oxysporum. The MIC values obtained for the recombinant Pen-3a and the [T8A]-Pen-3a analogue were found to be almost identical and revealed similar ranges of activity (Table II). The antimicrobial activity appeared to be target-specific with great inhibition of some of the Gram-positive bacteria strains such as A. viridans (with MIC less than 1.6 M). As for Pen-3a, an anti-Gram-negative activity was recorded for [T8A]-Pen-3a against E. coli, whereas no inhibition of the other bacterial strains tested could be observed for either peptides up to 50 M.
Thus, we concluded that the [T8A] mutation did not affect the antimicrobial activity against the selected microorganisms.
Attempts to Identify the Disulfide Pattern by Trypsin Cleavage and ESI-MS-The Pen-3a sequence of L. vannamei is characterized by a high content in glycine (17.5%), proline (15.9%), arginine (11.1%), serine (9.5%), and cysteine (9.5%). Together, these five amino acids account for 63.5% of the sequence. In contrast, Pen-3a does not contain either aspartic acid, or tryptophan, nor methionine. Class 3 sequences, which appear to be highly conserved within the shrimp species, significantly differ between species (ex: L. vannamei and L. setiferus) (13).
Concerning the location of the six cysteines, it should be noted that for all classes four cysteines belong to 2 doublets (Cys 47 , Cys 48 , and Cys 54 , Cys 55 for Pen-3a) and that the first two cysteines are separated by three residues (Pro 33 §NVal 34 -Ser 35 ) in class 3, whereas they are separated by only one for class 1 (Ala 26 ) or by two for class 2 (Asn 27 -Ala 28 ) and class 4 (Asp 24 -Val 25 ) (Fig. 1).
The arrangement of the three disulfide bonds in penaeidins was as yet unknown. The ESI-MS analysis of trypsin digests allowed us to identify fragments essentially resulting from cleavages after Lys 4 , Arg 9 , Arg 19 , Arg 37 , Arg 45 , Arg 50 , Arg 53 , and Lys 59 residues ( Fig. 1 and Table 1B (Fig. 1). These three-fragment peptides only differ by the presence of the Pro 10 -Arg 19 sequence. On the basis of the trypsin cleavage data, the Cys 32 -Cys 36 , the Cys 47 -Cys 48 , and the Cys 54 -Cys 55 disulfide bonds were excluded discarding seven arrangements among the 15 theoretically possible. Even using chymotrypsin, no more information could have been obtained from such an unusual disulfide pattern. In conclusion, the trypsin cleavage alone did not allow us to fully determine the disulfide bond pattern of Pen-3. Its determination has to be completed by modeling.
NMR Study-Preliminary NMR studies showed that Pen-3a gave a few broad resonances. The ESI-MS data showed that Pen-3a was contaminated by glycosylation on Thr 8 with polysaccharide chains from 3 to 15 sugar units. These polysaccharide units were also observed on the DQF-COSY experiment recorded in D 2 O, showing at least 12 typical signals of anomeric protons in the 4.8 -5.4 ppm range. Moreover, two spin systems of practically equal intensity were observed for the Tyr 7 , Thr 8 , and Arg 9 residues. For one spin system of Thr 8 , the NOE between the methyl group and the sugar protons (3.64 and 5.10 ppm) clearly indicated its glycosylation. Therefore, to avoid the glycosylation process, the [T8A]-Pen-3a analogue was expressed in yeast, purified to homogeneity, and used for the structural study.
Assignment-Two-dimensional spectra (TOCSY, DQF-COSY, and NOESY) of both Pen-3a and the [T8A]-Pen-3a analogue were recorded at several temperatures, ranging from 12 to 27°C. The similarity of their spectra indicated that the two peptides were similarly folded. The identification of all the spin systems of [T8A]-Pen-3a was obtained by analysis and comparison of DQF-COSY, TOCSY, and NOESY spectra according to the strat-egy described by Wü thrich (29). Two parts of the NOESY experiment, showing the dNN and the d␣(i)Ϫ␦␦ЈPro(iϩ1) dipolar correlations, are displayed in Fig. 2.
Due to the spreading of spin systems of the cysteine-rich domain, their assignment was straightforward (Fig. 2). In contrast, due to the unusual number of prolines (three are successive, Pro 14 -Pro 15 -Pro 16 ) and glycine residues along with several repetitive sequences (three Arg-Pro, (Arg 9 -Pro 10 , Arg 13 -Pro 14 , Arg 19 -Pro 20 ) and two Gly-Gly (Gly 5 -Gly 6 , Gly 23 -Gly 24 )), the assignment of the proline-rich domain was more difficult. However, the unambiguous assignment of the 10 prolines was obtained from the d␣(i)-␦␦ЈPro(iϩ1) dipolar correlations which were well resolved, due to the dispersion of the alpha proton resonances of the preceding residues (4.06 -5.08 ppm) (Fig. 2). When the TOCSY, DQF-COSY, and NOESY data recorded in H 2 O and D 2 O were combined, the full assignment of the peptide was obtained. Chemical shifts were deposited in the BMRB (accession number 5806).
To locate the elements of secondary structure, chemical shifts of the alpha protons were compared with statistical chemical shift values determined for random coil conformation to calculate the chemical shift indexes (CSI) (34,35). Indeed, successive positive and negative CSI are indicative of a ␤-stranded or a helical structure. The CSI are reported in Fig.  3. They indicate a helical structure for the Phe 41 -Arg 50 sequence and ␤-stranded structures for the Arg 13 -Pro 15 and Leu 51 -Arg 53 short sequences. The Val 34 , Ser 35 , Cys 36 , and Cys 55 negative indexes suggest that these residues belong to either a short helix or a turn, respectively. The largest positive chemical shift deviations measured for Ile 39 (19). Notice that penaeidins of class 3 significantly differ from the others, mainly by an insertion of 3-5 residues in the middle of the sequence and by its C-terminal extension. The numbering on the top is that of Pen-3a. The N-terminal proline-rich and the C-terminal cysteine-rich domains are boxed. Mutations are displayed in green and conserved cysteines in red. Lower part, to determine the disulfide pattern, a trypsin cleavage was carried out. The trypsin cleavage sites of Pen-3a are labeled by the numbers in italics. The three black lines represent the Pro 10 -Arg 37 /Ser 46 -Arg 50 /Cys 54 -Lys 59 heterotrimer cross-linked by three disulfide bonds (4133.07 Da) that was observed in the trypsin digests. The three gray lines are for a shorter peptide corresponding to the Arg 19 -Pro 20 cleavage (2976.33 Da). The disulfide pattern determined in the present study is displayed above the sequence.
were not taken into account in the preliminary calculations. These calculations revealed that the [T8A]-Pen-3a structure consists of an unconstrained and a constrained part, which roughly match with the proline-rich (1-28 residues) and the cysteine-rich (29 -58 residues) domains, respectively.
The global fold of the cysteine-rich domain is stabilized by three disulfide bonds whose arrangement has yet to be identified. As a result of trypsin cleavage experiment seven arrangements out of the 15 theoretically possible were withdrawn from consideration. From the global fold of the cysteine-rich domain, the eight arrangements still possible had to be taken into account by a statistical analysis of the mean distances between the beta carbons of cysteines. Therefore, an initial calculation of structures was carried out without any disulfide bond.
Such a distance analysis allowed us to unambiguously identify a first disulfide bond between Cys 48 and Cys 55 . However, the inter-cysteine d␣␤ NOEs, that usually characterize a disulfide bond, could not be observed due to the superimposition of the resonances of these protons. The mean distance between their C␤ was about 3.6 Å, whereas the distances with the other cysteine C␤ were in the range of 7.5-10.5 Å. Consequently, the two other disulfide bonds involve the four remaining cysteines (Cys 32 , Cys 36 , Cys 47 , and Cys 54 ) in three possible arrangements. In fact, in the preliminary calculations these four cys-teine C␤ occupied the corners of a square whose side is 4 -5 Å in length. Their location agrees well with NOEs observed between H␣ of Cys 32 and H␤ of Cys 47 , between H␣ of Cys 54 and H␤ of Cys 36 , and between H␤ of Cys 54 and H␤ of Cys 47 . Nevertheless, due to the packing, some NOEs must involve protons belonging to two disulfide bonds, thus making ambiguous the identification of the correct disulfide arrangement, which in this case remains a challenge.
To solve the disulfide arrangement, the three possible disulfide arrangements -Cys 32 -Cys 36 /Cys 47 -Cys 54 , Cys 32 -Cys 54 / Cys 36 -Cys 47 , and Cys 32 -Cys 47 /Cys 36 -Cys 54 -were taken into account in modeling and three calculations were carried out to evaluate their agreement with experimental data. As shown in Table III, for the Cys 32 -Cys 36 /Cys 47 -Cys 54 pattern, inconsistent with the trypsin cleavage; the Cys 54 side chain adopted an unfavorable rotamer (1 around 0°). In the structures obtained with the Cys 32 -Cys 54 /Cys 36 -Cys 47 forced pattern, the distance between the Cys 36 H␣ and Cys 32 H␤ protons and between the Cys 36 amide proton and the Cys 47 H␤ protons are too short (d␣␤ Յ 1.8 Å) and give rise to severe van der Waals contacts. These two strong contacts are inconsistent with the NOESY spectrum since no strong NOE between these protons was observed. Moreover, the resulting disulfide bonds interfere each other. All these inconsistencies as well as the NOEs vio-

FIG. 3. Distribution and summary of NOEs and CSI for [T8A]-Pen-3a.
Upper part, distribution of intraresidual (white), sequential (light gray), medium range (dark gray), and long range (black) constraints along the sequence. This plot clearly indicates that only the second half of the sequence, corresponding to the cysteine-rich domain, adopts a constrained structure. Middle part, only chemical shift deviations greater than 0.2 ppm were retained to plot the CSI of H␣ resonances as a function of the sequence. Random coil chemical shifts are those from Wishart & Case (35). Lower part, summary of the sequential, medium-range, and long range NOEs. The relative intensity of NOEs is represented by the thickness of the bars. Slowly exchanging amide protons are identified by filled squares. Solution Structure of Penaeidin-3 lations, allowed us to conclude that the likely disulfide bond pattern of [T8A]-Pen-3a is the third one (Cys 32 -Cys 47 and Cys 36 -Cys 54 ). Indeed, this arrangement, which is in agreement with the trypsin cleavage satisfies all the experimental NOEs since neither violation nor strong van der Waals contact was observed.
A final round of calculations including the Cys 32 -Cys 47 , Cys 36 -Cys 54 , and Cys 48 -Cys 55 disulfide bonds was carried out and the 20 lowest energy conformers out of the 60 refined structures were minimized and analyzed. The statistics for this ensemble of structures are summarized in Table IV. Since the proline-rich domain of [T8A]-Pen-3a was found to be unstructured, only the superimposition of conformers of the cysteinerich domain is shown in Fig. 4. This globular part is well defined and the pairwise mean rmsd value for the superimposition of the 30 -58 backbone atoms was 0.636 Ϯ 0.166 Å (0.533 Ϯ 0.154 Å for 31-57 backbone atoms). The distribution in the Ramachandran plot of all residues (except for the glycines and prolines) of the 20 best structures indicates their quality: 73.0% are located in the most favored regions, 24.1% in the additional allowed regions, and 1.5% in the generously allowed regions. 1.3% of the residues, mainly located in the unconstrained proline-rich domain, are located in the disallowed regions (32). Limits of secondary structure elements were determined by using STRIDE.
Finally, the main feature of the [T8A]-Pen-3a structure is that the proline-rich domain does not adopt a constrained structure, whereas the cysteine-rich domain adopts a welldefined structure that consists of a helix (41-50 residues) tightly bound to the upstream (27-40 residues) and the downstream (51-63 residues) coils by Cys 32 -Cys 47 and Cys 48 -Cys 55 disulfide bonds. In turn, these two coils are linked together by the third, Cys 36 -Cys 54 , disulfide bond (Fig. 4). The structure of the Pro 33 -Cys 36 and Cys 54 -Val 57 sequences consists of beta turn structures of type I and IV, respectively. Such a disulfide bond arrangement, which tightly links the three segments gives rise to a globular shape for the cysteine-rich domain (20 and 23 Å for its diameter and height, respectively). This is supported by the slow H-D exchange measured for eight amide protons involved in hydrogen bonds. Four hydrogen bonds between Ala 44 , Arg 45 , Cys 47 , and Cys 48 amide protons and Ser 40 , Phe 41 , Gln 43 , and Ala 44 carbonyl groups, respectively, are located in the helical part. Four other hydrogen bonds are located in the loop following the helix and involve Arg 50 , Leu 51 , Gly 52 , and Arg 53 amide protons and Ser 46 , Cys 47 , Cys 48 , and Cys 47 carbonyl groups, respectively.
Conformation of the Disulfide Bridges and of the X-Pro Amide Bonds-It is worthy to note that two of the three disulfide bridges are buried in the core of the molecule and significantly contribute to its hydrophobicity. In contrast, the third one (Cys 48 -Cys 55 ) is exposed to the solvent. Concerning their geometry, the statistical analysis shows that the Cys 32 -Cys 47 , Cys 36 -Cys 54 , and Cys 48 -Cys 55 disulfide bridges have 3 angle values of Ϫ105.6 Ϯ 15.0°, Ϫ110.31 Ϯ 12.5°, and 109.3 Ϯ 3.8°, respectively. These values are those measured for classical conformations of disulfide bridges ranging in the range of Ϯ 90 Ϯ 45° (36). Concerning the geometry of the ten X-Pro amide bonds (X for any residue), the observation of the typical d␣(i-1)-␦␦ЈPro(i) NOEs was indicative of a trans conformation for all of them (Fig. 2). DISCUSSION As a general feature, antimicrobial peptides are positively charged and most of them display an amphipathic character. According to their sequence or their three-dimensional structures, antimicrobial peptides are divided into four classes: the proline-rich, the glycine-rich, those which adopt a helical structure, and those which have disulfide bonds. Penaeidin sequences contain all these features. A sequence identity search indicated that neither the proline-rich nor the cysteine-rich domains of Pen-3a share significant identity with known antimicrobial proteins. Nevertheless, some low and limited similarities of the proline-rich sequence (Pro 12 -Arg 13 -Pro 14 -Pro 15 ) with apidaecin, drosocin, pyrrhocoricin, metalkowin and ( 9 RPIPRP 14 ) with the bactenecin-7 antimicrobial peptides have been reported (10). Concerning the cysteine-rich domain, some limited sequence similarities were observed with the chitin binding motif ( 47 CCS 49 sequence) and tachycitin (10,37).
The presence of these features in the same molecule might explain the multifonctional properties of the penaeidins and their broad spectrum of activity. Indeed, penaeidins display antimicrobial activity against Gram-positive bacteria, either  Solution Structure of Penaeidin-3 bactericidal or bacteriostatic depending on the strains, and a fungicidal effect activity against filamentous fungi (15). Additionally, a chitin-binding property has been experimentally evidenced for the peptide (16) as well as a capability to bind to Gram-negative bacteria of the genus Vibrio (38). Thus, to understand the function of penaeidins, the elucidation of their three-dimensional structure and organization of the two domains appeared to be of prime importance.
Since the amount of the native Pen-3a needed for a structural study was difficult to purify from shrimp, we expressed it in the yeast S. cerevisiae.
Whereas the native peptides purified from shrimp hemocytes are not post-translationally modified, the yeast-expressed Pen-3a appeared to be partially O-glycosylated on Thr 8 . Despite an intensive purification procedure to eliminate the glycosylated forms, traces of residual modified penaeidins rendered difficult the NMR study. To circumvent this O-glycosylation the [T8A]-Pen-3a analogue was expressed in yeast, purified, and used instead. No differences were seen between the antimicrobial activity of the analogue and its parent peptide. Their 1 H NMR spectra also appeared to be similar, thus suggesting an identical three-dimensional structure. Consequently, the structural study of the [T8A]-Pen-3a analogue was carried out by NMR to determine both its three-dimensional structure and the pattern of the three disulfide bonds. This structural study allowed us to gain insight into the amphipathic character of penaeidins and to initiate a structureactivity relationship analysis for this family of antimicrobial peptides.
Disulfide Bonds and Three-dimensional Structure-Because of the two cysteine doublets, the determination of the full disulfide bond arrangement by the enzymatic cleavage alone was destined to fail. Nevertheless, given the numerous trypsin cleavage sites in the molecule, the absence of fragments including the potential Cys 32 -Cys 36 , Cys 47 -Cys 48 , and Cys 54 -Cys 55 disulfide bonds suggested that these three disulfide bonds were not present in Pen-3a. The absence of disulfide bonds between two adjacent cysteines (Cys 47 -Cys 48 and Cys 54 -Cys 55 ) is not surprising since such a disulfide bond is known to be energetically unfavorable, although such a bond formation has been reported for a protein, the carboxypeptidase T (PDB code: 1obr) (39), and recently for the hepcidin peptide (PDB code: 1m4e.pdb and 1m4f.pdb), an iron-regulatory hormone which also displays antimicrobial activity (40). In contrast, the mass of two fragments from Pen-3a corresponding to a three fragment-peptide cross-linked by the three disulfide bonds was measured. Although these trypsin cleavage data did not solve the arrangement of the three disulfide bonds, they allowed us to reject seven out of the 15 disulfide bond patterns theoretically possible. For these three packed disulfide bonds, the arrangement proposed by the modeling step is Cys 32 -Cys 47 , Cys 36 -Cys 54 , and Cys 48 -Cys 55 . One disulfide bond (Cys 48 -Cys 55 ) is solventexposed, whereas the two others belong to the hydrophobic core of the cysteine-rich domain.
The established three-dimensional structure together with the sequence allowed us to more rationally define the limits of the proline-rich and cysteine-rich domains. The absence of NOEs between the Pro 28 , Gly 58 , and Lys 59 side chains with the core of the cysteine-rich domain and the presence of NOEs between side chains of Tyr 29 and Leu 51 and between Val 57 and Cys 36 were indicative of the fact that Tyr 29 and Val 57 belong to the cysteine-rich domain whereas Pro 28 and Gly 58 do not. Therefore the cysteine-rich domain spans residues Tyr 29 -Val 57 . This is clearly shown in Fig. 4. Therefore, concerning the proline-rich domain, which is mainly unconstrained, it spans the N-terminal part up to Pro 28 . Nevertheless, it should be noted that no proline is located in the first nine residues and that the two last prolines (Pro 25 and Pro 28 ) belong to an additional sequence that is absent in the three other classes of penaeidins (Pro 25 is present in Pen-4a). Moreover, this additional sequence contains three glycine residues (Gly 23 , Gly 24 , and Gly 27 ), which probably provide this sequence with high flexibility, which therefore could act as a hinge between the two domains.
The [T8A]-Pen-3 structure, as determined from the NMRderived constraints, is characterized by a typically unstructured part corresponding to the proline-rich domain (1-28 residues), which contrasts with the well-defined structure adopted by the cysteine-rich domain spanning 29 -57 residues. The global fold of this latter domain consists of a helix inserted between an upstream and a downstream coil antiparallel to the helix, all tightly bound together by the disulfide bonds. ing was proposed (15). The presence of the proline-and glycinerich features and the cysteine-rich feature gathered in penaeidins in unconstrained and highly constrained domains, respectively, suggests that each domain alone does not exhibit an efficient antimicrobial effect to be selected by evolution. Therefore, we hypothesize a complementary effect of the two domains, the proline-rich domain bearing the membrane-anchoring feature and the cysteine-rich domain responsible for the antimicrobial activity. Further studies are necessary to address the question of the respective involvement of the two domains in the delivery, targeting, or binding to membranes, and in the membrane permeabilization or peptide internalization character, leading to the death of microorganisms.