Semienzymatic Cyclization of Disulfide-rich Peptides Using Sortase A*

Background: Sortase A (SrtA) is a transpeptidase capable of catalyzing the formation of amide bonds. Results: SrtA was used to backbone-cyclize disulfide-rich peptides, including kalata B1, α-conotoxin Vc1.1, and SFTI-1. Conclusion: SrtA-mediated cyclization is applicable to small disulfide-rich peptides. Significance: SrtA-mediated cyclization is an alternative to native chemical ligation for the cyclization of small peptides of therapeutic interest. Disulfide-rich cyclic peptides have generated great interest in the development of peptide-based therapeutics due to their exceptional stability toward chemical, enzymatic, or thermal attack. In particular, they have been used as scaffolds onto which bioactive epitopes can be grafted to take advantage of the favorable biophysical properties of disulfide-rich cyclic peptides. To date, the most commonly used method for the head-to-tail cyclization of peptides has been native chemical ligation. In recent years, however, enzyme-mediated cyclization has become a promising new technology due to its efficiency, safety, and cost-effectiveness. Sortase A (SrtA) is a bacterial enzyme with transpeptidase activity. It recognizes a C-terminal penta-amino acid motif, LPXTG, and cleaves the amide bond between Thr and Gly to form a thioacyl-linked intermediate. This intermediate undergoes nucleophilic attack by an N-terminal poly-Gly sequence to form an amide bond between the Thr and N-terminal Gly. Here, we demonstrate that sortase A can successfully be used to cyclize a variety of small disulfide-rich peptides, including the cyclotide kalata B1, α-conotoxin Vc1.1, and sunflower trypsin inhibitor 1. These peptides range in size from 14 to 29 amino acids and contain three, two, or one disulfide bond, respectively, within their head-to-tail cyclic backbones. Our findings provide proof of concept for the potential broad applicability of enzymatic cyclization of disulfide-rich peptides with therapeutic potential.

Disulfide-rich cyclic peptides are generating great interest in the field of drug design because of their remarkable stability toward enzymatic, thermal, or chemical attack (1). Additionally, anecdotal evidence of some of them being orally bioavailable in indigenous medicine applications (2) has recently received support from studies demonstrating oral activity of engineered cyclic peptides in animal pain models (3,4). Several classes of cyclic peptides, including the sunflower trypsin inhibitors and cyclotides, have now been discovered in a variety of plant species (5). In addition to the plant-derived cyclic peptides, disulfide-rich peptides isolated from cone snails, including ␣-conotoxins, have successfully been cyclized (3,6) and are showing promise as potential treatments for neuropathic pain (3).
Sunflower trypsin inhibitor 1 (SFTI-1) 5 comprises 14 amino acids with a head-to-tail cyclized backbone. It was isolated from sunflower seeds and specifically inhibits trypsin activity at 0.1 nM (7), which makes it the smallest and most potent serine proteinase inhibitor known. It was also reported to have inhibition activity against the epithelial serine protease matriptase at subnanomolar concentration (8). Its structure consists of two antiparallel ␤-strands connected by an extended loop and a sharp hairpin turn containing a cis-proline residue. The ␤-strands are constrained by a single disulfide bond, which stabilizes the molecule and divides it into two regions: the active site loop and the cyclization loop (7) (Fig. 1A).
Cyclotides are characterized by the combination of head-totail backbone cyclization and six conserved Cys residues forming a distinctive disulfide linkage pattern in which one disulfide bond passes through a ring formed by two other disulfide bonds (9,10) (Fig. 1B). The combination of a cyclic backbone and a knotted arrangement of three disulfide bonds is known as the cyclic cystine knot motif, and this motif is believed to be responsible for the exceptional stability of cyclotides toward chemical, enzymatic, or thermal degradation (1). Although thought to have a natural function as defense molecules in plants via their pesticidal activities (11)(12)(13), cyclotides possess a range of pharmaceutically interesting biological activities, including uterotonic (2), anti-HIV (14), antimicrobial (15), and insecticidal (11,16) activities, and are cytotoxic to tumor cells (17,18). Thousands of cyclotides are thought to exist in nature, and more than 250 sequences have been described so far (see the CyBase Web site) (19), making them one of the most abundant plant protein families discovered.
Cyclotides are able to accommodate the synthetic introduction of a range of biologically active sequence motifs in the backbone loops between the conserved Cys residues while retaining thermal, chemical, and enzymatic stability. This tolerance to sequence substitutions makes them ideal scaffolds for the development of stable peptide-based drugs incorporating epitopes of therapeutic value. For example, a kB1 hybrid, with an anti-angiogenic epitope incorporated into loop 3, combined anti-angiogenic activity with the native stability of kB1 (20). Likewise, molecules possessing activity against foot-andmouth disease (21), inhibitory activity against ␤-tryptase and human leukocyte elastase (22), angiogenic activity (23), and the ability to antagonize intracellular p53 degradation (24) have also been achieved by grafting relevant bioactive epitopes into the cyclotide MCoTI-I or MCoTI-II. The capacity to target intracellular sites is consistent with a range of recent biophysical studies demonstrating the ability of disulfide-rich cyclic peptides to internalize into cells (24 -28).
As well as being valuable scaffolds, naturally occurring cyclic peptides, such as cyclotides and SFTI-1, have inspired the concept of using artificial head-to-tail backbone cyclization as a stabilizing strategy for peptides and proteins of therapeutic interest (3,6). For example, Clark et al. (3) showed that cyclization of ␣-conotoxin Vc1.1, a potential neuropathic pain treatment, endowed it with oral activity. Thus, great interest has been shown in the development of cost-effective and efficient methods for the N-to C-terminal backbone cyclization of synthesized or expressed proteins (29,30). In the case of cyclic Vc1.1 (cVc1.1), the termini of the naturally occurring 16-residue peptide were covalently linked using a 6-amino acid linker, as illustrated in Fig. 1C (3).
Currently, the most widely used strategy for backbone cyclization of synthetic peptides is native chemical ligation (NCL) (31,32), which utilizes an N-terminal cysteine and a functionalized C terminus to form a thioester-linked intermediate that undergoes an S,N-acyl migration to form a native peptide bond (32). In general, peptides are synthesized using solid phase peptide synthesis, using either t-butoxycarbonyl or Fmoc protection. Despite the usefulness of NCL for splicing or cyclizing protein, there is still a demand for alternative methods of amide bond formation that are chemoselective, rapid, cheap, safe, and waste-free (30). Moreover, the requirement for an N-terminal cysteine and a C-terminal thioester in NCL is not always compatible with Fmoc solid phase peptide synthesis or recombinant expression. Accordingly, a number of different approaches to backbone cyclization have been developed (29), including expressed protein ligation (33,34), intein-mediated protein trans-splicing (35), and genetic code reprogramming (36).
Recently, a family of thiol-containing transpeptidases, known as the sortases, were found to successfully cyclize linear proteins (37)(38)(39)(40). Sortases are found in most Gram-positive bacteria, where their primary function is to attach surface proteins to the bacterial cell wall (41). This ligation reaction has been exploited for a variety of protein engineering purposes, in most cases utilizing sortase A (SrtA) from Staphylococcus aureus (42). SrtA recognizes a 5-residue sequence motif, LPXTG, and cleaves the peptide bond between the Thr and Gly residues, forming an acyl enzyme intermediate between a cysteine at the active site of SrtA and the carboxylate at the truncated C terminus of the substrate (43,44). Although in a natural system, this covalent intermediate is resolved by nucleophilic attack from a pentaglycine side chain in a peptidoglycan precursor, a variety of glycine-based nucleophiles can activate SrtA-catalyzed transpeptidation (38). The result is the efficient attachment of various moieties to the protein substrates, and this process has been applied to site-specific protein labeling (45), PEGylation (46), protein thioester generation (47), and protein-protein fusion (48). When such glycine-based nucleophiles originate from the N terminus of the substrate, intramolecular transpeptidation reactions occur to yield covalently closed (circular) polypeptides by amide bond formation between the two termini (38 -40). So far, this approach has been examined only for a limited number of possible substrates and has not been explored for disulfide-rich peptides.
In the current study, we demonstrate the efficient enzymatic cyclization of disulfide-rich peptides, including SFTI-1, cVc1.1, and kalata B1 (kB1), containing one, two, or three disulfide bonds, respectively, by SrtA without the need for a thioester linker and using Fmoc chemistry. This study provides a proof of concept that SrtA-mediated ligation can be utilized to cyclize disulfide-rich peptides ranging from the simple one-disulfidecontaining SFTI-1 to the complex cyclic cystine knot-containing cyclotide structure of kB1 while retaining the native disulfide bond connectivity that is vital for the stability of these proteins. Fig. 1D summarizes the cyclization strategy for kB1. -Cys IV ) formation. Cleavage of the peptide from the resin was achieved using trifluoroacetic acid (TFA)/triisopropylsilane/water (95:2.5:2.5) at room temperature for 2 h. The TFA was evaporated, and the peptide was precipitated with ice-cold diethyl ether and dissolved in 50% acetonitrile (ACN) containing 0.05% TFA and lyophilized. Wild type Vc1.1 and SFTI-1 were prepared as described previously (3,49).

EXPERIMENTAL PROCEDURES
SrtA Expression-The plasmid of and recombinant expression protocol for S. aureus SrtA, which comprises the catalytic domain (residues 60 -206 with an N-terminal hexahistidine tag), was as described by Popp et al. (50). Briefly, an overnight culture of Escherichia coli BL21 (DE3) transformed with SrtA plasmid was cultured in the presence of kanamycin (50 g/ml) until A 600 reached ϳ0.7. The protein expression was induced by adding isopropyl ␤-D-thiogalactopyranoside to a final concentration of 1 mM for 3 h at 37°C. The cells were then harvested by centrifugation at 7000 ϫ g at 4°C for 10 min. Cells were then resuspended in ice-cold lysis buffer (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 25 mM imidazole, and 10% glycerol) and lysed by passing through a cell disrupter (TS Series Bench Top) operating at 26,000 p.s.i. The cell debris was removed by centrifugation at 12,000 ϫ g at 4°C for 30 min. The lysate was subjected to immobilized metal ion affinity chromatography using a 5-ml nickel-nitrilotriacetic acid FF column (GE Healthcare). The column was washed extensively with lysis buffer after sample loading, and SrtA was eluted with a linear gradient over 10 column volumes using elution buffer (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, and 500 mM immidazole). Fractions containing SrtA were buffer-exchanged (50 mM Tris-HCl, 150 mM NaCl, and 10% glycerol) and concentrated by using a centrifugal concentrator (Ͻ10 kDa). The SrtA was then stored at Ϫ80°C until further use. The protein concentration was calculated by its extinction coefficient at 280 nm.
Oxidative Folding-The linear precursor [GGG]kB1[TGG] was oxidized in 50% isopropyl alcohol (v/v), 0.1 M NH 4 HCO 3 with 1 mM reduced glutathione at pH 8 for 20 h at room temperature (51). The progress of oxidation was monitored by analytical reversed-phase high performance liquid chromatography (RP-HPLC) with a gradient from 0 to 50% ACN (0.05% TFA) in 50 min at a flow rate of 0.3 ml/min using an analytical C18 column (Agilent ZORBAX 300SB-C18, 5 m, 2.1 ϫ 150 mm). The oxidation yield was calculated based on HPLC profile. The mixture was then purified by RP-HPLC using a 0.5% gradient, and the molecular weight was confirmed by mass spectrometry. Native kB1 was isolated from Oldenlandia affinis as described previously (9,52 and cyclic) with one disulfide bond was then dissolved in 50% acetic acid to a final concentration of 0.5 mg/ml under N 2 . Excess iodine dissolved (10 eq with respect to the Acm-protected Cys) in 50% acetic acid was added until the color of the reaction became yellow, and the reaction was left at room temperature for 24 h. The reaction was quenched by the addition of ascorbic acid until the solution became colorless, followed by RP-HPLC purification using a gradient of 0 -60% acetonitrile in 60 min.
Cyclization-The concentration of modified kB1 was determined by its molar extinction coefficient at 280 nm using NanoDrop2000c (Thermo Scientific). Because there were no aromatic residues present, the concentration of SFTI-1 was determined at 214 nm using the NanoDrop2000c (Thermo Scientific) UV-visible function with the extinction coefficient calculated based on the equation, M SrtA were incubated in the reaction buffer at 37°C with pH adjusted to 8.5 to form a disulfide bond at the same time. The progress of cyclization was monitored using analytical RP-HPLC with a gradient from 0 to 50% ACN (0.05% TFA) in 50 min at a flow rate of 1 ml/min using an analytical C18 column (GraceSmart TM RP18, 5 m, 4.6 ϫ 150 mm). Nuclear to a final concentration of 1 mM and pH 6.1 (54). Cyclo- to a final concentration of ϳ1 mM and pH 3.1. All two-dimensional nuclear magnetic resonance (NMR) spectra were acquired at 298 K. Spectra recorded for assignments and H␣ secondary chemical shift analysis included TOCSY with an 80-ms mixing time and NOESY with a 200-ms mixing time. All spectra were acquired on a Bruker Avance 600-MHz NMR spectrometer equipped with a cryogenically cooled probe (Bruker, Karlsruhe, Germany). Water suppression for two-dimensional TOCSY and two-dimensional NOESY spectra were achieved by using excitation sculpting (55). Spectra were recorded with 2048 data points in the direct F 2 dimension and 600 increments in the indirect F 1 dimension. Additional spectra recorded for cyclo-[GGG]kB1[T] to be used in full structure calculations were NOESY with a 100-ms mixing time as well as E.COSY (56), DQF-COSY, and 13 C HSQC to produce the carbon chemical shifts. All spectra were processed using Topspin (Bruker) and analyzed by CCPNMR (57) and Xeasy (58). The temperature coefficients were determined from recording TOCSY spectra from 283 to 303 K in steps of 10 K. Chemical shifts were referenced to internal 2,2-dimethyl-2-silapentane-5-sulfonate (DSS) at 0.0 ppm (59). The peaks in the NOESY spectra with mixing times of 100 and 200 ms were manually picked, intraresidual and sequential NOEs were assigned, and a full list of interproton distances was generated using the AUTO function in CYANA (tolerances used for CYANA were set to 0.02 ppm for both indirect and direct 1 H dimension) (60). Several rounds of AUTO, including additional restraints, such as disulfide bonds, hydrogen bonds, and dihedral angle restraints derived from TALOSϩ, were run to ensure correct peak assignment. H␣, HN, C␣, and C␤ chemical shifts derived from 1 H-1 H NOESY and 1 H-13 C HSQC were used to generate and backbone dihedral angles using TALOSϩ (61). The ANNEAL function in CYANA was used to perform 10,000-step torsion angle dynamics to generate an ensemble from which 20 with the lowest penalty function were chosen for further analysis. Several rounds of ANNEAL were run to resolve distance and dihedral constraint violations. Using protocols from the RECOORD database (62), 50 structures were calculated using CNS (63) with force fields distributed with Haddock version 2.0 (64) and refined in a water shell (65) as described previously (66). A set of 20 structures with no NOE violations of greater than 0.3 Å and dihedral violations greater than 3.0º was chosen for MolProbity analysis (67). MOLMOL (68) was used for generating figures and final superimposition.
Hemolytic Activity Assay on [GGG]kB1[T]-The hemolytic assay was conducted as described previously (69). Briefly, eryth-rocytes were isolated from human blood and washed in phosphate-buffered saline (PBS, pH 7.4) with repeated centrifugation at 1500 ϫ g. The hemolytic activities of cyclo- [TGG], and kB1 were determined with eight concentration points from 300 to 1.17 M by serial dilution with a 2-fold interval with a final volume of 20 l/well in a 96-well plate. Triton X-100 solution (1% (v/v), 20 l) was used as a positive control to achieve the 100% lysis, and PBS solution (20 l) was used as a negative control. 100 l of erythrocyte stock solution was added to each well and incubated at 37°C for 1 h. After centrifugation of intact cells from the 96-well plate, the supernatant of each well was measured by visual absorption spectroscopy at 415 nm. Each experiment was performed in triplicate.

Serum Stability Assay on [GGG]kB1[T]-
The stability of cyclo-[GGG]kB1[T] in human serum was evaluated using a method described previously (23). Briefly, human serum from male AB plasma (Sigma-Aldrich) was centrifuged at 17,000 ϫ g for 10 min to remove the lipid component. The resultant supernatant was collected and incubated for 15 min at 37°C before the assay. , GGG-RCTKSIPPICFPDLPETGG (with the native sequences in boldface type)), were assembled using automated Fmoc chemistry, cleaved from the resin with TFA, and purified using RP-HPLC. No oxidation of cysteine residues was observed during the purification process, as determined using mass spectrometry. The molecular mass and purity were confirmed using mass spectrometry and analytical RP-HPLC, respectively.
Oxidation and Cyclization of [GGG]kB1[TGG]-In preliminary experiments two approaches were used to obtain oxidized, cyclic kB1 using SrtA-mediated cyclization. First, linear reduced [GGG]kB1[TGG] was cyclized using the SrtA reaction buffer but with the addition of a 2 mM concentration of the reducing agent TCEP. The peptide was then oxidized after completion of the cyclization reaction. Second, the linear peptide [GGG]kB1[TGG] was oxidized prior to cyclization with SrtA, which proceeded without the addition of reducing agent to the reaction buffer. Cyclization prior to oxidation resulted in higher yields of cyclic reduced peptide as determined by RP-HPLC (ϳ80%; data not shown), but subsequent oxidative refolding yielded several isomers (Ͻ10% yield for each isomer; data not shown), presumably due to misfolding during the oxidation reaction. Oxidation prior to cyclization resulted in a predominance of correctly folded peptide, and this approach was therefore used for subsequent experiments.
Theoxidativefoldingofthelinearprecursor[GGG]kB1[TGG]was monitored by RP-HPLC ( Fig. 2A), and the yield of correctly folded product was ϳ27% based on RP-HPLC peak integration. Following oxidative folding, linear [GGG]kB1[TGG] was puri-fied by RP-HPLC, and the correct folding was confirmed by NMR analysis (data not shown).
Cyclization of 150 M linear oxidized [GGG]kB1[TGG] with 50 M SrtA was monitored over 24 h using RP-HPLC (Fig. 2B). The reaction was not extended beyond 24 h due to the possibility of a competing hydrolysis reaction in which SrtA irreversibly hydrolyzes the LPVTG motif (70). After 24 h, ϳ49% of linear, oxidized [GGG]kB1[TGG] was converted to the cyclic product (highlighted in Fig. 2B). MALDI-TOF analysis of the highlighted peak fraction showed a monoisotopic mass (3162.92 Da) corresponding to that of cyclo-[GGG]kB1[T] (Fig. 2C), which was purified to Ͼ95% purity. Interestingly, the elution time of this peak (40.8 min) was similar to that of native kB1 (Fig. 2D) despite the additional amino acids in the sequence. Apart from this peak, two other peaks, at ϳ37 and 40 min, appeared to be intermediate species in the cyclization reaction because their abundance over time was negatively correlated with the abundance of correctly folded cyclo-[GGG]kB1[T] (Fig. 2B). A small peak appearing at ϳ42.8 min contained a peptide with the mass corresponding to that of cyclo-[GGG]kB1[T]. This peak also increased during the course of the experiment and is most probably an alternatively folded form of the peptide (Fig. 2B).

Characterization of the Cyclo-[GGG]kB1[T] by NMR-The NMR signals of cyclo-[GGG]kB1[T]
were well dispersed in the amide proton region, indicating that the peptide is folded and has a well defined structure. The individual amino acid spin systems were readily assigned using the sequential assignment procedure based on TOCSY and NOESY spectra (71). Cyclization was confirmed based on observation of a sequential ␣H i -HN i ϩ 1 NOE between Thr 30 and Gly 31 , indicative of the amide linkage of the C-terminal LPVT and N-terminal GGG motif (Fig. 3, supplemental Fig. 3).
Chemical shifts are extremely sensitive to the local chemical environment, and H␣ chemical shifts were thus used to compare the structure of cyclo-[GGG]kB1[T] with that of native kB1. Overall, the chemical shift differences between the two molecules are minimal (Fig. 3). With the exception of residues 1 and 27-33 (numbering system in Fig. 3), which are in loop 6 and either very close to or part of the SrtA sorting motif, no other cyclo-[GGG]kB1[T] residue exhibited significant deviations from the corresponding residue in native kB1. The very small shifts from random coil values for residues 1 and 27-33 of cyclo-[GGG]kB1 [T] indicate that this region is disordered, and this was confirmed in the three-dimensional structure of cyclo-[GGG]kB1[T] (Fig. 4A). The solution structure was calculated using CYANA (60) followed by refinement in a water shell using CNS (63) based on 388 NOE distance restraints; 53 dihedral angle restraints, including 26 , 20 , and 6 1 dihedral angle restraints; and five hydrogen bonds derived from temperature coefficient data ( Table 1). The 20 conformers with the lowest energy superimposed with a backbone root mean square deviation of 0.53 Ϯ 0.14 Å across resides 1-24 (Fig. 4A). Comparison of the cyclo-[GGG]kB1[T] structure with that of native kB1 (PDB code 1NB1) shows them to be very similar except for the more flexible loop 6 of cyclo-[GGG]kB1[T] (Fig. 4B).
Hemolytic Activity and Serum Stability Assay on Cyclo-[GGG]kB1[T]-Native kB1 possesses mild hemolytic activity (15,51), which is undesired in a pharmaceutical context. Thus, to assess the effects of SrtA cyclization on the novel kB1 analog, the hemolytic activity of cyclo-[GGG]kB1[T] and linear oxidized [GGG]kB1[TGG] was determined and compared with that of the native peptide. Compared with native kB1, both peptides exhibited significantly less hemolytic activity (Fig. 5A). After 24 h of incubation, 50 M native kB1 lysed 64% of human erythrocytes, whereas in the same time period, only 6 or 2% of erythrocytes were lysed by the same concentration of cyclo-[GGG]kB1[T] or linear oxidized [GGG]kB1[TGG], respectively. The reduced hemolytic activity is consistent with previous mutagenesis studies on kB1, which showed that linearization or strategic replacement of certain residues can abrogate hemolytic activity (72,73).
To determine the effects of SrtA-mediated cyclization on the stability of the peptide, cyclo-[GGG]kB1[T] was incubated in human serum. Encouragingly, its stability was comparable with that of native kB1, with only a slight reduction in peptide survival observed. After a 24-h incubation, 77.4 and 95.9% of the starting concentration of cyclo-[GGG]kB1[T] and kB1, respectively, was still present (Fig. 5B). Interestingly, 71% of linear oxidized [GGG]kB1[TGG] also remained in human serum after 24 h. In contrast, a linear 12-mer control peptide, without any disulfide bond, was completely degraded within the first 1 h.

Oxidation and Cyclization of [G]Vc1.1[GLPETGGS]
-␣-Conotoxin Vc1.1 was successfully synthesized with a SrtA sorting motif at the C terminus and an additional Gly residue at the N terminus. Acm protecting groups on Cys 4 and Cys 17 were used for selective disulfide bond formation of Cys II -Cys IV (Fig. 6A). As for cyclo-[GGG]kB1[T], two approaches, namely oxidation prior to cyclization and cyclization prior to oxidation, were attempted to obtain the cyclo-[G]Vc1.1 [GLPET]. The former approach gave rise to only a low yield (ϳ1%) of oxidized product following Acm removal. In contrast, isomerization during oxidative folding was averted by cyclizing the linear precursor  (Fig. 6B). The molecular mass and purity of the desired products for each step were confirmed by MALDI-TOF and RP-HPLC (Fig. 6B). SrtA Reaction for SFTI-1-For this single disulfide-bonded peptide, the SrtA reaction was performed at pH 8.5 to achieve cyclization and oxidation in a one-pot reaction. After 9 h, ϳ70% of the linear precursor had been converted to the cyclic oxi- dized form based on the RP-HPLC profile (peak at 20 min; Fig.  6C). The mass of cyclic oxidized [GG]SFTI-1[LPET] was confirmed by MALDI-TOF (Fig. 6C). After a 24-h incubation, the peak at 20 min decreased, and a peak at 22.8 min increased in height and was found to contain both cyclic oxidized and hydrolyzed linear [GG]SFTI-1 [LPET].

DISCUSSION
Disulfide-rich cyclic peptides offer an attractive framework for the development of stable bioactive peptides with pharmaceutical potential. They display a remarkable degree of functional plasticity and high stability. Accordingly, interest has been shown both in the cyclization of linear disulfide-rich peptides (especially those isolated from venoms) and the development of native cyclotides as therapeutics and in their use as scaffolds for the grafting of pharmaceutically relevant epitopes onto the native cyclotide framework (74,75). A bottleneck in the development of these peptides as potential therapeutics, however, is the production of linear precursors and their subsequent cyclization and folding. NCL (using t-butoxycarbonyl chemistry) has been the preferred strategy for the production of disulfide-rich cyclic peptides, but this approach requires reaction conditions that cannot always be achieved using Fmoc chemistry or recombinantly expressed proteins. We therefore investigated SrtA-mediated cyclization as a mechanism for the increased efficiency of one of the steps in cyclic peptide production, namely head-to-tail cyclization of synthetically pro-duced peptides. Specifically, we explored the possibility of using SrtA to enzymatically cyclize SFTI-1, Vc1.1 (with a linker), and kB1 and showed that SrtA offers an efficient, safe, and cost-effective method for the cyclization of disulfide-rich peptides ranging in size from 14 to 29 amino acids with 1-3 disulfide bonds.
The presence of a native Leu-Pro-Val in loop 6 of kB1 made this peptide an ideal model for cyclization using SrtA. These three residues fortuitously form a part of the penta-amino acid sorting motif, LPXTG, and cyclization was achieved with minimal disruption to the native sequence of kB1. In NCL-based synthesis of kB1, the cyclization and oxidation is carried out in a one-pot fashion. However, the reaction requires a two-step procedure when using SrtA for cyclization, with oxidation occurring either after or before SrtA-mediated cyclization. In our hands, the order of the reaction (i.e. oxidation/cyclization versus cyclization/oxidation) had a significant effect on the yield of correctly folded peptide. Despite the minimal changes introduced in kB1 by SrtA cyclization, no correctly folded peptide was produced when oxidation occurred after cyclization. Small changes in the amino acid composition of cyclotides can have profound effects on the refolding of synthetic forms (6,20), and it is possible that the introduction of the TGGG amino acid sequence in loop 6 of kB1 had a negative effect on the oxidative refolding of the cyclic form. The proximity of the cyclization point to Cys I (Fig. 1) might also have adversely affected oxidative folding. The only region of significant structural perturbation in cyclo-[GGG]kB1[T] was for the Leu-Pro-Val residues immediately preceding Cys I , and, when combined with restraints imposed by cyclization, conformational variations in this region of the reduced, cyclized peptide might have prevented correct disulfide formation.
Production of correctly folded cyclo-[GGG]kB1[T] was achieved by oxidative refolding prior to cyclization with SrtA. Monitoring of the cyclization reaction showed what appeared to be intermediates at various stages of the reaction (Fig. 2B), but by 24 h, the reaction had efficiently converted the linear peptides to the cyclic form in ϳ49% yield.
Several pieces of NMR evidence suggest that cyclo-[GGG]kB1[T] adopts the fold of native kB1: (i) Pro 20 adopts a cis-conformation as in native kB1; (ii) The HN of Trp 19 is not observed in TOCSY spectra but is observed in NOESY spectra, as characteristically applies to kB1; (iii) The HB of Pro 20 is shifted upfield to Ϫ0.26 ppm; (iv) the protonation/deprotonation of Glu 3 significantly affects the chemical shifts of HN of Asn 11 and Thr 12 ; (v) the 1 dihedral angles of all six cysteine residues agree with those for kB1, providing strong evidence for an identical disulfide connectivity. Consistent with this connectivity, NOEs between Cys 1 H␤2/H␤3 to Cys 15 H␤3 and Cys 10 H␤2/H␤3 to Cys 22 H␤2 were observed (NOEs between Cys 5 H␤ and Cys 17 H␤ could not be unambiguously identified due to extensive spectral overlap). Interestingly, NOEs between H␤ protons of non-connected cysteine residues, including Cys 10 -Cys 15 , Cys 1 -Cys 22 , and Cys 15 -Cys 22 , were also observed, but this is in complete accordance with a study on native kB1 by Rosengren et al. (54), where it was suggested that for cyclic cystine knot peptides, H␤ cross-peaks are not the preferred diagnostic tool for determining disulfide connectivity due to the close proximity of all disulfide bonds. From the extensive evidence noted above, especially the fact that the two peptides share identical 1 dihedral angles for all of the Cys residues, we were confident that we produced cyclo-[GGG]kB1[T] with the native disulfide connectivity. Thus, we calculated the structure of cyclo-[GGG]kB1[T] using the cysteine connectivities as native kB1 (PDB code 1NB1).
The hemolytic activity of cyclo-[GGG]kB1[T] was significantly reduced compared with native kB1. Because structural perturbation of cyclo-[GGG]kB1[T] was minimal (Fig. 5), the loss of hemolytic activity is presumably due to the effects of the residues introduced into loop 6 upon cyclization. Previous grafting experiments utilizing kB1 have also shown the abrogation of hemolytic activity after relatively minor changes to the primary sequence; for example, the grafting of the vascular endothelial growth factor sequence (RRKRRR) into loop 2, 3, 5, or 6 significantly reduced the hemolytic activity of kB1 (20). The effects of single point mutations on the hemolytic activity of kB1 have also been extensively studied (73). The hemolytic activity of kB1 was dramatically reduced by Ala substitution on a specific set of residues, including the last three residues of loop 6 (Leu 27 , Pro 28 , and Val 29 ). Similar results were observed during Lys substitution at the same positions (69). Barry et al. (72) also noted the complete loss of hemolytic activity of linearized kB1 with the RNGLP sequence deletion. All of the studies suggest that the hemolytic activity of kB1 can be diminished when a less hydrophobic residue is introduced into the last three positions of loop 6. In kalata B2, a close homolog of kB1, loop 6 forms part of a hydrophobic interface that mediates interactions between the cyclotide and the plasma membrane (76). In this context, the loss of hemolytic activity after the introduction of additional residues to loop 6 suggests the interruption of critical membrane/peptide interactions in this region of the peptide.  The residues introduced into loop 6 might have also contributed to the slight reduction in stability of cyclo-[GGG]kB1[T] relative to the native peptide. Chemical shifts for these residues differed little from random coil values, suggesting that structural disorder at the cyclization point made the peptide slightly more susceptible to plasma proteases than native kB1. This disorder was also confirmed in the NMR solution structure of [GGG]kB1[T], with loop 6 displaying markedly increased disorder compared with native kB1. The serum stability results show that the cystine knot formation is crucial for stability because linear oxidized [GGG]kB1[TGG] was largely (ϳ71%) intact after 24 h.
The sortase-catalyzed process requires modest modifications in order to render peptides amenable to cyclization. In this work, the cyclization point was chosen on the basis of a pre-existing Leu-Pro-Val motif in loop 6, which overlapped with the SrtA recognition motif. This fortuitous coincidence resulted in the introduction of three fewer non-native amino acids than would be required using a different cyclization point. Analysis of chemical shifts showed that apart from differences of ϳ0.5 ppm in the Leu-Pro-Val residues of the cyclization point, the remainder of the SrtA cyclized peptide exhibited very little structural perturbation. Loop 6 of cyclotides exhibits the greatest diversity in structure, sequence, and length (77), and the lack of structural perturbation in this case reflects the ability of this loop to accommodate a range of structural motifs without affecting the overall fold of the peptide. Accordingly, the use of this loop for cyclization should provide a convenient and simple mechanism for production of synthetic cyclotides for a variety of applications.
To test the generality of SrtA-mediated cyclization, we investigated if it could be applied to other small disulfide-rich peptides, such as ␣-conotoxin Vc1.1 and SFTI-1. MALDI-TOF and NMR analysis confirmed that their cyclized and oxidized structures were similar to the native structures. In contrast to kB1, 9 h of SrtA reaction appeared to be sufficient to obtain the maximum yield for cyclo-[GG]SFTI-1[LPET], presumably due to its smaller size. Beyond this incubation time, some hydrolysis of the peptide was found to occur, resulting in linearization because the final product still contains a LPETG motif. Interestingly, introduction of the sortase motif in SFTI-1 (cyclo-[GG]SFTI-1[LPET]) gave rise to cis-trans isomerization of Pro 8 , as highlighted by the observation of strong NOEs between both Ile 7 H␣ i to Pro 8 H␣ i ϩ1 (cis) and Ile 7 H␣ i to Pro 8 H␦ i ϩ 1 (trans). Although the LPXTG motif was introduced to the cyclization loop of SFTI-1, presumably placing the motif next to the Asp 14 made the structure of the peptide less constrained than the wild type. This is consistent with a previous study showing that Ile 7 , Pro 8 , Pro 9 , and Asp 14 are important for structural integrity (49). In that study, when Asp 14 was replaced with Ala, the turn region was destabilized, resulting in cis-trans isomerization of Pro 13 (49). Although the isomerization site is different, introduction of the SrtA sorting motif appears to have destabilized the turn region, giving rise to the minor conformation observed by NMR.
The work described here demonstrates the applicability of the SrtA method for the cyclization of different classes of disulfide-rich peptides without significant structural perturbation to the overall peptide fold. We have shown that this method is applicable whether the peptides are cyclic in their native form, such as SFTI-1 containing one disulfide bond or kB1 with a three-disulfide bond cyclic cystine knot motif, or whether the peptide is linear in its native form and the cyclization site is introduced by the addition of a linker, such as for cVc1.1. SrtAmediated head-to-tail cyclization works optimally with peptides larger than 12-mers, excluding the SrtA sorting signal (40). Cyclotides are typically 28 -37 amino acids in size, making all characterized cyclotides suitable for SrtA cyclization. In addition, other disulfide-rich peptides (e.g. conotoxins (78) or defensins (79,80)) are also within this size range, making SrtAmediated cyclization a possibility for a wide range of peptides of therapeutic interest. Importantly, the ability to place the SrtA tags at any position within a peptide sequence makes it possible to introduce cyclization points anywhere in the peptide of interest and thus minimize disruption of biologically or structurally important motifs. Thus, peptide bond closure of linear precursors through sortase-catalyzed circularization represents a simple, safe, and effective method not only for the production of synthetic cyclotides but also for the introduction of increased stability and bioavailability to other disulfide-rich peptides with therapeutic value.