Characterizing circular peptides in mixtures: sequence fragment assembly of cyclotides from a violet plant by MALDI-TOF/TOF mass spectrometry

Cyclotides are a very abundant class of plant peptides that display significant sequence variability around a conserved cystine-knot motif and a head-to-tail cyclized backbone conferring them with remarkable stability. Their intrinsic bioactivities combined with tools of peptide engineering make cyclotides an interesting template for the design of novel agrochemicals and pharmaceuticals. However, laborious isolation and purification prior to de novo sequencing limits their discovery and hence their use as scaffolds for peptide-based drug development. Here we extend the knowledge about their sequence diversity by analysing the cyclotide content of a violet species native to Western Asia and the Caucasus region. Using an experimental approach, which was named sequence fragment assembly by MALDI-TOF/TOF, it was possible to characterize 13 cyclotides from Viola ignobilis, whereof ten (vigno 1–10) display previously unknown sequences. Amino acid sequencing of various enzymatic digests of cyclotides allowed the accurate assembly and alignment of smaller fragments to elucidate their primary structure, even when analysing mixtures containing multiple peptides. As a model to further dissect the combinatorial nature of the cyclotide scaffold, we employed in vitro oxidative refolding of representative vigno cyclotides and confirmed the high dependency of folding yield on the inter-cysteine loop sequences. Overall this work highlights the immense structural diversity and plasticity of the unique cyclotide framework. The presented approach for the sequence analysis of peptide mixtures facilitates and accelerates the discovery of novel plant cyclotides.


Introduction
Cyclotides are a unique class of cysteine-rich macrocyclic mini-proteins of about 30 amino acids in size that are defined by a head-to-tail cyclized backbone and three disulfide bonds in a knotted arrangement referred to as cyclic cystine-knot (CCK) motif . Their knotted structure makes them exceptionally stable against thermal, chemical and enzymatic degradation (Colgrave and Craik 2004). Cyclotides have been discovered and isolated from plants of the violet (Violaceae), coffee (Rubiaceae), cucurbit (Cucurbitaceae) and legume family (Fabaceae) . Their distribution within the plant kingdom still remains unclear (Gruber 2010), but they are expected to be far more widespread and the number of different cyclotides may be around 50,000 (Gruber et al. 2008;Simonsen et al. 2005) making them one of the largest peptide classes within plants. In agreement with their anticipated number, recent studies report the presence of more than 70 different cyclotides within one single species (Seydel et al. 2007;Gründemann et al. 2012). The first cyclotide kalata B1 was discovered from ''kalata-kalata'', a decoction from leaves of Oldenlandia affinis, which has been used as a remedy during childbirth in African ethnomedicine due to its uterotonic activity (Gran 1970;Gruber and O'Brien 2011). In line with their reported antibacterial (Tam et al. 1999), antifouling (Göransson et al. 2004), anthelmintic (Colgrave et al. 2008) and insecticidal properties (Jennings et al. 2001;Gruber et al. 2007a; Barbeta et al. 2008) their native function seems to be part of the plant defence system.
As a key feature, cyclotides are amenable to various amino acid changes by peptide engineering, which highlights the flexibility and plasticity of the cyclotide framework (Clark et al. 2006). Thus, their high sequence diversity is extensively under investigation for being utilized as scaffolds in the development of agrochemicals and pharmaceuticals (Henriques and Craik 2010). Besides these distinct differences in the sequences of the so-called intercysteine loops, cyclotides can be divided into two subfamilies, i.e., Möbius or bracelet type cyclotides based on the presence or absence of a cis-Pro residue in loop 5 ( Fig. 1)  ). These differences have further implications regarding their physico-chemical properties. Whereas most Möbius cyclotides are slightly negatively charged or have an overall net-charge of zero, bracelet cyclotides are usually multiply positively charged. This ultimately influences their chemical behaviour and amenability to sequencing and oxidative folding, which are still challenges, in particular for bracelet cyclotides.
Usually, amino acid sequencing of cyclotides is performed after enzymatic digestion of peptides that have been laboriously purified by reversed-phase high performance liquid chromatography (RP-HPLC) to produce single linearized peptides that are amenable to tandem mass spectrometry (MS) analysis. However, the complexity of cyclotide plant extracts, which comprise dozens of distinct peptides, limits their analysis and characterization by standard MS analysis. Using endoproteinase GluC (endo-GluC), cyclotides are mostly cleaved to yield a single ('ring-opened') peptide fragment due to a conserved glutamic acid in loop 1, whereas the use of trypsin and chymotrypsin usually yields several fragments due to multiple cleavage sites. When applied to the analysis of cyclotide mixtures as they occur in plant extracts, mass spectra may be confusing and hard to evaluate caused by fragment ion overlays. Hence the application of combinations of digests to obtain peptide-specific fragments and the subsequent accurate assembly of sequence fragments may overcome this issue. Particularly for bracelet cyclotides this is of importance since until now the majority (*70 %) of more than 200 published cyclotide sequences accessible on CyBase (Wang et al. 2008b) belong to this subfamily.
Besides the complexity of cyclotide sequence analysis, another issue associated with their great diversity is their chemical and biological synthesis. Previous studies have shown that different enzymes seem to be involved in backbone cyclization and disulfide bond formation (Gruber et al. 2007b;Saska et al. 2007) during biosynthesis of these gene-encoded peptides in planta. However, the community still lacks clarity about this process, in particular with respect to the sequence-folding relationship, i.e., how the inter-cysteine sequences of different cyclotides can kalata B1 LPVCGETCVGGTCNTP--GCTCSWPVCTR cycloviolacin O2 IP-CGESCVWIPCISSAIGCSCKSKVCYR influence the formation of the native CCK-motif and hence determine their folding yield. As a consequence, in vitro oxidative folding is still a major challenge in cyclotide engineering. Whereas high-yield chemical synthesis and folding of Möbius cyclotides is possible , obtaining correctly folded bracelet cyclotides is much more difficult and yields of about 10-40 % or less of native peptide are common (Leta Aboye et al. 2008;Wong et al. 2011). However, chemical synthesis of cyclotides is an important tool to obtain sufficient peptide material for bioactivity studies.
Our aim is to characterize plant cyclotides from Viola ignobilis, a native Iranian species of the violet family that was recently discovered to contain cyclotides (Hashempour et al. 2011). As a rich source for representatives of both cyclotide subfamilies, we analysed cyclotide-containing fractions using matrix-assisted laser desorption ionizationtime of flight (MALDI-TOF) MS and tandem MS analytics. Together with the use of different enzymatic digests and detailed analysis of mass spectra, full peptide sequence coverage could be achieved by assembling and aligning various sequence fragments. This approach, which we called sequence fragment assembly turned out to be a powerful tool for cyclotide identification and de novo sequencing even when analysing mixtures. To dissect the influence of cyclotide sequence variability with respect to the formation of their native structure, we performed oxidative refolding experiments using representative vigno (V. ignobilis) cyclotides comprising distinct, but subtle differences in their inter-cysteine loop sequences. Overall the characterization of these novel cyclotides highlights their enormous sequence variability and the proposed sequencing methodology may overcome limitations in the discovery of novel representatives of this unique class of circular plant peptides.

Plant collection, extraction and RP-HPLC fractionation
Aerial parts of V. ignobilis Rupr. were collected in the mountains at an altitude of 1,500-2,500 m around the village of Negarestan in the region of East Azerbaijan (Iran) in spring 2010. A voucher specimen was identified and deposited at the Institute of Medicinal Plants and Drug Research, Iran (MPH-1917). The dried plant material (*500 g) was ground prior to solvent extraction with a mixture of MeOH:CH 2 Cl 2 (1:1; v/v) overnight under continuous agitation at 20°C. After adding of 0.5 volume water the aqueous phase was concentrated on a rotary evaporator prior to freeze drying, yielding what is further referred to as crude extract. The crude extract was dissolved in 0.1 M NH 4 HCO 3 buffer (pH * 8.1) and immediately used for solid-phase extraction (SPE). C 18 SPE cartridges (Macherey-Nagel, Chromabond; 10 g; 50 mL) were activated with 1 bed volume of MeOH and subsequently equilibrated with 1 bed volume of aqueous 1 % FA. After application of the extract, the cartridges were washed with 1 bed volume of 1 % FA. Putative cyclotide containing fractions of 50 and 80 % EtOH were collected and freeze dried. After dissolving in 1 % FA they were fractionated using preparative and semi-preparative RP-C 18 HPLC (Knauer, Eurospher I 5 lm; 250 9 16.1 mm; 100 Å ) using a Knauer 1200 series unit, with an isocratic flow of 30 % acetonitrile: H 2 O (v/v) at a flow rate of 8 mL min -1 . Fractions were collected manually by UV detection at 210 nm. All samples were extracted by avoiding prolonged exposure to high pH and sample heating to reduce the risk of Asn deamidation.

Reduction, alkylation and enzymatic digest
Prior to MS analysis, cyclotides were enzymatically digested to produce linearized fragments following reduction and alkylation of Cys-residues. Lyophilized samples (*0.5 lg peptide) were dissolved in 0.1 M NH 4 HCO 3 buffer (pH 8.2) and 20 lL aliquots were reduced by adding 2 lL of 10 mM dithiothreitol and were incubated at 20°C for 30 min. Alkylation was carried out by adding 4 lL of 100 mM iodoacetamide to the reduced samples and incubating for 10 min at 20°C. After a second incubation step for 10 min with 1 lL of 10 mM dithiothreitol to quench the reaction with iodoacetamide, 2 lL of trypsin, endo-GluC and/or chymotrypsin (all Sigma-Aldrich, Austria) at concentrations of 0.1-0.5 lg lL -1 were added. All digests were incubated at 37°C between 3 and 16 h, quenched with concentrated acetic acid (final concentration 3 %) and stored at 4°C-20°C until further analysis.

MALDI-TOF/TOF analysis and peptide sequencing
Analysis of crude, reduced/alkylated and digested samples were performed on a MALDI-TOF/TOF 4800 Analyser (AB Sciex, Canada) operated in reflector positive ion mode acquiring 2,000-3,600 total shots per spectrum with a laser intensity set between 3,200 and 3,800. MS and MS/MS experiments were carried out using a-cyano-hydroxyl-cinnamic acid matrix at a concentration of 5 mg mL -1 in 50 % (v/v) acetonitrile. 0.5 lL of each sample was mixed with 3 lL of matrix solution and the mixture was spotted onto the target plate. Tandem mass spectra were acquired using laser energy of 1 kV with and without the use of collision-induced dissociation and processed using the Data Explorer Software. Cyclotides were identified by sequence fragment assembly (as explained below) and manual peptide sequencing. Automated database searches using the ERA-tool  and DeNovoExplorer software were used to compare manual annotated sequences. The MS/MS spectra were examined and sequenced based on assignment of the N-terminal b-ion and C-terminal y-ion series. The disulfide connectivity of C I-IV , C II-V and C III-VI was assigned based on homology with published sequences.
Oxidative refolding of cyclotides Cyclotides were purified by RP-HPLC on a Dionex Ultimate 3000 HPLC unit (Dionex, Netherlands) using semipreparative (250 9 10 mm) and analytical (250 9 4.6 mm) Kromasil C 18 columns (5 lm; 100 Å ) with linear gradients of 0.1-1 % min -1 or isocratic flow of 25-35 % buffer B (90 % acetonitrile in ddH 2 O, 0.08 % TFA) at flow rates of 3 and 1 ml min -1 , respectively. The control peptide kalata B1 was isolated from Oldenlandia affinis extract as described earlier (Gründemann et al. 2012). The same procedure was applied for the purification of cycloviolacin O2 from Viola odorata. Reduction was performed as described above and stopped after 30 min incubation by adding concentrated TFA (Sigma-Aldrich, Austria) and samples were immediately subjected to HPLC purification. Folding of 60 lL aliquots, containing 2.5-10 lM peptide, was performed at final concentration of 2 mM reduced (GSH) and 0.1 mM oxidized (GSSG) glutathione (Sigma-Aldrich, Austria). Freeze-dried aliquots were resolved in three different folding-buffers, i.e., 25 and 75 % isopropanol (Roth, Germany) and 35 % DMSO/5 % dodecyl-bmaltoside (DBM) in 0.1 M NH 4 HCO 3 buffer (pH 8.2). For control experiments with cycloviolacin O2 the folding conditions included final concentrations of 2 mM GSH and 2 mM cystamine in 35 % DMSO/5 % DBM buffer and GSH/ cystamine (2/2 mM) in 0.1 M Tris-HCl buffer (pH 8.5) at 4°C and 20°C. Aliquots were analysed at several time points (15 min, 1 h and 24 h) after incubation at 20°C. Folding reaction was quenched by adding 1 lL of concentrated TFA and samples were analysed by RP-HPLC on an Aeris Peptide XB-C 18 (150 9 2.1 mm; 3.6 lm; 100 Å ) column (Phenomenex, Germany) at a flow rate of 0.3 ml min -1 with a gradient of 2 % min -1 buffer B. Folding yields were determined using the peak integration tool of Chromeleon software 6.8 with a peak detection limit set at 0.07 9 signal (mAU at 214 nm) 9 retention time (min). Folding kinetic graphs and calculations of rate constant and half-time were prepared using the one-phase association fit in GraphPad Prism 5 software.

Results and discussion
The discovery and hence the pharmaceutical value of cyclotides is limited by an efficient and reliable protocol for peptide sequence analysis, in particular in crude plant extracts and fractions that contain mixtures of different cyclotides. Therefore, the main goal of this study was to describe a robust method for cyclotide sequence characterization using MALDI-TOF/TOF analytics.

Identification of novel cyclotides from Viola ignobilis
Viola ignobilis, a violet plant that frequently occurs in mountainous regions of Iran has recently been described to contain cyclotides (Hashempour et al. 2011) and this is consistent with previous studies showing that all Violaceae plants analysed to date express this unique class of cyclic mini-proteins. According to established extraction protocols it was possible to identify and isolate several cyclotide-containing fractions from V. ignobilis. Figure 2 shows the seven fractions obtained by RP-HPLC fractionation of the initial two solid-phase extracts. Exemplarily, a MALDI-TOF spectrum of one fraction (Fig. 2c) indicates the presence of multiple cyclotides whereas standard HPLC analysis is not powerful enough to resolve those coeluting cyclotides (Fig. 2d). This is a common scenario and usually requires further purification prior to MS sequencing. However in the case of low sample amounts, laborious purification may lead to sample loss and the use of mixtures might be inevitable. Hence the aim was to develop a protocol to perform MALDI-TOF/TOF-based sequence characterization using semi-pure fractions containing at least two cyclotides. Subsequently it was possible to identify 13 cyclotides (vigno 1-10, varv A, cycloviolacin O2 and cO9) in seven HPLC fractions (Fig. 2a, b).
De novo cyclotide sequencing using 'sequence fragment assembly' Cyclotide-containing mixtures were chemically modified to yield S-carbamidomethylated Cys-residues and digested to produce linear peptides amenable to fragmentation by tandem MS. Completely reduced and alkylated samples were digested using single enzymes or combinations of trypsin, endo-GluC and chymotrypsin. Resulting mass spectra were manually analysed by assigning N-terminal band C-terminal y-ions. Most novel cyclotide sequences were independently confirmed by automatic database searches using the ERA-tool ) and DeNovoExplorer software.
Cyclotide sequences are generally obtained from pure peptides by enzymatic digestion and interpretation of tandem mass spectra. For co-eluting Möbius cyclotides in mixtures, such as vigno 1 and vigno 2 (Fig. 2), it was possible to obtain the full, unambiguous sequences by evaluating the ion-fragmentation pattern and by alignment of two independent digests using trypsin or endo-GluC ( Fig. 3; Supplementary Table S1). In cases where single digests yielded incomplete fragment ion coverage, a combination of trypsin and endo-GluC digests was applied to generate smaller fragments and their analysis enabled complete sequence interpretation as shown for vigno 3 and 4 (Fig. 4, Supplementary Fig. S1; Supplementary Table S1). This procedure of reduction, alkylation and enzymatic digest provided also good sequence coverage of other Möbius cyclotides in mixtures namely varv A and vigno 5 (Supplementary Fig. S2; Supplementary Table S1).
By contrast, this methodology did not yield evaluable mass spectra for bracelet cyclotides. A linearized bracelet cyclotide using endo-GluC very often results in incomplete fragmentation making it impossible to assign. Tryptic digests of fractions containing multiple bracelet cyclotides on the other hand resulted in unclear fragmentation patterns due to multiple enzyme cleavage sites and hence fragmention overlay. Since cyclotides of the bracelet subfamily often contain multiple basic residues (Arg and Lys), a tryptic digest typically yields fragments corresponding to single, double or multiple cleaved peptides (Supplementary  Table S2). Due to the high sequence homology of cyclotides, the same fragment sequence may appear in more than one peptide and, therefore, in cyclotide mixtures these fragments have the same mass. To circumvent this problem, single and double digests combining chymotrypsin and endo-GluC were performed. This resulted in peptide  Supplementary Fig. S4). Finally each sequence had to be confirmed by assignment of the linearized endo-GluC digested cyclotides. Exemplarily, a flowchart of the sequence fragment assembly approach and the sequence elucidation of the bracelet cyclotide vigno 6 have been presented (Figs. 5,6). The tryptic digest of fraction 5 or 6 ( Fig. 2b) leads to the two fragments of 2,711.9 and 3,290.1 Da that may originate from several cyclotides, e.g., vigno 6, vigno 8 or cycloviolacin O2 (Supplementary Table S2). On the other hand, the chymotrypsin digest and a combination of chymotrypsin and endo-GluC yields fragments with distinct molecular weights and by combining the sequence information from the tryptic peptides with alignment of the chymotrypsin/ endo-GluC fragments it was possible to assemble the full sequence. Table 1 shows all cyclotide sequences from V. ignobilis that were elucidated by the sequence fragment assembly approach. The combination of several single and double digests of fractions containing multiple cyclotides and the alignment of partial sequences and assembling of peptidespecific fragments allowed the discrimination and unambiguous assignment and elucidation of the distinct cyclotide sequences. This approach ultimately led to the identification of 13 novel cyclotides, whereof ten display previously unknown sequences (Table 1). The use of multiple enzymes and varying combinations thereof together with MALDI-TOF/TOF analysis will overcome major limitations of cyclotide de novo sequencing and facilitate the discovery of novel sequences within peptide mixtures. MS/MS spectra and summaries of all digested fragments used for the sequence determination of all cyclotides from V. ignobilis are available as Supplementary Data (Supplementary Fig. S1, Supplementary Fig. S2, Supplementary  Fig. S3, Supplementary Fig. S4, Supplementary Table S1,  Supplementary Table S2). To identify cyclotides in mixtures using MALDI-TOF/TOF analysis is a powerful tool for an efficient sequence elucidation and the discovery of novel cyclotide sequences. The identification of ten novel cyclotides from V. ignobilis supports the evidence that cyclotides are one of the largest peptide classes within plants with immense sequence diversity in their inter-cysteine loops built around the stable CCK frame (Supplementary Fig. S5).   NGLPLCGETCAGGTCNTPGCSCSWPVCVR   y4  y5  y6  y7  y8  y8  y10  y11  y12  y13  y14  y15  y16  y17  y18  y20  y22  y23  y24  y25  y26  y27  y29   b2  b4  b10 b11  b14b15  b24  b28 b29   y3 y2  y19  y21  y1  y28   b3 b13 b16 b17 b20  The 13 identified peptides from Viola ignobilis belong to both subfamilies (Table 1). One Möbius cyclotide, known as varv peptide A, has been previously isolated from Viola arvensis (Claeson et al. 1998), as well as two bracelet cyclotides, known as cycloviolacin O2 and O9, have been originally found in Viola odorata ). This is not surprising since some cyclotides such as varv peptide E (=cycloviolacin O12), occur in many different Viola species such as V. tricolor, V. odorata, V. arvensis, V. bashoanensis, V. yedoensis and V. abyssinica and, therefore, seem to be genus-specific. Besides these rather rare examples of inter-genus identity, each single plant species seems to express an abundant array of specific cyclotides. In V. ignobilis the most abundant Möbius cyclotides are vigno 1 and vigno 2. Sequence analysis of these two peptides revealed the presence of an AGGT motif in loop 2 which was recently also described for V. abyssinica cyclotides (Yeshak et al. 2011).
Besides the conserved six Cys-residues and the glutamic acid (E) in loop 1, all six Möbius cyclotides have the same typical GET motif in loop 1 and a serine in loop 4, which is conserved within all newly identified cyclotides from V. ignobilis. Furthermore, the GES motif in loop 1 and the VWIP motif in loop 2 are conserved within all bracelet cyclotides from V. ignobilis. The differences and novelties are within loop 3, 5 and 6 which are known to show the highest amino-acid variability (Supplementary Fig. S5; Table 1). For vigno 2 a novel sequence motif for loop 6 of cyclotides, VRDGSSPL, has been discovered. Although all amino acids are known to occur in this loop, the presence of two serine residues next to each other has not been reported hitherto. The presence of two serine residues and an aspartic acid makes this loop more hydrophilic and confers the peptide with an overall net charge of -1.   3.

2.
RNGIPCGE % Intensity (b) Fig. 5 Sequence fragment assembly approach for vigno 6. a An overview of the sequence fragment assembly workflow that has been used to elucidate cyclotide sequences in mixtures is presented. b The combination of single trypsin, chymotrypsin and endoproteinase GluC and a combination of chymotrypsin/endo-GluC digests of fractions containing multiple cyclotides together with the alignment of partial sequences and assembling of peptide-specific fragments allowed the discrimination and unambiguous assignment and elucidation of the cyclotide sequences. MS spectra of four digests using endo-GluC   To distinguish between Asn and Asp residues in loop 6 of vigno 2 and vigno 10, we have additionally analysed the molecular weight and isotopic distribution of diagnostic fragment ions ) (Supplementary Figure  S6). Further sequencing of the co-eluting Möbius cyclotides vigno 3 and vigno 4 (Fig. 2) revealed the presence of an alanine or glycine within loop 3 and corresponds to the mass difference of 14 Da in the crude sample (Fig. 4a). A combined digest using trypsin and endo-GluC yields the fragments of 2,416.9 and 2,430.9 Da that allowed the sequence determination and confirming the difference of a glycine (NTPG) and alanine (NTPA) at the last position of loop 3. This is to our knowledge the first report of an alanine residue at this position and expands the known possibilities at this position which was primarily thought to be a conserved glycine . Within the sequence of vigno 5 a glycine at the first position of loop 5 was found, which so far has only been shown for bracelet cyclotides such as cycloviolacin Y1-3 (Wang et al. 2008a) and tricyclon A and B (Mulvenna et al. 2005). Vigno 6 shows the very common KSKV sequence for loop 5, which is intersected by a glycine, KGSKV. The identification of ten novel peptides underlines the high flexibility and sequence variability caused by single amino acid changes at various positions. It is obvious that sequence variability accounts for different biological and chemical behaviour due to varying physico-chemical properties. As an example, we decided to characterize the in vitro oxidative refolding properties of representative vigno cyclotides, since it is a valuable model to analyse their sequence-folding relationship, which has broader implications on the synthesis and design of cyclotides as tools in pharmaceutical applications.

Sequence-folding relationship of novel vigno cyclotides
The influence of certain residues on oxidative folding and the correct formation of the native disulfide bonds of cyclotides are still not fully understood. In the current study it has been our particular interest to elucidate the sequencefolding relationships of three vigno cyclotides with respect to their yield and folding kinetics using different folding conditions. Therefore, the oxidative refolding of the most abundant Möbius cyclotides in V. ignobilis, vigno 1 and vigno 2 was studied, in comparison to the prototypic cyclotide kalata B1, and the bracelet cyclotide vigno 10 in comparison to cycloviolacin O2, a well-studied bracelet cyclotide isolated from V. odorata.
To provide material for the folding studies all three vigno cyclotides and the control cyclotides kalata B1 and cycloviolacin O2 were purified and reduced. All native and reduced peptides were analysed by MALDI-MS and RP-HPLC confirming their high purity ([95 %) and the cysteine oxidation state (Fig. 7, Supplementary Fig. S7, Supplementary Table S3). As expected the reduced cyclotides showed a mass shift of 6 Da on MALDI-MS, indicating the complete reduction of the three native disulfide bonds. This results in decreased overall hydrophobicity and hence the peptides eluted significantly earlier from the reversedphase column (Fig. 7f). Based on previous studies on kalata B1 ) buffers containing 0.1 M ammonium-bicarbonate buffer with low (25 %) and high (75 %) concentration of isopropanol were used in the initial folding studies. High concentrations of alcohol were recently shown to favour folding of a bracelet cyclotide (Wong et al. 2011). In addition, a buffer containing 35 % DMSO and 5 % dodecyl-b-maltoside (DBM) was used; it has been shown for cycloviolacin O2 that a buffer containing DMSO and a non-ionic detergent yields reasonable amounts of folded peptide (Leta Aboye et al. 2008). All buffers contained the disulfide shuffling redox agents reduced (GSH) and oxidized (GSSG) glutathione. After dissolving the reduced peptides, the refolding process was monitored by taking aliquots at several time-points (15 min, 1 and 24 h). After 24 h incubation at 20°C, refolding was stopped by quenching the reaction with concentrated TFA and samples were subjected to RP-HPLC ( Fig. 7; Supplementary Fig.  S7) and MALDI-MS analysis (Supplementary Table S3). Generally refolding starts immediately after the reduced peptides have been dissolved in the folding buffer, which can be determined by disappearance of the peak in the HPLC chromatograms corresponding to the reduced cyclotides. Refolding data were analysed by measuring the area under curve of the peaks corresponding to the reduced and native peptides and plotting the folding yields in percentage versus time of incubation (Fig. 7, Supplementary   Fig. S7; Table 2). In particular for the Möbius cyclotides one can observe an increase of the peak corresponding to the native cyclotide after 15 min and the final yield was measured after 24 h of folding. As listed in Table 2, folding of the Möbius cyclotides vigno 1, vigno 2 and kalata B1 led to respectable yields between *30 and 90 % of refolded peptides, favoured by higher isopropanol concentrations versus the DMSO/detergent buffer. This is consistent with previous studies and the fact that the hydrophobic solvent appears to stabilize the surface exposed hydrophobic patches of cyclotides . To compare the folding kinetics of each cyclotide the rate constant (k n ) and the half-time (t 1/2 ) of appearance of the native species were calculated. For the three Möbius cyclotides the half-time of appearance of the native peak was between 25 and 96 min ( Table 2). Folding of bracelet cyclotides appears to be more complex and difficult under in vitro conditions as has previously been reported (Aboye et al. 2011;Gunasekera et al. 2009). Accordingly, the overall refolding yield of the bracelet cyclotides vigno 10 and cycloviolacin O2 in the two isopropanol buffers or the standard DMSO/detergent buffer (=35 % DMSO/5 % DBM in 0.1 M NH 4 HCO 3 with GSH/GSSG) was significantly lower (*10-14 %) (  this study using the DMSO/detergent buffer (11.4 %). However, their folding conditions were slightly different, i.e., (1) use of Brij 35 instead of DBM, (2) use of 2 mM cystamine in addition to GSH as redox component, (3) use of Tris instead of NH 4 HCO 3 -buffer, (4) folding temperature of 4°C as well as 20°C and (5) folding was allowed to proceed for 48 h with addition of fresh redox components after 24 h. To further analyse the influence of these differences folding of the control cyclotide cycloviolacin O2 in each folding condition was performed and in agreement with Göransson et al. we obtained the highest folding yield at 4°C using the Tris-buffered 35 % DMSO/ 6 % Brij 35 after 48 h of folding (Table 2). In particular, the longer incubation time and lower temperature appeared to enhance the yield of natively folded cycloviolacin O2. Figure 7 shows the HPLC chromatograms of the highest yielding folding conditions for each cyclotide. Comparison of the sequences and structures provides insights into the differences observed in the folding efficiency in terms of yield and kinetics. Vigno 1 and vigno 2 have high sequence similarity to kalata B1, and all three peptides have high folding yields in 75 % isopropanol buffer (Table 2). However, there are differences in folding using the 25 % isopropanol and 35 % DMSO/detergent buffers ( Table 2). The three Möbius peptides differ in loop 6, i.e., VRNGLPL in vigno 1, VRDGSSPL in vigno 2 and TRNGLPV in kalata B1 (Table 2; Supplementary Fig. S8). The two adjacent serine residues and an aspartic acid in vigno 2 make this loop more hydrophilic and confers the peptide an overall single negative net charge. This may explain higher folding yields of vigno 2 for the 25 % isopropanol buffer and lower yields for the more hydrophobic 75 % isopropanol buffer as compared to vigno 1, a neutral cyclotide. The more hydrophilic nature of loop 6 of vigno 2 ( Supplementary Fig. S8) (Aboye et al. 2011) may also contribute to the different folding yields in the DMSO/ detergent buffer, which vary between 62.7 % for vigno 1, 30.5 % for vigno 2 and 87 % for the control peptide kalata  Supplementary Fig. S8), which probably also contributes to the folding differences and the overall later elution on RP-HPLC (Fig. 7f). Analysis of the folding of the bracelet cyclotide vigno 10 highlights the complexity of the oxidative folding of this cyclotide sub-family. The folding yields of vigno 10 and cycloviolacin O2 in the 75 % isopropanol buffer were comparable (13.5 vs. 11.4 %, respectively), but considerably lower than that observed for the Möbius cyclotides.
Interestingly, the use of the DMSO/detergent buffer did not increase the yield of correctly folded vigno 10, but in fact resulted in negligible amounts of the native fold. By contrast, this buffer resulted in approximately 11 % of the correctly folded cycloviolacin O2. Comparison of the sequences reveals that only loops 3 and 6 differ between vigno 10 and cycloviolacin O2. Loop 3 differs by two residues and loop 6 of vigno 10 is a hybrid between the loop 6 sequences of cycloviolacin O2 and kalata B8. Loop 3 forms a helical structure in bracelet cyclotides, and has previously been suggested to be important in the folding of cycloviolacin O2 (Göransson et al. 2009;Leta Aboye et al. 2008). The introduction of two valine residues in loop 3 of vigno 10, which does not favour the formation of helices, may account for slight distortion in this region of vigno 10 (Fig. 8). A study involving the synthesis of hybrids of kalata B1 and cycloviolacin O1 indicated that bracelet loops 2 and 6 significantly influence the folding (Gunasekera et al. 2009). Given the loop 2 sequences of vigno 10 and cycloviolacin O2 are identical, the differences in loop 6 are also likely to be involved in the folding differences observed for these two peptides. In the model of vigno 10 the backbone and side chain atoms of residues T2, K29 and D30 have different orientations as compared to residues R29 and N30 of cycloviolacin O2 (Fig. 8). These structural changes could contribute to the decreased folding yield in the DMSO/detergent buffer, by destabilizing a late folding intermediate, which shifts the equilibrium towards the nonnative conformation (Leta Aboye et al. 2008). Furthermore, loop 2 contains an isoleucine previously shown to detrimentally influence the oxidative folding and this residue may, in part, be responsible for the generally lower yields observed for vigno 10 and cycloviolacin O2 with respect to the Möbius cyclotides. Taken together, our folding studies of the novel vigno cyclotides compared to kalata B1 and cycloviolacin O2 confirms the significant difference of Möbius versus bracelet folding and the strong influence of the surface characteristics of cyclotides and solution conditions on their folding.

Conclusion
This work has broadened the knowledge about the immense sequence diversity of plant cyclotides, a unique class of naturally occurring backbone-cyclized peptides built around a conserved cyclic cystine-knot. By characterizing 13 sequences from an Iranian violet species it has been confirmed that cyclotides are one of the most abundant peptide class within the plant kingdom. The characterization of cyclotides in mixtures using MALDI-TOF/ TOF analytics may overcome laborious isolation and challenges in de novo peptide sequencing. The use of ) and the homology model of vigno 10 (green) were aligned using PyMOL (root-mean-square deviation = 1.324 Å ) and are shown in cartoon representation (a). The cyclotide loops and disulfide bonds (yellow) are indicated. Differences in the side-chain orientation of the distinct residues of both cyclotides of loop 3 (b) and loop 6 (c) are indicated in stick representation and amino acids are labelled in one-letter code, numbered according to their position in the cyclotide sequence (starting from G1, see d). Images have been prepared using PyMOL. (d) Sequence alignment of vigno 10 and cycloviolacin O2 with the three conserved disulfide bridges (shown in yellow) and the cyclized backbone (black dots and connecting line). Residues differing between those two cyclotides have been highlighted in red (color figure online) Characterizing circular peptides in mixtures 593 different proteases as well as the assembly and alignment of sequence fragments facilitates the discovery of novel cyclotide sequences. In addition, by performing oxidative refolding studies on representative cyclotides the knowledge of their in vitro oxidative folding behaviour was extended and this underlines the high dependency of folding yield to their inter-cysteine loop sequences and careful choice of the folding conditions. These studies have further implications taking into account that cyclotides have numerous bioactivities and hence display a scaffold that is extensively used for peptide-based drug design.