Figure 1.
Schematic illustration of cyclotide structure.
A 3D illustration of the cyclotide structure showing the CCK motif (with cysteines labeled I–IV), and the inter-cysteine loops (labeled 1–6). The CCK motif is characterized by a cyclic peptide backbone and the knotted disulfide bonds (I–IV, II–V, III–VI), which are indicated by thick brown bars. The structure usually contains antiparallel beta strands (indicated by large yellow arrows) and several tight turns.
Figure 2.
Flow chart depicting the methodology used for QSAR modeling.
Predicted structures for the cyclotides were obtained through homology modeling using the sequences of the 13 cyclotides whose solution-phase structures had previously been determined by NMR. A) To identify appropriate templates, each target sequence was aligned with those of the 13 peptides with known structures. A neighbor joining (NJ) tree was then constructed based on the multiple sequence alignment. The resulting cladogram was used to identify the closest relatives of the target sequence with a known structure, which were then used as templates for modeling the unknown structure of the target peptide. B) For each template sequence, 20 PDB structures were generated based on the pairwise sequence alignment of the target sequence with the selected templates. These 20 conformations were evaluated using the DOPE potential and GA341 score, and the best three conformations were selected. If two templates were chosen for a target sequence, a total of 6 PDB structures were selected. C), D) and E) The structures generated during step B were subjected to conformation searches. If structural refinement of the protein as a whole yielded a result with an unacceptable geometry or one for which the RMSD of the Cα-atoms was greater than 2.0 Å relative to the starting conformation, two loops were selected for loop sampling. Loops were selected based on their sequences, focusing on those containing residues with different physicochemical properties relative to those in the corresponding positions of the template sequence or those that aligned with gaps in the template sequence. After structure refinement, one conformation of each peptide sequence was selected for use in calculating the molecular descriptors. The molecular descriptors of the cyclotides, together with their activity data were used as variables in QSAR modeling.
Figure 3.
The sequences of the membrane-buried regions of selected cyclotides and their lipophilicity profiles.
The multiple sequence alignment shows all of the cyclotides for which 3D structures (based on solution-phase NMR studies) are available. These sequences were used as template candidates for homology modeling of the cyclotides of unknown structure. The cysteine of loop 1 was arbitrarily selected as the N-terminus for the multiple sequence alignments. The residues required for cyclization, i.e. the conserved C-terminal residue (Asn or Asp) and N-terminal residue (Gly), are positioned within loop 6. The membrane-buried residues are highlighted with a gray background. OPM was used to predict the orientation of the selected cyclotides on the membrane. Among the members of the Möbius subfamily, many of the loop 5 and 6 residues are membrane-buried. Among the bracelet cyclotides, large proportions of loops 2 and 3 are buried. Only a couple of residues from the cyclotides of the hybrid subfamily penetrate the membrane interface. The lipophilicity of the loops was calculated as the total lipophilicity (LS) of all of the residues within the loop. It should be noted that the membrane-buried regions identified by OPM are consistent with those exhibiting high lipophilicity.
Figure 4.
Template selection strategy based on the NJ tree.
This figure illustrates the strategy of a template selection using a cladogram in which the target sequence (a) relates differently to various sequences of known structure. The symbols R and R′ represent the other branches of the cladograms. These branches contain sequences of known structure. Our aim was to identify sequences of known structure that could reasonably be used as templates for the target sequence based on the bootstrapped consensus tree. The procedure used to select the template sequences most closely related to the target sequence from the bootstrap NJ tree depended on the structure of the tree and the position of the target sequence relative to the nearest neighbors of known structure. In cases such as that shown in subfigure A, a single peptide of known structure (b) was selected as the template for target sequence (a) if the bootstrap value for the two taxa, a and b, was greater than 50%. Two taxa, b and c were selected if the bootstrap value for taxa a and b was below 50%. In cases such as that shown in subfigure B, two taxa, b and c were selected as templates, regardless of the bootstrap value of the common ancestor of the three taxa, a, b and c. In case C, we randomly chose two templates from the sister taxa (R′) of the target sequence (a) if the sister taxa (R′) included two or more sequences of known structure.
Table 1.
Normalized lipophilicity and solvent accessible surface area (SASA) values for natural and modified amino acids, assuming maximal side-chain exposure.
Figure 5.
The calculation of lipophilic moment (LM).
Figure 6.
The calculation of the HBD amphipathic moment (EM).
Table 2.
Labels used to refer to each cyclotide considered in this work and their activities against H. contortus and cells from the U-937GTB line.
Figure 7.
Moment plot (LM, EM) for the studied cyclotides.
The cyclotides, labeled according to their subfamily (M for Möbius, B for bracelet, H for hybrid) and with their unique numbers, were plotted on a map based on their lipophilic moments (LM) and HBD amphipathic moments (EM). They cluster into two groups based on their moments: one group has low moments and exhibits low activity while members of the other group have high moments and exhibit moderate or high activity. The synthetic hybrid cyclotides, [W23K, P24N, V25K]-kB1 (H1), [P24K]-kB1 (H2) and [P24D, V25K]-kB1 (H3) cluster in low moment group. The activity of the cyclotides H1 and H3 against the U937GTB cell line and H. contortus is not known, but they are reported to be inactive against human type A erythrocytes [30].
Figure 8.
A dendrogram showing the hierarchical clustering of 75 cyclotides based on their molecular descriptors.
The descriptors considered include the lipophilic moment (LM), exclusive lipophilicity , positively charged surface area (ES) and HBD amphipathic moment (EM). The left column shows the cyclotides' labels, which indicate the subgroup to which they belong (indicated by the letter ‘G’), relative activity (indicated by the letter ‘A’) and a subfamily-specific letter (M for Möbius, B for bracelet, H for hybrid and L for linear). The relative distance measures the physicochemical dissimilarity between cyclotides in respect to the normalized scale. The relative activity was calculated with respect to the cytotoxicity of kB1, and three levels of activity were defined: A0 for peptides whose relative activity ranges of >1.0, A1 for those with relative activities between 0.2 and 1.0, and A2 for those with relative activities of <0.2. Those with unknown activity were labeled An. Aside from [Mee]-cyO2 and varv A, all of the cyclotides were assigned to the same activity groups for both cytotoxicity and anthelmintic activity. The cyclotide subfamilies clustered into different subgroups, with most of the Möbius cyclotides being in groups 1 (red) and 2 (green), and all of the bracelets being in group 3 (blue). Aside from psyle A (H5), all of the hybrid cyclotides clustered in group 2. Group 2 also contains some unusual cyclotides that exhibit high or intermediate activity, namely cyO14 (M33), cyO15 (M34), cyO16 (M35), [T20K, S22K]-kB1 (M19) and [S22K]-kB1 (M22). These cyclotides are all rich in lysines.
Figure 9.
Predicted membrane binding modes (left) and lipophilicity profiles (right) for kalata B1 (A), cycloviolacin O2 (B) and kalata B8 (C) in their native cyclic conformations and in hypothetically unfolded conformations.
The outer phosphate layer of a phospholipid membrane is represented by a red line in each case. Strained backbone regions with little structural mobility are indicated in blue. Backbone regions with a defined secondary structure are colored in yellow and red to indicate β-strands and α-helices, respectively. Only selected side chains are depicted, including those that form the hydrophobic patch, residues carrying positive charges, and the conserved glutamic acid in loop 1. The cyclotide structures were obtained from the PDB server; their PDB IDs are 1NB1 (kalata B1) [28], 2KNM (cycloviolacin O2) [33], and 2B38 (kalata B8) [16]. The lipophilicity profile of each residue was calculated for the natively folded cyclotides and for their unfolded conformations. The hypothetical unfolded cyclotide conformations were defined such that ψ = φ = 180°. It should be noted that in the natively folded conformations of kalata B1 (kB1) and cycloviolacin O2 (cyO2), the degree of solvent exposure for lipophilic residues is high while that for lipophobic residues is low. For example, in loop 2 of cyO2, the lipophilicity of Trp and Pro is higher in the native fold than in the unfolded conformation. Moreover, the lipophobicity values for polar residues are relatively high in natively folded kB8 compared to natively folded kB1 and cyO2. In contrast to the situation for kB1 and cyO2, the polar residues of kB8 exhibit similar lipophobicities in the native and unfolded conformations. Notably, the glutamic acid residue in loop 1 of kB8 participates in an internal hydrogen-bonding network within the core of the protein that effectively minimizes its SASA value. Consequently, it exhibits a much higher level of lipophobicity in the native fold than do the equivalent residues of kB1 and cyO2.
Table 3.
Molecular descriptors used in the QSAR models.
Figure 10.
Generation of QSAR models.
Table 4.
QSAR equations for activity against the U-937GTB line and H. contortus.
Figure 11.
Statistical parameters for the moment plots (LM, EM).
The grayscale values indicate the performance of the QSAR models based on the correlation coefficient (r2) and the cross-validated correlation coefficient for the moment plot. Panels A and B show the values of r2 and
for the model describing cyclotide cytotoxicity against the lymphoma cell line, while panels C and D show the r2 and
values for anthelmintic activity. Cyclotides whose moment values were below the critical point were placed in the low moment group, while those whose moments exceeded the critical point were placed in the high moment group. The critical points were determined to be (0.03750, 4.5000) and (0.02813, 5.6250) for activity against the U937GTB line and H. contortus, respectively.
Figure 12.
Chart of molecular weight and flexibility in cyclotides.
This chart illustrates the correlation between flexibility and molecular weight. Points where multiple cyclotides overlap are indicated by the letters A-E; the cyclotides at these points are: A (M42, M43), B (M5, M9, M12, M14, M17, M24, M25), C (M4, H2), D (M2, M6, M10, M11, M16) and E (B10, B21). The linear cyclotide (L1 and L2) is relatively flexible since it lacks a cyclic backbone. The members of the hybrid cyclotide subfamily (H4, H6) have high molecular weights and flexibility values.