Rare by Natural Selection: Tunable Disulde-bonded Supramolecular Antimicrobials Peptides

Short helical antimicrobial peptides forming inter-molecular disulfide bonds are selected against in nature, and were utilized here to design switchable antimicrobials via the formation of functional supramolecular fibrils. Specifically, using the available structural information on the stable fibril-forming human LL-37 17-29, we designed cysteine mutations and demonstrated position-dependent controllable antibacterial activity, mediated by their disulfide-dependent self-assembly into ordered fibrils, which proved sensitive to reducing conditions. The crystal structure of the LL-37 17-29 bearing a I24C substitution, located in a critical structural position, revealed disulfide-bonded dimers that further assembled into a fibrillar structure of densely packed helices. The native and mutant peptides both featured a fibril surface with zigzagged hydrophobic and positively charged belts, which likely underlie interactions with bacterial membranes. Yet, they differed in their helical packing arrangement, which corresponded with different levels of activity, with only the mutant being susceptible to reducing conditions. The presented findings promise to advance the design of novel antimicrobials resistant to harsh conditions for coating of surfaces susceptible to pathogens. with bacterial cells 41 . We designed eight single-point cysteine mutations in different structural locations in hLL-37 17-29 and analyzed their antibiotic activity against four bacterial strains, along with their sensitivity to reducing conditions. The observations were supported by cysteine substitutions in the amphibian uperin 3.5, another AMP that forms supramolecular structures, specifically of amyloid fibrils with a functional secondary structure switch between cross- a and cross- b configurations 42 . We further determined the atomic structure of a cysteine mutant at the center of the helical hydrophobic moment, the most deeply buried position in the assembly of hLL-37 17-2941 . Overall, this work reinforces the importance of self-assembly in enabling the antibacterial activity of hLL-37 17-29 and demonstrates redox-switchable AMP activity, mediated by regulation of intermolecular disulfide bonds and their central role in the formation of supramolecular structures. by hLL-37 17-29 and hLL-37 17-29 with one of eight single-point cysteine mutants, indicated by MIC values, tested up to 200 µ M. The cells are colored from most to least active (lowest to higest MICs) colored blue to red, respectively. The ability of DTT, at x10 molar ratio compared to peptide concentration, to abolish activity, is indicated by (+) or (-), as detailed in supplementary Figures 7-10. In cases of non-bactericidal peptides, sensitivity to DTT was not relevant and the effect is marked with (o). Cases in which the DTT was too toxic for the bacteria at the indicated concentrations are marked by N/A. Mutant sensitivity to DTT was manually scaled and is indicated is the left column using a gray color scale, with light to dark shades indicate low to high DTT sensitivity, respectively. The MIC determination and DTT sensitivity experiments were performed at least four times each, on different days. Error values of MIC values indicate the standard error of all repeat measurements. Light

Paradoxically, in the presence of elevated concentrations of a reduction agent, the antibiotic activity of the human b-defensin 3, which is a cysteine-rich AMP, decreased, while that of bdefensin 1 increased, indicating distinct structure-function relationships and physiological regulation of antimicrobial peptides via reductive pathways 28,29,31,32 .
In contrast to the group of cysteine-rich AMPs, which are predominantly b-rich in their secondary structure, and usually form intramolecular disulfide bonds, AMPs with single cysteines can only form intermolecular disulfide contacts, yielding covalently bonded dimers.
Since AMPs, especially helical amphipathic types, tend to self-assemble to enhance antimicrobial activity [33][34][35][36][37][38][39][40] , such covalent dimers are expected to affect aggregation, with functional implications. We further expected that an intermolecular disulfide bond dictates the fibrillar morphology of short helical AMPs that assemble into functional ordered supramolecular structures, which can be reversed by reducing conditions. To test this hypothesis, we utilized the structural knowledge obtained for the active core peptide of the human AMP LL-37 (hLL-3717-29), which forms highly stable supra-helical fibrils that interact with bacterial cells 41 . We designed eight single-point cysteine mutations in different structural locations in hLL-3717-29 and analyzed their antibiotic activity against four bacterial strains, along with their sensitivity to reducing conditions. The observations were supported by cysteine substitutions in the amphibian uperin 3.5, another AMP that forms supramolecular structures, specifically of amyloid fibrils with a functional secondary structure switch between cross-a and cross-b configurations 42 . We further determined the atomic structure of a cysteine mutant at the center of the helical hydrophobic moment, the most deeply buried position in the assembly of hLL-3717-29 41 . Overall, this work reinforces the importance of self-assembly in enabling the antibacterial activity of hLL-3717-29, and demonstrates redox-switchable AMP activity, mediated by regulation of intermolecular disulfide bonds and their central role in the formation of supramolecular structures.

Short amphipathic helical AMPs with an odd number of cysteines are rare
We sought to determine the abundance, among the thousands of AMPs included in the CAMPR3 database 43 , of short (<40 amino acids) amphipathic helical AMPs (sahAMP) which contain an odd number of cysteine residues, rendering them more likely to form dimers connected via intermolecular disulfide bonds. AMPs were defined helical according to a secondary structure prediction performed by the Jpred server 44 . AMPs were defined as amphipathic if the hydrophobic moment (µH) of the helical part of the sequence was above the average µH for all short helical AMPs. Despite the relatively high abundance of cysteine residues in AMPs, compared to proteins in general (Figure 1), the prevalence of sahAMPs with an odd number of cysteines was especially low (Supplementary Figure 1). Specifically, we found that among 662 sahAMPs, only 16 contained an odd number of cysteine residues (Supplementary Figure 1 and Supplementary Table 1). All but one of these 16 sequences had a single cysteine. In comparison, the prevalence of sahAMPs with an odd number of tyrosines, a residue which shows similar secondary structure propensities to cysteine 45   The figure outlines amino acid frequencies in four groups: proteins from the Swiss-Prot database 46 (n=563972), proteins shorter than 40 amino acids from the Swiss-Prot database (n=9526), AMPs shorter than 100 amino acids from the CAMPR3 database 43 (n=2916), and AMPs shorter than 40 amino acids from the CAMPR3 database 43 (n=2103). The amino acid frequencies are presented as percentage (%), indicating its relative abundance. The color code ranges from lowest to highest prevalence, colored in increasingly deep shades of blue to red, respectively.

The hLL-3717-29 I24C mutant forms supra-helical fibrils with inter-molecular disulfide bonds
In a recent characterization of the atomic details of stable supra-helical hLL-3717-29 fibrils 41 , we report on fibrils composed of a basic unit of four-helix bundles with a hydrophobic core, with Ile24 located at the center of the bundle, completely buried within this assembly 41 . Substituting this position with alanine (I24A), which is less bulky and hydrophobic, or with various polar residues, fully abolished the ability of hLL-3717-29 to inhibit the growth of M. luteus, as well as its ability to form ordered supramolecular structures 41 . We hypothesized that a substitution to cysteine will maintain inter-molecular associations via the formation of a stabilizing disulfide bond in the center of the hydrophobic face of the hLL-3717-29 amphipathic helix.
Solving the crystal structure of the hLL-3717-29 I24C mutant at 1.5Å resolution (PDB code 7NPQ, Supplementary Table 2

Figure 3. Disulfide-bound helical dimers of the hLL-3717-29 I24C mutant further assemble into a fibril via a network of cation-Π and hydrophobic interactions
The hLL37(17-29) I24C protofibril encapsulating the hydrophobic core is shown in a grey ribbon representation. (a-b) A view down the fibril axis. (c) A side-view along the fibril axis of the protofibril, rotated 90˚ compared to panels a-b. Phe17 side chains are colored pink, and Arg23 and Cys24 side chains are colored by atom type (nitrogen in blue and sulfur in yellow). Phe17 and Arg23 are presented as sticks (a) or as space-filled atoms (b&c). Cys24 is presented as sticks in all panels. Close contacts between the aromatic Phe17 and the side chain of Arg23 suggests Π-stacking and cation-Π interactions that further stabilize the protofibril core.
The  Figures 7-10). This is likely due to the ability of these mutants to form active supramolecular or aggregated structures at high concentrations, independent of the disulfide bond. -

Reduction of disulfide bonds dissembled hLL-3717-29 I24C but not the Q22C mutant
The effect of DTT on self-assembly was evaluated by assessing the particle size distribution of the peptides using dynamic light scattering (DLS) (Supplementary Figure 12 and Supplementary Table 4). The particle size population of native hLL-3717-29, with an average diameter of ~300 nm, indicating large assemblies, was not affected by the addition of DTT. In contrast, hLL-3717-29 I24C displayed a drastic DTT-dependent reduction in particle sizes, which had an average diameter of 1.3 nm, which roughly corresponds to a small oligomer of few subunits. Specifically, while untreated hLL-3717-29 I24C samples contained only ~34% small particles, and the remainder were large particles with an average size of ~340 nm, their abundance increased to ~100% on the addition of DTT. This indicates that self-assembly of I24C is very much dependent on disulfide bond formation, and can be reversed by a reducing agent. In contrast to the effect of DTT on I24C, it had no effect on the Q22C particle size distribution, which had an average diameter of ~360 nm, similar to the native hLL-3717-29, suggesting that a disulfide bond is either not involved, or not critical for the formation of supramolecular species.

Self-assembly in the presence of M. luteus bacterial cells
The   Table 5). While DTT had no effect on the activity of native uperin 3.5, it reduced that activity of both mutants ( Figure 6 and Supplementary Table 5), with a greater impact on the I13C mutant (located on the hydrophobic face) as compared to the S11C mutant (located on the hydrophilic face). This is likely due to more efficient supramolecular fibrillar structure formation when the disulfide bond strengthens the hydrophobic intermolecular interfaces.

Figure 6. Cysteine mutants of the sahAMP uperin 3.5 show DTT-sensitive antibiotic activity
The effect of uperin 3.5 and of its I13C and S11C mutants on M. luteus bacterial growth was tested in the presence and absence of DTT. Bars show the ratio of M. luteus growth in the presence of the indicated peptides, compared to the bacteria growth without the peptide. DTT was added at x10 molar ratio compared to peptide concentration. The experiments were performed at least three times, on different days. Error bars represent the standard deviation of the mean of all biological repeats.  Table 3). In accordance with its critical structural location, substitutions of Ile24 to polar residues indeed abolished its antibiotic activity against M. luteus 41 . Even a milder substitution to alanine, which is more hydrophobic than cysteine, but still less hydrophobic than isoleucine 51 , abolished antibiotic activity against M. luteus and the formation of ordered fibrils of hLL-3717-29 41 . It was therefore predicted that the activity of I24C, if any, will be dependent on the formation of disulfide bond-mediated assembly. The I24C mutant was not active against S. hominis, E. coli and P. fluorescens. It remained active against M. luteus, but only under oxidizing conditions, which support the formation of a disulfide bond and assembly of supramolecular fibrils, as shown by the crystal structure ( Figure 2). Reducing conditions indeed dissembled the ordered assembly of the I24C mutant, as demonstrated by the marked shift in particle size distribution towards small particles (Supplementary Figure 12 and Supplementary Table 4), and as visualized in Figure 5. The results further support the important role of self-assembly in hLL-3717-29 antibiotic activity 41 .
DTT was least effective in altering the activity of the mutants bearing K18C and Q22C substitutions, which are located on the polar face of the helix (Figure 4 and Supplementary Figure 6). Theoretically, a disulfide bond connecting the polar face of the helix could hinder the formation of ordered fibrillar structures featuring a hydrophobic core. Therefore, one conjecture is that the K18C and Q22C mutants are active, to the same extent, as monomers, covalent dimers, or as amorphous aggregate. Yet a more plausible hypothesize is that these mutants self-assemble regardless of the formation of a disulfide bond via an extensive fibrillar hydrophobic core featuring a more extensive surface area buried along with a network of interactions, as shown for hLL-3717-29 41 and its I24C mutant (Figures 2-3 and Supplementary   Figures 2-4), compared to an isolated covalent dimer. In accordance with this hypothesis, while DTT had a significant effect on I24C activity, it had no effect on Q22C particle size distribution, which remained primarily composed of large particles (Supplementary Figure 12 and Supplementary Table 4). To conclude, we demonstrated the feasibility of designing switchable antimicrobials that can be deactivated under reducing conditions, but remain active under oxidative conditions, such as in areas of inflammation [23][24][25] . Their dependence on oxidizing conditions can also be used to target activity of AMPs to act on cancerous cells 54 . It is possible that the observed selection against sahAMPs with an odd number of cysteine residues is due to the microorganisms' ability to express reducing factors [25][26][27] . Since AMP self-assembly can often bear functional relevance and enhance antimicrobial activity 33 , another explanation to this negative selection could be that the intermolecular disulfide bonds lead to a reduction in the number of conformational states, and to reduced entropy, similarly to disulfide-mediated protein folding 55-60 , thereby hindering peptide aggregation. The demonstrated control over AMP activity is enabled via regulation of its self-assembly into functional supramolecular structures, which can be used as scaffolds for a wide range of bio and nanotechnology, regenerative medicine and bioengineering applications 61 , with the invaluable advantage of an inherent antibacterial activity. Our findings can specifically advance the design of novel antimicrobials with stability under harsh conditions, for coating of medical devices, food packages, water pipes and other surfaces susceptible to aggressive and resistant pathogens.

Bacterial strains and culture media
Micrococcus luteus (M. luteus, an environmental isolate) was a kind gift from Prof. Charles Greenblatt from the Hebrew University of Jerusalem, Israel. An inoculum was grown in Luria-Bertani medium (LB), at 30 °C, 220 rpm shaking, 16 h 42 . Staphylococcus hominis (subsp.
Pseudomonas fluorescens (P. fluorescens) was a kind gift from Prof. Roi Kishoni from the Technion, Israel. An inoculum was grown in LB, at 28 °C, with 220 rpm shaking, for 16 h 62,63 .
Escherichia coli, subsp. k3 3106 (E. coli) was a kind gift from Prof. Ehud Gazit from Tel Aviv University, Israel. An inoculum was grown in LB, at 37 °C, with 220 rpm shaking, for 16 h.

Dynamic light scattering (DLS)
The particles size distribution of hLL-3717-29 and its cysteine mutants, Q22C and I24C, with or without DTT, was compared using DLS. Lyophilized peptides were dissolved in UPddw to a

Calculations of structural properties
The electrostatic potential map and hydrophobicity coloring presented in the figures were generated using Chimera 68 . The values of the hydrophobicity scale were according to Kyte and Doolittle 69 . The electrostatic potential was calculated using APBS-PDB2PQR 70 . Helix amphipathicity, and chemical and physical properties of sequences (Supplementary Figure 6) were calculated with HeliQuest 71 . The helical wheels were also generated by HeliQuest 71 .

Solvent-accessible surface area calculations
Solvent-accessible surface areas (SASAs) were calculated using AREAIMOL, with a probe radius of 1.4Å 72,73 , via the CCP4 package 66 . The solvent-accessible buried surface area of each chain in the asymmetric unit was calculated as the area difference between the isolated chain and the chain within the fibril assembly, and is presented as the percentage of the total SASA of the chain. The SASA per residue within different isolated helical assemblies is presented in Supplementary Table 3.

Databases and protein groups analyzed
To calculate amino acid frequency in different protein groups, we used the Swiss-Prot database of curated proteins 46 , and the (CAMPR3) database 43 for a collection of anti-microbial peptides. All 20 amino acid frequencies were calculated in four different protein groups: proteins from the Swiss-Prot database 46 (n=563972 protein sequences (PG1)), proteins shorter than 40 amino acids from the Swiss-Prot database (n=9526 sequences (PG2)), AMPs shorter than 100 amino acids from the CAMPR3 database 43 (n=2916 sequences (PG3)), and AMPs shorter than 40 amino acids from the CAMPR3 database 43 (n=2103 sequences (PG4)).
For each group, the frequency of each amino acid was calculated as follows: where PG# denotes the four different protein groups.

Secondary structure prediction and amphipathicity
The Jpred webserver 44  times to obtain a minimum length of 20 amino acids. We defined the secondary structure as helical using three threshold values: (1) minimal number of residues predicted as helix is eight, or the residues predicted as helix encompass at least 80% of the sequence length (2) Minimal gap of four residues between segments predicted as helix and (3) Minimal Jpred predicted score of the helix is at least two, on average, over the predicted residues encompassing the helix (this is considered a low threshold, over a range of probability scores going up to nine).
Thus, we were permissive in defining helical sequences. The hydrophobic moment (µH) was calculated as ] & , where N is the sequence length, . is the hydrophobicity of the n th amino acid in the sequence according to its octanol\water partition 74 and is the angle separating side chains along the backbone, with = 100° for an α-helix 75 . The µH threshold was determined as the mean µH calculated for all AMPs shorter than 40 amino acids that were predicted as helical, and helical sequences were defined as amphipathic in case the calculated µH was higher than the µH threshold.

Specific amino acid frequency calculations in sahAMPs
The number of sequences containing an odd number of cysteines was counted among AMPs shorter than 40 amino acids from the CAMPR3 database 43 , among short AMPs predicted as helical, and among those also defined as amphipathic. The numbers are presented in Supplementary Figure 1. As controls, the numbers of sequences from the same groups having an odd number of tyrosines, due to its similar helical propensity, according to the Cho-Fasman table 76 and of arginines, due to its similar frequency in sahAMPs as cysteine, were counted. To graphically present the number of AMPs containing an odd number of cysteine, tyrosine, or arginine (residue 'X'), normalized to their frequencies in AMPs shorter than 40 amino acids from the CAMPR3 database 43 , including 2103 sequences (PG4), we used :