Structural, Nanomechanical, and Computational Characterization of d , l -Cyclic Peptide Assemblies

The rigid geometry and tunable chemistry of D , L -cyclic peptides makes them an intriguing building-block for the rational design of nano- and microscale hierarchically structured materials. Herein, we utilize a combination of electron microscopy, nanomechanical characterization including depth sensing-based bending experiments, and molecular modeling methods to obtain the structural and mechanical characteristics of cyclo -[(Gln- D -Leu) 4 ] (QL4) assemblies. QL4 monomers assemble to form large, rod-like structures with diameters up to 2 µm and lengths of 10s to 100s of µm. Image analysis suggests that large assemblies are hierarchically organized from individual tubes that undergo bundling to form larger structures. With an elastic modulus of 11.3 ± 3.3 GPa, hardness of 387 ± 136 MPa and strength (bending) of 98 ± 19 MPa the peptide crystals are among the most robust known proteinaceous micro- and nano-fibers. The measured bending modulus of micron-scale fibrils (10.5 ± 0.9 GPa) is in the same range as the Young’s modulus measured by nanoindentation indicating that the robust nanoscale network from which the assembly derives its properties is preserved at larger length-scales. Materials selection charts are used to demonstrate the particularly robust properties of QL4 including its specific flexural modulus in which it outperforms a number of biological proteinaceous and non-proteinaceous materials including collagen and enamel. The facile synthesis, high modulus, and low density of QL4 fibers indicate that they may find utility as a filler material in a variety of high efficiency, biocompatible composite materials.

Cyclic peptides; Supramolecular; Molecular dynamics; Elastic modulus; Amyloid; Nanoindentation Research into the structure, chemistry and functionality of biomaterials has accelerated in recent years with the goal of creating environmentally benign materials that have optimized physical and mechanical properties. 1 Many of these materials (e.g. silks, 2,3 collagens 4,5 ) derive their properties from protein constituents, 6 which form specific secondary structures that assemble into higher order structures with robust yet efficient mechanical behavior. 7 Another class of proteinaceous materials, amyloid fibrils, are particularly interesting because they can be formed from many different protein sequences. 8 Amyloids were first studied in conjunction with neurodegenerative diseases such as Alzheimers, 9 but newly discovered amyloidogenic gene products are increasingly being associated with functional (as opposed to pathogenic) biological performance. 10 Amyloids are classified by their fibrillar structure in which β-strands associate to form an extended network of supramolecular β-sheets perpendicular to the long-axis of the fiber. 11 This structural motif is intriguing in that it is largely defined by the properties of the peptide backbone itself, namely the propensity for hydrogen bonding between carbonyl oxygen and amide hydrogen. 12 This allows a variety of different proteins, when subjected to various denaturing and assembly processes, to form amyloid-like structures. 13 These nanoscale fibrils, self-assembled from individual protein and peptide molecules, continue to grow and form robust fibers with hierarchical order. Indeed, the dense hydrogen bonding and ordered structure of amyloid materials renders them among the most robust known proteinaceous materials with experimentally determined Young's moduli at the nanoscale in the range of 2-4 GPa. 14,15 The combination of self-assembly, nanoscale order and high stiffness provides a breadth of potential applications including light-harvesting, 16 carbon capture, 17 drug delivery, 18  D,L-cyclic peptides (DLCPs) share certain characteristics with amyloid fibrils including self-assembly, dense intermolecular hydrogen bonding along the peptide backbone, and chemical and mechanical stability. However, their molecular structure and geometry allow for an enhanced ability to chemically modify the monomers and to guide assembly. 20 The peptide cycles are composed of six to ten amino acids with alternating D-and Lstereochemistry, causing a planar geometry in which the amino acid side chains radiate from the center of the ring and the amide backbone is perpendicular to the plane of the ring. This geometrical arrangement promotes their assembly into high aspect ratio nano-and microstructures through β-sheet-like hydrogen bonding in which each ring sits flat on the surface of another ring (Figure 1a). Unlike linear amyloids in which β-sheets are able to slide with respect to one another, 21 DLCP assemblies are only able to stack directly atop each other. Depending on the sequence of the peptides, the nanotubes may also associate laterally into bundles to create structures that can be microns in width and hundreds of microns in length. 22 The assembly of cyclic peptides was first theorized by De Santis et al. 23 and later demonstrated using a pH sensitive peptide 8-mer that was found to assemble into large, micron-scale fibers. 20 As DLCPs are synthesized via common solid-state peptide synthesis methods 24 , one has the freedom to explore limitless combinations of amino acids and postsynthetic modifications. 25 This versatility has been leveraged to study DLCPs in a variety of applications including structural antibiotics, 26 ion channels, 27 selective ion transporters, 28 and reinforcing agents. 29 Additional research has focused on the fundamental properties of cyclic peptides including the thermodynamics of assembly and the breadth of structural flexibility. 30 One area that remains largely unexplored is the utility of cyclic peptides as structural components of composite materials, perhaps because there is little information on the mechanical properties of DLCP structures themselves. While several computational analyses of DLCP structures indicate that they are quite stiff, [31][32] there have been no direct measurements of their mechanical properties. As amyloids represent some of the stiffest known protein-based materials, we hypothesized that the similarly structured DLCP assemblies would also exhibit high modulus and strength. The unusual ability of DLCPs to position functional groups precisely in space, coupled with their relatively straightforward synthesis, make them well suited to study the combined effects of geometry and sequence on mechanical properties. Such a study would provide essential information for downstream applications of DLCPs within medical devices and tissue scaffolds.
In this study we demonstrate that assemblies of the DLCP cyclo-[(Gln-D-Leu) 4 ], abbreviated here as QL4, are among the stiffest and strongest known peptide-based materials.
This was accomplished through computational, structural and nanomechanical analyses including molecular dynamics simulations, electron microscopy, nanoindentation, and micro-bending studies. QL4 was selected because it is known to form assemblies, herein referred to as fibers, that are large and stable enough to be probed with accessible nanomechanical probe tools. This particular DLCP has also been shown to increase the elastic modulus of polylactides when incorporated as the filler component of a composite fiber. 29 While previous attempts at molecular dynamics simulations have given clues regarding the elastic modulus of a DLCP assembly, 31 this study represents the first experimental characterization of the mechanical properties of DLCP fibers. the Amide bands of the Raman and FT-IR spectra, which is similar to that of β-sheets 22,33 ( Figure S2). SEM analysis of the QL4 fibers indicates that there are two classes of assembled structures, those which are not bundled and have diameters on the order of ~100 nm and those that are bundled and have diameters on the order of a few microns (Figure 2a). Present within the larger bundles are striations that indicate a bundle is composed of several ~100 nm assemblies. Furthermore, within larger bundles, one observes gaps in which fibers have broken free as is highlighted in Figure 2b, indicating that the ~100 nm assemblies are discrete elements in the larger bundle. We hypothesize that this bundling may occur naturally due to hydrophobic effects between the surfaces of QL4 assemblies in an increasingly aqueous environment during the assembly process or, alternatively, may be a function of the evaporation necessary in sample preparation. High-resolution TEM micrographs of fibers show striations with thicknesses of approximately 2 nm along the long-axis of the fiber, providing evidence that the QL4 fibers are composed of individual DLCP nanotubes ( Figure   2c and d). Taken together, this information suggests that there are three levels of hierarchy in a large QL4 assembly: Individual tube (~ 2 nm diameter), fiber (~ 100 nm diameter), and bundled fiber (~ 1 µm diameter) (Figure 2e). While it is possible to verify the existence of the fiber, and bundled fiber, we have not obtained evidence that individual QL4 tubes exist in solution. Prior work has shown that conjugation of polymers to the DLCP side chains can lead to stable assemblies of individual DLCP tubes, 28, 34 however, we have found that without modification, individual tubes continue to grow until a larger fiber or bundled fiber is formed. Through molecular dynamics simulations, we obtained a theoretically ideal structure for QL4 fibers and used this as a reference for both structural and mechanical analyses.

RESULTS AND DISCUSSION
Along the longitudinal axis of assembly, QL4 monomers assemble through the creation of 12 hydrogen bonds, 8 along the peptide backbone and 4 mediated by glutamine sidechains, locking the monomers together with a spacing of 9.62 Å ( Figure 3A) along the z-axis. Along the lateral axes of assembly (x and y) the fiber is most stable in a square-packed conformation with a unit cell spacing of 17.73 Å. In this orientation, glutamine residues and leucine residues provide interlaced hydrogen bond and hydrophobic interactions in the x-y plane. The density of the resultant peptide assembly is calculated to be 1000 kg/m 3 . For further detail on the computational method and structure, please refer to the supplemental figures S3-S6. One can appreciate a key difference in the geometry of QL4 versus a typical amyloid in that for QL4, hydrophobic interactions are interdigitated with the hydrogen bonds in the lateral direction, while in a typical amyloid fibers, hydrogen bonds and hydrophobic interactions (or other -R group-mediated interactions) are present in two separate, orthogonal planes ( Figure S7). 35 We hypothesize that these interlocking interactions are partially responsible for the solvent stability, large size, and general robustness of the assembled fibers-hydrogen bonds maintain their strength in the presence of nonpolar solvents while the interspersed leucine residues may act to prevent the ingress of polar solvents. We used the structural model to analyze the elastic modulus of the QL4 assembly in the X-, Y-and Z-axes as described in Figure 3. In the Z-axis, along the length of the tube, the theoretical modulus is 22.6 GPa, in line with the stiffest known protein-based materials. 8 In the X-and Y-axes, the modulus is approximately 2.7 GPa due to the decrease in hydrogen bonds per unit area within the structure ( Figures S3-S6). It is worth noting that even in the comparatively compliant axes, our model suggests that QL4 assemblies possess a stiffness in the range of 2-4 GPa, equal to the measured value in amyloid fibers. Along the stiff axis, the theoretical modulus of 22.6 GPa is significantly higher than what has been measured at the nanoscale for amyloid or other protein-based fibrils.
The direct mechanical characterization of QL4 fibers was performed experimentally through both nanoindentation and micro-scale bending experiments, which were conducted using a depth-sensing nanoindenter. Both analyses were performed on large QL4 assemblies with widths ranging from 1-2 µm. In indentation studies, QL4 samples were deposited on a  Figure S8). The average elastic modulus and hardness of a large QL4 fiber were determined to be 11.3 ± 3.3 GPa and 387 ± 136 MPa, respectively. It is intriguing to note that the average modulus (measured perpendicular to the fiber axis) is significant larger than the computed moduli in the x and y directions (2.72 GPa).
We attribute this discrepancy to two factors. First, the indentation stress field beneath the indenter is complex and characterized by principal tensile stresses whose direction lay almost perpendicular to the direction of indentation, 36 namely along the fiber axis. Since the computed modulus along the fiber axis is high (22.6 GPa), the results may be a consequence of some peptide nanotube sustaining tensile stresses during indentation, resulting in a higher apparent modulus. Second these results may also be a result of the strong lateral interactions between the individual peptide nanotubes, leading to increased mechanical stability, a hypothesis that is also supported by the micro-bending experiments presented next. Despite the large size of the fibers tested here, the measured modulus and hardness of QL4 fibers are also significantly higher than those measured for individual amyloid fibers by AFM-based methods. To elucidate the fracture strength of QL4 fibers and to gain a deeper understanding of the fiber modulus, fibers were subjected to a micro-scale bending analysis (Figure 5a). QL4 fibers were deposited on uncoated molybdenum TEM grids with 40 µm circular pores ( Figure 5b). After locating individual fibers spanning pores in the grid, the fiber was bent, orthogonal to the long axis, by a cono-spherical tip with a 5 µm nominal radius. To confirm that fiber loading neither compressed the fiber nor bent the grid, the same force was applied to fibers on a flat surface and the measured displacement was negligible. Because the fibers are very long in comparison to the grid pore, and because the ends of the fiber did not appear to move during or after the experiment, we have used a clamped beam as a model for our calculations ( Figure S9). The bending modulus and strength, 10.5 ± 0.9 GPa and 97.8 ± 18.8 MPa, respectively, were calculated from individual force versus displacement curves ( Figure   5e and f). In the displacement curve, partial fracture is indicated by sudden decreases in the measured load (Figure 5d). We hypothesize that these micro-fractures occur as individual fibers within the larger bundle are broken and/or delaminated from one another (Figure 5c). It is worth noting that microscale modulus values via bending analysis are nearly identical to those obtained through nanoindentation perpendicular to the fiber axis. Unlike many other molecularly self-assembled systems, QL4 DLCPs are able to grow to such large sizes that direct nano-and micro-scale measurements can be obtained on the same samples and compared, providing an understanding of the mechanical properties across higher length scales. Prior bending analyses of other tube-based assemblies including single-walled carbon nanotube ropes and microtubules have shown that the shear modulus of the bundle is much weaker than the axial (Young's) modulus. 37,38 This is due primarily to a lack of inter-tube interaction within the bundle, which leads to bending and flexing as tubes slide with respect to one another. Conversely, QL4 fibers contain a dense network of inter-tube hydrogen bonds, mediated by glutamine side chains that act to stabilize each tube with respect to its neighbors ( Figure S6). We suggest that these lateral interactions are directly responsible for the robust flexural and Young's moduli because they lead to more efficient load transfer between the tubes, which is also in line with the high modulus measured by nanoindentation.
Indeed high load transfer will enhance the development of a continuum-like stress field beneath the indenter that results in regions of the fibers to be under tensile stresses, as discussed above. Furthermore these interactions contribute to the remarkably rigid, rod-like morphology of QL4 fibers. The overall rigidity, characterized by the persistence length (l p = EI/k b T), where I is the moment of inertia of the fibers, is on the order of meters (assuming a circular cross-section), which explains the lack of any noticeable bending within the fibers observed to date. This brings to light an interesting lever with which one may be able to tune the properties of DLCP structures. By tuning the side chain chemistry of the individual DCLP monomer that composes the fiber, one may be able to tune the flexural modulus and control the general morphology and properties of the assembled fiber to be more rod-like or more flexible in nature. A summary of the mechanical results for both the computational and experimental analyses is presented in Figure 6. In order to place the modulus and strength of QL4 within the context of other natural and synthetic materials, we have constructed two material selection charts (Ashby plots): one modulus vs. density plot (Figure 7a), with the density inferred from the computational modeling, and one modulus vs. strength plot (Figure 7b). A subset of the data on the plots is also tabulated in numerical form in the Supplementary Information (Figure S10). Guidelines drawn on the modulus vs. density plot allows one to compare the specific performance of materials in tension (given by E/ρ ) or in bending (given by E 1/2 /ρ). 7,39 One sees that the efficiency of QL4 fibers in tension outpaces a variety of materials, including collagen, tendon, and cancellous and compact bone. The material outperformance is more significant when considering the material efficiency in bending, where QL4 fibers also outperform enamel and even steel. The relatively facile synthesis, high aspect ratio, high modulus, and low density of QL4 fibers indicate that they may find utility as a filler material in a variety of high efficiency, biocompatible composite materials. In comparing the stiffness and strength, it is evident that QL4 fibers exhibit similar behavior to bone and that their modulus is on par with the stiffest known proteinaceous materials including spider dragline silk.
The stiffness vs. strength selection chart of QL4 (Fig. 7b), together with the materials index σ f 2 /E (guidelines with slope of 2) allows one to compare the materials performance in terms of their ability to store elastic energy prior to fracture. 7,39 In both stiffness and strength, QL4 fibers are most directly comparable to bone. The stiffness of QL4 fibers exceed that of all but the most robust known proteinaceous materials such as spider dragline silk. Although the elastic energy to failure of silks is superior, as evidenced by their location along the farright guideline in Figure 7b, QL4 fibers exhibit a performance metric on par with commonly known biological materials such as collagen, tendon and keratin. Perhaps unexpectedly, Figure 7a and 7b support the claim that the mechanical response of QL4 fibers is more similar to 'light' bone than to other protein based materials.

CONCLUSION
Our multi-scale mechanical analysis of QL4 fibers by nanoindentation and microbending demonstrates that DLCP assemblies are comparable to the most robust known protein and peptide materials. Indeed, the properties exceed those of amyloid fibrils and are on par with those of silks, suckerins, 40,41 and diphenylalanine nanotubes, 42 the most robust known self-assembling peptide based materials. Furthermore, they maintain those properties even at the micron length scale. This is a noteworthy feature given that the larger scale assembly is not stabilized by covalent, electrostatic, or coordination bonds between the individual monomers. In terms of stiffness and strength, QL4 is similar to bone, despite being roughly half the density. We hypothesize that the stiffness in both indentation and flexural

Transmission electron microscopy of QL4 assemblies A suspension of QL4
assemblies in water was spotted on a quantifoil TEM grid, wicked away, and stained with uranyl acetate. Sample was viewed on a JEOL2100 TEM at 200kV. For higher resolution images, spots were selected in which the assembly was overlapping a pore in the quantifoil. Assemblies overlapping pores were bent using a 5 µm cono-spherical indenter tip. The force and displacement curves were used to identify the modulus and strength as well as the nature of fracture of the material.

Supporting Information
The Supporting Information Available: Diagram and mass spectrometric characterization of QL4, FTIR and Raman spectra, detailed description of molecular dynamics approach, schematic of β-sheet and DLCP structures, description of nanoindentation and bending. This material is available free of charge via the Internet at http://pubs.acs.org.

AUTHOR INFORMATION
Corresponding Author

Method for modeling peptide behavior
In this paper, we employed the COMPASS force field (Condensed-phase Optimized Molecular Potentials for Atomistic Simulation Studies) -developed by H. Sun 3 , to optimize the geometry and calculate the energy of all molecules. The COMPASS force field is based on stateof-the-art ab initio and empirical parametrization techniques. The valence parameters and atomic partial charges were supported by ab initio data, and the van der Waals (vdW) parameters were derived by fitting the experimental data of cohesive energies and equilibrium densities.
The convergence tolerance is 2x10 -5 kcal/mol for the energy, 0.001 kcal/mol/Å for the force, 0.001 GPa for the stress and 10 -5 Å for the displacement. The Ewald method is used for calculating the electrostatic and the van der Waals terms. The accuracy is 10 -5 kcal/mol. The repulsive cutoff is 6 Å for the van der Waals term. For the periodical structure, the box vectors are also optimized together with the molecules. The Mechanical Properties are calculated by applying strain using static approach. The elastic constants were calculated from a polynomial fit to the calculated stress-strain relation.  Table 1. The value of C33 is the highest since the cyclic peptides are assembled by strong H-bonds along the z direction. This is the major source for stabilizing the peptide in the nanotube form. The values of C11 and C22 are smaller compared with the C33 coefficient, which is due to the decrease in hydrogen bond density as well as the comparatively weaker interactions, such as van der Waals and electrostatic type, among cyclic peptides in the x-y plane. Hence, the anisotropy of the system is manifested by the more rigid character in the z direction. When taking the Cii (i=1,2,3) as the approximate Young's moduli, the data in Table 1 are in good agreement with previously reported results for similar peptide nanotubes and protein crystalline systems 4,1 and also with the experimental data measured in this work.  By rotating the cyclic peptide along z-axis, another conformer is found. (As shown in Figure S4). The lattice constants are calculated as a= 17.94 Å b= 18.11 Å c= 9.92 Å at the optimized structure. This conformer of cyclic peptide nanotubes is classified as anisotropic due to the differences in the mechanical properties along x and y direction. For example, the elastic constant C11 is 4.43 GPa for x direction and 1.36 for y direction. Along z direction, the interaction is comparatively weaker due to a smaller elastic constant C33 of 20.21 GPa than that of Model 1. The total energy of Model 2 is calculated to be 67.6 kcal/mol higher than that of Model 1. The reason why Model 2 is higher in energy can be explained by weakening hydrogen bonds along z-axis. As shown in Figure S5, in Model 2, the inter subunit hbonds increased by about 0.2 Å for each hydrogen bond. The side chain hydrogen bonds increase by 0.2 Å for each hydrogen bond. This is also consistent with the longer lattice constant c of 9.92 Å for Model 2 compared with the more compact model1 (c=9.62 Å). Meanwhile, leucine residues are packed tightly together between each pair of tubes, allowing for close contact and favorable van der Waals interactions (blue circle).     --Nanoindentation [11,12]