Integrating Computation, Experiment, and Machine Learning in the Design of Peptide-Based Supramolecular Materials and Systems.

Interest in peptide-based supramolecular materials has grown extensively since the 1980s and the application of computational methods has paralleled this. These methods contribute to the understanding of experimental observations based on interactions and inform the design of new supramolecular systems. They are also used to virtually screen and navigate these very large design spaces. Increasingly, the use of artificial intelligence is employed to screen far more candidates than traditional methods. Based on a brief history of computational and experimentally integrated investigations of peptide structures, we explore recent impactful examples of computationally driven investigation into peptide self-assembly, focusing on recent advances in methodology development. It is clear that the integration between experiment and computation to understand and design new systems is becoming near seamless in this growing field.

Guiding principles for peptide nanotechnology through directed discovery.

Life's diverse molecular functions are largely based on only a small number of highly conserved building blocks - the twenty canonical amino acids. These building blocks are chemically simple, but when they are organized in three-dimensional structures of tremendous complexity, new properties emerge. This review explores recent efforts in the directed discovery of functional nanoscale systems and materials based on these same amino acids, but that are not guided by copying or editing biological systems. The review summarises insights obtained using three complementary approaches of searching the sequence space to explore sequence-structure relationships for assembly, reactivity and complexation, namely: (i) strategic editing of short peptide sequences; (ii) computational approaches to predicting and comparing assembly behaviours; (iii) dynamic peptide libraries that explore the free energy landscape. These approaches give rise to guiding principles on controlling order/disorder, complexation and reactivity by peptide sequence design.

Molecular dynamics simulations reveal disruptive self-assembly in dynamic peptide libraries.

There is significant interest in the use of unmodified self-assembling peptides as building blocks for functional, supramolecular biomaterials. Recently, dynamic peptide libraries (DPLs) have been proposed to select self-assembling materials from dynamically exchanging mixtures of dipeptide inputs in the presence of a nonspecific protease enzyme, where peptide sequences are selected and amplified based on their self-assembling tendencies. It was shown that the results of the DPL of mixed sequences (e.g. starting from a mixture of dileucine, L2, and diphenylalanine, F2) did not give the same outcome as the separate L2 and F2 libraries (which give rise to the formation of F6 and L6), implying that interactions between these sequences could disrupt the self-assembly. In this study, coarse grained molecular dynamics (CG-MD) simulations are used to understand the DPL results for F2, L2 and mixed libraries. CG-MD simulations demonstrate that interactions between precursors can cause the low formation yield of hexapeptides in the mixtures of dipeptides and show that this ability to disrupt is influenced by the concentration of the different species in the DPL. The disrupting self-assembly effect between the species in the DPL is an important effect to take into account in dynamic combinatorial chemistry as it affects the possible discovery of new materials. This work shows that combined computational and experimental screening can be used complementarily and in combination providing a powerful means to discover new supramolecular peptide nanostructures.

Using experimental and computational energy equilibration to understand hierarchical self-assembly of Fmoc-dipeptide amphiphiles.

Despite progress, a fundamental understanding of the relationships between the molecular structure and self-assembly configuration of Fmoc-dipeptides is still in its infancy. In this work, we provide a combined experimental and computational approach that makes use of free energy equilibration of a number of related Fmoc-dipeptides to arrive at an atomistic model of Fmoc-threonine-phenylalanine-amide (Fmoc-TF-NH2) which forms twisted fibres. By using dynamic peptide libraries where closely related dipeptide sequences are dynamically exchanged to eventually favour the formation of the thermodynamically most stable configuration, the relative importance of C-terminus modifications (amide versus methyl ester) and contributions of aliphatic versus aromatic amino acids (phenylalanine F vs. leucine L) is determined (F > L and NH2 > OMe). The approach enables a comparative interpretation of spectroscopic data, which can then be used to aid the construction of the atomistic model of the most stable structure (Fmoc-TF-NH2). The comparison of the relative stabilities of the models using molecular dynamic simulations and the correlation with experimental data using dynamic peptide libraries and a range of spectroscopy methods (FTIR, CD, fluorescence) allow for the determination of the nanostructure with atomistic resolution. The final model obtained through this process is able to reproduce the experimentally observed formation of intertwining fibres for Fmoc-TF-NH2, providing information of the interactions involved in the hierarchical supramolecular self-assembly. The developed methodology and approach should be of general use for the characterization of supramolecular structures.

CHARMM force field parameterization protocol for self-assembling peptide amphiphiles: the Fmoc moiety.

Aromatic peptide amphiphiles are known to self-assemble into nanostructures but the molecular level structure and the mechanism of formation of these nanostructures is not yet understood in detail. Molecular dynamic simulations using the CHARMM force field have been applied to a wide variety of peptide-based systems to obtain molecular level details of processes that are inaccessible with experimental techniques. However, this force field does not include parameters for the aromatic moieties which dictate the self-assembly of these systems. The standard CHARMM force field parameterization protocol uses hydrophilic interactions for the non-bonding parameters evaluation. However, to effectively reproduce the self-assembling behaviour of these molecules, the balance between the hydrophilic and hydrophobic nature of the molecule is essential. In this work, a modified parameterization protocol for the CHARMM force field for these aromatic moieties is presented. This protocol is applied for the specific case of the Fmoc moiety. The resulting set of parameters satisfies the conformational and interactions analysis and is able to reproduce experimental results such as the Fmoc-S-OMe water/octanol partition free energy and the self-assembly of Fmoc-S-OH and Fmoc-Y-OH into spherical micelles and fibres, respectively, while also providing detailed information on the mechanism of these processes. The effectiveness of the parameters for the Fmoc moiety validates the protocol as a robust approach to paramterise this class of compounds.

Biocatalytic Self-Assembly Using Reversible and Irreversible Enzyme Immobilization.

Biocatalytic control of molecular self-assembly provides an effective approach for developing smart biomaterials, allowing versatile enzyme-mediated tuning of material structure and properties as well as enabling biomedical applications. We functionalized surfaces with bioinspired polydopamine and polyphenol coatings to study the effects of enzyme surface localization and surface release on the self-assembly process. We show how these coatings could be conveniently used to release enzymes for bulk gelation as well as to irreversibly immobilize enzymes for localizing the self-assembly to the surface. The results provide insights to the mode of action of biocatalytic self-assembly relevant to nanofabrication and enzyme-responsive materials.

Metastable hydrogels from aromatic dipeptides.

We demonstrate that the well-known self-assembling dipeptide diphenylalanine (FF) and its amidated derivative (FF-NH2) can form metastable hydrogels upon sonication of the dipeptide solutions. The hydrogels show instantaneous syneresis upon mechanical contact resulting in rapid expulsion of water and collapse into a semi-solid gel.

Insight into the esterase like activity demonstrated by an imidazole appended self-assembling hydrogelator.

A low molecular weight hydrogelator with a covalently appended imidazole moiety is reported. Capable of percolating water in the pH range of 6 to 8, it proves to be an efficient catalyst upon self-assembly, showing Michaelis-Menten type kinetics. Activities at different pH values correlated with dramatic structural changes were observed. It can hydrolyse p-nitrophenyl acetate (pNPA) as well as inactivated esters, and L and D-phenylalanine methyl esters. The enhanced activity can be related to the conglomeration of catalytic groups upon aggregation resulting in their close proximity and the formation of hydrophobic pockets.

Enzymatic synthesis of beta-lactam antibiotics via direct condensation.

In this paper, the feasibility of precipitation driven synthesis of acidic and zwitterionic beta-lactam antibiotics is studied. As an example of the first type, penicillin G was produced in good yield (160 mmol kg(-1)) directly from the free acid and amine aqueous substrate suspension, where the synthesis product precipitated. Such a precipitation driven synthesis via direct reversal of the hydrolytic reaction is thermodynamically unfavourable for zwitterionic beta-lactam antibiotics, such as amoxicillin. In this paper, a novel method is suggested to help favour precipitation of (poorly soluble) product salts by deliberate addition of certain counter-ions. After screening a number of different counter-ions, it was found that the amoxicillin anion forms a poorly soluble salt with Zn(2+). Despite increased beta-lactam degradation due to the presence of zinc ions, in a synthetic reaction with 0.1 M ZnSO(4) present the synthetic yield could be increased at least 30-fold.

Three-dimensional cell culture of chondrocytes on modified di-phenylalanine scaffolds.

The design of self-assembled peptide-based structures for three-dimensional cell culture and tissue repair has been a key objective in biomaterials science for decades. In search of the simplest possible peptide system that can self-assemble, we discovered that combinations of di-peptides that are modified with aromatic stacking ligands could form nanometre-sized fibres when exposed to physiological conditions. For example, we demonstrated that a number of Fmoc (fluoren-9-ylmethyloxycarbonyl) modified di- and tri-peptides form highly ordered hydrogels via hydrogen-bonding and pi-pi interactions from the fluorenyl rings. These highly hydrated gels allowed for cell proliferation of chondrocytes in three dimensions [Jayawarna, Ali, Jowitt, Miller, Saiani, Gough and Ulijn (2006) Adv. Mater. 18, 611-614]. We demonstrated that fibrous architecture and physical properties of the resulting materials were dictated by the nature of the amino acid building blocks. Here, we report the self-assembly process of three di-phenylalanine analogues, Fmoc-Phe-Phe-OH, Nap (naphthalene)-Phe-Phe-OH and Cbz (benzyloxycarbonyl)-Phe-Phe-OH, to compare and contrast the self-assembly properties and cell culture conditions attributable to their protecting group difference. Fibre morphology analysis of the three structures using cryo-SEM (scanning electron microscopy) and TEM (transmission electron microscopy) suggested fibrous structures with dramatically varying fibril dimensions, depending on the aromatic ligand used. CD and FTIR (Fourier-transform IR) data confirmed beta-sheet arrangements in all three samples in the gel state. The ability of these three new hydrogels to support cell proliferation of chondrocytes was confirmed for all three materials.

Comparison of methods for thermolysin-catalyzed peptide synthesis including a novel more active catalyst.

This is a comparative study of the performance of thermolysin for enzymatic peptide synthesis by reversed hydrolysis in several different reaction systems. Z-Gln-Leu-NH(2) was synthesized in acetonitrile containing 5% water (with various catalyst preparation methods) as well as by the "solid-to-solid" and frozen aqueous methods. Reaction rates (values in nanomoles per minute per milligram) in acetonitrile depended significantly on the method of addition of enzyme: (a) direct suspension in the reaction mixture as freeze-dried powders gave 60 to 95; (b) addition as an aqueous solution, so that enzyme precipitates on mixing with acetonitrile, gave 230; (c) addition as an aqueous suspension gave a remarkable increase in reaction rates (up to 780); (d) immobilized enzymes (adsorbed at saturating loading on celite, silica, Amberlite XAD-7, or polypropylene, then dried by propanol rinsing) all gave <230. It is postulated that, starting with the enzyme already in the form of solid particles in aqueous buffer, there is a minimum chance of alteration of its optimal conformation during transfer to the organic medium. For solid-to-solid synthesis with 10% water content we found initial rates of 670 under optimized conditions. In frozen aqueous synthesis, rates were <10. Equilibrium yields were always around 60% in low water organic solvent, whereas they were found to >80% in the aqueous systems studied.

Predicting when precipitation-driven synthesis is feasible: application to biocatalysis.

Precipitation-driven synthesis offers the possibility of obtaining high reaction yields using very low volume reactors and is finding increasing applications in biocatalysis. Here, a model that allows straightforward prediction of when such a precipitation-driven reaction will be thermodynamically feasible is presented. This requires comparison of the equilibrium constant, Keq, with the saturated mass action ratio, Zsat, defined as the ratio of product solubilities to reactant solubilities. A hypothetical thermodynamic cycle that can be used to accurately predict Zsat, in water is described. The cycle involves three main processes: fusion of a solid to a supercooled liquid, ideal mixing of the liquid with octanol, and partitioning from octanol to water. To obtain the saturated mass action ratio using this cycle, only the melting points of the reactants and products, and in certain cases the pKa of ionisable groups, are required as input parameters. The model was tested on a range of enzyme-catalysed peptide syntheses from the literature and found to predict accurately when precipitation-driven reaction was possible. The methodology employed is quite general and the model is therefore expected to be applicable to a wide range of other (bio)-catalysed reactions.

Monte Carlo simulation of the alpha-amylolysis of amylopectin potato starch. 2. alpha-amylolysis of amylopectin.

A model is presented that describes all the saccharides that are produced during the hydrolysis of starch by an alpha-amylase. Potato amylopectin, the substrate of the hydrolysis reaction, was modeled in a computer matrix. The four different subsite maps presented in literature for alpha-amylase originating from Bacillus amyloliquefaciens were used to describe the hydrolysis reaction in a Monte Carlo simulation. The saccharide composition predicted by the model was evaluated with experimental values. Overall, the model predictions were acceptable, but no single subsite map gave the best predictions for all saccharides produced. The influence of an alpha(1-->6) linkage on the rate of hydrolysis of nearby alpha(1-->4) linkages by the alpha-amylase was evaluated using various inhibition constants. For all the subsites considered the use of inhibition constants led to an improvement in the predictions (a decrease of residual sum of squares), indicating the validity of inhibition constants as such. As without inhibition constants, no single subsite map gave the best fit for all saccharides. The possibility of generating a hypothetical subsite map by fitting was therefore investigated. With a genetic algorithm it was possible to construct hypothetical subsite maps (with inhibition constants) that gave further improvements in the average prediction for all saccharides. The advantage of this type of modeling over a regular fit is the additional information about all the saccharides produced during hydrolysis, including the ones that are difficult to measure experimentally.