Predicting peptide self-assembly and phase transitions for the design of responsive biomaterials via molecular simulations and machine learning
Date
2022
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Recent advances in materials design, synthesis, and simulation have allowed the creation of biomimetic materials with responsive and controllable physicochemical properties. Such materials self-assemble into desired morphologies such as vesicles, fibrils and gels, and their ability to self-assemble can be tuned by applying external stimuli such as heat, light, pH, and salt for applications including drug delivery and tissue engineering. While experimental synthesis and characterization of biomaterials are often time consuming and limited in terms of resolution, simulations allow for efficient screening of broad design spaces while also giving insight into the molecular mechanisms and driving forces governing the complex phase behavior and assembly of responsive biomaterials. Therefore, there is a need to develop molecular models that can capture the thermodynamics and self-assembly of biopolymers to simulate experimentally relevant length scales and time scales. The overarching goal of my thesis is to use atomistic (AA) and coarse-grained (CG) molecular dynamics simulations and machine learning to study and design responsive, peptide-based biomaterials such as elastin-like peptides (ELP), collagen-like peptides (CLP), and ELP-CLP bioconjugates. ☐ In the first part of this thesis, I used atomistic and coarse-grained MD simulations to develop design rules for ELPs and ELP-CLP conjugates for the molecular design of thermoresponsive biomaterials. ELPs are biopolymers that have been shown to undergo a lower critical solution temperature (LCST)-like transition in aqueous solutions such that they are soluble below the transition temperature (Tt) and insoluble above Tt. Synergistic efforts involving MD simulations and experimental synthesis and characterization from the Kristi Kiick research group showed that amino acid substitutions involving bulky, hydrophobic amino acids such as phenylalanine (F) and tryptophan (W) and relatively hydrophilic amino acids such as tyrosine can be used to control the Tt of ELPs and ELP-CLP conjugates to obtain biomedically relevant transition temperatures. I used coarse-grained simulations along with atomistic simulations performed by Dr. Prhashanna Ammu to show that the incorporation of W residues impacted the hydrogen bonding via the formation of secondary structures and increased local chain stiffness, thus resulting in lower Tt. I also demonstrated using atomistic and coarse-grained simulations that substitutions of F with Y resulted in a higher propensity to form inter-peptide chain hydrogen bonds, and that the sequence order of Y impacted the π-π stacking in ELP-CLP systems such that the placement of Y residues near to the tethering point of ELP to CLP (terminus) discouraged favorable stacking interactions and resulted in poor self-assembly as reflected by higher Tt. ☐ In the second part of this thesis, I investigated the self-assembly of CLPs in aqueous solutions into triple helices, fibrils, and supramolecular networks (hydrogels). I used coarse-grained MD simulations, complemented by atomistic simulations by Dr. Francesca Stanzione and experimental synthesis and characterization of CLPs by the April Kloxin research group, to show that the incorporation of charged amino acids and nonnatural residues, such allyloxycarbonyl functionalized lysine resides (Kalloc) for crosslinking in CLP systems, result in thermally less stable triple helices (lower melting temperature) than canonical natural sequences. Using atomistic and coarse-grained simulations, I also explored the self-assembly of sticky ended CLP triple helices into fibrils and supramolecular networks and developed design rules for supramolecular assembly as a function of peptide design and solutions conditions. I also used machine learning models (artificial neural networks) to screen sticky ended CLP design and solutions conditions to predict CLP self-assembly into fibrils and supramolecular networks and minimum CLP concentrations for supramolecular network assembly. ☐ Overall, in this thesis, I present efforts involving the computational design of peptide-based biomaterials, linking molecular driving forces for self-assembly to their corresponding macroscopic structure and morphology. This work highlights the predictive and synergistic power of computational tools, such as simulations and machine learning, combined with experiments to design multifunctional soft biomaterials.
Description
Keywords
Lower critical solution temperature, Peptide self-assembly, Biomaterials, Machine learning, Amino acids