High throughput characterization and data-driven design of single-chain polymer nanoparticles
Description
TitleHigh throughput characterization and data-driven design of single-chain polymer nanoparticles
Date Created2022
Other Date2022-10 (degree)
Extent1 online resource (332 pages) : illustrations
DescriptionProteins are remarkable macromolecules that exhibit thermodynamic stability with an ability to fold into globular, compact structures. Over several decades of study and a wealth of knowledge stored in places such as the Protein Data Bank (PDB), these phenomena are well understood to be due to hydrophobic collapse, hydrogen bonding, and electrostatic interactions to provide stabilizing interaction forces and impart solubility. A major scientific challenge involves mimicking the compactness and flexibility of proteins in the laboratory utilizing synthetic materials in the form of single-chain polymer nanoparticles (SCNPs). SCNPs provide design flexibility with the ability to vary chain length, composition, architecture, and valency. However, a design paradigm for SCNPs for therapeutic or drug delivery applications is lacking. This is because traditional polymer chemistries required degassing and as a result limited throughput. More recently, our group with collaborators have developed techniques such as enzyme-assisted reversible addition-fragmentation chain-transfer polymerization (Enz-RAFT) and photoinduced electron/energy transfer RAFT (PET-RAFT) which have enabled automated polymer synthesis directly in 96-well plates. With these tools at our disposal, it is now more realistic to sample a large chemical landscape and combine high throughput experimentation (HTE) with data-driven design to realize quantitative structure-activity relationships (QSAR). In this dissertation, we hypothesize that a high throughput polymer synthesis and characterization approach informed by machine learning (ML) can be utilized to define an SCNP structural design paradigm related to bioactive polymers and drug formulation while improving experimental efficiency. To do so, we developed a combinatorial HTE strategy to synthesize, purify, and characterize SCNPs for compactness and flexibility and ability to formulate a hydrophobic small-molecule drug. We further incorporated ML to provide novel polymer designs to identify SCNPs with similar compactness and flexibility as proteins.
The first specific aim involves developing high throughput tools to prepare and characterize diverse SCNPs for compactness and flexibility. First, we validated a plate-based size-exclusion purification technique for removing small-molecule impurities associated with PET-RAFT purification and strain-promoted azide-alkyne cycloaddition (SPAAC) click chemistry. For a diverse library of polymers, we demonstrated the ability to remove small-molecule associated with SPAAC to >97% efficiency and retained polymers on average at >87% in a single step. Once this technique showed potential for combinatorial chemistry, we were able to synthesize and purify a library of over 450 polymers and polymer-polymer conjugates that could then be characterized by small-angle X-ray scattering (SAXS) and dynamic light scattering (DLS) to determine compactness and flexibility which are two important factors in optimizing ligand-receptor interactions for multivalent therapeutics. Interestingly, we identified two polymer-PEG conjugates that displayed similar features to that of traditionally folded protein bovine serum albumin (BSA). Since the polymer library had sufficient diversity, we also observed a relationship between polymer hydrophobicity, compactness, and flexibility. This initial effort illustrated the importance of HTE but also the space for rational and data-driven design of SCNPs.
The second specific aim is to rationally design and rapidly screen a library of SCNPs for hydrophobic oral drug formulation. In collaboration with the Preformulation Sciences group at Merck & Co., we designed a library of 25 random heteropolymers to inhibit precipitation of the poorly soluble model drug probucol. Since probucol crystallizes via hydrogen bonding interactions, our random heteropolymers contained one monomer with a hydroxyl side group, hydroxypropyl acrylate (HPA), along with a second hydrophobic monomer, either methyl acrylate (MA), butyl acrylate (BA), hexyl acrylate (HA), or cyclohexyl acrylate (CHA). Through a rapid screening assay, we found hydrophobicity plays a major role in achieving probucol supersaturation as heteropolymers containing BA and HA at intermediate mol% performed well. Meanwhile, side group geometry also is important since CHA-containing monomers (same level of hydrophobicity as HA) did not perform well compared to the simple homopolymer of HPA. A fascinating observation this combinatorial screening process provided was that the drug release kinetics associated with the BA- and HA-containing heteropolymers contrasted greatly compared to conventional polymer excipients hydroxypropyl methylcellulose acetate succinate (HPMCAS) and vinylpyrrolidone-vinyl acetate copolymer (PVP/VA).
The third specific aim combines combinatorial chemistry and HTE with data-driven design to demonstrate how implementing ML can aid in SCNP design. An initial library of over 1000 polymer backbones, polymer-PEG conjugates, and polymer-peptide conjugates was characterized by SAXS and DLS and then utilized to train an evidential neural network (ENN). This model displayed predictive capability for Porod exponent and radius of gyration (Rg), two values with implications for macromolecular compactness and flexibility. From a theoretical chemical space of over 500,000 polymer-peptide conjugates using the boundary conditions of the initial dataset, we obtained high confidence predictions for difficult to design compact SCNPs: Rg < 7 nm and Porod exponent from 3.6-4.0. Through our experimental validation, we remarkably found that 10/30 novel ML-proposed polymer conjugates displayed high compactness (Rg/Rh < 0.80) and low flexibility (Porod exponent > 3.0) comparable to BSA. This effort demonstrated that an emphasis on data-driven design can empower SCNP design for therapeutic applications and improve experimental efficiency.
In conclusion, this work represents a case study in how combinatorial chemistry, HTE, and data-driven design can be useful for nanoparticle structural design. This enabled us to uncover interesting relationships related to structural design and drug formulation, and it provided novel structures to target specific structural criteria. In the future, this methodology could be applied to define design criteria for therapeutic nanoparticles.
NotePh.D.
NoteIncludes bibliographical references
Genretheses
LanguageEnglish
CollectionSchool of Graduate Studies Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.