Masters thesis of engineering computational modelling of graphene oxide and amyloidogenic protein amylin

Computational Modelling of Graphene oxide and Amyloidogenic Protein Amylin Computational Modelling of Graphene Oxide and Amyloidogenic Protein Amylin A thesis submitted in fulfilment of the requiremen[.]

Computational Modelling of Graphene Oxide and Amyloidogenic Protein Amylin

A thesis submitted in fulfilment of the requirements for the degree of Master of Engineering

Enxi Peng Bachelor of Chemical & Materials Engineering (Hons)

University of Auckland

School of Engineering College of Science, Technology, Engineering and Maths

RMIT University

November 2020

i

Declaration of Candidature

I certify that except where due acknowledgement has been made, the work is that of the author alone; the work has not been submitted previously, in whole or in part, to qualify for any other academic award; the content of the thesis is the result of work which has been carried out since the official commencement date of the approved research program; any editorial work, paid or unpaid, carried out by a third party is acknowledged; and, ethics procedures and guidelines have been followed I acknowledge the support I have received for my research through the provision

of an Australian Government Research Training Program Scholarship

Enxi Peng

6/11/2020

I would also like to thank my fellow research group cohorts for providing the fun and stimulating discussions we have had together Thank you, Dr Andrew J Christofferson, Dr Adam Makarucha, Dr Mathew Penna, Dr George Yiapanis, Dr M Harunur Rashid, Dr Tu Le, Dr Patrick Charchar, Kamron Ley, Alan Bentvelzen, and Wenxuan Li

I acknowledge the support I have received for my research through the provision of an Australian Government Research Training Program Scholarship I also acknowledge the generous allocation of high-performance computational resources from the Australian National Computational Infrastructure (NCI), the Pawsey Supercomputing Centre, and Melbourne Bioinformatics

iii

Table of Contents

Declaration of Candidature i

Acknowledgments ii

1 Introduction 1

1.1 Overview 1

1.2 Proteins: Structure & Folding 1

1.3 Protein self-association & aggregation 6

1.3.1 History of amyloidogenic diseases 8

1.3.2 Structure formations of amyloidogenic fibrils 9

1.3.3 Islet amyloid polypeptide 12

1.4 Effects of Graphitic Nanomaterials on Amyloid Aggregation 12

1.4.1 Graphene and Graphite 12

1.4.2 Graphene Oxide 13

1.5 Project aims 17

2 Computational Modelling Techniques 18

2.1 Overview 18

2.2 Introduction 18

iv

2.3 Ab-initio Methods 20

2.4 Classical Molecular Dynamics 21

2.5 Conformational sampling 21

2.6 Replica Exchange with Solute Tempering 23

2.7 Simulation Procedures 25

3 Effects of forcefield and sampling method in all-atom simulations of inherently disordered proteins: Application to conformational preferences of human amylin 28

4 Effects of Size and Functionalisation on the Structure and Properties of Graphene Oxide Nanoflakes: An In Silico Investigation 41

5 Conclusions and Future Work 48

References 50

Appendix 54

Publication List 54

Peer-reviewed Publications 54

Conference Presentation 55

Supplementary Data 57

Effects of Forcefield and Sampling Method in All-atom Simulations of Inherently Disordered Proteins: Application to Conformational Preferences of Human Amylin 57

Comparing to BEMD and REMD 57

v

Free-Energy Calculations 59

Modified TIP3P Water Simulations 62

Effects of Size and Functionalisation on the Structure and Properties of Graphene Oxide Nanoflakes: An In Silico Investigation 63

vi

Abstract

Detailed understanding of the interactions between nanomaterials and biological molecules is crucial for the development of novel materials in modern medicine for targeted drug delivery and novel therapeutics On the other hand, the incorporation of various nanomaterials into everyday life is equally as progressive, which could lead to potential adverse implications on biological systems Recent studies have suggested that some carbon-based nanoparticles can promote, and others can inhibit fibril formation of amyloidogenic peptides and proteins These types of proteins can misfold and aggregate leading to the accumulation of insoluble fibril-like structures, which have been linked to diseases such as Alzheimer’s, atherosclerosis and type-II diabetes The mechanisms of interactions that render these nanomaterials as inhibiting or promoting of amyloid formation remain unclear However, with the aid of advanced computational resources, it is now possible to explore the conformational features and dynamical interactions of proteins with complex nanomaterials at atomistic details and time scales inaccessible by experiments In order to explore the interactions of proteins with nanomaterials it is important we have a detailed understanding of the dynamical behaviour and properties of the individual systems first With this in mind, theoretical computational approaches were utilised to investigate the structures and dynamics of the amyloidogenic protein amylin and Graphene Oxide (GO)

The thesis is organised as follows:

Chapter 1 comprises the literature review of protein structure, amyloid fibrils and of the effect

of nanomaterials on protein aggregation

Chapter 2 reviews the computational methods used to perform simulations and analysis Chapter 3 comprises my published article on benchmarking forcefields and sampling methods for disordered proteins using amylin as a case study

an introduction into proteins structures In section 1.3, a literature review on the current state

of understanding of protein self-assembly and aggregation is represented A detailed review of the effect of nanomaterials on biological systems is represented under section 1.4

1.2 Proteins: Structure & Folding

One crucial element of all organic life forms is presence of proteins, a biological macromolecule responsible for a wide array of roles inside living organisms The responsibilities of proteins include cell-signalling hormones, catalytic enzymes, immune responses, reproductive and metabolic cycles, as well as forming structural elements such as keratin or collagen

Proteins are polypeptide chain consisting of amino acid residues, linked together by peptide bonds Although there are 22 different amino acids throughout known life, only 20 of which are present in the genetic code Each of these amino acids consist of a central carbon atom (C) attached to an amino group (-NH2), a carboxyl group (-COOH), a hydrogen atom (-H) and finally

a unique side-chain group (-R) Due to the uniqueness of the side-chain group, each amino acid can be distinguished from each other, see Figure 1 Individual amino acids are joined together via a condensation reaction where they are joined between the amino and carboxyl group, producing a single water molecule from the reaction Almost all amino acids, with the exception

2

of Gly can form one of two enantiomers, which are mirror images of each other However, amino acids exist almost always in the L-configuration, while the D-configuration only exists in microorganisms and the cell walls of bacteria

Figure 1 Polypeptides are polymers composed of many amino acids linked together

by peptide bonds Peptide bond forms between carboxyl group of one amino acid and amino group of another, and it is a dehydration reaction (Image: useruploads.socratic.org)

There are 20 naturally occurring amino acids, where the side-chain group also attributes

to various properties to the amino acids, which can be divided into three groups Hydrophobic amino acids are made up of hydrophobic side-chains, which consists of: Alanine (Ala), Isoleucine (Ile), Leucine (Leu), Methionine (Met), Phenylalanine (Phe), Tryptophan (Trp), Tyrosine (Tyr), Proline (Pro) and Valine (Val) The next group is composed of charged residues, Arginine (Arg), Lysine (Lys), Aspartic Acid (Asp) and Glutamic Acid (Glu), and those with polar side-chain groups; Serine (Ser), Threonine (Thr), Asparagine (Asp), Glutamine (Gln), Cysteine (Cys), Histidine (His), and Asparagine (Asn) The last of the 20 amino acids is Glycine (Gly), which only has a hydrogen atom as the side-chain group, thus this amino acid possesses special properties that usually classifies it as a part of the hydrophobic group One particular amino acid group of interest is that with aromatic groups on its side-chains, which includes:

be found in section 1.3.3)

The process in which proteins are assembled is dictated by the genetic information encoded in an organism’s DNA This process is known as translation, where a set of three nucleotides, known as codons, are responsible for a specific amino acid The first step of translation is known as initiation, where the ribosome, responsible for binding transfer RNA (tRNA) anticodons to messenger RNA (mRNA) codons, assembles around the target mRNA to initiate the start codon The second stage is called elongation, where the tRNA transfers an amino acid to the next corresponding tRNA codon, as the ribosome translocates to the next mRNA codon, thus producing an amino acid chain The process concludes at the third stage known as termination, where a stop codon is reached, and the ribosome releases the assembled polypeptide After the protein is synthesised from the DNA sequence, the protein undergoes a process called post-translational modification to alter its physical and chemical properties As a result, the misfolding of certain proteins are widely accepted as the leading cause of a wide range of disease, which is one of the main aspects of this thesis The detailed discussions of protein folding, and leading diseases are discussed in sections 1.3

Proteins and their functional properties are determined via distinct level of dimensional structures, which are conveniently represented as four structural levels Illustrations of these structural levels are shown in Figure 2a The polypeptide chain made up

three-4

of amino acid sequences is known as the primary structure Local regions of a polypeptide chain can form regular secondary structures, such as α-helices and β-strands The α-helix secondary structural is commonly identified by its spiralling motif, whereas the β-strands tend

to self-associate or with other β-strands to form pleated β-sheets The stability of an α-helix is determined by the formation of parallel hydrogen bonds between the backbone of sequential amino acids within the polypeptide chain Three different types of α-helices can occur, which are: α-helix, 310-helix, and the α-helix, all due to the variances in the hydrogen bond coordination

of their backbones On the other hand, the β-strand secondary structures form pleated β-sheets

in either parallel or anti-parallel arrangements, shown in Figure 2a The tendencies for these strands to self-associate are directly correlated to the formation of amyloid fibrils, which are the known cause for a multitude of neurodegenerative and non-neuropathic disease However, Figure 2b [2] indicates that there can be several folding intermediate states all across the energy landscape of a particular protein The role of misfolding in aggregation diseases illustrates that proteins are highly at risk of the ‘black-pit’ of protein aggregation This ‘black-pit’ can readily outcompete folding once the scales are tipped towards such unfavourable intermolecular processes There is a delicate balance between functional requirements on proteins, and their susceptibility to competing misfolding derivatives Characteristics such as kinetically stable partial states, conformations with exposed interactive surfaces, or metastability required for function are all contributors towards the risk of protein misfolding [3]

β-Figure 2a Four levels of Protein Structure (Image: useruploads.socratic.org)

5

Figure 2b.Funnel diagram representation of the different energy landscapes of protein folding (a) A fast-folding protein indicated by the smooth folding landscape (b) A rugged energy landscape forming folding intermediates within kinetic traps

When individual secondary structures are folded into compact structures, this is known

as the tertiary structure, which gives rise to a proteins three-dimensional motif These secondary structures are held together by turns and loop regions, where a series of non-specific hydrophobic interactions reduces the overall solvent accessible surface area thus stabilising its conformation Specific interactions such as salt-bridges, disulphide bonds, and hydrogen bonding provide stabilising support to the protein structure When multiple folded proteins interact with each other and form a multi-subunit complex, this is known as a quaternary structure This multi-subunit complex is held together by the same specific interactions that confine proteins tertiary structures

It is imperative that the correct three-dimensional structure of a protein is maintained, where the resultant misfolded protein may become pathogenic An example of such misfolding are amyloid-β peptides, where the amyloidogenic species of this protein are responsible for Alzheimer’s’ disease [4] A detailed review of amyloidogenic proteins and disease is presented

in section 1.3

The capability in which proteins can fold de novo into their native conformations is a fundamental phenomenon crucial to the functionality of organic life It is paramount that the

6

understanding of protein folding should be developed to fully understand this biological process A major milestone to this field of study was achieved when Anfinsen et al proposed the theory that the native conformation of a protein is the most thermodynamically stable structure, which only depends on the polypeptide sequence and solution conditions, and not the kinetics of the folding [5] However, it is unfeasible to search for the native state by random search, as a 50-residue protein would require 350 configurations to be scanned, which could take years to complete Since then is was established that proteins achieve native conformation within the timeframes of microseconds to seconds, random searches are deemed ineffective, which was coined as the Levinthal’s paradox [6] However, it wasn’t long before computational and theoretical techniques were developed to investigate the underlying mechanisms of protein folding However, due to the sheer size of some functional protein 3D structures, even with the aid of advanced computational techniques, the process by which linear polypeptides fold into 3D functional structures remains poorly understood, and the pathway by which this process occurs eludes researchers till this day A review of current computational knowledge of amyloidogenic proteins of interest and its interactions with nanoparticles are discussed in section 1.4 and 1.5 The computational techniques that can be utilised to study the conformational landscape of the protein and its interactions with nanoparticles are detailed in Chapter 2

1.3 Protein self-association & aggregation

Even though all the information required for a protein to fold into its functional form is encoded within the amino acid sequence, several additional factors also play an important role in vivo, in determining its conformation Within a crowded cellular environment, newly born proteins face the challenge of finding the correct interaction in order to fold into their functional form It is not uncommon for nascent polypeptides to become trapped in metastable intermediate states, which are regulated by proteasomes and degraded, or refolded by chaperonins However, they

7

can misfold and self-associate resulting in amorphous clusters or elongated amyloid fibrils The accumulation of amyloid fibrils is the leading cause to a range of diseases in the human body, commonly known as amyloidosis A list of amyloidogenic diseases is compiled in Table 1, together with the amyloidogenic protein responsible and its native solution state conformation This form of non-communicable diseases (NCD) has a global presence and pose significant burdens on the socioeconomic infrastructures of the public health section in the 21st century This has resulted in a race to elucidate the fundamental understandings of protein aggregation mechanisms, with the goal of developing novel, targeted and effective drugs for amyloidogenic disorders

Table 1 A List of some amyloid diseases and their precursor polypeptides [7]

However, not all forms of protein misfolding are amyloid like It was highlighted in the comprehensive review by Gershenson et al that given the right conditions, all proteins can misfold, depending on its folding landscape [3] For example, the serpin superfamily which consists of ‘serine protease inhibitors’ juggles the sensitive balance between functional and folder-induced vulnerabilities Serpin misfolding can be attributed towards mutations which

8

results in its polymerization within the endoplasmic reticulum often mapped to structural regions important for conformational changes There is a strong overlap between serpin folding and functions, which suggest that some of these misfolding attributes are what allows serpin

to perform a range of roles [3]

1.3.1 History of amyloidogenic diseases

The earliest mentioning of amyloidosis can be traced back to Rudolf Virchow’s paper from

1854, who was the first to use the term “amyloid” for structural deposits found in human tissue specimens [8] A solution of iodine in water with hydrated sulphuric acid was used to stain the cellulose in the human specimen, which resulted in observing corpora amylacea in the ependymal and choroid plexus having a typical cellulose reaction to iodine Further testing was carried out on tissue samples by Virchow and Meckel, which exhibited what is known today as systemic amyloidosis, and the resulting observation was a similar reaction to iodine as that observed with corpora amylacea They concluded that these waxy deposits had a degenerative effect on the spleen, liver and kidney, with the “amyloids” present to be cellulose-like in nature When Friedreich and Kekulé conducted chemical analysis on the amyloid rich specimens of a spleen obtained from an amyloidosis patient, it was found that the deposit was in fact proteinaceous As a result, the earliest iodine staining method was replaced by metachromatic stains of crystal voilet, which was later replaced with the historical Congo red introduced by Benhold Originally widely used in the textile industry, Congo red became a useful tool in diagnosing amyloidosis due to its strong affinity to amyloid deposits, and further discoveries made by Divry and Florkin in 1927 [8] An observed double refraction or enhanced birefringence was present after staining amyloid deposits with Congo red, which was suggested due to the ordered arrangement of elongated Congo red molecules within the amyloid This suggested that amyloids were not amorphous as earlier studies have described, but rather organised structures The Congo red staining method was standardised by Puchtler et al and still remains

1.3.2 Structure formations of amyloidogenic fibrils

The structure of amyloid fibrils was investigated by Cohen and Cakins in 1959 to determine if amyloids possess an ordered sub-microscopic structure, as suggested by the birefringence of Congo red-stained amyloid deposits Their studies revealed that the translucent, structureless amyloid under light microscopy is a characteristic fibrillar ultrastructure when analysed using an electron microscope [10] This ground-breaking finding was then confirmed by a series of other studies [11-16] which gave rise to an in-depth knowledge of the specific organisational structuring of amyloid fibrils One conclusive result from these studies established that the starting amino acid sequence does not affect the

10

resultant amyloid fibril structure, as the final morphology of the fibrils were all very similar to one another This fibrillar structure consists of cross- structures with -strands perpendicular

to the fibril axis, with an overall structure resembling an unbranched filamentous structure only

a few nanometres in diameter but reaching several micrometres in length Ultimately, a mature fibril was observed to be formed from multiple protofilaments that twist around each other Significant strides in determining the structures of amyloid fibrils were made using experiment techniques in recent years The fibrillar structure of A(1-40), A (1-42), prion fibril,

2microglobulin and human IAPP have all been proposed using solid state NMR, and were proven to be consistent with molecular dynamics simulations [1, 17-20]

The exact process by which proteins aggregate into fibril is a complex process, involving competitive formation of amorphous species and fibrillar aggregates, which includes a range

of intermediate species, filamentous forms and conformational states It is believed that a series of distinct stages governs the fibril forming process that in general adheres to a

“nucleation dependent polymerisation” model, likening the process to crystallisation The nucleation point, or fibril seed, that kick starts the fibrillation process forms above a critical concentration, and the fibrils elongate by irreversibly binding monomers to their free ends The time it takes for monomers to convert into its fibrillar form can be experimentally measured using a ThT assay, where the fluorescence of the ThT marker corresponds to how many amyloid fibrils are formed An initial lag phase is present to allow the formation of a nucleus, which is then followed by a rapid growth phase by associating other monomers and/or oligomers with the nucleus As a result, fibril growth can be represented as a lag exponential growth curve, where the presence of a fibril seed will significantly shorten the lag phase if not abolish it all together Nevertheless, the processes for fibril formation all require a significant degree of conformational change

11

A growth process such as the fibrillogenesis of amyloids is an inherently complex one, including several complex stages; monomer activation, primary nucleation, secondary growth and elongation It is near impossible to for an analytical model to fully encompass this elaborate growth process As a result, a range of kinetic models were developed to elucidate the process

by which these fibrils are formed The list of models includes the previously discussed nucleation-dependant model, which describes the initial nucleation processes, the secondary growth process that explains the creations of new filaments, and finally the elongation process The many mechanisms of the primary nucleation-dependent model include monomer addition, coagulation and “lock and dock”, were all proposed to describe very first stage of fibril formation The addition of a single monomer to an established nucleus is referred to as monomer addition, which is thought to be a rudimentary process in amyloid fibril formation Experimental observations of end-to-end joining of two oligomers is known as coagulation, which is an important process if the system’s initial elongation is dominated by monomer addition On the other hand, the “lock and dock” mechanism relies on an initial “docking” interaction that enables subsequent “locking” of oligomers or fibrils onto the deposited monomer, whereas the “locking” process determines the kinetic rate of fibril formation

It is paramount to understand the underlying mechanisms of fibril formation starting from the monomeric structure, especially since the evidence suggests nonfibrillar intermediates being the most toxic species of amyloidosis Experimental characterisation of cytotoxicity of amyloid proteins such as IAPP, also known as amylin, have also suggested intermediate species were responsible for the death of pancreatic -cells [16, 18, 21-23] Therefore, in order to develop novel therapeutic techniques for the treatment of amyloidogenic diseases, it is important to obtain molecular detailed characterisations of the thermodynamics and kinetics of all conformational changes that link these different species Recent fields of research have been investigating the application of nanomaterials to act as novel therapeutics

12

in the treatment of amyloid diseases However, the lasting effects of nanomaterial deposits on human biology is yet to be elucidated A detailed section of nanomaterials in the biological environment is presented in section 1.4

1.3.3 Islet amyloid polypeptide

Amyloidogenic diseases are metabolic conformational diseases caused by misfolding and aggregation of soluble proteins into insoluble fibril like structures, which are hypothesised to cause cell damage and ultimately cell death Amylin or islet amyloid polypeptide (IAPP) is a 37-residue hormone co-secreted and co-expressed along with insulin by the pancreatic β-cells; and

is present in 95% of patients diagnosed with type-II diabetes The plaque-like behaviour of IAPP was first discovered by American physician Eugene Opie in 1901 While performing a post-mortem microscopic analysis of the pancreatic parenchymal cells from a diabetic patient, he discovered the presence of a glassy and translucent material [24] It was not until 1986 and

1987 that two separate groups managed to purify IAPP from human insulinoma and human islet amyloid [25] An in-depth review of IAPP will be presented in Chapter 3

1.4 Effects of Graphitic Nanomaterials on Amyloid Aggregation

1.4.1 Graphene and Graphite

The study of aggregation processes of peptides and proteins, through atomic force spectroscopy or scanning tunnelling microscopy, initially used graphite as a deposition substrate It was soon realised that graphite had also influenced the aggregation via both fibril kinetics and fibril morphologies Aggregation of amyloid- into -sheets were found to form crystallographic symmetries similar to that of the graphite surface From this symmetrical starting point, the proteins assemble into parallel fibrils with hydrophobic side chain pointed at the graphite surface, and hydrophilic side changes pointed towards the water [26-29]

13

On the other hand, specific surface effects of graphite demonstrated promotion of fibrillation， where -sheet fibrils assembled on graphite substrate but only oligomer or monomers were observed on mica This effect however is in direct contradiction of inhibitory effects of carbon nanotubes (CNTs), where the surface chemical composition is identical As a result, differences

in surface curvature was attributed to be the cause of this disparity between graphite and CNTs The perpendicular axis of the CNT nanoparticles could not induce fibrillation as the highly curved surfaced sequestered monomers & oligomers from solution, depleting its ability form larger elongated fibrils [30-32] Meanwhile, graphite nanoparticles provided an ideal surface for nucleation to start, and form into large elongated amyloid fibrils Interestingly enough, when graphite is oxidised to form graphene oxide, similar inhibitory effects were observed to that of CNTs, which leads on the main focus of this study

1.4.2 Graphene Oxide

The rise of graphitic nanomaterials has brought about an extensive research effort into the biomedical applications of graphene, including the development of targeted drug delivery and bio-sensing applications Recent work identified a potential for nanoparticles (including graphitic nanoparticles) to inhibit or promote amyloid fibril formation [28, 33-37] There are only few studies which have looked at the interactions between graphene oxide and amyloidogenic proteins One study for example, by Mahmoudi et al investigated the inhibition mechanism by graphene oxide of amyloid beta fibrillation [38] This experimental study had shown that GO and their protein-covered surface delay the amyloid beta fibrillation process by adsorption of amyloid monomers onto its surface This study also showed that by increasing the GO sheets

in vitro concentration from 10% to 100%, amyloid beta fibrillation is inhibited by slowing down

the kinetics of the aggregation These studies provided some grounds to further investigate the promising potential of GO to act as fibril inhibitor of amylin

14

Another literature report on simulating GO and protein interactions is the study of adsorption of GA modules onto graphene and graphene oxide This study looks into the adsorption mechanism of protein GA module (GA53 – G-related albumin-binding module, part

of a family of surface proteins of different bacterial species) onto graphene oxide, using atom molecular dynamics [39] It was found that the proteins strongly adsorbed to the GO surface, however the binding sites are not specific The main difference being that the secondary structure of GA53 is well preserved in protein-GO systems The functional groups, hydroxyl and epoxy groups increased the distance between protein and GO therefore weakening the vdW interactions between the two groups The binding affinities of Protein-GO are predominantly resulting from hydrogen bonding and electrostatic attractions In protein-graphene systems, the strong vdW force distorts the GA secondary structures Additionally, π-

all-π stacking between aromatic residues of GA with GO and graphene still exists; however, GO is the more biocompatible of the two nanomaterials and prevents secondary structure formation while promoting adsorption MD simulations are carried out using the GROMACS software package, with OPLS-AA as the primary forcefield, and chemical structures of GO are taken from the molecular formula of C10O1(OH)1(COOH)0.5 What is interesting is that the parameters for each of the functional groups of the GO sheet are taken from various other forcefields and combined into OPLS to allow GO to be simulated in GROMACS This approach was taken due

to the lack of rigorously tested forcefield parameters that illustrates the electrostatics and vdW forces of the epoxy groups in particular Using parameters from other forcefields for simulating the behavior of GO is one way to bypass the lack of a vigorously tested forcefield; however, it does have potential disadvantages when it comes to accuracy [40-43]

Continuing along the lines of building computational models of GO using combined parameters, Baweja et al has undertaken studies into the hydration patterns of graphene-based nanomaterials and its role in helical protein surface adsorption [44-46] This paper discusses

15

the importance of understanding the nanomaterial induced conformational changes of a positively charged cytoplasmic protein To model these interactions, the researchers used GO with deprotonated carboxyl groups mimicking a physiological pH of 7.4 and GO parameters taken from the previous study What differentiates this paper from the rest is the wide range of analysis techniques used to define the interactions between protein and nanomaterial Centre

of mass (COM) distance was used to determine the stability of binding and a gradual decrease

in COM distance between protein and graphene-related nanomaterials (GBNM) was present The COM distance at the end of a 2ns simulation period was 10, 8 and 7.7 angstroms for GO, reduced GO (rGO) and pristine graphene (PG) respectively On the other hand, RMSD and secondary structure analysis was used to evaluate the degree of conformational change in the protein when adsorbed onto nanomaterial surface Overall, PG induced the most significant changes to the protein relative to both GO and rGO, and that protein adsorbed onto GO showed the most stable conformation followed by rGO Furthermore, to analyse the hydration pattern

of these protein-nanomaterial systems, the solvent accessible surface area (SASA) of these systems were studied The results from this analysis indicated that SASA increased with respect to time and that unfolding of the protein on PG was shown by an increase in SASA Additionally, there was a considerable increase to the degree of hydrogen bonding between protein and water during its adsorption onto PG Even without the use of a well parameterised forcefield for simulating GO, this study was still able to provide sufficient insight into a protein’s behaviour when adsorbed onto a GBNM surface The alpha helical structure of the protein during GO-protein simulation is seen to be mostly preserved due to the hydration of GO and the lack of π-π interactions between GO and protein In addition, only positively charged amino acids and aromatics selectively bind with GO via H-bonding, therefore different surface functionalised GBNM can lead to different adsorption pathways

16

On the other hand, recent studies within our research group focused on the interactions between pristine graphene, C60 and carbon nanotubes on the aggregation propensity of apolipoprotein C-II [31, 32] Makarucha et al have shown that the dimensionality of graphitic surface is a key factor in inhibiting fibril formation In this case, the high curvature of C60 inhibited the formation of fibrils by causing increased mobility of the peptides However, graphene and the carbon nanotubes in the axial direction promoted peptide aggregation These studies have shed some light on the fundamental processes that occur during the early stages

of amyloid fibril formation

Furthermore, Guo et al employed classical MD simulations to investigate the adsorption

of β-sheet rich oligomers of IAPP peptides on carbon nanoparticles (graphene, nanotubes and C60) [47] This study found that the π-π stacking and hydrophobic interactions are different between peptides and various nanoparticles, and these subtle differences are due to the variations in curvature and contact area The fibrillation of IAPP22-28 may be inhibited at its early stage by graphene or nanotube reducing the potential to develop β-sheet formation However, this study did not take into account the effect of peptide concentration on their adsorption to the graphitic surface It is crucial to keep peptide concentration consistent across the systems,

as this factor will affect the adsorption pathway, thus leading to variations in the degree of secondary structures formed

Although experimental work had shown that various carbon-based nanoparticles could provide an inhibitory effect on amyloid propensity of human amylin, little is known regarding the fundamental interactions between these two systems Coupled with the fact that current computational studies of GO are limited by inadequate forcefield parameters, it can be clearly seen that a major gap in knowledge is present

17

1.5 Project aims

The purpose of this project is to utilise computational techniques to investigate the folding, misfolding and aggregation of human islet amyloid polypeptide in the presence of graphene oxide and its derivatives, as well as characterising the fundamental nanoscale interactions between the protein and the graphitic nanomaterials

Specific aims of this project are as follows:

1 To determine the optimal computational approach (protocols and parameters) that will provide the closest agreement with experimental data

2 To produce graphene oxide models that are in good agreement with experimental data

3 Determine the statistically most populated (preferred) conformation of amylin from the theoretically simulated ensemble of structures

4 Investigate the possibilities for amylin to form intermediate structures that could lead

to fibril formation

5 Investigate the forcefield effects on amylin behaviour and how this could impact the observed conformations of amylin

2.2 Introduction

Ever since the inception of digital electronic computers in the mid-20th century, computational power has grown exponentially following what is often referred to as Moore’s Law, in which the number of components within an integrated circuit doubles every 18 months Modern commercial central compute units (CPU) technologies have reached up to 19 billion transistor count across 32 cores, whilst modern graphics processing units (GPU) are reaching 28 billion transistor counts It has become increasingly efficient for computational modelling and simulations to take advantage of the raw processing power of modern computer architectures

19

Molecular modelling (MM) is the collective term of utilising computational resources to predict the behaviour of molecules by computing theoretical models Whilst experimental studies are extremely powerful at determining many properties of a biological system, they have limited capabilities for studying phenomena at an atomic resolution or short time scales under various environmental conditions On the other hand, computational modelling is able to model systems on the nanometre scale and time frames upwards of microseconds, or milliseconds

on specially designed hardware Not only can you explore time and length scales unreachable

by experimental techniques, MM methods has the ability to explore the time evolution of a controlled system at atomistic detail

The current knowledge landscape of computational modelling can be characterised into several distinct categories, Ab-initio (from first principles) and Density Functional Theory are the methods capable of the greatest accuracy, as they are calculated from first principles by considering the electronic structure of a system As a result, this method is computationally inefficient for modelling protein structures due to the sheer size of protein systems The next level of detail obtainable from MM are the semi-empirical methods, which allows larger systems

to be modelled at electronic level, calculating only the valence electrons and fitting certain parameters to experimental data This less computationally expensive method is mostly used for understanding bond formation and bond breaking, electron configuration and calculating force field parameters As systems sizes and time scale get larger, an atomistic force field method can be utilised This method is established upon the Born-Oppenheimer approximation where electronic movement can be eliminated from the Hamiltonian of the system, with only the nuclear variable to be calculated Only physical interactions are considered in molecular mechanics, as opposed to chemical interactions, which includes: bond stretching, angle bending, and dihedral torsions Dynamical information about a protein can be calculated using this method using only a fraction of the computing resources compared to ab-initio

20

calculations As for larger complexes and long timescale, coarse-graining is a method capable

of simulating such system sizes, which has gained significant traction in recent years Within a coarse-grain model, a small group of atoms are treated as a single interaction unit, whereas the dynamics of the system is governed by a simple force field Therefore, parameterising coarse-grain force fields are intrinsically difficult as a small number of parameters are responsible for the interactions of a range of diverse and complex systems This thesis employs molecular mechanics based methods to investigate nano-bio systems Detailed discussion of the principles behind this method is presented in section 2.5, 2.7 and 2.8

2.3 Ab-initio Methods

Ab-initio, the Latin term meaning “from the beginning”, is also the name given to the first principles approach to describe molecular systems, by calculating the Schrödinger equation to predetermined levels of accuracy This method was developed in the late 1920s by Douglas Hartree and Vladimir Fock which solves the first-principal equation

The Hartree-Fock approximation leads to a set of differentials equations, which represent the coordinates of single electrons In turn, molecular orbitals can be expressed as linear combinations of a finite set using Linear Combination of Atomic Orbitals, of a basis set function This principle states that by representing interactions as a linear combination of atomic orbitals, the many-body problem can be avoided Therefore, the energy calculated from an approximated true wavefunction will always be greater than that of the true energy By using the self-consistent-field procedure described above, the wave function can be improved to result in the lowest energy for molecular orbitals

An alternative approach to describing electron correlation will be to consider the sum of exchange and correlation energies of a uniform electron gas can be calculated by only knowing