ABSTRACT
Title of Dissertation: EXPLORING THE EFFECTS
OF PHYSIOLOGICAL ENVIRONMENT
ON AMYLOID AGGREGATION
Abhilash Sahoo
Doctor of Philosophy, 2022
Dissertation Directed by: Dr. Silvina Matysiak
Department of Bioengineering
Molecular level self-assembly/aggregation processes are common in biomolecular systems.
Specifically, aggregation of protein molecules results in formation of amyloid deposits, that
has been associated with neuronal dysfunction leading up to neurodegeneration. The protein
aggregation is often influenced by several external physiological features, which can modulate
this pathological process in a specific or non-specific manner. This thesis aims to elucidate
the role of such factors in amyloid aggregation in the context of neurodegeneration. As test
cases, we have focused on different fragments of Amyloid-? peptide and Huntingtin protein
and explored common interaction schemes in the presence of phospholipid membranes, solvated
glucose molecules and added trailing sequences.
Phospholipid membranes, composed of a heterogeneous distribution of lipid molecules,
serve as packaging envelopes in cellular systems. But several studies have suggested a role
of cellular membranes in abetting protein aggregation in neurodegenerative diseases. The first
section of this thesis explores A? 16-22 aggregation in the presence of membranes. Lipid
membranes have been shown to modulate peptide aggregation in a charge dependent manner
with anionic membranes promoting faster peptide aggregation into ordered fibrillar structures
compared to zwitterionic membranes. In this work, we evaluate the role of this electrostatic
membrane headgroup charge on A? 16-22 peptide aggregation with model lipid membranes
composed of POPC (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine) and POPS (1-palmitoyl-
2-oleoyl-sn-glycero-3-phosphoserine) lipids. Beyond, membrane charge, membrane?s physical
organization can also affect peptide-peptide and peptide-membrane interactions. Here, we have
curated the effects of applied surface-tension, as a proxy for membrane curvature, on peptide
fibrillation propensities.
Apart from ordered structures such as membranes, solvated small molecules are a large
class of molecules that can affect aggregation patterning by affecting peptides through both
specific and non-specific interactions. The second section of this thesis explores A? 16-22
aggregation in varying hyperglycemic conditions, to draw correlations between Alzheimer?s
disease and type 2 diabetes. Here, we discovered that the glucose prefers partitioning onto the
aggregate-water interface in a specific manner, leading to a loss in rotational entropy that propels
peptide aggregation.
In the final section, we discuss the case of pathological peptide aggregation in the case
of Huntington?s disease. Broadly, Huntinting protein?s N-terminal region which consists of 17-
residue N-terminal domain (N17) and the following Glutamine repeat tract (Poly-Q) are our
objects of interest and associated with pathological aggregation. The aggregation landscape of
N17 is analyzed in presence of added different lengths of trailing Poly-Q tract and the presence
of curved membranes.
We have approached our research through a computational lens using molecular dynamics
simulations. To address the relevant concerns of large spatio-temporal scales necessary to study
peptide aggregation systems with molecular simulations, we have developed a coarse-grained
forcefield (ProMPT: Protein Model with Polarizability and Transferability) that uses reduced
spatial resolution to accelerate phase-space exploration. The forcefield can capture secondary
and tertiary folding of protein structures with minimal constraints, and is transferable across
biomolecular systems without a need for re-parametrization.
My dissertation presents a holistic picture of peptide aggregation and various physiological
factors that affect it, with biomolecular simulation across multiple scales.
EXPLORING THE EFFECTS OF PHYSIOLOGICAL ENVIRONMENT
ON AMYLOID AGGREGATION
by
Abhilash Sahoo
Dissertation submitted to the Faculty of the Graduate School of the
University of Maryland, College Park in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
2022
Advisory Committee:
Dr. Silvina Matysiak, Chair/Advisor
Dr. Pratyush Tiwary
Dr. Jinwoo Lee
Dr. Yanxin Liu
Dr. Jeffery B. Klauda, Dean?s representative
? Copyright by
Abhilash Sahoo
2022
Acknowledgments
First, I would like to thank Dr. Silvina Matysiak for her supervision, patience and support
throughout my doctoral studies. She has helped me understand how to navigate academic and
scientific research, and I consider myself incredibly lucky to be a part of her lab for five years.
I am grateful to my committee members Dr. Jeffery Klauda, Dr. Pratyush Tiwary, Dr.
Yanxin Liu and Dr. Jinwoo Lee for their time and constructive feedback, that was really influential
in shaping my dissertation.
I would like to thank my research collaborators Pei-yin Lee, Neha Nanajkar and Dr. Hongcheng
Xu, for all your support and sticking with me over these years as a graduate student. Also, I am
grateful to other members of Dr. Matysiak?s lab ? Riya Samanta, Gregory Custer, Suhas Gotla,
Meenal Jain and Neel Sanghvi for all the discussions and overall making my graduate school
enjoyable.
My wife, Dr. Alisha Pradhan has been my psychological and intellectual support system
during these six years of graduate school. I am grateful to have her as my partner and completing
this Ph.D. journey together. Thank you for always motivating me, your unwavering belief, being
my sounding board and my best friend. I owe my deepest thanks to my parents and brother for
their unending love and support through graduate school as an international student. Thank you
for being there for me. Lastly, this journey would not have been possible without the wonderful
friends I made over these years ? Spandan, Pramod, Biswak, Vishnu, Akshita, Vikram, Utkarsh
and Teja, who made my time in college park memorable.
ii
Table of Contents
Acknowledgements ii
Table of Contents iii
List of Tables vi
List of Figures vii
List of Abbreviations xiii
Chapter 1: Introduction 1
1.1 Objective of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Effects of Membrane charge and structure on aggregation of Amyloid-
Beta (A?) peptide?s central hydrophobic core (A? 16-22) . . . . . . . . . 3
1.1.2 Effects of hyperglycemic conditions on aqueous aggregation of A? 16-22 7
1.1.3 Impact of membrane curvature and the presence of polyglutamine (QN)
repeats on aggregation behavior of Huntingtin protein?s N-terminal domain
(N17) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Outline of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Chapter 2: Water-Explicit Polarizable Coarse Grained Model - WEPCGM 13
2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Water-Explicit Polarizable Protein Model - WEPPROM . . . . . . . . . . . . . . 16
2.4 Water-Explicit Polarizable Membrane Model - WEPMEM . . . . . . . . . . . . 17
2.5 Validations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.6 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Chapter 3: Effects of membrane headgroup charge on Amyloid-? 16-22 aggregation 22
3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3.1 Peptide Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3.2 Lipid Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3.3 Simulation Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3.4 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
iii
3.4 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4.1 Rate of peptide aggregation . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4.2 Beta Sheet Content . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Chapter 4: Effects of Applied Surface-tension on Membrane-assisted A? Aggregation 43
4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3.1 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4.1 Impact of induced curvature on peptide aggregation . . . . . . . . . . . . 49
4.4.2 Effects of peptide aggregation on membrane structure . . . . . . . . . . . 52
4.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Chapter 5: Aggregation of A? 16-22 in Hyperglycemic Conditions 57
5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.2.1 Peptide and Glucose Model . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.2.2 Simulation Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.2.3 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.3.1 Impact of Glucose on A? 16-22 Aggregation . . . . . . . . . . . . . . . 63
5.3.2 Secondary Structure in Protein Aggregates . . . . . . . . . . . . . . . . . 65
5.3.3 Restricted rotation of interfacial glucose molecules . . . . . . . . . . . . 67
5.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
Chapter 6: Transferable and Polarizable Coarse grained model for Proteins - ProMPT 70
6.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
6.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.3.1 Non-Bonded Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . 78
6.3.2 Bonded Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
6.3.3 Simulation setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.3.4 Comparision with Atomistic Simulations - Replica Exchange with Solute
Tempering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.3.5 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.4 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.4.1 Simulations of Glycophorin-A and Mutants . . . . . . . . . . . . . . . . 101
6.4.2 Note on Computational Efficiency . . . . . . . . . . . . . . . . . . . . . 105
6.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Chapter 7: Effect of varying poly-Q tract and the presence of curved membranes on
aggregation of Huntingtin protein?s N-terminal domain 107
7.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
iv
7.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.3.1 Single Peptide (N17-Qn) simulations in aqueous solution . . . . . . . . . 115
7.3.2 Multi Peptide (N17-Qn) simulations in aqueous solution . . . . . . . . . 116
7.3.3 Single Peptide (N17) in presence of a planar membrane patch . . . . . . . 116
7.3.4 Multi Peptide (N17) simulations in presence of a curved membrane . . . 117
7.3.5 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
7.3.6 Area per lipid on membranes with varying curvature . . . . . . . . . . . 119
7.3.7 Fine-grained surface for density calculations . . . . . . . . . . . . . . . . 120
7.4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
7.4.1 Conformational landscape of N17-Qn in solution . . . . . . . . . . . . . 120
7.4.2 Aggregation of N17-Qn in solution . . . . . . . . . . . . . . . . . . . . . 124
7.4.3 Membrane interaction of a single N17 peptide . . . . . . . . . . . . . . . 128
7.4.4 Multi Peptide (N17) simulations in presence of a curved membrane . . . 135
7.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
Chapter 8: Thesis Summary 148
8.1 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
Appendix A: Results From Other Replica Simulations of N17 with Curved Membrane 159
Bibliography 161
v
List of Tables
2.1 Bond and angle parameters in WEPPRO model. BB: Backbone bead, D: dummy
particle, S1: Sidechain 1 bead, S2: Sidechain 2 bead. . . . . . . . . . . . . . . . 18
2.2 Non-bonded Lennard-Jones (LJ) interaction strengths in WEPCGM model. Unit
of interaction strength (?) is in kJ/mol. The radius (?) of all LJ interactions is 4.7A?. 18
3.1 Non-bonded Lennard-Jones (LJ) interaction strengths in WEPPRO model. Unit
of interaction strength (?) is in kJ/mol. The radius (?) of all LJ interactions is 4.7A?. 31
6.1 Charges and characteristic bonded potentials for dummy beads. Bond-length
is the length of the tether from the primary interaction-center to the charged
dummies. kangle is the spring constant preventing deviation of the angle between
charged dummies and the primary-interaction-center from 180 degrees. . . . . . . 76
6.2 Bonded interaction potentials between different primary CG interaction-centers.
BB: Backbone; S1: First Sidechain; S2: Second Sidechain . . . . . . . . . . . . 82
6.3 Bond lengths between primary coarse grained interaction sites. . . . . . . . . . . 83
6.4 Angular interaction potentials between different primary CG interaction-centers.
BB: Backbone; S1: First Sidechain; S2/S3/S4: Second/Third/Fourth Sidechain . . 83
6.5 Dihedral interaction potentials between different primary CG interaction-centers.
BB: Backbone; S1: First Sidechain; S2/S3/S4: Second/Third/Fourth Sidechain . . 83
6.6 Dihedral potential parameters for ?-helix and ?-sheet corresponding to equation
1 in the main text. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.7 Simulation setup for each protein. Sequence: KLVFFAE. . . . . . . . . . . . . . 89
vi
List of Figures
2.1 Spatio-temporal scales accessible to different biomolecular methods . . . . . . . 14
2.2 A schematic description of peptide coarse-grained model for K-L-V-F-F-A-E. . . 17
2.3 A schematic description of lipid coarse-grained model . . . . . . . . . . . . . . . 19
3.1 Cross-beta structure of A? 16-22 in solution. . . . . . . . . . . . . . . . . . . . 28
3.2 a) Variation of number of ?unabsorbed peptides? with time, averaged over two
replica-simulations. The variation of the number of partially absorbed aggregates
has been provided as inset. b,c,d) Different pathways for peptide absorption into
lipid bilayer. b) Single (monomeric) peptide absorption. c) Peptide absorption as
oligomeric aggregates. d) Peptide aggregation through dissociation and rearrangement
of partially absorbed aggregates. Coloring scheme: Light green beads - Sidechains
of Phenylalanines (F); Blue beads - Peptide backbones; Red region - Polar/charged
lipid headgroup; White region - Hydrophobic alkyl tails (Lipids). . . . . . . . . . 33
3.3 Residue-wise insertion of Backbone beads (BB) into different membranes. a)POPC
bilayer; b)POPS bilayer. The gray region describes the average location of bilayer
headgroup (PO4). The presented results have been averaged over both replica-
simulations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4 a,b) Last frame snapshot of peptide aggregates on two opposing leaflets of POPC
lipid membrane in simulation 1. c,d) Last frame snapshot of peptide aggregates
on two opposing leaflets of POPS lipid membrane in simulation 1. The red part in
this representation corresponds to polar headgroup and the white part corresponds
to hydrophobic tails. The blue connected beads represent peptide backbone. e)
Variation in the number of A? 16-22 aggregates over time, averaged over both
replica-simulations. Even connected components of size one (monomers) have
been designated as a single cluster. f) Integration of radial distribution function
between charged peptide sidechains (E/K-S2) and lipid headgroup (POPC:NC3/PO4,
POPS:CNO/PO4), averaged over both replica-simulations. . . . . . . . . . . . . 35
3.5 Variation of size of peptide aggregates with time. The colors of heatmap correspond
to frequency of particular sized aggregate. a)POPC bilayer; b)POPS bilayer. The
presented results have been averaged over both replica-simulations. . . . . . . . 36
vii
3.6 a) Time evolution of beta sheet fraction. b) Distribution of end-to-end length of
peptides over the last 200 ns. The gray line shows the end-to-end distance criteria
used to determine beta sheets. (inset)-single peptide representative snapshots
describing end-to-end lengths of peptides. Peptide backbone of A? 17-21 (LVFFA)
is presented in magenta, whereas residues K and E are represented by blue and
red respectively. The connected blue beads represent hydrophobic sidechains.
c) Density distribution of E-S2/K-S2 (POPC/POPS) along bilayer normal from
bilayer center over the last 200 ns of simulation time. The gray region describes
the average location of bilayer headgroup (PO4). d) Density distribution of F19-
S2/F20-S2 (POPC/POPS) beads along bilayer normal from bilayer center over
last 200 ns of simulation time. All the results presented here have been averaged
over all replica-simulations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.7 Distribution of peptide end-to-end distances over different atomistic simulations.
The end-to-end distance is defined as the distance between terminal nitrogen (N)
of K which is a part of peptide backbone and terminal carbon (C) of E which
is also a part of peptide backbone. Color scheme of VMD snapshots: Purple -
Peptide backbone, Yellow - F (Phenylalanine) . . . . . . . . . . . . . . . . . . . 38
3.8 a) Sanpshot of peptide aggregation on POPC bilayer at the end of extended
simulation. b) Snapshot of peptide aggregation on POPS bilayer at the end of
extended simulation. Coloring scheme of VMD snapshots: Light green beads -
Sidechains of Phenylalanines (F); Blue beads - Peptide backbones; Red region -
Polar lipid headgroup; White region - Hydrophobic alkyl tails (Lipids). Right-
Increase in size of clusters by addition of 48 new peptides and extension of
simulation for POPC (a) and POPS (b). The size of initial cluster increased due
to recruitment of peptides during 500 ns of extended simulation. . . . . . . . . . 40
4.1 Relationship between applied surface-tension and the area-per-lipid. Each point
denotes a simulation system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.2 Variation of peptide absorption and ordered aggregation with increasing surface-
tension. a) Absorption (Green) and ordered aggregation (Blue) among all the
peptides (in-solution + on-membrane). b) Ordered aggregation (Blue) among
peptides absorbed on the membrane only. These values are averaged over the last
200 ns of two independent replicates. . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3 (a) Hydrophobic solvent accessible surface-area of membranes in absence of
peptides. Snapshots of membranes with peptide aggregate. b - Simulation with
surface-tension of 71.5 dyne/cm. Coloring Scheme - Membrane components are
colored by their position along Z, from red to blue; Peptides: Magenta. c -
Simulation without surface-tension (Lateral view). Coloring scheme - Membranes:
Grey. d - Simulation without surface-tension (Top view). Coloring Scheme -
Membrane components are colored by their position along Z, from red to blue;
Peptides: Magenta; Hydrophobic groups: Lime . . . . . . . . . . . . . . . . . . 55
4.4 Mean squared deviation of a single peptide as a function of time-lag. . . . . . . . 56
viii
4.5 a - Lipid tail order with respect to interaction site at acyl tail, numbered starting
from the membrane interface. b - A snapshot of peptide aggregate on the membrane
for simulation with surface-tension of 71.5 dyne/cm. Coloring Scheme - Membrane
components are colored by their position along Z, from red to blue; Peptides:
Orange. c - Distribution of headgroup (PN-vector) tilt with-respect-to the bilayer
normal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.1 Geometry of the Glucose Molecule. The atomistic numbering is in black. B1:
Blue; B2: Green; B3: Orange . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.2 1a-d: Evolution of peptide aggregate sizes over time at varying concentration
of co-solvated glucose (1a) 0M; 1b) 1.98 mM; 1c) 9.8 mM and 1d) 19.8 mM)
1e-f: Structure of peptide aggregate (1e/1f - violet) and spatially close glucose
molecules (1f - orange) created from the 19.8 mM glucose simulation at the final
time-step . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.3 Evolution of ? sheet content over time. . . . . . . . . . . . . . . . . . . . . . . . 66
5.4 Relative enrichment of individual coarse-grained interaction site of glucose . . . 66
6.1 Schematic geometry of Martini polarizable water . . . . . . . . . . . . . . . . . 77
6.2 Schematic geometries of coarse-grained amino-acids . . . . . . . . . . . . . . . 77
6.3 Non-Bonded Interactions [POL-POL]. Refer to Fig. 6.2 for naming nomenclature 78
6.4 Non-Bonded Interactions [POL-HYD]. Refer to Fig. 6.2 for naming nomenclature 79
6.5 Non-Bonded Interactions [POL-Others]. Refer to Fig. 6.2 for naming nomenclature 79
6.6 Non-Bonded Interactions [HYD-HYD]. Refer to Fig. 6.2 for naming nomenclature 80
6.7 Non-Bonded Interactions [HYD-Other]. Refer to Fig. 6.2 for naming nomenclature 80
6.8 Non-Bonded Interactions [Other-Other]. Refer to Fig. 6.2 for naming nomenclature 81
6.9 BB(Previous amino acid)-BB-S1 tabulated angular potentials (Set 1) . . . . . . . 85
6.10 BB(Previous amino acid)-BB-S1 tabulated angular potentials (Set 2) . . . . . . . 86
6.11 BB(Previous amino acid)-BB-S1 tabulated angular potentials (Set 3) . . . . . . . 87
6.12 Secondary-structure specific dihedral potential used in the CG forcefield. The
tabulated potentials are fitted (for ?-helix and 3-10 helix) or derieved (?-sheet) to
capture maximum value in the dihedral probability distributions . . . . . . . . . 88
6.13 PMF for Trp-cage (a) with RMSD helix BB and native contact as the reaction
coordinates. PMF for Trpzip4 (b) with RMSD BB and native contact as the
reaction coordinates. Both PDB structure (left) and the representative structure
(right) from our model for Trp-cage (c) and Trpzip4 (d) are shown. Blue indicates
the specific secondary structure each protein exhibits. . . . . . . . . . . . . . . . 93
6.14 Free energy plot for Trp-cage at 290K from the REST simulations. The folded
basin has a free energy of 2.82 kT and the unfolded basin has a free energy of
3.98 kT. The ?G is estimated to be 1.17 kT (2.81 kJ/mol). . . . . . . . . . . . . 95
6.15 Free energy plot for Trpzip4 at 290K from the REST simulations. The folded
basin has a free energy of 2.12 kT and the unfolded basin has a free energy of
4.00 kT. The ?G is estimated to be 1.88 kT (4.53 kJ/mol). . . . . . . . . . . . . 96
6.16 PMF for villin with RMSD S1 and RMSD S2 as the reaction coordinates at
T?=0.52. The representative structure for each basin is shown as insets. . . . . . . 97
ix
6.17 RMSD BB time series for (a) WW-domain and (b) ?-?-? at T?=0.52 (blue) and
T?=0.82 (red). The PDB structure (left) and the representative structure (right)
are shown in (c) and (d) for WW-domain and ?-?-?, respectively. Blue indicates
the specific secondary structure each protein exhibits. . . . . . . . . . . . . . . . 99
6.18 Surface Plot of GB1 with a solvent probe of 1.4 A?. a- CG after 5 ns of simulation
starting from the folded state, b- PDB. The backbone is traced to show the tertiary
packing of the protein. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.19 Time series for the number of A? 16-22 peptide forming ?-sheets. An illustration
of ?-sheets aggregation is shown in the inset with yellow representing PHE residues.102
6.20 PMF for GpA with the average helical content and the number of BB contacts
as the reaction coordinates. The representative conformation is shown as inset-
figure. Color code: Thr (grey), Gly (green), Val (purple), Leu (orange). . . . . . . 103
6.21 Helical crossing angle between the two helices of Glycophorin A. The horizantal
lines reflect PDB values. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.22 The residue-residue contact maps for GpA . . . . . . . . . . . . . . . . . . . . . 104
7.1 Schematic of Huntingtin protein with focus on the N-terminus . . . . . . . . . . 110
7.2 A snapshot of the created curved membrane . . . . . . . . . . . . . . . . . . . . 118
7.3 Free energy landscape of a single N17 with varying Qs, in solution with asphericity
as the reaction coordinate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.4 a - Number of contacts between different fragments of the peptide. b - Number
of peptide-water contact per residue. The lighter shade is indicative of increase
in number. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
7.5 Reweighted contact (backbone CG interaction sites within 7 A?) map for different
peptides ? N17-7Q (a), N17-15Q (b), N17-35Q (c) and N17-45Q (d). The N17
and poly-Q regions are marked in green and black respectively. . . . . . . . . . . 123
7.6 Representative VMD snapshots at 300K for a-N17-7Q, b-N17-15Q, c-N17-35Q,
d-N17-45Q. Color scheme - N17 backbone: Red; Poly-Q backbone: cyan; PHE
sidechain: Orange . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
7.7 Contacts between different domains in the aggregate structure. a - Number of
contacts between N17-N17 (blue) and N17-Poly-Q (green). b - Number of GLN-
GLN contacts. c - Water solvation per residue of N17 (blue), poly-Q (green). d
- PHE-PHE contacts (blue) and PHE solvation (green). All the plots here are a
function of the length of polyglutamine tract. . . . . . . . . . . . . . . . . . . . . 126
7.8 Representative VMD snapshots of peptide aggregate for a-N17-7Q, b-N17-15Q,
c-N17-35Q, d-N17-45Q. Color scheme - N17 backbone: Red; Poly-Q backbone:
cyan; PHE sidechain: Orange . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
7.9 Re-weighted contact (backbone CG interaction sites within 7 A?) map for different
peptide aggregates ? N17-7Q (a), N17-15Q (b), N17-35Q (c) and N17-45Q (d).
The N17 and poly-Q regions are marked in green and black respectively. . . . . . 130
x
7.10 a, c, e - Boxplot for backbone dihedral angles, with median marked in orange
for different trajectory slices. The black horizontal line corresponds to 50.4
degrees. (a) 0-10 ns; (c)40-50 ns; (e)90-100 ns; b - A representative snapshot
corresponding to 0-10 ns; d - A representative snapshot corresponding to 40-50
ns; f - A representative snapshot corresponding to 90-100 ns Coloring scheme:
Magenta:peptide backbone; Orange: Phenylalanine; sidechain; Green:Other hydrophobic
groups sidechain; Cyan:Polar residue sidechains; Yellow: Lysine sidechains . . . 132
7.11 a, c - Boxplot for backbone dihedral angles, with median marked in orange for
different trajectory slices. The black horizontal line corresponds to 50.4 degrees.
(a) 140-150 ns; (c)190-200 ns; b - A representative snapshot corresponding to
140-150 ns; d - A representative snapshot corresponding to 190-200 ns Coloring
scheme: Magenta:peptide backbone; Orange: Phenylalanine; sidechain; Green:Other
hydrophobic groups sidechain; Cyan:Polar residue sidechains; Yellow: Lysine
sidechains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
7.12 Normal distance between peptide center-of-mass and the center-of-mass of lipid
headgroups. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
7.13 Distribution of helical-tilt with bilayer normal. Here the last 30 ns of trajectory
has been used for calculation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
7.14 Characterization of membrane-peptide interactions at 70-80 ns (a/b) and 150-160
ns (c/d) : a,c-Contact map between peptide and membrane (NC3: Choline; PO4:
Phosphate; GLE: Glycerol-Ester; Plus: Positive charged amino acid sidechain;
Minus: Negative charged amino acid sidechain; POL: Polarizable amino-acid
sidechains; Backbone: Peptide backbone; PHE: Phenylalanine sidechains; Hyd:
Other hydrophobic group sidechains) c,d-A representative snapshot of peptide
interacting with the membrane. Coloring scheme: blue/ochre/pink: membrane
headgroup; cyan:acyl tail; green: peptide backbone; red: Phenylalanine . . . . . . 137
7.15 Characterization of membrane-peptide interactions at 200-210 ns (a/b) and 330-
340 ns (c/d): a,b-Contact map between peptide and membrane (NC3: Choline;
PO4: Phosphate; GLE: Glycerol-Ester; Plus: Positive charged amino acid sidechain;
Minus: Negative charged amino acid sidechain; POL: Polarizable amino-acid
sidechains; Backbone: Peptide backbone; PHE: Phenylalanine sidechains; Hyd:
Other hydrophobic group sidechains) c,d-A representative snapshot of peptide
interacting with the membrane. Coloring scheme: blue/ochre/pink: membrane
headgroup; cyan:acyl tail; green: peptide backbone; red: Phenylalanine . . . . . . 138
7.16 Boxplot marking distribution of peptide backbone dihedrals a-Peptides absorbed
onto the membrane; b-Peptides as aggregates in solution . . . . . . . . . . . . . 140
7.17 Relative distance of different groups from the membrane surface (determined by
PO4). BB: Peptide Backbone; SC: Side-chain . . . . . . . . . . . . . . . . . . . 141
7.18 Snapshot of lipid density at 350 ns. a - Simulation with absorbed peptide. One
PHE sidechain per peptide marked with black. b - Control simulation without
peptides. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
7.19 Boxplot of area per lipid over simulation time for the simulation with peptides on
a curved membrane. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
7.20 Comparison of area-per-lipid between the simulation with peptide and the control
simulation without peptides. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
xi
7.21 VMD snapshot at 750 ns for a F11L/F17L simulation. Color scheme - peptide
backbone: Green; Leucines: Red; Membrane in surface representation: Black . . 144
7.22 Density of lipid groups with center-of-mass of absorbed peptides marked in black. 144
7.23 Boxplot of backbone dihedrals for a)peptides in contact with the membrane b)
peptides not in contact with the membrane. The black horizontal line at 50.4
degrees marks backbone dihedral to match alpha helix structure. . . . . . . . . . 145
A.1 A representative snapshot of peptide interacting with the membrane for replica-
simulation 1. Coloring scheme: blue/ochre/pink: membrane headgroup; cyan:acyl
tail; green: peptide backbone; red: Phenylalanine . . . . . . . . . . . . . . . . . 159
A.2 A representative snapshot of peptide interacting with the membrane for replica-
simulation 2. Coloring scheme: blue/ochre/pink: membrane headgroup; cyan:acyl
tail; green: peptide backbone; red: Phenylalanine . . . . . . . . . . . . . . . . . 160
xii
List of Abbreviations
Abeta Amyloid-Beta
AFM Atomic force microscopy
AWSEM Associative Memory, Water Mediated, Structure and Energy Model
CD Circular Dichorism
CG Coarse-Grained
DOPC 1,2-Dioleoyl-sn-glycero-3-phosphocholine
DOPG 1,2-dioleoyl-sn-glycero-3-phospho-(1?-rac-glycerol)
DPC Dodecyl Phosphocholine
EM Electron Microscopy
FG Fine Grained
GUV Giant Unilamellar Vesicles
htt Huntingtin
LUV Large Unilamellar Vesicle
MD Molecular Dynamics
N17 Nterminal-17
NMR Nuclear Magnetic Resonance
OPEP Optimized Potential for Efficient Protein structure prediction
Poly-P Poly-Proline
Poly-Q Poly-Glutamine
POPC 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine
POPS 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoserine
ProMPT Protein Model with Polarizability and Transferability
SUV Small Unilamellar Vesicle
TEM Transmission Electron Microscopy
TIRF Total Internal Reflection Fluorescence
UV Ultra Violet
WEPCGM Water Explicit Polarizable Coarse Grained Model
WEPMEM Water Explicit Polarizable Membrane Model
WEPPRO Water Explicit Polarizable Protein Model
xiii
Chapter 1: Introduction
1.1 Objective of Thesis
Peptide misfolding and accumulation of aberrant proteinaceous amyloid-like aggregates is
a recurrent theme in numerous diseases associated with neuronal dysfunction such as Alzheimer?s
disease (AD), Huntington?s disease (HD) and Parkinson?s disease (PD) [1]. In particular, with
progressive increase in the aging population, neurodegenerative diseases present a significant
current social and economic challenge.
In this thesis, I have focused on fragments of Amyloid-beta (A?) peptide associated with
Alzheimer?s disease and Huntingtin protein associated with Huntington?s disease as test cases to
understand interaction schemes that drive this self-association behavior. A detailed understanding
of the molecular mechanisms and pathological event pathways of peptide?s conformational changes
and self-association can aid towards development of rational therapeutics. But such aggregation
processes do not occur in isolation and are often affected by the presence of several physiological
factors such as biomembranes. Previous research has implicated the importance of environment
in protein folding and aggregation pathways [2?5]. Moreover, common pathways for amyloid
aggregation related cytotoxicity and inhibitory mechanisms involve biomolecular structures, such
as lipid membranes and the extracellular matrix [6?9]; and mutations/modifications. This thesis
explores the impact of membranes, varying concentration of glucose molecules and addition of
1
trailing lengths of glutamines on aggregation behavior of fragments of A? and Huntingtin protein.
While, wet-lab experiments are essential for our understanding of biological systems, they
can suffer from some specific disadvantages. Experimental investigations into structural features
of amyloid - environment interaction is limited due to extensive structural heterogeneity and
complex competing interactions. Moreover, the disordered nature of these peptides is highly
susceptible to local environmental alterations and generate experimental artifacts which often
remain unaccounted for, leading to controversial results. This has also led to diverging interpretations
to same/similar experimental results.
Computer simulations and molecular modeling, aided by statistical thermodynamics can
provide insights on how complex bio-molecular systems behave beyond what theory and experiments
can deliver separately. Molecular dynamics (MD) simulations involve analyzing the spatial and
temporal motion of atoms derived from a system?s model Hamiltonian (forcefield), to generate
both atomic-scale structural and kinetic insights. It can be particularly ideal for characterizing
transient peptide structures which are difficult to study experimentally and has been widely
used to evaluate heterogeneous ensemble of structures generated by amyloid peptides, with
appropriate controls for environmental factors [10?13]. While traditional atomistic simulations
provide higher resolution and more detailed insights about the peptide-based biomolecular systems,
the spatio-temporal scales to study conformational transitions and peptide/aggregate - environment
interactions cannot be reliably achieved by present-day computational machineries. This necessitates
the need for specialized tools to access such time scales and length scales. Towards this direction,
for advanced sampling protocols ? both biased and unbiased simulations have been developed.
Biased simulation methods such as metadynamics and umbrella sampling, improve exploration
of phase space along particular reaction coordinates, whereas unbiased methods such as replica
2
exchange (REMD, H-REMD) seek for a more general enhancement of sampling. But these
methods require significant computational resources and can be particularly unfeasible for large
spatiotemporal processes such as peptide aggregation [14?17]. On the other hand, due to fewer
number of particles ? resulting in lowered resolution and a smoother free-energy landscape,
coarse grained (CG) MD can provide a more holistic picture for such multi-agent phenomena
in larger biomolecular systems [18].
In my research, I have developed coarse-grained models and used molecular dynamics
with these CG models to curate how the presence of physiological features ? membranes, local
enhancement of cosolutes (glucose molecules) and added trailing polar sequences can affect
the structural properties and kinetics of peptide aggregation. This thesis address three primary
objectives.
? Explore the impact of membrane charge and membrane curvature on aggregation of Amyloid-
Beta (A?) peptide?s central hydrophobic core (A? 16-22).
? Understand the effects of hyperglycemic conditions on aqueous aggregation of A? CHC.
? Study the impact of membrane curvature and the presence of polyglutamine (Qn) repeats
on aggregation behavior of Huntingtin protein?s N-terminal domain (N17).
1.1.1 Effects of Membrane charge and structure on aggregation of Amyloid-
Beta (A?) peptide?s central hydrophobic core (A? 16-22)
In the first section of the thesis, I focused on how membranes can affect aggregation
patterning and kinetics of Amyloid-beta aggregation. Amyloid plaques and neurofibrillary tangles,
3
contributing to progressive cognitive decline have been established as hallmarks for Alzheimer?s
disease [1]. The amyloid cascade hypothesis, has been widely accepted by neuropathologists
as the primary model of AD pathogenesis. According to this hypothesis, oligomerization of
A? peptides initiates a cascade of events, culminating in neuronal dysfunction and dementia.
Pathogenic A? peptides are about 39?43 residue long intrinsically disordered peptide in aqueous
solution and ordered alpha helix rich structures in apolar environments, formed from successive
incisions by ?-secretase and ?-secretase, which can aggregate into ?-sheet rich aggregates.
Mammalian cell membranes are dynamic and complex microsystems of significant physiological
relevance. The lipid membranes are self-assembled biological structures composed of lipid
molecules ? amphiphilic molecule with a hydrophilic head and a hydrophobic alkyl tail. These
molecules are highly diverse in chemical specificity due to variations in the head and tail types
and therefore impart varying physico-chemical properties to membranes. Phospholipids are the
primary component of cellular membranes, marked by the presence of a phosphate group at the
?head?. The net charge on a lipid molecule has been shown to modulate membrane properties and
function. Therefore, although zwitterionic lipid molecules particularly 1-palmitoyl-2-oleoyl-sn-
glycero- 3-phosphocholine (POPC) forms a significant fraction of lipid molecules in mammalian
cells [19], anionic lipids such as 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-l-serine (POPS) [20,
21], which predominantly exist in the inner leaflet, plays a more significant role in cellular
signaling pathways and interactions with peptides. The production of A? peptides occurs in
a membranous environment, exposing the peptides to a number of lipid?peptide interactions.
Also, the perturbation of cellular membranes, followed by ion-dysregulation due to oligomeric
forms of A? peptides is hypothesized to be a central part of A? assisted AD pathology [22?
24]. Therefore, a mechanistic understanding of bio-mechanical interactions of A? peptides
4
with cellular membranes is necessary to gain insights into the amyloid cascade pathway. A?
peptides have been shown to exhibit varying aggregation patterns on lipid bilayers depending
on their structure and composition. CD and Thioflavin T assay studies of unilamellar vesicles
have revealed an accelerated aggregation of A? 16?28 peptides into ordered beta sheets on an
anionic bilayer (DPPG) as compared to a zwitterionic bilayer (DPPC) [25]. In addition, an AFM
experiment using supported bilayers has shown that the disruption of zwitterionic membranes
(DOPC) is higher than that of anionic membranes (DOPG) in presence of A? peptides [26].
Recent evidence from imaging studies using TEM, AFM and total internal reflection fluorescence
microscopy has suggested that small unilamellar vesicles (SUVs) with a larger curvature promotes
amyloid fibril formation when compared to large unilamellar vesicles [27, 28]. Therefore, the
interaction of A? peptides with cellular membrane can result in complex changes in energetics
and kinetics of structural transitions and can also result in significant membrane disruptions. The
interactions are highly heterogeneous with significant dependencies on membrane composition,
oligomer structure, peptide/lipid ratio and cellular environment.
Experimental characterization of peptide aggregates by X-ray diffraction has revealed common
structural features such as a cross beta sheet architecture [29, 30]. Structural studies of A?
peptides by solid state NMR, hydrogen?deuterium exchange and electron microscopy have also
shown the presence of similar cross beta sheet patterns [31]. The central hydrophobic core
(CHC), residues 17?21 (L?V?F?F?A) of the complete A?, is crucial for fibrillation [32?36].
In addition, solid state NMR studies have confirmed that A? 16?22 (K?L?V?F?F?A?E) is one
of the the smallest peptide sequences capable of forming highly ordered, stable beta sheet rich
fibrils at neutral pH [37]. Therefore studies of the structural and kinetic properties of a simpler
tailorable model peptide, A? 16?22, can provide a better understanding of the molecular forces
5
responsible for fibril formation/elongation [38?40].
A morphological characterization of A? oligomers is difficult due to its transient and
soluble nature. On the other hand, computational studies, particularly molecular dynamics (MD)
can be an ideal alternative to access this small time scale, transient behaviour. Atomistic simulations
often coupled with advanced sampling techniques have been extensively implemented to study
small-scale peptide oligomerization in solution. Some recent atomistic studies on peptide?lipid
interactions have investigated pre-formed membrane-inserted oligomers and the very initial phases
of oligomer?lipid interactions [11,13,41]. Due to high computational costs and sampling issues,
atomistic simulations have not been used to study peptide aggregation on lipid bilayers starting
from a solvated monomeric configuration. Coarse grained molecular dynamics (CG-MD), which
provides a reduced resolution description of a system and significantly improved sampling of a
protein conformational landscape is an effective tool to study complex systems with extended
spatio-temporal scales, specifically peptide aggregation starting from monomeric peptides.
With our in-house developed coarse grained model (WEPPRO/WEPCGM) we were able
to study the imapct of membrane charge through model membranes created with zwitterionic
(1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine: POPC) and anionic (1-palmitoyl-2-oleoyl-
sn-glycero-3-phospho-L-serine: POPS) membranes. We also curated the effects of membrane-
curvature (expressed through changes in applied surface-tension) on A? fragment?s aggregation
behavior.
6
1.1.2 Effects of hyperglycemic conditions on aqueous aggregation of A? 16-22
In the second section, I have focused on more disordered and non-specific interactions
associated with the presence of co-solvated sugar molecules. Several clinical and molecular
studies have outline pathological correlations between type II diabetes (T2D) and Alzheimer?s
disease [42?45]. Insulin resistance and improper sugar metabolism are the established outcomes
of T2D, which can result in increasing the blood sugar content (hyperglycemia). This has
engendered studies to understand this effect of hyperglycemia on Alzheimer?s disease. Recent
structural evidences have established that chemical crosslinks generated by glycation of lysines,
thereby preventing the peptide aggregates from dissolution as the primary correlation between
increased severity of Alzheimer?s in T2D patients [46?48]. But this idea was recently challenged
with experiments reporting a time-scale separation between increased aggregation in presence of
sugar molecules and formation of chemical crosslinks, which happen at a later point [49]. This
points to a possible alternate thermodynamic mechanism. In this section, we aim to investigate
this pathological correlation from a thermodynamic perspective with coarse-grained molecular
simulations.
1.1.3 Impact of membrane curvature and the presence of polyglutamine (QN)
repeats on aggregation behavior of Huntingtin protein?s N-terminal domain
(N17)
Another neurodegenerative disease that involves progressive amyloid deposition and associated
membrane disruption is Huntington?s disease [50?52]. The pathogenesis of dominantly inherited
7
Huntigton?s disease is linked to fibrillar nano-scale deposits of Huntingtin protein (htt). A mutant
htt gene (with multiple CAG repeats) encodes variants of htt protein with an anomalous expanded
homopolymeric Poly-Q sequence that aids the aggregation process. While the flanking amino
acid residues, particularly, the first 17 N-terminal amino acid residues (N17) modulate aggregation
behavior and lipid binding by formation of amphipathic alpha helix, the length of Poly-Q tract
directly participates in aggregation and generation of a variety of aggregate species ? oligomers
and larger fibrillar structures. Several reports indicate htt protein interacts with membrane, either
through intracellular vesicular transport, or by association with Endoplasmic reticulum and Golgi
apparatus [53]. In addition, the pathogenesis of Huntington?s disease is hypothesized to proceed
through mitochondrial dysfunction. But the membrane interactions of htt and polyQ deposits
have not been fully characterized.
Huntington?s disease is among a larger section of diseases marked by the presence of
expanded poly-glutamine tract [53?55]. While, some global common themes have been suggested
for structural details of N17, the stuctures at the poly-Q tract and the impact of poly-Q tract on
N17-poly-Q aggregation remains ill defined. As such, in this section, we focus on the impact
of poly-Q on the conformational landscape and aggregation of Huntingtin protein?s N-terminal
domain.
Moreover, similar to A?, both membranes and oligomers are affected by htt-membrane
association. Solution AFM studies have reported oligomeric and fibrillar deposits over mica
surface [56, 57]. Studies have demonstrated an alteration in htt alpha helical content in presence
of POPC and POPC/POPS SUVs [58]. AFM studies with Total Brain Lipid Extract (TBLE)
suggests local alterations of bilayer compressibility on interaction with htt oligomers [59]. Investigations
by Chaibva et al. suggested an enhanced peptide binding and aggregation propensities with
8
increase in membrane curvature [60].
The intrinsically disordered nature of N17, similar to A? peptides, in solution makes
experimental characterization difficult. The transience of peptide conformations has necessitated
the need for computational investigations. While scarce in number, atomistic molecular simulations
keep reiterating the importance of N17 in driving membrane interactions. Due to the associated
structural transitions and complexity at capturing accurate environmental effects (such as local
dielectrics), coarse-grained simulations of N17-membrane interaction has not been attempted yet.
A coarse grained forcefield capable of encoding environmental features and generating structural
transitions can be an essential tool to study N17?s behaviour in membranous environments.
The relation between neurodegeneration and membrane curvature has been established for
several peptide sequences ? Tau, A?, ?-synuclein and huntingtin [61, 62]. One recurring theme
among such peptides is the presence of a membrane binding amphipathic helical (AH) motif.
The AH motifs can bind to membranes and significantly retard peptide diffusion, which could
in turn promote peptide aggregation [63]. Due to shortness of its sequence, the AH motif of
the Huntingtin protein (Huntington?s disease) can be reliably used as a model to investigate
this behavior, and provide a general outline how neurodegenerative AHs can sense membrane
curvature. In addition, the amphipathic nature of this model N17 can be treated as a window into
the biophysical properties of similar amphipathic helical regions, often prevalent in neurodegenerative
peptides such as tau and ? synuclein [61].
9
1.2 Outline of Thesis
In chapter 2, I introduce the coarse grained model for proteins (Water Explicit Polarizable
PROtein Model: WEPPRO) and membrane (Water Explicit Polarizable MEMbrane Model: WEPMEM)
that has been developing in our lab over years. Here, I present an overview of the forcefield and
discuss its benefits and potential applications.
In chapter 3, I explored differential patterns of membrane-induced A? 16?22 (K?L?V?F?F?A?E)
aggregation from the microscopic perspective of molecular interactions. In agreement with
experimental observations, anionic POPS molecules promote extended configurations in A?
peptides that contribute towards faster emergence of ordered ?-sheet-rich peptide assemblies
compared to POPC, suggesting faster fibrillation. In addition, lower cumulative rates of peptide
aggregation in POPS due to higher peptide?lipid interactions and slower lipid diffusion result in
multiple distinct ordered peptide aggregates that can serve as nucleation seeds for subsequent A?
aggregation.
In chapter 4, I introduce the effects of POPC membrane curvature (using applied surface-
tension as proxy) on A? 16-22 aggregation. This work presents a mechanistic and morphological
overview of the relationships between the biomembrane local structure and organization, and A?
peptide aggregation.
In chapter 5, I adopt coarse-grained molecular dynamics simulations to investigate protein
aggregation behavior with A? 16-22 peptides in near-pathological hyperglycemia. Here, we
uncovered entropic effects that direct glucose concentration-induced increased aggregation of
peptides.
In chapter 6, I discuss the development of a new transferable CG forcefield ? ProMPT
10
with an explicit representation of the environment for accurate simulations with proteins. The
forcefield consists of a set of pseudo-atoms representing different chemical groups that can be
joined/associated together to create different biomolecular systems. This preserves the transferability
of the forcefield to multiple environments and simulation conditions. We have added electronic
polarization that can respond to environmental heterogeneity/fluctuations and couple it to protein?s
structural transitions. The non-bonded interactions are parametrized with physics-based features
such as solvation, and partitioning free energies determined by thermodynamic calculations and
matched with experiments and/or atomistic simulations. The bonded potentials are inferred from
corresponding distributions in non-redundant protein structure databases. We present validations
of the CG model with simulations of well-studied aqueous protein systems with specific protein
fold types- TRP-cage, TrpZip4, Villin, WW-domain and ?-?-?. We also explore the applications
of the forcefield to study aqueous aggregation of A? 16-22 and dimerization of Glycophorin A
in presence of Dodecylphosphocholine (DPC) micelles.
In chapter 7, we examine the impact of pathological glutamine repeats on peptide structure
and aggregation, with the coarse-grained model developed in chapter 6 ? ProMPT. Here, we
curated a list of structural features for N17 with different lengths of poly-glutamine repeats, both
as a single peptide, and the peptide in its aggregate form. Increase in the length of trailing
polyglutamine tract resulted in a reverse-micelle architecture with the aggregate core dominated
by fibrillar poly-glutamine tract, and a disruption of the traditional hydrophobic core formed by
bulky hydrophobic groups.
Moreover, we also studied the structural features of the POPC membrane and N17 peptide,
that drives the curvature sensing of the Huntingtin protein. Our simulations delineate gradual
progression of N17 from a dominantly unstructured peptide in solution, to a more structured
11
?-helical patch on the membrane. In presence of a curved membrane, we observed that N17
peptides preferentially interact with the curved region. Here, we noted that while the polar and
charged groups drive the initial membrane-peptide interaction, the membrane curvature sensing
is dominantly controlled by the bulky hydrophobic groups.
12
Chapter 2: Water-Explicit Polarizable Coarse Grained Model - WEPCGM
2.1 Overview
The application of classical molecular dynamics (MD) simulations at atomic resolution
(fine-grained level - FG), to most biomolecular processes, remains limited because of the associated
computational complexity of representing all the atoms. To address this, coarse-grained models
have been developed. In the past decade, the development of various coarse-grained models
has provided key insights into the driving forces in folding and aggregation. But, models in
literature are either implicit-solvent, non-transferable or cannot study structural transitions. Here,
we discuss the coarse-grained model that was developed by our lab to address these shortcomings.
2.2 Introduction
While, wet-lab experiments are essential for our understanding of biological systems, they
suffer from some disadvantages. Specifically for complex biological systems, wet-lab experiments
are complicated and isolating individual effects/interactions can be very difficult. This can be
crucial for biological systems with many inter-dependent entities. Moreover, experimental results
can have diverging interpretations and are susceptible to environmental fluctuations. As an
alternative, computer simulations can be used to explain and direct experimental research. It
13
can be used to characterize molecular biophysics with theoretical precision, while capturing the
impacts of all the interactions in the biological system. Classical molecular dynamics involves
direct numerical integration of Newton?s equations, aided by mathematical functions and parameters
? forcefield, that describe the dependence of system?s potential energy on individual atomic
positions to generate time evolution of molecules. The efficiency and accuracy of bio-molecular
simulations is dependent on mathematical forcefields and packaged molecular dynamics simulation
programs (GROMACS [?], CHARMM [64], NAMD [65] and DLPoly [66], etc.). Some of the
popular chemically specific peptide?lipid forcefield families are Assisted Model Building and
Energy Refinement [67, 68] (AMBER), Chemistry at HARvard Molecular Mechanics [69?71]
(CHARMM), GROningen MOlecular Simulation [72, 73] (GROMOS), Optimized Potential for
Liquid Simulations [74,75] (OPLS). Improvements in parallel computing architecture and use of
graphical processing units have enabled millisecond level atomistic simulations to study protein
folding and unfolding in an unbiased manner.
Figure 2.1: Spatio-temporal scales accessible to different biomolecular methods
14
Under rather typical experimental conditions, the time scale of biological events can be
orders of magnitude larger than the time scale achieved by classical molecular dynamics simulations(Fig.
2.1). To address this time-sale issue, many novel techniques have been developed. One such
technique ? spatial coarse-graining involves local averaging of atomic positions, which can
smooth out the free energy landscape, and provide computational speed-up. On the basis of
particle-based resolution, molecular dynamics simulations can be broadly classified into all-
atom/atomistic (AA), united-atom (UA) and coarse grained (CG). Coarse grained molecular
dynamics (CG-MD) simulations can be applied to access relevant length and timescales to study
biophysical features such as membrane remodeling, peptide folding and aggregation and membrane-
assisted peptide aggregate?s structural changes. Some novel and popular coarse grained forcefields
include PRIME-20 [76], AWESEM [77], OPEP [78] and MARTINI [79]. In current literature, the
transferable coarse grained forcefields are often either limited by the variations of biomolecule
species they can represent or by the ability to capture secondary/tertiary peptide structures from
primary sequence.
In this chapter, I elaborate on the coarse grained models for proteins (Water-Explicit Polarizable
Protein Model - WEPPROM) and lipid membrane (Water-Explicit Polarizable Membrane Model
- WEPMEM) that I have used in the upcoming chapters [80?82]. This family of coarse grained
forcefields (Water-Explicit Polarizable Coarse Grained Models - WEPCGM) address the current
shortcomings in the coarse-grained forcefields present in literature. These models have an explicit
representation of environment that allows studies with environment-effected biomolecular motion.
With the protein model, we can study structural transitions into secondary structures. Finally,
these coarse-grained models are transferable and do not require re-parametrizations in order to
study different biomolecular systems.
15
2.3 Water-Explicit Polarizable Protein Model - WEPPROM
Our protein model was first introduced by Ganesan et al. and has been going through
constant updates. The model can capture structural transitions from a primary sequence of amino-
acids without any added bias. The CG protein can consist of three types of beads - charged
(+/-), hydrophobic (H) and polar (P) mapped in an atomistic amino acid sequence (Fig. 2.2).
Each amino acid?s peptide backbone has been mapped into a single polarizable backbone bead
(BB) with structural polarization added through dummy positive/negative charges (BBp/BBm)
of equal absolute magnitude. These dummy charges are tethered to the central interaction site
(BB) through harmonic restraints. The dummies can interact with the environment through
electrostatic forces and that can result in an induced-dipole effect, adding directionality to peptide
backbone?backbone interactions, and generating secondary and super-secondary structures. The
side-chains are amino-acid length, hydrophobicity and charge specific. The current version of our
coarse-grained model does not account for chirality of protein backbones. The model has been
parametrized with Yesylevskyy et al. ?s [83] polarizable MARTINI water model. Please refer to
Ganesan et al. for more details on model parametrization and finer details of the model [80, 81].
A description of the non-bonded interactions between coarse-grained interaction sites is provided
in Table 2.2
In comparision to the previous implementation, the backbone dummy (D) angle (D-BB-D)
was modified from 0 degree to 180 degrees to mimic the structure of typical peptide bonds. The
modification increases the dipole moment of polarizable peptide backbone beads and stabilize the
formation of secondary structure as in agreement with quantum mechanical calculations . The
original BB-BB-S1 (Backbone-Backbone-first sidechain) angles were removed from the model
16
Figure 2.2: A schematic description of peptide coarse-grained model for K-L-V-F-F-A-E.
and replaced by non-bonded interactions between backbone beads to sidechain beads on adjacent
amino acids. Similar to the original implementation of this model, no external dihedral potential
has been used to maintain the secondary structure of the peptide.
2.4 Water-Explicit Polarizable Membrane Model - WEPMEM
Our lipid forcefield complements the protein model and uses same/similar interaction-
center types. Similar to the original MARTINI forcefield, lipids - POPC and POPS in our lipid
model ? adapted from WEPMEM [82, 84], are modelled by 13 CG beads each through a 4:1
mapping scheme (Fig. 2.3). The non-bonded interactions between interaction sites is presented in
Table 2.2. At the lipid-headgroup, phosphate (PO4) and choline (NC3) are mapped to a charged
17
Table 2.1: Bond and angle parameters in WEPPRO model. BB: Backbone bead, D: dummy
particle, S1: Sidechain 1 bead, S2: Sidechain 2 bead.
bonds Rbond (nm) Kbond (kJ mol?1 nm?2)
BB-BB 0.385 7500
BB-D 0.14 5000
BB-S1 0.25 5000
S1-S2 0.28 5000
angles ?0(deg) K ?1angle (kJ mol )
D-BB-D 180 7.2
BB-BB-BB 109 75
BB-S1-S2 (LYS, PHE) 151 25
BB-S1-S2 (GLU) 180 25
Table 2.2: Non-bonded Lennard-Jones (LJ) interaction strengths in WEPCGM model. Unit of
interaction strength (?) is in kJ/mol. The radius (?) of all LJ interactions is 4.7A?.
Beads BB (P5) H1 (C1) H2 (C3) C+ (Qd) C? (Qa) Water (POL)
BB (P5) 5.0 2.0 2.7 5.32 5.32 4.75
H1 (C1) 2.0 3.5 3.5 2.3 2.3 1.0
H2 (C3) 2.7 3.5 3.5 2.7 2.7 2.7
C+ (Qd) 5.32 2.3 2.7 3.5 4.0 5.0
C? (Qa) 5.32 2.3 2.7 4.0 3.5 5.0
Water (POL) 4.75 1.0 2.7 5.0 5.0 4.0
coarse grained bead, whereas glycerol-esters (GL1 and GL2) and serine (CNO) are represented
by polarizable beads with structural polarization generated through two dummy charges similar
to the peptide model. Lipid oleoyl tails are designed with five hydrophobic beads, whereas the
palmitoyl tails with four.
In comparision to previous implementation of WEPMEM, the LJ interactions between
peptide backbone bead type P5 (BB) and charged beads (Qd/Qa) were changed from 4 kJ/mol
to 5.32 kJ/mol to avoid overbinding. Also, if the interaction between hydrophobic beads and
water beads are too low, the hydrophobic beads tend to aggregate in the aqueous environment. To
maintain the intricate balance between apolar-water and apolar-apolar interactions, the interactions
between hydrophobic sidechain bead type C1 and water bead type POL were modified from 0.2
kJ/mol to 1.0 kJ/mol. Please refer to Ganesan et. al. for a more detailed validation of the
18
Figure 2.3: A schematic description of lipid coarse-grained model
forcefield and other model descriptions [81, 84]. As of now, we have representations for POPC
and POPS only.
2.5 Validations
Our membrane model was validated through several comparisions with atomistic simulations
and experimental reports [81, 82]. With simulations starting from a random distribution of
the lipid molecules in a simulation box, the lipid molecules conformed into well ordered lipid
bilayers, exhibiting reasonable descriptions of lipid behavior. The model WEPMEM membrane
matched the structural, dynamic and electrostatic properties, such as the cross-bilayer density
19
profile, area per lipid, bilayer thickness, mean squared displacement, line tension, dipole potential,
dielectric profile, head group orientation and formation of lipid clusters, compared to experiments
and/or all-atom simulations. Our model could also capture the right trend in membrane-interfacial
electric potential compared to the MARTINI model.
Similarly, our protein model could achieve realistic ?/? content with patterns created from
hydrophobic-hydrophilic residues [80]. This helped in charecterization of dipole?dipole and
dipole?charge interactions in shaping the secondary and supersecondary structure of proteins.
Formation of helix bundles and ?-strands. Moreover, the model could also achieve folding into
ordered ?-sheets for elastin like model octa-peptides in presence of a hydrophilic-hydrophobic
interface [85]. In another publication from the group, folding of anti-cancer peptide SVS1 was
studied in presence of model POPC and POPS membrane. Our model could reproduce the
expected folding of SVS1 in presence of anionic membrane into a ? hairpin structure [81].
2.6 Limitations
In the current form, WEPCGM suffers from several limitations. While the model has
been shown to reproduce common secondary structures of smaller peptides, it has not been
validated with larger protein structures with complex tertiary packing. In addition, we don?t have
representation for all amino acids with this model. We will address these limitations in the new
version of our CG model that we introduce in Chapter 6 ? Protein Model with Polarizability
and Transferability (ProMPT). Similarly, the repository of lipid molecules is also limited with
current representation for POPC and POPS only. Ongoing work in our lab aims to expand this
set to include POPG, POPE and sterols.
20
2.7 Conclusion
In this chapter, we introduce the coarse-grained models developed by our lab that I have
contributed to and applied in my research. The coarse-grained model features explicit representation
of the environment, that allows for studying environment assisted structural changes. The presence
of explicit solvent also aids in creating a transferable forcefield without a need for reparametrization
in order to study another biomolecular system. In the upcoming chapters we will use WEPPROM
in conjuction with WEPMEM to study the aggregation of a fragment A? peptide in presence of
the membrane and hyperglycemia.
21
Chapter 3: Effects of membrane headgroup charge on Amyloid-? 16-22 aggregation
3.1 Overview
This chapter is based on the author?s publication: Pathways of amyloid-beta absorption
and aggregation in a membranous environment. Abhilash Sahoo, Hongcheng Xu and Silvina
Matysiak. Physical Chemistry Chemical Physics, 2019.
Aggregation of misfolded oligomeric amyloid-beta (A?) peptides on lipid membranes has
been identified as a primary event in Alzheimer?s pathogenesis. However, the structural and
dynamical features of this membrane assisted A? aggregation have not been well characterized.
In this chapter, we explore differential patterns of membrane headgroup -induced A? 16?22
(K?L?V?F?F?A?E) aggregation from the microscopic perspective of molecular interactions.
Physics-based coarse-grained molecular dynamics simulations were employed to investigate the
effect of lipid headgroup charge ? zwitterionic (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine:
POPC) and anionic (1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine: POPS) ? on A? 16?22
peptide aggregation. Our analyses present an extensive overview of multiple pathways for peptide
absorption and biomechanical forces governing peptide folding and aggregation. In agreement
with experimental observations, anionic POPS molecules promote extended configurations in
A? peptides that contribute towards faster emergence of ordered ?-sheet-rich peptide assemblies
compared to POPC, suggesting faster fibrillation. In addition, lower cumulative rates of peptide
22
aggregation in POPS due to higher peptide?lipid interactions and slower lipid diffusion result in
multiple distinct ordered peptide aggregates that can serve as nucleation seeds for subsequent A?
aggregation. This study provides an in-silico assessment of experimentally observed aggregation
patterns, presents new morphological insights and highlights the importance of lipid headgroup
chemistry in modulating the peptide absorption and aggregation process.
3.2 Introduction
Aberrant aggregation of peptides and proteins on cellular membranes has been associated
with the pathogenesis of a number of neurodegenerative diseases such as Alzheimer?s (AD),
Parkinson?s (PD) and Huntington?s (HD) disease [86?88]. Alzheimer?s disease, characterized
by extra-cellular amyloid plaques [89?91] and intra-cellular neurofibrillary tangles [92, 93], is a
significant social challenge which has affected 5.7 million people in the United States [94]. The
amyloid cascade hypothesis for AD presents aggregation of a 39-43 residue long intrinsically
disordered peptide-amyloid beta (A?) as the trigger for a cascade of events culminating in neuronal
deaths and dementia [95, 96]. Multiple alloforms of A? peptides with an intrinsic tendency to
form fibrillar aggregates are created by successive excisions of Amyloid precursor proteins (APP)
by beta secratase in the endosomal pathway and gamma secretase in the plasma membrane
[97?99]. Recent evidences have implicated soluble, low molecular weight A? oligomers as
the primary cytotoxic agents, correlating strongly with cognitive defects [4, 5, 100, 101]. The
structural diversity of polymorphic A? oligomers contributes towards multiple pathways for A?
-induced neuronal toxicity [102].
A broad range of proteins and peptides, regardless of large variations in the amino acid
23
sequence, have been shown to form amyloid fibril at high concentrations [103?105]. Experimental
characterization of peptide aggregates through X-ray diffraction has revealed common structural
features such as the cross beta sheet architecture [30, 106]. Structural studies of A? peptides
through solid state NMR [31], hydrogen-deuterium exchange [107] and electron microscopy
[108?110] have also shown the presence of similar cross beta sheet patterns. The central hydrophobic
core (CHC), residues 17-21 (L-V-F-F-A) of the complete A? is crucial for fibrillation [32?36]. In
addition, solid state NMR studies have confirmed that A? 16-22 (K-L-V-F-F-A-E) is one of the
the smallest peptide sequence capable of forming highly ordered, stable beta sheet rich fibrils at
neutral pH [37]. Therefore studies of the structural and kinetic properties of a simpler tailorable
model peptide, A? 16-22, can provide a better understanding of the molecular forces responsible
for fibril formation/elongation [38?40].
The production of A? peptides occurs in a membranous environment, exposing the peptides
to a number of lipid-peptide interactions [22]. Also, the perturbation of cellular membranes,
followed by ion-dysregulation due to oligomeric forms of A? peptides is hypothesized to be
a central part of A? assisted AD pathology [22?24]. Therefore, a mechanistic understanding
of bio-mechanical interactions of A? peptides with cellular membranes is necessary to gain
insights on the A? cascade pathway. A? peptides have been shown to exhibit varying aggregation
patterns on lipid bilayers depending on their structure and composition [25,26,111?116]. CD and
Thioflavin T assay studies on unilamellar vesicles have revealed an accelerated aggregation of A?
16-28 peptides into ordered beta sheets on anionic bilayer (DPPG) as compared to zwitterionic
bilayer (DPPC) [25]. In addition, an AFM experiment using supported bilayers, has shown that
disruption of zwitterionic membranes (DOPC) is higher than anionic membranes (DOPG) in
presence of A? peptides [26]. Recent evidences from imaging studies using TEM, AFM and total
24
internal reflection fluorescence microscopy have suggested that small unilamellar vesicle (SUVs)
with a larger curvature promotes amyloid fibril formation when compared to large unilamellar
vesicles [27, 28].
A morphological characterization of A? oligomers is difficult due to its transient and
soluble nature [117?119]. On the other hand, computational studies, particularly molecular
dynamics (MD) can be an ideal alternative to access this small time scale, transient behaviour
[120]. Atomistic simulations, often coupled with advanced sampling techniques have been extensively
implemented to study small-scale peptide oligomerization in solution [121?128]. Some recent
atomistic studies on peptide-lipid interactions have investigated pre-formed membrane-inserted
oligomers and the very initial phases of oligomer-lipid interaction [11, 13, 41]. Due to high
computational costs and sampling issues, atomistic simulations have not been used to study
peptide aggregation on lipid bilayers starting from a solvated monomeric configuration. Coarse
grained molecular dynamics (CG-MD), which provides a reduced resolution description of a
system and significantly improved sampling of protein conformational landscape, is an effective
tool to study complex systems with extended spatio-temporal scales [129], specifically peptide
aggregation starting from monomeric peptides. Many novel coarse grained force-fields (In-
lattice and Off-lattice) have successfully characterized ordered amyloid aggregation [76, 130?
134]. PRIME 20, an intermediate resolution unbiased peptide coarse graining scheme has been
implemented with discontinuous molecular dynamics (DMD) on many amyloidogenic sequences
including A? peptides to study fibril formation in solution [135]. Zheng, et. al. explored the
aggregation free energy landscape of A? peptides using AWSEM-MD ? a predictive coarse
grained force-field [136]. A minimalistic, phenomenological model (Clafisch model) was implemented
with a variable dihedral term to generate peptide aggregation on model vesicles [131]. Recently,
25
another phenomenological coarse grained three bead-per-residue, amyloidogenic peptide model
(Shea model) was employed with implicit solvent to demonstrate spontaneous peptide aggregation
into beta sheets on model lipid membranes [132]. These coarse graining techniques, either have
been designed only for peptides (PRIME20-Hall model) or do not provide peptide sequence/lipid
type specificity (Clafisch and Shea models).
Here, we present a CG-MD study of partitioning, folding and aggregation dynamics of A?
16-22 peptides (K-L-V-F-F-A-E), in presence of model lipid bilayers composed with zwitterionic-
POPC (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine) and anionic-POPS (1-palmitoyl-2-
oleoyl-sn-glycero-3-phospho-L-serine) starting from their solvated monomeric state. We designed
the peptide using a modified version of prior-developed coarse grained model - Water-Explicit
Polarizable PROtein Model (WEPPROM) [80,81], which generated secondary structures of small
peptides from primary amino acid sequences without any built-in bias. Lipids were modeled
by another slightly altered variant of recently created Water-Explicit Polarizable MEMbrane
(WEPMEM) model that could accurately reproduce dielectric properties at the lipid-headgroup
(interface) region [81,84]. Physically, peptide aggregation involves an interplay between hydrophobic
and hydrophilic effects, indicating the importance of precise modeling of electrostatic interactions
[80, 85, 137]. The novelty of these models is the introduction of structural polarization at the
peptide-backbone and lipid-headgroup which is necessary for studying peptide-lipid interaction
and membrane induced peptide folding [81]. Both these models have been parameterized to use
Yesylevskyy?s polarizable water model [83].
Outer neuronal cell membranes are extremely diverse with a wide variety of glycerophospholipids
and ceramides [138]. A? peptides show an enhanced propensity to aggregate into ordered beta
sheets on negatively charged membranes composed with anionic lipids [25,137,139?142]. Also,
26
many experimental studies to understand the effect of lipid type on peptide aggregation have
used POPS as one of the anionic components [143?145]. This prompts our choice for using
POPS as the model anionic lipids to study the oligomerization of A? peptides in membraneous
environments. In this chapter we explore the differences in the morphology and kinetics of
membrane-assisted beta sheet formation by A? 16-22 on model lipid membranes using coarse
grained MD simulations. The chapter also presents a mechanistic explanation to experimentally
observed kinetic behaviour. To our knowledge, this is the first coarse grained simulation study
that captures the complex process of peptide aggregation in membraneous environment, from
peptides solvated in their monomeric state to formation of ordered secondary structures, while
maintaining peptide sequence and lipid type specificity.
3.3 Methods
3.3.1 Peptide Model
The A? peptide model is primarily derived from WEPPROM ? described in details in
Chapter 2 and previous publications [80?82]. The forcefield has been successfully applied to
study interfacial folding and aggregation of small peptides without any externally added bias
[80, 81]. The mapping scheme for the peptide A? 16-22 is provided as Fig. 2.2.
The peptide model was validated using experimental and computational evidences of A?
16-22 aggregation into ordered cross beta structures in aqueous solution [30,38,106,146]. Simulations
with 12 A? monomers, solvated in water (0.14 M) aggregated into stable beta sheet rich oligomers
(Fig. 3.1).
27
Figure 3.1: Cross-beta structure of A? 16-22 in solution.
3.3.2 Lipid Model
The lipid molecules in this work were modeled from WEPMEM forcefield, described in
detail in Chapter 1 and previous publications [81, 82, 84]. The forcefield could capture accurate
structural, dynamic and electrostatic features of the lipid membrane. Please refer to Fig. 2.3 for
details of the mapping scheme.
3.3.3 Simulation Protocol
The model POPC bilayer, composed with 240 lipids and POPS bilayer with 242 lipds
were solvated with CG water (lipid to water ratio fixed approximately at 20). The simulations
were performed on GROMACS 4.5.4 [147]. After the preliminary energy minimization, the
bilayers and counter-ions (in case of POPS bilayer) were equilibrated for 10 ns (time step, dt
= 10 fs) using NPT ensemble. Temperature was maintained at 300K through a Noose-Hoover
thermostat [148, 149] with a time constant of 1 ps. Parinello-Rahman barostat [150] with a time
28
constant of 1 ps and compressibility of 3 ? 10?5/bar was used alongside with semi-isotropic
pressure coupling to maintain a pressure of 1 bar. Particle mesh Ewald (PME) [151, 152] with a
relative dielectric constant of 2.5 and cutoff distance of 1.6 nm was used to compute long range
electrostatics. The Lennard-Jones interactions were modified starting from 0.9 nm to 0 at 1.2 nm
by the GROMACS shift scheme.
After creation of an equilibrated bilayer, 48 peptides (peptide to lipid ratio of about 0.2)
are randomly added into the solution. The peptide to lipid molar ratio of 1:5 has been previously
investigated in small (A? 25-35) peptide-membrane experiments [153]. The composite system is
then energy minimized and re-equilibrated for 50 ns with position restraints placed on 4th residue
backbone (on F-19 BB in x, y an z directions) in peptides and phosphate (on PO4 in z direction) in
lipids. The bilayers were simulated with a fixed surface tension (Berendsen barostat) to mimic an
enhanced area per lipid due to the outer membrane curvature of SUVs. The area per lipid of both
POPC and POPS bilayers were fixed at 95 A?2 to simulate outer membrane of a SUV with 13.4
nm diameter. These values were obtained through a WEPMEM simulation of a small unilamellar
vesicle system composed of 877 lipids and 61113 CG-water molecules and also verified against
other coarse grained molecular dynamics simulations with comparable vesicle sizes [154, 155].
All other simulation parameters were kept similar to the first equillibration step. This second step
allows us to equilibrate the monomeric peptides in the solution in presence of a lipid bilayer.
Finally, position restraints were removed and a production run of 1.5 ?s was carried out
using the remaining simulation parameters from the second equillibration step. To verify the
statistics presented in this chapter, we also simulated a replica of bilayer-peptide systems with
different initial states.
29
3.3.4 Analysis
Built-in functions of GROMACS, analysis modules of Visual Molecular Dynamics [156]
(VMD) and in-house developed scripts were used to analyze molecular simulation trajectory.
3.3.4.1 Peptide absorption
The relative position of two F-S2 beads from the lipid bilayer surface, described by locally
close (six nearest neighbors from center of mass) phosphate (PO4) beads has been used as
metrics to differentiate between absorbed and unabsorbed peptides. To accommodate for the
local curvature of lipid bilayers, the height of bilayer surface (h) is determined by the average
heights of six nearest PO4 beads to each individual peptide. The peptides are then categorized
into three classes - completely absorbed (CA), partially absorbed (PA) and unabsorbed (UA)
based on positions of Phenylalanine S2 (F-19/F-20 S2) on a single peptide chain. If the position
of both F-S2 along bilayer normal (z) is less or equal to h, the peptide is classified as completely
absorbed (CA) whereas if z > h, the peptides are considered unabsorbed (UA).
3.3.4.2 Peptide-aggregate Clusters
Peptide aggregation in the simulated systems was quantified by clustering connected peptides.
In this analysis, the peptides are described by main beads - BB, S1 and S2. This connectedness is
established using distance cutoffs determined from the first peak of radial distribution functions
(g(r)) between these beads. Two peptides are connected if they have at-least one pair of beads
within their respective cutoff distance. These cutoffs are detailed in the Table 1.
The number of peptide clusters, representing peptide aggregates at a particular time is
30
Table 3.1: Non-bonded Lennard-Jones (LJ) interaction strengths in WEPPRO model. Unit of
interaction strength (?) is in kJ/mol. The radius (?) of all LJ interactions is 4.7A?.
Interaction Cutoff (nm)
BB - BB 0.45
BB - S1 0.40
BB - S2 0.61
S1 - S1 0.65
S1 - S2 0.42
S2 - S2 0.70
computed by determining connected components. In addition, the peptide aggregates is also
categorized into three groups based on absorption classes of component peptides - completely
absorbed aggregates, partially absorbed aggregates and unabsorbed aggregates. A peptide-
aggregate has been listed as a completely absorbed aggregate, if all of its component peptides
can be classified as either completely absorbed or partially absorbed. On the other hand, an
aggregate is designated as a partially absorbed aggregate, if at-least one of its component peptide
can be classified as partially absorbed and at-least one another as unabsorbed. Finally, if all of
the component peptides are marked as unabsorbed, then the peptide-aggregate is categorized as
unabsorbed aggregate.
3.3.4.3 Beta Sheet Content
A backbone contact between two peptides is defined by alignment of back-bond dipoles
(BBm?BBp), characterized using a distance cut off of 2.5 nm between two oppositely charged
BB-dummies. This distance was determined from the interaction peak of radial distribution
function, g(r), between opposite charged dummy particles. We considered two peptides as
beta sheets, if they have at least five (71.5 %) such backbone-backbone contacts and end-to-end
distance greater than 1.2 nm, similar to beta sheet determination technique used by Lu et.al. [146].
31
The end-to-end length is the distance between the back-bone (BB) beads of flanking amino acids
- K and E. The fraction of peptides in mutual beta sheets constitute the total beta sheet content.
The relative position of two F-S2 beads from the lipid bilayer surface, described by locally
close (six nearest neighbors from center of mass) phosphate (PO4) beads has been used as
metrics to differentiate between absorbed and unabsorbed peptides. To accommodate for the
local curvature of lipid bilayers, the height of bilayer surface (h) is determined by the average
heights of six nearest PO4 beads to each individual peptide. The peptides are then categorized
into three classes - completely absorbed (CA), partially absorbed (PA) and unabsorbed (UA)
based on positions of Phenylalanine S2 (F-19/F-20 S2) on a single peptide chain. If the position
of both F-S2 along bilayer normal (z) is less or equal to h, the peptide is classified as completely
absorbed (CA) whereas if z > h, the peptides are considered unabsorbed (UA).
3.4 Results and discussion
Fig. 3.2a shows the number of unabsorbed peptides over time whereas Fig. 3.2a-inset
presents the variation in the number of partially absorbed aggregates with time. The peptides
can partition into lipid membranes as individual monomers (Fig. 3.2b), as small oligomeric
aggregates (Fig. 3.2c) or by slow dissociation of larger peptide aggregations (Fig. 3.2d) bound to
the lipid membrane. Previous AFM studies with full length A? peptide have also reported some
of these pathways ? absorption as monomers and oligomers into supported bilayers [26]. There
was a sudden significant decrease in the number of unabsorbed peptides within about first 15 ns
for both anionic-PS and zwitterionic-PC simulations (Fig. 3.2a). Following that, the absorption
slowed down due to formation of a number of partially absorbed aggregates. These aggregates
32
a b c
d
Figure 3.2: a) Variation of number of ?unabsorbed peptides? with time, averaged over two
replica-simulations. The variation of the number of partially absorbed aggregates has been
provided as inset. b,c,d) Different pathways for peptide absorption into lipid bilayer. b) Single
(monomeric) peptide absorption. c) Peptide absorption as oligomeric aggregates. d) Peptide
aggregation through dissociation and rearrangement of partially absorbed aggregates. Coloring
scheme: Light green beads - Sidechains of Phenylalanines (F); Blue beads - Peptide backbones;
Red region - Polar/charged lipid headgroup; White region - Hydrophobic alkyl tails (Lipids).
arranged themselves to protect the hydrophobic cores of their component peptides (Fig. 3.2d),
which afforded some stability and reduced absorption rates. Over time, these partially absorbed
aggregates continued to slowly lose peptides to the membrane and decrease in total number
and size. Particularly, in contrast to POPC where all inter-facial aggregates were completely
absorbed, one such aggregate on POPS rearranged and conformed into a stable layered beta sheet
structure (Fig. 3.2d) attached to the membrane, similar to structures observed in membrane-free
systems.
The backbones of the absorbed peptides had variable residue-wise insertion into a lipid
membrane modulated by their side chain hydrophobicity and relative position along the peptide
chain (Fig. 3.3). Throughout the 1.5 ?s simulation, the absorbed peptides remained close to
the bilayer headgroup region (Fig. 3.2d, S6). Once absorbed into the membrane, the peptides
(monomers and aggregates) start diffusing laterally on bilayer surface as shown in the supplementary
33
Figure 3.3: Residue-wise insertion of Backbone beads (BB) into different membranes. a)POPC
bilayer; b)POPS bilayer. The gray region describes the average location of bilayer headgroup
(PO4). The presented results have been averaged over both replica-simulations.
movie. In addition, as the peptides are pre-dominantly hydrophobic (71.43 %), the aggregation
on the membrane occurs primarily by pushing away polar/charged lipid heads thereby exposing
the hydrophobic tail region as is evident in Fig. 3.4a-d, where peptides have pre-dominantly
aggregated on top of the hydrophobic alkyl tails (white region).
3.4.1 Rate of peptide aggregation
The variation in cumulative number of peptide clusters (ordered + disordered) over time is
presented in Fig. 3.4e. Over time, due to continued aggregation, the number of peptide clusters
34
a b c
d e f
Figure 3.4: a,b) Last frame snapshot of peptide aggregates on two opposing leaflets of POPC
lipid membrane in simulation 1. c,d) Last frame snapshot of peptide aggregates on two opposing
leaflets of POPS lipid membrane in simulation 1. The red part in this representation corresponds
to polar headgroup and the white part corresponds to hydrophobic tails. The blue connected
beads represent peptide backbone. e) Variation in the number of A? 16-22 aggregates over time,
averaged over both replica-simulations. Even connected components of size one (monomers)
have been designated as a single cluster. f) Integration of radial distribution function between
charged peptide sidechains (E/K-S2) and lipid headgroup (POPC:NC3/PO4, POPS:CNO/PO4),
averaged over both replica-simulations.
on POPC steadily reduced in number (Fig. 3.4e) to about three clusters and increased in size
? number of peptides (Fig. 3.5). In contrast, on POPS membrane, the number of peptide
aggregates/clusters continued to be relatively high, featuring significant variations in aggregate
sizes, pointing to a comparatively slower aggregation rate. This disparity in aggregation patterns
can be explained through difference in diffusion rates of PC and PS. Lateral diffusion of lipids is
35
Figure 3.5: Variation of size of peptide aggregates with time. The colors of heatmap correspond
to frequency of particular sized aggregate. a)POPC bilayer; b)POPS bilayer. The presented
results have been averaged over both replica-simulations.
relatively slower in POPS with a lateral coarse grained diffusion constant of 0.0369 x 10?-5 cm2/s
as compared to 0.0899 x 10?-5 cm2/s in POPC. While coarse grained diffusion constant is not
directly comparable to experimental and atomistic results, it can capture relative trends. Similar
qualitative trend has also been observed in other reported values for lateral diffusion constants
[157]. The rigid headgroup of POPS due to intra-molecular pseudo-hydrogen bonds captured by
CNO dipole-dipole interactions in this coarse grained scheme, prevents faster diffusion for lipids.
This restricts the effective exclusion of lipid molecules [81], preventing peptide aggregation. In
addition, increased interaction between lipid head group (CNO/PO4) and peptides in POPS can
further result in better mixing of peptides and lipids, compared to POPC (Fig. 3.4f).
36
3.4.2 Beta Sheet Content
a b
c d
Figure 3.6: a) Time evolution of beta sheet fraction. b) Distribution of end-to-end length
of peptides over the last 200 ns. The gray line shows the end-to-end distance criteria used
to determine beta sheets. (inset)-single peptide representative snapshots describing end-to-end
lengths of peptides. Peptide backbone of A? 17-21 (LVFFA) is presented in magenta, whereas
residues K and E are represented by blue and red respectively. The connected blue beads
represent hydrophobic sidechains. c) Density distribution of E-S2/K-S2 (POPC/POPS) along
bilayer normal from bilayer center over the last 200 ns of simulation time. The gray region
describes the average location of bilayer headgroup (PO4). d) Density distribution of F19-
S2/F20-S2 (POPC/POPS) beads along bilayer normal from bilayer center over last 200 ns of
simulation time. All the results presented here have been averaged over all replica-simulations.
Over time, peptides rearranged into aggregates with high beta sheet content on both bilayers.
Fig. 3.6a presents time evolution of beta sheet content over time. Similar to reported CD and
37
Figure 3.7: Distribution of peptide end-to-end distances over different atomistic simulations. The
end-to-end distance is defined as the distance between terminal nitrogen (N) of K which is a part
of peptide backbone and terminal carbon (C) of E which is also a part of peptide backbone. Color
scheme of VMD snapshots: Purple - Peptide backbone, Yellow - F (Phenylalanine)
Thioflavin T assay studies [25] with A? 16-28, the beta sheet content in peptide aggregates was
significantly larger (approximately double) for POPS as compared to POPC. Even with larger
sized aggregates, peptides on PC bilayer were more unstructured. The higher beta sheet content
on PS bilayer can be explained by the distribution of end-to-end distance of peptides. Peptides
on PS are in general more elongated (Fig. 3.6b) which exposes their backbone to more peptide
backbone-backbone interactions, thereby increasing the overall beta sheet content. In agreement
with our coarse grained simulations, atomistic molecular dynamics simulations of single peptide-
membrane systems also reproduce a trend towards more elongated peptides in PS than PC (Fig.
3.7). This lower end-to-end distance of peptides in PC can be reasoned in terms of membrane
compressibility and peptide-lipid insertion. The higher compressibility of POPC membranes
as compared to POPS, that has been previously reported [158] and also captured by our CG
model [81], increases the relative penetration of peptides into POPC membranes. Fig. 3.6c
38
presents the distribution of F-S2, the ?highest inserted? side chain bead into the membrane. This
increased insertion of F-S2 on POPC membranes, coupled with charged residues at the ends
? K and E which prefer to stay close to bilayer surface, distorts the shape of the peptide into
a sharper ?U/O - like? shape (Fig. 3.6b) thereby decreasing the average end-to-end distances
and backbone-bone contacts. The higher membrane disruption for some zwitterionic lipids like
DOPC compared to anionic DOPG has also been recorded in AFM experiments by Hanes et.
al [26]. Moreover, similar to FTIR experiments on A? peptides in membranous environment by
Yu et. al. [159], a relatively higher insertion of F-19 as compared to F-20 is observed. In addition,
as apparent from Fig. 3.6d the sidechain of Glutamate (E-S2) in POPS is positioned slightly
away from the bilayer-headgroup as compared to POPC because of the interaction between
positively charged choline (NC3) and negatively charged E-S2 which is absent in POPS. This
further increases possibility of intra-peptide K-E interactions which results in a less expanded
peptide conformation.
In addition, due to higher number of total inter-peptide interactions in large amorphous
peptide aggregations on POPC, it is difficult for peptides to effectively rearrange into ordered
beta sheets necessary for efficient fibrillation. On the other hand, peptides on PS form numerous
slow diffusing smaller oligomeric aggregates which can reorganize comparatively easily into beta
sheet rich structures before coalescing into larger aggregates.
Oligomeric deposits on membranes can act as nucleation-seeds for subsequent peptide
aggregation from solution. To test this hypothesis, after the initial production run of 1.5 ? s,
we added 48 more peptides into both bilayer systems (with pre-existing peptide aggregates) and
recorded their dynamics for 500 ns. Similar to previous experimental observations [160], initial
pre-existing oligomers acted as nucleating sites, recruiting either individual monomers or smaller
39
Figure 3.8: a) Sanpshot of peptide aggregation on POPC bilayer at the end of extended
simulation. b) Snapshot of peptide aggregation on POPS bilayer at the end of extended
simulation. Coloring scheme of VMD snapshots: Light green beads - Sidechains of
Phenylalanines (F); Blue beads - Peptide backbones; Red region - Polar lipid headgroup; White
region - Hydrophobic alkyl tails (Lipids). Right-Increase in size of clusters by addition of 48
new peptides and extension of simulation for POPC (a) and POPS (b). The size of initial cluster
increased due to recruitment of peptides during 500 ns of extended simulation.
oligomers to create large fibrillar aggregates (3.8). This presents a possibility of multiple ordered
aggregates on POPS (larger than POPC) to act as nucleation sites that can progressively increase
subsequent aggregation rates. These fibrils are attached to the membranes primarily through
inter-peptide interactions with embedded peptides and extend into solution. The aggregate sizes
of fibrils in solution, attached to the bilayers is on average larger than oligomers lying flat on
membrane surfaces.
Studies with central hydrophobic core (CHC) provides insights about how structures develop
40
in peptide aggregates [161?163]. Faster development of ordered beta sheet rich structures in A?
16-22 can be corroborated with faster ordered fibrillation and resulting toxicity of full length
A? peptides in POPS. Similar to our observations, Lindberg et. al. [164] also found about
two fold increase in fibrillation of A? 1-42 in an anionic membrane (DOPS). This effect of
lipid headgroup chemistry has also been corroborated in previous work with surface plasmon
resonance and magnetic bead assay on A? 1-40 [165].
3.5 Conclusion
Our coarse grained molecular dynamics simulations reproduce and provide a mechanistic
explanation to a broad spectrum of experimental results [25,26,159,164,165]. We characterized
multiple pathways for peptide absorption into membranes composed of POPC and POPS. Both
lipid molecules have distinct effects on aggregation patterns of absorbed peptides. While rapid
cumulative aggregation (ordered + disordered) was observed in zwitterionic PC bilayer, the
emergence of ordered beta sheets and by extension, fibrillation was faster in presence of anionic
POPS lipids. The results are in agreement with previous experimental studies [25, 137, 139?
142, 164, 165] that had observed faster growth of amyloid fibrils in presence of anionic lipids.
The discrepancy in the cumulative aggregation rates is a consequence of faster lateral diffusion
of POPC lipid molecules compared to POPS. On the other hand, increased beta sheet content
in POPS membranes is due to the differences in membrane compressibility. Higher membrane
compressibility of POPC membrane compared to POPS, results in a relatively higher peptide
insertion into the bilayer. This distorts the geometry of individual peptide molecules which
hinders their participation in beta sheet formation. Some of the morphological aspects of membrane
41
assisted A? aggregation reported in this study such as relatively higher membrane insertion
of F19 compared to F20, have been supported by previous experimental evidences [159]. We
also revealed the propensity of initial oligomeric deposits to act as nucleation seeds to enhance
further fibrillation. Considering the presence of multiple, ordered aggregates in POPS due to slow
cumulative aggregation and peptide aggregates operating as nucleation seeds, POPS membranes
can have an increased progressive peptide aggregation rates. This work unravels how lipid
headgroup driven biochemical interactions in homogeneous model membranes shape peptide
absorption and aggregation.
42
Chapter 4: Effects of Applied Surface-tension on Membrane-assisted A? Aggregation
4.1 Overview
This chapter is based on the author?s publication: Effects of Applied Surface-tension on
Membrane-assisted A? Aggregation; Abhilash Sahoo, and Silvina Matysiak. Physical Chemistry
Chemical Physics, 2021.
In previous chapter, we focused on how lipid headgroups can impact A? aggregation.
Beyond, membrane charge, membrane?s physical organization can also affect peptide-peptide
and peptide-membrane interactions.
Studies connecting biological membrane organization and A? aggregation are limited due
to experimental and computational challenges. While experiments have suggested that an increased
membrane curvature results in faster A? peptide aggregation in the context of Alzheimer?s disease,
a mechanistic explanation for this relation is missing. In this work, we are leveraging molecular
simulations with a physics-based coarse grained model to address and understand relationships
between lipid packing in membranes and aggregation of a model template peptide A? 16-22. In
agreement with experimental results, our simulations also suggest a positive correlation between
increased peptide aggregation and lipid packing defects. More curved membranes have higher
lipid packing defects that engage peptide?s hydrophobic groups and promote faster diffusion
leading to the peptide?s fibrillar structures. In addition, we curated the effects of peptide aggregation
43
on membrane?s structure and organization. Interfacial peptide aggregation results in heterogeneous
headgroup-peptide interactions and an induced crowding effect at the lipid headgroup region,
leading to a more ordered headgroup region and disordered lipid-tails at the membrane core. This
work presents mechanistic and morphological overview of relationships between biomembrane?s
local structure and organization, and A? peptide aggregation.
4.2 Introduction
Alzheimer?s disease is characterized by extracellular self-assembly of A? protein segments
into ordered fibrillar aggregations [89?91]. Recent reports have pointed to amyloid-membrane
interactions as a crucial step in modulating the aggregation kinetics and associated disease pathogenesis
[22,111,116,166]. In particular, biophysical aspects of membrane organization such as membrane
order and charge can affect peptide aggregation kinetics. In addition, peptide aggregation can
also effect changes to membrane?s structural properties. Three experiment-derived hypothesis
detailing effects of peptide aggregation on membrane organization have been suggested ? carpeting
model, membrane-pore model and detergent model [167]. But the exact biophysical mechanism
characterizing this interaction is still missing. While significant research efforts have been applied
to understanding peptide aggregation in aqueous solvent, studies in presence of membranous
environments are rather limited. Research in this direction can impact our understanding of
event-pathways in Alzheimer?s disease.
Recent experimental reports have suggested that A? peptide can have differential aggregation
patterns when interacting with anionic membranes compared to zwitterionic/cationic ones [25,
26]. Here, peptide insertion and peptide aggregation in presence of zwitterionic membranes
44
were correlated to the increased membrane compressibility. Experimental studies have also
characterized the dynamic role of membrane curvature in driving the formation of and interaction
with amyloids [27, 28]. Peptide aggregation was studied with Phosphatidylcholine vesicles of
varying sizes using Thioflavin T fluorescence assay to track fibrillation. This study observed a
reduced lag time with smaller vesicles compared to larger ones. The authors also used isothermal
titration calorimetry to report endothermic binding between A? and small lamellar vesicles compared
to exothermic binding with large unilamellar vesicles. Such curvature associated modulations in
aggregation thermodynamics and kinetics has also been observed in the case of ?-Synuclein
(Parkinson?s disease) and Huntingtin protein (Huntington?s disease), suggesting a fundamental
nature in protein aggregate-membrane interactions [168,169]. This presents a need for biophysical
studies into membrane curvature/lipid-packing driven peptide aggregation.
While most studies of membrane-curvature induced peptide behaviour using atomistic
simulation have been limited to studies with a single peptide, some emerging work have reported
the correlated effects of membrane curvature and peptide aggregation [170, 171]. Physics-based
transferable coarse grained forcefields like MARTINI have enabled molecular simulations with
exact amino-acid sequences that have coupled protein aggregations to membrane-curvature [79,
172, 173]. But MARTINI involves restraining the secondary structure of each protein units,
thereby preventing studies of conformational transitions.
In this work, we studied the inter-dependent effects of membrane lipid packing and peptide
aggregation with 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) and A? 16-22. Here,
we have used POPC to create our model lipid membranes, as phosphatidylcholines are one of the
most abundant lipid molecules in mammalian cells and previous experimental studies have used
such membranes for peptide-aggregation studies [27].
45
4.3 Methods
The coarse grained simulation systems were created with Water-Explicit protein and membrane
model outlined in chapter 2 (Fig. 2.2, Fig. 2.3). Both the membrane and peptide models feature
local grouping of atoms to generate functional coarse-grained atom-types (CG-beads) that are
representative of the physics and chemistry of the atoms they represent. Previous research has
explored the use of these CG models to study membrane-induced peptide folding, calcium ion
driven lipid demixing and peptide aggregation in presence of membranes [81,82,84,85,85,174].
We validated our forcefield by simulations with A? 16-22 peptide in aqueous solution and in
presence of membranes composed of POPC and POPS lipids, where peptides were allowed
to interact with the membrane from solution and form ordered aggregates [82] (Described in
Chapter 3, Fig.3.1).
Similar to Terakawa et al. ?s experimental study on the effect of membrane-curvature
on peptide aggregation, our model lipid membranes are also composed of POPC lipids [27].
The initial frame was prepared with A? 16-22 peptides (with peptide to lipid ratio of 1:10,
similar to experimental systems outlined in Kandel et al. [153]) solvated randomly away from
the membrane. We instituted a hardcore repulsion with c6 and c12 terms set to 0 and 0.00247
respectively between lipid headgroups (phosphate and choline) in the lower leaflet and the peptide
backbone. This forces the peptides to exclusively interact with the upper leaflet. As previous
simulations have shown that this peptide fragment primarily stays at the membrane-water interface,
such applied repulsion between the lower leaflet and the peptide backbone would not affect the
peptide aggregation and adsorption properties. Each simulation system was composed with 242
POPC lipid molexules, 24 A? 16-22 peptides solvated in 5400 coarse grained water molecules.
46
The membranes were held with different values of surface-tension to simulate conditions with
varying curvature and membrane packing defects. Such an approach of using applied surface-
tension as proxies for membrane curvature has been previously used in atomistic simulations to
study membrane insertion of anti-microbial peptides [175]. Furthermore, A? 16-22 stays at the
membrane-interface, where membrane-curvature primarily manifests as an increased area-per-
lipid [82]. Fig. 4.1 shows the relationship between the area-per-lipid and the applied surface-
tension values in standalone simulations with POPC membranes in absence of peptides. The
area per lipid is computed simply by dividing the total membrane lateral area with the number
of lipid molecules in one leaflet. The highest surface-tension in our case corresponds to a small
unilamellar vesicle with a diameter of 13.4 nm, which was verified with a 100 ns molecular
simulation of complete vesicle (877 lipids) [82]. These values are similar to those reported
with coarse-grained and large all-atom simulations [154, 155, 176]. We also simulated a larger
membrane with 512 POPC lipid molecules, 50 A? 16-22 peptides and 9973 coarse-grained water
molecules at the highest value of surface-tension for 1.5 microseconds, to verify that our system
does not suffer from finite size effects. We note here that membrane curvature is a complex
process involving several structural features of lipid headgroup and tails. Here, we are presenting
a minimalist view of membrane curvature by focusing on the membrane packing defects arising
from increased area-per-lipid in membranes at high surface-tension.
The simulation protocol is identical to that mentioned in Sahoo et al. [82]. A brief overview
is presented here. The initial peptide-membrane system was created with peptides dispersed
randomly in solution, away from the membrane. The molecular system was then energy minimized
and equilibrated for a period of 50 ns, with position restraints on the peptides. After that, the
position restraints were removed and the peptides were allowed to interact among themselves and
47
Figure 4.1: Relationship between applied surface-tension and the area-per-lipid. Each point
denotes a simulation system.
with the membrane during the production phase of 1.5 microseconds. All the simulations were
performed with GROMACS 4.5.4 [177], with temperature maintained at 300K with a Nose-
Hoover thermostat [148] and pressure by Berendsen barostat [178]. Two independent replicas
were simulated for each simulation condition. The long range electrostatics is determined through
Particle-Mesh Ewald method with relative dielectric constant of 2.5.
4.3.1 Analysis
We have used visual molecular dynamics [156] and in-house developed scripts to analyze
our molecular simulations. Different geometric features of peptide and lipid organization was
used to define peptide absorption and ordered aggregation. We used the position of two second
sidechain beads (S2) of PHE to determine absorption of a peptide into the membrane. First,
six nearest lipid phosphate groups closest to the peptide center-of-mass were used to create a
local membrane plane for each peptide, thereby allowing accommodations for local membrane
undulations. Then, peptides were classified as absorbed if at least one of the PHE-S2 of the
peptides moved beyond that local plane in the z-direction. Here, we classified peptide aggregates
48
Figure 4.2: Variation of peptide absorption and ordered aggregation with increasing surface-
tension. a) Absorption (Green) and ordered aggregation (Blue) among all the peptides (in-
solution + on-membrane). b) Ordered aggregation (Blue) among peptides absorbed on the
membrane only. These values are averaged over the last 200 ns of two independent replicates.
as ?ordered? if each peptide in that aggregate participates in at least three backbone-driven dipole
orientations. Such dipole-dipole interactions can be construed as equivalent to hydrogen bonds
(as observed in atomistic representation of a ? sheet) in a coarse-grained sense.
4.4 Results and Discussion
4.4.1 Impact of induced curvature on peptide aggregation
To determine the effect of membrane curvature and lipid packing defects on peptide aggregation,
we characterized the structure of peptide aggregates in simulation with varying applied surface-
tension values. Fig. 4.2 records the overall behavior (in-solution + on-membrane) of peptide
aggregates, with measures for total ordered-aggregate fraction (Fig. 4.2a) and ordered-aggregate
fraction among the peptides absorbed on the membrane (Fig. 4.2b). Peptides can interact and
aggregate both in solution and on top of membranes. This competition between peptide-peptide
49
and peptide-membrane interaction shapes the peptide aggregation behavior. At surface-tension
values greater than 35 dyne/cm, peptide absorption progressively increases. This increase in
peptide absorption, in turn leads to an increased arrangement of the absorbed peptides into
ordered macro-structures on the membrane surface (Fig. 4.2b), which increases the overall
ordered content. It is worth noting that such increased fibrillation in more-curved membranes
are in line with experimental observation of increased beta-sheets in small unilamellar vesicles
compared to larger ones [27]. The computed fraction of absorbed peptides (1.0) and fraction
of total ordered aggregation (0.68, standard deviation 0.02) from a larger simulation system at
the highest surface-tension agrees with the values from the smaller system, suggesting no finite
size effects. Therefore all further analyses presented in this article have focused on the smaller
peptide-membrane systems.
To understand of this packing-induced peptide fibrillation, we used geometric and physical
metrics that quantify membrane?s structure and organization. Fig. 4.3a shows the distribution
of hydrophobic solvent-accessible-surface-area (HSASA) for membranes at different values of
surface-tension. An increased HSASA is indicative of more hydrophobic defects that can affect
membrane-peptide interactions. These evidences suggest, peptide absorption is driven by increased
hydrophobic defects in membranes with higher surface-tension. This is apparent in the molecular
snapshot (Fig. 4.3b) showing these peptide aggregations on top of an exposed hydrophobic patch
of the membrane at the highest surface-tension. Furthermore, increased area compressibility
of these membranes allow for easy rearrangement of lipid molecules to sustain larger absorbed
aggregates. Also, due to reduced crowding, peptide diffusion is faster in membranes with higher
surface-tension. To capture that, we simulated a single absorbed peptide in a membrane at several
different surface-tension values. Then we tracked the mean squared deviation of the peptide
50
center of mass with time lag (Fig. 4.4), which shows faster lateral diffusion with increasing
applied surface-tension.
Now, as we are working with a dominantly hydrophobic fragment of the A? peptide, the
hydrophobic side-chains of the absorbed peptides primarily interact with the membrane?s acyl
core. With increasing surface-tension values, these absorbed peptides with their side-chains
engaged can diffuse around faster. The interaction between the peptides are then driven by
backbone dipole alignments leading to ordered beta-sheet like structures. A parallel with Glycine-
Valine repeats at hexadecane-water interface can be drawn here, with fibrillation at the interface
region driven by buried hydrophobic sidechains into apolar media [80]. Therefore, this increased
fibrillation fraction at later surface-tension values can be attributed to the cumulative effects of
peptide absorption, higher membrane compressibility and faster peptide diffusion. It is worth
noting here that similar to earlier solid-state NMR experiments, our simulations also report
some ordered aggregations into cross-beta like structures in solution phase on membranes with
lower surface-tension values (Fig. 4.3 - c, d) [37]. Here the peptide aggregates primarily in
solution, tethered to the membrane (Fig. 4.3c). Such cross-beta structures are also noted for
our simulations in solution, in absence of any membranes (Chapter 3). But, the prevalence of
ordered aggregation is higher on ?more curved? membranes because of the specific topological
aspects (hydrophobic defects) that promote more absorption and backbone-driven inter-peptide
interactions.
51
4.4.2 Effects of peptide aggregation on membrane structure
Amyloid depositions can alternatively affect membrane?s structure which can in turn result
in functional changes. Several studies have reported A? assisted membrane damage with formation
of heterogeneous pores or detergent-like effects [166]. A correlation between membrane fluidity
and A? self-assembly was suggested in experiments using dynamic light scattering, fluorescence
and electron microscopy techniques [179]. This disruption of membranous environments is also
noted by Lemkul et al. for single A? peptide with atomistic simulations [180]. The authors
reported several structural impacts of A?-membrane interactions including local disordering of
lipid groups, and increased tail-tilts.
In this work, we also analyzed induced changes in membrane order and headgroup tilt due
to peptide aggregation. While the reported behavior exist in all simulation systems, we have
shown results from membranes with the highest surface-tension as that system has the highest
number of absorbed peptides, and therefore can be used to report statistically significant results.
The coarse-grained tail order parameter, 1 (3 ?cos2?? ? 1), quantifies disorder in acyl tails. Here ?
2
is the angle made by each bond vector in the acyl tail with the bilayer normal. The order parameter
varies from -0.5 to 1, reaching the maximum value when bond vectors are aligned parallel to
membrane normal. We classified lipid molecules in contact (any CG-bead within 7 angstroms)
from the peptide aggregate as ?close-to-aggregate? and others as ?away-from-aggregate?. Fig.
4.5a reports the tail order parameter at the interaction-sites of the acyl tail starting from the region
near the headgroups to the center of the membrane. As compared to lipid molecules that are away
from the aggregate, the lipid molecules closer to the aggregate are more ordered at the interfacial
region, and more disordered deeper inside the membrane. This observation can be explained in
52
terms of the location of the peptide aggregate along the Z-axis (Fig. 4.5b). Due to the presence of
peptide aggregates, the interfacial regions get more crowded, inducing increased order closer to
the aggregates (Fig. 4.5a). Farther inside the membrane, a hollow defect is generated by pushing
the headgroups away at the interface (Fig. 4.5b), which contributes to more disorder near the
bilayer center, closer to the aggregates. These observations align with the reported ?aggregate
induced order? in previous-mentioned experimental study [179]. We observed this order at the
interfacial region, and a more disordered acyl tail inside the membrane core as the short peptides
reside at the membrane interface.
Finally, we also looked at how the lipid headgroups behave closer to the aggregate as
compared to away from it. Differential arrangement of lipid headgroups can alter local membrane
potential and have implications in cellular signalling processes. The distribution of headgroup
(P?N ) tilt with respect to the bilayer normal is plotted in Fig 4.5c. P?N tilt is more dispersed
closer to the aggregate. This can be understood in terms of increased heterogeneity induced by
peptide-lipid interactions at the headgroups, closer to the aggregate.
4.5 Conclusions
Peptide aggregation into structured deposits have been associated with many neurodegenerative
diseases, particularly through interaction with cellular membranes. Experimental results have
suggested that both membranes and peptide can influence each other?s structure and dynamics.
Lipid packing defects associated with membrane curvature can promote extensive fibrillation,
with faster formation of ordered structures in presence of small unilamellar vesicles, compared to
larger ones. On the other hand, peptide aggregates can alter membrane?s structure and packing. In
53
this work we have investigated these intertwined effects with coarse-grained molecular simulations.
Our results agree with previous observations of curvature-driven peptide aggregation into ordered
structures, and suggest possible biophysical mechanisms for it [27,28]. Membranes with increased
curvatures lead to increased peptide absorption, due to more exposed hydrophobic defects. The
absorbed peptides can then laterally diffuse around, and interact through the peptide backbone
to form ordered fibrillar structures. The presence of such peptide aggregates also affects the
lipid membrane?s local structure. The lipid groups closer to the aggregates are more ordered
at the crowded interfacial region due to interfacial presence of peptides; and less ordered deep
inside the membrane core. These observations align with previous experiments and molecular
simulations that also highlighted membrane disruption by peptides [179,180]. Our results support
the previously suggested ?carpeting model? of membrane disruption, by increasing membrane
fluidity inside the membrane core [166, 181]. These locally close lipids also have a broad
distribution in P?N vector tilt from the membrane normal, suggesting the heterogeneous nature of
peptide-lipid interactions. Such local variations in P?N vector tilt can manifest in changes to the
membrane?s electrostatic potential, ion distributions and alter electrostatics-assisted membrane
signalling [182,183]. This study unravels the effects of lipid packing in aiding peptide aggregation,
and the effects of peptide aggregation in reshaping membrane?s local attributes.
54
Figure 4.3: (a) Hydrophobic solvent accessible surface-area of membranes in absence of peptides.
Snapshots of membranes with peptide aggregate. b - Simulation with surface-tension of 71.5
dyne/cm. Coloring Scheme - Membrane components are colored by their position along Z, from
red to blue; Peptides: Magenta. c - Simulation without surface-tension (Lateral view). Coloring
scheme - Membranes: Grey. d - Simulation without surface-tension (Top view). Coloring Scheme
- Membrane components are colored by their position along Z, from red to blue; Peptides:
Magenta; Hydrophobic groups: Lime
55
Figure 4.4: Mean squared deviation of a single peptide as a function of time-lag.
Figure 4.5: a - Lipid tail order with respect to interaction site at acyl tail, numbered starting from
the membrane interface. b - A snapshot of peptide aggregate on the membrane for simulation
with surface-tension of 71.5 dyne/cm. Coloring Scheme - Membrane components are colored by
their position along Z, from red to blue; Peptides: Orange. c - Distribution of headgroup (PN-
vector) tilt with-respect-to the bilayer normal.
56
Chapter 5: Aggregation of A? 16-22 in Hyperglycemic Conditions
5.1 Overview
Over chapter 3 and 4, we have been focusing on peptide membrane interactions and how
that affects specific details of peptide aggregation. Beyond cell membranes, there are several
other factors in the physiological environment that can shape peptide aggregation behavior. Co-
solutes are a large class of molecules that can affect aggregation patterning by affecting peptides
through both specific and non-specific interactions. In this chapter we explore how solvated
glucose molecules affect A? 16-22 aggregation in a concentration dependent manner.
Previous reports have drawn out correlations between pathogenesis of Alzheimer?s disease
and hyperglycemic conditions associated with type 2 diabetes. Previous research has proposed
chemical crosslinking, in presence of glucose molecules as responsible for increased toxicity. But
the time-scale separation between increased aggregation and formation of chemical cross-links
suggests the presence of an alternate thermodynamic mechanism, which we aim to explore in
this research. In this work, we apply coarse-grained molecular dynamics simulations to study the
how glucose molecules can drive A? 16-22 aggregation. Our analyses suggest that increasing the
concentration of solvated glucose molecules can result in faster aggregation of A? 16-22, without
any appreciable change in ?-sheet content. This change in aggregation rates can be explained in
terms of relative orientations of interfacial glucose molecules. The glucose molecules at the
57
peptide-water interface features a preferential orientation, resulting in loss of rotational entropy
in a concentration dependent manner. This can assist in faster aggregation of the peptide, to
reduce the cumulative availability of solvent accessible surface-area.
5.2 Introduction
Several studies have suggested pathological correlations and epidemiological linkages between
a very common metabolic disease, type II diabetes (T2D) with Alzheimer?s disease [42?45]. T2D
involves development of cellular insulin resistance, which leads to improper sugar metabolism
and high blood sugar concentration. Recent fMRI studies have suggested a declined cognitive
performance of patients suffering from T2D [45]. Another study has suggested that patients with
T2D have a 50% higher chance of suffering from AD [42]. But, the mechanistic event pathways
of this correlation is still not clear. While, numerous experimental and computational studies have
investigated A? conformational changes and aggregation in aqueous environments, investigations
into A?-sugar aggregate structures and sugar-induced peptide aggregation, at elevated levels of
physiological sugar concentrations in the case of T2D is limited [184].
Previous research efforts have postulated the post-translational glycation (Advanced glycation
end products - AGE) induced increased aggregation as possible pathological pathway correlating
AD and T2D [46?48]. These covalent modifications stabilize peptide aggregates against degradation,
linking it to neuronal dysfunction. But, in-vitro studies have reported that formation of AGE can
take upto a month with incubation in very high concentrations of glucose. At early stages of
A? aggregation formation of AGE is not reported [49]. Experiments with cellular culture has
revealed the upstream effects of glucose in production of A? peptides and preventing degradation
58
of Amyloid Precursor Protein. Glucose can also effect structural changes on peptide aggregates
and alter its interaction with physiological structures. Recent experiment by Kedia et. al. have
suggested that the presence of sugar molecules can lead to faster formation of toxic, unstructured
and membrane-active oligomers that can be taken up by the cells and can interfere with mitochondrial
activity [49].
The presence of co-solutes can modulate protein folding and peptide aggregation pathways,
prompting the use of simple saccharides as macromolecular crowding agents [185?187]. These
are often used to simulate cell-like crowded conditions in in-vitro experiments. Several experimental,
computational and theoretical studies have outlined the effect of such crowded conditions ?
from altering the thermodynamic phase space to modulating the kinetics of processes [188?
191]. This poses the question whether there are thermodynamic pathways that can explain
the behavior of A? peptides in hypoglycemic conditions, beyond the current understanding of
covalent modifications.
Computational studies on A? peptides in hyperglycemic conditions is limited. A recent
atomistic molecular dynamics with beyond-physiological ( 0.8 M) glucose and full-length A? 1-
42 revealed a caging effect of glucose molecules on peptide aggregates, which led to an increased
hydration of protein structures [192]. A complete biophysical picture of amyloid aggregation in
physiological/pathological hyperglycemic condition is missing.
In this work, we adopt coarse-grained molecular dynamics simulations to investigate protein
aggregation behavior with A? 16-22 peptides in near-pathological hyperglycemia. A? 16-22
peptides are recognized as one of the smallest fragments capable of forming fibrillar structures,
which has prompted research studies with this segment as templates [37?40]. With coarse-
graining, we reduce the number of interaction-centers in our simulation-system to access spatio-
59
temporal scales not typically accessible by traditional atomistic molecular dynamics simulations.
Here, we used the in-house developed coarse-grained protein model, Water-explicit Polarizable
Coarse-Grained Model (WEPCGM, Chapter 2) that can capture protein?s secondary structural
transitions starting from a primary sequence of amino acids, specifically in presence of external
stimulus. The coarse grained model can reproduce cross-beta like A? 16-22 aggregate structures
in aqueous solution, and ordered membrane-adsorbed beta sheet aggregates in presence of model
membranes (Chapter 3, Chapter 4). In this chapter, we present unique structural features of A?
16-22 and glucose co-aggregates and discuss mechanistic perspectives of glucose-accelerated A?
aggregation.
5.2.1 Peptide and Glucose Model
We refer readers to previous chapters [Chapter 2] and previous publications for details of
the CG model [80?82]. Please refer to the schematic figure 2.2 for details of the mapping scheme.
Figure 5.1: Geometry of the Glucose Molecule. The atomistic numbering is in black. B1: Blue;
B2: Green; B3: Orange
60
The geometry and cross-interactions terms for glucose molecules are borrowed directly
from the MARTINI model [Fig. 5.1] [193]. We verified the fidelity of the glucose coarse grained
model ? glucose-glucose and glucose-water interactions in polarizable water by computing
the osmotic second virial coefficient, B22, of the osmotic pressure that can be a measure of
deviation from ideality of solutions and matching with previous recorded values in experiments
and atomistic simulations. Here, we followed the simulation protocol outlined by Schmalhorst et
al.. A simulation box of (19)3nm3 was created with 420 randomly placed glucose molecules and
then solvated with Martini polarizable water [83]. Here, we followed an approximate strategy
to derive B22 using the radial-distribution-function, g(r) and cumulative number distribution
function N(r?) developed by Schmalhorst et al. [194]. We obtained an average B22 value of
0.119 (+/- 0.02) L/mol, close to the reported experimental B22 for glucose (0.117 L/mol).
Glucose molecules have carbon numbered from C1-C6 starting from the aldehyde (in open
conformation) group. Here carbon and the associated oxygen atoms for C1 and C2 are mapped
to CG interaction site B3; C3 and C4 mapped to B2; and C5 and C6 are mapped to B1. The sites
B1, B2 and B3 are all linked together forming a planar, triangular shape representing glucose in
its cyclic closed form.
5.2.2 Simulation Setup
The mixed systems ? glucose molecules, A? 16-22 and water molecules were created
by placing the groups randomly in a simulation box. We created three different compositions
(peptide-to-sugar ratios of 1.66, 0.33 and 0.16) for analyzing the effect of increasing sugar
concentration on A? aggregation. The concentration of glucose molecules were correspondingly
61
maintained at 1.98 mM, 9.8 mM and 19.8 mM respectively. The regular blood sugar concentration
in healthy individuals can range from 2.5-6.5 mM, and hyperglycemia associated with diabetes
has a concentration of 7-10 mM [195]. Emergency hospitalization is recommended for concentrations
higher than 16.7 mM.
After the preliminary energy minimization, the systems were equilibrated for 5 ns (time
step, dt = 10 fs) using NVT ensemble. Temperature was maintained at 300K through a Noose-
Hoover thermostat [148,149] with a time constant of 1 ps. Following this, production simulation
was run with constant-pressure ensemble. Parinello-Rahman barostat [150] with a time constant
of 1 ps and compressibility of 3? 10?5/bar was used alongside with isotropic pressure coupling
to maintain a pressure of 1 bar. Particle mesh Ewald (PME) [151, 152] with a relative dielectric
constant of 2.5 and cutoff distance of 1.6 nm was used to compute long range electrostatics. The
Lennard-Jones interactions were modified starting from 0.9 nm to 0 at 1.2 nm by the GROMACS
shift scheme.
5.2.3 Analysis
To calculate peptide aggregate sizes, we used an interaction cut-off of 7 A? from any CG
interaction site to determine if a peptide is part of the aggregate. We also quantified the structure
of these peptide aggregates through a ?-sheet metric. Peptides were considered as part of ? sheet
if greater than 57 % of the backbone dipoles aligned. These explicit dipole-dipole interactions
can be used as surrogates in our CG representation for hydrogen-bonds between peptides in ?
sheets. This analysis metric is specific to our coarse-grained forcefield, and we have used similar
metric in our previous publications to quantify ordered peptide aggregation.
62
Finally we also report the relative enrichment of glucose molecules as a function of distance
from the peptide aggregate.
ni(r) [Ni +Now]
Pi(r) =
Ni [ni(r) + now(r)]
Here, Ni and Now are the total number of groups of species i and water molecules respectively.
This metric has been applied to study solvation effect of proteins and protein aggregation [196?
199]. Here, we compared this relative enrichment of different parts of the glucose molecule (B1,
B2 and B3) near the peptide aggregate.
5.3 Results and Discussion
5.3.1 Impact of Glucose on A? 16-22 Aggregation
To characterize the effect of glucose on peptide aggregation, we analyzed the peptide
aggregate size in simulations with varying glucose concentrations. Fig.5.2 a-d shows the time-
series of aggregate sizes with different near-physiological sugar concentrations. In most (except
one replica for no glucose case) of the simulations, we recorded aggregation into larger aggregates
with all peptides involved (Snapshots - Fig. 5.2e, f). The scatter plots (Fig. 5.2a-d) have data
aggregated from two replica simulations for each peptide-sugar ratio. In our simulations, due
to high concentration of A? peptides ( 33mM), we do not observe the expected lag phase in
self-assembly processes, rather there is a fast initial evolution of aggregate sizes in all simulation
systems. We observed a concentration dependence in peptide aggregation kinetics, with faster
complete aggregation at increased sugar content and slower evolution for reduced concentrations.
63
Figure 5.2: 1a-d: Evolution of peptide aggregate sizes over time at varying concentration of co-
solvated glucose (1a) 0M; 1b) 1.98 mM; 1c) 9.8 mM and 1d) 19.8 mM) 1e-f: Structure of peptide
aggregate (1e/1f - violet) and spatially close glucose molecules (1f - orange) created from the
19.8 mM glucose simulation at the final time-step
64
This correlations between increased hyperglycemia and A? fibrillation has been previously noted
in experiments by Kedia et al. [49].
The glucose molecules were predominantly concentrated at the interface of the aggregate,
forming a cage-like structure at about 5-7 A? distance, with fast glucose exchanges between
interface and the bulk-solvent (Fig.5.2 f), similar to the organization suggested by Menon et al. in
their atomistic study with A? 1-42. Such, organization of glucose molecules can influence local
environmental alterations [192]. The authors here reported a decrease in protein dimerization
because of co-solvated glucose. But this can be attributed to the non-physiological concentration
of glucose molecules (0.8 M) in their simulation systems. This can be explained in terms of
slower peptide diffusion in these crowded environments. Slower peptide, and aggregate diffusion
can result in lower peptide-peptide interactions necessary for increase in aggregate size. Such
diffusion-limited peptide aggregation behavior is noted in previous research on macromolecular
crowding and aggregation [189]. Here, we need to note that physiological levels of glucose is
about 10 mM in the case of type 2 diabetes. Previous experiments investigating the impact of
glucose on A? aggregation have also adopted glucose concentrtions of 5 mM and 10 mM.
5.3.2 Secondary Structure in Protein Aggregates
Further, we analyzed ? sheet content in our protein aggregates and tracked its growth over
time (Fig. 5.3). Our simulations do not show any particular trend in the amount of ordered ?
sheet. There is a quick gain in ordered structures at the start of the simulation, followed by a
rough close to ?-sheet content of 0.4/0.5. The presence of glucose at near physiological levels
led to faster aggregation, but did not affect the fibrillation levels. Previous experimental reports
65
Figure 5.3: Evolution of ? sheet content over time.
by Kedia et. al. also reported this increased aggregation in presence of glucose without any
changes in circular dichorism [49].
Figure 5.4: Relative enrichment of individual coarse-grained interaction site of glucose
66
5.3.3 Restricted rotation of interfacial glucose molecules
Finally, we quantified the local enhancement of glucose molecules close to the peptide
aggregate. Fig. 5.4 is the relative enrichment (RE) of each CG interaction site of glucose,
compared to water molecules as a function of distance from the protein aggregate. As glucose
is an asymmetric molecule, here, we want to understand if there are any directional nature of
glucose-peptide interaction.
We found that the C1 and C2 groups of glucose (CG interaction site - B3) interacted
preferentially with the peptide, whereas B2 (representing C3 and C4) and B1 (representing C5
and C6) preferred interacting with water. Previous ab initio calculations of glucose?s hydrogen-
bonding patterns in water has noted a similar asymmetric set of interactions [200]. The authors
found that the oxygens at C5 and C6 have lower hydrogen bonding capabilities compared to other
oxygens. We observe a similar trend, with C5 and C6 partitioned closer to the peptide aggregate.
This preferential orientation of glucose molecules will result in restricted rotations, and therefore
a loss of rotational entropy. This can explain our concentration dependent increase in aggregation
rates. With increase in size of aggregate (concurrent increase in aggregation rates), the aggregate
surface-area accessible for interaction with glucose decreases, which in turn reduces this entropic
penalty. Similar macromolecular crowding and entropy associated effects (folding/denaturation)
are noted also for RNA and protein systems [201, 202].
While epidemiological studies have correlated AD and T2D, the primary reason for this has
been attributed to covalent modification of arginine and lysines (AGE), resulting in cross-linking
of amyloid aggregates. These covalent crosslinks prevent dissolution of peptide aggregates.
Reports have suggested of a possible temporal mismatch between peptide aggregation and chemical
67
modifications [49]. Here we uncover a complementary thermodynamic pathway for this observed
increase in peptide aggregation with glucose.
In the case of protein aggregation, research on macromolecular crowding effects has been
generally limited to non-interacting (steric) crowders and very high concentrations to mimic
physiological viscosity in biological environment [189, 203]. Latsaw et. al. reported that such
crowders can lead to decrease in effective sizes of aggregates resulting in increased number of
smaller oligomers. In some instances, addition of ad-hoc hydrophobic effect to the crowder
could shape the structural features of the aggregate. Finally, adding soft non-specific crowder
interactions have been shown to have unique implications, and reverse the effect of systems
with only steric crowders [204, 205]. In this work, we report a non-specific, but directional
interaction of glucose molecules, with a certain preferred orientation. This uncovers another
possible mechanism how specific co-solutes can affect biologically relevant processes.
5.4 Conclusion
Self-assembly of A? peptides into specific deposits is noted as an essential upstream process
in Alzheimer?s disease. Several recent studies have suggested interlinks between high blood sugar
levels and Alzheimer?s disease. This has led to biophysical research to uncover connections
between hyperglycemic conditions and A? aggregation. Several research studies have indicated
the ability of glucose to chemically modify amino acids and create covalent crosslinks, which are
resistant to dispersion. Currently, to our knowledge, there are limited studies aimed at uncovering
thermodynamics-derived mechanisms that can aid peptide aggregation in presence of glucose
molecules. In this work, we approach this research question from a computational lens ? using
68
coarse-grained molecular simulations to understand the aggregation of A? 16-22 with varying
concentrations of glucose (1.98 mM, 9.8 mM and 19.8 mM). Our analyses of coarse-grained
simulation trajectories also suggest a glucose concentration dependent increase in aggregation
rate. We did not see any changes in the amount of ?-sheet rich ordered structures in presence of
glucose molecules. The glucose molecules close to the peptide aggregate preferred a particular
orientation, with C5 and C6 partitioning closer to the aggregate, while other CG interaction center
associated with other carbon molecules preferring interaction with water. This restricted rotation
of this ring molecule would result in a rotational entropy loss, that can force a reduction of
accessible area for glucose molecules to interact. Therefore, hastening the peptide aggregation
process. This could be a possible thermodynamic mechanism for this concentration dependent
increase in aggregation rate of glucose.
69
Chapter 6: Transferable and Polarizable Coarse grained model for Proteins -
ProMPT
6.1 Overview
This chapter is based on the author?s publication: Transferable and Polarizable Coarse
grained model for Proteins - ProMPT; Abhilash Sahoo, Pei-Yin Lee and Silvina Matysiak. A.S.
and P.L. are co-first authors in this publication.
The application of classical molecular dynamics (MD) simulations at atomic resolution
(fine-grained level - FG), to most biomolecular processes, remains limited because of the associated
computational complexity of representing all the atoms. This problem is magnified in the presence
of protein-based biomolecular systems that have a very large conformational space and MD
simulations with fine-grained resolution have slow dynamics to explore this space. Current
transferable coarse-grained (CG) force fields in literature are either limited to only peptides with
the environment encoded in an implicit form or cannot capture transitions into secondary/tertiary
peptide structures from a primary sequence of amino acids. To address these constaints, we
initially came up with our in-house transferable coarse-grained model ? water-explicit polarizable
coarse grained model for proteins (WEPPRO) that I introduced in chapter 2 (methods). Although,
this model afforded us access to several biomolecular systems; it has several shortcomings.
70
Firstly, WEPPROM does not have representation for all the essential amino-acids. Secondly,
it has not been studied and validated for tertiary folding protein structures.
In this work, we present a transferable CG forcefield (Protein Model with Polarizability
and Transferability - ProMPT) with an explicit representation of the environment for accurate
simulations with proteins. The forcefield consists of a set of pseudo-atoms representing different
chemical groups that can be joined/associated together to create different biomolecular systems.
This preserves the transferability of the forcefield to multiple environments and simulation conditions.
We have added electronic polarization that can respond to environmental heterogeneity/fluctuations
and couple it to protein?s structural transitions. The non-bonded interactions are parametrized
with physics-based features such as solvation, and partitioning free energies determined by thermodynamic
calculations and matched with experiments and/or atomistic simulations. The bonded potentials
are inferred from corresponding distributions in non-redundant protein structure databases. We
present validations of the CG model with simulations of well-studied aqueous protein systems
with specific protein fold types- TRP-cage, TrpZip4, Villin, WW-domain and ?-?-?. We also
explore the applications of the forcefield to study aqueous aggregation of A? 16-22 peptides and
dimerization of glycophorin A (GPA) in micellar environments.
With this model, we primarily address the shortcomings of WEPPRO. ProMPT now has
access to all the amino acid representation. It has updated bonded and non-bonded interaction
parameters with which we validated the conformational landscape of several small proteins and
aggregates.
71
6.2 Introduction
Physiological functions of protein molecules are closely intertwined with their associated
structure and dynamics [206,207]. This complex macro-organization is shaped through microscopic
multi-body interactions that define a protein molecule?s conformational landscape. In this direction,
computer simulations, primarily molecular dynamics are being increasingly leveraged to understand
protein biochemistry and biophysics [208?210].
With recent advancements in dedicated high performance computing architectures and
graphical processing unit, it is now possible to run long simulations with small proteins for
multiple microseconds, at conventional all-atom resolution [211?214]. Such long molecular
simulations could capture a small number of protein folding-unfolding transitions in an unbiased
manner. But such processes require significant amount of computing resources and are still
not scalable to larger systems. Most biological systems of interest involve large biomolecules
that interact over long spatiotemporal scales. To alleviate some of these shortcomings, several
enhanced sampling approaches such as metadynamics and replica exchange molecular dynamics
have been proposed to allow faster exploration of conformational space with limited resources
[14?17]. But such approaches require extensive knowledge about the particular biomolecular
system and/or can be infeasible for larger systems.
Coarse-grained molecular dynamics (CG-MD) involves creating a simplified representation
or minimal model of biomolecules that can capture the essential biophysics [18]. This approach
allows efficient access to long spatio-temporal scales by directly reducing computational complexity
and allowing fast conformational transitions by smoothing the local free energy landscape. Early
coarse-grained (CG) models with simplified phenomenological potentials were instrumental in
72
establishing the foundations of energy landscape theory of protein folding [215?217]. The coarse
grained models require creating interaction sites that are representative of a particular molecule
and defining interaction schemes (potentials) that allow these interaction sites to communicate.
The interaction potentials, also known as forcefields can be variations of knowledge-based (developed
from analysis of statistical databases) or physics-derieved potentials (created for example to fit
free energies, partition coefficients, and biomolecular phases.) [?, 218?224]. Coarse grained
molecular simulations of proteins have played a significant role in shaping our understanding
of physiological processes such as protein folding, protein aggregation and membrane-protein
interactions [217, 225?231].
CG models with varying molecular resolutions and diverse coarse graining strategies have
been proposed. The levels of coarse-graining can vary from multiple-residues represented by a
single interaction site, to models with representation for all the heavy atoms in a biomolecular
system. Several of the CG models employ an implicit description of the solvent environment
through interaction potentials [220, 221, 224, 232?237]. These models have been used to study
several protein-based dynamical processes such as folding, adsorption, misfolding and aggregation.
While these models provide a significant reduction in system complexity, an implicit representation
of the environment cannot be used to study heterogeneous environmental effects that can be
essential in crowded physiological systems. Moreover, the role of solvent in governing thermodynamics
and kinetics of protein folding is well documented [238?240]. In some of the implicit-water
CG models, the dynamics is biased towards native state, and cannot capture non-native states
[216, 241?243].
MARTINI is a popular coarse grained forcefield with explicit description for solvents and
a modular architecture [79, 244]. The interactions are directed through a set of interaction-
73
levels created by leveraging environment-dependant free energies (solvation, vaporization and
partitioning free energies). Due to its reasonable accuracy and ease of use, particularly in heterogenous
and crowded environment, this forcefield has been widely adopted by the natural science and
engineering communities. In the case of proteins, MARTINI has been used to study ligand-
binding, aggregation, surface-absorption and membrane translocation [245?250]. However the
model relies on restraining protein?s secondary structures through artificial potentials, and the
global/tertiary structure also needs to be restrained to prevent spontaneous unfolding. This
prevents any study of dynamical changes to protein?s conformation over the simulation time using
MARTINI. While Go? model has been employed along with the MARTINI forcefield to study
large-amplitude conformational changes, it can suffer from several shortcomings including loss
of amino-acid identity, insensitive to point-mutations and environmental changes [225,251?253].
In our previous publications, we had introduced an explict-solvent polarizable CG model
for a selection of amino-acids to study secondary structure transitions starting from primary
sequence in presence of different environmental stimulus such as hydrophobic media, interfaces
and lipid bilayers [80?82, 85, 254]. In this work, we have formalized our approach to model
parametrization, introduced new bonded and non-bonded potentials, updated residue geometry;
and extended the representation to all natural amino acids to capture accurate protein tertiary
structures.
Our Protein Model with Polarizability and Transferability (ProMPT) consists of two types
of interaction sites (beads) ? primary coarse grained beads and dummy beads. The basic
biomolecule structure is created with primary interaction sites which feature modular architecture
and geometry similar to the MARTINI model that allows for easy transferability; and are parametrized
along the MARTINI scales [79, 255]. Through additional off-center dummy charges to the
74
primary sites that represent polarizable entities, we have introduced explicit local dipole moment,
that can result in anisotropic coulombic interactions similar to hydrogen bonds in protein?s secondary
structure. The angle and dihedral potentials are derived from statistical distributions of these
features from the protein data bank (PDB). ProMPT can capture secondary and tertiary conformational
transitions, along with appropriate intermediate conformations and accurate folding free energy
profiles. As such, this CG model can be applied to study spatiotemporally complex biological
phenomena and processes involving proteins. In this paper, we present model validations using
simulations of small protein structures in aqueous solvent, and aqueous protein aggregation. We
have also discussed future applications, and current shortcomings of this CG potential.
6.3 Methods
The coarse grained interaction sites follow a basic MARTINI-influenced nomenclature, with
atom-types grouped into polar, neutral, charged and hydrophobic beads [79,255]. The polar beads
have explicit electrostatic dipoles added through charged dummy particles, constrained to the
main interaction site (Fig. 6.2). For parametrization, we initially run a 50 ns atomistic simulations
(using CHARMM36m/TIP3P force field [256, 257] and CHARMM-GUI [258] equilibration
protocol) of relevant atoms that map to the particular polar CG interaction site tripeptides that
have the corresponding mapped regions in chemical space. The parameters (charge on the beads -
q and distance between the beads - r) are generated to match the dipole moment corresponding to
the maximum in dipole moment distribution of the mapped region from the atomistic simulation.
The charge and relevant geometry of these charged dummies are provided in Table 6.1.
The dummy charges interact with the local environmental through electrostatic interactions
75
Dummy Types Charge Bond-length kangle
Backbone ? 0.340 0.14 7.2
SER-Sidechains ? 0.144 0.14 7.2
THR-Sidechains ? 0.153 0.14 7.2
GLN-Sidechains ? 0.256 0.14 7.2
ASN-Sidechains ? 0.256 0.14 7.2
TYR-Sidechains ? 0.138 0.14 7.2
TRP-Sidechains ? 0.136 0.14 7.2
Table 6.1: Charges and characteristic bonded potentials for dummy beads. Bond-length is the
length of the tether from the primary interaction-center to the charged dummies. kangle is the
spring constant preventing deviation of the angle between charged dummies and the primary-
interaction-center from 180 degrees.
(such as change in dielectric constant at membrane-water interface), which then directs protein?s
structural changes. As these charges are placed off-center to the main interaction site, they can
introduce an asymmetry and directionality to interactions, and introduce spatial heterogeneity in
local charge distribution. Here, a parallel can be drawn between the direct dipole-dipole and
charge-dipole interactions in this CG model and electrostatic alignments in hydrogen bonds.
Previous publications with a much simplified variant of this CG model, could generate appropriate
sequence-specific secondary structures through alignment of these dipolar charges [80?82, 85,
254]. Several experiment and simulation-based studies have also underlined the importance of
these molecular dipoles in protein folding.
ProMPT has been parametrized to work with the Yesylevsky et al.?s polarizable MARTINI
water model [83]. This water model uses three coarse-grained (CG) interaction sites (beads)
that map to four water molecules. One of the beads is the primary/central interaction site with
the other two dummy interaction sites tethered to it [Fig. 6.1 - charges on dummies: +/- 0.46;
Bond-length: 0.14 nm]. The authors could capture effective orientational polarizability of real
water, that can modulate inhomogenous dielectric response. They applied the water model to
study ion permeation in lipid membranes, and electroporation of lipid membranes and oil-slabs.
76
This model could naturally capture relative dielectric changes due to the local environment.
Figure 6.1: Schematic geometry of Martini polarizable water
Figure 6.2: Schematic geometries of coarse-grained amino-acids
The amino-acid geometry defining the structures of the CG amino-acids are shown in
77
Figure 6.3: Non-Bonded Interactions [POL-POL]. Refer to Fig. 6.2 for naming nomenclature
Fig. 6.2. The interaction sites representing protein backbone and polar sidechains are assigned
polarizable atom-types, with off-center dummy charges. The primary CG interaction sites that
map to aromatic regions in the chemical space have a radius of 0.43 nm, compared to 0.47 nm
for all other interaction sites.
6.3.1 Non-Bonded Interactions
All the non-bonded Lennard-Jones (LJ) type interactions between the atom-types are parametrized
along the MARTINI interaction levels, which allows for easy transferability, and can therefore be
used with all biomolecular environments that can be represented with the MARTINI forcefield. A
complete description of the non-bonded interaction parameters is presented in through heatmaps
(Fig. 6.3-6.8).
78
Figure 6.4: Non-Bonded Interactions [POL-HYD]. Refer to Fig. 6.2 for naming nomenclature
Figure 6.5: Non-Bonded Interactions [POL-Others]. Refer to Fig. 6.2 for naming nomenclature
79
Figure 6.6: Non-Bonded Interactions [HYD-HYD]. Refer to Fig. 6.2 for naming nomenclature
Figure 6.7: Non-Bonded Interactions [HYD-Other]. Refer to Fig. 6.2 for naming nomenclature
80
Figure 6.8: Non-Bonded Interactions [Other-Other]. Refer to Fig. 6.2 for naming nomenclature
While, most of the cross-interactions between non-polarizable interaction sites are directly
borrowed from the MARTINI forcefield, the interactions between our polarizable groups had to
be re-parametrized to balance out the added electrostatic interactions from the charged dummies.
Similar to the MARTINI forcefield, the parametrization here aimed to fit non-bonded interaction
parameters to reproduce accurate free energies of solvation and partitioning. The interaction
level between hydrophobic groups and the interaction sites representing water is reduced by
upto 50% from the MARTINI level to reproduce appropriate environment-induced structural
transitions. These reductions balance out issues of over-polarization by backbone dipoles and
allow conformational switching from helices to ?-strands. We refer readers to the previous
publication by Ganesan et al. and Sahoo et al. regarding further details about non-bonded
parametrization [81, 82]. The electrostatic charges on specific dummy beads are enumerated
81
in the supplementary.
A set of special interactions were added in an ad-hoc manner to capture specific protein-
protein interactions. We applied specific attraction (? = 3.0 kJ/mol) between positively charged
groups (such as ions and cationic amino acid sidechains) and the aromatic rings to mimic cation-?
effects [259]. Similarly, interaction between aromatic rings, and proline sidechains with aromatic
rings were made attractive (? = 3.0 kJ/mol) to capture ?-? stacking and CH-? interactions that
have been highlighted in quantum mechanical calculations [260?262]. These electronic effects
are ubiquitous in protein folding. Additional dummy-dummy and dummy-charged group hard-
core repulsion (c12 ? 10?7 nm) were added to prevent over-interaction between CG groups
similar to methods to prevent polarization catastrophe in polarizable forcefields [263, 264]. This
allows for fast switching between conformations through binding/unbinding among charged dummies.
6.3.2 Bonded Interactions
Bonded interactions can be broadly categorized as bonds, angles and dihedrals. The features
describing bonds in ProMPT ? bond lengths and bond rigidity were borrowed directly from
the MARTINI forcefield. The corresponding parameters are enumerated below (Table 6.2-6.6).
Some angular (between backbone beads and first side-chain) and dihedral potentials (backbone
only) were informed from a statistical distributions of protein structures.
Bond-Types Bond-Length Bond-Rigidity/Constraints
A. Between backbone interaction-centers 0.385 7500
B. Between BB-S1 of non-aromatic amino-acids 0.250 5000
C. Between S1-S2 of non-aromatic amino-acids 0.280 5000
D. Bonds in planar rings of aromatic residues 0.270 Constraints
Table 6.2: Bonded interaction potentials between different primary CG interaction-centers. BB:
Backbone; S1: First Sidechain; S2: Second Sidechain
82
Type Bond-length
Backbone (BB) - Backbone (BB) 0.385 nm
Backbone (BB) - Sidechain1 (S1) 0.28 nm
Sidechain1 (S1) - Sidechain2 (S2) 0.25 nm
Between aromatic rings 0.27 nm
Table 6.3: Bond lengths between primary coarse grained interaction sites.
Angle-Types Angle (degrees) Angular-Rigidity
A. Between backbone interaction-centers 109 75
B. Between BB-S1-S2 of non-aromatic amino-acids 151 25
C. Between BB-S1-S2/S3 for F Y 150 50
D. Between BB-S1-S2 for W 210 50
E. Between BB-S1-S3 for W 90 50
Table 6.4: Angular interaction potentials between different primary CG interaction-centers. BB:
Backbone; S1: First Sidechain; S2/S3/S4: Second/Third/Fourth Sidechain
We used a non-redundancy p-value of 10?7 to create a database of about 14000 protein
structures from the protein data bank. The angular and dihedral potentials were based on their
corresponding normalized distributions from this database. The angle between the protein backbone
sites was universally set to 109o, restrained through a harmonic potential. The angular potential
between the backbone and the sidechain was created specific to each amino-acid by applying
Boltzmann inversion at 300K to each amino-acid specific distributions. These potential energy
functions corresponding to each amino-acid have been reported in Fig. 6.9-6.11. Finally, the
dihedral potentials between the backbone interaction-centers are secondary-structure specific,
with different tabulated potentials for ?-helix, 3-10 helix and ?-sheets. The functional forms
Angle-Types Dihedral (degrees) Dihedral-Rigidity
A. Between backbone interaction-centers Tabulated NA
B. Between BB-S2-S3-S1 for F Y Improper; 0 50
C. Between BB-S2-S3-S1 for W Improper; 0 50
D. Between S1-S2-S4-S3 for W Improper; 0 200
Table 6.5: Dihedral interaction potentials between different primary CG interaction-centers. BB:
Backbone; S1: First Sidechain; S2/S3/S4: Second/Third/Fourth Sidechain
83
Secondary-Structure A B C D ?0
?-helix 1.799 1.969 -1.199 -1.03 1.227
3-10 helix 1.799 1.969 -1.199 -1.03 1.969
Table 6.6: Dihedral potential parameters for ?-helix and ?-sheet corresponding to equation 1 in
the main text.
of ?-helix and 3-10 helix potentials were taken from previous publications of C? based coarse
grained models [265].
V (?;A,B,C,D)
= A [1 + cos(?+ ?0)]
+B [[1 + cos(?? ?0)] (6.1)+ C [1 + cos(3(?+ ?0))]? ]
+D 1 + cos(?+ ?0 + )
4
The fitted parameters are provided in Table 6.6. These periodic functions allow for introducing
multiple local/global minimums, in contrast to single and deep global minima by Boltzmann
inversion. We parametrized these functions to set the global functional minimum to match
the maximum value in the dihedral probability distributions of relevant secondary structures,
in addition to introducing several local minimas. These local extremas introduce additional
frustrations to the protein?s conformations and has been adopted in previous coarse grained
models [265]. The tabulated ? sheet potential is generated through a direct Boltzmann inversion
(at 300K) of dihedral distributions from regions specific to beta sheets in the protein structure
database. Fig. 6.12 shows the dihedral potentials employed in ProMPT.
6.3.3 Simulation setup
All simulations are performed with GROMACS 2019.4 [177]. The starting conformation
and the protein topology are generated with in-house codes, taking the amino acid sequence
84
Figure 6.9: BB(Previous amino acid)-BB-S1 tabulated angular potentials (Set 1)
85
Figure 6.10: BB(Previous amino acid)-BB-S1 tabulated angular potentials (Set 2)
86
Figure 6.11: BB(Previous amino acid)-BB-S1 tabulated angular potentials (Set 3)
87
Figure 6.12: Secondary-structure specific dihedral potential used in the CG forcefield. The
tabulated potentials are fitted (for ?-helix and 3-10 helix) or derieved (?-sheet) to capture
maximum value in the dihedral probability distributions
from the reference PDB structures listed in Table 6.7. The initial conformation for all proteins
is a random unfolded extended conformation. The setup for each protein system including the
number of protein beads, the number of polarizable water molecules, and temperature details
are also listed in Table 6.7. Adequate number of ions are added to neutralize the system. We
have used reduced temperature, T? = KbT/? as our temperature scale, where ? is the highest
interaction strength (5.6 kJ/mol) in our CG potential. Energy minimization is first conducted
with steepest descent. An NPT equilibration run at T?=0.52 for 50 ps is followed at 1 bar and
with time step 0.001 ps. The canonical production run is then performed with time step 0.01 ps
for the simulation time listed in Table 6.7 at different reduced temperatures. Since the box size is
88
constant, the solvent density remains the same across various temperatures. Electrostatic cutoff is
1.6 nm and particle-mesh Ewald (PME) method is applied for long range electrostatic interactions
[151]. Nose?-Hoover thermostat is used to maintain the system at the desired temperature [266].
Trp-cage Trpzip4 Villin WW-Domain ?-?-? A?16-22
PDB code 1L2Y 1LE3 1YRF 1E0L 2KI0 -a
Number of proteins 1 1 1 1 1 8
Number of water 1957 1610 26391 30136 24998 2584
Simulation time per
temperature (ns) 200 200 600 100 200 300
Temperature range
performed 0.52-1.04 0.52-1.04 0.52-1.04 0.52,0.82 0.52,0.82 0.52
Total number of temperature
simulated 15 15 19 2 2 1
Table 6.7: Simulation setup for each protein. Sequence: KLVFFAE.
To validate the use of ProMPT in heterogenous environmental conditions, we studied the
dimerization of Glycophorin A (GpA) monomers starting from their solvated-unfolded state. Two
CG GpA monomers are put into an 8 nm wide cubic box with a distance of 5.66 nm between the
two monomers pointing to the same direction. The box is then solvated with 80 DPC detergents
and 3000 CG water molecules. Energy minimization is first conducted with steepest descent. An
NPT equilibration run at 350K for 5 ns is followed at 1 bar and with a time step of 0.001 ps. The
compressibility is 3.50 ? 105 bar?1. The production run in an NPT ensemble is then performed
with a time step of 0.01 ps for the simulation time of 500 ns at 350K. Electrostatic cutoff is 1.6 nm
and particle-mesh Ewald (PME) method is applied for the long range electrostatic interactions
[151]. Nose?-Hoover thermostat is used to maintain the system at the desired temperature [266].
Six replicas are run for each system.
89
6.3.4 Comparision with Atomistic Simulations - Replica Exchange with Solute
Tempering
The replica exchange with solute tempering (REST) [267] simulations are performed for
Trp-cage and Trpzip4 in water and the Amber99sb [268] atomistic force field is used. For Trp-
cage, an unfolded coil is solvated with 2678 water molecules and neutralized with one chloride
ion. The system then goes through energy minimization, 200 ps of NVT equilibration and 200
ps of NPT equilibration. After equilibration, the REST protocol is used for the production runs
where 10 replicas are simulated spanning a temperature range from 290K to 540K. Each replica
is run for 700 ns and the resulting exchange rates are between 22% to 36%. For Trpzip4, an
unfolded coil is solvated with 3570 water molecules and neutralized with 3 sodium ions. The
system undergoes the same equilibration process and the same REST protocol as those for Trp-
cage. For Trpzip4, a longer simulation time is needed for convergence, each replica is run for
1400 ns and the resulting exchange rates are between 24% to 32%.
6.3.5 Analysis
We used potential of mean force (PMF) plots of Trp-cage, Trpzip4, and villin to validate the
ability of our CG force field to capture a protein?s conformational landscape. Backbone native
contact and root-mean-squared-deviation (RMSD) of peptide fragments were used as reaction
coordinates for PMF calculations. Backbone native contact measures the fraction of backbone
contacts with a cutoff of 7A? that are replicated in our CG trajectories that are native to the PDB-
converted-CG structure. RMSD backbone (BB) is a measure of relative deviation of our CG
90
backbone from the backbone beads in PDB-converted-CG structure. For all proteins, the first
80ns of trajectory-data is removed and the equilibrated data are collected at all temperatures
simulated and re-weighted through Multistate Bennett Acceptance Ratio (MBAR) method [269].
Here, we present the final PMF at T?=0.52. For villin, we used a cluster analysis tool from
GROMACS with a RMSD cutoff of 0.3 nm to generate representative conformation corresponding
to each basin in the PMF. For the aggregation simulations, two peptides are considered part of
?-sheet if more than three dummy-dummy interaction pairs are formed. A fraction of one means
that all the peptides (8 of them) are forming ?-sheet.
In this study we used potential of mean force to capture the folding and dimerization
landscape of our Glycophorin A. Here, we only used the trajectory after at least one CG interaction
site contact (cutoff=6A?). We computed helical content per monomer as number of native backbone
contacts (cutoff=6A?) compared to the PDB-converted-CG structure for the same sequence of the
wide type. We used average helical content and the number of backbone (BB) contacts are used
as the reaction coordinates for the PMF calculations.
6.4 Results and discussion
Mini proteins Trp-cage and Trpzip4 are selected to be our first test proteins because they
are already well-studied both computationally and experimentally [270?274]. Trp-cage has an
?-helix starting from the N-terminus and a 3-10 helix at the middle part, and an unstructured
C-terminal tail, which collapse together forming a hydrophobic core (Fig. 6.13c). Trpzip4 has
a ?-hairpin structure with four Trp residues packing together. Fig. 6.13a shows the PMF for
Trp-cage using native contact and RMSD helix BB as two reaction coordinates. A two-state
91
folding/unfolding path can be observed from the PMF, which agrees with experiments and REMD
simulation results [270, 275]. The folded basin at around 0.1 nm RMSD helix BB indicates a
perfect ?-helix. The distribution of native contact is relatively broader, from 0.6 to 0.7. The
unfolded basin has a a native contact of 0.4 to 0.5 and RMSD helix BB from 0.2 to 0.3 nm,
indicating some partial helicity in agreement with previous simulation reports [276]. The PDB
structure and a representative folded structure with our CG model for Trp-cage are shown in
Fig. 6.13c. Both ?-helix and 3-10 helix can be well reproduced at the correct positions. The
hydrophobic core formed by Tyr3, Trp6, and three proline residues at the C-terminus is also well
captured as seen in the experimental structure. Our estimation of ?G for protein folding is 3.2
kJ/mol, close to the experimental estimate of 2.6 kJ/mol [277]. We also perform a set of replica
exchange with solute tempering (REST) simulations to estimate the folding-unfolding free energy
difference for Trp-cage, from the atomistic simulations, the ?G is estimated to be around 2.81
kJ/mol at 290K (Fig. 6.14).
Fig. 6.13b shows the PMF for Trpzip4 with native contact and RMSD BB as the reaction
coordinates. The PMF of Trpzip4 also shows a two-state folding/unfolding path, which agrees
with experiments and atomistic simulations. The folded basin for Trpzip4 is located at 0.35 nm
RMSD BB with a native contact of around 0.8. The unfolded basin has a broad RMSD BB
distribution from 0.5 to 0.7 nm with a native contact of around 0.4. Fig. 6.13d shows both the
PDB structure (left) and a representative folded state for Trpzip4 with our model (right). The
?-sheet content can be well captured. Four tryptophan residues are interacting with each other
and forming a hydrophobic core as observed in the experimental structure. Here, our CG model
estimated a ?G of about 7.3 kJ/mol for protein folding, compared to 12.3 kJ/mol reported in
experiments [278]. From the REST simulations we run, a ?G of 4.53 kJ/mol is estimated (Fig.
92
6. CG Protein Model Paper Figure I
(a) (b)
(c) (d)
Figure 6.13: PMF for Trp-cage (a) with RMSD helix BB and native contact as the reaction
coordinates. PMF for Trpzip4 (b) with RMSD9B1 B and native contact as the reaction coordinates.
Both PDB structure (left) and the representative structure (right) from our model for Trp-cage (c)
and Trpzip4 (d) are shown. Blue indicates the specific secondary structure each protein exhibits.
93
6.15), which is lower than both the experimental data and the estimation from our CG model.
Our next target protein - villin has three helices with a hydrophobic core. Successful folding
of villin validates our CG protein model for general applications to proteins with tertiary packing
of homogeneous secondary structures. Fig. 6.16 shows the PMF for villin with RMSD strand 1
(S1) and RMSD strand 2 (S2) as two reaction coordinates, where the locations of the two strands
are highlighted in red. Strand 1 contains two helices from the N-terminus (residue 2-18), while
strand 2 only contains one long helix (residue 21-32) from the C-terminus. There are one native
state and two intermediate states populated at T?=0.52. At higher temperatures, the location of
the unfolded basin shifts to more extended conformations. All of the states populated at T?=0.52
have similar RMSD S1 values, but different RMSD S2 values, which correspond to different
tertiary structures. Indeed, it has been reported in experiments that although there are fluctuations
in the N-terminal part of villin, large-scale unfolding will not occur unless undergoing global
unfolding [279]. In our model, global unfolding happens at higher temperatures, where disruption
of strand 1. The native state is centered at RMSD S1 0.35 nm and RMSD S2 0.1 nm. The
representive structure of the folded basin with our model is very similar to the experimental
structure from PDB. For the intermediate states, one is centered at RMSD S1 0.35 nm and RMSD
S2 0.2 nm while the other one is centered at RMSD S1 0.35 nm and RMSD S2 0.35 nm. Both of
the intermediate basins show a decrease in helicity for strand 2 and losing of hydrophobic core,
while the helicity for the basin with larger RMSD S2 is reduced further more. The intermediate
states are similar to the states described in the experimental work from Reiner and co-workers,
where the intermediate states have an unfolded strand 2 that is undocked from the hydrophobic
core [279]. Both the native state and intermediate states are on the folded side of the folding-
unfolding landscape.
94
Free energy (kT)
1.2 10
9
1
8
0.8 7
6
0.6
5
0.4 4
3
0.2
2
1
0.7 0.8 0.9 1 1.1 1.2 1.3 1.4
Radius of gyration (nm)
Figure 6.14: Free energy plot for Trp-cage at 290K from the REST simulations. The folded basin
has a free energy of 2.82 kT and the unfolded basin has a free energy of 3.98 kT. The ?G is
estimated to be 1.17 kT (2.81 kJ/mol).
95
RMSD (nm)
Free energy (kT)
10
1.2 9
8
1
7
0.8 6
0.6 5
4
0.4
3
0.2 2
1
0.6 0.8 1 1.2 1.4
Radius of gyration (nm)
Figure 6.15: Free energy plot for Trpzip4 at 290K from the REST simulations. The folded basin
has a free energy of 2.12 kT and the unfolded basin has a free energy of 4.00 kT. The ?G is
estimated to be 1.88 kT (4.53 kJ/mol).
96
RMSD (nm)
Villin
Strand 1
Strand 2
32
Figure 6.16: PMF for villin with RMSD S1 and RMSD S2 as the reaction coordinates at T?=0.52.
The representative structure for each basin is shown as insets.
97
For the proteins having a tertiary structure with homogeneous ?-sheets, we used WW-
domain as our benchmark protein. The PDB structure is shown in Fig. 6.17c with three antiparallel
?-strands and two long tails at both ends. A representative folded structure with our model is
shown in Fig. 6.17c. In Fig. 6.17a, the RMSD BB time series without considering the loop
regions are shown at two temperatures. From the time series at T?=0.52 we observe that the
simulation converges at a value of RMSD BB around 0.3 nm. The simulation can reach the folded
state in about 20 ns. We could note frequent transitions between folded and unfolded states at
T?=0.82. The last protein we focus on ?-?-? (PDB code: 2KI0), which is a designed protein
that has a mixture of secondary structures (partially ?-helix and ?-sheet) with a defined tertiary
structure. Using DSSP [280] we assigned the secondary structures using one of the deposited
NMR structures in the PDB database. The antiparallel ?-strands only have four residues on each
strand. Fig. 6.17d shows a representative folded structure with our model, the ?-helix can be
well formed, while the ?-sheets are formed but shifted. Fig. 6.17b shows the RMSD BB time
series for ?-?-? at two temperatures, without considering the tails. The RMSD BB converges at
around 0.5 nm at T?=0.52. The time series at a higher temperature (T?=0.82) is shown in red,
which confirms the ability of ProMPT to capture frequent transitions between different states.
We have demonstrated that with our model we can fold a variety of proteins, which include
pure ?-sheets, pure helices, and proteins that have a defined tertiary structure composed with
homogeneous secondary structures. For proteins that have mixed secondary structures such as
?-?-?, the packing could not be perfectly reproduced. We also attempted simulating the folding
of another well-studied protein with heterogeneous secondary structures - GB1 (PDB code:
2J52.pdb). GB1 is a slightly larger protein with four ?-strands and one ?-helix. Although our
CG model could generate accurate secondary structure at appropriate regions, the overall tertiary
98
CG Protein Model Paper Figure
(a) (b)
(c) (d)
Figure 6.17: RMSD BB time series for (a) WW-domain and (b) ?-?-? at T?=0.52 (blue) and
T?=0.82 (red). The PDB structure (left) and the representative structure (right) are shown in (c)
and (d) for WW-domain and ?-?-?, respectively. B95lue indicates the specific secondary structure
each protein exhibits.
packing was not reproduced. We compared the molecular packing of amino-acid side-chains
in ProMPT for GB1 to GB1 PDB native structure mapped into our CG description. ProMPT
simulations, with backbone restrained to the native state, resulted in a protein core with packing
defects and voids (Fig. 6.18). On the other hand, CG structure mapped directly from PDB native
state showed no such anomalies. Therefore, we postulate that appropriate modifications to side-
chain size, can achieve adequate protein-core packing required for correct tertiary structures.
Proteins with mixed folds will be our next target proteins and could potentially be solved
by adjusting the size and shape of each amino acid in the CG model to better mimic the side-
chain geometry. Recent parametrization of MARTINI 3 force field has adopted this strategy and
included several smaller bead-types to capture the accurate geometrical shapes of biomolecules
[244]. This resulted in significant improvement in molecular packing, and could capture correct
99
Figure 6.18: Surface Plot of GB1 with a solvent probe of 1.4 A?. a- CG after 5 ns of simulation
starting from the folded state, b- PDB. The backbone is traced to show the tertiary packing of the
protein.
100
open and close conformations in transmembrane proteins.
Beyond studying protein folding, ProMPT can also be applied to study peptide aggregation.
Here we demonstrate the aggregation of A? 16-22 peptides in water at T?=0.52. The protein
concentration in the system is 0.0426 M. Fig. 6.19 shows the time series for ?-sheet fraction. The
?-sheet fraction fluctuates between 60% to 80%, indicating significant fibrillation. Occasionally
the ?-sheet fraction can reach 100%, where all the A? peptides are involved in the ?-sheet
formation. Similar to previous experimental and simulation studies, peptides organize into stable,
layered ?-sheet rich structures [189,281?284]. PHE residues are shown in yellow and a hydrophobic
core formed by PHE packing can be observed from the aggregate snapshot (inset of Fig. 6.19).
We observed anti-parallel ?-sheets in our simulations, similar to the structures reported by experiments
[285]. Nguyen et al. applied replica exchange MD on A? 16-22 monomer, dimer, and trimer
to study the effects of different atomistic force fields on the formation of ?-sheets [286]. While
AMBER99 and OPLS generated a diversity of conformations, only GROMOS96 could successfully
generate antiparallel ?-sheets. With ProMPT, we observe aggregation into ?-sheet rich, fibril-like
structures by 30 ns and can also qualitatively study the kinetics of this aggregation process.
6.4.1 Simulations of Glycophorin-A and Mutants
Fig. 6.20a shows the PMF of wild type GpA, where the most populated basin is observed
with a high average helical content and a broad range of number of BB contacts. A representative
snapshot of GPA dimerization is also presented. The WT-GPA dimerizes as a parallel dimer. In
addition, we could also reproduce C?-H contacts prevalent in the G79xxxG83 motif, along with
T87-T87 contact that is noted in previous research [287].
101
CG Protein Model Paper Figure IV
Figure 6.19: Time series for the number of A? 16-22 peptide forming ?-sheets. An illustration
of ?-sheets aggregation is shown in the inset with9y6ellow representing PHE residues.
The helical crossing angle is close to PDB deposited structures (marked by black horizantal
lines), with a strong peak between -50 and 20 degrees for WT-GpA [Fig. 6.21]. Here, we also
noted some anti-parallel orientations, which can be because of the less restrictive media of DPC
micelles compared to the membrane. Anti-parallel dimers have been reported before in atomistic
simulations in both implicit membrane and implicit cyclohexane [288].
The contact map shown in Fig. 6.22 shows that the majority of the wild type GpA dimers
are parallel, which agrees with the crossing angle result shown earlier. In addition, from the
residue-residue contacts it can be seen that the dimers are highly symmetric and specific. This
specific contact surface is also observed in CG simulations in membrane where a mobile helix is
found to be anisotropically distributed near the fixed helix [289] that is originated from the close
contacts of the GxxxG motif.
102
Figure 6.20: PMF for GpA with the average helical content and the number of BB contacts as the
reaction coordinates. The representative conformation is shown as inset-figure. Color code: Thr
(grey), Gly (green), Val (purple), Leu (orange).
103
Figure 6.21: Helical crossing angle between the two helices of Glycophorin A. The horizantal
lines reflect PDB values.
Figure 6.22: The residue-residue contact maps for GpA
104
6.4.2 Note on Computational Efficiency
To compare the computational efficiency gained by applying our CG model, we estimate
the computational resource used for obtaining a converged PMF for Trp-cage via REST and our
CG model. In order to get a converged PMF using REST atomistic simulations, 10 replicas at
different temperatures are needed with 700 ns simulation time for each replica. With optimized
simulation speed of 0.17 hours/ns and 20 cores-per-replica, the simulation required about 23.80
kSU (supercomputing units) of resources. For the converged PMF with our CG model, we used
15 replicas at different temperatures, with 200 ns simulation time per replica. With optimized
simulation speed of 0.028 hour/ns and 27 cores-per-replica, the simulation required about 2.27
kSU. In order to get a converged PMF using REST, several parameters need to be tested (such as
exchange frequency, number of replicas, range of temperatures), including the use of force field.
These will result in a much higher computational cost in practice.
The applications and validations of our CG model in this work was focused on capturing
accurate protein conformational landscape and structural diversity in aqueous environment. This
explicit treatment of environment can capture local electrostatic changes such as dielectric fluctuations
more effectively that can contribute towards protein?s structural changes. This also allows for
easy transferability of interactions, without a need for re-parametrization. Therefore, ProMPT
can have special applications in studies involving chemically heterogeneous biomolecular systems
due to its explicit description of environment (and solvent). Because of similar parametrization
of non-bonded interactions, ProMPT can be patched with our in-house coarse grained membrane
model ? Water Explicit Polarizable MEMbrane model (WEPMEM) to study membrane and
detergent assisted conformational transitions [81, 82, 84, 254]. Moreover, ProMPT can also
105
be patched with our CG model for divalent ions to study metal-assisted protein folding [174].
Finally, the modular architecture of the forcefield allows for easy parametrization of other biomolecules.
Future work in our lab will be focused on studies to elucidate this environment-driven conformational
changes in protein structure and dynamics with this forcefield.
6.5 Conclusion
In this article, we developed a polarizable coarse-grained model with explicit representation
for the environment ? ProMPT. The key objective here was to create a coarse-grained molecular
dynamics forcefield to capture tertiary folding of protein structures with minimal constraints.
The CG mapping scheme follows closely the MARTINI coarse grained forcefield geometry.
In ProMPT, the polar CG beads have explicit drude-like charges that can couple a protein?s
environment to its structure and dynamics. We parametrized the non-bonded interactions between
CG beads through free energies of their interaction with the environment. The bonded interaction
potentials were generated through analysis of corresponding normalized bonded-feature distributions
generated from a database of non-redundant protein structures from the protein data bank. This
CG forcefield was validated by reproducing the conformational free energy landscape of several
well-studied small protein systems. While ProMPT can accurately fold protein structures with
homogeneous secondary structures, we observed some side-chain packing defects in folding of
proteins composed with heterogenous secondary structures. Our future research efforts will be
focused on fixing sidechain packing through accurate mapping of sidechain geometry. Due to its
transferability, ProMPT can be patched with our previous in-house lipid and divalent ion models
to study protein folding assisted by physiological environment.
106
Chapter 7: Effect of varying poly-Q tract and the presence of curved membranes
on aggregation of Huntingtin protein?s N-terminal domain
7.1 Overview
In the previous chapters (chapter 3-5), we characterized the aggregation behavior of a
fragment of A? peptide. Along another direction in this chapter, we focus on Huntington?s
disease.
The initial 17 residues of the N-terminal and the following Glutamine repeats (poly-Q/GLN-
repeats) of the Huntingtin protein have been associated with the pathogenesis of Huntington?s
disease (HD). A direct correlation has been established between the length of poly-Q tract and
the severity of Huntington?s disease. Moreover, HD belongs to the larger class of diseases whose
pathogenesis involves poly-Q aggregation. While previous research have presented a consensus
overview of the structure and dynamics of N17 fragment, a similar overview of ther poly-Q
tract is missing. There have been competing evidences of secondary structure in the poly-Q
tract. Moreover, since this is a fast aggregating peptide, it is difficult to effectively discern
between individual peptides and peptides in their aggregate form through structural studies. In
this chapter, we aim to explore the conformations of N17-Qn (where n ? {7, 15, 35, 45}), both
as a single peptide and peptides in their aggregate form. For the single peptide, we noted a
107
trend towards more globular (decreasing asphericity) peptide structures with increasing lengths
of polyglutamine tract. We also observed a significant specificity towards Q-Q interactions also
observed in previous experimental reports. Finally, we also found a gradual change from helical-
rich structures to increasing sheet-like features with increase in number of trailing GLN residues.
For the case of peptide aggregates, we found that the GLN-repeats drove peptide aggregation.
While, for smaller poly-Q tract, the hydrophobic core led by bulky Phenylalanine groups on N17
are formed, but a similar core is absent for longer poly-Q tract.
N17 often structures into amphipathic helices that can tether the protein to the membrane
surface. Peptide anchoring to the membrane surface can have unique implications to disease
pathogenesis, such as the peptide diffusion can get arrested resulting in increased aggregation.
While some structural features of N17-membrane interaction are known, detailed biophysical
reports of peptide-membrane interactions are still missing. In this chapter we also aim to understand
these biomolecular details of peptide-membrane interaction and the molecular mechanism of
curvature sensing. Our simulations suggest a gradual progression from a dominantly unstructured
peptide in solution, to a more structured ?-helical patch on the membrane. In presence of a
curved membrane, we observed that N17 peptides preferentially interact with the curved region.
Here, we discovered that while the polar and charged groups drive the initial membrane-peptide
interaction, the membrane curvature sensing is dominantly controlled by the bulky hydrophobic
groups such as PHE. Omission of these bulky residues from the sequence leads to loss of this
curvature sensing ability of N17.
We will approach this research with the coarse grained protein model that we developed in
our previous chapter ? ProMPT.
108
7.2 Introduction
Huntington?s disease (HD) is a hyperkinetic movement disorder and is associated with rapid
neurodegeneration [290?293]. In recent years, this has emerged as a significant socio-economic
concern as this disease is incurable and invariably lethal. Structural and functional alterations of
Huntingtin protein is responsible the Huntington?s disease [294?296].
Huntingtin protein is a large ( 348 Kda) soluble cytosolic protein often found in the central
nervous system [294]. The wild-type Huntingtin protein consists of a 17 residue N-terminal
domain, followed by 35 or less Glutamine repeats, a poly-proline rich segment and a cluster
of HEAT repeats (?-helix-turn-?-helix unit) [294]. The final HEAT motifs are essential for
several protein-protein interactions associated with this protein?s functions. Abnormality in the
Glutamine repeats has been associated with the pathology of Huntington?s disease (HD). Fig. 7.1
is a schematic image highlighting these regions.
Studied have correlated the increase in polyglutamine lengths (mid 30s and higher) to
increased disease severity [297?299]. Now, HD is a part of a large number of diseases such
as spinocerebellar ataxia (SCA), and spinal and bulbar muscular atrophy (SBMA) are associated
with glutamine repeats and share several common pathological signatures [53?55]. Indications of
poly-Q diseases involve inclusion-bodies with several poly-Q stretches and progressive neuronal
cell deaths. Studies with animal models have reported that just the presence of long poly-
glutamine chains is enough for initiation of neuronal death and neurodegeneration [55].
For HD, clinical research found that mutants with expanded poly-glutamine tract can lead
to pathological protein aggregation, often associated with neurodegeneration [290?296]. These
findings have fueled increasing efforts to understand the biophysical properties of GLN that result
109
Figure 7.1: Schematic of Huntingtin protein with focus on the N-terminus
in aggregation of poly-Q. But, as early aggregation of poly-Q is fast and difficult to study by
traditional experimental methods, computational studies have often been applied at those stages.
Early computational and experimental studies found that water is a poor solvent for GLN and
the poly-glutamines prefer to self-interact and assemble into disordered structures, which can
over time evolve into fibrillar deposits [300?302]. Several in-vitro studies have also confirmed
that poly-Q can aggregate by itself into ordered and insoluble ?-sheet rich patches, among other
more disordered structures [298]. These insoluble aggregates were also confirmed in cell cultures
and mice models [55, 303]. Poly-Q, by itself, can be disordered in solution, and collapse into a
globular shape with increase in Q-repeats [55, 304].
While poly-Q forms the unifying theme for these glutamine repeat diseases, other fragments
can have unique implications, particularly in protein aggregation. In huntingtin protein, there are
two more fragments associated with the disease pathology ? the first 17 residues (N17) and the
110
38 residue poly-proline (polyP-C38) rich domain following poly-Q. While, polyP-C38 is known
to retard protein aggregation rates, N17 aggravates it [305?307].
N17 is intrinsically disordered in solution, but can form amphipathic helix-like structures in
hydrophobic and membranous environments [307]. Ampipathic helices (AH) are commonplace
in several biological systems, and have functional relevance to many physiological processes [63,
308, 309]. Particularly, these helices can help in tethering the protein to the cellular membranes.
This can help in protein co-localization and membrane permeation ? a strategy that is employed
by several anti-microbial peptides. Larger AHs such as Bar domain have been known to affect
membrane remodeling, lipid phase separation and local lipid diffusion [309, 310].
In the case of Huntington?s disease, recent biophysical evidences have suggested that N17
is the cis-acting switch for driving poly-Q aggregation [307,311]. Particularly these experiments
established the importance of hydrophobic residues. Point-mutations that broke the helix or
reduced hydrophobicity, in turn reduced the aggregation propensity. Even in the absence of poly-
Q, N17 is known to aggregate into peptide clusters.
Huntingtin protein is specifically associated with several membrane-related processes such
as vesicle trafficking and has been found to localize at the membranes of endoplasmic reticulum,
golgi apparatus and endosomal vesicles [53]. N17, being an AH forms the membrane binding
domain of huntingtin protein. But, in the case of pathologically mutated htt, membrane binding
can tether the peptide and arrest the dimensionality of diffusion, thus promoting aggregation. As
N17 plays a central role in aggregation both in solution and in presence of membranes, it can be
the key towards therapeutic intervention [307].
While several experimental research attempted to understand the biophysics of the N17
and N17-poly-Q domain, structural information remains limited due to highly dynamic and
111
transient (quick aggregating) nature of these biomolecules. Recent experimental characterization
revealed that with a particular variant of N17-poly-Q, the peptide has a propensity towards ?-
sheet structures in Q-rich regions [312, 313]. Experimental data by Langen and coworkers with
Time-Resolved Fluorescence Energy Transfer (TR-FRET) and Electron Paramagnetic Resonance
(EPR) suggest both ?-helix and ?-sheets within the poly-Q segment [305]. Another structural
study by Nagai et al. also reported an increased ?-sheet fraction with increase in Q-repeats [314].
Recently Michaelek et al. used two dimensional solution NMR to identify the lowest
energy structure of this N17 domain in micellar conditions [315]. In the same study, they also
delineated the structure in presence of POPC membranes. In another structural investigation, Tao
et al. used model small unilamellar vesicles to study the structure of N17-Q25 fragment [316].
Their study showed the propensity of this peptide to interact interfacially with the membrane,
with stable helicity in the N17 part. Chaibva et al. used atomic force microscopy with POPC
membranes restricted to a fixed curvature to record a interaction of N17-Q35-P10-KK fragment
dominantly with regions with positive curvature [60].
Only a few computational studies have been attempted to characterize these systems because
of the associated large system sizes and the intrinsic disordered nature of these peptides. Moreover,
discrepancies can be noted in these computational findings. Initial results by Kelly et al. with
simulated tempering suggested two stable populations of N17 structures ? the more stable
double helix (bent) and a single linear helical patch [317]. With Multiscale Coarse grained (MS-
CG) simulations by Wang and Voth reported quick interchange between folded-unfolded states
for lower Q-repeats (?28), and long stable helices for longer Q-repeats (?28) [318]. Similarly,
investigations by Dlugosz and Trylyska with atomistic simulations and replica exchange molecular
dynamics also suggested a helix-rich poly-Q (Q-55) domain with a solvent exposed N17 headpiece
112
[319]. On the other hand, Lakhani et al. with discrete molecular dynamics and an atomistic
resolution forcefield found an increased tendency to form ? sheet rich structures with expanded
lengths of Q [320]. Both computational and experimental structural studies centered around
understanding the morphology of N17-Poly-Q aggregates are scarce, and we have limited knowledge
of aggregate architecture.
Zhang et al. investigated the folding of N17 domain through a consensus forcefield approach,
using simulations from multiple different forcefields (OPLS-AA/L, CHARMM36, and AMBER99sb-
ILDNP) in presence of dodecylphosphocholine (DPC) micelles. They found a relatively low
barrier ( 3 kcal/mol) of transformation into a helical structure [321]. Cote et al. characterized the
membrane interaction of N17 through a combination of molecular simulations and experiments
[322]. Their analysis points to a single ?-helical structure, more stable than the micellar NMR
structure suggested by Michaelek et al. [315]. A more recent study elucidated correlation between
peptide structure and the length of Q-repeats for a single peptide. Here the authors report of a
super-compact and more globular structures with progressive increase in Q-repeats.
Studies on how N17 drives membrane interaction and curvature sensing is limited due to
compounded complexity of the presence of membranes, in addition to the intrinsically disordered
and highly transient peptide. While peptide aggregates are acknowledged as the primary pathological
entity in Huntington?s disease, computational research focus has been primarily on monomeric
structure and monomer-membrane interaction. Moreover, N17 being an amphipathic helix can
have preferential interaction with curved membrane as noted earlier by Chabaiva and coworkers
with atomic force microscopy [60]. But, the details of peptide structure that enable such curvature
sensing is still missing. In this work, we aim to expand our understanding to peptide aggregate
structures and how they drive membrane-protein interaction, particularly curvature sensing. Our
113
recently developed coarse grained forcefield ? ProMPT is uniquely poised to address this because
of the explicit-solvent and transferable nature of this CG forcefield. Also, coarse-graining affords
the access to length and time-scales beyond the current atomistic levels to study peptide-membrane
interactions in the case of aggregates.
The objectives of this research are two fold.
1. Understand the impact of pathological glutamine repeats on peptide structure and aggregation
- While previous research have presented a consensus overview of the structure and dynamics
of N17 fragment, a similar overview of ther poly-Q tract is missing. There have been
competing evidences of secondary structure in the poly-Q tract. Moreover, since this is
a fast aggregating peptide, it is difficult to effectively discern between individual peptides
and peptides in their aggregate form through structural studies. Therefore, in this work we
aim to explore the conformations of N17-Qn (where n ? {7, 15, 35, 45}), both as a single
peptide and peptides in their aggregate form.
2. Present a mechanistic view on how N17 peptides sense curvature on POPC membranes -
Previous research on peptide-membrane interactions has been limited both from a experimental
and computational perspective. While some structural features of N17-membrane interaction
are known, detailed biophysical reports of peptide-membrane interactions are still missing.
In this chapter we aim to understand these biomolecular details of peptide-membrane
interaction and the molecular mechanism of curvature sensing.
To our knowledge, this is the first computational research that studies unbiased peptide
aggregation and peptide-membrane interactions of N17/N17-Qn peptides starting from solvated
unstructured peptides.
114
7.3 Methods
For this work, we used the coarse-grained forcefield developed in chapter 5 ? ProMPT.
Due to the explicit-solvent nature of this forcefield, it can capture local environmental fluctuations
(such as local dielectrics) well. This can contribute towards environment-assisted structural
changes in protein structure. Moreover, this explicit representation of environment allows for
easy transferability across multiple biomolecular systems, without need for re-parametrization.
As such, we used ProMPT (see chapter 5) to represent the protein, WEPMEM (see chapter 2) to
represent membranes with the MARTINI polarizable water [81?83].
7.3.1 Single Peptide (N17-Qn) simulations in aqueous solution
To understand the folding landscape of N17 with multiple Q-repeats, we ran simulations
with 7, 15, 35 and 45 trailing glutamines. All simulations are performed with GROMACS 2019.4
[177]. The starting conformation and the protein topology are generated with in-house codes. The
initial conformation for all proteins is a random unfolded extended conformation. The simulation
box is adequately solvated and neutralized through addition of ions. 9 simulations were run
for each peptide system ranging from 300K (T?=0.445, previous chapter) to 460K (T?=0.682,
previous chapter). This allows us to escape local free energy barriers and have a better exploration
of conformational landscape.
Energy minimization is first conducted with steepest descent. An NPT equilibration run at
T=300K for 50 ps is followed at 1 bar and with time step 0.001 ps. The canonical production run
is then performed with time step 0.01 ps for 200ns at different temperatures. Since the box size is
constant, the solvent density remains the same across various temperatures. Electrostatic cutoff is
115
1.6 nm and particle-mesh Ewald (PME) method is applied for long range electrostatic interactions
[151]. Nose?-Hoover thermostat is used to maintain the system at the desired temperature [266].
7.3.2 Multi Peptide (N17-Qn) simulations in aqueous solution
We also did a similar exercise, but with multiple peptides. 6 peptides (? 6.9 mM), each
with trailing glutamine repeats of 7, 15, 35 and 45 were solvated in a polarizable water box, with
ions added to neutralize the box. In the initial conformation, all copies of peptides are randomly
distributed in a simulation box. The simulation procedure is identical to the single monomer
simulation.
7.3.3 Single Peptide (N17) in presence of a planar membrane patch
In this section, we characterize the curvature sensing nature of the N17 part of Huntingtin
protein. As membrane interaction is primarily driven by the N17 fragment of the peptide, here
we have focused on just that segment. To understand how single N17 peptide interacts with
the membrane, we created two replica simulations of a single N17 peptide in presence of a
smaller patch of planar lipid membrane. These simulations can be used as validations against
pre-existing results of N17-membrane interactions. We created the small membrane patch with
240 POPC lipids on each leaflet, and 12 nm (in the Z-direction) solvated box. After initial
energy minimization and 1 ns of equilibration, the production simulation was run for 200 ns
with constant-pressure ensemble. Temperature was maintained at 300K through a Noose-Hoover
thermostat [148, 149] with a time constant of 1 ps. Parinello-Rahman barostat [150] with a time
constant of 1 ps and compressibility of 3 ? 10?5/bar was used alongside with semi-isotropic
116
pressure coupling to maintain a pressure of 1 bar. Particle mesh Ewald (PME) [151, 152] with a
relative dielectric constant of 2.5 and cutoff distance of 1.6 nm was used to compute long range
electrostatics. The Lennard-Jones interactions were modified starting from 0.9 nm to 0 at 1.2 nm
by the GROMACS shift scheme.
7.3.4 Multi Peptide (N17) simulations in presence of a curved membrane
Next, we pivot onto curvature sensing aspect of N17 peptides. Towards this, we created an
artificial curved membrane with a diameter of 13 nm from POPC lipid molecules. The overall
architecture consists of a semi-sphere and a planar region. Fig. 7.2 is provided as a reference. The
curved membrane is created by first placing the POPC lipids as points, and growing them to their
individual sizes through slow-growth method of Gromacs. We followed the approach outlined
in BuMPY [323]. We also added dummy particles close to the lipid headgroup, and instituted a
hardcore repulsion (c12 ? 10?5) between lipid tails and the added dummies. This maintains the
membrane curvature throughout the simulation time.
This curved membrane was energy-minimized and then equilibrated for 1ns with a dt of
0.001 ps with a Berendsen thermostat at 300K. We followed this up with another dt=0.01 ps
equilibration for 1 ns with nose hoover thermostat. Finally, we ran a small production run for
20 ns with constant pressure ensemble. All the simulation protocol is identical to the planar
membrane case.
After, this 20 ns run, we added 36 peptides (? 8 mM), above the critical concentration for
Huntingtin protein aggregation; and 125 mM of NaCl ions [324]. The peptides were initially held
with position restraints at the backbone of residue 8, while initial minimization and equilibration.
117
Figure 7.2: A snapshot of the created curved membrane
The initial equilibration steps were conducted with dt=0.001 for gradually increasing number of
steps ? 0.5 ns, 1 ns and 5 ns. All the other parameters for simulation control are similar to the
curved bilayer creation stage. Finally, at the production stage the simulation was run for 400 ns.
We ran two independent replicas of this simulation system. For an appropriate control, we also
simulated a curved bilayer without any peptides. The amount of curved membrane surface area
available to the peptide is about 265.46 nm2, compared to 623.51 nm2 for the planar region. So,
without any preference for curved/planar region, the chance of interaction with curved region is
36 %, compared to 64 % on the planar region. Moreover, to decode the effects of Phenylalanine
(bulky hydrophobic group) from the other parts of the peptide, we also simulated (for 750 ns) a
peptide system with the two Phenylalanines mutated to Valines (F11V/F17V).
118
7.3.5 Analysis
The re-weighting for all the trajectories in the case of solvated peptide systems (single
peptide and peptide aggregate) were done with pyMBAR module [269]. This allows us to analyze
a particular reaction coordinate in a defined ensemble. The results shown in this chapter has
reweighted the reaction coordinate to distributions at 300K. Similarly, the contact plots shown in
this work is the expectation-value of a particular contact at 300K, along the generated distribution
of that contact.
Special care had to be taken for calculations on the curved membrane. This is because,
there is no uni-directional normal as is the case for a planar lipid bilayer. Therefore, we created
parallels for commonly used membrane metrics for our case, that takes into account the local
membrane curvature.
7.3.6 Area per lipid on membranes with varying curvature
Traditionally, area per lipid is calculated through a voronoi tesselletion of the membrane
surface, and ascribing each lipid with a defined area. But this calculation can be tricky for curved
membranes, as creating a convex hull in two-dimensions would not be appropriate, and a three
dimensional convex hull is often error-prone at our scales. Therefore, in this work, we apply a
local normal approach outlined by Buchoux in their implementation of FATSLiM [325]. First, the
locality set?Ai, for each lipid molecule, defined by the locally close phosphates to the phosphate
of ith lipid molecule is created. Now, the principal component analysis is used to determine
the local normal at ith point, and then identify the local tangent plane for ith lipid (Ti). The
local neighborhood of the ith lipid, Ai is projected onto this to create
?
Ti Ai. Finally, a voronoi
119
tesselletion is performed on ?Ai to define the local surface area.
7.3.7 Fine-grained surface for density calculations
For the density calculation, we interpolated the lipid membrane to create a super fine-
grained local grid of points along the membrane surface. Initially, we create a lateral grid of
about 40000 points placed in the simulation box along X and Y. Now, the Z direction is added
to this grid of points, by creating a set of weights based on two-dimensional distance from the
lipid-phosphates. This two dimensional distance is penalized by exponent of squared distance.
That way, the points closer get more weights and this reduces for points farther off. The approach
of including weights from local points to ascribe Z position of the grid, generates smooth local
curvature. Now, the distance for each specific species can be calculated by computing distance
from locally present grid points.
7.4 Results and Discussion
7.4.1 Conformational landscape of N17-Qn in solution
First, we attempt an exploration of the conformational landscape and structural features for
N17 with varying lengths of poly-Q tract. Towards this, we ran simulations with 7, 15, 35 and
45 trailing GLN in aqueous solution at varying temperatures. We reweighted the trajectory data
(by Multi-state Benett Acceptance Ratio - MBAR) at 300K, and created potential of mean force
plots using asphericity as the reaction coordinate [Fig. 7.3] [269]. Asphericity computes how
spherically asymmetric a particular structure is, and can be used as a measure for globular nature
of a structure. Asphericity, b can be measured in terms of principal moments of gyration tensor
120
(?x, ?y and ?z).
1
b = ?z ? (?x + ?y)
2
The values of asphericity can range from 0 being complete spherically symetric, and 1
being planar.
Figure 7.3: Free energy landscape of a single N17 with varying Qs, in solution with asphericity
as the reaction coordinate
Our simulations suggest a trend towards more globular (decreasing asphericity) peptide
structures with progressively increasing lengths of polyglutamine tract. This decreasing asphericity
with increasing Q-repeats was also recently noted in an atomistic study coupled with replica
exchange molecular dynamics [304].
121
Fig. 7.4 shows the overview of the conformations we capture with our simulations. Fig.
7.4-a presents the number of particular heavy atom contacts (cutoff = 7A?) between each fragment
of the peptide ? N17-N17, N17-poly-Q and poly-Q-poly-Q for varying trailing poly-Qs. Here,
the number of contacts between N17 and Poly-Q does not change with increasing number of
Qs, while poly-Q-poly-Q contacts increase. This suggests a significant specificity towards Q-Q
interactions that are also noted in previous experimental and computational reports [300?302].
Figure 7.4: a - Number of contacts between different fragments of the peptide. b - Number of
peptide-water contact per residue. The lighter shade is indicative of increase in number.
Fig. 7.4-b presents the average number of water contacts per each residue belonging to
a particular fragment of the peptide. There is an effective decrease in how solvated particular
regions (N17/poly-Q) of the peptide are, with the faster decrease for the poly-Q fragment. This
corroborates with previous experimental and computational evidences that suggest water can
behave as a poor solvent for GLN polymers [300?302].
For a more fine-grained perspective, we created contact maps of the backbone for each
peptide (Fig. 7.5). Here, a contact is defined by primary backbone interaction center within 7
A? of each other. We created these heatmaps by reweighting trajectories across the temperature
122
Figure 7.5: Reweighted contact (backbone CG interaction sites within 7 A?) map for different
peptides ? N17-7Q (a), N17-15Q (b), N17-35Q (c) and N17-45Q (d). The N17 and poly-Q
regions are marked in green and black respectively.
123
range through MBAR. Contiguous contacts with positive or negative slope (off the diagonal)
marks sheet-like features. While near-diagonal contacts (i,i+4) are suggest ?-helical structures.
We noted a gradual change from helical-rich structures to increasing sheet-like features with
increase in number of trailing GLN residues. Representative snapshots of single peptide structure
is provided in Fig 7.6
While computational methods are well-poised to study secondary structure of these peptides,
the results have been rather vague and contradictory. While some reports have pointed to primarily
helical structures in the complete exon, others have noted some sheet factions. Recent experiments
by Urbanec et al. used site-specifically labelled NMR samples to pose an ensemble view of
htt-exon1 structure [326]. They suggested an increase in propensity for extended structures at
later positions of the poly-Q tract, and a competition between helix and extended structures at
initial positions. This propensity for increased ?-sheets is also corroborated by a recent atomistic
Replica-exchange Molecular Dynamics study by Kang et al. [304]. Our observations suggest
support for the hypothesis of increase in sheet-like structures, following the addition of GLN
residues.
7.4.2 Aggregation of N17-Qn in solution
Similarly, we also explored the conformational landscape of large protein aggregates, through
simulations of six copies (? 6.9 mM) of each peptide (N17-7Q, N17-15Q, N17-35Q and N17-
45Q). The plots on Fig. 7.7 present inferences on aggregate?s structure, with information re-
weighted across multiple temperature to 300K by MBAR. Representative VMD snapshots are
provided in Fig. 7.8.
124
Figure 7.6: Representative VMD snapshots at 300K for a-N17-7Q, b-N17-15Q, c-N17-35Q, d-
N17-45Q. Color scheme - N17 backbone: Red; Poly-Q backbone: cyan; PHE sidechain: Orange
125
Figure 7.7: Contacts between different domains in the aggregate structure. a - Number of contacts
between N17-N17 (blue) and N17-Poly-Q (green). b - Number of GLN-GLN contacts. c - Water
solvation per residue of N17 (blue), poly-Q (green). d - PHE-PHE contacts (blue) and PHE
solvation (green). All the plots here are a function of the length of polyglutamine tract.
126
Fig. 7.7-a shows the the number of N17-N17 (blue) and N17-Q (green) contacts with
increase in the length of poly-Q. Here, we observed a significant decrease in the number of N17-
N17 contacts, with progressive increase in poly-Q. This decrease is compensated by an increase in
N17-Q contacts. Fig. 7.7-b shows the increase in the number of GLN-GLN contacts with increase
in GLN lengths. We note an up to eight fold increase in the number of GLN-GLN contacts (also
noted from VMD snapshots of Fig. 7.8), which similar to the single peptide case also suggests the
specificity of interactions between glutamines. This specificity was also noted in simulations by
Chen et al. [324]. Fig. 7.7-c shows the hydration per residue of N17 (blue) and poly-Q (green).
While, N17 shows a slow dehydration with increase in the length pf poly-Q tract, the solvation
of the poly-Q tract itself is not affected. The overall solvation is significantly higher for the N17
compared to the poly-Q, which is also an indication of water being a poor solvent, and the N17-
repeats situated closer to the periphery compared to the larger core created by the poly-Q repeats.
This feature is also evident in Fig. 7.7-d, which shows the number of Phenylalanine residue
contacts (blue lineplot) in different simulations. We observe a significant drop in the number
of PHE-contacts from N17-7Q/N17-15Q to more pathological N17-35Q/N17-45Q. Moreover
the number of PHE-water contacts (Fig. 7.7-d, green) increases, which establishes N17 as an
interfacial domain (also Fig. 7.8-d) in the case of aggregate structures for longer poly-Q tracts.
These structural measures suggest that there is a reduction in the size of the hydrophobic core,
with N17 distributed at the interfacial region of the aggregate. This is in contrast to the traditional
view of hydrophobic cores for aggregates/micelles. Here, at longer lengths of the poly-Q tract, the
glutamines drive aggregation and a polar core develops with GLN at the aggregate-center, similar
to the case of reverse micelles [327]. This disrupts the number/size of the hydrophobic core led
by bulky PHE groups and delegates N17 to the periphery of the peptide, where hydrophobic
127
groups are more water exposed.
To understand backbone-backbone interactions and get a fine-grained perspective of secondary
structures in our protein aggregates, we created re-weighted backbone contact-maps (Fig 7.9).
These contact maps show the expectation value for each possible inter and intra peptide backbone-
backbone contact (cutoff 7 A?) re-weighted to 300K. The contacts between groups which have
bonds (i, i?1) have been removed.
With increasing length of poly-Q tract we note an increase in sheet-like features. Contiguous
contacts with positive or negative slope (off the diagonal) marks sheet-like features. While
near-diagonal contacts (i,i+4) are ?-helical structures. Similar to the single peptide simulations,
here we also note a change from helical-rich structures to sheet-like features with increase in
number of trailing GLN residues. This fibrillation is particularly high for the pathological N17-
45Q fragment, in line with the current expectation of increased fibrillar patches at longer and
pathological poly-Q fragments [297?299].
7.4.3 Membrane interaction of a single N17 peptide
Next, we probe into peptide membrane interaction of Huntingtin protein. Since, N17 is the
membrane binding domain, we would focus on these initial 17 residues only. As a first set of
simulations, we analyze the interaction of a single N17 peptide with a planar membrane. These
simulations are aimed to understand the pathways of peptide?s partitioning onto the membrane,
and peptide folding as a consequence of peptide-membrane interaction.
We created the dihedral angle box-plot for the peptide backbone with different slices of
the trajectory [Fig. 7.10a/b - 0-10 ns; Fig. 7.10c/d - 40-50 ns, Fig. 7.10e/f - 90-100 ns; Fig.
128
Figure 7.8: Representative VMD snapshots of peptide aggregate for a-N17-7Q, b-N17-15Q,
c-N17-35Q, d-N17-45Q. Color scheme - N17 backbone: Red; Poly-Q backbone: cyan; PHE
sidechain: Orange
129
Figure 7.9: Re-weighted contact (backbone CG interaction sites within 7 A?) map for different
peptide aggregates ? N17-7Q (a), N17-15Q (b), N17-35Q (c) and N17-45Q (d). The N17 and
poly-Q regions are marked in green and black respectively.
130
7.11a/b - 140-150 ns, Fig. 7.11c/d - 190-200 ns]. The horizontal black line on the boxplots mark
50.4 degrees, which is the preferred backbone dihedral angle for ?-helices as noted from the
non-redundant protein structures we scraped from the protein data bank [see Chapter 5, section
6.3.2].
Our simulations show a gradual progression from a dominantly unstructured peptide in
solution, to a more structured ?-helical patch on the membrane. This can be noted from a gradual
reduction in variation of backbone dihedral with time, moving closer to 50.4 degrees. The peptide
forms fast initial contact with the membrane within 15-20 ns, followed by gradual membrane
absorption by 45-50 ns [Fig. 7.12]. This initial contact is led by THR (cyan), MET (green) and
to some extent LYS (yellow) co-located at the N-terminus of the peptide. The peptide, while
disordered, forms a hydrophobic core in solution through the phenylalanine groups. Now, after
the initial contact, between 20-40 ns, these PHE groups partition onto the membrane. This starts
off a slow secondary structure re-adjustment phase, completing the helical transformation by 160-
170 ns. Here, we record that significant rearrangement needs to be done at A10-L14 fragment
to break the PHE-PHE interaction in favor of interaction between PHE and the acyl tails, that
reshapes the peptide into a proper helix.
This preference towards structured ? helices has been noted for N17 in both membranous
and micellar environments. Michaelek et al noted a pronounced transition into helical conformation
in presence of DPC micelle (2D solution NMR) and POPC membrane (solid state NMR) [315].
Stable membrane associated ?-helix for N17 is also corroborated by several atomistic and coarse-
grained computational work [317,318]. Our work, validates this observation of increased helicity
in presence of membranes and points to a possible mechanism for this structural transition.
Fig. 7.13 shows the distribution of the angle that the regression line fitted to the peptide
131
Figure 7.10: a, c, e - Boxplot for backbone dihedral angles, with median marked in orange
for different trajectory slices. The black horizontal line corresponds to 50.4 degrees. (a) 0-10
ns; (c)40-50 ns; (e)90-100 ns; b - A representative snapshot corresponding to 0-10 ns; d - A
representative snapshot corresponding to 40-50 ns; f - A representative snapshot corresponding
to 90-100 ns Coloring scheme: Magenta:peptide backbone; Orange: Phenylalanine; sidechain;
Green:Other hydrophobic groups sidechain; Cyan:Polar residue sidechains; Yellow: Lysine
sidechains
132
Figure 7.11: a, c - Boxplot for backbone dihedral angles, with median marked in orange
for different trajectory slices. The black horizontal line corresponds to 50.4 degrees. (a)
140-150 ns; (c)190-200 ns; b - A representative snapshot corresponding to 140-150 ns; d
- A representative snapshot corresponding to 190-200 ns Coloring scheme: Magenta:peptide
backbone; Orange: Phenylalanine; sidechain; Green:Other hydrophobic groups sidechain;
Cyan:Polar residue sidechains; Yellow: Lysine sidechains
133
Figure 7.12: Normal distance between peptide center-of-mass and the center-of-mass of lipid
headgroups.
Figure 7.13: Distribution of helical-tilt with bilayer normal. Here the last 30 ns of trajectory has
been used for calculation.
134
backbone makes with the bilayer normal, with the data from the last 30 ns (when the peptide
has transformed into more structured ?-helix). The peptide stays in an almost-lateral orientation
with the lipid membrane, with hydrophobic groups facing the membrane and hydrophobic/polar
groups more solvent exposed.
This more inserted C-terminus structure, and the relatively tilted (more than 90 degrees)
orientation of N17 is noted in previous two dimensional solid state NMR spectra by Michaelek et
al. [315]. In our work, we could note similar structures for membrane-associated N17. Moreover,
we also delineate pathways for membrane association, which is led by the charged/polar N-
terminal residues, and final membrane insertion and restructuring by the hydrophobic groups in
the C-terminus.
These simulations also serve as a validation of the use of our CG model (ProMPT) towards
peptide-membrane simulations; and membrane-assisted structural transitions.
7.4.4 Multi Peptide (N17) simulations in presence of a curved membrane
In the next phase, we extended our scope to study peptide aggregation in presence of a
curved membranes. We started from peptides solvated in a grid-like fashion away from the
membrane, with minimum peptide-membrane separation greater than 6 nm. The peptides could
interact with each other, and the membrane in an unbiased manner. Three independent replicas
were simulated for this biomolecular system for 300-350 ns. While, the number of absorbed
peptides were different across the simulations, the observed biophysical mechanism were similar.
We have added the VMD snapshot for replica 2 (Fig. A.1) and 3 (Fig. A.2) in Appendix A. The
following analysis in this chapter is for the first replica simulation of N17 peptide in presence of
135
curved membrane.
Fig. 7.14-7.15 shows a gradual progression of the peptide from its solvated state to membrane-
bound state, with the relevant lipid-peptide contacts (a) and the corresponding VMD snapshots
(b). The peptides are absorbed primarily on top of the curved region as aggregates, and with some
aggregates in solution. In another replica simulation, one of the solvated aggregate was in contact
with the planar membrane, but most of the peptides still interacted exclusively with the curved
region of the membrane. Here, we need to note that the absorbed peptides primarily reside on the
curved region of the membrane, while 64 % of the available to interact membrane surface area is
planar.
This shows the curvature sensing ability of N17 peptide previously discovered in the AFM
study with nanospheres that enforced membrane curvature [60]. The initial interactions between
the peptide and the membrane is driven through the polar and charged groups on the peptide (Fig.
7.14), with the peptides forming an aggregate in solution. The peptide aggregate consists of a
hydrophobic core driven by the hydrophobic groups and the amphipathic face of N17. Over time
(Fig. 7.14: 70-80 ns and 150-16 ns; Fig. 7.15: 200-210 ns and 330-340 ns, we note an increase in
absorption with the larger PHE groups partitioning onto the membrane and interacting with the
acyl core. Peptide-peptide interaction here can be between one aggregate solvated in solution,
and another tethered to the membrane, or through peptides/aggregates laterally diffusing on top
of the membrane. We noted similar trends for A? 16-22 peptide in Chapter 3.
To understand the structural features of the peptides absorbed on the membrane, in contrast
to the aggregate in solution, we computed the distribution of each backbone dihedral angle
(Fig. 7.16). Here, an angle of 50.4 degrees marks perfect ? helix (Chapter 6). Overall, we
note helical structures both in the solvated peptide aggregate and membrane-absorbed aggregate,
136
Figure 7.14: Characterization of membrane-peptide interactions at 70-80 ns (a/b) and 150-160
ns (c/d) : a,c-Contact map between peptide and membrane (NC3: Choline; PO4: Phosphate;
GLE: Glycerol-Ester; Plus: Positive charged amino acid sidechain; Minus: Negative charged
amino acid sidechain; POL: Polarizable amino-acid sidechains; Backbone: Peptide backbone;
PHE: Phenylalanine sidechains; Hyd: Other hydrophobic group sidechains) c,d-A representative
snapshot of peptide interacting with the membrane. Coloring scheme: blue/ochre/pink:
membrane headgroup; cyan:acyl tail; green: peptide backbone; red: Phenylalanine
137
Figure 7.15: Characterization of membrane-peptide interactions at 200-210 ns (a/b) and 330-340
ns (c/d): a,b-Contact map between peptide and membrane (NC3: Choline; PO4: Phosphate;
GLE: Glycerol-Ester; Plus: Positive charged amino acid sidechain; Minus: Negative charged
amino acid sidechain; POL: Polarizable amino-acid sidechains; Backbone: Peptide backbone;
PHE: Phenylalanine sidechains; Hyd: Other hydrophobic group sidechains) c,d-A representative
snapshot of peptide interacting with the membrane. Coloring scheme: blue/ochre/pink:
membrane headgroup; cyan:acyl tail; green: peptide backbone; red: Phenylalanine
138
with more disordered structures in solvated aggregates. Previous experiments have outlined the
presence of helix-rich structures of N17 in their aggregate form. N17 forms amphipathic helical
structures, with the hydrophobic face contributing to the hydrophobic core of the aggregate,
and the hydrophilic side facing the solvent. This forces the initiation of peptide-membrane and
aggregate-aggregate interaction through the hydrophilic residues.
On the membrane, although the helices are more structured, irregularities can be marked
near A10-S13 region. We had previously noted the role of this fragment with peptide absorption
onto the membrane. This can point to a somewhat bent helix being preferred on membranes.
Early computational investigations by Kelly et al. with simulated tempering had observed such a
population of bent double helix for N17 [317].
Fig. 7.17 shows the relative insertion of the peptides into the membrane, with distance
calculated from a membrane surface constructed with phosphate headgroup. Only peptides with
at-least one contact (? 7 A?) with the phosphate group are considered here. We observe the
highest membrane insertion for PHE sidechains, followed by other hydrophobic groups. The
least partitioned groups are the positive and negative charged sidechains. The polar backbone
and sidechains of THR and SER are sandwiched between these extremes.
Fig. 7.18 shows the lipid density (Number of lipid headgroups per unit surface area) at
330 ns for a - simulation with N17 peptides; b - control curved membrane simulation without
any peptides. The lipid density is lower in the curved membrane which results in peptides
preferentially interacting with the curved region to engage with the membrane defects. Fig. 7.19
is the box-plot of area-per-lipid calculated at each time interval. We also mark the slow growth
in these defects, with larger gaps at increased times when lateral absorption and aggregation of
peptides is complete. This can also be noted in the long tailed distribution of the area-per-lipid
139
Figure 7.16: Boxplot marking distribution of peptide backbone dihedrals a-Peptides absorbed
onto the membrane; b-Peptides as aggregates in solution
140
Figure 7.17: Relative distance of different groups from the membrane surface (determined by
PO4). BB: Peptide Backbone; SC: Side-chain
between the control simulation and the simulation with peptide present (Fig. 7.20).
Finally, to further elucidate the role of bulky hydrophobic groups that we characterized as
an essential part in membrane partitioning, we utilized a simulation of the F11L/F17L. A VMD
snapshot of the membrane with the peptide aggregate is provided for 750 ns (Fig. 7.21). In our
simulation, we did not observe a preference towards the curved membrane region here, with one
peptide absorbed on the curved region compared to 6 on the planar region (Fig. 7.22). Instead, the
peptide/peptide aggregate interacted both on the planar and the curved membrane. The peptides
here, also could either partition onto the membrane as a single peptide or as a solvated aggregate.
The helicity of the peptides are not affected (Fig. 7.23).
This could be because of the larger size of PHE groups, which makes them difficult to
partition into membranes with smaller defects. Therefore, PHE prefers larger defects in the
141
Figure 7.18: Snapshot of lipid density at 350 ns. a - Simulation with absorbed peptide. One PHE
sidechain per peptide marked with black. b - Control simulation without peptides.
142
Figure 7.19: Boxplot of area per lipid over simulation time for the simulation with peptides on a
curved membrane.
Figure 7.20: Comparison of area-per-lipid between the simulation with peptide and the control
simulation without peptides.
143
Figure 7.21: VMD snapshot at 750 ns for a F11L/F17L simulation. Color scheme - peptide
backbone: Green; Leucines: Red; Membrane in surface representation: Black
Figure 7.22: Density of lipid groups with center-of-mass of absorbed peptides marked in black.
144
Figure 7.23: Boxplot of backbone dihedrals for a)peptides in contact with the membrane b)
peptides not in contact with the membrane. The black horizontal line at 50.4 degrees marks
backbone dihedral to match alpha helix structure.
145
curved region. On the other hand, the smaller LEU sidechain could partition into both the larger
and smaller defects without any issues. This establishes the role of bulky groups such as PHE in
curvature sensing of huntingtin protein. Two modes of amphipathic helices sensing curvature has
been outlined in literature for smaller peptides ? imbalance between polar/hydrophobic groups
such as for ?-synuclein and presence of bulky groups in the hydrophobic face for peptides such as
ALPS and ArfGAP1 [168,328]. In this work, we position N17 into the second bin. This presence
of PHE groups at the later part of N17 domain drives the curvature sensing nature of the peptide.
7.5 Conclusion
In this chapter, we focused on a slightly complex peptide sequence ? Huntingtin Exon
1 (htt) with the updated CG model (ProMPT - that we developed in the previous chapter) to
elucidate the role of flanking sequence and membrane curvature. Huntington?s disease belongs
to the larger class of diseases formed due to mutations in the poly-Q tract. The Huntingtin
protein?s N-terminal domain is characterized by an initial 17 residue N17 domain followed by
a poly-Q tract. While previous research have presented a consensus overview of the structure
and dynamics of N17 fragment, a similar overview of ther poly-Q tract is missing. There have
been competing evidences of secondary structure in the poly-Q tract. Moreover, since this is
a fast aggregating peptide, it is difficult to effectively discern between individual peptides and
peptides in their aggregate form through structural studies. Here, we explored the conformations
of N17-Qn (where n ? {7, 15, 35, 45}), both as a single peptide and peptides in their aggregate
form. For the single peptide, we noted a trend towards more globular (decreasing asphericity)
peptide structures with increasing lengths of polyglutamine tract. We also observed a significant
146
specificity towards Q-Q interactions also observed in previous experimental reports. Finally, we
also found a gradual change from helical-rich structures to increasing sheet-like features with
increase in number of trailing GLN residues. For the case of peptide aggregates, we found
that the GLN-repeats drove peptide aggregation. The aggregate core was made with the GLN-
GLN interactions that disrupted the number of N17-N17 contacts, with N17 situated more at
the periphery of the aggregate. In another thread here, we also studied the structural features
of the POPC membrane and N17 peptide, that drives the curvature sensing of the Huntingtin
protein. Previous research on peptide-membrane interactions has been limited both from a
experimental and computational perspective. While some structural features of N17-membrane
interaction are known, detailed biophysical reports of peptide-membrane interactions are still
missing. In this chapter we aim to understand these biomolecular details of peptide-membrane
interaction and the molecular mechanism of curvature sensing. Our simulations suggest a gradual
progression from a dominantly unstructured peptide in solution, to a more structured ?-helical
patch on the membrane. In presence of a curved membrane, we observed that N17 peptides
preferentially interact with the curved region. While the polar and charged groups drive the
initial membrane-peptide interaction, the membrane curvature sensing is dominantly controlled
by the bulky hydrophobic groups such as PHE. Replacing these bulky residues from the sequence
leads to loss of this curvature sensing ability of N17.
147
Chapter 8: Thesis Summary
The objective of this thesis was to understand various aspects of protein aggregation associated
with neurodegenerative diseases. In particular, I explored how physiological biomolecules such
as the presence of a membrane or solvated small bio-molecules and genetic mutations can affect/alter
aggregation behavior. As case studies, I focused on Amyloid-beta (A?, associated with the
Alzheimer?s disease) and Huntingtin (htt, associated with the Huntington?s disease). In this work,
I explored various modalities on how protein aggregation is impacted by external stimulus such
as membrane headgroup charge, applied surface-tension, hyperglycemic conditions, membrane
curvature and added trailing GLN repeats.
We approached this research through the lens of computer simulations. Molecular dynamics
is a class of computational methods that use statistical physics based underpinnings to present a
description of a biomolecular system. Here, we represent a molecule as a collection of sites,
that interact among themselves through a defined potential function. This generates a trajectory
of moving sites, which map to the motions in that molecule. Molecular dynamics simulations
aim to complement biomolecular experiments by addressing certain limitations that experiments
pose such as structural transience and competing explanations of experimental observations.
Among classical molecular dynamics simulation methods, all-atom simulations are the most
commonplace. Here, all the atoms (or all heavy atoms) in a biomolecular system are represented
148
as interaction sites. But, this method can be inaccessible for studying larger spatiotemporal
systems. To address this, we have followed a more reduced resolution approach of spatial
coarse-graining. Here, we spatially group atoms into larger interaction site, and define interaction
potential between them. This allows for a significant reduction in computational effort, allowing
studies of complex and large systems.
In this thesis I modified and applied the coarse-grained model that was developed in our
lab ? Water Explicit Polarizable Coarse-Grained Model (WEPCGM) to study the effects of
membrane headgroup charge, applied surface-tension and hyperglycemia on A? 16-22 aggregation.
In chapter 2, I introduced the WEPCGM and detailed its properties. The coarse-grained model
features explicit representation of the environment, that allows for studying environment assisted
structural changes. The presence of explicit solvent also aids in creating a transferable forcefield
without a need for reparameterization in order to study another biomolecular system. The current
version of this forcefield consists of two segments ? Water Explicit Polarizable MEMbrane
Model (WEPMEM) and Water Explicit Polarizable PROtein Model (WEPPRO) [80?82]. In
the next chapters (Chapter 3, 4 and 5) we applied WEPPROM in conjunction with WEPMEM
towards presenting a biophysical picture of A? 16-22 aggregation.
In chapter 3, we used WEPCGM to to investigate the effect of lipid headgroup charge ?
zwitterionic (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine: POPC) and anionic (1-palmitoyl-
2-oleoyl-sn-glycero-3-phospho-L-serine: POPS) ? on A? 16?22 peptide aggregation. Our coarse
grained molecular dynamics simulations provided a mechanistic explanation to a broad spectrum
of experimental results [25, 26, 159, 164, 165]. We characterized multiple pathways for peptide
absorption into membranes composed of POPC and POPS. Both lipid molecules have distinct
effects on aggregation patterns of absorbed peptides. While rapid cumulative aggregation (ordered
149
+ disordered) was observed in zwitterionic PC bilayer, the emergence of ordered beta sheets
and by extension, fibrillation was faster in presence of anionic POPS lipids. The results are
in agreement with previous experimental studies [25, 137, 139?142, 164, 165] that had observed
faster growth of amyloid fibrils in presence of anionic lipids. The discrepancy in the cumulative
aggregation rates is a consequence of faster lateral diffusion of POPC lipid molecules compared
to POPS. On the other hand, increased beta sheet content in POPS membranes is due to the
differences in membrane compressibility. Higher membrane compressibility of POPC membrane
compared to POPS, results in a relatively higher peptide insertion into the bilayer. This distorts
the geometry of individual peptide molecules which hinders their participation in beta sheet
formation. Some of the morphological aspects of membrane assisted A? aggregation reported
in this study such as relatively higher membrane insertion of F19 compared to F20, have been
supported by previous experimental evidences [159]. We also revealed the propensity of initial
oligomeric deposits to act as nucleation seeds to enhance further fibrillation. Considering the
presence of multiple, ordered aggregates in POPS due to slow cumulative aggregation and peptide
aggregates operating as nucleation seeds, POPS membranes can have an increased progressive
peptide aggregation rates. This work unravels how lipid headgroup driven biochemical interactions
in homogeneous model membranes shape peptide absorption and aggregation.
In chapter 4, we leveraged molecular simulations to address and understand relationships
between curved cellular membranes and aggregation of a model template peptide A? 16-22.
Experimental results have suggested that both membranes and peptide can influence each other?s
structure and dynamics. Membrane curvature can promote extensive fibrillation, with faster
formation of ordered structures in presence of small unilamellar vesicles, compared to larger
ones. On the other hand, peptide aggregates can alter membrane?s structure and packing. In
150
this work we investigated these intertwined effects with coarse-grained molecular simulations.
Our results agree with previous observations of curvature-driven peptide aggregation into ordered
structures, and suggest possible biophysical mechanisms for it [27,28]. Membranes with increased
curvatures lead to increased peptide absorption, due to more exposed hydrophobic defects. The
absorbed peptides can then laterally diffuse around, and interact through the peptide backbone to
form ordered fibrillar structures. The presence of such peptide aggregates also affects the lipid
membrane?s local structure. The lipid groups closer to the aggregates are more ordered at the
crowded interfacial region due to interfacial presence of peptides; and less ordered deep inside the
membrane core. These observations align with previous experiments and molecular simulations
that also highlighted membrane disruption by peptides [179, 180]. Our results supported the
previously suggested ?carpeting model? of membrane disruption, by increasing membrane fluidity
inside the membrane core [166, 181]. These locally close lipids also have a broad distribution in
P?N vector tilt from the membrane normal, suggesting the heterogeneous nature of peptide-lipid
interactions. Such local variations in P?N vector tilt can manifest in changes to the membrane?s
electrostatic potential, ion distributions and alter electrostatics-assisted membrane signalling [182,
183]. This study unravels the effects of membrane curvature in aiding peptide aggregation, and
the effects of peptide aggregation in reshaping membrane?s local attributes.
In chapter 5, we explored the mechanisms through which solvated glucose molecules
affect A? 16-22 aggregation in a concentration dependant manner. Previous reports have drawn
correlations between pathogenesis of Alzheimer?s disease and hyperglycemic conditions associated
with type 2 diabetes. Experimental research has proposed chemical crosslinking, in presence of
glucose molecules as responsible for increased toxicity. But the time-scale separation between
increased aggregation and formation of chemical cross-links suggests the presence of an alternate
151
thermodynamic mechanism, which we explored in this chapter. We applied coarse-grained
molecular dynamics simulations to study the how glucose molecules can drive A? 16-22 aggregation.
We noted that that increasing the concentration of solvated glucose molecules can result in faster
aggregation of A? 16-22, without any appreciable change in ?-sheet content. This change
in aggregation rates can be explained in terms of relative orientations of interfacial glucose
molecules. The glucose molecules at the peptide-water interface feature a preferential orientation,
resulting in loss of rotational entropy in a concentration dependent manner. This can assist
in faster aggregation of the peptide, to reduce the cumulative availability of solvent accessible
surface-area.
In chapter6, we revisited our coarse-grained model again and outlined its limitations. Firstly,
WEPPROM does not have representation for all the essential amino-acids. Secondly, it has not
been studied and validated for tertiary folding protein structures. To address these factors, we
developed a transferable CG forcefield (Protein Model with Polarizability and Transferability
- ProMPT) with an explicit representation of the environment for accurate simulations with
proteins. The forcefield consists of a set of pseudo-atoms representing different chemical groups
that can be joined/associated together to create different biomolecular systems. This preserves
the transferability of the forcefield to multiple environments and simulation conditions. We added
electronic polarization that can respond to environmental heterogeneity/fluctuations and couple
it to protein?s structural transitions. The non-bonded interactions are parametrized with physics-
based features such as solvation, and partitioning free energies determined by thermodynamic
calculations and matched with experiments and/or atomistic simulations. The bonded potentials
are inferred from corresponding distributions in non-redundant protein structure databases. We
presented validations of the CG model with simulations of well-studied aqueous protein systems
152
with specific protein fold types- TRP-cage, TrpZip4, Villin, WW-domain and ?-?-?. We also
explored the applications of the forcefield to study aqueous aggregation of A? 16-22 peptides
and dimerization of glycophorin A (GPA) in micellar environments.
With this model we primarily address the shortcomings of WEPPROM. ProMPT now has
access to all the amino acid representation. It has updated bonded and non-bonded interaction
parameters with which we validated the conformational landscape of several small proteins and
aggregates.
In chapter 7, I focused on a more complex peptide sequence ? Huntingtin protein?s N-
terminal domain (htt) with the updated CG model (ProMPT) to elucidate the role of varying
lengths of trailing sequence and membrane curvature. Huntington?s disease belongs to the larger
class of diseases formed due to mutations in the poly-Q tract. The htt is characterized by an initial
17 residue N17 domain followed by a poly-Q tract. While previous research have presented a
consensus overview of the structure and dynamics of N17 fragment, a similar overview of ther
poly-Q tract is missing [315]. There have been competing evidences of secondary structure in
the poly-Q tract [304, 318]. Moreover, since this is a fast aggregating peptide, it is difficult
to effectively discern between individual peptides and peptides in their aggregate form through
structural studies. Here, we aim to explore the conformations of N17-Qn (where n ? 7, 15, 35,
45), both as a single peptide and peptides in their aggregate form.
For the single peptide, we could mark a trend towards more globular (decreasing asphericity)
peptide structures with increasing lengths of polyglutamine tract. We also observed a significant
specificity towards Q-Q interactions also observed in previous experimental reports. Finally,
we also found a gradual change from helical-rich structures to the growth of sheet-like features
with increase in number of trailing GLN residues. For the the case of peptide aggregates,
153
we could establish that the GLN-repeats drove peptide aggregation. With longer poly-Q tract,
the balance between hydrophobic interactions and polar interactions changes with GLN-GLN
contacts creating a polar core of the aggregate.
In another thread here, we also studied the structural features of the POPC membrane and
N17 peptide, that drives the curvature sensing of the Huntingtin protein. Previous research on
peptide-membrane interactions has been limited both from a experimental and computational
perspective. While some structural features of N17-membrane interaction are known, detailed
biophysical reports of peptide-membrane interactions are still missing [60, 315]. In this chapter
we aim to understand these biomolecular details of peptide-membrane interaction and the molecular
mechanism of curvature sensing.
Our simulations propose a gradual progression from a dominantly unstructured peptide
(with some helicity) in solution, to a more structured ?-helical patch on the membrane. In
presence of a curved membrane, we observed that N17 peptides preferentially interact with
the curved region. Here, we noted that while the polar and charged groups drive the initial
membrane-peptide interaction, the membrane curvature sensing is dominantly controlled by the
bulky hydrophobic groups such as PHE. Omission of these bulky residues from the sequence
leads to loss of this curvature sensing ability.
In summary, this thesis examined aggregation of neurodegenerative peptides associated
with Alzheimer?s disease (A?) and Huntington?s (htt) disease and how this self-assembly is
impacted by external factors. These factors can change both the structure and kinetics of peptide
aggregation. In the case of both A? 16-22 and N17, the initial aggregation was driven by the
hydrophobic core of the peptide. But the presence of a membrane initiated a competition between
peptide-peptide and peptide-membrane interactions. In the case of A? 16-22 this membrane
154
interaction resulted in fibrillar structures, while in the case of N17 the peptide conformed into
structured amphipathic interfacial helices. This suggests the ability of the membrane-water
interface to render particular secondary structures and tertiary arrangements to these intrinsically
disordered fragments.
In addition, we also discovered the importance of bulky hydrophobic groups in defining
overall organization of the aggregate. For A? 16-22, these groups were essential for engaging
with the membrane core, thereby allowing peptides to initiate contacts through the peptide backbone.
On the other hand, for htt, these bulky groups were instrumental in orienting the peptides appropriately
on membrane surface and for curvature sensing. These groups can be essential targets for rational
design of therapeutics. For both N17 and A? 16-22, the aggregate core is primarily driven by the
hydrophobic groups. In contrast, the presence of poly-Q segment initiated another set of strong
competing interactions through the trailing GLN. These led to these GLN-repeats replacing the
hydrophobic groups in forming the primary aggregate core at longer lengths of poly-Q tract,
similar to a reverse micelle structure. This increased GLN repeats formed highly stable ?-sheet
rich structures, contrary to previously unstructured, and non-specific hydrophobic core.
Beyond specific interactions with the membrane, we also curated the effects of non-specific
interactions that solvated small molecules have on aggregation rates. Here, we observed a
concentration dependent effect driven by a specific way in which glucose molecules partition
onto the aggregate surface.
Another contribution of this thesis is the development of a polarizable coarse-grained
model for simulation of proteins in explicit environment. The key aspect of this forcefield
is that it is transferable across biomolecular systems and does not require fitting to explicit
biomolecular systems. This allows to decouple and study the effects of the environment on
155
biophysical processes. As such, it can be a adopted by the structural and molecular biophysics
community towards large spatiotemporal studies of biomolecular systems of increased complexity.
8.1 Future Work
In this thesis we explored the effects of model membranes made with a single type of
lipids (either 100% POPC or 100% POPS) to study peptide aggregation. To present a somewhat
realistic scenario, the next objective can be to extend this study to binary mixture of lipid groups
at varying ratios. This would present an in-depth overview of gradual introduction of membrane
charge on aggregation behavior.
Beyond charge on lipid molecules, solvated divalent cations have been shown to affect
peptide aggregation directly. Studies with surface plasmon resonance (SPR) and atomic force
microscopy (AFM) have suggested the role of calcium ions in bridging Glutamate (residue 22)
and the phosphate of lipids, anchoring the peptide to membrane-surface and driving aggregation
[329]. Also, the local presence of calcium has been known to induce mesoscopic changes in
bilayer organization and geometry. The binding of calcium ion with the model membranes such
as vesicles has been investigated primarily with thermodynamic, microscopic, and spectroscopic
techniques [330?335]. Several reports suggest lateral reorganization of lipid mixtures as a response
to the presence of calcium ions. Doosti et al. used the localized application of calcium ions on
vesicles enriched with negatively charged lipids to generate tubular protrusions [333]. Recently
Graber et al. showed the development of negative curvature as a consequence of anionic lipid?
phosphatidylinositol-4,5-bisphosphate (PI(4, 5)P2) and phosphatidylserine (PS) clustering in the
presence of calcium ions [334]. Therefore, presence of calcium can result in local enrichment and
156
curvature changes, which in-turn can alter membrane?s interaction pattern with peptides. These
physico-chemical changes of lipid molecules in response to calcium ions can therefore lead to
differential peptide aggregation properties.
In addition, investigations into a more realistic ternary ? membrane, A? 16-22 peptide
and calcium ions system is limited due to associated computational complexities. CG-MD can
be an essential tool to investigate the complex effects of membrane assisted peptide aggregation
in presence of calcium ions. We have already developed a coarse-grained model for calcium
ions to work with WEPMEM [174], that can capture calcium driven demixing of binary lipid
mixture (POPC and POPS). We plan to apply these coarse-grained models towards elucidating
the microscopic interactions that shape aggregation in such complex systems.
The studies of peptide self-assembly, in this thesis, have always considered crowding
biomolecules in isolation. The specific and non-specific interactions common in crowded systems
such as an extra-cellular matrix (ECM) would be a more reasonable description of the fibrillation
process. Hence, there is a great deal of interest in studying the aggregation process in with a
realistic model for that environment. Recent reports have suggested of a particular biomolecule
? chitosan (CHT), that can act as a mimetic for the extra-cellular matrix [336, 337]. Chitosan
is a linear co-polymer composed of ?-(1, 4) linked N-acetylglucosamine (GlcAc) units and
glucosamine (GlcN) units [337].
A complete mechanistic picture of CHT self-assembly and CHT-A? interactions is missing.
Molecular simulations can be an useful approach to connect microscopic level details to macroscopic
statistical properties. Recent efforts in this direction has yielded several insights into atomic
details of CHT self-assembly. But the associated length scales have made it difficult to study
CHT self-assembly processes with atomistic molecular dynamics. Our lab has recently published
157
a coarse-grained model for CHT molecule that can capture both atomic micro-interactions and
macroscopic mechanical properties of CHT hydrogels [338]. This positions us to study CHT-A?
interactions by decoupling individual effects such as specificity of monomeric interactions and
effects of polymeric structures.
For Huntington?s disease, we explored the curvature sensing nature of Huntington?s disease
and the impact different lengths of poly-Q fragment has on peptide aggregation. Beyond that, we
plan to study how poly-Q repeats could affect membrane interactions and membrane assisted
aggregation. Moreover, we plan to increase the peptide length to study proline rich domains that
have been known to have specific impacts on inhibiting peptide aggregation. Future plans also
involve studying similar other amphipathic helices to provide an in-depth perspective of curvature
sensing.
The coarse-grained models developed in our lab is always under constant updates. For
the protein model, we are planning appropriate modifications to side-chain size, can achieve
adequate protein-core packing required for correct tertiary structures. Along other directions,
the model is currently being used to understand ion-assisted folding and unfolding of proteins
due to explicit representation of the environment and transferability of interaction potential.
The membrane model is also currently being expanded to include several other lipid and sterol
varieties. Moreover our lab is also working on creating a representation for non-traditional
solvents such as ionic liquids, as models for the environment.
158
Appendix A: Results From Other Replica Simulations of N17 with Curved
Membrane
Figure A.1: A representative snapshot of peptide interacting with the membrane for replica-
simulation 1. Coloring scheme: blue/ochre/pink: membrane headgroup; cyan:acyl tail; green:
peptide backbone; red: Phenylalanine
159
Figure A.2: A representative snapshot of peptide interacting with the membrane for replica-
simulation 2. Coloring scheme: blue/ochre/pink: membrane headgroup; cyan:acyl tail; green:
peptide backbone; red: Phenylalanine
160
Bibliography
[1] Christopher A Ross and Michelle A Poirier. Protein aggregation and neurodegenerative
disease. Nat. Med., 10(S7):S10?S17, 2004.
[2] Meenakshi Verma, Abhishek Vats, and Vibha Taneja. Toxic species in amyloid disorders:
Oligomers or mature fibrils. Ann Indian Acad Neurol, 18(2):138?45, 2015.
[3] Fernanda G De Felice and Se?rgio T Ferreira. Beta-amyloid production, aggregation, and
clearance as targets for therapy in Alzheimer?s disease. Cell. Mol. Neurobiol., 22(5-
6):545?63, 2002.
[4] William L Klein, Grant A Krafft, and Caleb E Finch. Targeting small A? oligomers: the
solution to an Alzheimer?s disease conundrum? Trends Neurosci, 24(4):219?224, 2001.
[5] Dominic M. Walsh and Dennis J. Selkoe. A? Oligomers ? a decade of discovery. J.
Neurochem., 101(5):1172?1184, 2007.
[6] Hai Lin, Rajinder Bhatia, and Ratneswar Lal. Amyloid ? protein forms ion channels:
implications for Alzheimer?s disease pathophysiology. FASEB J., 15(13):2433?2444,
2001.
[7] Sara M. Butterfield and Hilal A. Lashuel. Amyloidogenic Protein-Membrane Interactions:
Mechanistic Insight from Model Systems. Angew. Chem., 49(33):5628?5654, 2010.
[8] N Arispe, H B Pollard, E Rojas, Rushana Azimova, Douglas Ng, Blas Frangione, Bruce
Kagan, Jorge Ghiso, and Ratnesh Lal. Giant multilevel cation channels formed by
Alzheimer disease amyloid beta-protein [A beta P-(1-40)] in bilayer membranes. Proc.
Natl. Acad. Sci. U.S.A., 90(22):10573?7, 1993.
[9] T. R. Serio, A G Cashikar, A S Kowal, G J Sawicki, J J Moslehi, L Serpell, M F
Arnsdorf, and S L Lindquist. Nucleated Conformational Conversion and the Replication
of Conformational Information by a Prion Determinant. Science, 289(5483):1317?1321,
2000.
161
[10] Hyunbum Jang, Fernando Teran Arce, Srinivasan Ramachandran, Ricardo Capone,
Rushana Azimova, Bruce L Kagan, Ruth Nussinov, and Ratnesh Lal. Truncated beta-
amyloid peptide channels provide an alternative mechanism for Alzheimer?s Disease and
Down syndrome. Proc. Natl. Acad. Sci. U.S.A., 107(14):6538?43, 2010.
[11] Anne M. Brown and David R. Bevan. Molecular Dynamics Simulations of Amyloid ? -
Peptide (1-42): Tetramer Formation and Membrane Interactions. Biophys. J., 111(5):937?
949, 2016.
[12] Jo V. Rushworth and Nigel M. Hooper. Lipid Rafts: Linking Alzheimer?s Amyloid- ?
Production, Aggregation, and Toxicity at Neuronal Membranes. Int J Alzheimers Dis,
2011:1?14, 2011.
[13] Charles H. Davis and Max L. Berkowitz. Interaction Between Amyloid-? (1?42) Peptide
and Phospholipid Bilayers: A Molecular Dynamics Study. Biophys. J., 96(3):785?797,
2009.
[14] Giovanni Bussi and Alessandro Laio. Using metadynamics to explore complex free-energy
landscapes. Nat. Rev. Phys., 2(4):200?212, April 2020.
[15] A. Laio and M. Parrinello. Escaping free-energy minima. Proc. Natl. Acad. Sci. U.S.A.,
99(20):12562?12566, October 2002.
[16] Alessandro Barducci, Massimiliano Bonomi, and Michele Parrinello. Metadynamics.
Wiley Interdiscip. Rev. Comput. Mol. Sci., 1(5):826?843, September 2011.
[17] Yuji Sugita and Yuko Okamoto. Replica-exchange molecular dynamics method for protein
folding. Chem. Phys. Lett., 314(1-2):141?151, November 1999.
[18] Sebastian Kmiecik, Dominik Gront, Michal Kolinski, Lukasz Wieteska,
Aleksandra Elzbieta Dawid, and Andrzej Kolinski. Coarse-Grained Protein Models
and Their Applications. Chem. Rev., 116(14):7898?7936, July 2016.
[19] Sanja Po?yry and Ilpo Vattulainen. Role of charged lipids in membrane structures ? Insight
given by simulations. Biochim. Biophys. Acta, Biomembr., 1858(10):2322?2333, 2016.
[20] Jason G. Kay and Sergio Grinstein. Phosphatidylserine-Mediated Cellular Signaling. In
Daniel G.S. Capelluto, editor, Lipid-mediated Protein Signaling, volume 991, pages 177?
193. Springer Netherlands, Dordrecht, 2013.
[21] Akiko Shiratsuchi, Masato Umeda, Yoshiki Ohba, and Yoshinobu Nakanishi. Recognition
of Phosphatidylserine on the Surface of Apoptotic Spermatogenic Cells and Subsequent
Phagocytosis by Sertoli Cells of the Rat. Journal of Biological Chemistry, 272(4):2354?
2358, 1997.
[22] M Paul Murphy and Harry III LeVine. Alzheimer?s disease and the amyloid-beta peptide.
J. Alzheimers Dis., 19(1):311?323, 2010.
162
[23] Anna Itkin, Vincent Dupres, Yves F Dufre?ne, Burkhard Bechinger, Jean-Marie
Ruysschaert, and Vincent Raussens. Calcium ions promote formation of amyloid ?-
peptide (1-40) oligomers causally implicated in neuronal toxicity of Alzheimer?s disease.
PLoS ONE, 6(3):e18250, 2011.
[24] Angelo Demuro, Ian Parker, and Grace E. Stutzmann. Calcium Signaling and Amyloid
Toxicity in Alzheimer Disease. J. Biol. Chem., 285(17):12463?12468, 2010.
[25] Nagarajan Sureshbabu, R. Kirubagaran, H. Thangarajah, E. J. Padma Malar, and
R. Jayakumar. Lipid-Induced Conformational Transition of Amyloid ? Peptide Fragments.
J. Mol. Neurosci., 41(3):368?382, 2010.
[26] Francis Hane, Elizabeth Drolle, Ravi Gaikwad, Erin Faught, and Zoya Leonenko.
Amyloid-? Aggregation on Model Lipid Membranes: An Atomic Force Microscopy
Study. J. Alzheimers Dis., 26(3):485?494, 2011.
[27] Mayu S. Terakawa, Yuxi Lin, Misaki Kinoshita, Shingo Kanemura, Dai Itoh, Toshihiko
Sugiki, Masaki Okumura, and Young-Ho Lee. Impact of membrane curvature on amyloid
aggregation. Biochim. Biophys. Acta, Biomembr., 1860(9):1741?1764, 2018.
[28] Yuuki Sugiura, Keisuke Ikeda, and Minoru Nakano. High Membrane Curvature Enhances
Binding, Conformational Changes, and Fibrillation of Amyloid-? on Lipid Bilayer
Surfaces. Langmuir, 31(42):11549?11557, 2015.
[29] Kyle L. Morris and Louise C. Serpell. X-Ray Fibre Diffraction Studies of Amyloid Fibrils.
In Einar M. Sigurdsson, Miguel Calero, and Mar??a Gasset, editors, Amyloid Proteins:
Methods and Protocols, Methods in Molecular Biology, pages 121?135. Humana Press,
Totowa, NJ, 2012.
[30] Margaret Sunde, Louise C Serpell, Mark Bartlam, Paul E Fraser, Mark B Pepys, and
Colin C.F Blake. Common core structure of amyloid fibrils by synchrotron X-ray
diffraction. J. Mol. Biol., 273(3):729?739, 1997.
[31] Holger A Scheidt, Isabel Morgado, Sven Rothemund, and Daniel Huster. Dynamics of
amyloid ? fibrils revealed by solid-state NMR. J. Biol. Chem., 287(3):2017?2021, 2012.
[32] C Hilbich, B Kisters-Woike, J Reed, C L Masters, and K Beyreuther. Substitutions of
hydrophobic amino acids reduce the amyloidogenicity of Alzheimer?s disease beta A4
peptides. J. Mol. Biol., 228(2):460?473, 1992.
[33] H Inouye, P E Fraser, and D A Kirschner. Structure of beta-crystallite assemblies formed
by Alzheimer beta-amyloid protein analogues: analysis by x-ray diffraction. Biophys. J.,
64(2):502?519, 1993.
[34] D A Kirschner, H Inouye, L K Duffy, A Sinclair, M Lind, and D J Selkoe. Synthetic
peptide homologous to beta protein from Alzheimer disease forms amyloid-like fibrils in
vitro. Proc. Natl. Acad. Sci. U.S.A., 84(19):6953?6957, 1987.
163
[35] S J Wood, R Wetzel, J D Martin, and M R Hurle. Prolines and amyloidogenicity in
fragments of the Alzheimer?s peptide beta/A4. Biochemistry, 34(3):724?730, 1995.
[36] K. Aurelia Ball, Aaron H. Phillips, Paul S. Nerenberg, Nicolas L. Fawzi, David E.
Wemmer, and Teresa Head-Gordon. Homogeneous and Heterogeneous Tertiary Structure
Ensembles of Amyloid-? Peptides. Biochemistry, 50(35):7612?7628, 2011.
[37] J J Balbach, Y Ishii, O N Antzutkin, R D Leapman, N W Rizzo, F Dyda, J Reed, and
R Tycko. Amyloid fibril formation by A beta 16-22, a seven-residue fragment of the
Alzheimer?s beta-amyloid peptide, and structural characterization by solid state NMR.
Biochemistry, 39(45):13748?13759, 2000.
[38] Anil K. Mehta, Kun Lu, W. Seth Childers, Yan Liang, Steven N. Dublin, Jijun Dong,
James P. Snyder, Sai Venkatesh Pingali, Pappannan Thiyagarajan, and David G. Lynn.
Facial Symmetry in Protein Self-Assembly. J. Am. Chem. Soc., 130(30):9829?9835, 2008.
[39] Kun Lu, Jaby Jacob, P. Thiyagarajan, V. P. Conticello, and D. G. Lynn. Exploiting Amyloid
Fibril Lamination for Nanotube Self-Assembly. J. Am. Chem. Soc., 2003.
[40] Kai Tao, Jiqian Wang, Peng Zhou, Chengdong Wang, Hai Xu, Xiubo Zhao, and Jian R. Lu.
Self-Assembly of Short A?(16?22) Peptides: Effect of Terminal Capping and the Role of
Electrostatic Interaction. Langmuir, 27(6):2723?2730, 2011.
[41] Li Na Zhao, See-Wing Chiu, Je?ro?me Benoit, Lock Yue Chew, and Yuguang Mu. Amyloid
? Peptides Aggregation in a Mixed Membrane Bilayer: A Molecular Dynamics Study. J.
Phys. Chem. B, 115(42):12247?12256, 2011.
[42] Anusha Jayaraman and Christian J. Pike. Alzheimer?s Disease and Type 2 Diabetes:
Multiple Mechanisms Contribute to Interactions. Curr Diab Rep, 14(4):476, April 2014.
[43] Wenqing Xia, Shaohua Wang, Zilin Sun, Feng Bai, Yi Zhou, Yue Yang, Pin Wang, Yan
Huang, and Yang Yuan. Altered baseline brain activity in type 2 diabetes: A resting-state
fMRI study. Psychoneuroendocrinology, 38(11):2493?2501, November 2013.
[44] Chris Moran, Thanh G. Phan, Jian Chen, Leigh Blizzard, Richard Beare, Alison Venn,
Gerald Mu?nch, Amanda G. Wood, Josephine Forbes, Timothy M. Greenaway, Susan
Pearson, and Velandai Srikanth. Brain Atrophy in Type 2 Diabetes. Diabetes Care,
36(12):4036?4042, December 2013.
[45] Sanne M. Manschot, Augustina M.A. Brands, Jeroen van der Grond, Roy P.C. Kessels,
Ale Algra, L. Jaap Kappelle, Geert Jan Biessels, and on behalf of the Utrecht Diabetic
Encephalopathy Study Group. Brain Magnetic Resonance Imaging Correlates of Impaired
Cognition in Patients With Type 2 Diabetes. Diabetes, 55(4):1106?1113, April 2006.
[46] Nobuyuki Sasaki, Ryo Fukatsu, Kayo Tsuzuki, Yorihide Hayashi, Taku Yoshida, Nobuhiro
Fujii, Takao Koike, Ikuro Wakayama, Richard Yanagihara, Ralph Garruto, Naoji Amano,
and Zenji Makita. Advanced Glycation End Products in Alzheimer?s Disease and Other
Neurodegenerative Diseases. The American Journal of Pathology, 153(4):1149?1155,
October 1998.
164
[47] Yanyan Kong, Fushuai Wang, Jiao Wang, Cuiping Liu, Yinping Zhou, Zhengqin Xu,
Chencheng Zhang, Bomin Sun, and Yihui Guan. Pathological Mechanisms Linking
Diabetes Mellitus and Alzheimer?s Disease: the Receptor for Advanced Glycation End
Products (RAGE). Front. Aging Neurosci., 12:217, July 2020.
[48] Shun-Yao Ko, Hshin-An Ko, Kuo-Hsiung Chu, Tzong-Ming Shieh, Tzong-Cherng Chi,
Hong-I Chen, Weng-Cheng Chang, and Shu-Shing Chang. The Possible Mechanism
of Advanced Glycation End Products (AGEs) for Alzheimer?s Disease. PLoS ONE,
10(11):e0143345, November 2015.
[49] Niraja Kedia, Michael Almisry, and Jan Bieschke. Glucose directs amyloid-beta into
membrane-active oligomers. Phys. Chem. Chem. Phys., 19(27):18036?18046, 2017.
[50] Joana M. Gil and Ana Cristina Rego. Mechanisms of neurodegeneration in Huntington?s
disease. Eur. J. Neurosci., 27(11):2803?2820, 2008.
[51] Kimberly B Kegel-Gleason. Huntingtin interactions with membrane phospholipids:
strategic targets for therapeutic intervention? J. Huntington?s Dis., 2(3):239?50, 2013.
[52] Gillian Bates. Huntingtin aggregation and toxicity in Huntington?s disease. Lancet,
361(9369):1642?1644, 2003.
[53] Henry L. Paulson, Vikram G. Shakkottai, H. Brent Clark, and Harry T. Orr. Polyglutamine
spinocerebellar ataxias ? from genes to potential treatments. Nat Rev Neurosci,
18(10):613?626, October 2017.
[54] Craig S. McIntosh, Dunhui Li, Steve D. Wilton, and May T. Aung-Htut. Polyglutamine
Ataxias: Our Current Molecular Understanding and What the Future Holds for Antisense
Therapies. Biomedicines, 9(11):1499, October 2021.
[55] Maciej Figiel, Wojciech J. Szlachcic, Pawel M. Switonski, Agnieszka Gabka, and
Wlodzimierz J. Krzyzosiak. Mouse Models of Polyglutamine Diseases: Review and Data
Table. Part I. Mol Neurobiol, 46(2):393?429, October 2012.
[56] Kathleen A. Burke, Jordan Godbey, and Justin Legleiter. Assessing mutant huntingtin
fragment and polyglutamine aggregation by atomic force microscopy. Methods,
53(3):275?284, 2011.
[57] Justin Legleiter, Gregor P. Lotz, Jason Miller, Jan Ko, Cheping Ng, Geneva L. Williams,
Steve Finkbeiner, Paul H. Patterson, and Paul J. Muchowski. Monoclonal Antibodies
Recognize Distinct Conformational Epitopes Formed by Polyglutamine in a Mutant
Huntingtin Fragment. J. Biol. Chem., 284(32):21647?21658, 2009.
[58] Randy Singh Atwal, Jianrun Xia, Deborah Pinchev, Jillian Taylor, Richard M. Epand, and
Ray Truant. Huntingtin has a membrane association signal that can modulate huntingtin
aggregation, nuclear entry and toxicity. Hum. Mol. Genet., 16(21):2600?2615, 2007.
165
[59] Kathleen A. Burke, Kaitlin M. Hensal, C. Samuel Umbaugh, Maxmore Chaibva, and Justin
Legleiter. Huntingtin disrupts lipid bilayers in a polyQ-length dependent manner. Biochim
Biophys Acta Biomembr, 1828(8):1953?1961, 2013.
[60] Maxmore Chaibva, Kathleen A. Burke, and Justin Legleiter. Curvature Enhances Binding
and Aggregation of Huntingtin at Lipid Membranes. Biochemistry, 53(14):2355?2365,
2014.
[61] Anthony R. Braun, Michael M. Lacy, Vanessa C. Ducas, Elizabeth Rhoades, and
Jonathan N. Sachs. Alpha-Synuclein?s uniquely long amphipathic helix enhances
its membrane binding and remodeling capacity. The Journal of membrane biology,
250(2):183?193, 2017.
[62] Elka R. Georgieva, Shifeng Xiao, Peter P. Borbat, Jack H. Freed, and David Eliezer.
Tau Binds to Lipid Membrane Surfaces via Short Amphipathic Helices Located in Its
Microtubule-Binding Repeats. Biophysical Journal, 107(6):1441?1452, 2014.
[63] Manuel Gime?nez-Andre?s, Alenka C?opic?, and Bruno Antonny. The Many Faces of
Amphipathic Helices. Biomolecules, 8(3), 2018.
[64] B. R. Brooks, C. L. Brooks, A. D. Mackerell, L. Nilsson, R. J. Petrella, B. Roux, Y. Won,
G. Archontis, C. Bartels, S. Boresch, A. Caflisch, L. Caves, Q. Cui, A. R. Dinner, M. Feig,
S. Fischer, J. Gao, M. Hodoscek, W. Im, K. Kuczera, T. Lazaridis, J. Ma, V. Ovchinnikov,
E. Paci, R. W. Pastor, C. B. Post, J. Z. Pu, M. Schaefer, B. Tidor, R. M. Venable, H. L.
Woodcock, X. Wu, W. Yang, D. M. York, and M. Karplus. CHARMM: The biomolecular
simulation program. Journal of Computational Chemistry, 30(10):1545?1614, 2009.
[65] James C. Phillips, Rosemary Braun, Wei Wang, James Gumbart, Emad Tajkhorshid,
Elizabeth Villa, Christophe Chipot, Robert D. Skeel, Laxmikant Kale?, and Klaus Schulten.
Scalable molecular dynamics with NAMD. Journal of Computational Chemistry,
26(16):1781?1802, 2005.
[66] Ilian T. Todorov, William Smith, Kostya Trachenko, and Martin T. Dove. DL POLY 3:
new dimensions in molecular dynamics simulations via massive parallelism. Journal of
Materials Chemistry, 16(20):1911, 2006.
[67] James A. Maier, Carmenza Martinez, Koushik Kasavajhala, Lauren Wickstrom, Kevin E.
Hauser, and Carlos Simmerling. ff14SB: Improving the Accuracy of Protein Side Chain
and Backbone Parameters from ff99SB. J. Chem. Theory Comput., 11(8):3696?3713,
2015.
[68] Callum J. Dickson, Benjamin D. Madej, A?ge A. Skjevik, Robin M. Betz, Knut Teigen,
Ian R. Gould, and Ross C. Walker. Lipid14: The Amber Lipid Force Field. J. Chem.
Theory Comput., 10(2):865?879, 2014.
[69] Robert B. Best, Xiao Zhu, Jihyun Shim, Pedro E. M. Lopes, Jeetain Mittal, Michael Feig,
and Alexander D. MacKerell. Optimization of the Additive CHARMM All-Atom Protein
Force Field Targeting Improved Sampling of the Backbone?, ? and Side-Chain ?1 and ?2
Dihedral Angles. J. Chem. Theory Comput., 8(9):3257?3273, 2012.
166
[70] R. W. Pastor and A. D. MacKerell. Development of the CHARMM Force Field for Lipids.
J. Phys. Chem. Lett., 2(13):1526?1532, 2011.
[71] Jeffery B. Klauda, Viviana Monje, Taehoon Kim, and Wonpil Im. Improving the
CHARMM Force Field for Polyunsaturated Fatty Acid Chains. J. Phys. Chem. B,
116(31):9424?9431, 2012.
[72] Chris Oostenbrink, Alessandra Villa, Alan E. Mark, and Wilfred F. Van Gunsteren. A
biomolecular force field based on the free enthalpy of hydration and solvation: The
GROMOS force-field parameter sets 53A5 and 53A6. J. Comput. Chem., 25(13):1656?
1676, 2004.
[73] See-Wing Chiu, Sagar A. Pandit, H. L. Scott, and Eric Jakobsson. An Improved United
Atom Force Field for Simulation of Mixed Lipid Bilayers. J. Phys. Chem. B, 113(9):2748?
2763, 2009.
[74] William L. Jorgensen and Julian Tirado-Rives. The OPLS [optimized potentials for liquid
simulations] potential functions for proteins, energy minimizations for crystals of cyclic
peptides and crambin. J. Am. Chem. Soc., 110(6):1657?1666, 1988.
[75] David S. Maxwell William L. Jorgensen and Julian Tirado-Rives. Development and
Testing of the OPLS All-Atom Force Field on Conformational Energetics and Properties
of Organic Liquids. J. Am. Chem. Soc., pages 11225?11236, 1996.
[76] Mookyung Cheon, Iksoo Chang, and Carol K Hall. Extending the PRIME model for
protein aggregation to all 20 amino acids. Proteins, 78(14):2950?2960, 2010.
[77] Aram Davtyan, Nicholas P. Schafer, Weihua Zheng, Cecilia Clementi, Peter G. Wolynes,
and Garegin A. Papoian. AWSEM-MD: Protein Structure Prediction Using Coarse-grained
Physical Potentials and Bioinformatically Based Local Structure Biasing. J Phys Chem B,
116(29):8494, 2012.
[78] Fabio Sterpone, Simone Melchionna, Pierre Tuffery, Samuela Pasquali, Normand
Mousseau, Tristan Cragnolini, Yassmine Chebaro, Jean-Francois St-Pierre, Maria
Kalimeri, Alessandro Barducci, Yoann Laurin, Alex Tek, Marc Baaden, Phuong Hoang
Nguyen, and Philippe Derreumaux. The OPEP protein model: from single molecules,
amyloid formation, crowding and hydrodynamics to DNA/RNA systems. Chem. Soc. Rev.,
43(13):4871?4893, 2014.
[79] Siewert J. Marrink, H. Jelger Risselada, Serge Yefimov, D. Peter Tieleman, and Alex H.
de Vries. The MARTINI Force Field: Coarse Grained Model for Biomolecular
Simulations. J. Phys. Chem. B, 2007.
[80] Sai J. Ganesan and S. Matysiak. Role of Backbone Dipole Interactions in the Formation
of Secondary and Supersecondary Structures of Proteins. J. Chem. Theory Comput.,
10(6):2569?2576, June 2014.
167
[81] Sai J. Ganesan and Silvina Matysiak. Interplay between the hydrophobic effect and dipole
interactions in peptide aggregation at interfaces. Phys. Chem. Chem. Phys., 18(4):2449?
2458, 2016.
[82] Abhilash Sahoo, Hongcheng Xu, and Silvina Matysiak. Pathways of amyloid-beta
absorption and aggregation in a membranous environment. Phys. Chem. Chem. Phys.,
21(16):8559?8568, April 2019. Publisher: The Royal Society of Chemistry.
[83] Semen O. Yesylevskyy, Lars V. Scha?fer, Durba Sengupta, and Siewert J. Marrink.
Polarizable Water Model for the Coarse-Grained MARTINI Force Field. PLoS Comput
Biol, 6(6):e1000810, June 2010.
[84] Sai J. Ganesan, Hongcheng Xu, and Silvina Matysiak. Influence of Monovalent Cation
Size on Nanodomain Formation in Anionic?Zwitterionic Mixed Bilayers. J. Phys. Chem.
B, 121(4):787?799, February 2017.
[85] Sai J. Ganesan, Hongcheng Xu, and Silvina Matysiak. Effect of lipid head group
interactions on membrane properties and membrane-induced cationic ?-hairpin folding.
Phys. Chem. Chem. Phys., 18(27):17836?17850, 2016.
[86] David R. Howlett, Kevin H. Jennings, David C. Lee, Michael S.G. Clark, Frank Brown,
Ronald Wetzel, Stephen J. Wood, Patrick Camilleri, and Gareth W. Roberts. Aggregation
State and Neurotoxic Properties of Alzheimer Beta-Amyloid Peptide. Neurodegeneration,
4(1):23?32, 1995.
[87] G Brent Irvine, Omar M El-Agnaf, Ganesh M Shankar, and Dominic M Walsh. Protein
aggregation in the brain: the molecular basis for Alzheimer?s and Parkinson?s diseases.
Mol. Med., 14(7-8):451?64, 2008.
[88] Montserrat Arrasate and Steven Finkbeiner. Protein aggregates in Huntington?s disease.
Exp Neurol., 238(1):1?11, 2012.
[89] George G. Glenner and Caine W. Wong. Alzheimer?s disease and Down?s syndrome:
Sharing of a unique cerebrovascular amyloid fibril protein. Biochem. Biophys. Res.
Commun., 122(3):1131?1135, 1984.
[90] George G. Glenner and Caine W. Wong. Alzheimer?s disease: Initial report of the
purification and characterization of a novel cerebrovascular amyloid protein. Biochem.
Biophys. Res. Commun., 120(3):885?890, 1984.
[91] C L Masters, G Simms, N A Weinman, G Multhaup, B L McDonald, and K Beyreuther.
Amyloid plaque core protein in Alzheimer disease and Down syndrome. Proc. Natl. Acad.
Sci. U.S.A., 82(12):4245?9, 1985.
[92] Carlo Ballatore, Virginia M.-Y. Lee, and John Q. Trojanowski. Tau-mediated
neurodegeneration in Alzheimer?s disease and related disorders. Nat. Rev. Neurosci.,
8(9):663?672, 2007.
168
[93] H. Braak and E. Braak. Evolution of neuronal changes in the course of Alzheimer?s
disease. pages 127?140. Springer, Vienna, 1998.
[94] 2018 Alzheimer?s disease facts and figures. Alzheimers Dement, 14(3):367?429, 2018.
[95] B. Yankner, L. Dawes, S Fisher, L Villa-Komaroff, M. Oster-Granite, and R. Neve.
Neurotoxicity of a fragment of the amyloid precursor associated with Alzheimer?s disease.
Science, 245(4916):417?420, 1989.
[96] Dennis J. Selkoe. The molecular pathology of Alzheimer?s disease. Neuron, 6(4):487?498,
1991.
[97] M. Goedert and M. G. Spillantini. A Century of Alzheimer?s Disease. Science,
314(5800):777?781, 2006.
[98] Dennis J. Selkoe. Alzheimer?s Disease: Genes, Proteins, and Therapy. Physiol. Rev.,
81(2):741?766, 2001.
[99] R. Tanzi, J. Gusella, P. Watkins, G. Bruns, P St George-Hyslop, M. Van Keuren,
D Patterson, S Pagan, D. Kurnit, R. Neve, and al. Et. Amyloid beta protein gene:
cDNA, mRNA distribution, and genetic linkage near the Alzheimer locus. Science,
235(4791):880?884, 1987.
[100] Dominic M. Walsh and Dennis J. Selkoe. Deciphering the Molecular Basis of Memory
Failure in Alzheimer?s Disease. Neuron, 44(1):181?193, 2004.
[101] Charles C. Glabe. Amyloid accumulation and pathogensis of alzheimer?s disease:
Significance of monomeric, oligomeric and fibrillar a?. In Amyloid Accumulation and
Pathogensis of Alzheimer?s Disease: Significance of Monomeric, Oligomeric and Fibrillar
A?, pages 167?177. Springer US, 2005.
[102] Jessica Meinhardt, Carsten Sachse, Peter Hortschansky, Nikolaus Grigorieff, and Marcus
Fa?ndrich. A?(1-40) Fibril Polymorphism Implies Diverse Interaction Patterns in Amyloid
Fibrils. J. Mol. Biol., 386(3):869?877, 2009.
[103] Julian D. Gillmore, Philip N. Hawkins, and Mark B. Pepys. Amyloidosis: A review of
recent diagnostic and therapeutic developments. Br. J. Haematol., 99(2):245?256, 1997.
[104] E H Koo, P T Lansbury, and J W Kelly. Amyloid diseases: abnormal protein aggregation
in neurodegeneration. Proc. Natl. Acad. Sci. U.S.A., 96(18):9989?9990, 1999.
[105] Roma N Rambaran and Louise C Serpell. Amyloid fibrils: abnormal protein assembly.
Prion, 2(3):112?117, 2008.
[106] Kyle L. Morris and Louise C. Serpell. X-Ray Fibre Diffraction Studies of Amyloid Fibrils.
In Methods in molecular biology (Clifton, N.J.), volume 849, pages 121?135. 2012.
169
[107] Anders Olofsson, Malin Lindhagen-Persson, Monika Vestling, A. Elisabeth Sauer-
Eriksson, and Anders O?hman. Quenched hydrogen/deuterium exchange NMR
characterization of amyloid-? peptide aggregates formed in the presence of Cu2+ or Zn2+
. FEBS J., 276(15):4051?4060, 2009.
[108] Valerie L Anderson and Watt W Webb. Transmission electron microscopy characterization
of fluorescently labelled amyloid ? 1-40 and ?-synuclein aggregates. BMC Biotechnol,
11:125, 2011.
[109] Carsten Sachse, Marcus Fa?ndrich, and Nikolaus Grigorieff. Paired beta-sheet structure
of an Abeta(1-40) amyloid fibril revealed by electron microscopy. Proc. Natl. Acad. Sci.
U.S.A., 105(21):7462?7466, 2008.
[110] Lothar Gremer, Daniel Scho?lzel, Carla Schenk, Elke Reinartz, Jo?rg Labahn, Raimond
B. G. Ravelli, Markus Tusche, Carmen Lopez-Iglesias, Wolfgang Hoyer, Henrike Heise,
Dieter Willbold, and Gunnar F. Schro?der. Fibril structure of amyloid-?(1?42) by
cryo?electron microscopy. Science, 358(6359):116?119, 2017.
[111] Elizabeth Drolle, Alexander Negoda, Keely Hammond, Evgeny Pavlov, and Zoya
Leonenko. Changes in lipid membranes may trigger amyloid toxicity in Alzheimer?s
disease. PloS ONE, 12(8):e0182194, 2017.
[112] Mun?delanji Vestergaard, Tsutomu Hamada, and Masahiro Takagi. Using model
membranes for the study of amyloid beta:lipid interactions and neurotoxicity. Biotechnol.
Bioeng., 99(4):753?763, 2008.
[113] Justin A. Lemkul and David R. Bevan. Lipid composition influences the release of
Alzheimer?s amyloid ?-peptide from membranes. Protein Sci., 20(9):1530?1545, 2011.
[114] Isabel Morgado and Megan Garvey. Lipids in Amyloid-? Processing, Aggregation, and
Toxicity. pages 67?94. Springer, Cham, 2015.
[115] Laura Dominguez, Leigh Foster, John E. Straub, and D. Thirumalai. Impact of membrane
lipid composition on the structure and stability of the transmembrane domain of amyloid
precursor protein. Proc. Natl. Acad. Sci. U.S.A., 113(36):E5281?E5287, 2016.
[116] Katsumi Matsuzaki. Physicochemical interactions of amyloid ?-peptide with lipid
bilayers. Biochim. Biophys. Acta, Biomembr., 1768(8):1935?1942, 2007.
[117] Wenxin Yu, Kun Zou, Jian Sheng Gong, Mihee Ko, Katsuhiko Yanagisawa, and Makoto
Michikawa. Oligomerization of amyloid ?-protein occurs during the isolation of lipid
rafts. J. Neurosci. Res., 80(1):114?119, 2005.
[118] Dominic M. Walsh, Aleksey Lomakin, George B. Benedek, Margaret M. Condron,
and David B. Teplow. Amyloid ?-protein fibrillogenesis: Detection of a protofibrillar
intermediate. J. Biol. Chem., 272(35):22364?22372, 1997.
170
[119] Rakez Kayed, Elizabeth Head, Jennifer L. Thompson, Theresa M. McIntire, Saskia C.
Milton, Carl W. Cotman, and Charles G. Glabel. Common structure of soluble amyloid
oligomers implies common mechanism of pathogenesis. Science, 300(5618):486?489,
2003.
[120] Payel Das, Silvina Matysiak, and Jeetain Mittal. Looking at the Disordered Proteins
through the Computational Microscope. ACS Cent Sci, 4(5):534?542, 2018.
[121] Rongliang Wu, Jun Liu, Xinlong Qiu, and Manli Deng. Molecular dynamics simulation
of the nanofibrils formed by amyloid-based peptide amphiphiles. Mol Simul, 43(13-
16):1227?1239, 2017.
[122] B. Urbanc, L. Cruz, F. Ding, D. Sammond, S. Khare, S.V. Buldyrev, H.E. Stanley, and N.V.
Dokholyan. Molecular Dynamics Simulation of Amyloid ? Dimer Formation. Biophys.
J., 87(4):2310?2321, 2004.
[123] Matthew Biancalana and Shohei Koide. Molecular mechanism of Thioflavin-T binding to
amyloid fibrils. Biochim. Biophys. Acta, Proteins Proteomics, 1804(7):1405?1412, 2010.
[124] Nikolaos G. Sgourakis, Myrna Merced-Serrano, Christos Boutsidis, Petros Drineas,
Zheming Du, Chunyu Wang, and Angel E. Garcia. Atomic-Level Characterization of
the Ensemble of the A?(1?42) Monomer in Water Using Unbiased Molecular Dynamics
Simulations and Spectral Algorithms. J. Mol. Biol., 405(2):570?583, 2011.
[125] Nikolaos G. Sgourakis, Yilin Yan, Scott A. McCallum, Chunyu Wang, and Angel E.
Garcia. The Alzheimer?s Peptides A?40 and 42 Adopt Distinct Conformations in Water:
A Combined MD / NMR Study. J. Mol. Biol., 368(5):1448?1457, 2007.
[126] Naohiro Nishikawa, Phuong H. Nguyen, Philippe Derreumaux, and Yuko Okamoto.
Replica-exchange molecular dynamics simulation for understanding the initial process of
amyloid peptide aggregation. Mol Simul, 41(10-12):1041?1044, 2015.
[127] Dmitri K Klimov and D Thirumalai. Dissecting the assembly of Abeta16-22 amyloid
peptides into antiparallel beta sheets. Structure, 11(3):295?307, 2003.
[128] Se?bastien Santini, Normand Mousseau, and Philippe Derreumaux. In Silico Assembly of
Alzheimer?s A? 16-22 Peptide into ?-Sheets. J. Am. Chem. Soc., 126(37):11509?11516,
2004.
[129] Sebastian Kmiecik, Dominik Gront, Michal Kolinski, Lukasz Wieteska,
Aleksandra Elzbieta Dawid, and Andrzej Kolinski. Coarse-Grained Protein Models
and Their Applications. Chem. Rev., 116(14):7898?7936, 2016.
[130] Sanne Abeln, Michele Vendruscolo, Christopher M. Dobson, and Daan Frenkel. A Simple
Lattice Model That Captures Protein Folding, Aggregation and Amyloid Formation. PLoS
ONE, 9(1):e85185, 2014.
[131] Ran Friedman, Riccardo Pellarin, and Amedeo Caflisch. Amyloid Aggregation on Lipid
Bilayers and Its Impact on Membrane Permeability. J. Mol. Biol., 387(2):407?415, 2009.
171
[132] Alex Morriss-Andrews, Frank L. H. Brown, and Joan-Emma Shea. A Coarse-Grained
Model for Peptide Aggregation on a Membrane Surface. J. Phys. Chem. B, 118(28):8420?
8432, 2014.
[133] Mara Chiricotto, Thanh Thuy Tran, Phuong H. Nguyen, Simone Melchionna, Fabio
Sterpone, and Philippe Derreumaux. Coarse-grained and All-atom Simulations towards
the Early and Late Steps of Amyloid Fibril Formation. Isr. J. Chem., 57(7-8):564?573,
2017.
[134] Yassmine Chebaro, Samuela Pasquali, and Philippe Derreumaux. The Coarse-Grained
OPEP Force Field for Non-Amyloid and Amyloid Proteins. J. Phys. Chem. B,
116(30):8741?8752, 2012.
[135] Mookyung Cheon, Mooseok Kang, and Iksoo Chang. Polymorphism of fibrillar structures
depending on the size of assembled A?17-42 peptides. Sci. Rep., 6(1):38196, 2016.
[136] Weihua Zheng, Min-Yeh Tsai, Mingchen Chen, and Peter G Wolynes. Exploring the
aggregation free energy landscape of the amyloid-? protein (1-40). Proc. Natl. Acad. Sci.
U.S.A., 113(42):11835?11840, 2016.
[137] Canay Ege, Jaroslaw Majewski, Guohui Wu, Kristian Kjaer, and Ka Yee C.
Lee. Templating Effect of Lipid Membranes on Alzheimer?s Amyloid Beta Peptide.
ChemPhysChem, 6(2):226?229, 2005.
[138] R. O. Calderon, B. Attema, and G. H. DeVries. Lipid Composition of Neuronal Cell
Bodies and Neurites from Cultured Dorsal Root Ganglia. J. Neurochem., 64(1):424?429,
2002.
[139] Xiang Yu, Qiuming Wang, Qingfen Pan, Feimeng Zhou, and Jie Zheng. Molecular
interactions of Alzheimer amyloid-? oligomers with neutral and negatively charged lipid
bilayers. Phys. Chem. Chem. Phys, 15(23):8878, 2013.
[140] JoAnne McLaurin and Avijit Chakrabartty. Characterization of the Interactions of
Alzheimer beta-Amyloid Peptides with Phospholipid Membranes. Eur. J. Biochem.,
245(2):355?363, 1997.
[141] Evelyne Terzi, Guenter Hoelzemann, and Joachim Seelig. Alzheimer .beta.-Amyloid
Peptide 25-35: Electrostatic Interactions with Phospholipid Membranes. Biochemistry,
33(23):7434?7441, 1994.
[142] Evelyne Terzi, Gu?nter Ho?lzemann, and Joachim Seelig. Interaction of Alzheimer ?-
Amyloid Peptide(140) with Lipid Membranes?. Biochemistry, 1997.
[143] Claudio Canale, Silvia Seghezza, Silvia Vilasi, Rita Carrotta, Donatella Bulone, Alberto
Diaspro, Pier Luigi San Biagio, and Silvia Dante. Different effects of Alzheimer?s peptide
A?(1?40) oligomers and fibrils on supported lipid membranes. Biophys. Chem., 182:23?
29, 2013.
172
[144] Marc-Antoine Sani, John D. Gehman, and Frances Separovic. Lipid matrix plays a role in
Abeta fibril kinetics and morphology. FEBS Lett., 585(5):749?754, 2011.
[145] Tong-Lay Lau, John D. Gehman, John D. Wade, Keyla Perez, Colin L. Masters, Kevin J.
Barnham, and Frances Separovic. Membrane interactions and the effect of metal ions
of the amyloidogenic fragment A?(25-35) in comparison to A?(1?42). Biochim Biophys
Acta Biomembr, 1768(10):2400?2408, 2007.
[146] Yan Lu, Philippe Derreumaux, Zhi Guo, Normand Mousseau, and Guanghong Wei.
Thermodynamics and dynamics of amyloid peptide oligomerization are sequence
dependent. Proteins, 75(4):954?963, 2009.
[147] Sander Pronk, Szila?rd Pa?ll, Roland Schulz, Per Larsson, Pa?r Bjelkmar, Rossen Apostolov,
Michael R. Shirts, Jeremy C. Smith, Peter M. Kasson, David van der Spoel, Berk Hess,
and Erik Lindahl. GROMACS 4.5: A high-throughput and highly parallel open source
molecular simulation toolkit. Bioinformatics, 2013.
[148] Shu?ichi Nose?. A molecular dynamics method for simulations in the canonical ensemble.
Mol. Phys., 52(2):255?268, 1984.
[149] Hoover. Canonical dynamics: Equilibrium phase-space distributions. Phys. Rev. A,
31(3):1695?1697, 1985.
[150] M. Parrinello and A. Rahman. Polymorphic transitions in single crystals: A new molecular
dynamics method. J. Appl. Phys., 52(12):7182?7190, 1981.
[151] Tom Darden, Darrin York, and Lee Pedersen. Particle mesh ewald: An nlogn method for
ewald sums in large systems. J. Chem. Phys., 98(12):10089?10092, 1993.
[152] Ulrich Essmann, Lalith Perera, Max L. Berkowitz, Tom Darden, Hsing Lee, and Lee G.
Pedersen. A smooth particle mesh Ewald method. J. Chem. Phys., 103(19):8577?8593,
1995.
[153] Nabin Kandel, Tianyu Zheng, Qun Huo, and Suren A. Tatulian. Membrane Binding
and Pore Formation by a Cytotoxic Fragment of Amyloid ? Peptide. J. Phys. Chem.
B, 121(45):10293?10305, 2017.
[154] Siewert J Marrink and Alan E Mark. Molecular Dynamics Simulation of the Formation,
Structure, and Dynamics of Small Phospholipid Vesicles. J. Chem. Phys., 2003.
[155] Chun-Min Lin, Chun-Shian Li, Yu-Jane Sheng, David T Wu, and Heng-Kwong Tsao. Size-
Dependent Properties of Small Unilamellar Vesicles Formed by Model Lipids. Langmuir,
28:689?700, 2012.
[156] William Humphrey, Andrew Dalke, and Klaus Schulten. VMD ? Visual Molecular
Dynamics. J Mol Graph, 14:33?38, 1996.
[157] Fernando E. Herrera and Sergio Pantano. Structure and dynamics of nano-sized raft-like
domains on the plasma membrane. J. Chem. Phys., 136(1):015103, 2012.
173
[158] Jianjun Pan, Xiaolin Cheng, Luca Monticelli, Frederick A. Heberle, Norbert Kuc?erka,
D. Peter Tieleman, and John Katsaras. The molecular structure of a phosphatidylserine
bilayer determined by scattering and molecular dynamics simulations. Soft Matter,
10(21):3716, 2014.
[159] Yi Hu, Pascal Kienlen-Campard, Tzu-Chun Tang, Florian Perrin, Re?mi Opsomer, Marie
Decock, Xiaoshu Pan, Jean-Noel Octave, Stefan N. Constantinescu, and Steven O. Smith.
?-Sheet Structure within the Extracellular Domain of C99 Regulates Amyloidogenic
Processing. Sci. Rep., 7(1):17159, 2017.
[160] Neil R. Anthony, Anil K. Mehta, David G. Lynn, and Keith M. Berland. Mapping
amyloid-?(16-22) nucleation pathways using fluorescence lifetime imaging microscopy.
Soft Matter, 10(23):4162?4172, 2014.
[161] Sandra Chimon, Medhat A Shaibat, Christopher R Jones, Diana C Calero, Buzulagu
Aizezi, and Yoshitaka Ishii. Evidence of fibril-like ?-sheet structures in a neurotoxic
amyloid intermediate of Alzheimer?s ?-amyloid. Nat. Struct. Mol. Biol., 14(12):1157?
1164, 2007.
[162] Nicolae-Viorel Buchete and Gerhard Hummer. Structure and dynamics of parallel beta-
sheets, hydrophobic core, and loops in Alzheimer?s A beta fibrils. Biophys. J., 92(9):3032?
3039, 2007.
[163] Jie Li, Ruiwu Liu, Kit S Lam, Lee-Way Jin, and Yong Duan. Alzheimer?s disease drug
candidates stabilize A-? protein native structure by interacting with the hydrophobic core.
Biophys. J., 100(4):1076?1082, 2011.
[164] David J. Lindberg, Emelie Wese?n, Johan Bjo?rkeroth, Sandra Rocha, and Elin K.
Esbjo?rner. Lipid membranes catalyse the fibril formation of the amyloid-? (1?42) peptide
through lipid-fibril interactions that reinforce secondary pathways. Biochim Biophys Acta
Biomembr, 1859(10):1921?1929, 2017.
[165] John J Kremer and Regina M Murphy. Kinetics of adsorption of beta-amyloid peptide
Abeta(1-40) to lipid bilayers. J. Biochem. Biophys. Methods, 57(2):159?169, 2003.
[166] Abhilash Sahoo and Silvina Matysiak. Computational insights into lipid assisted
peptide misfolding and aggregation in neurodegeneration. Phys. Chem. Chem. Phys.,
21(41):22679?22694, October 2019.
[167] Karine Berthelot, Christophe Cullin, and Sophie Lecomte. What does make an amyloid
toxic: Morphology, structure or interaction with membrane? Biochimie, 95(1):12?19,
2013.
[168] Iwona M. Pranke, Vincent Morello, Joe?lle Bigay, Kimberley Gibson, Jean-Marc
Verbavatz, Bruno Antonny, and Catherine L. Jackson. ?-Synuclein and ALPS motifs
are membrane curvature sensors whose contrasting chemistry mediates selective vesicle
binding. J. Cell Biol., 194(1):89, July 2011. Publisher: The Rockefeller University Press.
174
[169] Maxmore Chaibva, Kathleen A. Burke, and Justin Legleiter. Curvature enhances binding
and aggregation of huntingtin at lipid membranes. Biochemistry, 53(14):2355?2365, April
2014.
[170] G. C. A. da Hora, N. L. Archilha, J. L. S. Lopes, D. M. Mu?ller, K. Coutinho, R. Itri, and
T. A. Soares. Membrane negative curvature induced by a hybrid peptide from pediocin
PA-1 and plantaricin 149 as revealed by atomistic molecular dynamics simulations. Soft
Matter, 12(43):8884?8898, November 2016.
[171] Denys E. S. Santos, Lae?rcio Pol-Fachin, Roberto D. Lins, and Thereza A. Soares.
Polymyxin Binding to the Bacterial Outer Membrane Reveals Cation Displacement and
Increasing Membrane Curvature in Susceptible but Not in Resistant Lipopolysaccharide
Chemotypes. J. Chem. Inf. Model, 57(9):2181?2193, September 2017.
[172] Federico El??as-Wolff, Martin Linde?n, Alexander P. Lyubartsev, and Erik G. Brandt.
Curvature sensing by cardiolipin in simulated buckled membranes. Soft Matter,
15(4):792?802, 2019.
[173] Jordi Go?mez-Llobregat, Federico El??as-Wolff, and Martin Linde?n. Anisotropic Membrane
Curvature Sensing by Amphipathic Peptides. Biophys. J., 110(1):197?204, January 2016.
[174] Abhilash Sahoo and Silvina Matysiak. Microscopic picture of calcium-assisted lipid
demixing and membrane remodeling using multiscale simulations. J. Phys. Chem. B,
124(34):7327?7335, August 2020.
[175] Keon A. Reid, Caitlin M. Davis, R. Brian Dyer, and James T. Kindt. Binding, folding and
insertion of a beta-hairpin peptide at a lipid bilayer surface: Influence of electrostatics and
lipid tail packing. Biochimica et Biophysica Acta (BBA) - Biomembranes, 1860(3):792?
800, 2018.
[176] Semen O. Yesylevskyy, Timothe?e Rivel, and Christophe Ramseyer. The influence of
curvature on the properties of the plasma membrane. Insights from atomistic molecular
dynamics simulations. Sci. Rep, 7(1):16078, December 2017.
[177] Mark James Abraham, Teemu Murtola, Roland Schulz, Szila?rd Pa?ll, Jeremy C. Smith,
Berk Hess, and Erik Lindahl. GROMACS: High performance molecular simulations
through multi-level parallelism from laptops to supercomputers. SoftwareX, 1-2:19?25,
September 2015.
[178] H. J. C. Berendsen, J. P. M. Postma, W. F. van Gunsteren, A. DiNola, and J. R. Haak.
Molecular dynamics with coupling to an external bath. J. Chem. Phys., 81(8):3684?3690,
October 1984.
[179] Michele F.M. Sciacca, Carmelo Tempra, Federica Scollo, Danilo Milardi, and Carmelo
La Rosa. Amyloid growth and membrane damage: Current themes and emerging
perspectives from theory and experiments on A? and hIAPP. Biochim Biophys Acta
Biomembr, 1860(9):1625?1638, September 2018.
175
[180] Justin A. Lemkul and David R. Bevan. Perturbation of membranes by the amyloid ?-
peptide ? a molecular dynamics study. FEBS J., 276(11):3060?3075, 2009.
[181] Thomas L. Williams and Louise C. Serpell. Membrane and surface interactions of
Alzheimer?s A? peptide ? insights into the mechanism of cytotoxicity. FEBS J.,
278(20):3905?3917, 2011.
[182] Xubo Lin, Vinay Nair, Yong Zhou, and Alemayehu A. Gorfe. Membrane potential and
dynamics in a ternary lipid mixture: insights from molecular dynamics simulations. Phys.
Chem. Chem. Phys., 20(23):15841?15851, 2018.
[183] Carl-Johan Ho?gberg and Alexander P. Lyubartsev. Effect of Local Anesthetic Lidocaine
on Electrostatic Properties of a Lipid Bilayer. Biophys. J., 94(2):525?531, January 2008.
[184] Guo-fang Chen, Ting-hai Xu, Yan Yan, Yu-ren Zhou, Yi Jiang, Karsten Melcher, and
H Eric Xu. Amyloid beta: structure, biology and structure-based therapeutic development.
Acta Pharmacol Sin, 38(9):1205?1235, September 2017.
[185] Chetan N. Patel, Schroeder M. Noble, Gresham T. Weatherly, Ashutosh Tripathy, Donald J.
Winzor, and Gary J. Pielak. Effects of molecular crowding by saccharides on ?-
chymotrypsin dimerization. Protein Sci., 11(5):997?1003, May 2002.
[186] Sumra Shahid, Ikramul Hasan, Faizan Ahmad, Md. Imtaiyaz Hassan, and Asimul
Islam. Carbohydrate-Based Macromolecular Crowding-Induced Stabilization of Proteins:
Towards Understanding the Significance of the Size of the Crowder. Biomolecules,
9(9):477, September 2019.
[187] Lihui Bai, Xiaocui Guo, Xue Zhang, Wenting Yu, and Dayong Yang. Saccharides
Create a Crowding Environment for Gene Expression in Cell-Free Systems. Langmuir,
35(17):5931?5936, April 2019.
[188] Irina Kuznetsova, Konstantin Turoverov, and Vladimir Uversky. What Macromolecular
Crowding Can Do to a Protein. IJMS, 15(12):23090?23140, December 2014.
[189] David C. Latshaw, Mookyung Cheon, and Carol K. Hall. Effects of Macromolecular
Crowding on Amyloid Beta (16?22) Aggregation Using Coarse-Grained Simulations. J.
Phys. Chem. B, 118(47):13513?13526, November 2014.
[190] F. Musiani and A. Giorgetti. Protein Aggregation and Molecular Crowding. In
International Review of Cell and Molecular Biology, volume 329, pages 49?77. Elsevier,
2017.
[191] Larissa A. Munishkina, Atta Ahmad, Anthony L. Fink, and Vladimir N. Uversky. Guiding
Protein Aggregation with Macromolecular Crowding. Biochemistry, 47(34):8993?9006,
August 2008.
[192] Sneha Menon and Neelanjana Sengupta. Influence of Hyperglycemic Conditions on Self-
Association of the Alzheimer?s Amyloid ? (A? 1?42 ) Peptide. ACS Omega, 2(5):2134?
2147, May 2017.
176
[193] Cesar A. Lo?pez, Andrzej J. Rzepiela, Alex H. de Vries, Lubbert Dijkhuizen, Philippe H.
Hu?nenberger, and Siewert J. Marrink. Martini Coarse-Grained Force Field: Extension to
Carbohydrates. J. Chem. Theory Comput., 5(12):3195?3210, December 2009.
[194] Philipp S. Schmalhorst, Felix Deluweit, Roger Scherrers, Carl-Philipp Heisenberg, and
Mateusz Sikora. Overcoming the Limitations of the MARTINI Force Field in Simulations
of Polysaccharides. J. Chem. Theory Comput., 13(10):5039?5053, October 2017.
[195] Moritoki Egi, Rinaldo Bellomo, Edward Stachowski, Craig J. French, Graeme K. Hart,
Colin Hegarty, and Michael Bailey. Blood glucose concentration and outcome of critical
illness: The impact of diabetes*:. Critical Care Medicine, 36(8):2249?2255, August 2008.
[196] A. Lerbret, F. Affouard, P. Bordat, A. He?doux, Y. Guinet, and M. Descamps. Molecular
dynamics simulations of lysozyme in water/sugar solutions. Chemical Physics, 345(2-
3):267?274, April 2008.
[197] Na Zhang, Fu-Feng Liu, Xiao-Yan Dong, and Yan Sun. Molecular Insight into the
Counteraction of Trehalose on Urea-Induced Protein Denaturation Using Molecular
Dynamics Simulation. J. Phys. Chem. B, 116(24):7040?7047, June 2012.
[198] A. Lerbret, P. Bordat, F. Affouard, A. He?doux, Y. Guinet, and M. Descamps. How Do
Trehalose, Maltose, and Sucrose Influence Some Structural and Dynamical Properties
of Lysozyme? Insight from Molecular Dynamics Simulations. J. Phys. Chem. B,
111(31):9410?9420, August 2007.
[199] Fu-Feng Liu, Luo Ji, Xiao-Yan Dong, and Yan Sun. Molecular Insight into the Inhibition
Effect of Trehalose on the Nucleation and Elongation of Amyloid ?-Peptide Oligomers. J.
Phys. Chem. B, 113(32):11320?11329, August 2009.
[200] C. Molteni and M. Parrinello. Glucose in Aqueous Solution by First Principles Molecular
Dynamics. J. Am. Chem. Soc., 120(9):2168?2171, March 1998.
[201] Nicholas F. Dupuis, Erik D. Holmstrom, and David J. Nesbitt. Molecular-crowding effects
on single-molecule RNA folding/unfolding thermodynamics and kinetics. Proc. Natl.
Acad. Sci. U.S.A., 111(23):8464?8469, June 2014.
[202] Michael Senske, Lisa To?rk, Benjamin Born, Martina Havenith, Christian Herrmann, and
Simon Ebbinghaus. Protein Stabilization by Macromolecular Crowding through Enthalpy
Rather Than Entropy. J. Am. Chem. Soc., 136(25):9036?9041, June 2014.
[203] Kim A. Sharp. Analysis of the size dependence of macromolecular crowding shows that
smaller is better. Proc. Natl. Acad. Sci. U.S.A., 112(26):7990?7995, June 2015.
[204] Mohona Sarkar, Conggang Li, and Gary J. Pielak. Soft interactions and crowding. Biophys
Rev, 5(2):187?194, June 2013.
[205] William M. Aumiller, Bradley W. Davis, Emmanuel Hatzakis, and Christine D. Keating.
Interactions of Macromolecular Crowding Agents and Cosolutes with Small-Molecule
Substrates: Effect on Horseradish Peroxidase Activity with Two Different Substrates. J.
Phys. Chem. B, 118(36):10624?10632, September 2014.
177
[206] A D McLachlan. Protein Structure and Function. Annu. Rev. Phys. Chem., 23(1):165?192,
October 1972.
[207] Joseph A. Marsh and Sarah A. Teichmann. Structure, Dynamics, Assembly, and Evolution
of Protein Complexes. Annu. Rev. Biochem., 84(1):551?575, June 2015.
[208] Freddie R Salsbury Jr. Molecular dynamics simulations of protein dynamics and their
relevance to drug discovery. Curr Opin Pharmacol, 10(6):738?744, December 2010.
[209] M. Karplus and J. Kuriyan. Molecular dynamics and protein function. Proc. Natl. Acad.
Sci. U.S.A., 102(19):6679?6685, May 2005.
[210] Ron O. Dror, Robert M. Dirks, J.P. Grossman, Huafeng Xu, and David E. Shaw.
Biomolecular Simulation: A Computational Microscope for Molecular Biology. Annu.
Rev. Biophys., 41(1):429?452, June 2012.
[211] Paul Maragakis, Kresten Lindorff-Larsen, Michael P. Eastwood, Ron O. Dror, John L.
Klepeis, Isaiah T. Arkin, Morten. Jensen, Huafeng Xu, Nikola Trbovic, Richard A.
Friesner, Arthur G. Palmer, and David E. Shaw. Microsecond Molecular Dynamics
Simulation Shows Effect of Slow Loop Dynamics on Backbone Amide Order Parameters
of Proteins. J. Phys. Chem. B, 112(19):6155?6158, May 2008.
[212] Peter L. Freddolino, Feng Liu, Martin Gruebele, and Klaus Schulten. Ten-Microsecond
Molecular Dynamics Simulation of a Fast-Folding WW Domain. Biophys. J., 94(10):L75?
L77, May 2008.
[213] K. Lindorff-Larsen, S. Piana, R. O. Dror, and D. E. Shaw. How Fast-Folding Proteins
Fold. Science, 334(6055):517?520, October 2011.
[214] Kresten Lindorff-Larsen, Paul Maragakis, Stefano Piana, and David E. Shaw. Picosecond
to Millisecond Structural Dynamics in Human Ubiquitin. J. Phys. Chem. B, 120(33):8313?
8320, August 2016.
[215] Jose? Nelson Onuchic, Zaida Luthey-Schulten, and Peter G. Wolynes. THEORY OF
PROTEIN FOLDING: The Energy Landscape Perspective. Annu. Rev. Phys. Chem.,
48(1):545?600, October 1997.
[216] Ronald Hills and Charles Brooks. Insights from Coarse-Grained Go? Models for Protein
Folding and Dynamics. Int. J. Mol. Sci., 10(3):889?905, March 2009.
[217] Cecilia Clementi. Coarse-grained models of protein folding: toy models or predictive
tools? Curr. Opin. Struct. Biol., 18(1):10?15, February 2008.
[218] Ram Samudrala and John Moult. An all-atom distance-dependent conditional probability
discriminatory function for protein structure prediction 1 1Edited by F. Cohen. J. Mol.
Biol., 275(5):895?916, February 1998.
[219] Parimal Kar, Srinivasa Murthy Gopal, Yi-Ming Cheng, Alexander Predeus, and Michael
Feig. PRIMO: A Transferable Coarse-Grained Force Field for Proteins. J. Chem. Theory
Comput., 9(8):3769?3788, August 2013.
178
[220] Aram Davtyan, Nicholas P. Schafer, Weihua Zheng, Cecilia Clementi, Peter G. Wolynes,
and Garegin A. Papoian. AWSEM-MD: Protein Structure Prediction Using Coarse-
Grained Physical Potentials and Bioinformatically Based Local Structure Biasing. J. Phys.
Chem. B, 116(29):8494?8503, July 2012.
[221] Yassmine Chebaro, Samuela Pasquali, and Philippe Derreumaux. The Coarse-Grained
OPEP Force Field for Non-Amyloid and Amyloid Proteins. J. Phys. Chem. B,
116(30):8741?8752, August 2012.
[222] Karolina Zieba, Magdalena S?lusarz, Rafa? Slusarz, Adam Liwo, Cezary Czaplewski, and
Adam K. Sieradzan. Extension of the UNRES Coarse-Grained Force Field to Membrane
Proteins in the Lipid Bilayer. J. Phys. Chem. B, 123(37):7829?7839, September 2019.
[223] Henry Chan, Mathew J. Cherukara, Badri Narayanan, Troy D. Loeffler, Chris Benmore,
Stephen K. Gray, and Subramanian K. R. S. Sankaranarayanan. Machine learning coarse
grained models for water. Nat. Commun., 10(1):379, December 2019.
[224] Jiang Wang, Simon Olsson, Christoph Wehmeyer, Adria? Pe?rez, Nicholas E. Charron,
Gianni de Fabritiis, Frank Noe?, and Cecilia Clementi. Machine Learning of Coarse-
Grained Molecular Dynamics Force Fields. ACS Cent. Sci., 5(5):755?767, May 2019.
[225] Cecilia Clementi, Hugh Nymeyer, and Jose? Nelson Onuchic. Topological and energetic
factors: what determines the structural details of the transition state ensemble and ?en-
route? intermediates for protein folding? an investigation for small globular proteins. J.
Mol. Biol., 298(5):937?953, May 2000.
[226] Scott Brown, Nicolas J. Fawzi, and Teresa Head-Gordon. Coarse-grained sequences
for protein folding and design. Proc. Natl. Acad. Sci. U.S.A., 100(19):10712?10717,
September 2003.
[227] Alex Morriss-Andrews, Frank L. H. Brown, and Joan-Emma Shea. A Coarse-Grained
Model for Peptide Aggregation on a Membrane Surface. J. Phys. Chem. B, 118(28):8420?
8432, July 2014.
[228] Payel Das, Silvina Matysiak, and Cecilia Clementi. Balancing energy and entropy: A
minimalist model for the characterization of protein folding landscapes. Proc. Natl. Acad.
Sci. U.S.A., 102(29):10141?10146, July 2005.
[229] Hao Wu, Yamini Dalal, and Garegin A. Papoian. Binding Dynamics of Disordered Linker
Histone H1 with a Nucleosomal Particle. J. Mol. Biol., 433(6):166881, March 2021.
[230] David S. Yang, Arash Saeedi, Aram Davtyan, Mohsen Fathi, Michael B. Sherman,
Mohammad S. Safari, Alena Klindziuk, Michelle C. Barton, Navin Varadarajan,
Anatoly B. Kolomeisky, and Peter G. Vekilov. Mesoscopic protein-rich clusters
host the nucleation of mutant p53 amyloid fibrils. Proc. Natl. Acad. Sci. U.S.A.,
118(10):e2015618118, March 2021.
179
[231] Scott P. O. Danielsen, James McCarty, Joan-Emma Shea, Kris T. Delaney, and Glenn H.
Fredrickson. Molecular design of self-coacervation phenomena in block polyampholytes.
Proc. Natl. Acad. Sci. U.S.A., 116(17):8224?8232, April 2019.
[232] Bogdan Barz and Brigita Urbanc. Minimal Model of Self-Assembly: Emergence of
Diversity and Complexity. J. Phys. Chem. B, 118(14):3761?3770, April 2014.
[233] Stefan Auer, Filip Meersman, Christopher M. Dobson, and Michele Vendruscolo.
A Generic Mechanism of Emergence of Amyloid Protofilaments from Disordered
Oligomeric Aggregates. PLoS Comput. Biol., 4(11):e1000222, November 2008.
[234] K. Yue, K. M. Fiebig, P. D. Thomas, H. S. Chan, E. I. Shakhnovich, and K. A. Dill. A
test of lattice protein folding algorithms. Proc. Natl. Acad. Sci. U.S.A., 92(1):325?329,
January 1995.
[235] Jian Zhou, Ian F. Thorpe, Sergey Izvekov, and Gregory A. Voth. Coarse-Grained Peptide
Modeling Using a Systematic Multiscale Approach. Biophys. J., 92(12):4289?4303, June
2007.
[236] S. Y. Mashayak, Mara N. Jochum, Konstantin Koschke, N. R. Aluru, Victor Ru?hle, and
Christoph Junghans. Relative Entropy and Optimization-Driven Coarse-Graining Methods
in VOTCA. PLOS ONE, 10(7):e0131754, 2015. Publisher: Public Library of Science.
[237] Sergei Izvekov and Gregory A. Voth. A Multiscale Coarse-Graining Method for
Biomolecular Systems. J. Phys. Chem. B, 109(7):2469?2473, February 2005.
[238] S.M. Vaiana, M. Manno, A. Emanuele, M.B. Palma-Vittorelli, and M.U. Palma. The role
of solvent in protein folding and in aggregation. J. Biol. Phys., 27(2/3):133?145, 2001.
[239] Hiraku Oshima and Masahiro Kinoshita. Essential roles of protein-solvent many-body
correlation in solvent-entropy effect on protein folding and denaturation: Comparison
between hard-sphere solvent and water. J. Chem. Phys., 142(14):145103, April 2015.
[240] C. Nick Pace, Saul Trevin?o, Erode Prabhakaran, and J. Martin Scholtz. Protein structure,
stability and solubility in water and other solvents. Philos. Trans. R. Soc. Lond., B, Biol.
Sci., 359(1448):1225?1235, August 2004.
[241] D. K. Klimov and D. Thirumalai. Mechanisms and kinetics of beta -hairpin formation.
Proc. Natl. Acad. Sci. U.S.A., 97(6):2544?2549, March 2000.
[242] D. K. Klimov and D. Thirumalai. Linking rates of folding in lattice models of proteins
with underlying thermodynamic characteristics. J. Chem. Phys., 109(10):4119?4125,
September 1998.
[243] Yuzo Ueda, Hiroshi Taketomi, and Nobuhiro G? Studies on protein folding, unfolding, and
fluctuations by computer simulation. II. A. Three-dimensional lattice model of lysozyme.
Biopolymers, 17(6):1531?1548, June 1978.
180
[244] Paulo C. T. Souza, Riccardo Alessandri, Jonathan Barnoud, Sebastian Thallmair, Ignacio
Faustino, Fabian Gru?newald, Ilias Patmanidis, Haleh Abdizadeh, Bart M. H. Bruininks,
Tsjerk A. Wassenaar, Peter C. Kroon, Josef Melcr, Vincent Nieto, Valentina Corradi,
Hanif M. Khan, Jan Doman?ski, Matti Javanainen, Hector Martinez-Seara, Nathalie Reuter,
Robert B. Best, Ilpo Vattulainen, Luca Monticelli, Xavier Periole, D. Peter Tieleman,
Alex H. de Vries, and Siewert J. Marrink. Martini 3: a general purpose force field for
coarse-grained molecular dynamics. Nat. Methods, 18(4):382?388, April 2021.
[245] Paulo C. T. Souza, Sebastian Thallmair, Paolo Conflitti, Carlos Ram??rez-Palacios,
Riccardo Alessandri, Stefano Raniolo, Vittorio Limongelli, and Siewert J. Marrink.
Protein?ligand binding with the coarse-grained Martini model. Nat Commun, 11(1):3714,
December 2020.
[246] Paulo C. T. Souza, Vittorio Limongelli, Sangwook Wu, Siewert J. Marrink, and Luca
Monticelli. Perspectives on High-Throughput Ligand/Protein Docking With Martini MD
Simulations. Front. Mol. Biosci., 8:657222, March 2021.
[247] Cristian Gobbo, Isabelle Beurroies, David de Ridder, Rienk Eelkema, Siewert J. Marrink,
Steven De Feyter, Jan H. van Esch, and Alex H. de Vries. MARTINI Model for
Physisorption of Organic Molecules on Graphite. J. Phys. Chem. C, 117(30):15623?
15631, August 2013.
[248] C. Arnarez, S. J. Marrink, and X. Periole. Molecular mechanism of cardiolipin-mediated
assembly of respiratory chain supercomplexes. Chem. Sci., 7(7):4435?4443, 2016.
[249] Fude Sun, Carsten F.E. Schroer, Lida Xu, Huiwei Yin, Siewert J. Marrink, and Shi-Zhong
Luo. Molecular Dynamics of the Association of L-Selectin and FERM Regulated by PIP2.
Biophys. J., 114(8):1858?1868, April 2018.
[250] Tsjerk A. Wassenaar, Kristyna Pluhackova, Anastassiia Moussatova, Durba Sengupta,
Siewert J. Marrink, D. Peter Tieleman, and Rainer A. Bo?ckmann. High-Throughput
Simulations of Dimer and Trimer Assembly of Membrane Proteins. The DAFT Approach.
J. Chem. Theory Comput., 11(5):2278?2291, May 2015.
[251] John Karanicolas and Charles L. Brooks. The origins of asymmetry in the folding
transition states of protein L and protein G. Protein Science, 11(10):2351?2361, April
2009.
[252] Adolfo B. Poma, Marek Cieplak, and Panagiotis E. Theodorakis. Combining the
MARTINI and Structure-Based Coarse-Grained Approaches for the Molecular Dynamics
Studies of Conformational Transitions in Proteins. J. Chem. Theory Comput., 13(3):1366?
1374, March 2017.
[253] Md. Iqbal Mahmood, Adolfo B. Poma, and Kei-ichi Okazaki. Optimizing Go?-MARTINI
Coarse-Grained Model for F-BAR Protein on Lipid Membrane. Front. Mol. Biosci.,
8:619381, February 2021.
181
[254] Abhilash Sahoo and Silvina Matysiak. Effects of applied surface-tension on membrane-
assisted a? aggregation. Phys. Chem. Chem. Phys., 23(36):20627?20633, 2021.
[255] Luca Monticelli, Senthil K. Kandasamy, Xavier Periole, Ronald G. Larson, D. Peter
Tieleman, and Siewert-Jan Marrink. The MARTINI Coarse-Grained Force Field:
Extension to Proteins. J. Chem. Theory Comput., 4(5):819?834, May 2008.
[256] Jing Huang, Sarah Rauscher, Grzegorz Nawrocki, Ting Ran, Michael Feig, Bert L
de Groot, Helmut Grubmu?ller, and Alexander D MacKerell. CHARMM36m: an improved
force field for folded and intrinsically disordered proteins. Nat Methods, 14(1):71?73,
January 2017.
[257] William L. Jorgensen, Jayaraman Chandrasekhar, Jeffry D. Madura, Roger W. Impey, and
Michael L. Klein. Comparison of simple potential functions for simulating liquid water.
J. Chem. Phys., 79(2):926?935, 1983.
[258] Sunhwan Jo, Taehoon Kim, Vidyashankara G. Iyer, and Wonpil Im. CHARMM-GUI: A
web-based graphical user interface for CHARMM. J. Comput. Chem., 29(11):1859?1865,
August 2008.
[259] J. P. Gallivan and D. A. Dougherty. Cation-pi interactions in structural biology. Proc. Natl.
Acad. Sci. U.S.A., 96(17):9459?9464, August 1999.
[260] Neal J. Zondlo. Aromatic?Proline Interactions: Electronically Tunable CH/? Interactions.
Acc. Chem. Res., 46(4):1039?1049, April 2013.
[261] Emily G Baker, Christopher Williams, Kieran L Hudson, Gail J Bartlett, Jack W Heal,
Kathryn L Porter Goff, Richard B Sessions, Matthew P Crump, and Derek N Woolfson.
Engineering protein stability with atomic precision in a monomeric miniprotein. Nat Chem
Biol, 13(7):764?770, July 2017.
[262] Motohiro Nishio, Yoji Umezawa, Jacques Fantini, Manfred S. Weiss, and Pinak
Chakrabarti. Ch?? hydrogen bonds in biological macromolecules. Phys. Chem. Chem.
Phys., 16(25):12648?12683, 2014.
[263] Jing Huang, Pedro E. M. Lopes, Beno??t Roux, and Alexander D. MacKerell. Recent
Advances in Polarizable Force Fields for Macromolecules: Microsecond Simulations of
Proteins Using the Classical Drude Oscillator Model. J. Phys. Chem. Lett., 5(18):3144?
3150, September 2014.
[264] Haibo Yu, Troy W. Whitfield, Edward Harder, Guillaume Lamoureux, Igor Vorobyov,
Victor M. Anisimov, Alexander D. MacKerell, and Beno??t Roux. Simulating Monovalent
and Divalent Ions in Aqueous Solution Using a Drude Polarizable Force Field. J. Chem.
Theory Comput., 6(3):774?786, March 2010.
[265] Eng-Hui Yap, Nicolas Lux Fawzi, and Teresa Head-Gordon. A coarse-grained ?-carbon
protein model with anisotropic hydrogen-bonding. Proteins, 70(3):626?638, February
2008.
182
[266] Harald A. Posch, William G. Hoover, and Franz J. Vesely. Canonical dynamics of the
Nose? oscillator: Stability, order, and chaos. Phys. Rev. A, 33(6):4253?4265, June 1986.
[267] Lingle Wang, Richard A Friesner, and BJ Berne. Replica exchange with solute scaling: a
more efficient version of replica exchange with solute tempering (rest2). The Journal of
Physical Chemistry B, 115(30):9431?9438, 2011.
[268] Viktor Hornak, Robert Abel, Asim Okur, Bentley Strockbine, Adrian Roitberg, and Carlos
Simmerling. Comparison of multiple amber force fields and development of improved
protein backbone parameters. Proteins: Structure, Function, and Bioinformatics,
65(3):712?725, 2006.
[269] Michael R Shirts and John D Chodera. Statistically optimal analysis of samples from
multiple equilibrium states. J. Chem. Phys., 129(12):124105, 2008.
[270] Dietmar Paschek, Sascha Hempel, and Angel E Garc??a. Computing the stability diagram
of the trp-cage miniprotein. Proc. Natl. Acad. Sci. U.S.A., 105(46):17754?17759, 2008.
[271] Jia Lin Huang, Michael E Noss, Karson M Schmidt, Leigh Murray, and Michelle R
Bunagan. The effect of neat ionic liquid on the folding of short peptides. Chemical
Communications, 47(28):8007?8009, 2011.
[272] Ryan Day, Dietmar Paschek, and Angel E Garcia. Microsecond simulations of the
folding/unfolding thermodynamics of the trp-cage miniprotein. Proteins: Structure,
Function, and Bioinformatics, 78(8):1889?1899, 2010.
[273] Andrea G Cochran, Nicholas J Skelton, and Melissa A Starovasnik. Tryptophan zippers:
Stable, monomeric ?-hairpins. Proc. Natl. Acad. Sci. U.S.A., 98(10):5578?5583, 2001.
[274] Deguo Du, Yongjin Zhu, Cheng-Yen Huang, and Feng Gai. Understanding the key factors
that control the rate of ?-hairpin folding. Proc. Natl. Acad. Sci. U.S.A., 101(45):15915?
15920, 2004.
[275] Aimee Byrne, D Victoria Williams, Bipasha Barua, Stephen J Hagen, Brandon L Kier,
and Niels H Andersen. Folding dynamics and pathways of the trp-cage miniproteins.
Biochemistry, 53(38):6011?6021, 2014.
[276] Dietmar Paschek, Hugh Nymeyer, and Angel E Garc??a. Replica exchange simulation
of reversible folding/unfolding of the trp-cage miniprotein in explicit solvent: On the
structure and possible role of internal water. J. Mol. Biol., 157(3):524?533, 2007.
[277] Lucas N. R. Wafer, Werner W. Streicher, and George I. Makhatadze. Thermodynamics of
the Trp-cage Miniprotein Unfolding in Urea. Proteins, 78(6):1376?1381, May 2010.
[278] Beatrice N. Markiewicz, Lijiang Yang, Robert M. Culik, Yi Qin Gao, and Feng Gai. How
Quickly Can a ?-Hairpin Fold from Its Transition State? J. Phys. Chem. B, 118(12):3317?
3325, March 2014.
183
[279] Andreas Reiner, Peter Henklein, and Thomas Kiefhaber. An unlocking/relocking barrier in
conformational fluctuations of villin headpiece subdomain. Proc. Natl. Acad. Sci. U.S.A.,
107(11):4955?4960, 2010.
[280] Wolfgang Kabsch and Christian Sander. Dictionary of protein secondary structure:
pattern recognition of hydrogen-bonded and geometrical features. Biopolymers: Original
Research on Biomolecules, 22(12):2577?2637, 1983.
[281] Yiming Wang, Samuel J. Bunce, Sheena E. Radford, Andrew J. Wilson, Stefan Auer, and
Carol K. Hall. Thermodynamic phase diagram of amyloid-? (16?22) peptide. Proc. Natl.
Acad. Sci. U.S.A., 116(6):2091?2096, February 2019.
[282] Dongdong Lin, Yin Luo, Shan Wu, Qianqian Ma, Guanghong Wei, and Xinju
Yang. Investigation of the Aggregation Process of Amyloid-?-(16-22) Peptides and the
Dissolution of Intermediate Aggregates. Langmuir, 30(11):3170?3175, March 2014.
[283] Suman Samantray, Feng Yin, Batuhan Kav, and Birgit Strodel. Different Force Fields Give
Rise to Different Amyloid Aggregation Pathways in Molecular Dynamics Simulations. J.
Chem. Inf. Model., 60(12):6462?6475, December 2020.
[284] John J Balbach, Yoshitaka Ishii, Oleg N Antzutkin, Richard D Leapman, Nancy W
Rizzo, Fred Dyda, Jennifer Reed, and Robert Tycko. Amyloid fibril formation by
a?16-22, a seven-residue fragment of the alzheimer?s ?-amyloid peptide, and structural
characterization by solid state nmr. Biochemistry, 39(45):13748?13759, 2000.
[285] John J. Balbach, Yoshitaka Ishii, Oleg N. Antzutkin, Richard D. Leapman, Nancy W.
Rizzo, Fred Dyda, Jennifer Reed, and Robert Tycko. Amyloid Fibril Formation by A?
16-22 , a Seven-Residue Fragment of the Alzheimer?s ?-Amyloid Peptide, and Structural
Characterization by Solid State NMR ?. Biochemistry, 39(45):13748?13759, November
2000.
[286] Phuong H Nguyen, Mai Suan Li, and Philippe Derreumaux. Effects of all-atom force fields
on amyloid oligomerization: Replica exchange molecular dynamics simulations of the a?
16?22 dimer and trimer. Phys. Chem. Chem. Phys., 13(20):9778?9788, 2011.
[287] Kevin R MacKenzie, James H Prestegard, and Donald M Engelman. A transmembrane
helix dimer: structure and implications. Science, 276(5309):131?133, 1997.
[288] Madhusoodanan Mottamal, Jinming Zhang, and Themis Lazaridis. Energetics of the
native and non-native states of the glycophorin transmembrane helix dimer. PROTEINS:
Structure, Function, and Bioinformatics, 62(4):996?1009, 2006.
[289] Emi Psachoulia, Philip W Fowler, Peter J Bond, and Mark SP Sansom. Helix- helix
interactions in membrane proteins: Coarse-grained simulations of glycophorin a helix
dimerization. Biochemistry, 47(40):10503?10512, 2008.
[290] P. McColgan and S. J. Tabrizi. Huntington?s disease: a clinical review. Eur J Neurol,
25(1):24?34, January 2018.
184
[291] Christopher A Ross and Sarah J Tabrizi. Huntington?s disease: from molecular
pathogenesis to clinical treatment. The Lancet Neurology, 10(1):83?98, January 2011.
[292] Gillian Bates. Huntingtin aggregation and toxicity in Huntington?s disease. The Lancet,
361(9369):1642?1644, May 2003.
[293] Francis O Walker. Huntington?s disease. The Lancet, 369(9557):218?228, January 2007.
[294] Matthew P. Parsons and Lynn A. Raymond. Huntington Disease. In Neurobiology of Brain
Disorders, pages 303?320. Elsevier, 2015.
[295] Hueng-Chuen Fan, Li-Ing Ho, Ching-Shiang Chi, Shyi-Jou Chen, Giia-Sheun Peng, Tzu-
Min Chan, Shinn-Zong Lin, and Horng-Jyh Harn. Polyglutamine (PolyQ) Diseases:
Genetics to Treatments. Cell Transplant, 23(4-5):441?458, May 2014.
[296] J. Shao and M. I. Diamond. Polyglutamine diseases: emerging concepts in pathogenesis
and therapy. Human Molecular Genetics, 16(R2):R115?R123, July 2007.
[297] Sunjay Barton, Ron Jacak, Sagar D. Khare, Feng Ding, and Nikolay V. Dokholyan. The
Length Dependence of the PolyQ-mediated Protein Aggregation. Journal of Biological
Chemistry, 282(35):25487?25492, August 2007.
[298] Tetyana Yushchenko, Elke Deuerling, and Karin Hauser. Insights into the Aggregation
Mechanism of PolyQ Proteins with Different Glutamine Repeat Lengths. Biophysical
Journal, 114(8):1847?1857, April 2018.
[299] Keizo Sugaya, Shiro Matsubara, Yasuhiro Kagamihara, Akihiro Kawata, and Hideaki
Hayashi. Polyglutamine Expansion Mutation Yields a Pathological Epitope Linked to
Nucleation of Protein Aggregate: Determinant of Huntington?s Disease Onset. PLoS ONE,
2(7):e635, July 2007.
[300] Andreas Vitalis, Xiaoling Wang, and Rohit V. Pappu. Quantitative Characterization of
Intrinsic Disorder in Polyglutamine: Insights from Analysis Based on Polymer Theories.
Biophysical Journal, 93(6):1923?1937, September 2007.
[301] Scott L. Crick, Murali Jayaraman, Carl Frieden, Ronald Wetzel, and Rohit V. Pappu.
Fluorescence correlation spectroscopy shows that monomeric polyglutamine molecules
form collapsed structures in aqueous solutions. Proc. Natl. Acad. Sci. U.S.A.,
103(45):16764?16769, November 2006.
[302] Natasha H. Rhys, Alan K. Soper, and Lorna Dougan. Hydrophilic Association in a Dilute
Glutamine Solution Persists Independent of Increasing Temperature. J. Phys. Chem. B,
119(51):15644?15651, December 2015.
[303] Jeremy T Lant, Rashmi Kiri, Martin L Duennwald, and Patrick O?Donoghue. Formation
and persistence of polyglutamine aggregates in mistranslating cells. Nucleic Acids
Research, 49(20):11883?11899, November 2021.
185
[304] Hongsuk Kang, Francisco X. Va?zquez, Leili Zhang, Payel Das, Leticia Toledo-Sherman,
Binquan Luan, Michael Levitt, and Ruhong Zhou. Emerging ?-Sheet Rich Conformations
in Supercompact Huntingtin Exon-1 Mutant Structures. J. Am. Chem. Soc., 139(26):8820?
8827, July 2017.
[305] Nitin K. Pandey, J. Mario Isas, Anoop Rawat, Rachel V. Lee, Jennifer Langen, Priyatama
Pandey, and Ralf Langen. The 17-residue-long N terminus in huntingtin controls stepwise
aggregation in solution and on membranes via different mechanisms. Journal of Biological
Chemistry, 293(7):2597?2605, February 2018.
[306] Maria Lucia Pigazzini, Mandy Lawrenz, Anca Margineanu, Gabriele S.
Kaminski Schierle, and Janine Kirstein. An Expanded Polyproline Domain Maintains
Mutant Huntingtin Soluble in vivo and During Aging. Front. Mol. Neurosci., 14:721749,
October 2021.
[307] James R. Arndt, Maxmore Chaibva, and Justin Legleiter. The emerging role of the first 17
amino acids of huntingtin in Huntington?s disease. Biomolecular Concepts, 6(1):33?46,
March 2015.
[308] Guillaume Drin and Bruno Antonny. Amphipathic helices and membrane curvature. FEBS
Letters, 584(9):1840?1847, May 2010.
[309] Nikos S Hatzakis, Vikram K Bhatia, Jannik Larsen, Kenneth L Madsen, Pierre-Yves
Bolinger, Andreas H Kunding, John Castillo, Ulrik Gether, Per Hedega?rd, and Dimitrios
Stamou. How curved membranes recruit amphipathic helices and protein anchoring
motifs. Nat Chem Biol, 5(11):835?841, November 2009.
[310] Anna Giangaspero, Luca Sandri, and Alessandro Tossi. Amphipathic ? helical
antimicrobial peptides.: A systematic study of the effects of structural and physical
properties on biological activity. European Journal of Biochemistry, 268(21):5589?5600,
November 2001.
[311] Ashwani K Thakur, Murali Jayaraman, Rakesh Mishra, Monika Thakur, Veronique M
Chellgren, In-Ja L Byeon, Dalaver H Anjum, Ravindra Kodali, Trevor P Creamer, James F
Conway, Angela M Gronenborn, and Ronald Wetzel. Polyglutamine disruption of the
huntingtin exon 1 N terminus triggers a complex aggregation mechanism. Nat Struct Mol
Biol, 16(4):380?389, April 2009.
[312] Cody L. Hoop, Hsiang-Kai Lin, Karunakar Kar, Ga?bor Magyarfalvi, Jonathan M. Lamley,
Jennifer C. Boatz, Abhishek Mandal, Jo?zef R. Lewandowski, Ronald Wetzel, and Patrick
C. A. van der Wel. Huntingtin exon 1 fibrils feature an interdigitated ?-hairpin?based
polyglutamine core. Proc. Natl. Acad. Sci. U.S.A., 113(6):1546?1551, February 2016.
[313] Valentina Fodale, Natalie C. Kegulian, Margherita Verani, Cristina Cariulo, Lucia
Azzollini, Lara Petricca, Manuel Daldin, Roberto Boggio, Alessandro Padova, Rainer
Kuhn, Robert Pacifici, Douglas Macdonald, Ryan C. Schoenfeld, Hyunsun Park,
J. Mario Isas, Ralf Langen, Andreas Weiss, and Andrea Caricasole. Polyglutamine-
and Temperature-Dependent Conformational Rigidity in Mutant Huntingtin Revealed
186
by Immunoassays and Circular Dichroism Spectroscopy. PLoS ONE, 9(12):e112262,
December 2014.
[314] Yoshitaka Nagai, Takashi Inui, H Akiko Popiel, Nobuhiro Fujikake, Kazuhiro Hasegawa,
Yoshihiro Urade, Yuji Goto, Hironobu Naiki, and Tatsushi Toda. A toxic monomeric
conformer of the polyglutamine protein. Nat Struct Mol Biol, 14(4):332?340, April 2007.
[315] Matthias Michalek, Evgeniy S. Salnikov, and Burkhard Bechinger. Structure and Topology
of the Huntingtin 1?17 Membrane Anchor by a Combined Solution and Solid-State NMR
Approach. Biophysical Journal, 105(3):699?710, 2013.
[316] Meixin Tao, Nitin K. Pandey, Ryan Barnes, Songi Han, and Ralf Langen. Structure of
Membrane-Bound Huntingtin Exon 1 Reveals Membrane Interaction and Aggregation
Mechanisms. Structure, 27(10):1570?1580.e4, October 2019.
[317] Nicholas W. Kelley, Xuhui Huang, Stephen Tam, Christoph Spiess, Judith Frydman, and
Vijay S. Pande. The Predicted Structure of the Headpiece of the Huntingtin Protein and Its
Implications on Huntingtin Aggregation. Journal of Molecular Biology, 388(5):919?927,
May 2009.
[318] Yanting Wang and Gregory A. Voth. Molecular Dynamics Simulations of Polyglutamine
Aggregation Using Solvent-Free Multiscale Coarse-Grained Models. J. Phys. Chem. B,
114(26):8735?8743, July 2010.
[319] Maciej D?ugosz and Joanna Trylska. Secondary Structures of Native and Pathogenic
Huntingtin N-Terminal Fragments. J. Phys. Chem. B, 115(40):11597?11608, October
2011.
[320] Vinal V. Lakhani, Feng Ding, and Nikolay V. Dokholyan. Polyglutamine Induced
Misfolding of Huntingtin Exon1 is Modulated by the Flanking Sequences. PLoS Comput
Biol, 6(4):e1000772, April 2010.
[321] Leili Zhang, Hongsuk Kang, Francisco X. Va?zquez, Leticia Toledo-Sherman, Binquan
Luan, and Ruhong Zhou. Molecular Mechanism of Stabilizing the Helical Structure
of Huntingtin N17 in a Micellar Environment. The Journal of Physical Chemistry B,
121(18):4713?4721, 2017. Publisher: American Chemical Society.
[322] Se?bastien Co?te?, Vincent Binette, Evgeniy S. Salnikov, Burkhard Bechinger, and Normand
Mousseau. Probing the Huntingtin 1-17 Membrane Anchor on a Phospholipid Bilayer by
Using All-Atom Simulations. Biophysical Journal, 108(5):1187?1198, 2015.
[323] Kevin J. Boyd and Eric R. May. BUMPy: A model-independent tool for constructing
lipid bilayers of varying curvature and composition. Journal of chemical theory and
computation, 14(12):6642?6652, 2018.
[324] Mingchen Chen and Peter G. Wolynes. Aggregation landscapes of Huntingtin exon 1
protein fragments and the critical repeat length for the onset of Huntington?s disease. Proc.
Natl. Acad. Sci. U.S.A., 114(17):4406?4411, April 2017.
187
[325] Se?bastien Buchoux. FATSLiM: a fast and robust software to analyze MD simulations of
membranes. Bioinformatics, 33(1):133?134, January 2017.
[326] Annika Urbanek, Matija Popovic, Anna Morato?, Alejandro Estan?a, Carlos A. Elena-Real,
Pablo Mier, Aure?lie Fournet, Fre?de?ric Allemand, Stephane Delbecq, Miguel A. Andrade-
Navarro, Juan Corte?s, Nathalie Sibille, and Pau Bernado?. Flanking Regions Determine
the Structure of the Poly-Glutamine in Huntingtin through Mechanisms Common among
Glutamine-Rich Human Proteins. Structure, 28(7):733?746.e5, July 2020.
[327] E.P Melo, M.R Aires-Barros, and J.M.S Cabral. Reverse micelles and protein
biotechnology. In Biotechnology Annual Review, volume 7, pages 87?129. Elsevier, 2001.
[328] Joe?lle Bigay, Jean-Franc?ois Casella, Guillaume Drin, Bruno Mesmin, and Bruno Antonny.
ArfGAP1 responds to membrane curvature through the folding of a lipid packing sensor
motif. EMBO J, 24(13):2244?2253, July 2005.
[329] Xinyao Yi, Yi Zhang, Ming Gong, Xiang Yu, Narek Darabedian, Jie Zheng, and Feimeng
Zhou. Ca2+ Interacts with Glu-22 of Abeta(1?42) and Phospholipid Bilayers to Accelerate
the Abeta(1?42) Aggregation Below the Critical Micelle Concentration. Biochemistry,
54(41):6323?6332, 2015. Publisher: American Chemical Society.
[330] Marina E. Monteiro, Maria J. Sarmento, and Fa?bio Fernandes. Role of calcium in
membrane interactions by PI(4,5)P2-binding proteins, 2014.
[331] M. Mirza, Y. Guo, K. Arnold, C. J. Van Oss, and S. Ohki. Hydrophobizing effect of cations
on acidic phospholipid membranes. J Dispers Sci Technol, 19(6-7):951?962, 1998.
[332] H Hauser and R M Dawson. The binding of calcium at lipid-water interfaces. Eur. J.
Biochem., 1(1):61?619, 1967.
[333] Baharan Ali Doosti, Weria Pezeshkian, Dennis S. Bruhn, John H. Ipsen, Himanshu
Khandelia, Gavin D.M. Jeffries, and Tatsiana Lobovkina. Membrane Tubulation in Lipid
Vesicles Triggered by the Local Application of Calcium Ions. Langmuir, 33(41):11010?
11017, 2017.
[334] Z. T. Graber, Z. Shi, and T. Baumgart. Cations induce shape remodeling of negatively
charged phospholipid membranes. Phys. Chem. Chem. Phys., 19(23):15285?15295, 2017.
[335] Mijo Simunovic, Ka Yee C. Lee, and Patricia Bassereau. Celebrating Soft Matter?s 10th
anniversary: screening of the calcium-induced spontaneous curvature of lipid membranes.
Soft Matter, 11(25):5030?5036, 2015.
[336] Ewa Stodolak-Zych, Piotr Jelen?, Ewa Dzierzkowska, Ma?gorzata Krok-Borkowicz,
?ukasz Zych, Maciej Bogun?, Alicja Rapacz-Kmita, and Beata Kolesin?ska. Modification
of chitosan fibers with short peptides as a model of synthetic extracellular matrix. Journal
of Molecular Structure, 1211:128061, 2020.
188
[337] A. Muxika, A. Etxabide, J. Uranga, P. Guerrero, and K. de la Caba. Chitosan as a bioactive
polymer: Processing, properties and applications. International Journal of Biological
Macromolecules, 105:1358?1368, 2017.
[338] Hongcheng Xu and Silvina Matysiak. Effect of pH on chitosan hydrogel polymer network
structure. Chemical Communications, 53(53):7373?7376, 2017. Publisher: The Royal
Society of Chemistry.
189