UNDERSTANDING SPECIFICITY OF SMALL MOLECULE INHIBITORS OF REGULATORS OF

G-PROTEIN SIGNALING (RGS) PROTEINS

By

Vincent Shaw

A DISSERTATION

Michigan State University

in partial fulfillment of the requirements

Submitted to

for the degree of

Pharmacology and Toxicology—Doctor of Philosophy

2019

ABSTRACT

UNDERSTANDING SPECIFICITY OF SMALL MOLECULE INHIBITORS OF REGULATORS OF

G-PROTEIN SIGNALING (RGS) PROTEINS

By

Vincent Shaw

Regulators of G-protein Signaling (RGS) proteins terminate G-Protein Coupled Receptor

(GPCR) signaling by binding to active Gα subunits and accelerating hydrolysis of GTP. Targeting

RGS proteins with inhibitors is a strategy to increase receptor-mediated signaling. There are sev-

eral existing RGS inhibitors, which including the thiadiazolidinones (TDZDs). All RGS inhibitors

discovered to date are covalent modifiers of cysteine residues and these act preferentially on RGS4

over other RGS isoforms. To widen the scope of therapeutic potential of RGS inhibitors, it would

be useful to have inhibitors with specificities for other isoforms. To aid in the development of

new inhibitors, it will be important to understand what factors are responsible for RGS isoform

selectivity. While RGS isoforms vary in their number and location of cysteines, cysteines that

are shared among most RGS proteins are buried beneath the protein surface. We hypothesize

that there is a dual role for cysteine complement and protein dynamics that drives specificity of

TDZD inhibitors.

Interestingly, representative RGS proteins RGS4, RGS8, and RGS19 have dramatic differ-

ences in potency of inhibition when mutated to contain a single cysteine. Hydrogen-deuterium

exchange (HDX) was used to evaluate differences in flexibility among RGS proteins, and deu-

terium incorporation was found to be correlated with TDZD potency. Molecular dynamics stud-

ies supported these differences in flexibility, and illustrated that flexibility differences may un-

derlie solvent accessibility of shared cysteines. To understand what structural elements control

RGS domain flexibility, we focused on interhelical salt bridge-forming residues that differ among

the RGS isoforms. Mutations that induced salt bridge formation in RGS19 decreased its flexi-

bility and decreased potency of TDZD inhibition, while salt bridge removal in RGS8 and RGS4

increased flexibility and increased potency of inhibition. This suggests a causative relationship

between protein dynamics and inhibitor potency. The movements observed in these proteins

suggest that cysteines may be exposed to solvent by formation of a transient pocket, which may

be taken advantage of in the design of non-covalent inhibitors. Finally, the role of individual

conserved cysteines was evaluated. NMR studies of single-cysteine RGS8 mutants demonstrated

that inhibitors can interact with either cysteine. Mass spectrometry studies showed that a TDZD

inhibitor may mediate an interaction between the α4 and α7 cysteines in WT RGS8 by formation

of a disulfide bond. As a whole, this work demonstrates a role for both cysteine interaction and

protein dynamics in the control of RGS inhibitor selectivity.

ACKNOWLEDGEMENTS

I have a lot of people to thank that have supported me in the past five years. All of the lab

members, current and former, were there to offer advice on countless occasions. These include,

but are by no means limited to: Erika, Benita, Jeff, Kate, Behirda, Tom, Sean, Jade, Cassie, Hoa,

Yajing, Maria, Nils, Maja, Indi, Zhangzhe, Clarissa, Charuta, and Melissa. Their presence has

kept lab life fun and spirits high. Special thanks also to Josiah, my undergraduate mentee who

has helped with countless protein preps and experiments. Thanks to the guidance of Dr. Benita

Sjögren, Dr. Harish Vashisth, Dr. Karen Liby, and Dr. Jon Kaguni.

In addition, having such a

supportive department has been invaluable to getting by as a graduate student.

I’m grateful

that all of the other grad students, our GSO, the faculty, our administrative staff, and Dr. Anne

Dorrance, our Graduate Program Director, have always been in my corner.

Much of this work was made possible through collaboration. This thesis features contribu-

tions that include MD simulations and analysis done by Drs. Mohammadjavad Mohammadi and

Hossein Mohammadiarani in the lab of Dr. Harish Vashisth, UNH; NMR work done in conjunc-

tion with Ryan Puterbaugh and Dr. Krisztina Varga, also at UNH; and ongoing screening efforts

from Dr. Arzu Uyar and Dr. Alex Dickson at MSU. I was helped greatly by Dr. Schilmiller and

Dr. Jones in the RTSF Mass Spectrometry and Metabolomics Core and Dr. Sundari Chodavarapu

in the lab of Dr. Jon Kaguni, who helped me develop a workflow for mass spectrometry detection

of hydrogen/deuterium exchange.

I’m lucky to have had friends close at hand, including Alex, Hannah, Charlotte, Erin,

Kelsey, Lynne, Dan, Shane, Kim, Sarah, Tim, Megan, and Steven. It’s a wonder I got anything

done with Steven roping me into side projects. I’m also very fortunate to have family that dou-

iv

ble as my closest friends: Mom, Dad, Andy, and Margaret; Rich, Sandy, Patrick, and Laurel; and

especially, my wife Kate.

Finally, I owe a lot of thanks to my mentor, Dr. Richard Neubig. Rick’s Socratic method

always helped me arrive at a clearer understanding and led me to ask better scientific questions

on my own. I always came away from our meetings with solidified plans for further experiments

and feeling motivated.

v

TABLE OF CONTENTS

LIST OF TABLES

LIST OF FIGURES

KEY TO ABBREVIATIONS

CHAPTER 1: Introduction
Challenging drug targets

.

.

.

.

.

.

.

.

.
.

.
.

.
.
.
.
.
.
.

.
RGS proteins as therapeutic targets

.
.
Protein-protein interactions
.
Importance of protein dynamics .
Covalent modifiers
.
.

.
.
.
.
.
.
.
.
.
.
.
The G-protein signaling pathway .
.
RGS protein diversity .
.
Physiology of RGS proteins in disease .
.
.
.
.

.
.
Recent discovery efforts .
.
Thiadiazolidinone characterization .
.

Contribution of this work .

RGS inhibitors .

.
.
.
.

.
.

.
.

.
.

.
.

.
.

.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

CHAPTER 2: Differential Protein Dynamics of Regulators of G-Protein Signaling:

.

.

.

.

.
.

.
.

.
.

.
.

.
.

.
.

.
.

.
.

.
.

.
.

.
.
.

.
.
.

Role in Specificity of Small-Molecule Inhibitors
.
Introduction .
.
.
Materials and Methods
.
.
.
.
.
.
.
.

.
.
.
.
Protein expression and purification .
.
Flow cytometry protein interaction assay .
.
.
Hydrogen/deuterium exchange .
.
.
System setup and simulation details .
.
.
RMSD, RMSF, and SASA Measurements .
.
.
.
.
.
.

Results
.
Discussion .
Conclusions

.
.
.
.
.
.
.
.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.

.

.

.

.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

CHAPTER 3: An Interhelical Salt Bridge Controls Flexibility and Inhibitor Po-

.

.

.

.

tency For Regulators of G-protein Signaling (RGS) Proteins 4, 8, and 19
.
.
Introduction .
.
.
.
Materials and Methods
.
.
.
.
.
.
.
.
.

.
.
.
.
.
Materials .
.
.
Protein Expression and Purification .
Differential Scanning Fluorimetry .
.
.
Flow Cytometry Protein Interaction Assay (FCPIA)

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.

.

.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

vi

ix

x

xxi

1
2
3
5
7
8
8
12
14
17
17
19
22

23
24
25
25
26
26
29
29
30
38
41

42
43
44
44
44
45
45

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.

.

.

.
Hydrogen-Deuterium Exchange .
Molecular Dynamics (MD) Simulation .
.
Dynamic cross-correlation analysis
.
Analysis of salt-bridge interactions
Statistical Analysis
.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.

.
.

.
.

.
.

.
.

.
.

.

Results
.
Discussion .

.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

CHAPTER 4: Distinct Roles of Individual Cysteines in Covalent Inhibition of RGS

Proteins
Introduction .
.
Materials and Methods

.

.

.

Results and Discussion .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.
.

.

.
.

.

.
.

.
.

.
.

.
.

.
.

.
.

.
.

.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
Protein purification and expression .
.
NMR Spectroscopy .
.
.
.
Iodoacetamide alkylation and trypsin digestion .
.
Protection of RGS8 from iodoacetamide akylation by CCG-203769 .
.
Protein mass spectrometry .
.
Flow cytometry protein interaction assay .
.
.
.
Non-reducing SDS-PAGE .
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Cys148 in RGS4 is more accessible to a covalent modifier than Cys95 .
.
.
CCG-203769 can directly act upon either cysteine in RGS8 .
.
.
.
Cys160 RGS8 is more sensitive to compound-induced denaturation than WT.
.
.
Functional inhibition by CCG-203769 is altered in cysteine mutants .
.
CCG-203769 induces an intra-protein disulfide in WT RGS8.
.
.
Among single-cysteine RGS8 mutants (Cys107 and Cys160), CCG-203769 induces
.
.
.

.
CCG-203769 induces inter-protein disulfide in RGS4 .
.

dimerization via an inter-protein disulfide. .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.
.

.
.

.
.

.
.

.
.

.
.

.
.

.
.

.
.

.
.

.
.

.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Conclusions

.

.

.

.

.

.

.

.
Introduction .
Approach and Results .

CHAPTER 5: Identification of Transient Pockets in RGS4 and RGS19
.
.
.
.
.
.

.
.
Pocket Identification .
Pocket Clustering .
.
Frames for screening .
.

Conclusions

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.

.
.

.

.

.

.

.

.

.

.

.

CHAPTER 6: Conclusions and Future Directions

.
Role of individual cysteines in action of inhibitors .
.
Role of protein dynamics in RGS inhibitor selectivity .
.
Future research in understanding action of TDZD inhibitors
Continuing discovery of non-covalent inhibitors .
.

.
.

.

.

.

.

.

.
.

.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.
.
.

.
.
.
.

.
.
.
.
.
.

.
.
.
.

.
.
.
.
.
.

.
.
.
.

.
.
.
.
.
.

.
.
.
.

.
.
.
.
.
.

.
.
.
.

.
.
.
.
.
.

.
.
.
.

.
.
.
.
.
.

.
.
.
.

.
.
.
.
.
.

.
.
.
.

APPENDIX

vii

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.

.
.
.
.
.
.

.
.
.
.

46
46
47
48
48
49
61

64
65
66
66
67
68
69
69
69
70
70
70
71
75
75
76

78
79
80

83
84
86
86
88
90
91

93
94
94
96
98

99

REFERENCES

158

viii

LIST OF TABLES

Table 2-1:

Summary of MD simulations.

Table 3-2:

Details of MD simulations.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Table 3-3: The salt-bridge interaction within the α4-α7 bundle of helices in single-
cysteine structure of RGS4, RGS8, and RGS19 from MD simulations and
potency of CCG-50014 inhibition of single-cysteine RGS proteins in our
.
previous work.144

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Table 3-4:

Interaction affinities between Gαo and RGS proteins and mutants

.

.

.

.

.

Table A-1:

Model definitions and corresponding metrics. Among models reported in
the literature are models M1 through M6 (empirical models) and the model
M7 (a fractional population model). For models reported in this work, M8
is an empirical model and M9 is a fractional population model. Additional
details on models M8 and M9 are provided in supporting information. .
.

.

.

.

.

.

34

47

50

58

. 104

Table A-2:

Summary of MD simulations.

.

Table A-3:

Models proposed in this work .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

. 124

. 127

Table A-4:

Details on all protection factor correlation models with the default and re-
optimized values of their parameters. Optimized values based upon simu-
lations conducted using CHARMM and AMBER force-fields are listed with
superscripts ch and am, respectively. In addition, details on two new models
.
M8 and M9 proposed in this work are listed.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

. 128

ix

LIST OF FIGURES

Figure 1-1:

Activation of G-protein signaling upon agonist binding at GPCR.

.

.

.

.

.

Figure 1-2:

Figure 1-3:

Activation of different signaling pathways is mediated by different
G-protein subtypes.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

RGS are GTPase-Activating Proteins (GAPs). They terminate G-protein sig-
naling by catalyzing hydrolysis of GTP on Gα.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure 1-4: The circuitry of the motor pathway. RGS4 is expressed in the striatum. In
the Parkinson’s disease state, dopaminergic input from the substantia nigra
to the striatum is lost. .
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure 1-5: The role of RGS4 in response to dopamine signaling in the indirect and direct
.

pathway spiny projection neurons of the striatum. .

.

.

.

.

.

.

.

.

.

.

.

.

Figure 1-6: Thiadiazolidinones CCG-50014, the lead compound, and CCG-203769, an
.

analog with improved specificity for RGS4.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure 1-7:

Figure 2-1:

Figure 2-2:

Figure 2-3:

Locations of cysteines in RGS protein. Gα is shown in gray spheres, RGS in
shown in light blue. Cysteines 71 and 132 in RGS4 (red) are not conserved
among RGS proteins. Cysteine 148 in RGS4 (blue) is shared by RGS8 and
RGS4. Cysteine 95 in RGS4 (green) is the best conserved cysteine among
RGS proteins, found in all isoforms except RGS6 and RGS7. .
.

.

.

.

.

.

.

.

Alignment of fragments observed by mass spectrometry following cleavage
of RGS proteins by pepsin. Horizontal bars indicate length and position of
observed fragments. The two N-terminal residues of each fragment were
excluded from analysis due to rapid back-exchange. Vertical gray boxes
.
indicate approximate positions of helices within the RGS domain.

.

.

.

.

(A) Locations of cysteines in RGS4, RGS8, and RGS19. (B) Potency of CCG-
50014 against RGS19, which has only one cysteine, and mutant RGS4 and
.
RGS8 containing only the shared α4 helix cysteine. n=3.

.

.

.

.

.

.

.

.

.

(A-E) Kinetics of deuterium exchange in selected protein fragments from
(A) α4, (B) α5, (C) α5-α6 interhelical region, (D) α6 and (E) α7. Sequences
of observed fragments are aligned with residue numbers of each fragment
.
indicated. Cysteine locations are marked in red. n=3.

.

.

.

.

.

.

.

.

.

.

.

x

.

.

.

.

.

.

.

.

.

.

10

11

12

16

18

20

21

28

31

32

Figure 2-4:

Figure 2-5:

Figure 2-6:

Figure 2-7:

Figure 2-8:

Figure 2-9:

Figure 3-1:

Figure 3-2:

(A) Global kinetics of deuterium exchange. Deuterium incorporation (DI) is
expressed as a percent of exchangeable amide hydrogen positions. Where
fragments overlap, data is displayed as average DI of observed fragments.
(B) Degree of DI at 300 minutes in 90% D2O is mapped onto protein structure
of RGS4, RGS8, and RGS19. n=3.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Root mean squared fluctuations (RMSF) per residue during 2 μs MD simula-
tions of (A) RGS4 (PDB: 1AGR), (B) RGS8 (PDB: 2ODE), and (C) RGS19 (PDB:
1CMZ). The RMSF trends for each protein for the simulation set 2 are shown
in Fig. 2-6. Gray bars indicate helical regions.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Root mean squared fluctuations across protein sequence during 3 μs MD
simulations of (A) RGS4 (PDB: 1AGR), (B) RGS8 (PDB: 2ODE), and (C) RGS19
(PDB: 1CMZ). Gray bars indicate helical regions. .
.

.

.

.

.

.

.

.

.

.

.

.

.

.

Solvent-accessible surface areas (SASA) are shown for sulfur atoms in
shared cysteines on α4 helix for simulation set 1 (A) and set 2 (B) in RGS4,
RGS8, and RGS19, and for shared cysteines on α6-α7 interhelical loop in
simulation set 1 (C) and set 2 (D) in RGS4 and RGS8. .
.

.

.

.

.

.

.

.

.

.

.

.

Conformational changes during molecular dynamics simulations. Root
mean square deviations of α6 helix and α6-α7 loop, starting conformation,
and a snapshot conformation during MD simulation are shown for (A, D,
G) RGS4, (B, E, H) RGS8, and (C, F, I) RGS19. Protein regions plotted in
MD trajectories are depicted in color in protein structures. Arrows indicate
.
locations of notable solvent exposure during simulation.

.

.

.

.

.

.

.

.

.

Snapshot of RGS19 from simulation set 2 at 240 ns. Cleft opening observed
.
in simulation set 1 (Fig 6I) was recapitulated in this simulation.

.

.

.

.

.

.

.

.

.

.

.

(A) Alignment of RGS19, RGS4, and RGS8 sequences in α4-α7 helix bundle.
Charged residues that make interhelical contacts are indicated in red and
blue. RGS19 has 1, RGS4 has 3, and RGS8 has 4 salt bridges. Structural
alignments of α4-α5 (B), α5-α6 (C), and α6-α7 (D) helix pairs are shown, with
highlighted residues in panel a rendered as sticks. RGS19 (PDB 1CMZ) is
in green, RGS4 (PDB 1AGR) is in yellow, and RGS8 (PDB 5DO9) is in cyan.
Black brackets in panel A indicate residues depicted in panels B, C, and D .

33

35

35

36

37

38

50

L118D mutation increases thermal stability of RGS19, but Q183K mutation
has no significant effect (n = 3, 1-way ANOVA with Sidak’s multiple compar-
ison test. ****p < 0.001). L118D mutation in RGS19 has reduced potency of
inhibition of CCG-50014, but Q183K mutation does not. Ki, calculated using
a Cheng-Prusoff correction,232 is reported to account for effect of mutations
.
in RGS on Gαo affinity.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

51

xi

Figure 3-3: Thermal stability was determined by differential scanning fluorimetry. (A)
The L118D mutation in RGS19 increased melting temperature by 7 ℃ com-
pared to WT. (B) The E84L mutation in RGS8 decreased melting temperature
by 8 ℃. (C) The RGS4 D90L mutation introduced a biphasic melt curve and
increased melting temperature by 5 ℃. For each pair, the three replicate
derivative melt curves are shown on the left and average melt temperatures
are shown on the right. Error bars represent SD. n = 3. Analyzed by 1-way
ANOVA with Sidak’s Multiple Comparisons test. ****p < 0.0001 .
.

.

.

.

.

Figure 3-4: The traces of root-mean-squared-deviation (RMSD) vs. simulation time (μs)
for (a) RGS4 D90L, (b) RGS8 E84L, and (c) RGS19 L118D. Two independent
simulation runs for each structure are presented, and the wild-type runs are
.
presented from our previous work.144

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure 3-5:

ΔRMSF in between WT and mutant simulation trajectories .

.

.

.

.

.

.

.

.

Figure 3-6:

Figure 3-7:

Figure 3-8:

Figure 4-1:

Dynamic cross correlation matrix calculated for the Cα atoms of (A)
RGS19/RGS19 L118D, (B) RGS8/RGS8 E84L, (C) RGS4/RGS4 D90L. Hor-
izontal dotted lines indicate the regions of the α4 helix, while vertical
solid lines indicate the regions of the α5 helix for each protein. The color
scheme ranges from anticorrelation (-1.0, blue), no correlation (0, green),
and positive correlation (+1.0, red). Values are the average for the two
.
independent simulation runs.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Difference in %deuterium incorporation (Δ%DI) between mutated and un-
mutated proteins in RGS19 L118D (A), RGS8 E84L (B), and RGS4 D90L (C)
fragments, as measured by HDX. Red arrows indicate fragments containing
mutated residue, and black arrows indicate fragments containing conserved
α4 cysteine. Kinetics of deuterium incorporation in these fragments for indi-
vidual constructs are shown below. n = 3. Error bars represent SD. Analyzed
by 2-way ANOVA with Sidak’s multiple comparisons test. *p < 0.05, **p <
.
0.01, ****p < 0.0001.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Potency of inhibition of CCG-50014 against α 4 is altered in salt bridge
mutants of RGS proteins. (A) RGS4 IC50: 8.8 µM, RGS4 D90L IC50: 2.2 µM.
(B) RGS8 IC50: 29 µM, RGS8 E84L IC50: 4.6 µM. (C) RGS19 IC50: 7.0 µM,
.
RGS19 L118D IC50: 1.1 µM. n=3. .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

(A) Locations of cysteines in RGS protein based on structure of RGS4 (PDB:
1AGR). α4 and α7 cysteines, conserved across multiple RGS proteins, are
marked in blue. The α3 and α6 helix cysteines, unique to RGS4, are marked
in red. (B) Degree of IAA alkylation at Cys71 (α3), Cys95 (α4), Cys132 (α6),
.
and Cys148 (α7) in RGS4.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

xii

.

.

.

53

54

55

.

57

.

.

.

59

60

71

Figure 4-2: WT RGS8 protein NMR spectra. (A) 1H-15N HSQC NMR spectrum of WT
RGS8. (B) The structure of ligand CCG-203769. (C) Overlay of 1H-15N HSQC
NMR spectra of WT RGS8 before (red spectrum) and after the addition of
its ligand CCG-203769 at 1:1, 1:2, and 1:4 RGS8:ligand ratio (grey spectra).
Shifted residues are highlighted in the zoomed spectrum. Spectra were ac-
quired at 25 ℃ on a Bruker AVANCE III HD 800 MHz NMR spectrometer
equipped with a TCI Cryoprobe at the CUNY Advanced Science Research
Center NMR facility.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

73

Figure 4-3:

Figure 4-4:

Figure 4-5:

Figure 4-6:

Chemical shift perturbation of WT and single-cysteine RGS8 protein NMR
spectra upon the addition of ligand CCG-203769 1H-15N HSQC NMR spec-
tra of RGS8 were overlaid before (red spectrum) and after the addition of
its ligand CCG-203769 at 1:1 RGS8:ligand ratio (black spectra) for (A) WT
RGS8 (B) Cys107 RGS8, and (C) Cys160 RGS8. (D) The magnitude of chemical
shift perturbation. Spectra were acquired at 25 ℃ on a Bruker AVANCE III
HD 800 MHz (WT and Cys107 RGS8) or a Bruker AVANCE III HD 700 MHz
(Cys160 RGS8) NMR spectrometers equipped with Cryoprobes at the CUNY
Advanced Science Research Center NMR facility. .
.

.

.

.

.

.

.

.

.

.

.

.

.

.

Inhibition of RGS-Gα binding for WT, Cys160, and Cys107 RGS8 in response
to increasing concentrations of CCG-203769 was measured by FCPIA. WT
IC50 = 25 μM), Cys160 IC50 = 2.2 μM, and Cys107 was not inhibited.
.

.

.

.

.

CCG-203769 masks cysteine alkylation by IAA by inducing disulfide bond.
(A) Deconvoluted mass spectra of WT RGS8 (first column), Cys160 RGS8 (sec-
ond column), and Cys107 RGS8 (third column). Spectra were taken before
treatment (first row), after excess of of IAA (second row), and pretreated
with CCG-203769 before addition of IAA (third row). (B) WT, Cys160, and
Cys107 RGS8 mass changes analyzed by SDS-PAGE after treatment with vehi-
cle, 250 μM CCG-203769, or CCG-203769 followed by 1 mM DTT. Monomer
mass indicated with black arrow and dimer mass indicated with red arrow.

RGS4, RGS8, and RGS19 mass changes analyzed by SDS-PAGE after treat-
ment with vehicle, 250 μM CCG-203769, or CCG-203769 followed by 1 mM
.
DTT.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure 4-7:

Proposed mechanism of disulfide bond induction by CCG-203769 in RGS8 .

Figure 5-1:

Figure 5-2:

Locations of pocket-forming residues in RGS4 (top) and RGS19 (bottom).
Color indicates frequency with which each atom touches a pocket alpha
sphere. Blue is less frequent and red is more frequent. .
.

.

.

.

.

.

.

.

.

.

.

Pocket volume and mean local hydrophobic density (MLHD) plotted over
the simulation trajectory for RGS19 (A) and RGS4 (B). Pockets in RGS19
.
were larger and more frequent than those in RGS4.

.

.

.

.

.

.

.

.

.

.

.

.

xiii

.

.

74

76

77

80

81

87

88

Figure 5-3:

Figure 5-4:

Figure A-1:

Figure A-2:

Figure A-3:

Figure A-4:

Clustering of pocket states for RGS19 (A) and RGS4 (B). Volume is plotted
against MLHD, and color indicates distinct clusters. An ensemble of pockets
representing clusters with high MLHD and a variety of pocket volumes were
selected for structure based screening.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Pocket states that are representative of cluster 4 and 7 in RGS19 (A and B)
and cluster 2 and 6 in RGS4 (C and D). Pocket-forming atoms illustrated
with white surface.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Kinetic scheme for HDX is highlighted. A conformational fluctuation in the
protein exposes buried amide groups (blue) (closed state) to solvent (open
state) where amide hydrogens (white) are exchanged by deuterium (yellow)
.
with an intrinsic rate constant kint.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Sequence and structural views of RGS proteins. (A) Sequence alignment of
RGS4, RGS8, and RGS19 is shown with conserved residues highlighted in
red; blue boxes indicate residues that are conserved between at least two
among three RGS proteins. (B) Shown are front and back views of the over-
lay of RGS4 (PDB code 1AGR), RGS8 (PDB code 2ODE), and RGS19 (PDB
code 1CMZ) structures with each of the nine helices uniquely colored. Re-
gions rendered as white cartoons are interhelical loops.
.

.

.

.

.

.

.

.

.

.

.

n

n

Comparisons of model predictions of HDX-MS data across all three RGS
proteins. Performance metrics (relative error, E, and correlation coefficient,
CC) for different models are shown based upon data averaged from all
trajectories of RGS4, RGS8, and RGS19 conducted with the CHARMM-FF
∑
∑
(data in panels A and B) and the AMBER-FF (data in panels C and D). (A,
C) The relative error between the predicted and observed %DI [E(x, y) =
∑
i=0 |xi−yi|/
i=0 yi]. (B, D) Correlation coefficient between the predicted
(yi − ¯y)2].
and observed %DI [CC(x, y) =
Gray bars are for models with the default parameters reported in the litera-
ture, blue bars are their re-optimized versions based upon our experimental
data, and red bars are for new models proposed in this work. No perfor-
mance data for the original model M5 are reported because the parameter
values were not available from the original work,42 but the performance
data are reported for the optimized version of this model (M5*) based upon
.
our experimental data.

√∑
(xi−¯x)(yi−¯y)/

(xi − ¯x)2

∑

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Comparisons of model predictions of HDX-MS data for each RGS protein.
The definitions of E and CC, and other details are the same as in Figure 3.
Colored bars distinguish data for each RGS protein: black bars, RGS4; blue
bars, RGS8; and magenta bars, RGS19 .
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

xiv

.

.

90

92

. 102

. 105

. 109

. 110

Figure A-5: The exposure of amide hydrogens in the NMR structures of RGS proteins.
Shown are the maximum (open circles) and the average (solid circles) val-
ues of the solvent accessible surface area for all amide hydrogens in the
NMR structures of RGS4 (panel A) and RGS19 (panel B). In both panels, the
absence of filled circles for certain amides as well as the absence of open
circles in panel B, is due to the approximately nil SASA values for those
amides. The absence of open circles for RGS4 in panel A is due to the lack
of availability of more than 1 conformer in the NMR structure of RGS4 as
.
opposed to 20 conformers in the NMR structure of RGS19.

.

.

.

.

.

.

.

.

Figure A-6: Mean residence times for the open and closed states of amide hydrogens.
Data are shown from all simulations of RGS4, RGS8, and RGS19 conducted
with the CHARMM-FF (panel A) and the AMBER-FF (panel B). The MRT cal-
culations were carried out using our proposed fractional population model
M9 that showed consistent predictions with the HDX-MS data. .
.

.

.

.

.

.

Figure A-7:

Figure A-8:

Experimentally measured percentage deuterium incorporation (%DI) of frag-
ments in RGS proteins at t = 0, 3, 10, 30, 100, 300, and 1000 minutes (RGS4:
.
top row; RGS8: middle row; RGS19: bottom row).

.

.

.

.

.

.

.

.

.

.

.

.

.

Definitions of fragments for each RGS protein. Each fragment comprises
residues whose color determines their location in nine α helices of each
RGS protein. Residue names in connecting loops are highlighted in black,
but shown as white cartoons in the protein structure. All helices are colored
.
and labeled in the protein rendering.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-9: Modeled deuterium incorporation of fragments in RGS4. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with default parameters (orange). This figure shows the MD simulation re-
.
sults for PDB:1AGR and AMBER force-field.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-10: Modeled deuterium incorporation of fragments in RGS4. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with default parameters (orange). This figure shows the MD simulation re-
sults for PDB:1EZT and AMBER force-field .
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-11: Modeled deuterium incorporation of fragments in RGS8. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with default parameters (orange). This figure shows the MD simulation re-
.
sults for PDB:2IHD and AMBER force-field .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-12: Modeled deuterium incorporation of fragments in RGS8. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with default parameters (orange). This figure shows the MD simulation re-
sults for PDB:2ODE and AMBER force-field .
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

xv

. 116

. 118

. 129

. 129

. 130

. 130

. 131

. 131

Figure A-13: Modeled deuterium incorporation of fragments in RGS19. The HDX exper-
iment (blue) is shown seven discrete times, alongside each different model
with default parameters (orange). This figure shows the MD simulation re-
sults for PDB:1CMZ and AMBER force-field.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-14: Modeled deuterium incorporation of fragments in RGS4. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with optimized parameters (orange). This figure shows the MD simulation
.
results for PDB:1AGR and AMBER Force-field .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-15: Modeled deuterium incorporation of fragments in RGS4. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with optimized parameters (orange). This figure shows the MD simulation
results for PDB:1EZT and AMBER Force-field .
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-16: Modeled deuterium incorporation of fragments in RGS8. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with optimized parameters (orange). This figure shows the MD simulation
results for PDB:2IHD and AMBER Force-field.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-17: Modeled deuterium incorporation of fragments in RGS8. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with optimized parameters (orange). This figure shows the MD simulation
.
results for PDB:2ODE and AMBER Force-field.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-18: Modeled deuterium incorporation of fragments in RGS19. The HDX exper-
iment (blue) is shown seven discrete times, alongside each different model
with optimized parameters (orange). This figure shows the MD simulation
results for PDB:1CMZ and AMBER Force-field. .
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-19: Modeled deuterium incorporation of fragments in RGS4. The HDX exper-
iment (blue) is shown twice, alongside new models (M8, M9) with opti-
mized parameters (orange). This figure shows the MD simulation results
.
for PDB:1AGR and AMBER Force-field.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-20: Modeled deuterium incorporation of fragments in RGS4. The HDX exper-
iment (blue) is shown twice, alongside new models (M8, M9) with opti-
mized parameters (orange). This figure shows the MD simulation results
.
for PDB:1EZT and AMBER Force-field.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-21: Modeled deuterium incorporation of fragments in RGS8. The HDX exper-
iment (blue) is shown twice, alongside new models (M8, M9) with opti-
mized parameters (orange). This figure shows the MD simulation results
.
for PDB:2IHD and AMBER Force-field.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

xvi

. 132

. 132

. 133

. 134

. 135

. 136

. 136

. 136

. 137

Figure A-22: Modeled deuterium incorporation of fragments in RGS8. The HDX exper-
iment (blue) is shown twice, alongside new models (M8, M9) with opti-
mized parameters (orange). This figure shows the MD simulation results
for PDB:2ODE and AMBER Force-field.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-23: Modeled deuterium incorporation of fragments in RGS19. The HDX ex-
periment (blue) is shown twice, alongside new models (M8, M9) with op-
timized parameters (orange). This figure shows the MD simulation results
.
for PDB:1CMZ and AMBER Force-field. .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-24: Modeled deuterium incorporation of fragments in RGS4. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with default parameters (orange). This figure shows the MD simulation re-
sults for PDB:1AGR and CHARMM Force-field.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-25: Modeled deuterium incorporation of fragments in RGS4. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with default parameters (orange). This figure shows the MD simulation re-
sults for PDB:1EZT and CHARMM Force-field. .
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-26: Modeled deuterium incorporation of fragments in RGS8. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with default parameters (orange). This figure shows the MD simulation re-
sults for PDB:2IHD and CHARMM Force-field. .
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-27: Modeled deuterium incorporation of fragments in RGS8. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with default parameters (orange). This figure shows the MD simulation re-
.
sults for PDB:2ODE and CHARMM Force-field.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-28: Modeled deuterium incorporation of fragments in RGS19. The HDX exper-
iment (blue) is shown seven discrete times, alongside each different model
with default parameters (orange). This figure shows the MD simulation re-
.
sults for PDB:1CMZ and CHARMM Force-field.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-29: Modeled deuterium incorporation of fragments in RGS4. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with optimized parameters (orange). This figure shows the MD simulation
.
results for PDB:1AGR and CHARMM Force-field.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-30: Modeled deuterium incorporation of fragments in RGS4. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with optimized parameters (orange). This figure shows the MD simulation
.
results for PDB:1EZT and CHARMM Force-field. .

.

.

.

.

.

.

.

.

.

.

.

.

.

xvii

. 137

. 137

. 138

. 138

. 139

. 139

. 140

. 140

. 141

Figure A-31: Modeled deuterium incorporation of fragments in RGS8. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with optimized parameters (orange). This figure shows the MD simulation
results for PDB:2IHD and CHARMM Force-field. .
.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-32: Modeled deuterium incorporation of fragments in RGS8. The HDX experi-
ment (blue) is shown seven discrete times, alongside each different model
with optimized parameters (orange). This figure shows the MD simulation
.
results for PDB:2ODE and CHARMM Force-field.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-33: Modeled deuterium incorporation of fragments in RGS19. The HDX exper-
iment (blue) is shown seven discrete times, alongside each different model
with optimized parameters (orange). This figure shows the MD simulation
results for PDB:1CMZ and CHARMM Force-field.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-34: Modeled deuterium incorporation of fragments in RGS4. The HDX exper-
iment (blue) is shown twice, alongside new models (M8, M9) with opti-
mized parameters (orange). This figure shows the MD simulation results
for PDB:1AGR and CHARMM Force-field.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-35: Modeled deuterium incorporation of fragments in RGS4. The HDX exper-
iment (blue) is shown twice, alongside new models (M8, M9) with opti-
mized parameters (orange). This figure shows the MD simulation results
for PDB:1EZT and CHARMM Force-field. .
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-36: Modeled deuterium incorporation of fragments in RGS8. The HDX exper-
iment (blue) is shown twice, alongside new models (M8, M9) with opti-
mized parameters (orange). This figure shows the MD simulation results
.
for PDB:2IHD and CHARMM Force- field.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-37: Modeled deuterium incorporation of fragments in RGS8. The HDX exper-
iment (blue) is shown twice, alongside new models (M8, M9) with opti-
mized parameters (orange). This figure shows the MD simulation results
.
for PDB:2ODE and CHARMM Force-field.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-38: Modeled deuterium incorporation of fragments in RGS19. The HDX ex-
periment (blue) is shown twice, alongside new models (M8, M9) with op-
timized parameters (orange). This figure shows the MD simulation results
.
for PDB:1CMZ and CHARMM Force-field.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

. 142

. 143

. 144

. 144

. 144

. 145

. 145

. 145

Figure A-39: Deuterium incorporation is mapped on RGS proteins at t = 1000 min as ob-
served in experiments and as predicted by the models M7, M8, and M9. Data
are presented for the CHARMM-FF simulations of RGS4, RGS8, and RGS19.

146

xviii

Figure A-40: Root mean squared fluctuations (RMSF) per residue across protein se-
quences are shown from 2-μs long MD simulations of (A) RGS4 (PDB:
1AGR, 1EZT), (B) RGS8 (PDB: 2IHD, 2ODE), and (C) RGS19 (PDB: 1CMZ).
Color bars indicate helical regions.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-41: Modeled deuterium incorporation at t = 1000 min at a single-residue resolu-
.

tion (RGS4, CHARMM-FF).

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-42: Modeled deuterium incorporation at t = 1000 min at a single-residue resolu-
.

tion (RGS8, CHARMM-FF).

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-43: Modeled deuterium incorporation at t = 1000 min at a single-residue resolu-
.

tion (RGS4, AMBER-FF).

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-44: Modeled deuterium incorporation at t = 1000 min at a single-residue resolu-
.

tion (RGS8, AMBER-FF).

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-45: Modeled deuterium incorporation at t = 1000 min at a single-residue resolu-
.

tion (RGS19, CHARMM-FF).

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-46: Modeled deuterium incorporation at t = 1000 min at a single-residue resolu-
.

tion (RGS19, AMBER-FF). .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-47: The residues protected by hydrogen-bonds or salt-bridging interactions are
highlighted (panels A and B). The traces for distances between the centers-
of-masses of residue pairs are shown in panel C (S120-Q122) and panel D
.
(E84-R119 and E111-R119).

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-48:

SASA data similar to Fig. A-6 are shown from MD simulations of all RGS
proteins for both force-fields (CHARMM-FF, panel A; AMBER-FF, panel B).
Color and labeling details are similar to Fig. A-6 .
.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-49: Corrected mean residence times for open-states of amide hydrogens are
.

shown. Other details are similar to Fig. A-6.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-50: Residue-residue correlations among open states of all amide-hydrogens
(CHARMM- FF, RGS4 (PDB code 1AGR), model M7). The correlation matrix
is calculated based on the probability that two amide hydrogens simultane-
ously explore open states; C(i, j) = (P(i, j) − P(i)P(j))/(P(i)P(j)(1 −
P(i))(1 − P(j)))0.5 .
.
.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Figure A-51: Data similar to A-50 are shown for RGS8 (CHARMM-FF, RGS8 (PDB code
.

2ODE), model M7).

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

xix

. 147

. 148

. 149

. 150

. 151

. 151

. 152

. 152

. 153

. 154

. 155

. 156

Figure A-52: Probability of a closed to open transition in a given amide vs. simulation
length (μs) is presented based upon Poisson statistics. Data are shown for
PFs = 102, 104, 106, and 1011 with τO = 20 ps and 100 ps.
.

.

.

.

.

.

.

.

.

.

. 157

xx

KEY TO ABBREVIATIONS

2-AG—2-arachadonylglycerol

5HT—5-hydroxytryptamine (serotonin)

6-OHDA—6-hydroxydopamine

AC—adenylyl cyclase

BME—2-mercaptoethanol

BPTI—bovine pancreatic trypsin inhibitor

cAMP—cyclic adenosine monophosphate

DAG—diacylglycerol

DCC—dynamic cross-correlation

DEP—Dishevelled, Egl-10, Pleckstrin)

DI—deuterium incorporation

DMSO—dimethyl sulfoxide

dSPN—direct pathway spiny projection neuron

DSS—4,4-dimethyl-4-silapentane

DTT—dithiothreitol

eCB—endocannabinoid

FCPIA—flow cytometry protein interaction assay

GAIP—G alpha interacting protein

GAP—GTPase-activating protein

GDI—guanine nucleotide dissociation inhibitors

GDP—guanosine diphosphate

xxi

GEF—guanine nucleotide exchange factor

GGL—G gamma-like

GPCR—G-protein coupled receptor

GPe—globus pallidus externus

GPi—globus pallidus internus

GTP—guanosine triphosphate

HDX—hydrogen/deuterium exchange

HEPES—4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid

HSQC—heteronuclear single quantum coherence

IAA—iodoacetamide

IP3—inositol trisphosphate

iSPN—indirect pathway spiny projection neuron

LTD—long-term depression

LTP—long-term potentiation

MC—Monte Carlo

MD—molecular dynamics

MLHD—mean local hydrophobic density

MRT—mean residence time

MS—mass spectrometry

NMR—nuclear magnetic resonance

NPT—number of particles, pressure, temperature

NVT—number of particles, volume, temperature

PDB—protein data bank

xxii

PEPCK—phosphoenolpyruvate carboxykinase

PF—protection factor

PIP2—phosphatidylinositol 4,5-bisphosphate

PLC—phospholipase C

PKA—protein kinase A

RGS—regulator of G-protein signaling

RH—RGS Homology

RMSD—root-mean-square deviation

RMSF—root-Mean-Square Fluctuation

SASA—solvent-accessible surface area

SDS-PAGE—sodium dodecyl sulfate polyacrlyamide gel electrophoresis

SPN—spiny projection neuron

TDZD—thiadiazolidinone

WT—wild type

xxiii

CHAPTER 1:

Introduction

1

Protein dynamics play a major role in protein-ligand interactions.1–3 While dynamics have

a role to play in virtually all molecular interactions,1,2,4–6 this facet of binding may be especially

worth considering in the context of inhibition of challenging targets such as intracellular protein-

protein interactions, where there are generally no binding pockets evolutionarily built for receiv-

ing small-molecule signals.7,8 In this thesis, I discuss the role of protein dynamics in specificity of

an interaction between a series of covalently-acting small molecules and their targets, Regulators

of G-protein Signaling (RGS) proteins. These compounds inhibit by disrupting the interaction be-

tween RGS proteins and their binding partner, the alpha subunit of the heterotrimeric G-protein

(Gα). While this system may be unique in many respects, giving similar consideration to pro-

tein dynamics when probing the mechanism of other biologically active chemicals will provide

valuable insight into the drivers of drug specificity.

Challenging drug targets

Intermolecular interactions are behind the function all biological molecules. Historically,

the field of pharmacology has been built on studying receptors and molecules that can bind to

them.9–11 Receptors are proteins that have evolved to receive chemical signals from the outside of

the cell and relay them to the interior. As such, this obviates the need for the stimulus molecule

to enter the cell itself. In addition, receptors generally have a ready-made binding pocket, making

them convenient to target with exogenous chemicals. These pockets take very specific shapes,

which will only allow certain molecules to bind. Ligand specificity can be compared to key in a

lock: the compound must be just the right shape, or it will not fit in the binding pocket.

Receptors have historically been thought of as readily druggable targets: proteins for

which small molecules that bind will be relatively easy to find. However, only a small portion of

2

all proteins are considered druggable, and only a subset of these may have any medical utility.12

Most existing drugs target enzymes or receptors, which have pockets for binding substrates or

external chemicals respectively. As this low-hanging fruit gets exhausted, however, fewer and

fewer drugs are developed for new targets.12,13 If we can find ways to inhibit unique candidate

proteins, it will increase the potential for continued discovery of small-molecule therapeutics.

One alternative to traditional targeting of cell-surface receptors, while still using small

molecules, is to develop compounds that target intracellular signaling proteins. This is often

more difficult because these proteins have not eveolved to resopnd directly to chemical signals,

so they may lack dedicated binding pockets. Some intracellular proteins, however, do have bind-

ing pockets and are considered receptors. These include nuclear receptors, a receptor family

noted for mediating endocrine signals such as androgens, estrogen, thyroid hormones, and more,

including many yet-undiscovered ligands.14 Still other intracellular proteins, while not binding

external chemical signals as a part of their canonical function, bind to one another to mediate sig-

naling cascades. These may still have cavities that can be exploited in developing inhibitors. One

example is kinases, a protein family for which there there has been a sudden rise in the number of

available inhibitors, most of which act by competitive inhibition of ATP at the nucleotide binding

site.15,16 The ability to pursue intracellular proteins as targets will widen the scope of druggable

candidates, and greatly expand the possibilities for pharmacological modification of disease.

Protein-protein interactions

One key way in which proteins transmit signals is by binding to one another. By mod-

ulating protein-protein interactions (PPIs) among signal transducing proteins, these signals can

be tweaked. However, protein-protein interactions can be difficult to target. This is evidenced

3

by the observation that high throughput screening libraries have had lower success rates in iden-

tifying PPI inhibitors than in identifying inhibitors of traditional receptors.17 One reason is that

the interface between proteins is quite large and often relatively flat. The average interface size

between a small molecule of 500 Da and the binding pocket on its protein target is about 300

Å2,18 an order of magnitude smaller than protein-protein interfaces, which range in area from

1500-3000 Å2.19,20 Despite these interfaces being flat and lacking deep binding pockets, there are

now several examples of molecules that bind to these interfaces.21–23 However, the binding sur-

face between proteins and inhibitors acting at protein-protein interfaces are more spread out,

necessitating larger molecules to maintain the same number of contacts.23,24 One difficulty in the

identification of new inhibitors of PPIs is that most discovery efforts, including high-throughput

screens, use compound libraries that are biased toward smaller molecules similar to those that

bind existing receptors.25,26

Protein function may also be tweaked by binding of a molecule at a location distant from

its protein-protein binding interface, its substrate binding pocket, or orthosteric small-molecule

binding site.2,27,28 This mechanism is dependent on protein allostery, where a conformational

change at one part of the protein may induce conformational changes at an active site on an-

other part of the protein. An argument in favor of the use of allosteric regulators is that they

do not necessarily preclude binding of an endogenous ligand, substrate, or protein binding part-

ner, so they may modulate the intensity of existing signaling when and where it already occurs

rather than induce or block signaling globally.29 There are many examples of allosteric modula-

tors, both for receptors30 and for other protein targets.31,32 This may be important when targeting

protein-protein interactions because any cavity on a protein surface may be sufficient to alter

the protein’s function, even if a binding pocket is not available directly at the protein-protein

4

interaction surface.

Importance of protein dynamics

Our understanding of the shape that a ligand and receptor take on upon binding, and the

fit of the former into the latter, is driven by the field of structural biology. Advances in technol-

ogy using such techniques as X-ray crystallography, cryogenic electron microscopy (cryo-EM),

and nuclear magnetic resonance (NMR) have allowed high resolution determination of three-

dimensional structures of receptor-ligand complexes. Using these protein structural models, a

pharmacologist may understand how shape and molecular interactions drive binding, and may

hypothesize ways the ligand might be altered to improve affinity or specificity.33,34 However, at-

tempts to predict ligand binding using in silico docking techniques may be limited when using a

single static structure, because they do not account for protein flexibility.35–37 The role of dynam-

ics may explain why expected binding results obtained by in silico docking and virtual screening

techniques using a static structure are often far removed from affinities and structures that are

experimentally observed.38 Solution NMR offers an ensemble of structural states, but these still

may not be representative of the wide variety of transient movements a protein may make in

solution. Although they are more computationally intensive, there are now methods for in silico

evaluation of binding that account for protein flexibility.37,39,40 Protein dynamics are well worth

considering, as these might have a strong influence on ligand binding kinetics and affinity.3

There are two models for the role dynamics might play in binding of a molecule in a pocket,

called conformational selection and induced fit.41 In the conformational selection model, a pro-

tein’s conformation in solution, including the shape and properties of its binding pocket, are

undergoing continuous fluctuations. Occasionally, this pocket will be amenable to compound

5

binding. Therefore, a ligand’s on-rate may be influenced by how frequently the protein takes a

certain conformation. Many apo-proteins exhibit conformations in solution similar to conforma-

tions found when ligand is bound. One such protein is adenylate kinase (AdK). By locking the

protein AdK in a conformation similar to it’s ligand-bound state, ligand affinity was drastically

increased, providing evidence for this model.42

In the induced fit model, a compound’s interaction with the protein surface or pocket

may induce a conformational change that allows the compound to bind with higher affinity.1

Proteins in which an induced fit-like mode of ligand binding apply include lid-gated enzymes,

in which the active site is covered by a “lid” that closes around it. It would be difficult for such

a complex to be compatible with a pure conformational selection model, since even if the apo-

protein sampled a conformation similar to the substrate-bound one in solution, the lid would

sterically occlude compound entry.43,44 Induced fit would require an initial binding event in which

the compound first makes contact with the protein before inducing a conformational change.

Structural evidence for such an “encounter complex” has been seen in the phosphoenolpyruvate

carboxykinase (PEPCK) enzyme.45

The actual behavior of a protein-compound interaction may not purely fit with one or the

other of these models. Some would consider the two models less different than they appear, or

in fact just different perspectives of the same mechanism.46 For example, a combination could

exist in which a certain conformation of the apo-protein is required for a compound to bind,

and as the compound enters, it pushes on the residues forming the pocket to cause a secondary

conformational adjustment.47 Importantly, regardless of which model is more relevant to a par-

ticular protein-compound interaction, failure to account for protein dynamics may prevent a full

structural understanding of ligand binding.

6

Some pockets that are potentially druggable may in fact not be present in the crystal struc-

tures, but will open occasionally as the protein moves in solution. These are referred to as tran-

sient pockets. A related concept is that of cryptic pockets, which are pockets that only become

apparent once a compound is bound. As with conformational selection and induced fit, there

may be much overlap between transient and cryptic pockets. Many cryptic pockets may in fact

be more flexible than the surrounding residues and sample open-like states, even in the absence of

ligand.48 It may be helpful to identify transient pockets to develop PPI inhibitors that act directly

at the protein interface.7,8 However, it is also possible to find transient or cryptic pockets that

affect protein activity allosterically. A great example is in the case of K-Ras, which is frequently

mutated in cancers to become constitutively active and makes a very desirable drug target. Many

attempts at finding inhibitors have failed and K-Ras was long considered undruggable.49 How-

ever, a covalent inhibitor was discovered that binds to the cysteine in a G12C K-Ras mutant.50

Analysis of the protein-adduct structure revealed that the compound resided in a cryptic pocket:

one that was not previously apparent in apo structures of K-Ras. Additionally, this inhibitor acts

allosterically, binding adjacent to, as opposed to obscuring, the GTP binding site.50–52 As such,

the role of dynamics in the formation of transient or cryptic pockets has gained recognition for

its importance in discovery of molecules that bind to difficult targets.

Covalent modifiers

Another mechanism of inhibition is covalent modification. Covalent binding of an in-

hibitor may improve potency by reducing the off-rate to a negligible level, and there are several

examples of well-known drugs that act by covalent mechanisms. In fact, some of the most suc-

cessful and widely used drugs are covalent inhibitors.53 Some classic examples include aspirin,

7

which inhibits cyclooxygenase (COX) by acetylation of a serine in the active site,54 and cloprid-

grel, which is converted to an active metabolite by liver enzymes and inhibits P2Y12 adenosine

receptors by thiol-based cysteine modification.55 In more recent years, success in development

of covalent inhibitors has been met in the field of kinase inhibitors, with several covalent drugs

finding FDA approval.56–59

However, concerns about off-target effects and toxicity can be a barrier to development

of covalent modifiers, and these compounds are generally avoided by the pharmaceutical

industry.60,61 One concern is development of an immune response to a covalent drug-protein

adduct.62,63 For example, allergy to penicillin is mediated by IgE and T-cell responses to penicillin-

modified peptides.64 This does not occur in all patients, but there is a fear that such idiosyncratic

reactions may only be discovered once the drug is brought to a larger patient pool. Although

there are some cases where covalent modification is tolerable or even useful in an inhibitor,

finding non-covalent inhibitors is highly desirable to reduce risk and ease drug development.

RGS proteins as therapeutic targets

The G-protein signaling pathway

Perhaps the most pharmacologically important class of cell surface receptors is G-Protein

Coupled Receptors, or GPCRs. GPCRs are not unique to humans; they have been playing a role in

responding to stimuli since the dawn of multicelluar organisms. They are found not only among

animals, but in other kingdoms including fungi65 and possibly in plants.66 The human genome

contains over 800 GPCRs,67 which play a myriad of physiological roles, from the nervous system

(including neurotransmission, sensation of taste, pain, and vision) to cell to cell signals in other

8

physiological systems such as cytokines signaling in the immune system and hormones in the

endocrine system. As such, GPCRs and their associated signaling partners make attractive drug

targets for multiple applications.

The GPCR protein has seven transmembrane domains. The extracellular loops between

these helices form a binding pocket for a ligand. Generally ligand binding is a reversible, non-

covalent interaction, and the receptor may become active when a ligand is bound. Much cell-cell

signaling throughout the body, and particularly in the CNS, is mediated by such reversible ligand-

receptor interactions with GPCRs. These include glutamate with mGluRs,68,69 GABA at GABAB

receptors,70,71 monoamines at their respective receptors receptors,72–74 and more. However, di-

verse variations also exist. For example, retinal is a ligand for the opsin family of GPCRs. It can

remain bound to the receptor, and change conformation to activate the receptor in the presence

of light. Another example is protease-activated receptors, a family with an N-terminal tail that

may be cleaved to form a tethered ligand: a ligand that is a part of the receptor rather than an

external signal.75

The intracellular loops form a binding site for an intracellular effector: the heterotrimeric

G-protein. The G-protein is comprised of the alpha subunit (denoted Gα), which binds a gua-

nine nucleotide, and the beta and gamma subunits, which form an obligate dimer (Gβγ). The

C-terminal tail of the of a GDP-bound alpha subunit may dock to the intracellular side of the

receptor.76 Once an agonist binds to the receptor, a conformational change in the receptor allows

the GDP to dissociate from the Gα subunit and GTP to bind in its place. The GTP-bound G-

protein is now said to be in its active conformation. It dissociates from the receptor and the alpha

subunit dissociates from the beta-gamma dimer, allowing each to initiate signaling by binding to

downstream receptors.

9

Figure 1-1: Activation of G-protein signaling upon agonist binding at GPCR.

Gα proteins are divided into four main categories: the Gs, the Gi/o the Gq, and the G12 fam-

ilies. Gs proteins have adenylyl cyclase (AC) as their effector, binding to AC and stimulating pro-

duction of cyclic AMP, or cAMP. cAMP is a necessary signaling molecule for a signaling cascade

that starts with activation of Protein Kinase A (PKA), which is also known as cAMP-dependent

protein kinase. Thus, in the extracellular presence of an agonist for Gs coupled GPCRs, Gs can

elicit an increase in intracelluar cAMP. Gi/o proteins have the opposite effect: when activated,

they reduce intracellular cAMP. This effect may be mediated by the Gβγ subunit dimer rather

than the Gi/o alpha subunit itself. In fact, Gβγ is responsible for much of the G-protein signaling.

In the case of Gi/o, the “active” Gα could exert its effects merely by releasing Gβγ.77,78 It should

be noted that Gβγ is released during Gs signaling as well, but any inhibitory effects of the Gβγ

dimer, if present, are overshadowed by the activation induced by Gs. Gq and G12 families act

through cAMP-independent mechanisms. Gq, when active, is known for induction of calcium

release from the endoplasmic reticulum (ER). Gq binds to Phospholipase C (PLC), which cleaves

the phospholipid phosphatidylinositol 4,5-bisphosphate (PIP2) into diacylglycerol (DAG) and in-

ositol trisphosphate (IP3). IP3 binds to the IP3 receptor on the ER, causing calcium release into the

cytosol.79 One well-known example of Gq mediated signaling is endocrine regulation of smooth

muscle, where calcium release is necessary for contraction of actin and myosin.80 It should also

10

Figure 1-2: Activation of different signaling pathways is mediated by different G-protein sub-
types.

be noted that DAG, the other product of phospholipid cleavage by PLC, can also go on to initiate

signaling of its own. Finally, active G12 family proteins activate RhoGEFs, nucleotide exchange

factors for small Rho family GTPase proteins (as opposed to heterotrimeric G-proteins). These

small G-proteins go on to induce phosphorylation cascades and cause changes in cytoskeleton reg-

ulation and gene transcription.81 Although other receptor types exist, heterotrimeric G-proteins

and their receptors are a signaling powerhouse capable of initiating a vast array of cellular func-

tions.

The Regulators of G-protein Signaling (RGS) proteins are negative modulators of the G-

protein pathway. They bind to active Gα subunit, increasing the rate of GTP hydrolysis. This

is known as GTPase-Activating Protein, or GAP, activity. Although the Gα subunit has some

intrinsic ability to mediate hydrolysis, it is greatly accelerated when bound to an RGS protein.

By returning the alpha subunit to its GDP-bound state, it allows reassociation of the complex

11

Figure 1-3: RGS are GTPase-Activating Proteins (GAPs). They terminate G-protein signaling by
catalyzing hydrolysis of GTP on Gα.

between the alpha subunit, the beta-gamma subunits, and the receptor. G-protein signaling may

be thought of as a cycle, in which the receptor acts as a Guanine nucleotide Exchange Factor

(GEF), activating the G protein, and the RGS protein acts as a GAP, deactivating the G-protein

(Fig. 1-3). Many drugs exist that manipulate the G-protein cycle from the GPCR side, by positively

or negatively altering G-protein activation, but compounds that manipulate G-protein signaling

by altering GAP activity remain largely unexplored.

RGS protein diversity

The GAP activity of RGS proteins is carried out by the RGS homology (RH) domain, also

referred to as the RGS domain or RGS box. The RGS domain is comprised of about 130 conserved

amino acids, which form nine helices.82,83 While more proteins with RGS domains may exist, there

are twenty that are considered canonical RGS proteins. These RGS isoforms are numbered 1-21,

with no RGS15. These in turn are divided into four families: R4, R7, R12, and RZ. R4 is the largest

12

family, encompassing RGS isoforms 1-5, 8, 13, 16, 18, and 21. These proteins are relatively small,

with no additional domains other beyond the RGS domain. However, the N-terminal tail, while

not large, may play a role in targeting the RGS protein to certain receptors.84 The R4 family act

as GAPs for most Gi/o and Gq proteins. One notable exception is RGS2, which binds only to Gq

and not Gi/o.85

The R7 family consists of RGS6, 7, 9, and 11. These proteins are unique in that they con-

tain two more domains, DEP and GGL, that confer additional functions beyond GAP activity.

The DEP domain (Dishevelled, Egl-10, and Pleckstrin; named for the proteins in which it was

first identified) is a small domain that may help target the whole protein to specific subcellular

locations by binding to membrane anchor proteins.86–88 GGL (G-gamma-like) is a domain that, as

its name implies, shares sequence identity and structural features with Gγ. This domain binds to

Gβ5,89 and may help increase GAP activity by colocalizing the RGS protein with Gαo.90,91

The R12 family consists of RGS10, RGS12, and RGS14. These proteins are grouped

based on their sequence identity, but vary considerably in their number of domains. RGS10 is

relatively short, without additional well-characterized domains. RGS12 and RGS14 both have

tandem repeats of Ras-binding domains, allowing them to scaffold with small GTPase and MAP

kinase proteins, and a G-protein Regulatory (GPR) motif, also known as a GoLoco motif, which

bind to G alphai/o subunits and act as guanine nucleotide dissociation inhibitors (GDIs).92–94

RGS12 is the longest full-length RGS protein, at 1387 amino acids, containing in addition a PTB

(phosphotyrosine-binding) and PDZ domain, which assists in protein localization.93,95

The RZ family contains RGS17, RGS19, and RGS20. RGS19 is also known as GAIP (G alpha

interacting protein). Like most other RGS proteins, they are capable of GAP activity toward Gi

and Go, and are, in many respects, similar to the R4 family. They are relatively small, having no

13

additional domains other than a cysteine-rich region on the C-terminal tail. However, they are

unique in their ability to act as GAPs for Gαz, a G protein that is a part of the Gi/o family, but not

affected by other RGS proteins. This aspect gives the family its name.

In addition to differences in molecular function, RGS isoforms also differ in their tissue

distribution,96,97 allowing each isoform to play a unique physiological role. A specific RGS in-

hibitor will act only where it’s target isoform is expressed, improving tissue specificity beyond

that which could be achieved by a GPCR agonist distributed throughout the body. In addition,

rather than inducing signaling at the GPCR, use of an RGS inhibitor may prolong endogenous

signaling where it is already occurring, further reducing off-target effects.

Physiology of RGS proteins in disease

With RGS proteins playing such diverse roles, it is not surprising that they are involved

in the pathogenesis or modulation of many disease states. Since many RGS proteins are highly

expressed in the brain,96 there is high potential for modulation of RGS proteins in treatment of

CNS disorders. For example, RGS4 has been implicated in seizures. Endogenous adenosine has

a protective effect on kainate induced seizures, which may be reduced by negative regulation by

scaffolding between the A1 receptor, neurabin, and RGS4. Both neurabin knockout and RGS4

inhibition are able to reduce kainate-induced seizures.98 RGS4 has also been implicated in reward

and addiction. Male RGS4 knockout mice have reduced cocaine induced reward effects compared

to WT.99 RGS proteins may also have use in treatment of depression. Mice expressing RGS insensi-

tive Gai show an antidepression-like phenotype. This is likely mediated by the 5HT-1A receptor,

as the effect is blocked by a 5HT1A antagonist.100 RGS19 is capable of attenuating 5HT-1A re-

ceptor signaling, indicating it may be a potential target for treatment of depression.101 RGS19

14

attenuates mu-opioid signaling, indicating it may play a role in pain modulation as well.102

RGS17 is a potential target in treatment of certain cancer types.103 One example is in lung

cancer: lung cancer susceptibility has been associated with mutations in the first intron of RGS17.

RGS17 is heavily upregulated in as many as 80% of lung cancers, while RGS17 knockdown reduced

the rate of proliferation in a human lung tumor cell line.104 Similarly, RGS17 is also upregulated

in prostate cancers,105 which could be another indication for RGS17 inhibition.

One of the most promising applications for targeting an RGS protein is RGS4 in Parkin-

son’s disease. RGS4 is very highly expressed in the striatum,97 which is a critical part of the

motor control pathway. Motor signals originate in the cortex and are modulated by parts of the

the basal ganglia, specifically the substantia nigra, the striatum, and the globus pallidus. The

striatum receives glutamatergic input from the cortex and dopaminergic input from the substan-

tia nigra. There are two types of spiny projection neurons (SPNs) in the striatum, those that form

the direct pathway (dSPN) and those forming the indirect pathway (iSPN). The dSPNs express

excitatory Golf coupled (D1-type) dopamine receptors, and project to the internal globus pallidus

(GPi). Meanwhile, the iSPNs express inhibitory Gi coupled (D2-type) dopamine receptors, and

project to the external globus pallidus (GPe), which in turn projects to the GPi (Fig. 1-4).

In

the Parkinson’s disease state, dopamine-producing neurons in the substantia nigra die, and total

dopamine input is reduced. This causes disinhibition of the indirect pathway and reduced ex-

citation of direct pathway. This imbalance between direct and indirect pathways is thought to

underlie the motor deficits observed in Parkinson’s disease.106

Synaptic plasticity may play a significant role in regulation of motor control. Long-term

depression (LTD) is a form of synaptic plasticity mediated by the release of endocannabinoids

(eCBs) such as 2-arachadonylglycerol (2-AG) and anandamide. CB1 cannabinoid receptors are

15

Figure 1-4: The circuitry of the motor pathway. RGS4 is expressed in the striatum. In the Parkin-
son’s disease state, dopaminergic input from the substantia nigra to the striatum is lost.

expressed in glutamatergic nerve terminals projecting to the striatum,107 and anandamide is re-

leased in the striatum upon activation of D2-like but not D1-like receptors.108 In the Parkinson’s

disease state, lack of dopamine may reduce endocannabinoid release from iSPNs, thus disinhibit-

ing the indirect pathway and contributing to the imbalance between direct and indirect pathways.

Lerner and Kreitzer (2012) proposed a model in which RGS4 acts as a link between D2

receptor activity and synaptic plasticity in iSPNs.109 In this model, endocannabinoid release is

stimulated in response to glutamatergic signaling through Gαq-coupled mGluR receptors. RGS4

acts as a GAP for Gαq, negatively modulating endocannabinoid release. RGS4 activity is en-

hanced upon phosphorylation by PKA.110 Dopamine D2 receptor activity inhibits cAMP produc-

tion, thus reducing PKA activity and RGS4 phosphorylation (Fig. 1-5A). In a Parkinson’s disease

state, lack of dopamine may allow unchecked cAMP and PKA activity, leading to excessive activ-

ity of RGS4 and reduced endocannabinoid-mediated long-term depression. Indeed, RGS4-/- mice

16

have increased LTD over wild-type mice, even in the presence of a D2 antagonist. In addition,

in a 6-Hydroxydopamine (6-OHDA) lesion model of Parkinson’s disease, RGS4-/- mice were less

susceptible than wild-type mice to Parkinson’s-like motor deficits.109

A major problem in treatment of Parkinson’s disease is dyskinesia. L-DOPA, a precursor

to dopamine, is a standard way of treating motor symptoms in Parkinson’s patients. However, a

major problem with L-DOPA is that with continued use, its efficacy wanes and dosage is increased,

causing dyskinesia. There is a different form of synaptic plasticity that may also play a role in this

L-DOPA-induced dyskinesia. While dopamine (and L-DOPA) activity causes LTD in iSPNS, it can

induce long-term potentiation (LTP) of dSPN activity. This effect is dependent on D1 receptors;

application of D1 antagonists blocks LTP.111 Shen et al. (2015) showed that application of an RGS4

inhibitor can induce LTD in dSPNs. Therefore, RGS4 inhibition will not contribute to, and may

functionally oppose, the dyskinetic effect of D1 receptor-dependent LTP (Fig. 1-5B),112 which

would otherwise be induced upon L-DOPA administration.

RGS4 inhibitors may shine as a combination therapy with L- DOPA. In the indirect path-

way, they will complement the effect of L-DOPA in dampening excessive activity. In the direct

pathway, however, they may counteract the long-term potentiation responsible for L-DOPA in-

duced dyskinesia.

RGS inhibitors

Recent discovery efforts

In light of this developing rationale for targeting RGS proteins, there have been efforts to

develop inhibitors. Roman et al., 2007 performed a high throughput screening campaign run to

17

Figure 1-5: The role of RGS4 in response to dopamine signaling in the indirect and direct pathway
spiny projection neurons of the striatum.

18

discover inhibitors of the RGS/Gα interaction. This led to the discovery of CCG-4986, the first re-

ported RGS inhibitor, which has specificity for RGS4 over RGS8.113 Further investigation revealed

that this compound acted by covalent modification of cysteine residues. Interestingly, some cys-

teines are not located near the RGS/Gα interface, but binding at these cysteines is still sufficient

for inhibition of binding between RGS and Gα. This indicates that these compounds inhibit the

protein-protein interaction allosterically.31 A later screen, multiplexed to determine effects on

different RGS proteins, led to the discovery of CCG-50014, the lead compound of the thiadiazo-

lidinones (TDZDs).114 This compound also acts by covalent modification of cysteine residues, and

like CCG-4986, is specific for RGS4. To date, no noncovalent RGS inhibitors have been discovered.

Thiadiazolidinone characterization

RGS inhibitor CCG-50014 has been fairly well characterized in its action against RGS8.

Blazer et al., 2011, showed that at least one cysteine is necessary for the compound to inhibit the

RGS-Gα interaction, and an adduct between the compound and the protein can be detected by

mass spectrometry. Thus, it is well established that the thiadiazolidinone inhibition is mediated

by covalent modification.115 Interestingly, it was also shown that general cysteine modifiers such

as iodoacetamide and n-ethyl maleimide act far less potently on RGS proteins than the thiadiazo-

lidinones. Further, on a cysteine-dependent protease, general cysteine alkylators acted far more

potently than CCG-50014.115 These results suggest that the interaction between CCG-50014 and

RGS proteins is unique, and lends strength to the concept that a cysteine modifier is capable of

acting specifically, without indiscriminate adducts at other proteins.

Because of the reduced potential for adverse effects in with a highly isoform-specific in-

hibitor, a further effort has been made to develop thiadiazolidinones with improved specificity

19

Figure 1-6: Thiadiazolidinones CCG-50014, the lead compound, and CCG-203769, an analog with
improved specificity for RGS4.

for RGS4 over other isoforms. This resulted in the discovery of another thadiazolidinone, CCG-

203769, which has aliphatic chains in place of the aromatic rings found in CCG-50014. It is more

RGS4 selective than CCG-50014 because it is less potent against RGS8.116

Because CCG-203769 is more selective for RGS4 than other TDZDs, it may have use as

a treatment for disease states in which reduction of RGS4 activity is desirable, most notably,

Parkinson’s disease. Blazer et al., 2015 demonstrated that this compound has in vivo effects on

motor coordination. In this study, Parkinson’s-like bradykinesia (slowness of movement) was

induced using the D2-type receptor antagonist raclopride. Two tests were used to analyze reversal

of this impairment by CCG-203769: the drag test, which counts steps taken by a mouse as drawn

backward by the tail, and the bar test, in which mice were evaluated for the latency in removing

their forepaws from an elevated block. In each of these, a dose of as low as 0.1 mg/kg CCG-203769

was sufficient to induce reversal of the bradykinetic effect of raclopride.117

20

Figure 1-7: Locations of cysteines in RGS protein. Gα is shown in gray spheres, RGS in shown in
light blue. Cysteines 71 and 132 in RGS4 (red) are not conserved among RGS proteins. Cysteine
148 in RGS4 (blue) is shared by RGS8 and RGS4. Cysteine 95 in RGS4 (green) is the best conserved
cysteine among RGS proteins, found in all isoforms except RGS6 and RGS7.

21

Contribution of this work

While several RGS inhibitors have now been discovered, all are most potent for either

RGS4 or RGS1.118 This is not too surprising, given that these isoforms contain relatively high

numbers of cysteines in the RGS domain compared to other isoforms, with four and three cys-

teines respectively. In light of this, it is likely that the RGS4 or RGS1 selectivity is driven primarily

by cysteine complement. Interestingly, however, TDZDs can also act on many RGS proteins, even

those with only one cysteine. For example, RGS19 contains one cysteine, but CCG-50014 can still

inhibit its interaction with Gα with an IC50 of 120 nM.115 This suggests that the factors influenc-

ing TDZD selectivity are more complex than the simple quantity of cysteines in the RGS domain.

The cysteine found in RGS19 is well conserved, found in 18 out of the 20 canonical RGS proteins.

From its position on the α4 helix, it angles toward the center of the α4-α7 helical bundle, so it is

buried rather than at the protein surface (green spheres in Fig. 1-7). In order for a covalent in-

hibitor to access this cysteine, it may be necessary for the protein to undergo a motion exposing

this cysteine to the solvent. This implies that protein dynamics is an important yet unexplored

factor in RGS inhibitor specificity.

This work aims to develop a better understanding of other factors that drive TDZD selec-

tivity, especially at a structural and dynamic level, which will enable the development of nonco-

valent inhibitors.

22

CHAPTER 2:

Differential Protein Dynamics of Regulators of G-Protein Signaling: Role in Specificity

of Small-Molecule Inhibitors

Reprinted with permission from J. Am. Chem. Soc. 2018, 140, 3454-3460

Copyright 2018 American Chemical Society

Vincent S Shaw*, Hossein Mohammadiarani*, Harish Vashisth, Richard R Neubig

*Co-first authors
V.S. expressed protein and performed HDX-MS and FCPIA. H.M. performed MD simulations and
calculated RMSF, RMSD, and SASA.

23

Introduction

Protein-protein interactions (PPIs) remain a poorly tapped pool of potential targets for

small-molecule inhibitors. Targeting PPIs has been challenging because many protein-protein

interfaces are flat and lack a dedicated small-molecule binding pocket.23,119,120 However, it may be

possible to interrupt PPIs by binding to transiently exposed pockets,121,122 either at the protein-

protein interface7 or at allosteric sites.32,123 Targeting of allosteric sites, as they are less evolution-

arily conserved, may confer better specificity than directly targeting interfaces.124 In addition,

there may be variation in dynamic exposure of allosteric pockets among members of a protein

family. Such differences in protein dynamics could drive inhibitor specificity.125

G-protein signaling is critical in pharmacology. Approximately thirty percent of marketed

drugs target GPCRs and related pathways.12 Regulators of G-protein Signaling (RGS) proteins

control GPCR signaling by binding to active, GTP-bound Gα subunits, thereby accelerating GTP

hydrolysis. This terminates G-protein signaling. Inhibition of an RGS protein can amplify signal-

ing through a GPCR. We previously identified thiadiazolidinone (TDZD) inhibitors of the RGS-Gα

interaction in a high-throughput screen.116 They allosterically inhibit RGS proteins by covalent

modification of cysteine residues at sites distant from the RGS-Gα interface. The TDZD inhibitor

CCG-50014 is most potent against RGS4, followed by RGS19 and distantly by RGS8.115 RGS4 in-

hibitors may be valuable as therapeutics for Parkinson’s disease. RGS4 is highly expressed in

the striatum,96,97 where it regulates synaptic plasticity in response to dopamine signaling.109,112

A TDZD inhibitor with enhanced specificity for RGS4, CCG-203769, reduces bradykinesia in a

raclopride model of certain Parkinson’s-like motor deficits in mice.117

The RGS domain, which is responsible for the GTP-ase accelerating activity of RGS pro-

teins, is present in 20 human RGS proteins as well as some proteins with a similar fold that lack

24

Gα binding properties.126 The RGS domain is a 120-amino acid domain consisting of nine α-helices

(Fig. 2-2A).126,127 Differences in TDZD potency may be due to different locations or numbers of

cysteines among RGS isoforms or to differential transient cysteine exposure. RGS4, RGS8, and

RGS19 all share an α4 helix cysteine, while RGS4 and RGS8 share one on the α6-α7 interheli-

cal loop (Fig. 2-2A). Notably, these cysteines are buried beneath the protein surface in crystal

structures.83,128 Therefore, it may be necessary for dynamic pockets to open to expose these cys-

teines for TDZD interaction. Understanding dynamic pockets will be beneficial, as such a pocket

may be exploited in rational design of novel non-covalent inhibitors using a docking-based vir-

tual screen. We previously showed that the α5-α6 helical pair is flexible using enhanced sampling

MD simulations.129 Covalent modification by TDZD inhibitors could lock the α5-α6 interhelical

loop in a position that prevents the RGS interaction with Gα proteins. We hypothesize that dif-

ferential transient exposure of buried cysteine residues drives TDZD selectivity. Here, we used

hydrogen/deuterium exchange with mass spectroscopy (HDX-MS) and long time-scale classical

unbiased molecular dynamics (MD) studies to examine differences in dynamics between RGS4,

RGS8, and RGS19. These RGS protein isoforms represent a range of potencies of TDZD inhibitors

(RGS4 > RGS19 > RGS8). HDX-MS and MD studies make a powerful combined experimental and

computational approach for evaluating protein dynamics.130,131 These revealed a dual role of pro-

tein dynamics and cysteine complement in the selectivity of TDZDs against RGS proteins.

Materials and Methods

Protein expression and purification

N-terminally truncated (Δ51) rat RGS4 with 6xHis tag in pET23d vector, RGS homology

25

domain of human RGS8 (42-173) with 6xHis tag in pQE80 vector, RGS homology domain of human

RGS19 (89-206) with 6xHis tag in pET15b vector, and Gαo with 6xHis tag in pQE-6 vector132 were

individually transformed into BL21(DE3) E. coli. Single-cysteine mutant RGS protein constructs

were generated by mutating cysteines to alanines using QuikChange II mutagenesis kit (Agilent).

Protein expression was induced by addition of 200 µM IPTG. Expression was carried out overnight

at 25℃ and cells were harvested by centrifugation. Pellets were resuspended in 50 mM HEPES

100 mM NaCl pH7.4 and lysed by sonication. Lysate was centrifuged, supernatants were applied

to nickel affinity column, and protein was eluted with 300 mM imidazole. RGS4 was further

purified by SP sepharose column. Column was equilibrated with 50 mM Na Acetate, 40 mM

NaCl, 1 mM DTT, 1 mM EDTA, and 1mM EGTA (pH 5.5) and protein eluted by linear gradient

to buffer including 1M NaCl. RGS8 and RGS19 were purified by Q sepharose column. Column

was equilibrated with 20 mM NaCl, 50 mM Tris, and 1 mM DTT (pH 8.0), and protein eluted with

linear gradient to buffer including 1M NaCl.

Flow cytometry protein interaction assay

FCPIA was performed as previously described.133 Briefly, RGS proteins were biotinylated

and bound to xMAP LumAvidin microspheres (Luminex). Gαo protein labeled with AlexaFluor-

532 was exposed to beads in presence of GDP and aluminum fluoride to stabilize the transition

state. Bead fluorescence was read using Luminex 200 flow cytometer.

Hydrogen/deuterium exchange

HDX-MS was performed as described in Chodavarapu et al., 2015.134 In principle, after ex-

posure to D2O for different times, the exchange of amide hydrogens for deuterium was quenched

by acidification then samples were digested with pepsin and separated by LC-MS for analysis

26

of deuterium content. Specifically, proteins were incubated on ice for desired time in 90% D2O

containing 100 mM NaCl and 5 mM HEPES, pH 7.4. All columns and valves were kept on ice to

reduce back exchange. Exchange was quenched by 1:1 addition of ice cold 1% (v/v) formic acid

in H2O, bringing the pH to 2.5. 100 µl samples were immediately loaded at 0.1 ml/min, using

external pump (LC-20AD; Shimadzu), to an Enzymate pepsin column (2.1 x 30 mm, Waters) equi-

librated with cold 0.1% formic acid in H2O. After 1 min, the pump was stopped and proteins were

digested on-column for 1 min (See pepsin cleavage pattern in Fig. 2-6). Following digestion, the

resulting peptides were eluted at 0.5 ml/min onto an Xbridge BEH C18 VanGuard trap column

(2.1 x 5 mm, Waters). The peptides were then eluted from the trap column by valve switching

of liquid flow, using a Waters 2777c autosampler, onto an Ascentis Express Peptide ES-C18 col-

umn (2.1 x 50 mm, Supelco). Flow through the 2777c autosampler valve and the Peptide ES-C18

column was controlled by a Waters Acquity Binary Solvent Manager. The peptides were initially

washed for 1 min at 0.3 ml/min with 99% solvent A (0.1% formic acid in H2O) and 1% solvent B

(acetonitrile). Peptides were then separated by elution with a gradient from 1% B to 30% B at 3

min, then to 99 % B at 6 min and held at 99% B for 1 min. Eluted peptides were analyzed using a

Xevo G2-XS QToF mass spectrometer (Waters) by electrospray ionization operating in positive-

ion mode. Fragments observed following cleavage of RGS proteins by pepsin are shown in Fig.

2-1. Mass spectra were acquired in continuum mode over m/z 100-2000. Data were analyzed

using Microsoft Excel, HX Express,135 and GraphPad Prism software. Deuterium incorporation

was determined by the increase in centroid mass of each peptide’s isotope distribution compared

to undeuterated control.

27

Figure 2-1: Alignment of fragments observed by mass spectrometry following cleavage of RGS
proteins by pepsin. Horizontal bars indicate length and position of observed fragments. The two
N-terminal residues of each fragment were excluded from analysis due to rapid back-exchange.
Vertical gray boxes indicate approximate positions of helices within the RGS domain.

28

System setup and simulation details

Molecular dynamics (MD) simulations were carried out by the Vashisth Lab at UNH. Tra-

jectory calculations and their analysis were done using NAMD/VMD software suite136,137 with

the CHARMM force-field and CMAP correction.138,139 The initial coordinates for RGS4, RGS8,

and RGS19, respectively, were taken from the protein data bank entries 1AGR, 2ODE, and 1CMZ.

Each protein was initially modeled using the psfgen tool in VMD and further solvated in a sim-

ulation box (~65Å × ~70Å × ~65Å) of TIP3P water molecules and charge-neutralized with NaCl.

The final solvated and ionized simulation domains contained 28160 atoms (RGS4), 30731 atoms

(RGS8), and 29560 atoms (RGS19), respectively. The box volume was then optimized in the NPT

ensemble by initially applying 500 cycles of a conjugate-gradient minimization scheme followed

by a short 40 ps MD run with a 2 fs time step in which the temperature was controlled using the

Langevin thermostat and the pressure was controlled by the Nose-Hoover barostat. We carried

out all simulations using periodic boundary conditions. These briefly equilibrated systems of all

RGS proteins were further subjected to two independent MD simulation sets in the NVT ensem-

ble. The first set (Set 1 in Table 2-1) of simulations were 2 µs-long for each RGS protein, and the

second set of simulations (Set 2 in Table 2-1) were 3 µs-long for each RGS protein. Results from

Set 1 are discussed in Fig. 2-5 and 2-7, and from Set 2 are shown in Fig. 2-7, 2-6, and 2-9. Compu-

tations were performed on Trillian, a Cray XE6m-200 Supercomputer and using Premise, a UNH

in-house GPU based cluster. In addition, this work used the Extreme Science and Engineering

Discovery Environment (XSEDE).140

RMSD, RMSF, and SASA Measurements

We carried out the analyses on per-residue root-mean-squared-fluctuation (RMSF) and

29

root-mean-squared-deviation (RMSD), as reported in Fig. 2-4 and 2-5, by aligning each frame

of MD trajectories to the initial frame based upon all Cα atoms of each protein. The solvent-

accessible surface area (SASA) of sulfur atoms in cysteines were calculated using a probe radius

of 1.4 Å.

Results

Previous work has demonstrated a role for the number and position of cysteine residues

in the potency of RGS inhibitors.31 To eliminate this confounding variable and allow better assess-

ment of the role of protein dynamics, the potency of CCG-50014 was compared among RGS19

and mutants of RGS4 and RGS8 containing only the shared α4 cysteine. These mutants are termed

RGS4 95C and RGS8 107C respectively. While removal of additional cysteines reduced potency in

both RGS4 and RGS8, dramatic differences in TDZD potency still exist among single-cysteine pro-

teins. RGS19 was most potently inhibited with an IC50 of 1.1 μM, while RGS4 95C was inhibited

with an IC50 of 8.5 μM, and RGS8 107C had an IC50 of >100 μM (Fig. 2-2B).

To compare solvent exposure kinetics on the α4 helix, we performed HDX-MS on RGS4,

RGS8, and RGS19 apo-proteins. A map of pepsin cleavage fragments observed in each protein is

shown in Fig. 2-1. Consistent with the higher potency of inhibition by the TDZD, the cysteine-

containing fragment from α4 (residues 92-97) in RGS4 shows significantly higher exchange than

that from RGS8 (residues 86-91). After a 1000-minute incubation in D2O, the 92-97 fragment of

RGS4 had 35% deuterium incorporation (DI), while the analogous fragment in RGS8 had only 8%

DI. Further strengthening the correlation of dynamics with selectivity, RGS19 had much faster

exchange than RGS4 or RGS8 in the α4 helix. It reached 48% DI by only 100 minutes, while RGS4

and RGS8 had 9% and 1% respectively (Fig. 2-3A). A similar trend was observed in the α5 helix.

30

Figure 2-2: (A) Locations of cysteines in RGS4, RGS8, and RGS19.
(B) Potency of CCG-50014
against RGS19, which has only one cysteine, and mutant RGS4 and RGS8 containing only the
shared α4 helix cysteine. n=3.

RGS8 had the least exchange after 1000 minutes (24% DI), followed by RGS4 and RGS19 (38%

and 49% DI, respectively, Fig. 2-3B). One pattern consistent among all three isoforms is high

exchange in the α5-α6 interhelical loop, indicating that RGS proteins are flexible in this region.

Those fragments in all three proteins exceeded 50% DI by 100 minutes (Fig. 2-3C). This was not

surprising, as the α5-α6 loop is the longest unstructured region within the RGS domain. In the α6

helix, RGS19 again had higher exchange than RGS8 and RGS4. RGS8 was particularly protected in

the residue 126-136 fragment, reaching only 7% DI after 1000 minutes. However, higher exchange

was observed in the residue 130-140 fragment of RGS8, likely because this fragment also contains

residues that are a part of the α6-α7 loop (Fig. 2-3D). A similar effect was seen in RGS4 near the α7

helix, in which a fragment wholly within α7 (residues 150-159) had much slower exchange than

a fragment partially overlapping the α6-α7 loop (residues 143-151) (Fig. 2-3E).

According to these results, RGS8 had low deuterium exchange relative to other RGS pro-

teins throughout the helices surrounding its cysteines. This is indicative of rigidity of these he-

31

Figure 2-3: (A-E) Kinetics of deuterium exchange in selected protein fragments from (A) α4, (B)
α5, (C) α5-α6 interhelical region, (D) α6 and (E) α7. Sequences of observed fragments are aligned
with residue numbers of each fragment indicated. Cysteine locations are marked in red. n=3.

32

Figure 2-4: (A) Global kinetics of deuterium exchange. Deuterium incorporation (DI) is expressed
as a percent of exchangeable amide hydrogen positions. Where fragments overlap, data is dis-
played as average DI of observed fragments.
(B) Degree of DI at 300 minutes in 90% D2O is
mapped onto protein structure of RGS4, RGS8, and RGS19. n=3.

lices in RGS8, which likely prevents exposure of cysteines to solvent. This observation also could

explain the low potency of TDZDs against RGS8 relative to other RGS isoforms. The α6 helix

of RGS4 has more deuterium exchange than the α4, α5, and α7 helices (Fig. 2-4A and B). Rapid

exchange in the α6 helix may be due to movement away from neighboring helices or unfolding

of the helix itself. Such a movement could increase solvent exposure of the otherwise buried

cysteine 148 on the α6-α7 loop. This would allow access by TDZD inhibitors. Because the higher

exchange on α6 compared to other nearby helices is unique to RGS4, this potentially contributes

to the increased potency of TDZDs against wild type RGS4 versus RGS8. In the α4, α5, and α6

helices, RGS19 shows higher deuterium exchange than RGS4 or RGS8, indicating that RGS19 is

highly dynamic. For example, in a fragment of the α5 helix, RGS19 had 51% DI after 30 minutes,

while similar fragments in RGS4 and RGS8 had 15% and 17% incorporation, respectively (Fig. 2-

33

Protein initial coordinates
RGS4
RGS8
RGS19

PDB: 1AGR
PDB: 2ODE
PDB: 1CMZ

# of atoms

28160
30731
29560

trajectory length
set1
2 μs
2 μs
2 μs

set 2
3 μs
3 μs
3 μs

Table 2-1: Summary of MD simulations.

3B). This fits with functional data showing that RGS19 is more potently inhibited by CCG-50014

than single-cysteine RGS4 and RGS8 (2-2B). Although RGS19 lacks cysteines on the α6 helix and

α6-α7 loop which may contribute to potency of inhibition of RGS4 by TDZDs (Cys 132 and Cys

148 in RGS4), it has the highest potency of inhibition among single-cysteine RGS proteins. This

may be due to a pronounced movement of the α4, α5, and α7 helices, allowing TDZDs to access

RGS19’s cysteine on the α4 helix.

To probe the molecular details of dynamic motions in RGS4, RGS8, and RGS19 that un-

derlie the flexibility differences observed in HDX-MS as well as to evaluate possible routes of

access to cysteines by TDZDs, we performed long time-scale classical MD simulations in explicit-

solvent. Our previous short time-scale classical MD simulations did not show any major confor-

mational changes; but enhanced sampling simulations did show changes.129 Here, we conducted

microsecond time-scale classical MD simulations through which the flexibility in key helices be-

came apparent.

The first set of simulations that were 2 µs long (set 1 in Table 2-1) showed regions of

pronounced movement in all three proteins. RGS4 showed unique motions within the α6 helix

(Fig. 2-5A), while in RGS8 and RGS19, movement was primarily within the α6-α7 interhelical loop

(Fig. 2-5B and C). A second independent set of simulations that were 3 µs long (set 2 in Table 2-1)

showed the largest movement in RGS19, again particularly prominent in the α6-α7 interhelical

loop, with the α6 helix and α5-α6 interhelical loop also relatively flexible (Fig 2-6). However, RGS4

34

Figure 2-5: Root mean squared fluctuations (RMSF) per residue during 2 μs MD simulations of
(A) RGS4 (PDB: 1AGR), (B) RGS8 (PDB: 2ODE), and (C) RGS19 (PDB: 1CMZ). The RMSF trends
for each protein for the simulation set 2 are shown in Fig. 2-6. Gray bars indicate helical regions.

Figure 2-6: Root mean squared fluctuations across protein sequence during 3 μs MD simulations
of (A) RGS4 (PDB: 1AGR), (B) RGS8 (PDB: 2ODE), and (C) RGS19 (PDB: 1CMZ). Gray bars indicate
helical regions.

35

Figure 2-7: Solvent-accessible surface areas (SASA) are shown for sulfur atoms in shared cysteines
on α4 helix for simulation set 1 (A) and set 2 (B) in RGS4, RGS8, and RGS19, and for shared
cysteines on α6-α7 interhelical loop in simulation set 1 (C) and set 2 (D) in RGS4 and RGS8.

and RGS8 were relatively stable; simulation set 2 did not recapture the α6 helix movement in RGS4

observed in simulation set 1. This illustrates a limitation of MD simulations in observation of

movements that occur infrequently or on long time scales. Taken together, these simulation sets

indicate highest flexibility in RGS19, with potential for flexibility in distinct regions in RGS8 and

RGS4. In all simulations for each protein, pronounced movements also occurred in the residues

located in terminal helices. This is likely an effect of free terminal ends; residues outside of the

RGS homology domains were not included in the simulations.

Analysis of solvent exposure of sulfur atoms reveals exposure of initially buried cysteines.

(Fig. 2-7). Cys123 in RGS19 is more exposed than analogous cysteines in RGS4 and RGS8 in the

2 μs simulation set (Fig. 2-7A) and again in the 3 μs simulation set (Fig. 2-7B). This may explain

the potency of RGS19 relative to the analogous single-cysteine RGS4 and RGS8. Pronounced

exposure of the α6-α7 interhelical Cys160 in RGS8 was observed in both sets of simulations (Fig.

36

Figure 2-8: Conformational changes during molecular dynamics simulations. Root mean square
deviations of α6 helix and α6-α7 loop, starting conformation, and a snapshot conformation during
MD simulation are shown for (A, D, G) RGS4, (B, E, H) RGS8, and (C, F, I) RGS19. Protein regions
plotted in MD trajectories are depicted in color in protein structures. Arrows indicate locations
of notable solvent exposure during simulation.

2-7C and D).

In addition, the conformations observed during movements of the α6 helix and α6-α7 loop

show distinct routes of cysteine exposure among the three RGS proteins. In the RGS4 crystal

structure (PDB: 1AGR),83 Asn140 occludes Cys148 from exposure to the protein surface (Fig. 2-

8D). In the MD simulation set 1 using 1AGR as initial coordinates, a transient movement of the

α6 helix was observed, reaching 15.1 Å between α-carbons at 1.24 μs (Fig. 2-8G), versus 5.9 Å at

baseline. This movement coincided with a high solvent exposure of Cys148 (Fig. 2-7C). In MD

simulation set 1 of RGS8 (using PDB code 2ODE141 as initial coordinates), helices α4, α5, α6, and α7

37

Figure 2-9: Snapshot of RGS19 from simulation set 2 at 240 ns. Cleft opening observed in simula-
tion set 1 (Fig 6I) was recapitulated in this simulation.

were stable relative to the same helices in other proteins tested. However, the α6-α7 interhelical

region, which includes cysteine 160, underwent a pronounced movement (Fig. 2-8B). Cys160

rotated toward the protein surface at 1 μs, and remained exposed to solvent for the remainder of

the trajectory (Fig. 2-7B and 2-8H). This cysteine exposure was observed again for the duration of

simulation set 2 (Fig. 2-7D). RGS19 lacks the cysteine in the α6-α7 interhelical loop, having only

Cys123 on α4. Both MD simulations of RGS19 (starting with the PDB code 1CMZ142) revealed a

movement of the α6-α7 interhelical loop away from the α4 and α5 helices, resulting in an open

groove in the protein surface (arrow in Fig. 2-8I and 2-9). This observation likely explains the

higher observed DI of α4 and α5 helices in RGS19 compared to RGS4 and RGS8, but additional

changes, perhaps induced by compound binding, may be required for full exposure of Cys123.

Discussion

RGS protein flexibility, as measured both by DI and solvent exposure of the α4 cysteine

in MD simulations, is correlated with the potency of TDZDs to inhibit RGS proteins containing

38

only a single shared cysteine. RGS19 had the most pronounced DI throughout the α4-α7 helix

bundle, and it was more potently inhibited by CCG-50014 than single-cysteine RGS4 or RGS8.

Such flexibility could result in increased likelihood of binding of TDZDs at the α4 cysteine. This

would be expected to lead to perturbation of residues involved in G-protein binding, as suggested

by previous NMR experiments.129

There was also good concordance between regional protein flexibility in the HDX-MS

studies and in MD simulations. In RGS8, helices α4, α5, α6, and α7 were protected from deuterium

exchange and were also stable during MD simulations. The dramatic movement of the RGS4 α6

helix in simulation set 1 mirrors its high solvent exposure in HDX studies. This suggests that

movement of the α6 helix is likely responsible for solvent exposure of Cys 148 in RGS4, providing

a plausible route of access by TDZD inhibitors.

Indeed, cysteine 148 was the most important

single cysteine for inhibition of RGS4 by our other cys-linking inhibitor, CCG-4986.31

Deuterium exchange is measured on a much longer timescale than is accessible by MD sim-

ulations. In order for exchange to occur, amide hydrogens must be in a conformation amenable

to exchange, requiring both interruption of H-bonds and proximity of solvent waters. These

exchange-competent states are short lived, often existing on a 10-100 picosecond timescale.131

They are frequent enough to be readily observed in microsecond timescale simulations; however,

the rate of intrinsic hydrogen exchange is much slower than the rate of hydrogen solvent expo-

sure. This is termed EX2 kinetics, in which an amide hydrogen may make multiple visits to a

solvent-exposed state before an exchange event occurs.143 While exchange is still representative

of the time spent in an open state, this allows observation of exchange on much longer timescales

than those of dynamic motions.

Interestingly, dynamic cysteine exposure varied among protein isoforms. In RGS4, move-

39

ment of helix 6 exposed the α6-α7 cysteine, while in RGS8, helix 6 was stable and that cysteine

rotated toward solvent in during a movement of the α6-α7 loop. RGS19 lacks a cysteine on the α6-

α7 loop, but opens a cleft toward a deeply buried α4 helix cysteine. These results suggest that the

route of modification by covalent inhibitors varies among RGS isoforms, even at shared cysteine

locations.

These differences in dynamic motions among RGS isoforms may contribute to differences

in potency of TDZD inhibition by two ways. First, differences in the rate of covalent modifi-

cation or the magnitude of effect on Gα binding may be driven by differences between RGS

isoforms in the direction of cysteine solvent exposure. Second, distinct transient conformations

occurring more frequently in certain RGS isoforms may permit unique non-covalent docking to

drive covalent modification. In such a scenario, the open state could be taken advantage of in a

docking-based virtual screen, permitting the discovery of non-covalent RGS inhibitors. Although

additional future work is required to fully understand the inhibitor access routes and mechanisms

(e.g. conformational selection versus induced fit), we have previously shown129 using nuclear mag-

netic resonance (NMR) and MD simulation analyses that an open conformation of RGS4 facilitates

covalent docking of CCG-50014 and leads to significant perturbations in residues near the bind-

ing pocket and at the protein-protein interface. This is because inhibitor binding only allows a

partial recovery of the open conformation to an apo-like conformation as opposed to a nearly

complete recovery in the absence of the inhibitor. Because conformational changes induced by

compound binding may be a factor in inhibition, we aim to undertake studies involving docking

of other TDZD and non-TDZD analogs116 using conformations of RGS proteins reported in this

work. These possibilities remain an object of future investigations.

40

Conclusions

The application of HDX-MS and MD methods reveal that RGS isoforms differ in their

mechanism of transient cysteine exposure, suggesting distinct routes of access by covalent in-

hibitors. These differences are potentially responsible for the selective potency of TDZD in-

hibitors among RGS isoforms. Importantly, the conformations of RGS proteins in which cysteine

residues are transiently exposed could be potentially useful for designing the next generation of

inhibitory small-molecules.

41

CHAPTER 3:

An Interhelical Salt Bridge Controls Flexibility and Inhibitor Potency For Regulators

of G-protein Signaling (RGS) Proteins 4, 8, and 19

Vincent Shaw performed protein expression, DSF, HDX, and FCPIA. Mohammadjavad Moham-
madi contributed MD simulations and RMSF, RMSD, and DCC analyses. Josiah Quinn assisted
with mutagenesis and protein expression and performed DSF.

42

Introduction

Drug specificity is often considered to be like a key fitting into a complementary shaped

lock.

It has become clear recently that protein dynamics can play in important role in drug

discovery.3 Regulators of G-protein Signaling (RGS) proteins bind to activated Gα subunits of G-

proteins, thereby accelerating GTP hydrolysis and attenuating G-protein signaling. In regulating

GPCR signaling, RGS proteins play a role in the physiology of numerous systems. By inhibiting

RGS proteins, signaling via a GPCR may be enhanced. There are twenty RGS isoforms, each

with a different tissue distribution. Combination of GPCR agonists with inhibitors specific for

a single RGS isoform should limit effects on GPCR signaling to the subset of target tissues with

intersecting distributions of the RGS isoform and the GPCR. This has the potential to reduce

agonist off-target effects and makes RGS proteins an attractive target for modulation of GPCR

signaling.

The potent RGS inhibitors discovered to date are all covalent modifiers of cysteine residues

and are selective for RGS4 and RGS1.31,116,118 These proteins have four and three cysteines, respec-

tively, in the RGS homology domain, which is more than most other RGS proteins. RGS4 has

been linked to nervous system related disease states in which RGS4 inhibition may be desirable,

including seizures98 and Parkinson’s disease.109,112,117 Continued efforts to seek non-covalent in-

hibitors are worth pursuing, because the lower risk associated with non-covalent inhibitors is

considered safer and may facilitate further development.60 In addition, it would be valuable to

discover RGS inhibitors with other specificities since other RGS proteins which are not potently

inhibited by covalent modifiers have been implicated as potential targets, including RGS17 in

cancer103,105 and RGS19 in depression.101 To identify noncovalent inhibitors with novel specifici-

ties, it will be useful to understand what factors apart from the number of cysteines in the RGS

43

domain drive selectivity of RGS inhibitors.

The RGS homology domain contains nine alpha helices. A cysteine residue on the α4 helix,

which faces the interior of the α4-α7 helical bundle, is conserved among 18 of the 20 RGS isoforms,

excepting only RGS6 and RGS7.126 Interestingly, when RGS proteins are mutated to contain only

this single, shared cysteine, there are still dramatic differences in the potency by which different

isoforms are inhibited.144 RGS19, which contains only the shared α4 cysteine, is more potently

inhibited than single-cysteine versions of RGS4 and RGS8.144,145

Previously, we found using molecular dynamics (MD) simulations that RGS19 is more flex-

ible than RGS4 and RGS8.144 In these modeling studies, we also found that salt bridge interactions

were perturbed in response to inhibitor binding146 In this work, we hypothesized that mutations

that alter salt bridge interactions will both enhance RGS protein flexibility and increase the po-

tency of RGS inhibitors such as CCG-50014.

Materials and Methods

Materials

Chemicals were purchased from Sigma-Aldrich (St. Louis, MO). QuikChange II Mutagene-

sis kit was purchased from Agilent (Santa Clara, CA). BL21(DE2) competent cells and Protein Ther-

mal Shift Dye Kit was purchased from Thermo Fisher Scientific (Waltham, MA). Lumavidin Mi-

crospheres were purchased from Luminex (Austin, TX). CCG-50014 {4-[(4- fluorophenyl)methyl]-

2-(4-methylphenyl)-1,2,4-thiadiazolidine-3,5-dione} was synthesized as previously described.115

Protein Expression and Purification

RGS proteins were produced as previously described.144 Briefly, a his-tagged RGS domain

44

of RGS8 in a pQE80 vector, a his-tagged RGS domain of RGS19 in a pET15b vector, and a his-

tagged Δ51 N-terminally truncated RGS4 in a pET23d vector were transformed into BL21(DE3)

competent E. coli cells (Sigma-Aldrich, St. Louis, MO). At an OD600 of 2.0, protein production

was induced by addition of 200 µM IPTG, and incubation was continued at 25 ℃ for 16 hours.

Cells were lysed and the protein was purified by nickel affinity chromatography. Mutations were

induced with a QuikChange mutagenesis kit (Agilent) and verified by Sanger sequencing. All

RGS proteins, including those with mutations in salt bridge-forming residues, were produced on

a single-cysteine background (WT RGS19, C160A RGS8, and C74A C132A C148A RGS4). Gαo

protein was expressed and purified as described.132

Differential Scanning Fluorimetry

Differential scanning fluorimetry was performed using the Protein Thermal Shift Dye Kit

(ThermoFisher Scientific, Waltham, MA). Dye was added at 1X to 10 µM protein samples in 50

mM HEPES and 100 mM NaCl buffer, pH 7.4 in a volume of 20 µL. Fluorescence was read using

a QuantStudio 7 Flex Real-Time PCR System while the temperature was ramped from 20 ℃ to 80

℃ at a rate of 0.05 ℃/s. Peak melting temperatures were defined as the point of fastest increase

in fluorescence with respect to temperature. Data was analyzed using Protein Thermal Shift

software v1.3 (Thermo Fisher Scientific, Waltham, MA) and Prism 7 (GraphPad Inc, LaJolla, CA).

Flow Cytometry Protein Interaction Assay (FCPIA)

FCPIA was performed as described133 with minor modifications. RGS proteins were bi-

otinylated by incubation at 1:1 molar ratio with EZ-link NHS-LC-biotin (Thermo Fisher Scien-

tific) for two hours on ice, then excess biotin was removed using Amicon Ultra centrifugal filters

(catalog no. UFC501096, Millipore, Burlington, MA). RGS proteins at 50 nM were incubated with

45

xMAP LumAvidin beads (Luminex, Austin, TX) while shaking at room temperature for 1 hour.

Beads were washed and incubated with varying concentrations of CCG-50014, followed by addi-

tion of 50 nM Gαo labeled with AF-532.133 Samples were read in a Luminex 200 flow cytometer

as described133 and analysis performed in GraphPad Prism 7.

Hydrogen-Deuterium Exchange

Hydrogen-deuterium exchange was performed as previously described.134,144 Briefly, pro-

teins were incubated on ice at 1.2 µM in 90% D2O solvent with 5 mM HEPES and 100 mM NaCl,

pH 7.4 for the desired time (1, 3, 10, 30, or 100 minutes). Exchange was quenched by 1:1 addition

of ice cold 1% formic acid. A Shimadzu pump was used to load 100 µL of each sample onto a

pepsin column (Waters, Milford, MA) followed by incubation for 1 minute for digestion. Samples

were the loaded to an Xbridge BEH C18 VanGuard trap column (Waters) and eluted and separated

using an Ascentis Express Peptide ES-C18 column (Sigma-Aldrich) with a gradient of 0.1% formic

acid to acetonitrile. All columns and solvents were maintained on ice. Peaks were detected with

a Xevo G2-XS QToF mass spectrometer (Waters). Data were analyzed using MassLynx (Waters),

HX-Express2,135 and GraphPad Prism 7.

Molecular Dynamics (MD) Simulation

The Vashisth Lab at UNH performed two sets of classical all-atom and explicit-solvent

MD simulations for single-cysteine RGS4 and RGS4 D90L, single-cysteine RGS8 and RGS8

E84L, and WT RGS19 and RGS19 L118D (Table 3-2) using the NAMD software136 on a high-

performance computing cluster (Towns et al., 2014) using the CHARMM force-field with the

CMAP correction.138,139 We used Visual Molecular Dynamics (VMD) for system creation and

post-simulation analysis.137 The initial coordinates were obtained from the protein data bank

46

Run No.

1
2
3
4
5
6

Initial
structure
System
1AGR
RGS4 D90L
1AGR
RGS4
2ODE
RGS8 E84L
RGS8
2ODE
RGS19 L118D 1CMZ
RGS19
1CMZ

Run length
(μs)
1
1
1
1
1
1

System size
(atoms)
30031
30031
32257
32257
25077
25077

No. of runs

2
2
2
2
2
2

Table 3-2: Details of MD simulations.

files with codes 1AGR (RGS4), 2DOE (RGS8), and 1CMZ (RGS19). Except for Cys95 in RGS4

and Cys89 in RGS8, all cysteines were changed to alanines. Each protein was then solvated in a

simulation box of TIP3P water molecules147 and charge-neutralized with NaCl. The final solvated

and ionized simulation domains contained 30031 atoms (RGS4), 32257 atoms (RGS8), and 25077

atoms (RGS19). Each solvated and ionized system was energy minimized for ∼500-1000 cycles

via conjugate-gradient optimization, then equilibrated via 1 μs MD simulations conducted with

a time-step (Δt) of 2 fs. The NPT (constant number, pressure, temperature) ensemble with a

Langevin thermostat and a damping coefficient of 5 ps-1 was used for temperature control and

the Nosé-Hoover barostat was used for pressure control. Periodic boundary conditions were

used throughout; non-bonded interactions were accounted for with a cut-off of 10 Å where

smooth switching was initiated at 8 Å. Long-range electrostatic interactions were handled using

the Particle Mesh Ewald (PME) method.

Dynamic cross-correlation analysis

The dynamic cross-correlation (DCC) maps of each system were calculated based on the

Cα atoms of residues using the MD-TASK package.148 Each cell value (Cij) in the matrix of the

DCC map was calculated using the following formula:

47

(√

Cij =

)

⟩

⟨∆ri·∆rj⟩
√
⟨∆r2
⟨∆r2

⟩·

i

j

Where Δri represents the displacement from the mean position of atom i, and < > denotes

the time average over the whole trajectory. Positive values of Cij show correlated motion between

residues i and j, moving in the same direction, whereas negative values of Cij show anti-correlated

motion between residues i and j, moving in the opposite direction.

Analysis of salt-bridge interactions

Salt-bridge interaction analysis was carried out using VMD based on a distance criterion

uniformly applied to determine the existence of salt-bridges for each frame in all trajectories.

Specifically, salt-bridge interactions were considered to be formed if the distance between any

of the oxygen atoms of acidic residues and the nitrogen atoms of basic residues were within a

cut-off distance of 4 Å.

Statistical Analysis

All deuterium exchange and functional inhibition data were done with an n of 3 indepen-

dent experiments. Sample sizes were predetermined before experiments were done. Changes in

thermal stability were analyzed by 1-way ANOVA with Sidak’s multiple comparisons post-test.

H0: There is no difference in thermal stability between WT and mutant RGS proteins. Differences

in deuterium incorporation were analyzed using 2-way ANOVA with Sidak’s multiple compar-

isons post-test. H0: There is no difference in deuterium incorporation between WT and mutant

RGS proteins. Error bars represent means ±SD. In saturation binding experiments, RGS-Gα inhibi-

tion was determined by fitting total and nonspecific binding. In functional inhibition experiments,

IC50 was determined by fitting a four-parameter logistic curve. All curve fitting and statistical

analysis was done using GraphPad Prism 7 (GraphPad Inc, LaJolla, CA).

48

Results

Comparison of the structures for RGS19 (PDB 1CMZ),142 RGS4 (PDB 1AGR),83 and RGS8

(PDB 5DO9)128 shows that there are differing numbers of interhelical salt bridges on the exteriors

of their α4-α7 helix bundles. Some of these may contribute to differences in stability and dynamics

among the RGS isoforms.

RGS19 has only one interhelical salt bridge in this bundle, between E125 (α4) and K138

(α5) (Fig. 3-1A and B). However, this salt bridge is well conserved among all three proteins (Fig. 3-

1A-D), so it is unlikely to contribute to observed differences in flexibility.144 A salt bridge network

that connects α4, the α5-α6 interhelical loop, and α5 is present in RGS8 (E84-R119-E111) and RGS4

(D90-K125-E117) but absent in RGS19 (Fig. 3-1A and B). The residues that form this network are

present in 7 of the 20 RGS protein family members, all in the R4 subfamily. Between the α5 and

α6 helices, a salt bridge is present in RGS8 (D114-R132), but absent in both RGS4 and RGS19 (Fig.

3-1A and C). Finally, a charged pair between the α6 and α7 helices is present in RGS8 (E91-K104)

and RGS4 (D130-K155), but is absent in RGS19 (Fig. 3-1A and D).

To estimate the relevance of each of these salt bridges in maintenance of helix bundle

rigidity, the time each amino acid in a charged pair spent within a 4Å of one another over the

course of a long timescale (2 μs) MD simulation144 was measured. The α6-α7 salt bridge, which

is present in RGS4 and RGS8 but absent in RGS19, occupied a salt bridge-forming distance for

31.5% of the simulation in RGS4 and 36.1% in RGS8. The salt bridge interaction between residues

of α4 and α5-α6 interhelical loop, also not present in RGS19, was maintained for 58.7% of time in

RGS4 and 44.2% in RGS8 (Table 3-3). The charged pair that is unique to RGS8 between α5 and α6

helices remained in contact for 47.5% of the simulation.

We elected to make mutations that altered interhelical salt bridges to test their functional

49

Figure 3-1: (A) Alignment of RGS19, RGS4, and RGS8 sequences in α4-α7 helix bundle. Charged
residues that make interhelical contacts are indicated in red and blue. RGS19 has 1, RGS4 has
3, and RGS8 has 4 salt bridges. Structural alignments of α4-α5 (B), α5-α6 (C), and α6-α7 (D) helix
pairs are shown, with highlighted residues in panel a rendered as sticks. RGS19 (PDB 1CMZ) is in
green, RGS4 (PDB 1AGR) is in yellow, and RGS8 (PDB 5DO9) is in cyan. Black brackets in panel
A indicate residues depicted in panels B, C, and D

% of sim
within
4Å
58.7
44.2
-

% of sim
within
4Å
-
47.5
-

α5-α6

S120
S138
D114 R128
S148 N166

% of sim
within
4Å
31.5
36.1
-

CCG-
50014
IC50
(μM)
8.5
>1000
1.1

α4-α5

α6-α7

D90
E84

RGS4
K125
RGS8
R119
RGS19 L118 K153
Table 3-3: The salt-bridge interaction within the α4-α7 bundle of helices in single-cysteine struc-
ture of RGS4, RGS8, and RGS19 from MD simulations and potency of CCG-50014 inhibition of
single-cysteine RGS proteins in our previous work.144

D130 K155
D124 K149
D158 Q183

50

Figure 3-2: L118D mutation increases thermal stability of RGS19, but Q183K mutation has no
significant effect (n = 3, 1-way ANOVA with Sidak’s multiple comparison test. ****p < 0.001).
L118D mutation in RGS19 has reduced potency of inhibition of CCG-50014, but Q183K mutation
does not. Ki, calculated using a Cheng-Prusoff correction,232 is reported to account for effect of
mutations in RGS on Gαo affinity.

roles. There are two positions at which interhelical salt bridges are shared by RGS4 and RGS8 but

are absent in RGS19: α4-α5 (Fig. 3-1B) and α6-α7 (Fig. 3-1D). In the α4 helix of RGS19, L118 was

mutated to an aspartate to introduce the α4-α5 salt bridge found in RGS4 and RGS8 (Fig 3-1B).

In helix α7 of RGS19, Q183 was mutated to a lysine to introduce the α6-α7 salt bridge found in

RGS4 and RGS8 (Fig 3-1D). In order to eliminate confounding effects due to multiple cysteines

in inhibitor potency experiments, all proteins, with and without salt-bridge mutations, used a

single-cysteine protein background. Each construct has only the conserved cysteine in helix α4

of the RGS domain.

To determine how disruption or addition of a salt bridge may alter protein structure or

dynamics, thermal stability was measured by differential scanning fluorimetry. Addition of a

salt bridge in RGS19 by the L118D mutation caused a 7 ℃ increase in thermal stability compared

51

to WT (Fig 3-3A). In contrast, the Q183K mutation in RGS19 did not alter thermal stability or

inhibitor potency (Fig. 3-2). Removal of a salt bridge in RGS8 by the E84L mutation caused an 8

℃ decrease in thermal stability (Fig 3-3B). Unexpectedly, RGS4 showed a more complex pattern

in which the D90L mutation resulted in a biphasic melt curve and a 5 ℃ increase in melting

temperature rather than a decrease (Fig 3-3C).

To probe the molecular details of changes in structural flexibility in the mutant proteins,

we conducted microsecond timescale classical MD simulations in explicit-solvent for RGS19

L118D, RGS8 E84L, and RGS4 D90L. Root-mean-square deviations (RMSDs) of these simulations

are shown in Fig. 3-4. To understand the effect of the mutations on the protein structures,

particularly in helices in the vicinity of the mutated site, we computed the root-mean-square

fluctuation (RMSF) per residue from two independent MD simulations of mutated and WT RGS4,

RGS8, and RGS19. The calculated change in RMSF per residue of the mutant RGS19 L118D from

wild-type RGS19 reveals a strong stabilization and decrease in fluctuations of residues located

in helices α4-α7 and in the interhelical loops between these helices. There is a particularly

pronounced decrease in motion in the α5-α6 interhelical loop (Fig. 3-5A). We find a modest

increase in fluctuation of residues in mutant RGS8 E84L vs. the wild-type structure (Fig. 3-5B).

These changes are in the loop region connecting helices α5 and α6, the α6 helix, and the loop

connecting helices α6 and α7. Similar changes but of lesser extent were found in the mutant

RGS4 D90L (Fig. 3-5C). Additionally, small decreases were observed in the RMSF values of

residues in helices α3 and α8 of the mutated RGS19 (Fig. 3-5A), but not in the mutated RGS8 and

RGS4 (Fig. 3-5B and C).

To further investigate whether salt bridge-modifying mutations in RGS4, RGS8, and

RGS19 affect residue-residue interactions, we calculated dynamic cross-correlation matrices

52

Figure 3-3: Thermal stability was determined by differential scanning fluorimetry. (A) The L118D
mutation in RGS19 increased melting temperature by 7 ℃ compared to WT. (B) The E84L muta-
tion in RGS8 decreased melting temperature by 8 ℃. (C) The RGS4 D90L mutation introduced
a biphasic melt curve and increased melting temperature by 5 ℃. For each pair, the three repli-
cate derivative melt curves are shown on the left and average melt temperatures are shown on
the right. Error bars represent SD. n = 3. Analyzed by 1-way ANOVA with Sidak’s Multiple
Comparisons test. ****p < 0.0001

53

Figure 3-4: The traces of root-mean-squared-deviation (RMSD) vs. simulation time (μs) for (a)
RGS4 D90L, (b) RGS8 E84L, and (c) RGS19 L118D. Two independent simulation runs for each
structure are presented, and the wild-type runs are presented from our previous work.144

54

Figure 3-5: Change in RMSF per residue (ΔRMSF) between wild-type RGS proteins and RGS
proteins with mutation in the α4-α5 salt bridge forming residue. (A) L118D in RGS19 (B) E84L
in RGS8 and (C) D90L in RGS4. Data represent differences in RMSF from two independent MD
simulations of the mutated forms of RGS proteins.

55

for the Cα atoms in all MD trajectories. For WT RGS19, RGS8 and RGS4, there is a modest

positive correlation between the motions of residues of the α4 helix and the residues of the α5

helix (Fig. 3-6A-C). For the RGS19 L118D mutant, we find higher residue-residue correlations

between helices α4 and α5 in comparison to unmutated RGS19 (see arrows, Fig 3-6A). There was

no appreciable change between WT and mutant RGS4 (Fig 3-6B). For wild-type RGS8, we find

that the motions of residues in the α4 helix (aa 79-93) and the α5 helix (aa 97-113) are marginally

positively correlated (see arrows, Fig. 3-6C). This positive correlation between the α4 and α5

helices remains in the RGS8 E84L mutant, but shows a modest shift in areas of correlation away

from the loop connecting α4-α5 to mid-regions of the α4 and α5 helices (see arrows, Fig. 3-6C).

In order to experimentally determine which regions in WT and mutant proteins were

affected by the salt bridge mutations, hydrogen-deuterium exchange studies were performed. Af-

ter exposure to solvent containing 90% D2O, proteins were digested with pepsin and deuterium

incorporation (DI) was measured by mass spectrometry as previously reported.144 In RGS4, the

fragment surrounding the salt-bridge mutation site (aa 88-91) took up deuterium very slowly in

both the WT and D90L mutant constructs, reaching 8.1% and 6.7% DI, respectively. However,

the D90L mutation led to a substantial increase in deuterium exchange in the 92-97 fragment sur-

rounding Cys95, from 17.5% to 37.0% DI. The RGS4 D90L mutant also trended toward increased

DI across all protein fragments compared to WT RGS4, especially at higher timepoints (Fig. 3-

7A). In RGS8, removal of the salt-bridge forming residue by the E84L mutation did not cause a

significant change in DI in either of the fragments of the α4 helix but trended toward a global

increase in DI throughout the protein (Fig. 3-7B). In RGS19, mutation of L118 to a salt bridge-

forming residue, aspartic acid, caused significant decreases in DI in both α4 helical fragments, aa

116-119 and aa 120-125. In the 116-119 fragment, WT RGS19 had reached 43.1% DI by 10 minutes,

56

Figure 3-6: Dynamic cross correlation matrix calculated for the Cα atoms of (A) RGS19/RGS19
L118D, (B) RGS8/RGS8 E84L, (C) RGS4/RGS4 D90L. Horizontal dotted lines indicate the regions
of the α4 helix, while vertical solid lines indicate the regions of the α5 helix for each protein.
The color scheme ranges from anticorrelation (-1.0, blue), no correlation (0, green), and positive
correlation (+1.0, red). Values are the average for the two independent simulation runs.

57

Gαo KD (nM) CCG-50014 IC50 (μM) CCG-50014 Ki (μM)

RGS19
RGS19 L118D
RGS8
RGS8 E84L
RGS4
RGS4 D90L

16.6
20.2
5.9
4.8
5.2
3.9

1.1
7.0
29.0
4.6
8.8
2.2

0.27
2.01
3.06
0.40
0.83
0.16

Table 3-4: Interaction affinities between Gαo and RGS proteins and mutants, and IC50 and Ki
of inhibition of RGS-Gαo binding by CCG-50014. Ki values were calculated by Cheng-Prusoff
correction of the IC50 values.

while the RGS19 L118D mutant showed less than half as much DI (18.7%). In fragment 120-125,

WT RGS19 reached 18.5% DI at 10 minutes, while the RGS19 L118D mutant reached only 6.2%.

Unlike RGS4 and RGS8, the RGS19 L118D mutant’s changes in DI were more restricted to frag-

ments from helices neighboring the mutation site, and were most pronounced in the early (1 to

10 minute) timescale (Fig. 3-3C).

Finally, to assess the functional relevance of the α4 salt-bridge forming residues, we used

a flow-cytometry based protein-protein interaction assay (FCPIA)113,133 to measure the binding of

RGS proteins to Gαo and the potency of inhibition by CCG-50014. The L118D mutation in RGS19

induced an increase in IC50 from 1.1 µM (WT) to 7.0 µM (L118D) (Fig. 3-8A). Conversely, removal

of this charged α4 residue in RGS4 and RGS8 induced a decrease in IC50 (Fig. 3-8B and C). CCG-

50014 inhibited the RGS-Gα interaction with an IC50 of 8.8 µM for WT RGS4 and 2.2 µM for the

RGS4 D90L mutant. It showed a potency of 29 µM for WT RGS8 and 4.6 µM for the RGS8 E84L

mutant. None of the mutations to salt bridge-forming residues on the α4 helix caused notable

changes in affinity between Gαo and RGS proteins. The L118D mutation in RGS19 shifted the Kd

of the Gαo interaction from 17 nM to 20 nM, the E84L mutation in RGS8 shifted the Kd from 5.9

nM to 4.8 nM, and the D90L mutation in RGS4 shifted the Kd from 5.2 nM to 3.9 nM (Table 3-4).

58

Figure 3-7: Difference in %deuterium incorporation (Δ%DI) between mutated and unmutated pro-
teins in RGS19 L118D (A), RGS8 E84L (B), and RGS4 D90L (C) fragments, as measured by HDX.
Red arrows indicate fragments containing mutated residue, and black arrows indicate fragments
containing conserved α4 cysteine. Kinetics of deuterium incorporation in these fragments for in-
dividual constructs are shown below. n = 3. Error bars represent SD. Analyzed by 2-way ANOVA
with Sidak’s multiple comparisons test. *p < 0.05, **p < 0.01, ****p < 0.0001.

59

Figure 3-8: Potency of inhibition of CCG-50014 against α 4 is altered in salt bridge mutants of
RGS proteins. (A) RGS4 IC50: 8.8 µM, RGS4 D90L IC50: 2.2 µM. (B) RGS8 IC50: 29 µM, RGS8
E84L IC50: 4.6 µM. (C) RGS19 IC50: 7.0 µM, RGS19 L118D IC50: 1.1 µM. n=3.

60

Discussion

A comparison of the crystal structures of the three RGS proteins studied here revealed

several differences in charged residue contacts among the proteins. We first observed that RGS19

has fewer interhelical salt bridges in its α4-α7 helical bundle than RGS4 or RGS8. This may be

responsible for the high flexibility previously observed in WT RGS19.144 RGS8 has four distinct

interhelical salt bridges within the helical bundle, while RGS4 has three and RGS19 has one (Fig 3-

1A), correlating with previously observed flexibility differences. RGS19 is most flexible, followed

by RGS4 and RGS8.144 This further supports a role of salt bridges in RGS protein flexibility.

The changes in thermal stability in response to mutations in the α4 helix salt bridge-

forming residues suggest that this location may be responsible for differences in stability and

dynamics among the isoforms. This is supported by the increase in thermal stability in response

to the L118D mutation in RGS19, and destabilization in RGS8 response to the E84L mutation.

While the D90L mutation altered thermal stability in RGS4, it stabilized rather than destabilized

the protein. The biphasic melt curves in D90L RGS4 make the thermal stability data difficult to

interpret. HDX clarifies the effect of the D90L mutation in RGS4 by showing localized increases

flexibility of the protein. The lack of effect on thermal stability with the Q183K mutation in

RGS19 correlates with the observation that the α6-α7 salt bridges in RGS4 and RGS8 were less

stably maintained in simulations than were the α4-α5 salt bridges. In light of these results, we

found it unlikely that the difference between Q183 in α6 of RGS19 and the lysines found in RGS4

and RGS8 (K155 and K149 respectively) play a major role in the flexibility differences between

these proteins. Rather, the salt bridge-forming residue on α4 is a stronger driver of differences in

protein flexibility.

To determine the effects of mutations in salt bridge-forming residues on protein dynamics,

61

both an in silico approach (all-atom MD simulations) and an experimental approach (hydrogen-

deuterium exchange) were employed. In simulations, the increase in positive correlation between

residues in the α4 and α5 helices in the RGS19 L118D mutant likely results from the introduced

interhelical salt-bridge. The decrease in DI in the α4 helix of RGS19 in the HDX studies is consis-

tent with reduced solvent exposure. This is of particular interest given that the Cys123 target of

the TDZD compounds is located in that helix. Conversely, mutations that eliminated salt bridges

in RGS4 and RGS8 increased DI in some fragments from their α4 helices (Fig. 3-7A and B), sug-

gesting that this results in increased solvent exposure and greater compound accessibility at the

buried cysteine. Surprisingly, the RGS4 D90L mutant did not have increased DI in the fragment

spanning the mutation site (Fig. 3-7C). In addition, the μs timescale MD simulations captured

positive residue-residue (Cα-Cα) correlations between the α4 and α5 helices of that were similar

in WT and mutated RGS4 D90L. This fits with the thermal stability data and suggests that the

effect of the D90L mutation in RGS4 is more complex than simple disruption of an ionic contact.

In MD simulations, the RGS4 D90L and RGS8 E84L mutations did not have as large an

effect on the magnitude of residue fluctuations as did the L118D mutation in RGS19 (Fig. 3-5A

and B). This may be because differences become apparent on shorter timescales in RGS19 than in

RGS4 and RGS8, so simulations on μs timescales may not have captured all of the differences in

dynamics caused by mutations in RGS4 D90L and RGS8 E84L. Indeed, in HDX studies, stronger

differences in DI were observed between RGS19 and RGS19 L118D at shorter timepoints (1 and 3

minutes) than in RGS4 D90L and RGS8 E84L (Fig 3-7A-C).

Finally, to determine how changes in protein flexibility affected the potency of inhibition

by an RGS inhibitor, we used FCPIA to evaluate the inhibition of Gα binding by CCG-50014. Im-

portantly, manipulation of RGS protein flexibility induced the expected changes in the potency

62

of inhibition by TDZD covalent modifiers. Thus, enhancing flexibility by removal of salt bridge-

forming residues increased the potency of inhibition by CCG-50014 while reducing protein flex-

ibility reduced potency of inhibition by CCG-50014. These results support a causal relationship

between RGS protein flexibility and potency of inhibition.

In conclusion, differences in flexibility among RGS isoforms appear to drive differences

in the potency of a covalent inhibitor, CCG-50014. The differences in isoform flexibility in turn

are strongly influenced by the presence or absence of an α4-α5 salt bridge and manipulation of

this salt bridge is sufficient to induce changes in inhibitor potency among single-cysteine RGS

proteins. Developing a deeper understanding of these differences in flexibility may enable the

development of a new generation of RGS inhibitors with novel specificities.

63

Distinct Roles of Individual Cysteines in Covalent Inhibition of RGS Proteins

CHAPTER 4:

Vincent Shaw expressed proteins, performed FCPIA, MS, and SDS-PAGE, and analyzed data. Ryan
Puterbaugh and Dr. Kriszina Varga performed NMR and prepared spectra.

64

Introduction

Signaling via heterotrimeric G-proteins is a pathway critical to pharmacology.12 G-

proteins are activated upon agonist binding to GPCR, allowing GDP release from the Gα

subunit and GTP association. This puts the G-protein in its active conformation, initiating

downstream signaling via the Gα subunit and the Gβγ dimer. Regulators of G-protein signaling

(RGS) proteins end signaling through the G-protein by binding to the active Gα subunit and

accelerating hydrolysis of GTP. This GTPase-Activating Protein (GAP) activity is mediated by

the RGS domain, a 130 aa domain with nine α helices.82,83

There has been interest in targeting RGS proteins as a strategy for modulating G-protein

signaling. By inhibiting the GAP activity of RGS proteins, GPCR-mediated signaling may be

increased. Because there are many RGS isoforms with unique physiological roles, isoform se-

lectivity will be particularly important to limit off-target effects in the therapeutic use of RGS

inhibitors.149 Covalently acting inhibitors have been developed that prevent binding between the

RGS domain and the Gα subunit by modification of cysteines in the RGS domain.113,114,116,150–152

These include CCG-203769, a thiadiazolidinone (TDZD) inhibitor that is selective for RGS4 and

may hold promise for treatment of Parkinson’s disease.109,117 A better understanding of the role of

individual cysteines in RGS inhibition by CCG-203769 will help define the molecular mechanism

of isoform specificity.

Previous work demonstrates a relationship between RGS isoform dynamics and potency

of inhibition among three RGS proteins (RGS4, RGS8, and RGS19) when mutated to contain a

single cysteine.129,144,146 However, many RGS isoforms contain additional cysteines that may in-

fluence the potency of covalent modifiers. A cysteine on the α4 helix is very well conserved,

shared by all of the RGS proteins with the exception of RGS6 and RGS7. This cysteine is found

65

in all members of the R4 and RZ families. Only one other cysteine, on the α7 helix, is conserved

among the some of the RGS domains of RGS proteins. This is present in eight of the ten R4 fam-

ily members, including RGS4 and RGS8. It is not found in RZ family members, such as RGS19.

Although the α4 and α7 cysteines are near one another on adjacent helices in the 3D structure,

existing crystal structural information does not indicate the presence of a disulfide bond in the

apo structure.83,128,141,153

Our previous work demonstrates that protein dynamics plays a role in the isoform speci-

ficity of TDZDs when compound action is restricted to a shared, single cysteine on the α4 helix.144

However, many RGS proteins, including RGS4 and RGS8, have additional cysteines in the RGS

domain that may contribute to potency of covalent modifiers. The RGS proteins most potently

inhibited by the TDZD CCG-50014 are RGS4 and RGS1, both of which have additional cysteines

in the RGS domain beyond the well-conserved α4 and α7 cysteines.118. While a correlation be-

tween number of cysteines and potency of inhibition has been noted among TDZDs and several

other inhibitors,31,118 the way that individual cysteines contribute to compound-induced changes

in protein conformation have not been fully elucidated, particularly at cysteines beyond the con-

served α4 cysteine.

In this work, we provide evidence that the TDZD CCG-203769 can act at

multiple cysteines, and may mediate a unique interaction leading to the induction of a disulfide

bond between cysteines common to many RGS isoforms on the α4 helix and the α7 helix.

Materials and Methods

Protein purification and expression

Single-cysteine constructs of RGS8 containing either Cys107 or Cys160 were generated

66

by individual mutation of each cysteine to serine using QuikChange mutagenesis (Agilent,

Santa Clara, CA). An RGS8 C160S mutant is termed Cys107 RGS8 and an RGS8 C107S mutant

is termed Cys160 RGS8. His-tagged expression constructs of the RGS domain of RGS8 and

RGS19, a Δ51 N-terminally truncated RGS4, and Gαo were used to prepare the tagged proteins as

previously described.132,144 Isotopically labeled proteins were expressed by plasmid transfection

into BL21(DE3) competent cells (Sigma-Aldrich, St. Louis, MO). These were grown at 37 ℃ in LB

to OD600 0.7, followed by centrifugation at 8000×g for 20 minutes and resuspension at 25% of the

original volume in phosphate-buffered minimal media (recipe described by Storaska and Neubig,

2013),154 supplemented with 4 g/L D-Glucose and 1 g/L (15NH4)2SO4 (Sigma-Aldrich, St. Louis,

MO). Cells were incubated in minimal media for 30 minutes at 37℃, 200 μM IPTG was added,

and the temperature was lowered to 25 ℃ and protein was induced for 12 hrs. Cells were lysed

by sonication and centrifuged at 120,000×g for 1 hr. The cell lysates were batch purified on a

nickel affinity column and eluted with 300 mM imidazole in 50 mM HEPES and 500 mM NaCl,

pH 7.4. Protein was further purified by cation exchange chromatography. An SP sepharose

column (GE, Chicago, IL) was equilibrated with 50 mM sodium phosphate, 40 mM NaCl, and 1

mM DTT (pH 6.9) and protein was eluted using linear gradient to buffer including 1M NaCl.

NMR Spectroscopy

Protein was dialyzed to buffer containing 50 mM sodium phosphate, 40 mM NaCl, pH

6.0 and concentrated to 50 μM using Amicon 10,000 Da MWCO centrifugal filter columns (Milli-

pore, Burlington, MA). D2O, NaN3, and 4,4-dimethyl-4-silapentane (DSS) were added to achieve

5% v/v, 4 mM, and 0.2 mM final concentration, respectively. Titrations of the protein with the

CCG-203769 ligand were prepared, with the ligand protein-ligand concentration of 1:1, 1:2 and 1:4

67

(molar ratio). Samples were then packed into Shigemi NMR tubes, and 1H-15N HSQC (Heteronu-

clear Single Quantum Correlation) NMR spectra were collected at 25 ℃ at the CUNY Advanced

Science Research Center (ASRC). WT RGS8 and Cys107 RGS8 data were collected using a Bruker

AVANCE III HD 800 MHz NMR spectrometer equipped with a Bruker Ascend UltraShield Plus

18.8 Tesla standard bore magnet and a TCI Cryoprobe. Cys160 RGS8 data were collected using

a Bruker AVANCE III HD 700 MHz NMR spectrometer equipped with a Bruker UltraShield 16.4

Tesla standard bore magnet and a QCI-F CryoProbe. All NMR data were processed using the

programs Bruker Topspin and NMRpipe.155 All processed NMR data were analyzed using the

program NMRFAM Sparky.156

Iodoacetamide alkylation and trypsin digestion

RGS4 was treated with varying concentrations of iodoacetamide (IAA) in 50 mM HEPES

and 100 mM NaCl buffer, pH 7.4 and incubated at room temperature while shaking, protected

from light. Free IAA was removed by buffer exchange using Amicon 10,000 Da MWCO centrifu-

gal filter columns (Millipore) into digestion buffer (400 mM ammonium bicarbonate, 5 mM DTT,

pH 7.5) with 8 M urea. Samples were diluted to 0.5 M urea in digestion buffer. Trypsin from

porcine pancreas (Sigma-Aldrich catalog no. T0134) was added at a ratio of 1:1 protein:trypsin.

The mixture was incubated at 37 ℃ overnight before analysis by mass spectrometry (see methods

below). Peak height may be used as measurement of peptide quantity.157 The peak intensities of

iodoacetamide-alkylated peptides were determined for the most abundant RGS4 fragments result-

ing from trypsin digestion that contained one cysteine. The four cysteines in the RGS domain

of RGS4 and the associated fragments are: Cys71, aa 59-76; Cys95, aa 78-99; Cys132, aa 126-134;

and Cys148, aa 140-155. The intensities of alkylated fragments are expressed as a percent of the

68

sum of the intensities of alkylated and unalkylated fragments.

Protection of RGS8 from iodoacetamide akylation by CCG-203769

RGS8 WT, Cys160, or Cys107 proteins at 50 μM in 100 mM HEPES and 100 mM NaCl (pH

7.4) were treated with 100 μM CCG-203769 or DMSO vehicle (final concentration 1%) at room

temperature for 1 hr. An excess of iodoacetamide (500 μM) was added to quench the action of

CCG-203769 by alkylation of any free cysteine thiols. The mixture was incubated in the dark at

room temperature for 1 hr, then diluted 10-fold in urea (final concentration 9 M) to ensure access

of iodoacetamide to free cysteines.

Protein mass spectrometry

Samples were injected using a Waters 2777c autosampler and desalted by trapping on a

Hypersil Gold CN guard column (1 x 10 mm, Thermo Fisher Scientific, Waltham, MA) for full

proteins or separated using an Ascentis Express Peptide ES-C18 column (2.1 x 50 mm, Supelco,

Bellefonte, PA) for protein fragments using a gradient of 0.1% formic acid in H2O and acetonitrile.

Proteins were ionized by electrospray ionization using a Xevo G2-XS QToF mass spectrometer

(Waters, Milford, MA) in positive ion mode, collecting data in continuum mode over m/z 100-

2000. Full length protein spectra were deconvoluted and analyzed using the MaxEnt1 algorithm

in MassLynx (Waters).

Flow cytometry protein interaction assay

The flow cytometry protein interaction assay (FCPIA) was performed as previously

described133 to measure RGS-Gα binding. Briefly, biotinylated WT, Cys107, or Cys160 RGS8 was

linked to Lumavidin microspheres (Luminex, Austin, TX). These were incubated with varying

69

concentrations of CCG-203769 or vehicle (DMSO) for 30 minutes. AF-532-labeled Gαo (50

nM final concentration) was added and bead fluorescence was read using a Luminex 200 flow

cytometer.

Non-reducing SDS-PAGE

Protein samples at 5 μM were pretreated with vehicle or 250 μM CCG-203769. Where

indicated, disulfides were reduced by addition of 1 mM dithiothreitol (DTT). Samples were mixed

with SDS sample buffer (Bio-Rad) devoid of BME or other reducing agent, and separated by SDS-

PAGE using a 15% polyacrylamide gel. Proteins were visualized using Coomassie Blue stain.

Results and Discussion

Cys148 in RGS4 is more accessible to a covalent modifier than Cys95

Differences in accessibility of the cysteines to the solvent may contribute to variation

in TDZD action at different cysteines in RGS domain proteins. There are two cysteines that are

conserved among R4 family members, one on each of the the α4 and α7 helices of the RGS domain

(Fig. 4-1A). RGS4 is a representative RGS protein that has both of these cysteines, but also has

two others. Previous studies have suggested mechanisms by which otherwise buried cysteines

on RGS proteins may access solvent.129,144 In MD simulations, the α7 helix cysteine was found to

have more solvent accessible surface area than the α4 cysteine in both RGS4 and RGS8.144

To verify this experimentally, cysteine accessibility in RGS4 was measured by assessing

the degree of modification by a general cysteine alkylator, iodoacetamide. By fragmenting the

protein with trypsin, the degree of modification at each fragment can provide an indication of

the relative exposure of individual cysteines; however, the rate of alkylation may also be affected

70

Figure 4-1: (A) Locations of cysteines in RGS protein based on structure of RGS4 (PDB: 1AGR).
α4 and α7 cysteines, conserved across multiple RGS proteins, are marked in blue. The α3 and α6
helix cysteines, unique to RGS4, are marked in red. (B) Degree of IAA alkylation at Cys71 (α3),
Cys95 (α4), Cys132 (α6), and Cys148 (α7) in RGS4.

by factors such as how readily cysteine thiols convert to the thiolate anion.158 Cysteines on the α3

and α6 helices appear to be more exposed, as they became 97.4% and 99.8% alkylated respectively

when exposed to 250 μM IAA (Fig. 4-1B). This is consistent with their high degree of solvent

exposure in crystal structures (Fig. 4-1A).83,153 Cysteines on α4 and α7, which are conserved in

RGS8 and other R4 family proteins, were less readily alkylated. The α7 cysteine reached 60.4%

alkylation with 250 μM IAA, while the α4 cysteine reached only 30.3% alkylation (Fig. 4-1B).

This indicates that the α7 cysteine may be more solvent-exposed than the α4 cysteine, which is

consistent with our previous modeling data showing that the α4 cysteine has less solvent-exposed

surface area than the α7 cysteine in both RGS4 and RGS8.144

CCG-203769 can directly act upon either cysteine in RGS8

To determine the individual roles of the two conserved cysteines (on α4 and α7 helices), we

used the RGS8 protein and CCG-203769 ligand in 1H-15N HSQC NMR spectroscopy. This protein

was chosen because of the evidence supporting the stability of RGS8.144 Some RGS proteins may

71

be sensitive to denaturation upon interaction with small molecules.154 RGS8 has been observed to

have high thermal stability relative to other RGS proteins (Fig. 3-2), which may make it a better

candidate for studies requiring a long durations in solution or at elevated temperatures. CCG-

203769 was chosen as ligand. Despite inhibiting RGS8 with a lower potency than CCG-50014,

CCG-203769 is more soluble in aqueous solution.116

High quality 2D 1H-15N HSQC NMR spectra of the RGS8 proteins (WT, Cys107, and Cys160)

were obtained. In the WT spectrum (Fig. 4-2A), peak count corresponds well with the expected

1H-15N correlations. The wide chemical shift dispersion, well-defined peaks with roughly uniform

intensities and line shapes reflect a folded, homogeneous protein tertiary structure.

To probe the effect of ligand binding with RGS8, WT RGS8 (15N-enriched) was mixed with

CCG-203769 (unlabeled) at 1:1, 1:2, and 1:4 RGS8:ligand ratio. The ligand induced changes in the

RGS8 1H-15N HSQC spectra (Fig. 4-2C and 4-3A) and chemical shift perturbations increased in

magnitude with increasing concentrations of compound, indicating binding of CCG-203769 to

the protein. Once the assignments of the RGS8 spectra are completed, we will identify which

residues in RGS8 are perturbed in response to the small molecule inhibitor.

The titration of the Cys107 RGS8 with the ligand also yielded high quality spectra (Fig.

4-3B). Similarly, several peaks in Cys107 RGS8 were perturbed in response to CCG-203769, indi-

cating that CCG-203769 can also act at Cys107. Interestingly, many of the same peaks that were

perturbed in WT RGS8 were also perturbed in Cys107 RGS8 (Fig. 4-3B and D), indicating that

CCG-203769 affects protein conformation similarly between WT and Cys107 RGS8. This suggests

that Cys107 likely is involved in inhibition of WT protein function.

72

Figure 4-2: WT RGS8 protein NMR spectra. (A) 1H-15N HSQC NMR spectrum of WT RGS8. (B)
The structure of ligand CCG-203769. (C) Overlay of 1H-15N HSQC NMR spectra of WT RGS8 be-
fore (red spectrum) and after the addition of its ligand CCG-203769 at 1:1, 1:2, and 1:4 RGS8:ligand
ratio (grey spectra). Shifted residues are highlighted in the zoomed spectrum. Spectra were ac-
quired at 25 ℃ on a Bruker AVANCE III HD 800 MHz NMR spectrometer equipped with a TCI
Cryoprobe at the CUNY Advanced Science Research Center NMR facility.

73

Figure 4-3: Chemical shift perturbation of WT and single-cysteine RGS8 protein NMR spectra
upon the addition of ligand CCG-203769 1H-15N HSQC NMR spectra of RGS8 were overlaid before
(red spectrum) and after the addition of its ligand CCG-203769 at 1:1 RGS8:ligand ratio (black
spectra) for (A) WT RGS8 (B) Cys107 RGS8, and (C) Cys160 RGS8. (D) The magnitude of chemical
shift perturbation. Spectra were acquired at 25 ℃ on a Bruker AVANCE III HD 800 MHz (WT and
Cys107 RGS8) or a Bruker AVANCE III HD 700 MHz (Cys160 RGS8) NMR spectrometers equipped
with Cryoprobes at the CUNY Advanced Science Research Center NMR facility.

74

Cys160 RGS8 is more sensitive to compound-induced denaturation than WT.

To determine effects of CCG-203769 on protein conformation mediated by the α7 cysteine,

Cys160 RGS8 was also exposed to compound in 1H-15N HSQC NMR studies. Some chemical shift

perturbations were observed at 1:1 and 1:2 molar ratios of Cys160 RGS8 to CCG-203769. However,

higher concentrations of compound resulted in signal loss at protein:ligand ratios above 1:2, most

likely due to protein denaturation. The loss of signal with increasing concentrations of CCG-

203769 precluded measurement of peak perturbations. (Fig. 4-3D) The decrease in Cys160 RGS8

stability in the presence of CCG-203769 compared to WT RGS8 is surprising, given that there is

only one cysteine at which the compound may act. However, this fits with data demonstrating

that Cys160 RGS8 is more potently inhibited than WT in functional inhibition studies with CCG-

203769 (Fig. 4-4) and with CCG-50014.115

Functional inhibition by CCG-203769 is altered in cysteine mutants

Previous studies have illustrated that manipulation of cysteines alters potency of inhi-

bition by covalent modifiers.

In RGS4, removal of individual cysteines leads to a decrease in

potency of inhibition by CCG-498631 and by CCG-50014.144 Interestingly, however, mutation of

RGS8 to the single cysteine Cys160 has been shown to cause an increase in the potency of CCG-

50014 compared to WT, while mutation to the single cysteine Cys107 RGS8 caused a decrease in

potency.133 This is consistent with the observation that Cys160 RGS8 is more prone to denaturation

in response to compound exposure in NMR studies.

To test whether CCG-203769 acting at individual cysteines inhibits RGS-Gα binding sim-

ilarly to CCG-50014, a flow cytometry-based protein-protein interaction assay was used. As ex-

pected, the Cys107 RGS8 mutant was minimally inhibited by CCG-203769, retaining 87% of Gα

75

Figure 4-4: Inhibition of RGS-Gα binding for WT, Cys160, and Cys107 RGS8 in response to increas-
ing concentrations of CCG-203769 was measured by FCPIA. WT IC50 = 25 μM), Cys160 IC50 = 2.2
μM, and Cys107 was not inhibited.

binding at 100 μM CCG-203769, while WT RGS8 only had 19% binding remaining at that con-

centration. Cys160 RGS8 showed only partial inhibition of Gα binding (retaining 52% Gα binding)

even at the highest CCG-203769 concentration used (Fig. 4-4). CCG-203769 inhibited Cys160 RGS8

with an increased potency (IC50 = 2.2 μM) compared to WT (IC50 = 25 μM), which is consistent

with that previously seen with CCG-50014.133 In addition, while Cys160 RGS8 was inhibited, Cys107

RGS8 was minimally affected even at the highest concentration of CCG-203769 (Fig. 4-4). This

suggests that Cys160 is more readily acted upon than Cys107.

CCG-203769 induces an intra-protein disulfide in WT RGS8.

No mass adduct was directly observed upon incubation of CCG-203769 with RGS8, so an

excess of IAA was used to label free cysteine thiols. In protein not treated with CCG-203769, IAA

caused a mass increase of 114.5 Da (2 times the mass of the acetamide adduct), indicating that

iodoacetamide accesses and forms an adduct at both cysteines. When protein was pretreated with

CCG-203769, this 114.5 Da mass increase was largely absent, indicating that CCG-203769 protects

76

Figure 4-5: CCG-203769 masks cysteine alkylation by IAA by inducing disulfide bond. (A) Decon-
voluted mass spectra of WT RGS8 (first column), Cys160 RGS8 (second column), and Cys107 RGS8
(third column). Spectra were taken before treatment (first row), after excess of of IAA (second
row), and pretreated with CCG-203769 before addition of IAA (third row). (B) WT, Cys160, and
Cys107 RGS8 mass changes analyzed by SDS-PAGE after treatment with vehicle, 250 μM CCG-
203769, or CCG-203769 followed by 1 mM DTT. Monomer mass indicated with black arrow and
dimer mass indicated with red arrow.

77

RGS8 from IAA alkylation. Surprisingly, however, a peak of the expected mass of a protein-

compound adduct with CCG-203769 was not detected. In fact, there was a 2 Da decrease from

16930.5 to 16928.5 (Fig. 4-5A). This suggests that two hydrogens were lost, which is consistent

with the formation of a disulfide bond between the two cysteines.

Mass accuracy in protein MS is only about 0.01%,159,160 so it is difficult to draw defini-

tive conclusions from a 2 Da loss in mass in a 17 kDa protein. Another method for detecting

disulfides is labeling of free cysteines with IAA; cysteines participating in a disulfide bond are

unavailable for alkylation. Pretreatment of RGS8 with CCG-203769 only partially protects the

protein from IAA alkylation (Fig. 4-5A). However, in the smaller population of protein with cys-

teines accessible to IAA, both cysteines were alkylated, with no population of protein having a

mass corresponding to a single alkylation (Fig. 4-5A). This “all-or-nothing” response to alkylation

of the two cysteines by IAA after CCG-203769 pretreatment is also consistent with induction of

a disulfide bond in the population of protein that was not modified by IAA.

Among single-cysteine RGS8 mutants (Cys107 and Cys160), CCG-203769 induces dimerization via an
inter-protein disulfide.

To determine how CCG-203769 may act differently on distinct cysteines in RGS8, the

single-cysteine mutant proteins Cys107 RGS8 and Cys160 RGS8 were also tested to determine

whether CCG-203769 protected individual cysteines from IAA adduct formation. As expected,

treatment of proteins with an excess of IAA caused an increase in mass of 57 Da in both Cys107

and Cys160 RGS8, consistent with a single alkylation at each mutant’s only cysteine. When each

single-cysteine protein was pretreated with CCG-203769, there was a decrease in the amount of

protein alkylated by IAA, but no corresponding increase in the mass of unmodified protein. In-

stead, a mass appeared that corresponds to two times the mass of the protein (Fig. 4-5A). This

78

suggests that compound may be inducing a covalently linked dimer.

To test whether the dimer-inducing effect of CCG-203769 is mediated by a disulfide bond,

protein treated with compound was analyzed by SDS-PAGE with reducing agent absent from the

sample buffer. As observed by mass spectrometry, addition of CCG-203769 caused a dimer-sized

mass in single-cysteine but not WT RGS8. This was reversed by addition of dithiothreitol to the

CCG-203769-treated Cys107 and Cys160 RGS8 (Fig. 4-5B), consistent with a dimer mediated by a

disulfide bond. A slight difference was observed between band positions of Cys107 and Cys160

RGS8 dimer masses (Fig. 4-5B). This is likely due to differently positioned disulfides altering the

shape of the denatured protein, resulting in a gel shift.

Both single-cysteine mutants were sensitive to compound-induced dimerization. How-

ever, the Cys107 RGS8 was only partially dimerized in response to CCG-203769 addition, while

a large population was still alkylated by IAA. Cys160 RGS8 pretreated with CCG-203769 had a

larger proportion of the dimer mass (Fig. 4-5A). This, combined with data indicating that Cys160

RGS8 is more susceptible to compound-induced denaturation, suggests that Cys160 is more read-

ily dimerized than Cys107. This also fits with data indicating that the α7 cysteine is more readily

alkylated than the α4 cysteine in RGS4 (Fig. 4-1).

CCG-203769 induces inter-protein disulfide in RGS4

To determine whether CCG-203769 exhibits disulfide-inducing behavior against other

RGS proteins, RGS4 and RGS19 were also tested for an increase in size mediated by CCG-203769.

RGS19 has only one cysteine, Cys123, the α4 cysteine analogous to Cys107 in RGS8. As antici-

pated, it behaves much like Cys107 RGS8; CCG-203769 induces a mass double the size of monomer

protein, and this is reversible by addition of 1 mM DTT (Fig. 4-6). Interestingly, RGS4 also formed

79

Figure 4-6: RGS4, RGS8, and RGS19 mass changes analyzed by SDS-PAGE after treatment with
vehicle, 250 μM CCG-203769, or CCG-203769 followed by 1 mM DTT.

a dimer after addition of CCG-203769, despite having multiple additional cysteines. These also

were reversible by DTT, indicating they are disulfide-mediated. Of the proteins tested, RGS4 was

the only protein with multiple cysteines that was susceptible to covalent dimer formation. It is

possible that this is mediated by one or more of the cysteines unique to RGS4, namely Cys71 on

α3 and Cys132 on α6.

Conclusions

This work demonstrates that both of the conserved cysteines in RGS proteins can play a

role in inhibition by CCG-203769. Both Cys107 and Cys160 RGS8 exhibited chemical shift pertur-

bations in response to CCG-203769 in 1H-15N HSQC studies. They were also prone to compound-

80

Figure 4-7: Proposed mechanism of disulfide bond induction by CCG-203769 in RGS8

induced formation of a disulfide-linked dimer. The α7 cysteine is more readily alkylated than

the α4 cysteine in RGS4 (Fig. 4-1) and mutant RGS8 containing only this cysteine (Cys160) was

more susceptible to both denaturation and inhibition of Gα binding than was Cys107 RGS8. This,

combined with earlier data showing that the α7 cysteine has more solvent exposed surface area in

MD simulations of RGS4 and RGS8,144 suggests that the α7 cysteine is more likely to be the site of

primary compound adduct formation than the α4 cysteine. Importantly, CCG-203769 was found

to be capable of inducing a disulfide bond between free cysteines, a mechanism not previously

known to be effected by these compounds.

In single-cysteine proteins, this effect resulted in

formation of a disulfide-linked dimer which likely causes the instability observed in Cys160 RGS8.

We propose a model in which CCG-203769 first forms an adduct at the more accessible α7

cysteine. The protein-compound adduct contains a disulfide bond which is then displaced by the

free thiol of the α4 cysteine (Fig. 4-7), leaving an internal disulfide bond between the α4 and α7

helices. In single-cysteine RGS proteins, a similar mechanism may mediate disulfide formation

with initial CCG-203769 adduct taking place on the only available cysteine, and the disulfide may

81

be formed by a thiol from a separate molecule.

Interestingly, these compounds have activity in cells and even in in vivo.112,117 This occurs

despite a reductive intracellular environment due to the presence of glutathione and other reduc-

ing agents.161–163 It is possible that in the cell, the disulfide-forming behavior of these compounds

is reversible. Compound-induced dimers between single-cysteine RGS proteins likely are an ar-

tifact of the in vitro environment, where the only free thiols available for disulfide formation are

on other RGS proteins. However, it is possible that an intra-protein disulfide between Cys107

and Cys160 in WT RGS8 is responsible for a conformational change in RGS8 that prevents the

RGS-Gα interaction, both in vitro and in living systems. Further work will be necessary to deter-

mine whether a disulfide bond mediates inhibition by CCG-203769 in the cellular environment,

how such a disulfide affects activity of other RGS proteins, and whether other cysteine-modifying

inhibitors may act by the same mechanism.

82

CHAPTER 5:

Identification of Transient Pockets in RGS4 and RGS19

Vincent Shaw performed pocket identification, analysis, and clustering. Hossein Mohammadi-
arani performed MD simulations and Mohammadjavad Mohammadi prepared trajectory files.

83

Introduction

Protein dynamics play a critical role in molecular recognition.3,164 Whether a ligand selects

a specific protein conformation that alters its function or induces a functional change by pushing

a protein into a unique conformation, proteins do not remain stationary in solution. The dynamic

motions of a target protein play a key role in the specificity of its ligand.7,165,166

In some cases, pockets that might be capable of binding small molecules are not visible

in available structures, but exist in conformations taken by the protein in solution.8,122 It may be

possible to design drugs that take advantage of transient pockets, which are not present in static

structures but are sampled by a protein in solution; or cryptic pockets, which become apparent

once a ligand is bound.50 There are several proteins with dynamically fluctuating pockets that

have been targeted using virtual screens informed by molecular dynamics.32,167

Heterotrimeric G-protein signaling is pathway of enormous pharmacological significance,

with a high proportion of known drugs targeting G-proteins or related pathways.12 Agonist bind-

ing to a GPCR results in dissociation of the GDP nucleotide of the G-protein alpha subunit. This

allows GTP to bind, putting the G-protein in its active conformation. The Gα and Gβγ subunits

dissociate, each mediating downstream signaling. A key part of this pathway is termination of

signaling, which is catalyzed by Regulators of G-Protein Signaling (RGS) proteins. These bind

to the active Gα subunits and accelerate hydrolysis of the GTP to GDP, allowing a return to the

inactive form and re-recruitment of Gβγ.

Because of their status as critical component of the G-protein cycle, RGS proteins make

attractive drug targets. There are 20 canonical RGS isoforms, with yet more proteins having

RGS homology domains. Each has different tissue distributions and diverse physiological roles.

Identification of inhibitors with high isoform specificity will permit targeting of certain pathways

84

and disease states with reduced off-target effects. While existing inhibitors that are selective for

RGS4 may be useful in treatment of Parkinson’s disease,109,117 there may be uses for targeting of

other RGS proteins as well. One potential target is RGS19, which has been implicated in pain

regulation102 and depression.101

Several series of RGS inhibitors have been already identified, and all act by covalent

modification.113,114 Interestingly, these inhibitors do show isoform specificity, but all are selective

for RGS4 and/or RGS1. Most likely, this specificity is largely due to differences in the number

of cysteines among RGS isoforms; both RGS4 and RGS1 have additional cysteines that are not

well-conserved among RGS proteins.118 While this inhibition of RGS4 or RGS1 may be desirable

for treatment of some disease states, it may make targeting other RGS proteins with a smaller cys-

teine complement impossible as long as we are limited to use of covalent modifiers. Discovery of

non-covalent RGS inhibitors may open the doors to drugs with novel specificities in addition to

reducing any toxicity risk associated with the use of covalent drugs.

Previously, we identified a role for protein dynamics in the specificity of inhibitors acting

at a cysteine on the α4 helix.129,144,146 MD studies from this work suggested that the structure

of the RGS domain may open sufficiently to allow covalent inhibitors to access this otherwise

buried cysteine. Also, differences in flexibility among isoforms play a role in driving inhibitor

selectivity.144 If the apo-protein forms a binding pocket with sufficient frequency and druggability,

it may be exploited in the design of non-covalent inhibitors. In this work, we seek to identify

transient binding pockets in RGS proteins. This will permit discovery of new compounds by

virtual screening.

85

Approach and Results

Pocket Identification

Previous studies have showed flexibility in RGS4 and and RGS19.144 In MD simulations,

RGS4 showed a pronounced movement in the α6 helix, in which it partially lost its helical structure

and moved away from the helical bundle. RGS19 also showed a dramatic movement, in which the

α6-α7 loop separates from the α4 and α5 helices, creating a groove.144 In either case, compound

access may be permitted by the development of a transient pocket. If a pocket conformation can

be identified that is both amenable to small-molecule binding and frequently occurring, it will be

useful in the rational design of non-covalently binding inhibitors.

Pockets were found using MDpocket,168 a part of the Fpocket suite of pocket-finder

software.169 Fpocket defines pockets by filling cavities with alpha spheres: spheres with an

external boundary touching four atoms, with no atoms inside them.170 Sphere radii are restricted

to between 3 and 6 Å to prevent large spaces (like external surfaces) or small spaces from being

included as part of pockets.169

In RGS4, frequent pockets were found to occur between the α5 and α6 helices (Fig. 5-1).

This was expected, considering the flexibility and movement previously observed in α6 of RGS4.144

In RGS19, MDpocket most frequently identified pockets formed by atoms in the α6-α7 loop (Fig. 5-

1). This makes sense, considering the groove formed where the α6-α7 loop separates from the rest

of the α4-α7 helix bundle.144 Because pockets were identified with the highest frequency on these

parts of the structures, these locations were used for extracting descriptors of pocket shape and

characteristics, clustering of pocket conformations, and choosing a state for virtual screening.

Descriptors of the pocket for each frame of the trajectory were generated by MDpocket.

86

Figure 5-1: Locations of pocket-forming residues in RGS4 (top) and RGS19 (bottom). Color indi-
cates frequency with which each atom touches a pocket alpha sphere. Blue is less frequent and
red is more frequent.

87

Figure 5-2: Pocket volume and mean local hydrophobic density (MLHD) plotted over the simu-
lation trajectory for RGS19 (A) and RGS4 (B). Pockets in RGS19 were larger and more frequent
than those in RGS4.

By plotting these parameters over time, trends in the pocket’s size, shape, and druggable potential

can be identified.

Pocket Clustering

To use a conformation from an MD simulation in a virtual screen, we wanted to choose

a conformation that was both druggable (i.e. amenable to small molecule binding) and represen-

tative of a frequently occurring state. One option is to choose a state that has the highest mean

local hydrophobic density, an index closely correlated with a pocket’s druggability.168,171 Although

there is a precedent for a successful virtual screen being performed with such a strategy,167 it may

not be ideal for choosing a transient pocket from a long time scale simulation. Even if a compound

88

were identified that had a favorable energy of binding in the static structure of the most drug-

gable conformation sampled in a long time scale simulation, if that state was very rarely sampled

by the protein in solution it might have a low on-rate and therefore low affinity. Therefore, we

clustered pocket states to identify populations of distinct conformations taken by the protein in

solution. By choosing states that are most similar to cluster centers, we identify states that are

most representative of a conformational group, ensuring use of a state that is druggable but not

anomalous.

Clustering of pocket states was performed using R and Rstudio. First, pocket descriptors

were normalized to equal scales to avoid uneven weighting of pocket parameters. The pocket

states were clustered based on seven pocket descriptors: pocket volume, nonpolar surface area,

polar surface area, number of alpha spheres, average alpha sphere radius, maximum distance

between any two alpha spheres, and mean local hydrophobic density. These descriptors were

chosen to separate states based on their size, shape, and complexity. Mean local hydrophobic

density is a measurement of how densely packed hydrophobic areas are, based on the degree

with which nonpolar alpha spheres overlap one another. This descriptor has been found to be

closely related to druggability.168,171

The NbClust package172 was used to determine the optimal number of clusters. In order

to limit the number of groups of pocket states, the cluster numbers were limited to between 5

and 10. In the RGS4 trajectory, the optimal number of clusters was 6 according to a plurality

of indices, while in RGS19, the optimal number of clusters was 7. States were clustered using

kmeans clustering.

89

Figure 5-3: Clustering of pocket states for RGS19 (A) and RGS4 (B). Volume is plotted against
MLHD, and color indicates distinct clusters. An ensemble of pockets representing clusters with
high MLHD and a variety of pocket volumes were selected for structure based screening.

Frames for screening

For structure-based screening, it will be beneficial to use states that are amenable to small

molecule binding. Because mean local hydrophobic density has been found to be a strong indica-

tor of small-molecule binding potential,168,171 clusters that are high in this index were chosen. In

addition, pocket volume may play a role in which compounds may bind. The volumes of pockets

in a large set of ligand binding proteins in the protein data bank has been found to cover a wide

variety of pocket volumes.173 The median volume was 536 Å3, with first and third quartiles at 375

and 715 Å3 respectively,173 suggesting that pockets that are excessively large or small may not

make ideal drug targets. As such, the pocket populations that form clusters 4 and 7 in RGS19 and

clusters 2 and 6 in RGS4, which have high MLHDs and differing but moderate pocket volumes,

may make the most promising populations of pockets for virtual screening.

States most similar to the cluster center will best represent the actual pocket conformation

in solution. The transient pocket conformations shown in Figure 5-4 are the trajectory frames

most similar to cluster centers 2 and 6 in RGS4 and 4 and 7 in RGS19. These states will go on to

90

be used for virtual screening to identify non-covalent compounds that bind to RGS4 and RGS19.

One limitation of virtual screening methodology is usage of static structures, where the

conformation used may not be representative of those occurring in solution. The approach used

in this work partially addresses this problem by clustering populations of similar pocket-like

states, and choosing only the most representative conformation of each cluster for use in virtual

screening. However, compound docking methods still often use a static protein structure. Even

if the conformation used realistically occurs in solution, effects of the compound binding on the

protein structure are not accounted for. This may bias screening results toward compounds that

bind in an conformational selection mode rather than an induced fit mode, resulting in a smaller

proportion of hits that validate experimentally. It may be possible to resolve this issue by using

flexible docking, in which protein flexibility is taken into account. This may improve the quality

of hits and likelihood of successful identification of new chemical matter by virtual screen.

Conclusions

This work identifies transient pockets between α5 and α6 helices of RGS4 and RGS19. This

will enable identification of compounds that bind non-covalently by virtual screen. Rather than

consideration of static structures only, which may not be representative of conformations taken

by a protein in solution, this work chooses druggable, transient pockets that are representative

of frequently occurring conformations, and may be useful in the discovery of non-covalent com-

pounds.

91

Figure 5-4: Pocket states that are representative of cluster 4 and 7 in RGS19 (A and B) and cluster
2 and 6 in RGS4 (C and D). Pocket-forming atoms illustrated with white surface.

92

CHAPTER 6:

Conclusions and Future Directions

93

In this thesis, I aimed to understand the drivers of selectivity of covalent modifiers of RGS

proteins. This led to the pursuit of two main hypotheses: that RGS inhibitor isoform specificity

is determined by cysteine complement, and that RGS inhibitor isoform specificity is determined

by protein flexibility. While these hypotheses at first appear to be in conflict, this body of work

demonstrates that both are essential pieces of the full picture of isoform selectivity.

Role of individual cysteines in action of inhibitors

A primary determinant of selectivity among covalent modifiers of RGS proteins is the

number and location of cysteines in the RGS domain. Previous work has shown that conserved

cysteines on the α4 and α7 helices each are capable of mediating inhibition by CCG-50014.115 In

Chapter 4, I propose that the TDZD CCG-203769 induces a disulfide bond between these two

cysteines. NMR studies have demonstrated that CCG-203769 perturbs protein structure in both

single-cysteine mutants of RGS8, indicating it can act at either cysteine. Mass spectrometry stud-

ies indicated that CCG-203769 could induce a dimer sized mass among single-cysteine RGS8. This

was reversible by DTT, indicating it is disulfide mediated. Interestingly, without forming any ap-

parent adduct in WT RGS8, CCG-203769 prevented iodoacetamide alkylation and induced a 2 Da

reduction in mass, suggesting it may be inducing a disulfide between cysteines. This is a unique

covalent interaction between the two well-conserved cysteines present in RGS8 that is induced

by the inhibitor.

Role of protein dynamics in RGS inhibitor selectivity

In the course of this work, it also became apparent that there were other drivers of selectiv-

ity beyond the cysteine complement, namely protein flexibility. While the effects of differences

94

in flexibility on potency of inhibition may be largely masked by differences in cysteine comple-

ment among covalent inhibitors, it will be particularly important to understand protein dynamics

in order to identify inhibitors that act non-covalently. By eliminating the reliance on covalent

interactions, it may be possible to target RGS proteins that have fewer cysteines.

In Chapter 2, I demonstrate a correlation between protein flexibility and potency of the

TDZD CCG-50014 when it acts at a single, shared cysteine. Among RGS4, RGS8, and RGS19 mu-

tants that contain only the shared α4 cysteine, RGS19 is most potently inhibited (IC50 of 1.1 μM),

followed by RGS4 and RGS8 (IC50s of 8.5 μM and >100 μM respectively). When solvent exposure

was measured by deuterium exchange, RGS19 was found to have faster deuterium exchange in

the α4 helix, followed by RGS4 and RGS8. In addition, MD simulations shed light on movements

that may lead to these differences in solvent exposure, and how cysteines may may become ac-

cessible covalent inhibitors. In RGS19, MD simulations showed movement throughout the α4-α7

helix bundle, opening a cleft between α4-α5 and α6-α7 helices. This was supported by HDX data

showing high deuterium incorporation throughout this helix bundle in RGS19. Likewise, an α7

helical cysteine in RGS4 was exposed by an outward movement of the α6 helix in simulations,

which matched the high deuterium exchange observed in this helix. Finally, RGS8 was least flexi-

ble in MD simulations and had least deuterium exchange, which may explain the limited potency

with which it is inhibited.

While this work identified a correlative relationship between protein flexibility and po-

tency of inhibition, we wanted to go beyond correlation and demonstrate that direct manipula-

tion of flexibility could induce changes in potency of inhibition. To this end, we sought to identify

interacting residues within RGS proteins that are responsible for differences in flexibility among

isoforms. Of particular interest is a salt bridge network linking the α4 helix, the α5-α6 interhe-

95

lical loop, and the α5 helix. This network is shared by RGS4 and RGS8, but absent in RGS19,

which is lacks a charged residue on α4, having instead a leucine. This lack of a salt bridge may

be responsible for the observed flexibility of RGS19. Mutation in this residue in RGS19 to add

a salt bridge-forming residue (aspartate) increased thermal stability, reduced deuterium incorpo-

ration, and, importantly, decreased potency of inhibition by CCG-50014. Conversely, mutation

to remove the salt bridge in RGS4 and RGS8 increased deuterium incorporation and increased

the potency of CCG-50014. This strongly supports a causative relationship between RGS protein

flexibility and potency of inhibition.

Access of inhibitors to buried cysteines hinted at the existence of transient pockets. Pock-

ets were identified from MD simulations of RGS4 and RGS19. In RGS4 a pocket opens between

the α5 and α6 helices, and in RGS19, a cleft opened between the α4-α5 and α6-α7 sets of helices. In

order to move forward with virtual screening targeting these pockets, conformations were clus-

tered to ensure selection of a structure that was both druggable and representative of frequently

occurring states. Identification of transient pockets will enable rational design of non-covalent

inhibitors by structure-based screening.

Future research in understanding action of TDZD inhibitors

While some questions on the mode of compound action and the basis for specificity have

been answered, new avenues for future research have opened. One area is in better understanding

the role of the disulfide bonding induced by CCG-203769. While this may occur in RGS proteins in

these assay conditions, it remains to be understood how readily such a change can affect G-protein

binding and GAP activity, both in biochemical assays as well as in the cellular environment.

Experimental efforts to understand the molecular mechanism of the interaction between

96

RGS proteins and TDZD inhibitors is being undertaken in collaboration with Dr. Krisztina Varga

at the University of New Hampshire. While we already have demonstrated that HSQC spectra

in RGS proteins are perturbed by CCG-203769, it will be useful to understand which parts of

the protein are altered in response to inhibitor action. To this end, I have produced uniformly

labeled 13C 15N-RGS8, and efforts are already underway using these samples to assign HSQC

spectra peaks to individual amides on the protein backbone.

One open question that remains is whether the induction of a disulfide bond by CCG-

203769 is sufficient to prevent binding between RGS proteins and Gα. We would hypothesize

that peaks corresponding to amides of residues near cysteines 107 and 160 in RGS will be per-

turbed, but it will be interesting to see if there are also peaks perturbed corresponding to residues

involved in Gα binding. MD simulations also provide a useful avenue for answering these ques-

tions, work which is currently being carried forward by the Vashisth lab at the University of

New Hampshire. In particular, simulations that illustrate the effect of compound adduct at each

cysteine in RGS8 will be valuable for comparison with the HSQC NMR studies of single cysteine

mutants. Simulation work may also be able to shed light on how protein conformation may be

altered upon induction of a disulfide bond.

It would be interesting to see whether the compound-induced disulfide in RGS8 can occur

in cells. It may be possible to test whether RGS8 protein expressed in mammalian cells treated

with CCG-203769 are also protected from IAA alkylation, and whether this is reversible by DTT.

Finally, another open question is whether the disulfide bond-inducing behavior of CCG-203769

is relevant to other TDZDs, other covalent inhibitors, and among different RGS proteins. For

example, RGS19 is inhibited by CCG-203769 and other TDZDs, but because it lacks a cysteine on

the α7 helix, this cannot be mediated by an intraprotein disulfide. It remains to be seen whether

97

an adduct between a TDZD and a single-cysteine RGS proteins such as RZ family member is

maintained or displaced by a free thiol, in both in vitro and in cell environments.

Continuing discovery of non-covalent inhibitors

Research should continue in discovery of non-covalent inhibitors. In collaboration with

the Dickson Lab, virtual screening efforts are under way using the transient pockets described

in Chapter 5. Using a pharmacophore-based screening campaign, a library of compounds will

be extracted from the Zinc library that block interactions between residues that make contact in

the closed state but are separated in the open-like state. These compounds will then be docked

against the open states in a structure-based screen using Schrödinger Glide. Because docking

against a static structure does not account for protein movement induced by compound docking,

this may bias discovery against compounds that bind in an induced-fit-like mode. To combat

this, hits may be further refined with flexible docking, in which movement of both protein and

compound are simulated during binding. Finally, hit compounds will be ordered or synthesized

and tested for activity in inhibition of Gα binding or inhibition of GAP activity. This work may

yield new non-covalent inhibitors that take advantage of transient pockets in RGS proteins that

we have defined here.

In conclusion, this work shows a dual role for number of cysteines and protein dynamics

in specificity of RGS protein activity. Understanding drivers of RGS protein selectivity will allow

future researchers to better predict the action of current inhibitors as well as develop chemical

matter with new specificities. By laying out a path forward for targeting a transient pocket in

RGS proteins, we may be able to break the cysteine dependence of RGS inhibitors, allowing novel

selectivities and opening the doors to new applications as chemical probes or therapeutics.

98

APPENDIX

99

Interpreting Hydrogen-Deuterium Exchange Events in Proteins Using Atomistic

Simulations: Case Studies on Regulators of G-protein Signaling Proteins

Reprinted with permission from J. Phys. Chem. B 2018, 122, 40, 9314-9323

Copyright 2018 American Chemical Society

Hossein Mohammadiarani*, Vincent Shaw*, Richard R Neubig, Harish Vashisth

*Co-first authors
H.M. performed MD simulations and developed computational models. V.S. expressed protein
and performed HDX-MS.

100

Introduction

Hydrogen-deuterium exchange (HDX) is a widely used protein labeling reaction in which

an amide hydrogen in the backbone of amino acids in proteins is exchanged with a deuterium

atom. To probe the locations of exchanged hydrogens in the protein backbone, HDX is often ac-

companied by other techniques including nuclear magnetic resonance (NMR) spectroscopy and

mass-spectrometry (MS).174 HDX methodologies have been successfully applied to understand

protein-protein interactions,175–177 conformational changes in proteins,178–182 protein folding,180

and ligand binding.183,184 Early applications of HDX on the A-chain of hormone insulin showed

that intramolecular hydrogen bonds were a hindrance for hydrogen exchange because of their

role in stabilization of the helical structure.185 Since then many investigations have been con-

ducted to characterize the mechanism of exchange events. These include studies of: deuterium

exchange of poly-DL-alanine in aqueous solution at different temperatures and pH,186,187 the influ-

ence of residue side chains on the HDX rate of peptide groups,188 modeling amides and peptides

in a chemical exchange step,189–191 development of empirical rules for acid and base catalytic

rate constants,188,192 development of general models for recognizing hydrogen exchange process

between the folded states and the unfolded states using temperature variation,193–197 the nega-

tive effect of static solvent accessibility on exchanging protons,198 and the correlation between

apparent adiabatic compressibility and hydrogen exchange rates.199 Bai et al.200 carried out experi-

ments to formulate inductive and steric blocking effects of neighboring amino acids on the amide

group hydrogen exchange. Their comprehensive dipeptide models included all 20 amino acids

and have informed values of intrinsic kinetic rates used in many previous studies.130,131,201 The

qualitative and quantitative interpretation of HDX events is becoming an increasingly important

tool for studying dynamics in proteins which are challenging to study using other experimen-

101

Figure A-1: Kinetic scheme for HDX is highlighted. A conformational fluctuation in the protein
exposes buried amide groups (blue) (closed state) to solvent (open state) where amide hydrogens
(white) are exchanged by deuterium (yellow) with an intrinsic rate constant kint.

tal methods.130,202–204 These investigations, over the past half-century, have resulted in various

interpretations of the HDX mechanism204–207 primarily via different models used to rationalize

exchange events.130,131,200,208–212 The general mechanism of HDX is described by a dynamic equilib-

rium between closed and open states (Figure A-1) of amide hydrogens with rate constants kc and

ko, respectively, and a first order reaction in the exchange competent or open state130 (denoted

as an intrinsic rate constant, kint, in Figure A-1). The normal exchange mode for proteins that do
not undergo global unfolding events is the EX2 exchange limit, in which kc ≫ kint.191 This mecha-
nism suggests that steric hindrance protects amide hydrogens from exchanging with deuterium.

In addition to the physical protection, amide hydrogens that are involved in hydrogen-bonded (H-

bonded) structures are protected and show decreased exchange rates.205,207,213,214 Therefore, HDX

rates implicitly involve structural changes and dynamics in proteins.130 A variety of models have

been used to determine protein conformational states using Monte Carlo (MC)208,215 or molecular

dynamics (MD)130,131,201,209,210,212,216–224 approaches.

In these models, the protection factor (PF) (ranging between 0 and 1010) is a key parameter

that correlates conformational dynamics in proteins with the overall HDX rate (khdx).225 In Table

102

A-1, we summarize various PF correlations for seven different models (M1 through M7) that have

been proposed previously; detailed descriptions of these models are provided in the supplemen-

tal introduction. The parameters and criteria in PF correlations can be tuned either using MD

simulations210 or using structures refined from experiments (e.g. the NMR method). There are

two general approaches to obtain the PFs for amide hydrogens by sampling conformations using

simulation methods.

In the first approach, PFs empirically correlate to metrics of the protein

structure (e.g. models M1 to M6 in Table A-1). In the second approach,131 the PF is defined as a

fractional population of the closed state to the open state for each amide hydrogen (e.g. model

M7 Table A-1). As a complement to HDX experiments, MD simulations not only provide details

on exchanging amide hydrogens, but also capture frequencies of open states which may occur on

a much shorter time scale than the hydrogen exchange itself.131,210 As it remains challenging to

conduct long time-scale atomistic MD simulations, the modeling of hydrogen exchange using MD

simulations has generally been limited to coarse-grained and/or empirical models with implicit

solvent.208,217,226 Several studies have employed short time scale MD simulations to predict HDX

rates.130,131,201,227 To date, only Persson et al.131 used a millisecond long MD simulation228 for HDX

analysis of a 58-residue protein, bovine pancreatic trypsin inhibitor (BPTI). They suggest that the

mean residence times for the open states of all amides in BPTI are on the sub 100 ps time-scale.

However, the ability of existing models of PF correlations (Table A-1) to predict HDX

trends, when applied to identical experimental dataset(s), is yet to be systematically analyzed.

Furthermore, it would be useful to determine whether any of the existing models (based upon

their default or reoptimized parameters) can faithfully distinguish differences in HDX patterns

of homologous proteins. Finally, comparing the predictive performance of various models for

widely used interatomic potentials (force-fields) for proteins (e.g. CHARMM and AMBER) will

103

criteria
5
4

6

7

8

✓

3

✓

✓

Model
2
1
M1ref. 225 ✓ ✓
M2ref. 208 ✓
M3ref. 209 ✓
M4ref. 210
M4ref. 212 ✓
M6ref. 130 ✓
M7ref. 131
M8†
M9†
the vicinity; 5RMSF; 6# of waters in the vicinity; 7polar atoms in the vicinity; 8SASA; †new

Protection Factor Definitions
log(P Fi) = u · (SAi) + v/(HBi)
ln(P Fi) = (βcN C
ln(P Fi) = (βcN C
ln(P Fi) = (βcN C
P Fi = (CoN H sol
P Fi = base/(1 + (
P Fi = τC/τO
✓ ✓ ln(P Fi) = (βsSASA
✓ ✓ P Fi = τC/τO

base)1−N Hstati
−γp
−γs
i + βpD
i

1Hydrogen bond; 2Distance from the surface; 3# of residues in the vicinity; 4# of heavy atoms in

i + βhN h
i )
i )
i + βhN h
i )−1)
i + βr(N r
i + CcN H β
i )/CN H sol

√

i

✓ ✓

✓

model proposed in this work.

Table A-1: Model definitions and corresponding metrics. Among models reported in the literature
are models M1 through M6 (empirical models) and the model M7 (a fractional population model).
For models reported in this work, M8 is an empirical model and M9 is a fractional population
model. Additional details on models M8 and M9 are provided in supporting information.

likely provide further guidance for future studies combining MD simulations and HDX experi-

ments. In this work, we have investigated these issues by conducting a series of atomistic MD

simulations of three homologous regulators of G-protein signaling (RGS) proteins (RGS4, RGS8,

and RGS19) (Fig. A-2) using CHARMM and AMBER force-fields (CHARMM-FF and AMBER-FF).

We compared the predictive performance of seven existing models (Table A-1 with our recently

reported HDX-MS data for all three proteins,144 and reoptimized parameters of these existing

models for improved predictions. We also found solvent accessible surface area (SASA) as a use-

ful metric to better predict protection factors in combination with the open-state definition of

Persson et al.131 This was surprising because some existing models have reported SASA as a poor

predictor. Based upon this latter observation, we derived two new models (M8 and M9; see sup-

plemental methods and Table A-3, A-4) for better reproducing our experimentally observed HDX

trends in three RGS proteins.

104

Figure A-2: Sequence and structural views of RGS proteins. (A) Sequence alignment of RGS4,
RGS8, and RGS19 is shown with conserved residues highlighted in red; blue boxes indicate
residues that are conserved between at least two among three RGS proteins. (B) Shown are front
and back views of the overlay of RGS4 (PDB code 1AGR), RGS8 (PDB code 2ODE), and RGS19
(PDB code 1CMZ) structures with each of the nine helices uniquely colored. Regions rendered as
white cartoons are interhelical loops.

105

Materials and Methods

We carried out all MD simulation trajectories and their analyses using NAMD and VMD

software suite136,137 as well as python,229 and used both the CHARMM36 force-field with the

CMAP correction138,139 and the AMBER force-field (ff14SB).230 For all MD trajectories, 50000

frames were generated for each μs of dynamics. For RGS4 and RGS8, simulations were con-

ducted with two different initial coordinates, while for RGS19 only one experimental structure

is currently known, the coordinates of which were used in simulations. In particular, the initial

coordinates for RGS4, RGS8, and RGS19, respectively, were taken from the following protein

data bank entries (RGS4: 1AGR and 1EZT; RGS8: 2IHD, 2ODE; RGS19: 1CMZ). Each protein was

initially modeled using the psfgen tool in VMD, and then further solvated in a simulation box

(~65 Å x ~70 Å x ~65 Å) of TIP3P water molecules and charge-neutralized with NaCl. All system

sizes are provided in Table A-2 The volume of simulation domains was then optimized in the NPT

ensemble by initially applying 500 cycles of a conjugate-gradient minimization scheme followed

by a short 40 ps MD run with a 2 fs time step in which the temperature was controlled at 310K

using the Langevin thermostat and the pressure was controlled by the Nose-Hoover barostat.

We carried out all simulations using periodic boundary conditions. These briefly equilibrated

systems of all RGS proteins were further subjected to long time scale (2 μs for each protein) MD

simulations in the NVT ensemble. For all proteins and both force-fields, we generated 10 total

MD simulations with 20 μs of MD simulation data (Table A-2). All details on protein expression,

purification, and data collection using HDX-MS are provided in our previous work.144 Briefly,

deuterium incorporation (DI) for RGS4, RGS8 and RGS19 was measured at a fragment resolution

using HDX-MS experiments at t = 1, 3, 10, 30, 100, 300, and 1000 minutes (Fig. A-7 and A-8).144

We note that incubations were carried out in a 90% D2O solution containing 5 mM HEPES and

106

100 mM NaCl. We provide further description of protocols for HDX modeling in supplemental

methods.

Results and Discussion

Comparison of predicted and experimentally-observed deuterium incorporation trends

for RGS4, RGS8, and RGS19: To evaluate the predictive performance of various existing models

for PF correlations (see Table A-2 and Model Details), we conducted 10 independent all-atom,

explicit-solvent, and μs-timescale MD simulations for all RGS proteins (Table A-2 and supple-

mental methods). For each 2 μs timescale simulation, we analyzed 100,000 conformations of each

protein by applying criteria reported previously for each model (Table A-4) and combined cal-

culations on those metrics to obtain protection factors (PFs) for each residue. These PFs, when

combined with the intrinsic exchange rates,200 were then used to predict and compare the per-

centage of deuterium incorporation (%DI) at t= 0, 3, 10, 30, 100, 300, and 1000 minutes for each

experimentally observed fragment of each protein (Fig. A-7 and A-8).144 Then, we reoptimized

parameters of models M1 through M7 (the reoptimized models hereafter are referred to as M1*

through M7*) by minimizing a fragment-based objective function that compares the predicted

and measured values of DI (see supplemental methods). The reoptimization procedure was car-

ried out for simulations conducted with both force-fields (CHARMM-FF and AMBER-FF). The

default as well as re-optimized parameters of all 9 models are listed in Table A-4.

We quantified the comparisons between the predicted and experimentally measured deu-

terium incorporation (%DI) using the relative error (E) and correlation-coefficient (CC) analyses.

E measures the discrepancy between the exact values of DI that were measured via HDX-MS

experiments and the values that were calculated from MD simulations. However, CC measures

107

the linear relationship between the measured DI and the modeled DI. It is a measurement of the

interdependence or association of two variables and ranges between -1 (negative correlation) and

1 (positive correlation). Therefore, both E and CC are taken into account for the evaluation of

each model. In Fig. A-3 and Fig. A-4, we present the statistics of performance of each model via

calculations on E and CC for the CHARMM-FF and the AMBER-FF. Specifically, Fig. A-3 shows

the performance metrics computed by averaging over data from all MD simulations of all RGS

proteins (RGS4, RGS8, and RGS19), while Figure 4 shows the same metrics computed by averag-

ing over all MD simulations of each RGS protein. For additional details, we show the traces of the

predicted vs. measured %DI for all fragments of each RGS protein for both force-fields (Figure S3

to Figure S32).

For discussion in the following, we refer to models M1 through M6 as empirical models,

and the model M7 as a fractional population model (see supplemental introduction). Overall, we

observe that the models M1 through M6 show larger errors and lower correlations in comparison

to the model M7 for simulations with both force-fields (gray bars in Fig. A-3). Among empirical

models, the model M6 has the smallest error for simulations with the CHARMM-FF (Fig. A-3A),

while the model M4 has the smallest error for simulations with the AMBER-FF (Fig. A-3B). The

CC values are comparable for the model M6 in the CHARMM-FF and for the model M4 in the

AMBER-FF. After re-optimizing the parameters for these models (see supplemental methods and

Table A-4), the models M1* and M2* showed significant improvement (lower E and higher CC)

for both force-fields in comparison to other models (M3* to M6*), that only moderately improved

(blue bars in Fig. A-3). After the reoptimization, even though the E values for the model M7*

marginally decreased in comparison to the model M7 (with default parameters), the CC values

are similar in both force-fields. The E and CC values for our proposed models (M8 and M9), both

108

√∑
(xi − ¯x)2

Figure A-3: Comparisons of model predictions of HDX-MS data across all three RGS proteins.
Performance metrics (relative error, E, and correlation coefficient, CC) for different models are
∑
∑
shown based upon data averaged from all trajectories of RGS4, RGS8, and RGS19 conducted with
∑
the CHARMM-FF (data in panels A and B) and the AMBER-FF (data in panels C and D). (A, C) The
i=0 |xi − yi|/
i=0 yi]. (B,
relative error between the predicted and observed %DI [E(x, y) =
∑
(xi − ¯x)(yi −
D) Correlation coefficient between the predicted and observed %DI [CC(x, y) =
(yi − ¯y)2]. Gray bars are for models with the default parameters reported in
¯y)/
the literature, blue bars are their re-optimized versions based upon our experimental data, and
red bars are for new models proposed in this work. No performance data for the original model
M5 are reported because the parameter values were not available from the original work,42 but
the performance data are reported for the optimized version of this model (M5*) based upon our
experimental data.

n

n

109

Figure A-4: Comparisons of model predictions of HDX-MS data for each RGS protein. The defi-
nitions of E and CC, and other details are the same as in Figure 3. Colored bars distinguish data
for each RGS protein: black bars, RGS4; blue bars, RGS8; and magenta bars, RGS19

of which are based on the SASA of each amide hydrogen and its distance from the first polar atom

(see supplemental methods), show results comparable to the fractional population model M7 and

its reoptimized version M7*. Both of our proposed models consistently predict DI trends with

lower E values and higher CC values for both force-fields. Taken together, these data suggest

that the proposed models M8 and M9 as well as the models M7 and M7* predict experimentally

observed HDX trends better than the other models (M1/M1* through M6/M6*).

On comparing the performance of all empirical models for each RGS protein (Fig. A-4),

we observe that the DI trends in RGS4 and RGS8 for the CHARMM-FF are best described (lower E

and higher CC values) by the model M6, and for the AMBER-FF are best described by the model

M4 (for RGS4) and equally well described by the models 4 and 6 (for RGS8). For RGS19, the

110

model M1 captures DI trends better than other empirical models (M2 through M6) for both force-

fields, but this model is a poor predictor for RGS4 and RGS8. We also observe that the model M2

poorly predicts DI trends (higher E and lower CC values) for all three proteins, and the model

M7, a fractional population model, consistently shows better predictions (lower E and higher CC

values) for both force-fields. On re-optimizing, all empirical models (M1* through M6*) show

improvement (lower E and higher CC values) over their default parameter versions (M1 through

M6), but both versions of the fractional population model (M7 and M7*) provide consistently

better predictions than the empirical models. The performance of our proposed models M8 and

M9 is comparable to the model M7*, but for all three models (M7*, M8, and M9) the performance

is marginally poorer (i.e. E values are marginally higher and CC values marginally lower) for

RGS19 in comparison to RGS4 and RGS8.

The time-dependence of model predictions contributes significantly to differences in the

ability of each model to predict HDX-DI results for each experimentally observed fragment (24

fragments for RGS4, 38 fragments for RGS8, and 26 fragments for RGS19; Fig. A-8).144 The models

show significant variation between shorter time points (t= 0, 3, 10, 30, and 100 minutes) and

longer time points (t= 300 and 1000 minutes) when comparing predicted DI trends at the level

of individual fragments for both force-fields (Fig. A-9 to Fig. A-38). For example, models M3,

M4, and M6 under-predicted experimentally observed DI trends at shorter time points, but the

trends at longer time points are predicted reasonably well (Fig. A-24 and A-25). Similarly, the

re-optimized models including M2* through M6* under-predicted DI trends at shorter time points

for RGS4 simulations (Fig. A-30 and A-35). Unlike these models, our proposed models M8 and

M9 overall show better agreement with the HDX data across all time points and fragments for

RGS4 and RGS8 with both force-fields (Fig. A-19 to A-22 and A-34 to A-37). However, for RGS19,

111

except fragments 18 to 26, each model under-predicts DI trends for both force-fields (Fig. A-23

and A-38).

Our HDX-MS data showed that the amide hydrogens exchanged rapidly in RGS19 in com-

parison to RGS4 and RGS8 (Fig. A-7), especially in helices α4, α5, and α6 (fragments 10 to 23; Fig.

A-8).144 At t= 1000 minutes and for models M7, M8, and M9, the mapping of the predicted vs. mea-

sured DI on protein structures (Fig. A-39) shows that these models under-predicted DI trends in

the α4 helix of RGS19, but predicted well in the α6 helix as well as in the α5-α6/α6-α7 interhelical

loops. Importantly, the structural motifs in RGS proteins that showed poor agreement between

the predicted and measured DI trends also showed significantly lower residue fluctuations in MD

simulations (Fig. A-40) in comparison to those motifs that showed higher fluctuations and as a

result better agreement with the experiments.

In summary, each model has unique metrics for estimating the PFs and some of these

metrics are shared among different models. For example, the number of polar atoms or residues in

the vicinity of an amide hydrogen indirectly assess the likelihood of existence of hydrogen bonds

between amide hydrogens and other atoms in proteins. Therefore, different models are directly

or indirectly correlated to hydrogen bonds. Our analyses show that the fractional population

modeling (e.g. models M7/M7* and M9) is more robust than empirical approaches. In particular,

the fractional population models are broadly applicable to newer systems without reoptimization

of parameters (e.g. the model M7 makes reasonably accurate predictions both before or after

optimization). In our new models (M8 and M9), combining two metrics, SASA and the number

of polar protein atoms in the vicinity of a given amide hydrogen, shows better predictions both

for the empirical model (M8) and the fractional population model (M9). We also suggest that our

new models are potentially applicable to other protein systems for efficient interpretation of HDX

112

data because these models only require coordinates of the protein atoms. These can be readily

extracted from the solvated simulation trajectories for rapid analysis.

Comparison of predicted and measured HDX data at a single-residue resolution

Our HDX-MS data was collected at a fragment resolution for each protein (Fig. A-7 and

A-8),144 but atomistic MD simulations complement these data by providing additional details on

the protections of amide hydrogens at a single-residue resolution. At t = 1000 minutes for mod-

els M7, M7*, M8, and M9, we show in Fig. A-41 to A-46 a color-coded mapping of DI trends for

each residue of RGS4, RGS8, and RGS19 for both force-fields. These data show that the amide

hydrogens in the N-terminus of the α3 helix (containing 12 residues; see Fig. A-2) are fully ex-

changed and some residues are partially exchanged. MD simulations show that the unexchanged

or partially exchanged amide hydrogens are participating in hydrogen bonds and are therefore

largely protected. Consistent with HDX experiments, these protection effects are observed in

fragments 2 and 3 in RGS4 (Fig. A-29), fragments 8, 10, 11 in RGS8 (Fig. A-31), and the fragment

6 in RGS19 (Fig. A-33). In HDX-MS experiments, we observed that the residues in the N-terminus

of the α4-helix show high exchange propensity in all RGS systems which is accurately predicted

by models M7, M7*, and M8. However, all models underpredicted amide hydrogen exchanges in

other parts of the α4helix (e.g. fragment 6 in RGS4, fragments 14, 15, 16 in RGS8, and fragments

11 and 12 in RGS19) (Fig. A-30, A-31, A-33, A-34, A-36, and A-38). Analyses of our MD simula-

tions showed that the amide hydrogens in these fragments are strongly protected via hydrogen

bonds, and therefore local unfolding of the helical structure, even if very transiently, is perhaps

required to facilitate any exchange event. Through MD simulations, similar protection effects

were identified in the α5 helix of RGS8 (fragments 24 and 25) (Fig. A-31, A-32, A-36, and A-37)

113

and RGS19 (fragment 18) (Fig. A-33 and A-38).

The models accurately predicted experimentally-observed exchanges in amide hydrogens

in the connecting loops between helices, particularly for the α5-α6 loop (e.g. fragments 12 and

13 for RGS4 in Fig. A-30, A-31, A-34, and A-35; fragment 27 for RGS8 in Fig. A-31, A-32, A-36,

and A-37; and fragments 20, 21, and 22 for RGS19 in Fig. A-33 and A-38) which is the longest

unstructured region in RGS proteins (Fig. A-2). However, our models showed partial protection

for the amide hydrogen of Q122, a residue located in the α5-α6 interhelical loop of RGS4, even

though the side chain of this residue is solvent exposed. The amide hydrogen in Q122 forms a

long lasting hydrogen bond with S120 leading to a significant protection of this amide hydrogen

(Fig. A-47A and C). We also observed complete protection of the amide hydrogen in the residue

R119 of RGS8, which is located in the α5-α6 interhelical loop (Fig. A-42). We attribute this to

strong salt bridging interactions between the residue R119 and residues E84/E111 (Fig. A-47B

and D). For residues located near the protein surface as well as in flexible loops, the ability to

remain protected is consistent with the earlier observations on Staphylococcal nuclease211 show-

ing that the proximity to the surface of the protein does not usually produce fast exchange and

therefore a detailed hydrogen by hydrogen analysis is needed, as we have carried out here via MD

simulations. These results also provide testable predictions for future HDX-NMR studies aimed at

resolving residue-level exchanges since HDX-MS results only provide fragment-level resolution.

Solvent accessible surface area as a metric

In our proposed models M8 and M9, SASA is a key metric in determination of the expo-

sure of amide hydrogens to solvent that consequently contributes to the calculation of protection

factors. Since the hydrogen atoms are resolved in the NMR structures of RGS4 (PDB code 1EZT

114

containing only 1 conformer) and RGS19 (PDB code 1CMZ containing 20 conformers), we com-

puted the maximum and average SASA of all amide hydrogens from the NMR structures (Fig.

A-5). Given that all missing hydrogens are included in our MD simulations, we also calculated

similar SASA measures of all amide hydrogens of RGS4, RGS8, and RGS19 from all MD trajec-

tories conducted using both force-fields (Fig. A-48). The NMR structures show that only a few

amide hydrogens are exposed to solvent and those are located in the connection loops between

helices. The maximum values of SASA among all amide hydrogens are ~8Å2 and ~14 Å2 for RGS4

(PDB code 1EZT) and RGS19 (PDB code 1CMZ), respectively.

Our model M9 showed that the SASA threshold beyond which the experimental HDX

trends are well predicted are 8.02 Å2 and 9.15 Å2 for CHARMM and AMBER force-fields, respec-

tively. Given these values, none of the residues in the NMR structure of RGS4, and only 4 residues

in the NMR structure of RGS19 have enough exposure for competent exchange. However, amide

hydrogens show larger exposure to solvent in MD simulations (Fig. A-48) with maximum val-

ues up to ~20 Å2. For interhelical loops, the average SASA of amide hydrogens in simulations is

about two times that of helical motifs in RGS proteins. The residues within well-folded and stable

helices never adopt SASA values beyond the threshold SASA values (vide supra), thereby suggest-

ing strong protection effects for these amide hydrogens. Given that the SASA values of amide

hydrogens in the initial structures of RGS proteins (Fig. A-5) and in MD simulations (Fig. A-48)

are different as well as given the consistent performance of our SASA-based proposed models

(M8 and M9; Fig. A-3 and A-4), we find SASA computed from MD simulations as a useful metric

in modeling of HDX-MS data.

115

Figure A-5: The exposure of amide hydrogens in the NMR structures of RGS proteins. Shown are
the maximum (open circles) and the average (solid circles) values of the solvent accessible surface
area for all amide hydrogens in the NMR structures of RGS4 (panel A) and RGS19 (panel B). In
both panels, the absence of filled circles for certain amides as well as the absence of open circles
in panel B, is due to the approximately nil SASA values for those amides. The absence of open
circles for RGS4 in panel A is due to the lack of availability of more than 1 conformer in the NMR
structure of RGS4 as opposed to 20 conformers in the NMR structure of RGS19.

116

Mean residence times and cooperativity of amide hydrogens in the open and closed states

In the fractional-population models (M7/M7* and M9), the kinetics of fluctuations between

the open and closed states are characterized by the mean residence time (MRT) which is defined,

in an MD simulation, as the average number of consecutive frames in each state multiplied by

the time-step.131 Therefore, computing the MRT at residue-resolution provides information on

the tendency of each amide hydrogen to be in the open and the closed state. Two specific criteria

(Table A-4) were evaluated to classify amides as being in the open or closed states for each frame

in MD trajectories. Then, the MRT values of the closed state and the open state are used to

calculate the protection factors (PF = τC/τO). To calculate the PF for model M9, we divided the

number of frames in which an amide hydrogen is in a closed state (NFC) by the number of frames

in which an amide hydrogen is in an open state (NFO). If NO and NC are the number of visits to

the open state and the closed state during the MD trajectory, respectively, and TO and TC are the

total time that each amide is in the open or the closed state, respectively, it can be written that

TO = NFO∆τ = NOτO and TC = NFC∆τ = NCτC, where ∆τ is the time-step (which is 2 fs in our MD

simulations). This results in the protection factor, PF = TC/TO by assuming that NO = NC−1.30

In Fig. A-6, we show the MRT values of the open and the closed states of all residues from MD

trajectories of all proteins conducted using the CHARMM-FF and the AMBER-FF. These values

were calculated using equations: τO = NFO∆τ/NO and τC = NFC∆τ/NC.

Since the open states of amide hydrogens may occur at time scales shorter than the time-

step (∆τ) used in MD simulations, it was previously shown that the MRT values can be quantita-

tively corrected to account for the sampling-resolution systematic binning error. The corrected

values are given by τc

O = −∆τ/ln(1−NO/NFO) and τc

C = NFC∆τ/1−NFOln(1−NO/NFO).131 We show the

corrected MRT values in Fig. A-49. These data show that τO ranges between 20 to 50 ps while τc

O

117

Figure A-6: Mean residence times for the open and closed states of amide hydrogens. Data are
shown from all simulations of RGS4, RGS8, and RGS19 conducted with the CHARMM-FF (panel
A) and the AMBER-FF (panel B). The MRT calculations were carried out using our proposed
fractional population model M9 that showed consistent predictions with the HDX-MS data.

118

ranges between 5 to 50 ps and τC ranges between 170 ps to 2 μs while τc

C ranges between 110 ps

to 2 μs. The observation that the open states of amides occur on a sub-100 ps time scale is con-

sistent with similar earlier observations on the protein BPTI.131 As suggested previously,131 these

time scales are orders of magnitude shorter than the MRT values of globally unfolded proteins

and therefore highlight the concept that amides can exchange by highly localized and short-lived

fluctuations without the need for global unfolding. We further examined whether the open states

of amide hydrogens are truly localized or if they are allosterically coupled and cooperative. Specif-

ically, we computed the open state residue-residue correlation matrix for two simulations that

have shown significant per-residue fluctuations in RGS4 (PDB:1AGR) and RGS8 (PDB:2ODE) us-

ing the CHARMM-FF. We observed that the correlation matrix varies in a short-range for both

systems (Fig. A-50 and A-51) indicating that the open states for amides are largely uncorrelated

between residue pairs, as also has been previously observed for BPTI.131 These observations are

consistent with the amide hydrogen exchanges occurring in the EX2 exchange limit.191 Further-

more, the probability of observing open states of amides for a trajectory of given length can be

analyzed using Poisson statistics.131 We present this analysis in Fig. A-52 for the PF-values of

102, 104, 106, and 1011 with τO = 20 ps and 100 ps. The analysis shows that the open states of

amides with the PFs ranging between 102 and 106 can be observed in MD trajectories of simu-

lation lengths ranging between 10−3 μs and 10 μs. This is consistent with the results on the DI

observed in experiments and predicted by simulations for RGS proteins. However, the amides

that are highly protected and are not observed to exchange in experiments likely have protection

factors of 1011 or higher (as predicted by our simulations) and would require trajectories on time

scales of millisecond or higher for observing open states. We suggest that the probability of ob-

serving sufficient opening events for amides can be further enhanced by conducting simulations

119

with multiple force-fields and different initial structures of proteins, as we have carried out in

this work for RGS proteins.

Conclusion

We used MD simulations to study hydrogen-deuterium exchange events in three isoforms

of RGS proteins. Specifically, we analyzed various existing models from the literature to assess

their ability in accurately predicting experimentally observed exchange patterns in these homol-

ogous RGS proteins. These analyses revealed significant variation among models in accuracy of

predictions and showed that empirical models (termed models M1 through M6 in Table A-1) with

their previously reported criteria made inconsistent predictions, while a fractional population

model (Model M7) predicted experimentally-observed trends with good accuracy. Even though

we found that reoptimizing previous empirical models using our data on RGS proteins improves

their prediction accuracy, the performance of the fractional population model is less sensitive to

parameters. We further assessed the usefulness of a previously ignored metric, SASA of amide

hydrogens determined from MD simulations, and combined it with the distance of a given amide

hydrogen from the first polar atoms in proteins to propose two new models (models M8 and M9)

that show good predictions for observed HDX patterns. Importantly, the proposed models only

require the coordinates of protein atoms from solvated trajectories providing improved compu-

tational efficiency. We also find that the amide hydrogens often transiently visit open states on

sub-100 ps time scales, which is significantly shorter than time scales for global unfolding. This

therefore suggests that there is localized exposure of the amide-hydrogens, especially given that

open states among amide hydrogens of a given protein are uncorrelated.

120

Model Details

In the following, we provide details on seven existing models for protection factor (PF)

correlations, as shown in Table 1.

Model M1:

Resing et al.225 conducted early studies to predict exchange rates

in a kinase protein (ERK2) by fitting protection factors to an equation of
the form
log(P Fi) = log(kint/khdx) = u · (SAi) + v/(HBi), where khdx is the experimentally
measured exchange rate of an amide hydrogen, kint is the intrinsic exchange rate calculated

according to Bai et al.,200 SAi is the distance of each amide hydrogen from the surface of protein

in Å, and HBi is the hydrogen bond length of backbone amide nitrogens to an acceptor. They

also used deuterium exchange rates measured by Milne et al.231 for horse heart cytochrome c.

Model M2: Vendruscolo et al.208 proposed a model for predictions of HDX rates based

on the exploration of conformations using Monte Carlo (MC) sampling biased by experimental

data. They speculated that the protection of amide hydrogens comes from buried part of the

amide group and also from the hydrogen bonding in the secondary structure which resulted in a

phenomenological expression including the number of contacts of residue i with other residues

(N c

i ) and the number of hydrogen bonds formed by the amide hydrogens of residues (N h

i ), respec-

tively. According to their definition, hydrogen bonds are present if the angle between the NH

vector and the OH vector is below 0.7 rad and the OH distance is below 2.4 Å. Also, two residues

are in contact if any pair of their atoms are closer than 8.5 Å.

Model M3: Best et al.209 used the same phenomenological expression that Vendruscolo et

al.208 had proposed but with minor changes in definition of N c

i and N h

i . The contribution of burial

in the model is the number of heavy atoms within a distance of 6.5 Å from the amide nitrogen. A

cutoff of 2.4 Å between the donor hydrogen and the acceptor was used for identifying a hydrogen

121

bond without an angle criterion. They optimized the parameters of their model using experimen-

tal protection factors and the corresponding protection factors from a 1 ns conventional MD

simulation of seven different proteins. They acknowledge that major protein fluctuations were

elusive from short MD simulations which motivated them to conduct a biased simulation of the

protein bovine pancreatic trypsin inhibitor (BPTI) by using hydrogen exchange restraints with

varying values of the parameters.

Model M4: Kieseritzky et al.210 used MD simulations as a complement for hydrogen ex-

change experiments. They simulated oxidized c-type cytochrome under native conditions (PDB

code 1K3H) with the CHARMM22 force-field using explicit water molecules modeled using the

TIP3P water model. The simulation was 3 ns long. They proposed a protection factor definition

based on a linear combination of protection factors log(P Fi) = log(kint/khdx) = β1P F E1 +

β2P F E2. They optimized parameters β1 and β1 to arrive at an agreement between computed

(based on MD simulation data) and measured hydrogen exchange protection factors. The nine

different protection factor correlations in their paper show varieties of error and Pearson’s cor-

relation coefficient out of which P F E1 = [the number of residues which are in contact with

corresponding residue] and P F E2 = [the inverse of the backbone atom RMSF] show the least

error and the best correlation.

Model M5: A model was suggested by Ma et al.212 where N H β

i

is the average number

of hydrogen bonds between the NH atom of residue i and C=O backbone oxygen within 2.6 Å

distance, and N H sol

i

is the average number of hydrogen bonds between NH and water oxygen

within 3.0 Å distance of residue i. In the original model, N H β is measured in β-sheets and the

correlation is marginal P Fi = (N H sol

i + N H β

i )/C.N H sol

. They used CHARMM27 force-field to

i

do MD simulations of different β-sheet conformations, each of which was for 60 ns.

122

Model M6: Park et al.130 recently developed a novel model based on a comprehensive

HDX-MS experimental data using Amber 11 ff99SB force-field and a 100 ns long simulation. Their

logistics growth function HDX model consist of one fitting parameter called “base”. N Hstati is

defined as ([the number of snapshots showing H-bonding of amide hydrogen to protein]-[the

number of snapshots showing H-bonding amide hydrogen to water])/[the total number of snap-

shots]. They provided three amide hydrogen bond models out of which model HB2 has been

compared with other models in their work. In the HB2 model, H-bonding of a given amide hy-

drogen to the side chain as well as C=O group in the backbone are counted as H-bonding of amide

hydrogens to protein. The fraction of deuterium incorporation (DI) for each amide hydrogen was

computed by the first order reaction kinetics DI res

i = 1 − exp(−kint,i t/P Fi).

Model M7: Persson et al.131 used a significantly long MD simulation of protein BPTI (0.262

ms long) generated by Shaw et al.228 using Amber ff99SB-I/TIP4P-Ew force-field. They start with a

description of the standard model in which each amide can be exposed to solvent in an open state
or buried within the protein by a closed state: (N − H)c
kint−→ (N − D)o in which
HDX rate is given as khdx = kokint/(ko+kc+kint). The assumption of kint ≪ kc+ko, which is an
applicable assumption for HDX experiment, results in a simple and practical phenomenological

(N − H)o

ko←→

kc

model khdx = kint/(P F +1). The protection factor here is the key for the calculation of hydrogen-

deuterium exchange rate and it is defined as the ratio of residence time in the closed state to

residence time in the open state which is applicable to MD simulations. The criteria for the open

state and the closed state play an important role in computing protection factors. They speculate

that a direct access to external solvent and disruption of any intramolecular H-bond with the N-H

group are key factors in defining the open state. A residue is in an open state when the amide

hydrogen has at least two water oxygens within 2.6 Å and that the amide hydrogen has no other

123

PDB
1AGR
1EZT
2IHD
2ODE
1CMZ

RGS4

RGS8
RGS19

system size (atoms)

28160
29275
27490
30731
29560

force-field (trajectory length)

CHARMM36 (2 μs), AMBER (2 μs)
CHARMM36 (2 μs), AMBER (2 μs)
CHARMM36 (2 μs), AMBER (2 μs)
CHARMM36 (2 μs), AMBER (2 μs)
CHARMM36 (2 μs), AMBER (2 μs)

Table A-2: Summary of MD simulations.

polar protein atoms (except in neighboring residues) within 2.6 Å.

Other studies: In addition to models highlighted above, Craig et al.217 modeled deuterium

incorporation of three different proteins using coarse-grained MD simulations. The open state cri-

teria were evaluated by the number of contacts per residue and the distance changes between the

H- bonded residues compared to their native conformations. Petruk et al218 studied a kinase pro-

tein (ERK2MAP) using all-atom explicit-water MD simulations and showed that both the whole

dynamically averaged solvent accessible surface area (SASA) and the number of waters in the

first solvation shell of each amide nitrogen can be used as metrics for predicting deuterium incor-

poration. Recently, Adhikary et al.201 have modeled deuterium incorporation using multiple MD

simulations (each 450 ns long) of neurotransmitter sodium symporters.

Supplemental Methods

System Setup: MD simulations

A summary of all MD simulations for RGS4, RGS8, and RGS19 is provided in Table A-

2. Specifically, 10 independent MD simulations, each 2 μs long, were conducted using both

CHARMM and AMBER force-fields for all apo RGS proteins.

124

Protocols for HDX Modeling

The HDX-MS experiments provided fragment-based DI whereas in MD simulations, it is

feasible to calculate DI at a residue resolution.

In Fig. A-8, we show details on all fragments

and their residues for RGS4, RGS8, and RGS19. To compare DI between experiments and simula-

tions, DI of residues (except prolines that do not have amide hydrogens) were averaged over the

corresponding fragment using Eq. (1):

∑
j=1,̸=P RO DI res

m

j

m

DI f rag

i

=

(1)

where m is the number of residues in the fragment.

For all models, we calculated the intrinsic HDX kinetic rates per Bai et al.200 at 273 K, the

temperature at which our HDX-MS experiments were conducted. Initially, we analyzed 100,000

frames for each 2 μs MD trajectory by applying the default criteria reported in the literature for

models M1 through M7 to compute PFs of amides for all RGS proteins. We then re-optimized the

parameters of all models by minimizing an objective function (Eq. (2)) which incorporates HDX-

MS data and MD simulations of all RGS proteins. It should be noted here that the optimization

of parameters were carried out separately for each force-field due to the fact that CHARMM and

AMBER force-fields are parameterized differently for studies of protein dynamics.

5∑

n∑

OF =

(
SY S=1
f rag=1

(cid:12)(cid:12)(cid:12)DI f rag

exp

(cid:12)(cid:12)(cid:12))SY S

− DI f rag

sim

(2)

where OF is the objective function, DI is deuterium incorporation, SY S is the number of sim-

ulations for an RGS protein using the same force-field, and f rag is the fragment number. All

default and re-optimized parameters of models M1 through M7 are listed in Table A-4.

125

In addition to existing models (M1-M7), we revisited and evaluated SASA of amide hy-

drogens as a metric in prediction of amide PFs because contradictory observations regarding

the use of SASA as a metric have been proposed in the literature. Published studies indicate

that SASA of amide hydrogens reasonably predicts the number of exchanged hydrogens218 or is

an even better indicator for protected hydrogens than using H-bonds.221 Contrary to this view,

a lack of agreement between HDX experiments and MD simulations based on SASA has been

reported.131 Besides, although anticorrelations between the SASA of amide hydrogens and the

residue-resolution protection factors from experiments existed, Park et al.130 chose H-bonds as

a metric for HDX modeling to overcome the limitation of using SASA and they concluded that

H-bonds are a generic and suitable metric for the estimation of PFs.

We therefore developed two new models (listed as M8 and M9 in Table A-3 and A-4) using

the distances of amide hydrogens from the first polar atom as an alternative metric along with

SASA of each amide hydrogen to comply with the theory of HDX in which a residue may be pro-

tected by polar atoms despite having large enough SASA.131,205 This assertion comes from the fact

that surface exposed hydrogens (with higher values of SASA) can be significantly protected from

hydrogen exchange.225 Surprisingly, these two metrics in combination have resulted in trends

and values consistent with experiments.

Specifically, model M8 is an empirical model (similar to models M1 through M6) based

upon SASA of amide hydrogens and distances of amide hydrogens to the first polar atom (except

in the neighboring residues) (Di) and ln(P Fi) is a power function of SASAi and Di. However,

model M9 is a fractional population model131 where the same metrics (SASAi and Di) were used

for distinguishing between the open and closed states of amides. We define the open state in

model M9 for each amide hydrogen when its SASA crosses a threshold value (dsasa) and that the

126

Model Protection factor criteria
−γs
i + βpD
M8
M9
Table A-3: Models proposed in this work.

ln(P Fi) = (βsSASA
P Fi = τC/τO

−γp
i

amide hydrogen has no other polar protein atom (except in neighboring residues) within a thresh-

old distance (dp). The values of thresholds/cut-offs in model M9 and four correlation coefficients

in model M8 are obtained by minimizing the objective function in Eq. (2). The intrinsic exchange

rates in new models were also calculated according to Bai et al.200

127

Model
M1

(1999-Resing)

Criteria
log(P Fi) = u · (SAi) + v/(HBi)
u = 0.76, v = 8.2
uch = 6.15, vch = 5.32
uam = 5.18, vam = 4.92
ln(P Fi) = (βcN C

i + βhN h
i )

M2

M4

M5

M6

M7

M8

M9

(2011-Ma)

i + βhN h
i )

M3

(2006-Best)

i + βr(N r

i )−1)

(2006-Kieseritzky)

h = 0.85
h = 0.9

h = 5.40
h = 4.00

(2003-Vendruscolo) βc = 1, βh = 5
c = 0.49, βch
βch
c = 0.5, βam
βam
ln(P Fi) = (βcN C
βc = 0.35, βh = 2
c = 0.23, βch
βch
c = 0.23, βam
βam
ln(P Fi) = (βcN C
βc = 0.5, βr = 0.9
r = 1.31
c = 0.45, βch
βch
r = 6.45
c = 0.19, βam
βam
i + CcN H β
P Fi = (CoN H sol
−6, C ch
c = 2.50
o = 8.48e
C ch
√
c = 1.47e4
o = 0.15, C am
C am
base)1−N Hstati
P Fi = base/(1 + (
base = 108
basech = 1.3e8
baseam = 0.4e8
P Fi = τC/τO
dw = 2.60, dp = 2.60
p = 2.73
w = 2.43, dch
dch
p = 2.73
w = 2.40, dam
dam
−γs
ln(P Fi) = (βsSASA
i + βpD
p = 2.60e1
s = 0.72, βch
βch
p = 0.99
s = 0.53, γch
γch
−3, βam
s = 1.30e
βam
p = 1.27
s = 2.64, γam
γam
P Fi = τC/τO
sasa = 9.152, dch
dch
sasa = 8.022, dam
dam

p = 3.00
p = 2.99

(2015-Persson)

p = 3.65e1

(2011-Park)

i )/CN H sol

i

−γp
i

Table A-4: Details on all protection factor correlation models with the default and reoptimized
values of their parameters. Optimized values based upon simulations conducted using CHARMM
and AMBER force-fields are listed with superscripts ch and am, respectively. In addition, details
on two new models M8 and M9 proposed in this work are listed.

128

Figure A-7: Experimentally measured percentage deuterium incorporation (%DI) of fragments in
RGS proteins at t = 0, 3, 10, 30, 100, 300, and 1000 minutes (RGS4: top row; RGS8: middle row;
RGS19: bottom row).

Figure A-8: Definitions of fragments for each RGS protein. Each fragment comprises residues
whose color determines their location in nine α helices of each RGS protein. Residue names in
connecting loops are highlighted in black, but shown as white cartoons in the protein structure.
All helices are colored and labeled in the protein rendering.

129

Figure A-9: Modeled deuterium incorporation of fragments in RGS4. The HDX experiment (blue)
is shown seven discrete times, alongside each different model with default parameters (orange).
This figure shows the MD simulation results for PDB:1AGR and AMBER force-field.

Figure A-10: Modeled deuterium incorporation of fragments in RGS4. The HDX experiment (blue)
is shown seven discrete times, alongside each different model with default parameters (orange).
This figure shows the MD simulation results for PDB:1EZT and AMBER force-field

130

Figure A-11: Modeled deuterium incorporation of fragments in RGS8. The HDX experiment (blue)
is shown seven discrete times, alongside each different model with default parameters (orange).
This figure shows the MD simulation results for PDB:2IHD and AMBER force-field

Figure A-12: Modeled deuterium incorporation of fragments in RGS8. The HDX experiment (blue)
is shown seven discrete times, alongside each different model with default parameters (orange).
This figure shows the MD simulation results for PDB:2ODE and AMBER force-field

131

Figure A-13: Modeled deuterium incorporation of fragments in RGS19. The HDX experiment
(blue) is shown seven discrete times, alongside each different model with default parameters
(orange). This figure shows the MD simulation results for PDB:1CMZ and AMBER force-field.

Figure A-14: Modeled deuterium incorporation of fragments in RGS4. The HDX experiment
(blue) is shown seven discrete times, alongside each different model with optimized parameters
(orange). This figure shows the MD simulation results for PDB:1AGR and AMBER Force-field

132

Figure A-15: Modeled deuterium incorporation of fragments in RGS4. The HDX experiment
(blue) is shown seven discrete times, alongside each different model with optimized parameters
(orange). This figure shows the MD simulation results for PDB:1EZT and AMBER Force-field

133

Figure A-16: Modeled deuterium incorporation of fragments in RGS8. The HDX experiment
(blue) is shown seven discrete times, alongside each different model with optimized parameters
(orange). This figure shows the MD simulation results for PDB:2IHD and AMBER Force-field.

134

Figure A-17: Modeled deuterium incorporation of fragments in RGS8. The HDX experiment
(blue) is shown seven discrete times, alongside each different model with optimized parameters
(orange). This figure shows the MD simulation results for PDB:2ODE and AMBER Force-field.

135

Figure A-18: Modeled deuterium incorporation of fragments in RGS19. The HDX experiment
(blue) is shown seven discrete times, alongside each different model with optimized parameters
(orange). This figure shows the MD simulation results for PDB:1CMZ and AMBER Force-field.

Figure A-19: Modeled deuterium incorporation of fragments in RGS4. The HDX experiment (blue)
is shown twice, alongside new models (M8, M9) with optimized parameters (orange). This figure
shows the MD simulation results for PDB:1AGR and AMBER Force-field.

Figure A-20: Modeled deuterium incorporation of fragments in RGS4. The HDX experiment (blue)
is shown twice, alongside new models (M8, M9) with optimized parameters (orange). This figure
shows the MD simulation results for PDB:1EZT and AMBER Force-field.

136

Figure A-21: Modeled deuterium incorporation of fragments in RGS8. The HDX experiment (blue)
is shown twice, alongside new models (M8, M9) with optimized parameters (orange). This figure
shows the MD simulation results for PDB:2IHD and AMBER Force-field.

Figure A-22: Modeled deuterium incorporation of fragments in RGS8. The HDX experiment (blue)
is shown twice, alongside new models (M8, M9) with optimized parameters (orange). This figure
shows the MD simulation results for PDB:2ODE and AMBER Force-field.

Figure A-23: Modeled deuterium incorporation of fragments in RGS19. The HDX experiment
(blue) is shown twice, alongside new models (M8, M9) with optimized parameters (orange). This
figure shows the MD simulation results for PDB:1CMZ and AMBER Force-field.

137

Figure A-24: Modeled deuterium incorporation of fragments in RGS4. The HDX experiment (blue)
is shown seven discrete times, alongside each different model with default parameters (orange).
This figure shows the MD simulation results for PDB:1AGR and CHARMM Force-field.

Figure A-25: Modeled deuterium incorporation of fragments in RGS4. The HDX experiment (blue)
is shown seven discrete times, alongside each different model with default parameters (orange).
This figure shows the MD simulation results for PDB:1EZT and CHARMM Force-field.

138

Figure A-26: Modeled deuterium incorporation of fragments in RGS8. The HDX experiment (blue)
is shown seven discrete times, alongside each different model with default parameters (orange).
This figure shows the MD simulation results for PDB:2IHD and CHARMM Force-field.

Figure A-27: Modeled deuterium incorporation of fragments in RGS8. The HDX experiment (blue)
is shown seven discrete times, alongside each different model with default parameters (orange).
This figure shows the MD simulation results for PDB:2ODE and CHARMM Force-field.

139

Figure A-28: Modeled deuterium incorporation of fragments in RGS19. The HDX experiment
(blue) is shown seven discrete times, alongside each different model with default parameters
(orange). This figure shows the MD simulation results for PDB:1CMZ and CHARMM Force-field.

Figure A-29: Modeled deuterium incorporation of fragments in RGS4. The HDX experiment
(blue) is shown seven discrete times, alongside each different model with optimized parameters
(orange). This figure shows the MD simulation results for PDB:1AGR and CHARMM Force-field.

140

Figure A-30: Modeled deuterium incorporation of fragments in RGS4. The HDX experiment
(blue) is shown seven discrete times, alongside each different model with optimized parameters
(orange). This figure shows the MD simulation results for PDB:1EZT and CHARMM Force-field.

141

Figure A-31: Modeled deuterium incorporation of fragments in RGS8. The HDX experiment
(blue) is shown seven discrete times, alongside each different model with optimized parameters
(orange). This figure shows the MD simulation results for PDB:2IHD and CHARMM Force-field.

142

Figure A-32: Modeled deuterium incorporation of fragments in RGS8. The HDX experiment
(blue) is shown seven discrete times, alongside each different model with optimized parameters
(orange). This figure shows the MD simulation results for PDB:2ODE and CHARMM Force-field.

143

Figure A-33: Modeled deuterium incorporation of fragments in RGS19. The HDX experiment
(blue) is shown seven discrete times, alongside each different model with optimized parameters
(orange). This figure shows the MD simulation results for PDB:1CMZ and CHARMM Force-field.

Figure A-34: Modeled deuterium incorporation of fragments in RGS4. The HDX experiment (blue)
is shown twice, alongside new models (M8, M9) with optimized parameters (orange). This figure
shows the MD simulation results for PDB:1AGR and CHARMM Force-field.

Figure A-35: Modeled deuterium incorporation of fragments in RGS4. The HDX experiment (blue)
is shown twice, alongside new models (M8, M9) with optimized parameters (orange). This figure
shows the MD simulation results for PDB:1EZT and CHARMM Force-field.

144

Figure A-36: Modeled deuterium incorporation of fragments in RGS8. The HDX experiment (blue)
is shown twice, alongside new models (M8, M9) with optimized parameters (orange). This figure
shows the MD simulation results for PDB:2IHD and CHARMM Force- field.

Figure A-37: Modeled deuterium incorporation of fragments in RGS8. The HDX experiment (blue)
is shown twice, alongside new models (M8, M9) with optimized parameters (orange). This figure
shows the MD simulation results for PDB:2ODE and CHARMM Force-field.

Figure A-38: Modeled deuterium incorporation of fragments in RGS19. The HDX experiment
(blue) is shown twice, alongside new models (M8, M9) with optimized parameters (orange). This
figure shows the MD simulation results for PDB:1CMZ and CHARMM Force-field.

145

Figure A-39: Deuterium incorporation is mapped on RGS proteins at t = 1000 min as observed
in experiments and as predicted by the models M7, M8, and M9. Data are presented for the
CHARMM-FF simulations of RGS4, RGS8, and RGS19.

146

Figure A-40: Root mean squared fluctuations (RMSF) per residue across protein sequences are
shown from 2-μs long MD simulations of (A) RGS4 (PDB: 1AGR, 1EZT), (B) RGS8 (PDB: 2IHD,
2ODE), and (C) RGS19 (PDB: 1CMZ). Color bars indicate helical regions.

147

Figure A-41: Modeled deuterium incorporation at t = 1000 min at a single-residue resolution
(RGS4, CHARMM-FF).

148

Figure A-42: Modeled deuterium incorporation at t = 1000 min at a single-residue resolution
(RGS8, CHARMM-FF).

149

Figure A-43: Modeled deuterium incorporation at t = 1000 min at a single-residue resolution
(RGS4, AMBER-FF).

150

Figure A-44: Modeled deuterium incorporation at t = 1000 min at a single-residue resolution
(RGS8, AMBER-FF).

Figure A-45: Modeled deuterium incorporation at t = 1000 min at a single-residue resolution
(RGS19, CHARMM-FF).

151

Figure A-46: Modeled deuterium incorporation at t = 1000 min at a single-residue resolution
(RGS19, AMBER-FF).

Figure A-47: The residues protected by hydrogen-bonds or salt-bridging interactions are high-
lighted (panels A and B). The traces for distances between the centers-of-masses of residue pairs
are shown in panel C (S120-Q122) and panel D (E84-R119 and E111-R119).

152

Figure A-48: SASA data similar to Fig. A-6 are shown from MD simulations of all RGS proteins
for both force-fields (CHARMM-FF, panel A; AMBER-FF, panel B). Color and labeling details are
similar to Fig. A-6

153

Figure A-49: Corrected mean residence times for open-states of amide hydrogens are shown.
Other details are similar to Fig. A-6.

154

Figure A-50: Residue-residue correlations among open states of all amide-hydrogens (CHARMM-
FF, RGS4 (PDB code 1AGR), model M7). The correlation matrix is calculated based on the prob-
ability that two amide hydrogens simultaneously explore open states; C(i, j) = (P(i, j) −
P(i)P(j))/(P(i)P(j)(1 − P(i))(1 − P(j)))0.5

155

Figure A-51: Data similar to A-50 are shown for RGS8 (CHARMM-FF, RGS8 (PDB code 2ODE),
model M7).

156

Figure A-52: Probability of a closed to open transition in a given amide vs. simulation length (μs)
is presented based upon Poisson statistics. Data are shown for PFs = 102, 104, 106, and 1011 with
τO = 20 ps and 100 ps.

157

REFERENCES

158

REFERENCES

1. Koshland, D. E. Application of a Theory of Enzyme Specificity to Protein Synthesis.
Proceedings of the National Academy of Sciences of the United States of America 44, 98–104 (1958).

2. Monod, J., Wyman, J. & Changeux, J.-P. On the nature of allosteric transitions: A

plausible model. Journal of Molecular Biology 12, 88–118 (1965).

3. Feixas, F., Lindert, S., Sinko, W. & McCammon, J. A. Exploring the role of receptor

flexibility in structure-based drug discovery. Biophysical Chemistry 186, 31–45 (2014).

4. Tobi, D. & Bahar, I. Structural changes involved in protein binding correlate with in-
trinsic motions of proteins in the unbound state. Proceedings of the National Academy of Sciences
102, 18908–18913 (2005).

5. Mittag, T., Kay, L. E. & Forman-Kaya, J. D. Protein dynamics and conformational disor-

der in molecular recognition. Journal of Molecular Recognition 23, 105–116 (2010).

6. Latorraca, N. R., Venkatakrishnan, A. J. & Dror, R. O. GPCR Dynamics: Structures in

Motion. Chemical Reviews 117, 139–155 (2017).

7. Eyrisch, S. & Helms, V. Transient pockets on protein surfaces involved in protein- pro-

tein interaction. Journal of Medicinal Chemistry 50, 3457–3464 (2007).

8. Metz, A. et al. Hot Spots and Transient Pockets: Predicting the Determinants of Small-
Molecule Binding to a Protein–Protein Interface. Journal of Chemical Information and Modeling
52, 120–133 (2012).

9. Langley, J. N. On the reaction of cells and of nerve-endings to certain poisons, chiefly
as regards the reaction of striated muscle to nicotine and to curari. The Journal of Physiology 33,
374–413 (1905).

10. Parascandola, J. Origins of the receptor theory. Trends in Pharmacological Sciences 1,

189–192 (1979).

11. Maehle, A.-H. A binding question: the evolution of the receptor concept. Endeavour

33, 135–40 (2009).

12. Hopkins, A. L. & Groom, C. R. The druggable genome. Nature Reviews Drug Discovery

1, 727–730 (2002).

13. Taylor, D. Fewer new drugs from the pharmaceutical industry. Britich Medical Journal

159

(Clinical Research Edition) 326, 408–9 (2003).

14. Evans, R. M. & Mangelsdorf, D. J. Nuclear Receptors, RXR, and the Big Bang. Cell 157,

255–66 (2014).

15. Cohen, P. & Alessi, D. R. Kinase drug discovery - What’s next in the field? 8, 96–104

(2013).

16. Botta, M. New Frontiers in Kinases: Special Issue. ACS Medicinal Chemistry Letters 5,

270–270 (2014).

17. Sperandio, O., Reynès, C. H., Camproux, A.-C. & Villoutreix, B. O. Rationalizing
the chemical space of protein–protein interaction inhibitors. Drug Discovery Today 15, 220–229
(2010).

18. Cheng, A. C. et al. Structure-based maximal affinity model predicts small-molecule

druggability. Nature Biotechnology 25, 71–75 (2007).

19. Wells, J. A. & McClendon, C. L. Reaching for high-hanging fruit in drug discovery at

protein–protein interfaces. Nature 450, 1001–1009 (2007).

20. Conte, L. L., Chothia, C. & Janin, J. The atomic structure of protein-protein recognition

sites. Journal of Molecular Biology 285, 2177–98 (1999).

21. Stanton, B. Z. et al. A small molecule that binds Hedgehog and blocks its signaling in

human cells. Nature Chemical Biology 5, 154–156 (2009).

22. Ran, X. & Gestwicki, J. E. Inhibitors of protein–protein interactions (PPIs): an analysis
of scaffold choices and buried surface area. Current Opinion in Chemical Biology 44, 75–86 (2018).

23. Arkin, M. R., Tang, Y. & Wells, J. A. Small-molecule inhibitors of protein-protein

interactions: Progressing toward the reality. Chemistry and Biology 21, 1102–1114 (2014).

24. Gowthaman, R., Deeds, E. J. & Karanicolas, J. Structural Properties of Non-Traditional
Drug Targets Present New Challenges for Virtual Screening. Journal of Chemical Information and
Modeling 53, 2073–2081 (2013).

25. Dandapani, S. & Marcaurelle, L. A. Grand Challenge Commentary: Accessing new

chemical space for ’undruggable’ targets. Nature Chemical Biology 6, 861–863 (2010).

26. Bauer, R. A., Wurst, J. M. & Tan, D. S. Expanding the range of ’druggable’ targets with
natural product-based libraries: an academic perspective. Current Opinion in Chemical Biology
14, 308–14 (2010).

160

27. Changeux, J.-P. & Edelstein, S. J. Allosteric mechanisms of signal transduction. Science

308, 1424–8 (2005).

28. Tsai, C.-J., Sol, A. del & Nussinov, R. Protein allostery, signal transmission and dynam-

ics: a classification scheme of allosteric mechanisms. Molecular Biosystems 5, 207 (2009).

29. Conn, P. J., Christopoulos, A. & Lindsley, C. W. Allosteric modulators of GPCRs: a
novel approach for the treatment of CNS disorders. Nature Reviews Drug Discovery 8, 41–54
(2009).

30. May, L. T., Leach, K., Sexton, P. M. & Christopoulos, A. Allosteric Modulation of G

Protein–Coupled Receptors. Annual Review of Pharmacology and Toxicology 47, 1–51 (2007).

31. Roman, D. L., Blazer, L. L., Monroy, C. A. & Neubig, R. R. Allosteric inhibition of
the regulator of G protein signaling-Galpha protein-protein interaction by CCG-4986. Molecular
Pharmacology 78, 360–365 (2010).

32. Kunze, J. et al. Targeting dynamic pockets of HIV-1 protease by structure-based com-
putational screening for allosteric inhibitors. Journal of Chemical Information and Modeling 54,
987–991 (2014).

33. Appelt, K. et al. Design of enzyme inhibitors using iterative protein crystallographic

analysis. Journal of Medicinal Chemistry 34, 1925–34 (1991).

34. Kuntz, I. D. Structure-Based Strategies for Drug Design and Discovery. Science 257,

1078–1082 (1992).

35. Warren, G. L. et al. A Critical Assessment of Docking Programs and Scoring Functions.

Journal of Medicinal Chemistry 49, 5912–5931 (2006).

36. Englebienne, P. & Moitessier, N. Docking Ligands into Flexible and Solvated Macro-
molecules. 4. Are Popular Scoring Functions Accurate for this Class of Proteins? Journal of
Chemical Information and Modeling 49, 1568–1580 (2009).

37. Lexa, K. W. & Carlson, H. A. Protein flexibility in docking and surface mapping. Quar-

terly Reviews of Biophysics 45, 301–43 (2012).

38. Pantsar, T. & Poso, A. Binding Affinity via Docking: Fact and Fiction. Molecules 23,

(2018).

39.

Jain, A. N. Effects of protein conformation in docking:

improved pose prediction
through protein pocket adaptation. Journal of Computer-Aided Molecular Design 23, 355–74
(2009).

161

40. Subramanian, J., Sharma, S. & B-Rao, C. A Novel Computational Analysis of Ligand-
Induced Conformational Changes in the ATP Binding Sites of Cyclin Dependent Kinases. Journal
of Medicinal Chemistry 49, 5434–5441 (2006).

41. Hammes, G. G., Chang, Y.-C. & Oas, T. G. Conformational selection or induced fit:
A flux description of reaction mechanism. Proceedings of the National Academy of Sciences 106,
13737–13741 (2009).

42. Kovermann, M., Grundström, C., Sauer-Eriksson, A. E., Sauer, U. H. & Wolf-Watz, M.
Structural basis for ligand binding to an enzyme by a conformational selection pathway. Proceed-
ings of the National Academy of Sciences of the United States of America 114, 6298 (2017).

43. Fersht, A. R., Knill-Jones, J. W., Bedouelle, H. & Winter, G. Reconstruction by site-
directed mutagenesis of the transition state for the activation of tyrosine by the tyrosyl-tRNA
synthetase: a mobile loop envelopes the transition state in an induced-fit mechanism. Biochem-
istry 27, 1581–1587 (1988).

44. Joseph, D., Petsko, G. & Karplus, M. Anatomy of a conformational change: hinged

”lid” motion of the triosephosphate isomerase loop. Science 249, 1425–1428 (1990).

45. Sullivan, S. M. & Holyoak, T. Enzymes with lid-gated active sites must operate by an
induced fit mechanism instead of conformational selection. Proceedings of the National Academy
of Sciences of the United States of America 105, 13829–34 (2008).

46. Weikl, T. R. & Paul, F. Conformational selection in protein binding and function. Pro-

tein Science: A Publication of the Protein Society 23, 1508–18 (2014).

47. Csermely, P., Palotai, R. & Nussinov, R. Induced fit, conformational selection and inde-
pendent dynamic segments: an extended view of binding events. Trends in Biochemical Sciences
35, 539–46 (2010).

48. Beglov, D. et al. Exploring the structural origins of cryptic sites on proteins. Pro-
ceedings of the National Academy of Sciences of the United States of America 115, E3416–E3425
(2018).

49. O’Bryan, J. P. Pharmacological targeting of RAS: Recent success with direct inhibitors.

Pharmacological research 139, 503–511 (2019).

50. Ostrem, J. M., Peters, U., Sos, M. L., Wells, J. A. & Shokat, K. M. K-Ras(G12C) inhibitors

allosterically control GTP affinity and effector interactions. Nature 503, 548–51 (2013).

51. Milburn, M. V. et al. Molecular switch for signal transduction: structural differences

between active and inactive forms of protooncogenic ras proteins. Science 247, 939–45 (1990).

162

52. Comitani, F. & Gervasio, F. L. Exploring Cryptic Pockets Formation in Targets of
Pharmaceutical Interest with SWISH. Journal of Chemical Theory and Computation 14, 3321–3331
(2018).

53. Singh, J., Petter, R. C., Baillie, T. A. & Whitty, A. The resurgence of covalent drugs.

Nature Reviews Drug Discovery 10, 307–317 (2011).

54. Roth, G. J., Stanford, N. & Majerus, P. W. Acetylation of prostaglandin synthase by
aspirin. Proceedings of the National Academy of Sciences of the United States of America 72, 3073
(1975).

55. Savi, P. et al. Identification and biological activity of the active metabolite of clopido-

grel. Journal of Thrombosis and Haemostasis 84, 891–6 (2000).

56. Pan, Z. et al. Discovery of selective irreversible inhibitors for Bruton’s tyrosine kinase.

ChemMedChem 2, 58–61 (2007).

57. Miller, V. A. et al. Afatinib versus placebo for patients with advanced, metastatic
non-small-cell lung cancer after failure of erlotinib, gefitinib, or both, and one or two lines of
chemotherapy (LUX-Lung 1): a phase 2b/3 randomised trial. The Lancet Oncology 13, 528–538
(2012).

58. Yver, A. Osimertinib (AZD9291)—a science-driven, collaborative approach to rapid

drug design and development. Annals of Oncology 27, 1165–1170 (2016).

59. Zhao, Z. & Bourne, P. E. Progress with covalent small-molecule kinase inhibitors. Drug

Discovery Today 23, 727–735 (2018).

60. Potashman, M. H. & Duggan, M. E. Covalent modifiers: an orthogonal approach to

drug design. Journal of Medicinal Chemistry 52, 1231–1246 (2009).

61. Evans, D. C., Watt, A. P., Nicoll-Griffith, D. A. & Baillie, T. A. Drug-Protein Adducts:
An Industry Perspective on Minimizing the Potential for Drug Bioactivation in Drug Discovery
and Development. 17, 3–16 (2004).

62. Naisbitt, D. J., Gordon, S., Pirmohamed, M. & Park, B. Immunological Principles of

Adverse Drug Reactions. Drug Safety 23, 483–507 (2000).

63. Johnson, D. S., Weerapana, E. & Cravatt, B. F. Strategies for discovering and derisking

covalent, irreversible enzyme inhibitors. Future Medicinal Chemistry 2, 949–964 (2010).

64. Padovan, E. T-cell response in penicillin allergy. Clinical and experimental allergy :

journal of the British Society for Allergy and Clinical Immunology 28 Suppl 4, 33–6 (1998).

163

65. Krishnan, A., Almén, M. S., Fredriksson, R. & Schiöth, H. B. The origin of GPCRs:
identification of mammalian like Rhodopsin, Adhesion, Glutamate and Frizzled GPCRs in fungi.
PLOS ONE 7, e29817 (2012).

66. Taddese, B. et al. Do plants contain g protein-coupled receptors? Plant Physiology

164, 287–307 (2014).

67. Lagerström, M. C. & Schiöth, H. B. Structural diversity of G protein-coupled receptors

and significance for drug discovery. Nature Reviews Drug Discovery 7, 339–357 (2008).

68. Sugiyama, H., Ito, I. & Hirono, C. A new type of glutamate receptor linked to inositol

phospholipid metabolism. Nature 325, 531–533 (1987).

69. Conn, P. J. & Pin, J.-P. PHARMACOLOGY AND FUNCTIONS OF METABOTROPIC
GLUTAMATE RECEPTORS. Annual Review of Pharmacology and Toxicology 37, 205–237 (1997).

70. Dunlap, K. Functional and pharmacological differences between two types of GABA

receptor on embryonic chick sensory neurons. Neuroscience Letters 47, 265–270 (1984).

71. Andrade, R., Malenka, R. C. & Nicoll, R. A. A G protein couples serotonin and GABAB

receptors to the same channels in hippocampus. Science 234, 1261–5 (1986).

72. Goldstein, D. S., Eisenhofer, G. & McCarty, R. Catecholamines : bridging basic science

with clinical medicine. 1084 (Academic Press, 1998).

73. Missale, C., Nash, S. R., Robinson, S. W., Jaber, M. & Caron, M. G. Dopamine Receptors:

From Structure to Function. Physiological Reviews 78, 189–225 (1998).

74. Barnes, N. M. & Sharp, T. A review of central 5-HT receptors and their function.

Neuropharmacology 38, 1083–1152 (1999).

75. Chen, J., Ishii, M., Wang, L., Ishii, K. & Coughlin, S. R. Thrombin receptor activation.
Confirmation of the intramolecular tethered liganding hypothesis and discovery of an alternative
intermolecular liganding mode. The Journal of Biological Chemistry 269, 16041–5 (1994).

76. Rasmussen, S. G. F. et al. Crystal structure of the β2 adrenergic receptor-Gs protein

complex. Nature 477, 549–55 (2011).

77. Gilman, A. G. G Proteins and Dual Control of Adenylate Cyclase. 36, 577–579 (1984).

78. Tang, W. J. & Gilman, A. G. Type-specific regulation of adenylyl cyclase by G protein

beta gamma subunits. Science 254, 1500–3 (1991).

164

79. Berridge, M. J. Inositol trisphosphate and calcium signalling. Nature 361, 315–25

(1993).

80. Wynne, B. M., Chiao, C.-W. & Webb, R. C. Vascular Smooth Muscle Cell Signaling
Mechanisms for Contraction to Angiotensin II and Endothelin-1. Journal of the American Society
of Hypertension 3, 84–95 (2009).

81. Siehler, S. Regulation of RhoGEF proteins by G12/13-coupled receptors. British Journal

of Pharmacology 158, 41–9 (2009).

82. Druey, K. M., Blumer, K. J., Kang, V. H. & Kehrl, J. H. Inhibition of G-protein-mediated

MAP kinase activation by a new mammalian gene family. Nature 379, 742–746 (1996).

83. Tesmer, J. J. G., Berman, D. M., Gilman, A. G. & Sprang, S. R. Structure of RGS4 bound
to AlF4–activated G(i alpha1): stabilization of the transition state for GTP hydrolysis. Cell 89,
251–261 (1997).

84. Zeng, W. et al. The N-terminal domain of RGS4 confers receptor-selective inhibition

of G protein signaling. The Journal of Biological Chemistry 273, 34687–90 (1998).

85. Heximer, S. P., Watson, N., Linder, M. E., Blumer, K. J. & Hepler, J. R. RGS2/G0S8 is
a selective inhibitor of Gqalpha function. Proceedings of the National Academy of Sciences of the
United States of America 94, 14389–93 (1997).

86. Patil, D. N. et al. Structural organization of a major neuronal G protein regulator, the

RGS7-Gβ5-R7BP complex. eLife 7, (2018).

87. Hu, G. & Wensel, T. G. R9AP, a membrane anchor for the photoreceptor GTPase
accelerating protein, RGS9-1. Proceedings of the National Academy of Sciences of the United States
of America 99, 9755–60 (2002).

88. Drenan, R. M. et al. Palmitoylation regulates plasma membrane-nuclear shuttling of
R7BP, a novel membrane anchor for the RGS7 family. Journal of Cell Biology 169, 623–633 (2005).

89. Snow, B. E. et al. A G protein gamma subunit-like domain shared between RGS11 and
other RGS proteins specifies binding to Gbeta5 subunits. Proceedings of the National Academy of
Sciences of the United States of America 95, 13307–12 (1998).

90. Kovoor, A. et al. Co-expression of Gβ5 Enhances the Function of Two Gγ Subunit-like
Domain-containing Regulators of G Protein Signaling Proteins. Journal of Biological Chemistry
275, 3397–3402 (2000).

91. Sondek, J. & Siderovski, D. P. Gγ-like (ggl) domains: new frontiers in g-protein signal-

ing and β-propeller scaffolding. Biochemical Pharmacology 61, 1329–1337 (2001).

165

92. Siderovski, D. P., Diversé-Pierluissi, M. A. & De Vries, L. The GoLoco motif: a Gαi/o
binding motif and potential guanine-nucleotide exchange factor. Trends in Biochemical Sciences
24, 340–341 (1999).

93. Willard, M. D. et al. Selective role for RGS12 as a Ras/Raf/MEK scaffold in nerve

growth factor-mediated differentiation. The EMBO Journal 26, 2029–2040 (2007).

94. Kimple, R. J. et al. RGS12 and RGS14 GoLoco Motifs Are Gα <sub>i</sub> Interaction
Sites with Guanine Nucleotide Dissociation Inhibitor Activity. Journal of Biological Chemistry
276, 29275–29281 (2001).

95. Harris, B. Z. & Lim, W. A. Mechanism and role of PDZ domains in signaling complex

assembly. Journal of Cell Science 114, 3219–31 (2001).

96. Gold, S. J., Ni, Y. G., Dohlman, H. G. & Nestler, E. J. Regulators of G-protein signaling
(RGS) proteins: region-specific expression of nine subtypes in rat brain. The Journal of Neuro-
science 17, 8024–8037 (1997).

97. Larminie, C. et al. Selective expression of regulators of G-protein signaling (RGS) in

the human central nervous system. Molecular Brain Research 122, 24–34 (2004).

98. Chen, Y. et al. Neurabin Scaffolding of Adenosine Receptor and RGS4 Regulates Anti-

Seizure Effect of Endogenous Adenosine. Journal of Neuroscience 32, 2683–2695 (2012).

99. Rorabaugh, B. R. et al. Regulators of G-protein signaling 2 and 4 differentially regulate

cocaine-induced rewarding effects. Physiology & Behavior 195, 9–19 (2018).

100. Talbot, J. N. et al. RGS inhibition at G(alpha)i2 selectively potentiates 5-HT1A-
mediated antidepressant effects. Proceedings of the National Academy of Sciences of the United
States of America 107, 11086–91 (2010).

101. Wang, Q., Terauchi, A., Yee, C. H., Umemori, H. & Traynor, J. R. 5-HT1A receptor-
mediated phosphorylation of extracellular signal-regulated kinases (ERK1/2) is modulated by reg-
ulator of G protein signaling protein 19. Cellular Signalling 26, 1846–1852 (2014).

102. Wang, Q. & Traynor, J. R. Modulation of µ-opioid receptor signaling by RGS19 in

SH-SY5Y cells. Molecular Pharmacology 83, 512–20 (2013).

103. Bodle, C. R., Mackie, D. I. & Roman, D. L. RGS17: an emerging therapeutic target for

lung and prostate cancers. Future Medicinal Chemistry 5, 995–1007 (2013).

104. You, M. et al. Fine Mapping of Chromosome 6q23-25 Region in Familial Lung Cancer
Families Reveals RGS17 as a Likely Candidate Gene. Clinical Cancer Research 15, 2666–2674
(2009).

166

105. James, M. A., Lu, Y., Liu, Y., Vikis, H. G. & You, M. RGS17, an Overexpressed Gene
in Human Lung and Prostate Cancer, Induces Tumor Cell Proliferation Through the Cyclic AMP-
PKA-CREB Pathway. Cancer Research 69, 2108–2116 (2009).

106. DeLong, M. R. Primate models of movement disorders of basal ganglia origin. Trends

in Neurosciences 13, 281–285 (1990).

107. Morera-Herreras, T., Miguelez, C., Aristieta, A., Ruiz-Ortega, J. Á. & Ugedo, L. Endo-
cannabinoid modulation of dopaminergic motor circuits. Frontiers in Pharmacology 3, 110 (2012).

108. Giuffrida, A. et al. Dopamine activation of endogenous cannabinoid signaling in

dorsal striatum. Nature Neuroscience 2, 358–363 (1999).

109. Lerner, T. N. & Kreitzer, A. C. RGS4 is required for dopaminergic control of striatal

LTD and susceptibility to parkinsonian motor deficits. Neuron 73, 347–359 (2012).

110. Huang, J., Zhou, H., Mahavadi, S., Sriwai, W. & Murthy, K. S. Inhibition of Gα
<sub>q</sub> -dependent PLC-β1 activity by PKG and PKA is mediated by phosphorylation of
RGS4 and GRK2. American Journal of Physiology-Cell Physiology 292, C200–C208 (2007).

111. Picconi, B. et al. Loss of bidirectional striatal synaptic plasticity in L-DOPA–induced

dyskinesia. Nature Neuroscience 6, 501–506 (2003).

112. Shen, W. et al. M4 muscarinic receptor signaling ameliorates striatal plasticity deficits

in models of L-DOPA-induced dyskinesia. Neuron 88, 762–773 (2015).

113. Roman, D. L. et al.

Identification of small-molecule inhibitors of RGS4 using a
high-throughput flow cytometry protein interaction assay. Molecular Pharmacology 71, 169–175
(2007).

114. Roman, D. L., Ota, S. & Neubig, R. R. Polyplexed Flow Cytometry Protein Interaction
Assay: A Novel High-Throughput Screening Paradigm for RGS Protein Inhibitors. Journal of
Biomolecular Screening 14, 610–619 (2009).

115. Blazer, L. L., Zhang, H., Casey, E. M., Husbands, S. M. & Neubig, R. R. A nanomolar-
potency small molecule inhibitor of regulator of G-protein signaling proteins. Biochemistry 50,
3181–3192 (2011).

116. Turner, E. M., Blazer, L. L., Neubig, R. R. & Husbands, S. M. Small Molecule Inhibitors
of Regulators of G Protein Signaling (RGS) Proteins. ACS Medicinal Chemistry Letters 3, 146–150
(2011).

117. Blazer, L. L. et al. Selectivity and Anti-Parkinson’s Potential of Thiadiazolidinone

RGS4 Inhibitors. ACS Chemical Neuroscience 6, 911–919 (2015).

167

118. Hayes, M. P., Bodle, C. R. & Roman, D. L. Evaluation of the Selectivity and Cysteine
Dependence of Inhibitors across the Regulator of G Protein-Signaling Family. Molecular Pharma-
cology 93, 25–35 (2018).

119. Whitty, A. & Kumaravel, G. Between a rock and a hard place? Nature Chemical

Biology 2, 112–118 (2006).

120. Jin, L., Wang, W. & Fang, G. Targeting Protein-Protein Interaction by Small Molecules.

Annual Review of Pharmacology and Toxicology 54, 435–456 (2014).

121. Oleinikovas, V., Saladino, G., Cossins, B. P. & Gervasio, F. L. Understanding cryptic
pocket formation in protein targets by enhanced sampling simulations. Journal of the American
Chemical Society 138, 14257–14263 (2016).

122. Stank, A., Kokh, D. B., Fuller, J. C. & Wade, R. C. Protein Binding Pocket Dynamics.

Accounts of Chemical Research 49, 809–815 (2016).

123. Gunasekaran, K., Ma, B. & Nussinov, R. Is allostery an intrinsic property of all dy-

namic proteins? Proteins: Structure, Function, and Bioinformatics 57, 433–443 (2004).

124. Christopoulos, A. Allosteric binding sites on cell-surface receptors: novel targets for

drug discovery. Nature Reviews Drug Discovery 1, 198–210 (2002).

125. Mittal, A. & Johnson, M. E. Conformational diversity of bacterial FabH: Implications
for molecular recognition specificity. Journal of Molecular Graphics and Modelling 55, 115–122
(2015).

126. Tesmer, J. J. G. Structure and function of regulator of G protein signaling homology

domains. Progress in Molecular Biology and Translational Science 86, 75–113 (2009).

127. Neubig, R. R. & Siderovski, D. P. Regulators of G-Protein Signalling As New Central

Nervous System Drug Targets. Nature Reviews Drug Discovery 1, 187–197 (2002).

128. Taylor, V. G., Bommarito, P. A. & Tesmer, J. J. G. Structure of the Regulator of G
Protein Signaling 8 (RGS8)-Gαq Complex: MOLECULAR BASIS FOR Gα SELECTIVITY. Journal
of Biological Chemistry 291, 5138–5145 (2016).

129. Vashisth, H., Storaska, A. J., Neubig, R. R. & Brooks 3rd, C. L. Conformational dynam-
ics of a regulator of G-protein signaling protein reveals a mechanism of allosteric inhibition by a
small molecule. ACS Chemical Biology 8, 2778–2784 (2013).

130. Park, I.-H. et al. Estimation of hydrogen-exchange protection factors from MD simu-
lation based on amide hydrogen bonding analysis. Journal of Chemical Information and Modeling
55, 1914–1925 (2015).

168

131. Persson, F. & Halle, B. How amide hydrogens exchange in native proteins. Proceed-
ings of the National Academy of Sciences of the United States of America 112, 10383–10388 (2015).

132. Lee, E., Linder, M. E. & Gilman, A. G. Expression of G-protein alpha subunits in

Escherichia coli. Methods in Enzymology 237, 146–164 (1994).

133. Blazer, L. L., Roman, D. L., Muxlow, M. R. & Neubig, R. R. Use of flow cytometric
methods to quantify protein-protein interactions. Current Protocols in Cytometry 11–13 (2010).
doi:10.1002/0471142956.cy1311s51

134. Chodavarapu, S., Jones, A. D., Feig, M. & Kaguni, J. M. DnaC traps DnaB as an open

ring and remodels the domain that binds primase. Nucleic Acids Research 44, 210–220 (2015).

135. Guttman, M., Weis, D. D., Engen, J. R. & Lee, K. K. Analysis of overlapped and noisy
hydrogen/deuterium exchange mass spectra. Journal of the American Society for Mass Spectrome-
try 24, 1906–1912 (2013).

136. Phillips, J. C. et al. Scalable molecular dynamics with NAMD. Journal of Computa-

tional Chemistry 26, 1781–1802 (2005).

137. Humphrey, W., Dalke, A. & Schulten, K. VMD: visual molecular dynamics. Journal

of Molecular Graphics 14, 33–38 (1996).

138. MacKerell Jr, A. D. et al. All-atom empirical potential for molecular modeling and

dynamics studies of proteins. The Journal of Physical Chemistry B 102, 3586–3616 (1998).

139. MacKerell Jr, A. D., Feig, M. & Brooks, C. L. Extending the treatment of backbone
energetics in protein force fields:
limitations of gas-phase quantum with crystallography and
computation: defining accurate macromolecular structures, conformations and assemblies in so-
lution. Quarterly Reviews of Biophysics 40, 191–285 (2004).

140. Towns, J. et al. XSEDE: accelerating scientific discovery. Computing in Science and

Engineering 16, 62–74 (2014).

141. Soundararajan, M. et al. Structural diversity in the RGS domain and its interaction
with heterotrimeric G protein α-subunits. Proceedings of the National Academy of Sciences of the
United States of America 105, 6457–6462 (2008).

142. Alba, E. de, De Vries, L., Farquhar, M. G. & Tjandra, N. Solution structure of human
GAIP (Gα interacting protein): a regulator of G protein signaling. Journal of Molecular Biology
291, 927–939 (1999).

143. Weis, D. D., Wales, T. E., Engen, J. R., Hotchko, M. & Ten Eyck, L. F. Identification and
Characterization of EX1 Kinetics in H/D Exchange Mass Spectrometry by Peak Width Analysis.

169

Journal of the American Society for Mass Spectrometry 17, 1498–1509 (2006).

144. Shaw, V. S., Mohammadiarani, H., Vashisth, H. & Neubig, R. R. Differential Protein
Dynamics of Regulators of G-Protein Signaling: Role in Specificity of Small-Molecule Inhibitors.
Journal of the American Chemical Society 140, 3454–3460 (2018).

145. Mohammadiarani, H., Shaw, V. S., Neubig, R. R. & Vashisth, H. Interpreting
Hydrogen–Deuterium Exchange Events in Proteins Using Atomistic Simulations: Case Studies
on Regulators of G-Protein Signaling Proteins. The Journal of Physical Chemistry B 122,
9314–9323 (2018).

146. Mohammadi, M., Mohammadiarani, H., Shaw, V. S., Neubig, R. R. & Vashisth, H.
Interplay of cysteine exposure and global protein dynamics in small-molecule recognition by a
regulator of G-protein signaling protein. Proteins: Structure, Function, and Bioinformatics (2018).
doi:10.1002/prot.25642

147. Jorgensen, W. L., Chandrasekhar, J., Madura, J. D., Impey, R. W. & Klein, M. L. Com-
parison of simple potential functions for simulating liquid water. The Journal of Chemical Physics
79, 926–935 (1983).

148. Brown, D. K. et al. MD-TASK: a software suite for analyzing molecular dynamics

trajectories. Bioinformatics 33, 2768–2771 (2017).

149. Roman, D. L. & Traynor, J. R. Regulators of G protein signaling (RGS) proteins as drug
targets: modulating G-protein-coupled receptor (GPCR) signal transduction. Journal of Medicinal
Chemistry 54, 7433–40 (2011).

150. Blazer, L. L. et al. Reversible, allosteric small-molecule inhibitors of regulator of G

protein signaling proteins. Molecular Pharmacology 78, 524–33 (2010).

151. Storaska, A. J. et al. Reversible inhibitors of regulators of G-protein signaling iden-
tified in a high-throughput cell-based calcium signaling assay. Cellular Signalling 25, 2848–55
(2013).

152. Monroy, C. A., Mackie, D. I. & Roman, D. L. A High Throughput Screen for RGS

Proteins Using Steady State Monitoring of Free Phosphate Formation. PLOS ONE 8, (2013).

153. Moy, F. J. et al. NMR structure of free RGS4 reveals an induced conformational change

upon binding Galpha. Biochemistry 39, 7063–73 (2000).

154. Storaska, A. J. & Neubig, R. R. NMR Methods for Detection of Small Molecule Binding

to RGS4. Methods in Enzymology 522, 133–152 (2013).

155. Delaglio, F. et al. NMRPipe: A multidimensional spectral processing system based

170

on UNIX pipes. Journal of Biomolecular NMR 6, 277–293 (1995).

156. Lee, W., Tonelli, M. & Markley, J. L. NMRFAM-SPARKY: enhanced software for

biomolecular NMR spectroscopy. Bioinformatics 31, 1325–1327 (2015).

157. Zhang, G. et al. Protein Quantitation Using Mass Spectrometry. Methods in Molecular

Biology 673, 211 (2010).

158. Britto, P. J., Knipling, L. & Wolff, J. The local electrostatic environment determines

cysteine reactivity of tubulin. The Journal of Biological Chemistry 277, 29018–27 (2002).

159. Jardine, I. Molecular weight analysis of proteins. Methods in Enzymology 193, 441–55

(1990).

160. Ho, C. S. et al. Electrospray ionisation mass spectrometry: principles and clinical

applications. The Clinical Biochemist. Reviews 24, 3–12 (2003).

161. Meister, A. & Anderson, M. E. Glutathione. Annual Review of Biochemistry 52, 711–

760 (1983).

162. Holmgren, A. Thioredoxin and glutaredoxin systems. The Journal of Biological Chem-

istry 264, 13963–6 (1989).

163. Yang, J., Chen, H., Vlahov, I. R., Cheng, J.-X. & Low, P. S. Evaluation of disulfide reduc-
tion during receptor-mediated endocytosis by using FRET imaging. Proceedings of the National
Academy of Sciences 103, 13872–13877 (2006).

164. Boehr, D. D., Nussinov, R. & Wright, P. E. The role of dynamic conformational en-

sembles in biomolecular recognition. Nature Chemical Biology 5, 789–96 (2009).

165. McCammon, J. A. Target flexibility in molecular recognition. Biochimica et Biophysica

Acta - Proteins and Proteomics 1754, 221–224 (2005).

166. Cozzini, P. et al. Target flexibility: an emerging consideration in drug discovery and

design. Journal of Medicinal Chemistry 51, 6237–55 (2008).

167. Rahimova, R. et al. Identification of allosteric inhibitors of the ecto-5’-nucleotidase

(CD73) targeting the dimer interface. PLOS Computational Biology 14, e1005943 (2018).

168. Schmidtke, P., Bidon-Chanal, A., Luque, F. J. & Barril, X. MDpocket: open-source cav-
ity detection and characterization on molecular dynamics trajectories. Bioinformatics 27, 3276–
3285 (2011).

171

169. Le Guilloux, V., Schmidtke, P. & Tuffery, P. Fpocket: an open source platform for

ligand pocket detection. BMC bioinformatics 10, 168 (2009).

170. Liang, J., Woodward, C. & Edelsbrunner, H. Anatomy of protein pockets and cavities:
Measurement of binding site geometry and implications for ligand design. Protein Science 7,
1884–1897 (1998).

171. Schmidtke, P. & Barril, X. Understanding and Predicting Druggability. A High-
Throughput Method for Detection of Drug Binding Sites. Journal of Medicinal Chemistry 53,
5858–5867 (2010).

172. Charrad, M., Ghazzali, N., Boiteau, V. & Niknafs, A. NbClust: An R Package for
Determining the Relevant Number of Clusters in a Data Set. Journal of Statistical Software 61,
1–36 (2014).

173. Sheridan, R. P., Maiorov, V. N., Holloway, M. K., Cornell, W. D. & Gao, Y.-D. Drug-like
Density: A Method of Quantifying the ‘Bindability’ of a Protein Target Based on a Very Large Set
of Pockets and Drug-like Ligands from the Protein Data Bank. Journal of Chemical Information
and Modeling 50, 2029–2040 (2010).

174. Konermann, L., Vahidi, S. & Sowole, M. A. Mass Spectrometry Methods for Studying
Structure and Dynamics of Biological Macromolecules. Analytical Chemistry 86, 213–232 (2014).

175. Mandell, J. G., Baerga-Ortiz, A., Falick, A. M. & Komives, E. A. in Protein-ligand

interactions 305, 065–080 (Humana Press, 2005).

176. Mandell, J. G., Falick, A. M. & Komives, E. A. Identification of protein-protein inter-
faces by decreased amide proton solvent accessibility. Proceedings of the National Academy of
Sciences of the United States of America 95, 14705–10 (1998).

177. Lee, T. et al. Docking motif interactions in MAP kinases revealed by hydrogen ex-

change mass spectrometry. Molecular Cell 14, 43–55 (2004).

178. Bonnington, L. et al. Application of Hydrogen/Deuterium Exchange-Mass Spectrom-
etry to Biopharmaceutical Development Requirements: Improved Sensitivity to Detection of Con-
formational Changes. Analytical Chemistry 89, 8233–8237 (2017).

179. Houde, D., Berkowitz, S. A. & Engen, J. R. The Utility of Hydrogen/Deuterium Ex-
change Mass Spectrometry in Biopharmaceutical Comparability Studies. Journal of Pharmaceuti-
cal Sciences 100, 2071–2086 (2011).

180. Pirrone, G. F., Iacob, R. E. & Engen, J. R. Applications of Hydrogen/Deuterium Ex-

change MS from 2012 to 2014. Analytical Chemistry 87, 99–118 (2015).

172

181. Bobst, C. E. et al. Detection and Characterization of Altered Conformations of Pro-
tein Pharmaceuticals Using Complementary Mass Spectrometry-Based Approaches. Analytical
Chemistry 80, 7473–7481 (2008).

182. Huang, R. Y.-C. & Chen, G. Higher order structure characterization of protein thera-
peutics by hydrogen/deuterium exchange mass spectrometry. Analytical and Bioanalytical Chem-
istry 406, 6541–6558 (2014).

183. Barksdale, A. D. & Rosenberg, A. Acquisition and interpretation of hydrogen ex-
change data from peptides, polymers, and proteins. Methods of Biochemical Analysis 28, 1–113
(1982).

184. Abaturov, L., Jinoria, K., Varshavsky, Y. & Yakobashvily, N. Effect of ligand and heme
on conformational stability (intramolecular conformational motility) of hemoglobin as revealed
by hydrogen exchange. FEBS Letters 77, 103–106 (1977).

185. Hvidt, A. & Linderstrøm-Lang, K. The kinetics of the deuterium exchange of insulin

with D2O. An amendment. Biochimica et Biophysica Acta 16, 168–169 (1955).

186. Deuterium Exchange of Poly-DL-alanine in Aqueous Solution. 69, 106–118 (1957).

187. Bryan, W. P. & Nielsen, S. O. Hydrogen-deuterium exchange of poly-dl-alanine in

aqueous solution. Biochimica et Biophysica Acta 42, 552–553 (1960).

188. Molday, R. S., Englander, S. W. & Kallen, R. G. Primary structure effects on peptide

group hydrogen exchange. Biochemistry 11, 150–158 (1972).

189. Woodward, C. K. & Hilton, B. D. Hydrogen Exchange Kinetics and Internal Motions
in Proteins and Nucleic Acids. Annual Review of Biophysics and Bioengineering 8, 99–127 (1979).

190. Englander, S. W., Downer, N. W. & Teitelbaum, H. Hydrogen Exchange. Annual

Review of Biochemistry 41, 903–924 (1972).

191. Hvidt, A. & Nielsen, S. O. Hydrogen Exchange in Proteins. Advances in Protein Chem-

istry 21, 287–386 (1966).

192. Takahashi, T., Nakanishi, M. & Tsuboi, M. Hydrogen–deuterium exchange study of
amino acids and proteins by 200- to 230-nm spectroscopy. Analytical Biochemistry 110, 242–9
(1981).

193. Hilton, B. D. & Woodward, C. K. On the mechanism of isotope exchange kinetics of

single protons in bovine pancreatic trypsin inhibitor. Biochemistry 18, 5834–41 (1979).

173

194. Ellis, L. M., Bloomfield, V. A. & Woodward, C. K. Hydrogen-tritium exchange kinetics
of soybean trypsin inhibitor (Kunitz). Solvent accessibility in the folded conformation. Biochem-
istry 14, 3413–9 (1975).

195. Woodward, C. K., Ellis, L. M. & Rosenberg, A. Solvent accessibility in folded proteins.
Studies of hydrogen exchange in trypsin. The Journal of Biological Chemistry 250, 432–9 (1975).

196. Woodward, C. K. & Rosenberg, A. Studies of hydrogen exchange in proteins. VI. Urea
effects on ribonuclease exchange kinetics leading to a general model for hydrogen exchange from
folded proteins. The Journal of Biological Chemistry 246, 4114–21 (1971).

197. Rosenberg, A. & Chakravarti, K. Studies of hydrogen exchange in proteins. I. The
exchange kinetics of bovine carbonic anhydrase. The Journal of Biological Chemistry 243, 5193–
201 (1968).

198. Levitt, M. HYDROGEN BOND AND INTERNAL SOLVENT DYNAMICS OF BPTI

PROTEIN. Annals of the New York Academy of Sciences 367, 162–181 (1981).

199. Eden, D., Matthew, J. B., Rosa, J. J. & Richards, F. M. Increase in apparent compress-
ibility of cytochrome c upon oxidation. Proceedings of the National Academy of Sciences of the
United States of America 79, 815–9 (1982).

200. Bai, Y., Milne, J. S., Mayne, L. & Englander, S. W. Primary structure effects on peptide

group hydrogen exchange. Proteins: Structure, Function, and Genetics 17, 75–86 (1993).

201. Adhikary, S. et al. Conformational dynamics of a neurotransmitter:sodium symporter

in a lipid bilayer. Proceedings of the National Academy of Sciences 114, E1786–E1795 (2017).

202. Brock, A. Fragmentation hydrogen exchange mass spectrometry: A review of

methodology and applications. Protein Expression and Purification 84, 19–37 (2012).

203. Zhang, Q. et al. Epitope Mapping of a 95 kDa Antigen in Complex with Antibody by
Solution-Phase Amide Backbone Hydrogen/Deuterium Exchange Monitored by Fourier Trans-
form Ion Cyclotron Resonance Mass Spectrometry. Analytical Chemistry 83, 7129–7136 (2011).

204. Englander, S. W., Sosnick, T. R., Englander, J. J. & Mayne, L. Mechanisms and uses of

hydrogen exchange. Current Opinion in Structural Biology 6, 18–23 (1996).

205. Maity, H., Lim, W. K., Rumbley, J. N. & Englander, S. W. Protein hydrogen exchange
mechanism: Local fluctuations. Protein Science: A Publication of the Protein Society 12, 153 (2003).

206. Wyttenbach, T. & Bowers, M. T. Gas phase conformations of biological molecules: the
hydrogen/deuterium exchange mechanism. Journal of the American Society for Mass Spectrometry
10, 9–14 (1999).

174

207. Miller, D. W. & Dill, K. A. A statistical mechanical model for hydrogen exchange in

globular proteins. Protein Science: a Publication of the Protein Society 4, 1860–73 (1995).

208. Vendruscolo, M., Paci, E., Dobson, C. M. & Karplus, M. Rare Fluctuations of Native
Proteins Sampled by Equilibrium Hydrogen Exchange. Journal of the American Chemical Society
125, 15686–15687 (2003).

209. Best, R. B. & Vendruscolo, M. Structural Interpretation of Hydrogen Exchange Pro-
tection Factors in Proteins: Characterization of the Native State Fluctuations of CI2. Structure 14,
97–106 (2006).

210. Kieseritzky, G., Morra, G. & Knapp, E.-W. Stability and fluctuations of amide hy-
drogen bonds in a bacterial cytochrome c: a molecular dynamics study. Journal of Biological
Inorganic Chemistry 11, 26–40 (2006).

211. Skinner, J. J., Lim, W. K., Bédard, S., Black, B. E. & Englander, S. W. Protein hydrogen

exchange: Testing current models. Protein Science 21, 987–995 (2012).

212. Ma, B. & Nussinov, R. Polymorphic triple beta-sheet structures contribute to amide
hydrogen/deuterium (H/D) exchange protection in the Alzheimer amyloid beta42 peptide. The
Journal of Biological Chemistry 286, 34244–53 (2011).

213. Bentley, G. A. et al. Exchange of individual hydrogens for a protein in a crystal and

in solution. Journal of Molecular Biology 170, 243–7 (1983).

214. Clarke, J., Itzhaki, L. S. & Fersht, A. R. Hydrogen exchange at equilibrium: a short

cut for analysing protein-folding pathways? Trends in Biochemical Sciences 22, 284–7 (1997).

215. Liu, T. et al. Quantitative Assessment of Protein Structural Models by Comparison
of H/D Exchange MS Data with Exchange Behavior Accurately Predicted by DXCOREX. Journal
of The American Society for Mass Spectrometry 23, 43–56 (2012).

216. Kuwata, K. et al. NMR-detected hydrogen exchange and molecular dynamics simula-
tions provide structural insight into fibril formation of prion protein fragment 106-126. Proceed-
ings of the National Academy of Sciences 100, 14790–14795 (2003).

217. Craig, P. O. et al. Prediction of Native-State Hydrogen Exchange from Perfectly
Funneled Energy Landscapes. Journal of the American Chemical Society 133, 17463–17472 (2011).

218. Petruk, A. A. et al. Molecular Dynamics Simulations Provide Atomistic Insight into
Hydrogen Exchange Mass Spectrometry Experiments. Journal of Chemical Theory and Computa-
tion 9, 658–669 (2013).

219. Hsu, Y.-H. et al. Fluoroketone Inhibition of Ca2+ -Independent Phospholipase A2

175

through Binding Pocket Association Defined by Hydrogen/Deuterium Exchange and Molecular
Dynamics. Journal of the American Chemical Society 135, 1330–1337 (2013).

220. Radou, G., Dreyer, F. N., Tuma, R. & Paci, E. Functional Dynamics of Hexameric
Helicase Probed by Hydrogen Exchange and Simulation. Biophysical Journal 107, 983–990 (2014).

221. McAllister, R. G. & Konermann, L. Challenges in the Interpretation of Protein H/D Ex-
change Data: A Molecular Dynamics Simulation Perspective. Biochemistry 54, 2683–2692 (2015).

222. McAllister, R. G. & Lars Konermann, S. From Solution Into the Gas Phase: Studying
Protein Hydrogen Exchange and Electrospray Ionization Using Molecular Dynamics Simulation.
(The University of Western Ontario, 2015).

223. Mao, Y. et al. Hydrogen/Deuterium Exchange and Molecular Dynamics Analysis of
Amyloid Fibrils Formed by a D69K Charge-Pair Mutant of Human Apolipoprotein C-II. Biochem-
istry 54, 4805–4814 (2015).

224. Khakinejad, M., Kondalaji, S. G., Donohoe, G. C. & Valentine, S. J. Ion Mobility
Spectrometry-Hydrogen Deuterium Exchange Mass Spectrometry of Anions: Part 2. Assessing
Charge Site Location and Isotope Scrambling. Journal of the American Society for Mass Spectrom-
etry 27, 451–61 (2016).

225. Resing, K. A., Hoofnagle, A. N. & Ahn, N. G. Modeling deuterium exchange behavior
of ERK2 using pepsin mapping to probe secondary structure. Journal of the American Society for
Mass Spectrometry 10, 685–702 (1999).

226. Hilser, V. J. & Freire, E. Structure-based Calculation of the Equilibrium Folding Path-
way of Proteins. Correlation with Hydrogen Exchange Protection Factors. Journal of Molecular
Biology 262, 756–772 (1996).

227. García, A. E. & Hummer, G. Conformational dynamics of cytochrome c: correlation

to hydrogen exchange. Proteins 36, 175–91 (1999).

228. Shaw, D. E. et al. Atomic-Level Characterization of the Structural Dynamics of Pro-

teins. Science 330, 341–346 (2010).

229. Van Der Walt, S., Colbert, S. C. & Varoquaux, G. The NumPy array: a structure for

efficient numerical computation. Computing in Science and Engineering 13, 22–30 (2011).

230. Maier, J. A. et al. ff14SB: Improving the Accuracy of Protein Side Chain and Backbone

Parameters from ff99SB. Journal of Chemical Theory and Computation 11, 3696–3713 (2015).

231. Milne, J. S., Mayne, L., Roder, H., Wand, A. J. & Englander, S. W. Determinants of
protein hydrogen exchange studied in equine cytochrome c. Protein Science: A Publication of the

176

Protein Society 7, 739–45 (1998).

232. Yung-Chi, C. & Prusoff, W. H. Relationship between the inhibition constant (KI) and
the concentration of inhibitor which causes 50 per cent inhibition (I50) of an enzymatic reaction.
Biochemical Pharmacology 22, 3099–3108 (1973).

177