Synthetic flavonoid derivatives targeting the glycogen phosphorylase inhibitor site: QM/MM-PBSA motivated synthesis of substituted 5,7-dihydroxyflavones, crystallography, in vitro kinetics and ex-vivo cellular experiments reveal novel potent inhibitors

Article abstract of DOI:10.1016/j.bioorg.2020.104003

Glycogen phosphorylase (GP) is an important target for the development of new anti-hyperglycaemic agents. Flavonoids are novel inhibitors of GP, but their mode of action is unspecific in terms of the GP binding sites involved. Towards design of synthetic

Full text of DOI:10.1016/j.bioorg.2020.104003

Journal Pre-proofs
Synthetic flavonoid derivatives targeting the glycogen phosphorylase inhibitor
site: QM/MM-PBSA motivated synthesis of substituted 5,7-dihydroxyfla-
vones, crystallography, in vitro kinetics and ex-vivo cellular experiments re-
veal novel potent inhibitors
Ben A. Chetter, Efthimios Kyriakis, Daniel Barr, Aikaterini G. Karra,
Elisabeth Katsidou, Symeon M. Koulas, Vassiliki T. Skamnaki, Timothy J.
Snape, Anna-Maria G. Psarra, Demetres D. Leonidas, Joseph M. Hayes
PII:
S0045-2068(20)31300-6
DOI:
https://doi.org/10.1016/j.bioorg.2020.104003
YBIOO 104003
Reference:
To appear in:
Bioorganic Chemistry
Received Date:
Revised Date:
Accepted Date:
16 April 2020
2 June 2020
7 June 2020
Please cite this article as: B.A. Chetter, E. Kyriakis, D. Barr, A.G. Karra, E. Katsidou, S.M. Koulas, V.T.
Skamnaki, T.J. Snape, A.G. Psarra, D.D. Leonidas, J.M. Hayes, Synthetic flavonoid derivatives targeting the
glycogen phosphorylase inhibitor site: QM/MM-PBSA motivated synthesis of substituted 5,7-dihydroxyflavones,
crystallography, in vitro kinetics and ex-vivo cellular experiments reveal novel potent inhibitors, Bioorganic
Chemistry (2020), doi: https://doi.org/10.1016/j.bioorg.2020.104003
This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover
page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version
will undergo additional copyediting, typesetting and review before it is published in its final form, but we are
providing this version to give early visibility of the article. Please note that, during the production process, errors
may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
© 2020 Elsevier Inc. All rights reserved.

Synthetic flavonoid derivatives targeting the glycogen phosphorylase
inhibitor site: QM/MM-PBSA motivated synthesis of substituted 5,7-
dihydroxyflavones, crystallography, in vitro kinetics and ex-vivo cellular
experiments reveal novel potent inhibitors
Ben A. Chetter,a# Efthimios Kyriakis,b# Daniel Barr,a# Aikaterini G. Karra,b Elisabeth
Katsidou,b Symeon M. Koulas,b Vassiliki T. Skamnaki,b Timothy J. Snape,*c Anna-Maria G.
Psarra*b, Demetres D. Leonidas,*b Joseph M. Hayes*c
a School of Physical Sciences & Computing, University of Central Lancashire, Preston PR1
2HE, United Kingdom
b Department of Biochemistry & Biotechnology, University of Thessaly, Biopolis, 41500
Larissa, Greece
c School of Pharmacy & Biomedical Sciences, University of Central Lancashire, Preston PR1
2HE, United Kingdom
* Corresponding authors
tel: +441772894334 (JMH), fax: +441772894981; email: jhayes@uclan.ac.uk
tel: +302410565278 (DDL), fax: +302410565290; email: ddleonidas@bio.uth.gr
tel: +302410565221 (AMP), fax: +302410565290; email: ampsarra@bio.uth.gr
tel: +441772895805 (TJS), fax: +441772894981; email: tjsnape@uclan.ac.uk
# Equal contribution.
1

Abstract
Glycogen phosphorylase (GP) is an important target for the development of new anti-
hyperglycaemic agents. Flavonoids are novel inhibitors of GP, but their mode of action is
unspecific in terms of the GP binding sites involved. Towards design of synthetic flavonoid
analogues acting specifically at the inhibitor site and to exploit the site’s hydrophobic pocket,
chrysin has been employed as a lead compound for the in silico screening of 1169 new
analogues with different B ring substitutions. QM/MM-PBSA binding free energy
calculations guided the final selection of eight compounds, subsequently synthesised using a
Baker-Venkataraman rearrangement-cyclisation approach. Kinetics experiments against
rabbit muscle GPa and GPb together with human liver GPa, revealed three of these
compounds (11, 20 and 43) among the most potent that bind at the site (Ki s < 4 µM for all
three isoforms), and more potent than previously reported natural flavonoid inhibitors.
Multiple inhibition studies revealed binding exclusively at the inhibitor site. The binding is
synergistic with glucose suggesting that inhibition could be regulated by blood glucose levels
and would decrease as normoglycaemia is achieved. Compound 43 was an effective inhibitor
of glycogenolysis in hepatocytes (IC50 = 64 µM), further promoting these compounds for
optimization of their drug-like potential. X-ray crystallography studies revealed the B-ring
interactions responsible for the observed potencies.
Keywords: Flavonoids; halogens; natural products; QM/MM-PBSA; X-ray crystallography;
type 2 diabetes.
2

1. Introduction
Type 2 Diabetes (T2D) is a chronic heterogeneous disease characterised by hyperglycaemia.
Its prevalence has been rising rapidly, but particularly in low- and middle-income countries
[1]. T2D is characterized as a disorder of insulin secretion, insulin resistance and by increased
production of hepatic glucose. Glycogenolysis may account for more than 70% of the hepatic
glucose production. Additionally, a substantial amount of glucose from gluconeogenesis is
cycled through the glycogen pool prior to efflux from the liver cells [2, 3]. The rate-
determining enzyme in the glycogenolysis pathway is glycogen phosphorylase (GP; EC
2.4.1.1), rendering GP a very promising and validated target for the development of much
needed more effective T2D treatments [4]. However, GP inhibition has also been recognised
as having significant potential for the treatment of other conditions such as cancer,
myocardial and cerebral ischemias [5-7].
GP is an allosteric enzyme and has a number of different binding sites [8]: the catalytic,
allosteric, new allosteric, inhibitor, quercetin binding,[9] benzimidazole[10] and the glycogen
storage sites. It exists in two interconvertible forms, the phosphorylated GPa form which is
the predominantly active R state and with high substrate affinity, and the unphosphorylated
GPb form which is the predominantly inactive T state (low substrate affinity) [4]. Phytogenic
chemicals are a promising source of compounds with the potential to inhibit GP [11, 12].
Natural flavonoids have been revealed as a good source of GP inhibitors in a number of
studies [13-16], which have recently been reviewed [11, 12]. They are ubiquitous
components of fruit and vegetables and are consumed in our daily diets. Significantly,
studies in diabetic and non-diabetic rats have shown that the anti-hyperglycaemic properties
of flavonoids are due to their effect on glycogen metabolism [17-19]. Jakobs et al. extensive
study of 24 flavonoids demonstrated that most of the compounds investigated inhibited rabbit
muscle GPa (rmGPa), and that inhibition of GPa was two to four times better than for rmGPb
[13]. This inhibition of the active GPa form is favourable in terms of potential for blood
glucose regulation in T2D patients.
Structurally, flavonoids have a general scaffold consisting of two aromatic rings (A and B),
linked by three carbon atoms that are usually in a heterocyclic ring (C) with oxygen (c.f.
chrysin, Figure 1). The differences in ring C leads to the structural classification of flavonoids
as flavones, flavanols, flavanones, anthocyanidins and isoflavonoids [11]. The best flavonoid
inhibitors of GP are on the low µM range, some of which are shown in Figure 1. These
3

include the flavonols quercetin and quercetagetin; flavones such as 6-hydroxyluteolin and
chrysin. Certain anthocyanidins such as cyanadin and catechins such as epigallocatechin
gallate (EGCG), are also potent inhibitors. However, the inhibitory potential of different
flavonoids is very sensitive to the substitution patterns within the different structural classes
[11].
3'
O H
O H
O H
O H
2'
4'
5'
B
HO
O
O
HO
O
C
HO
HO
O
O
6'
A
O H
O H
O H
O
O H
O H
Chrysin
IC50(GPa) = 27.5 [13]
quercetin
IC50(GPa) = 4.8 [13]
quercetagetin
IC50(GPb) = 9.7 [14], (3.5(Ki)) [14]
IC50 (GPb) = 15.3 [13], (19.01
(Ki)) [16]
Binding site: inhibitor site
IC50(GPb) = 20.9 [13], 33.5 [14]
Binding site: quercetin [9]
Binding site: allosteric
O H
O H
O H
O H
O H
HO
O
O H
+
HO
HO
O
O
HO
O
O H
O
O H
O H
O
O H
O H
O H
O H
O H
6-hydroxyluteolin
IC50(GPb) = 11.6 [13]
Binding site: unknown
cyanidin
EGCG
IC50(GPa) = 7.7 [13]
IC50(GPb) = 33.9 [13], 34.0 [20]
Binding site: unknown
IC50(GPa) = 3.0 [13]
IC50(GPb) = 9.0 [13]
Binding site: unknown
OH OH
O H O H
OH OH
Cl
Cl
HO
O
HO
O
HO
O
O H
N
N
N
C H3
C H3
C H3
Flavopiridol
Ki (GPb) = 1.16 [21], 1.24 [16]
Binding site: inhibitor site
flavopiridol analogue
Ki (GPb) = 1.28[22]
Binding site: inhibitor site*
flavopiridol analogue
Ki (GPb) = 0.83[22]
Binding site: inhibitor site*
4

Figure 1. Inhibition constants (µM) for some of the most potent natural flavonoid inhibitors
of glycogen phosphorylase known to date together with their inhibition data and binding
sites. The flavonoid ring labels A-C and atom numbering scheme used throughout text is
shown for chrysin. Also displayed is the synthetic flavonoid derivative flavopiridol, together
with two of its analogues. (* = assumed binding site). The GPb dimer is also shown with the
various binding sites indicated.
Previously it was assumed that flavonoids may display their inhibitory potential of GP by a
similar mode of action. However, it is now evident from recent studies [9, 14-16] that their
binding site is dependent on the flavonoid specific substitution patterns, more specifically the
presence or absence of different hydroxyl groups. Indeed, the GP binding site of most
flavonoids is unknown [11]. However, the binding of chrysin to GP has been studied in
detail; its solved crystal structure complex with rmGPb revealed binding exclusively at the
inhibitor site [16].
The inhibitor site of GP, also known as the caffeine (Ki ~ 0.1 mM [23, 24]) binding site, is
found on the surface of the enzyme, approximately 12 Å from the catalytic site. When the
enzyme is in the less active T state, Phe285, from the 280s loop (residues 282-287) is stacked
close to Tyr613, from the α19 helix (residues 613-631). These two aromatic residues form the
5

core of the inhibitor site. The ligands that bind to this site are considerably influenced by
these residues due to the potential for an inhibitor to form a sandwich-like complex with GP,
exploiting π-stacking interactions with both Phe285 and the Tyr613 sidechains (Fig. 2). This
promotes the T state conformation of the enzyme via stabilisation of the closed conformation
of the 280s loop, which in turn blocks access to the catalytic site. Inhibition of the binding
site generally acts synergistically with glucose, which suggests that inhibition could be
regulated by blood glucose levels, and would decrease as normoglycaemia is achieved
without causing hypoglycaemia.[25] Because of this property, design of inhibitors targeting
the site is particularly attractive from a drug design perspective. The most potent reported
inhibitor binding at the site is flavopiridol (Figure 1; Ki ~ 1.2 µM [16, 21]) with slightly better
inhibition obtained in a study of its analogues (best Ki = 0.83 µM) [22]. Natural products such
as flavonoids provide scaffolds for design of potent GP inhibitors but the only flavonoid
ligand confirmed to bind at the site is chrysin, with a previously reported Ki of 19.01 µM
[16].
Figure 2. Binding of chrysin at the inhibitor site of rmGPb (PDB entry: 3EBO[16])
highlighting the key π-stacking interactions involved with residues Phe285 and Tyr613 and
also the hydrophobic pocket key residues Ile380, Tyr573 and Phe771 in close proximity to
the phenyl ring B of chrysin.
Of particular importance to binding of flavonoids at this site is a hydrophobic pocket lined by
Ile380, Tyr573 and Phe771, as highlighted in Fig. 2. We have previously proposed that
flavonoids with ring B polar OH substituents (in particular at position 4’) are unfavourable
for inhibitor site binding on the basis of quantum mechanics – molecular mechanics/Poisson
6

Boltzmann surface area (QM/MM-PBSA) binding free energy (BFE;
푏푖푛푑) calculations
∆퐺
[16]. It was demonstrated that quercetagetin with 3’ and 4’ OH substituents (Fig. 1) displayed
poor binding at the site based on its calculated binding affinity, and this was in agreement
with multiple inhibition studies proposing allosteric site binding [14]. This theory was further
validated by the subsequent discovery of the quercetin binding site (PDB entry 4MRA)
discovered through the screening of polyphenolic extracts from the vinification byproducts of
Vitis vinifera [9]. These binding features have motivated the current work to design more
potent synthetic flavonoid analogues of the lead compound chrysin with different ring B that
act specifically at the inhibitor site.
In this paper, we present the virtual screening of a library of 1169 compounds exploiting the
ZINC database (http://zinc.docking.org)[26] using chrysin as the lead compound and its 7-
dihydroxy-4H-chromen-4-one moiety as the core scaffold for the design of more effective B
ring substituents (Fig. 3). Initial screening was performed using Glide docking [27] in both
standard (Glide-SP) and extra precision (Glide-XP) modes but combining the results in a
consensus approach both in terms of predicted binding affinities [28, 29] and binding modes
[30], aimed at reducing the number of false positive predictions. The top-ranked predicted
compounds considered to be synthetically viable were then post-processed using QM/MM-
PBSA binding free energy calculations. Eight candidates were selected for synthesis and
extensive kinetics studies performed. Crystallographic studies revealed the interactions that
govern the observed potencies. Additionally, the effectiveness of the most potent inhibitor for
reduction of glycogenolysis at the cellular level was explored.
7

HO
O
O
O
Cl
+
R
O H
5,7-dihydroxy-4H-chromen-4-one
core
Benzoyl chloride substructure
(1169 structures)
CombiGlide
↓
R
HO
O
O H
O
Figure 3. The 7-dihydroxy-4H-chromen-4-one moiety of chrysin and the benzoyl chloride
substructure hits from searching the ZINC docking database[26] were combined using
CombiGlide to yield the final set of 1169 ligands for virtual screening.
2. Results and Discussions
2.1 In Silico Screening
The aim of the calculations was to screen 5,7-dihydroxyflavones for favourable substitutions
on ring B. Using the ZINC database [26] and CombiGlide program [27], we generated 1169
unique ligands for initial docking calculations (Fig. 3). The ligands were docked to two
models of GPb, model 1 from its solved crystallographic complex with chrysin (PDB entry
3EBO[16]) and model 2, an induced-fit docking GPb model for the predicted complex with
5,7-dihydroxy-2-(naphthalen-2-yl)-4H-chromen-4-one (B phenyl ring of chrysin substituted
by a 2-naphthalene group). The latter model aimed to refine the hydrophobic pocket region to
accept larger B ring substituents (Fig. 2). For each protein model, the ligands were docked
using both Glide-SP and –XP [27], and the predicted potency ranks from each method
combined using consensus scoring. Additional consensus of the Glide-SP and –XP docking
poses in terms of ligand binding modes was used to further filter ligands, as described in the
experimental section, given its potential/propensity to further reduce false positive
8

predictions [30]. This approach predicted 79 and 257 analogues as more potent than chrysin
using models 1 and 2, respectively. As anticipated, model 2 allowed larger substituents on the
B ring to occupy a more open hydrophobic pocket.
End point methods such as MM-PB(GB)SA are an effective post-docking strategy for the
calculation of binding free energies [31]. QM/MM-PB(GB)SA methods, however, are
gaining more prominence due to potential for better description of binding site interactions
using QM methods [32-35], and have previously been successfully applied to flavonoid
binding at the GP inhibitor site where a QM description was necessary to describe the π-
stacking interactions of ligands with residues Phe285 and Tyr613 [16]. Here, preliminary
QM/MM-PBSA calculations on the top-ranked 200 ligands from docking for each of model 1
and model 2 were performed. Protein-ligand complex geometries that gave
푏푖푛푑 values < -
∆퐺
11 kcal/mol (a threshold chosen based on its proximity to the
value of -12.35
∆퐺푏푖푛푑
kcal/mol for chrysin) were refined using Prime 4.2 [27] (58 ligands) before a final set of
QM/MM-PBSA binding free energy calculations. As expected, the final values were
∆퐺푏푖푛푑
lower for these optimized complex structures and 36 analogues were predicted as more potent
than the lead compound chrysin (-14.12 kcal/mol) (Table 1).
Analysis of the 36 flavonoid derivatives in Table 1 revealed 12, 22 and 2 ligands having
mono-, di- and tri-substituted B rings, respectively. These substitutions are mainly in the B
ring ortho (21 ligands) and meta (25 ligands) positions, with 10 ligands having para
substitutions. The majority of substitutions are with halogens (28 ligands), particularly with
fluorine (15 ligands). 5 ligands have di-fluoro B ring substitutions. Fluorine is widely used in
the lead optimization phase of drug design projects due to its potential to enhance drug-like
properties such as permeability [36]. This has particular significance for flavonoids where
oral bioavailability remains one of the most critical issues for their role in nutrition and health
[37]. 10, 9 and 2 derivatives contained one Cl, Br or I substituent, respectively. A smaller size
halogen, therefore, appeared to be potentially more favourable. One compound had di-chloro
substitutions (meta- and para-positions). Other B ring substituents for the 36 compounds
were methyl (2 ligands), amine (2 ligands), aldehyde (CHO) group (1 ligand) and hydroxyl (7
ligands).
Analysing the breakdown of
푏푖푛푑 (Eq.(1); c.f. Experimental Section) for the top 36 ligands
∆퐺
(Table 1), the 푄푀/푀푀 values representing the gas phase interaction energy values dominate
∆퐸
with values between -55.03 and -79.51 kcal/mol. The strongest
푄푀/푀푀 interactions
∆퐸
9

(compound 16) have polar hydroxyl groups in the 2’ and 3’ positions; the corresponding least
favourable (compound 29) has fluorine at the meta positions (3’ and 5’). For ligands that are
halogen substituted only, the best 푄푀/푀푀 values are -60.71 kcal/mol for compound 10 with
∆퐸
fluorine (3’) and bromine (4’) substitutions, and -59.63 kcal/mol for compound 11 with
fluorine (3’) and chlorine (4’) substitutions. However, more favourable
values
∆퐸푄푀/푀푀
generally had less favourable solvation free energy changes ( 푠표푙푣) values (range of +21.23
∆퐺
– +45.77 kcal/mol). The larger desolvation penalties were associated with the more polar
groups, as demonstrated by the larger polar contributions to the solvation free energy changes
푃퐵
푃퐵
(
term, predominantly electrostatic solvation enthalpy effects).
values ranged
푠표푙푣
∆퐺푠표푙푣
∆퐺
from +26.09 (compound 1) to +50.38 kcal/mol (compound 16). Meanwhile, the non-polar
푆퐴
contributions ( 푠표푙푣) to the solvation free energy changes were small and similar for all
∆퐺
ligands (range -4.39 – -5.11 kcal/mol). A sensitive balance between
and
푠표푙푣 for
∆퐺
∆퐸푄푀/푀푀
the identification of new potent inhibitors was therefore observed. For example, compound
16 with 2’ and 3’ OH substitutions had the most favourable 푄푀/푀푀 value (-79.51 kcal/mol)
∆퐸
value (+45.77 kcal/mol) and was ranked sixteen; ligand (1)
but the least favourable
∆퐺푠표푙푣
with a 4’ Br substitution had the least desolvation penalty ( 푠표푙푣= 21.23 kcal/mol) but much
∆퐺
less favourable 푄푀/푀푀 value (-59.29 kcal/mol) yet was ranked first. Entropy changes, TΔS,
∆퐸
on the other hand had a range of just ~ 2 kcal/mol (-13.37 to -15.99 kcal/mol) and did not
influence as much the ligand rankings.
Considering synthetic viability as well as the availability of reagents, eight compounds were
selected for synthesis and subsequent experimental validation, none of which gave warnings
for Pan Assay Interference Compounds (PAINS) [38], as determined using the ZINC on-line
filter (http://zinc15.docking.org/patterns/home/). Predicted inhibitors ranked 1 (4’ = Br) and 2
(2’ = OH) were chosen, together with fluoro-substituted 9 (3’ = F) and 20 (4’ = F). Two di-
substituted ligands ranked 10 (3’ = Br; 4’ = F), and 11 (3’ = Cl; 4’ = F) were also selected.
Although not in the top 36, we additionally chose two other mono-substituted ligands with
methyl substituents in the meta 3’ (43) and para 4’ positions (52). Their
푏푖푛푑 values were -
∆퐺
13.47 kcal/mol and -12.01 kcal/mol, respectively. They were ranked in ~ the top 50 from an
initial 1169 ligands and their simple non-polar substituents considered to have potential to
exploit the hydrophobic cavity from visual inspection of the predicted complexes.
10

a
∆퐺
Table 1: QM/MM-PBSA results showing the breakdown of contributions to
.
푏푖푛푑
3'
4'
5'
2'
HO
O
QM/MM-PBSA
(kcal/mol)
Ligand
/ rankb
6'
O H
O
푃퐵
푆퐴
푡표푡푎푙
∆퐸푄푀/푀푀
-59.29
-70.83
-57.88
-59.96
-65.72
-65.65
-59.07
-57.08
-56.08
-60.71
-59.63
-57.21
-57.34
-57.07
-71.17
-79.51
-59.11
-56.89
-59.52
-59.04
-56.32
-57.31
-56.43
-58.70
-56.23
-71.56
-71.88
-55.13
-55.03
-59.81
-56.20
-58.38
-70.06
-58.52
-55.62
-57.39
∆퐺푏푖푛푑
-23.37
-22.74
-21.82
-21.50
-20.89
-19.29
-19.18
-19.13
-19.02
-18.87
-18.73
-18.39
-18.38
-18.15
-17.78
-17.74
-17.19
-17.12
-17.01
-16.93
-16.92
-16.89
-16.54
-15.87
-15.76
-15.65
-15.53
-15.27
-15.03
-14.87
-14.86
-14.65
-14.45
-14.36
-14.29
-14.25
∆퐺
∆퐺
∆퐺
푠표푙푣
2’
H
3’
H
H
H
H
4’
5’
H
H
H
Br
H
Cl
H
H
H
H
H
H
I
6’
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
F
푇∆푆
푠표푙푣
푠표푙푣
1*
2*
3
Br
H
H
H
H
H
H
Cl
H
F
26.09
38.18
27.20
28.66
34.82
35.80
30.53
28.14
28.45
32.43
31.45
29.59
29.39
29.55
43.26
50.38
32.04
30.59
32.78
32.80
29.18
31.07
30.63
33.86
30.86
45.10
46.16
30.35
30.39
34.58
32.22
33.94
45.21
34.48
31.98
33.50
-4.86
-4.78
-4.98
-4.93
-4.96
-4.58
-4.55
-4.69
-4.76
-5.14
-4.79
-4.47
-5.06
-4.95
-5.11
-4.61
-4.91
-4.94
-4.69
-4.76
-4.42
-4.76
-4.39
-5.03
-4.73
-4.62
-4.67
-4.92
-4.60
-4.96
-4.86
-4.85
-4.67
-4.67
-4.92
-4.71
21.23
33.40
22.21
23.73
29.86
31.22
25.98
23.45
23.69
27.29
26.66
25.13
24.33
24.60
38.15
45.77
27.13
25.66
28.09
28.04
24.76
26.31
26.24
28.83
26.13
40.48
41.49
25.43
25.79
29.62
27.36
29.09
40.54
29.81
27.06
28.79
-14.69
-14.69
-13.85
-14.73
-14.96
-15.14
-13.90
-14.49
-13.37
-14.55
-14.24
-13.69
-14.63
-14.32
-15.24
-15.99
-14.79
-14.11
-14.42
-14.07
-14.64
-14.11
-13.64
-14.00
-14.33
-15.43
-14.86
-14.43
-14.20
-15.33
-13.98
-14.64
-15.07
-14.36
-14.27
-14.34
OH
Cl
F
4
5
OH
NH2
Br
F
CH3
H
6
7
H
8
H
9*
10*
11*
12
13
14
15
16
17
18
19
20*
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
H
F
H
Br
Cl
H
H
F
F
F
F
H
H
H
H
H
Cl
H
H
F
F
H
Cl
Cl
H
H
H
H
H
F
OH
OH
H
H
OH
Cl
Cl
NH2
H
H
H
H
H
Br
Br
H
H
H
H
F
H
H
H
H
H
H
F
H
F
F
Br
H
H
Cl
H
H
H
F
OH
H
OH
H
OH
Br
F
F
H
F
H
H
H
H
H
H
H
H
F
H
CH3
F
F
H
H
H
H
H
H
H
Cl
H
H
Br
H
OH
H
CHO
F
H
Cl
I
H
H
H
11

37
H
H
H
H
CH3
H
H
H
H
H
H
H
H
H
-56.24
-57.77
-50.53
33.30
34.64
28.42
-4.82
-4.86
-4.59
28.48
29.78
23.83
-13.64
-14.52
-14.68
-14.12
-13.47
-12.01
43*
52*
CH3
a c.f. Eq.’s (1) and (2). Chrysin is highlighted in bold italics and the eight ligands selected for synthesis
b
∆퐺
highlighted with an asterisk (*). Compounds were numbered by rank of predicted
.
푏푖푛푑
2.2 Synthesis
In general, compounds 1, 9, 10, 11, 20, 43 and 52 were synthesised in four steps (Scheme 1)
from commercially available starting materials. The synthetic process started with a union
between 2’-hydroxy-4’,6’-dimethoxyacetophenone (A) and the benzoyl chloride of choice
identified from the CombiGlide results (B), to yield ester (C) in good yields. A base-
promoted Baker-Venkataraman rearrangement followed in which the aroyl group migrated
from oxygen onto carbon giving rise to diketone (D). An acid catalysed cyclization reaction
yielded the core flavonoid structure (E), which following demethylation, gave the desired
products (F). For the synthesis of compound 2 (R=2'-OH), the same sequence of steps were
repeated (Scheme 1), but instead of using 2’-hydroxy-4’,6’-dimethoxyacetophenone (A), 2'-
hydroxyacetophenone was used instead for a reaction with 2,4,6-trimethoxybenzoyl chloride.
In this latter case, compound G was isolated as the product of step 3, which after treatment
with hydroiodic acid/PhOH under reflux then resulted in de-methylation followed by a ring-
opening/ring-closing isomerisation to provide compound 2 (R=2'-OH).
12

Scheme 1. Synthesis of the eight substituted 5,7-dihydroxyflavones analogues tested.
2.3 Kinetics
The inhibitory potency of 1, 2, 9, 10, 11, 20, 43 and 52 was assessed against human liver GPa
(hlGPa; the pharmaceutical target), and rmGPa and rmGPb for comparison reasons to the
structural studies (performed with rmGPb). Since the Ki of quercetin has not been reported
and the Ki of chrysin was reported for the glycogen breakdown [16], we determined the
inhibition constants for these compounds against all three enzymes for comparison. All
results are presented in Table 2. Quercetin displays a preference for rmGPa over rmGPb, as
suggested previously by the IC50 values [13], but it is not as significant as implied (2-fold vs
5-fold). Nevertheless, quercetin is almost equipotent for rmGPa and hlGPa. Compounds 1, 2,
and 10 displayed only 14-27% inhibition at 100 μM for rmGPb and thus, we did not perform
any further kinetic studies due to the observed weak inhibition.
To establish the binding site of 9, 11, 20, 43 and 52, the inhibition behaviour of the five
compounds was investigated with respect to caffeine (inhibitor site) and AMP (allosteric
site). The kinetic data were analysed by Dixon plots from which the interaction constant (α)
between multiple inhibitors can be estimated [39]. Thus, with varying concentrations of
caffeine and 9, 11, 20, 43 or 52, and constant concentrations of AMP (1 mM) and substrate
(Glc-1-P; 20 mM) the Dixon plots, yielded a pattern of parallel lines (interaction constant
approaches infinity), indicating that the binding of 9, 11, 20, 43 or 52 and caffeine is mutually
exclusive (i.e. each one competes with caffeine for the same binding site) [39]. Similar
experiments with AMP and constant concentrations of Glc-1-P (2.5 mM) yielded intersected
lines in the Dixon plots, signifying that 9, 11, 20, 43 or 52 and AMP are not mutually
exclusive (each of the inhibitors and AMP can simultaneously bind at the enzyme) [39].
These experiments indicated that all of the compounds (9, 11, 20, 43 and 52) bind to the
inhibitor binding site. Compounds 11, 20 and 43 are significantly more potent inhibitors than
quercetin and chrysin, with Ki values in the range of lower μM (< 4 µM for all isoforms),
while 9 is equipotent to chrysin (Table 2). 52 is more potent than quercetin but less potent
than chrysin. 43 is the most potent of the designed compounds and the second most potent
inhibitor reported for the inhibitor site of GP, while 11 and 20 are almost equipotent and also
among the most potent compounds discovered to date.
13

We then investigated whether the binding of the most potent inhibitor 43 to GPb interacts
with that of glucose, like caffeine [40] which displays an interaction constant α=0.2. Initial
velocity studies were performed by varying 43 and glucose concentrations at fixed
concentrations of the substrates Glc-1-P (2.5 mM) and glycogen (0.2%). Plotting 1/v vs
glucose concentration for various 43 concentrations yields a set of lines intersecting at the left
of the y axis (Fig. S4). The binding of 43 interacts with glucose in a synergistic manner (α =
0.65). Thus, 43 and glucose are able to bind to the enzyme at the same time [39], while each
compound promotes the binding of the other, like caffeine and glucose [40] or ellagic acid
and glucose [41].
Comparing the phenyl B ring substitutions (Table 2), it seems that a halogen atom at position
3' confers to inhibition with chlorine being the best, followed closely by fluorine, but a larger
bromine (10) diminishes inhibitory potency. Position 4' seems to have a significant impact,
and a fluorine atom at this position increases potency (11 and 20), while a methyl group (52)
has a somewhat negative effect, but again a bromine diminishes potency (1). Finally, a single
methyl group at the 3' (or 5') position for 43 leads to the most potent inhibitor of all
compounds studied here. The inhibitory potency of the compounds seems to follow the same
pattern for the three enzymes (i.e. Ki values for hlGPa > rmGPa > rmGPb).
Table 2. The inhibitory potency the compounds studied here for glycogen phosphorylases. The
crystallographic atom numbering is shown in the inhibitors.
Ki [µM]
Compound
rmGPb
rmGPa
hlGPa
O H
O H
HO
O
O
Quercetin
69.12 ± 4.71
32.93 ± 2.49 43.52 ± 1.65
O H
O H
3'
2'
4'
5'
B
HO
O
Chrysin
6'
7.56 ± 0.81
5.14 ± 0.06
7.28 ± 0.09
C
A
O H
O
Br
HO
O
14 % inhibition at
1
-
-
100 μM
OH
O
14

OH
HO
O
O
16 % inhibition at
100 µM
2
9
-
-
OH
F
HO
O
7.26 ± 0.60
8.55 ± 0.57
7.39 ± 0.09
OH
O
Br
F
27 % inhibition at
100 µM
O
HO
10
11
-
-
O
OH
Cl
F
O
HO
1.89 ± 0.05
2.31 ± 0.19
2.27 ± 0.08
3.39 ± 0.22
O
OH
C14
C11
F1 F
C13
C15
O2
O1
C1
C10
C9
HO
O
C6
C12
C2
20
3.46 ± 0.13
1.75 ± 0.17
3.89 ± 0.08
2.23 ± 0.08
C8
C3
C5
C7
C4
OH
O
O7
O4
C14
C11
C15
C13
C12
O2
O1
C1
C10
C9
HO
O
C6
C2
C16
43
52
1.00 ± 0.05
C8
C3
C5
C7
C4
OH
O
O7
O4
C16
C14
C11
C15
C13
C12
O2
O1
C1
C10
C9
HO
O
C6
C2
11.40 ± 0.41
17.79 ± 2.37 21.36 ± 0.98
C8
C3
C5
C7
C4
OH
O
O7
O4
2.4 X-ray Crystallography
The crystal structures of rmGPb in complex with 20, 43 and 52 were determined to elucidate
the structural basis for their strong inhibition potency. Rabbit muscle and human liver GP
share 97% sequence homology and both enzymes are fully conserved in terms of sequence
and structure at the inhibitor binding site, thus any structural analysis on rmGPb can also
apply for hlGPa. All three compounds were found bound, as expected, at the inhibitor site of
the enzyme with all their atoms well defined within the electron density maps (Fig. 4). All
three inhibitors bind very similar. Their binding modes are also similar to those of chrysin
[16] and flavopiridol [21]. The RMSD of 0.8 Å between the free rmGPb structure and those
15

of 20, 43 and 52 complexes with GPb, for all atoms of well-defined residues (18-249, 262-
312, 326-829), indicates that the binding of the inhibitors does not trigger any significant
conformational change.
Figure 4. The Phenix polder omit difference electron density maps [42] of the ligands at the
inhibitor site. Maps are contoured at 3.0 σ while the final models of the inhibitors are shown.
The A and C rings (Table 2) are packed between the aromatic side-chain phenyls of Phe285
and Tyr613 engaging in π-stacking (parallel) interactions. Para-fluorine atom of 20 is at
distances of 3.1 and 3.2 Å from Arg770 (NH1) and Glu382 (OE2), respectively, forming two
fluorine bonds [43]. In all of the three inhibitor complexes, the hydroxyl group (O4) of the
inhibitor molecule (Table 2), engages in two water mediated interactions with the main chain
nitrogen atoms of Pro611 and Gly612, while the carbonyl oxygen (O7) forms two water
mediated interactions with the main chain carbonyl oxygen and nitrogen of Asp283 (Fig. 5).
Overall, 20 is involved in 101 (58 nonpolar/nonpolar, 6 polar/polar, 37 nonpolar/polar) van
der Waals interactions mainly with Phe285 and Tyr613 (64 contacts), and with Asn282,
Glu382, His571, Ala610, Gly612, and Arg770. 43 forms 91 (58 nonpolar/nonpolar, 4
polar/polar, 29 nonpolar/polar) van der Waals interactions mainly with Phe285 and Tyr613
(64 contacts), and with Asn282, Asn284, Glu382, Glu572, Ala610, Gly612, and Arg770. 52
engages in 89 (55 nonpolar/nonpolar, 4 polar/polar, 30 nonpolar/polar) van der Waals
interactions mainly with Phe285 and Tyr613 (59 contacts), and with Asn282, Asn284,
Glu382, His571, Glu572, Ala610, Gly612, Arg770, and Phe771 (Fig. 5). The solvent
accessibilities of the free 20, 43 and 52 are 432, 450, and 448 Å2, respectively. When bound
these values become 83, 88, and 91 Å2 indicating that approximately 80% of their surface
area becomes inaccessible to the solvent. The greatest contribution comes favorably from the
non-polar rmGPb residues that contribute 77%, 81%, and 86% of the protein surface which
becomes inaccessible upon binding of 20, 43 and 52, respectively. The fluorine atom of 20 is
involved in 7 (2 polar/polar, 5 polar/non-polar) van der Waals interactions with Glu382,
Arg770, and Phe771. The methyl group of 43 and 52, is involved 4 (2 polar/polar, 2
16

polar/nonpolar) and 5 (3 polar/polar, 5 polar/nonpolar) van der Waals interactions,
respectively with Asn284, Leu380 and Glu382 (43) or Glu382, Arg770, and Phe771 (52).
Figure 5. Stereo diagrams of the interactions of the ligands 20, 43 and 52 at the inhibitor site.
The inhibitor molecules are shown as thick bonds, water molecules as cyan spheres and
hydrogen bonds as dashed lines and van der Waals interactions of the B ring as mesh cloud.
Comparative structural analysis of the three inhibitor complexes reveals that while 20 and 43
bind at the same location within the inhibitor binding site of rmGPb the location of 52 differs
by approximately 0.4 Å, and 52 is more exposed to the solvent (or less buried within the
protein structure). The significant difference between the inhibitory potencies of 20 and 52
(Table 1) might be attributed to the two fluorine bonds and the 12 additional van der Waals
17

interactions of 20 with respect to the interactions of 52. Although, 20 forms two fluorine
bonds and 10 van der Waals interactions in addition to those of 43, both inhibitors are almost
equipotent (Kis 2.3 and 1.0 μM, respectively). The presence of the fluorine in 20 triggers a
small but significant conformational change of Glu382 (RMSD between the two
conformations is 1.1 Å). The conformation adopted by the side chain of Glu382 in the
rmGPb-20 complex is very similar to the one that it adopts in the rmGPb-52 complex (RMSD
between the two conformations is 0.2 Å). Therefore, it seems that the energy cost associated
with the conformational change of Glu382 that accompanies the binding of 20, is
compensated by the additional interactions of 20, and hence 20 and 43 are almost equipotent.
Superposition of three inhibitor complexes onto the rmGPb – chrysin complex [16] (Fig. 6)
reveals that although chrysin binds very similarly to the new inhibitors, its position differs by
0.4, 0.6, and 0.8 Å from the binding locations of 20, 43, and 52, respectively. Upon binding
to rmGPb, chrysin forms 96 van der Waals interactions (54 nonpolar/nonpolar, 4 polar/polar,
and 38 polar/nonpolar) and two water mediated hydrogen bonds [16]. The difference in the
inhibitory potency between chrysin and 20 might be attributed to the fluorine bonds formed
by the latter compound. However, the significant difference between the inhibitory potencies
of both 43 and 52 with that of chrysin cannot be attributed to the small shift in their binding
location within the inhibitor site.
Figure 6. Stereo diagram of the superposition of the 20, 43, and 52 complexes onto the
rmGPb-chrysin complex [16].
2.5 Ex-vivo Cellular Studies
18

The efficacy of inhibitor 43 to inhibit glycogen phosphorylase was also evaluated in
hepatocarcinoma HepG2 cells. For that purpose, HepG2 cells were precultured in high
glucose medium (25 mM glucose), in the presence of 10 nM insulin, to activate glycogen
synthesis. Upon inhibitor 43 treatment, cells were cultured in low glucose medium (5 mM
glucose), in the presence of 10 nM DEX, conditions appropriate to induce glycogenolysis.
Inhibitor 43 was checked at a concentrations range of 55 μM to 330 μM. Assessment of the
enzymatic activity of GP in cellular extracts of HepG2 cells treated with the inhibitor 43
revealed high potency of inhibitor 43 to suppress GP activity, exhibiting IC50 of 70 µM (Fig.
7).
Figure 7. GP inhibition by inhibitor 43 in HepG2 cells. GP activity was assayed in lysates
after 3 h incubation of the cells with 54.5, 109, 218 and 327 μΜ of 43 as described in the
Experimental Section. The value of IC50 was calculated as 70 ± 2.5 μΜ using the program
GRAFIT and the “explicit weight” function to take into account all experimental error. GP
activity was measured at 4 different times and determined using the same method as for the in
vitro assays.
3. Conclusions
In silico screening for GP inhibitors acting at the inhibitor site has been performed using
chrysin as the lead compound[16] and post-docking QM/MM-PBSA binding free energy
calculations. By retaining the 5,7-dihydroxy-4H-chromen-4-one ligand core of chrysin and
screening for different B ring substitutions exploiting the ZINC database[26], eight
19

candidates from the in silico screening were selected for synthesis with the potential for more
favourable interactions at the inhibitor site hydrophobic pocket. The target compounds were
synthesized using a Baker-Venkataraman rearrangement-cyclization strategy, yielding the
compounds in modest yields over the four step procedure. Kinetics experiments against
rmGPb revealed compounds 11 (Ki = 1.9 µM), 20 (Ki = 2.3 µM) and 43 (Ki = 1.0 µM) among
the most potent inhibitor site compounds known to date and more potent than any previously
reported natural flavonoid inhibitor; multiple inhibition studies revealed binding exclusively
at the site, and this was confirmed by X-ray crystallographic studies. The crystal structure
GPb – inhibitor complexes revealed the interactions responsible for the observed potencies,
which included favourable halogen B ring interactions at positions 3' and 4', but most
significantly a nonpolar substituent (particularly methyl) at the meta position. Ex-vivo
experiments for the most potent inhibitor (43) were also encouraging and revealed the
compound to have strong inhibitory effects on endogenous GP activity in HepG2 cells in
culture (IC50 = 70 µM). In addition, optimization of the pharmacological potential of the
compounds can include modification of A and C ring properties and substituents. The
increased potency of flavopiridol compared to chrysin with A ring substitutions supports such
an approach [16, 22]. Additionally, while natural flavonoids have quite low oral
bioavailability, flavopiridol does not [44], and together with the promising cellular results
obtained here for 43, highlights the potential of synthetic flavonoid analogues in drug
discovery projects such as this.
4. Experimental Section
4.1. Computational Details
4.1.1. Protein Preparation
The GPb protein was prepared for calculations using the solved 1.9 Å resolution co-
crystallized complex with chrysin (PDB code 3EBO) and Schrödinger’s Protein Preparation
Wizard [45]. The waters within 5 Å of chrysin were retained, bond orders assigned and
hydrogens added, with protonation states for basic and acidic residues based on pKa’s at a pH
of 7 calculated using PROPKA[46]. Subsequent optimization of hydroxyl groups, histidine
protonation states and C/N atom flips, and side-chain O/N atom flips of Asn and Gln residues
was based on optimizing hydrogen bonding patterns. Finally, the system was softly
20

minimised using OPLS-AA (2005) forcefield with the RMSD of heavy atoms kept within 0.3
Å of the crystallographic positions.
4.1.2. Ligand Database Preparation
Using the ZINC docking database (http://zinc.docking.org/)[26], benzoyl chloride derivatives
were filtered so that the final set of screening compounds fulfilled certain criteria (MW,
HBAs, HBDs) among Lipinski’s ‘rule of five’ [47] for oral bioavailability [MW < 500 Da,
hydrogen bond acceptor (HBA) ≤ 10, hydrogen bond donor (HBD) ≤ 5 and log P < 5]. The
formal charge of the benzoyl chlorides was set to be 0 (more suitable for synthesis) and also a
limit set on the ligand flexibility for final compounds (number of rotatable bonds ≤ 5) to
limit entropy loss on binding; a larger number of rotatable bonds (≥ 10) has also been
associated with poor oral bioavailability [48]. Including these restrictions yielded 1169
benzoyl chloride derivatives. These were joined to the 5,7-dihydroxy-4H-chromen-4-one core
(Figure 3) using CombiGlide 3.9 with LigPrep [27]. This protocol yielded 1239 minimised
structures including potential tautomers/ionization states ready for virtual screening.
4.1.3. Docking
Docking of ligands was performed to the prepared protein conformation from its complex
with chrysin (c.f. 4.1.1.) (model 1) and also to an induced-fit docking (IFD) GPb model
(model 2) of its complex with 5,7-dihydroxy-2-(naphthalen-2-yl)-4H-chromen-4-one (B ring
as a 2-naphthyl). In Stage 1 of the IFD, the 5,7-dihydroxy-2-(naphthalen-2-yl)-4H-chromen-
4-one was docked to the inhibitor site of model 1 using Glide 6.9 with the SP scoring function
and up to 20 poses per ligand saved. The core of the ligand (A and C ring atoms) involved in
the π-stacking was restrained to its original position (PDB code: 3EBO) using core
constraints (tolerance 0.10 Å). The hydrophobic pocket Leu380 sidechain was temporarily
trimmed in this stage (mutated to Ala), as well as scaling of the van der Waals radii of non-
polar atoms by 0.5. Stage 2 was a Prime 4.2 refinement, with residues in the hydrophobic
pocket refined. For Stage III, up to 20 protein conformations within 30 kcal/mol of the
lowest-energy structure were used for Glide-SP re-docking. The top-ranked complex based
on IFDscore was selected as model 2 for subsequent docking.
In the docking calculations of screening compounds to GPb models 1 and 2 with Glide
6.9[49] [50], the shape and properties of the inhibitor binding site were mapped onto grids
with dimensions of 23.1 × 23.1 × 23.1 Å centred on the bound ligands. Standard parameters
21

were applied including van der Waals scaling for non-polar atoms (by 0.8) to include modest
‘induced-fit’ effects. Core constraints (RMSD of 0.10 Å) were applied on the A and C ring
atoms. Docking calculations were performed using both Glide-SP and -XP, both of which
included post-docking minimization with strain correction. All 1169 ligands were docked
with 1 pose per ligand saved for each docking calculation, and compounds ranked by
GlideScore. For each GP model (models 1 and 2), ligands were re-ranked by consensus
scoring using a Simple Sum Rank [28] of the Glide-SP and –XP ranks. RMSDs between
ligand atom positions for compounds that gave both SP and XP results were calculated to
determine if similar (consensus) poses for were obtained [30]. A RMSD cut-off < 1.0 Å was
applied for consensus pose agreement and retention of ligands. Additionally, ligands not
feasible with the Baker-Venkataraman rearrangement and/or with B ring substituents
considered too bulky were filtered. Based on this process, the top 200 ligands for each of the
two GPb conformation models 1 and 2 proceeded to the QM/MM-PBSA binding free energy
calculations.
4.1.4. QM/MM-PBSA Calculations
Glide-SP poses for the top 200 ligands from each GPb model for docking were used directly
as input for QM/MM-PBSA calculations. Binding free energies were calculated as the
difference in energies between the bound and the unbound states of the protein-ligand
complexes using Eq. (1).
Eq. (1)
푄푀/푀푀 and 푀푀 represent the total molecular mechanics energy (internal, electrostatic and
∆퐺푏푖푛푑 = ∆퐸푄푀/푀푀 + ∆퐺푠표푙푣 ― 푇∆푆푀푀
퐸
푆
van der Waals) and an entropy estimate, respectively; 푠표푙푣, the solvation free energy. DFT
퐺
with the M06-2X functional[51] and the LACVP*+ basis set [52, 53] was used for the QM
region (ligand, Phe 285 and Tyr 613 sidechains) and the rest of the protein was described
using MM with the OPLS-AA(2005) forcefield [54]. Effectively, the MM region polarizes
the QM region, with electrostatic effects between the MM point charges and the QM
wavefunction, and the van der Waals interactions between QM and MM atoms accounted
for.[54] No cut-off for non-bond interactions was employed. Corrections for basis set
superposition error (BSSE) were not included given that the key QM interaction region atoms
were the same for all ligands (rings A and C with Phe285 and Tyr613) and resulting
cancellation of errors [16]. Additionally, while the magnitude of BSSE is dependent on
intermolecular distance [55], the core constraints employed in the docking phase counteracted
22

this effect. Bulk solvation effects were included using Poisson-Boltzmann Surface Area
(PBSA) [56], with the default solute (internal) dielectric constant of 1.0 employed. The
푠표푙푣 term was calculated as two components:
∆퐺
푠표푙푣
∆퐺푠표푙푣 = ∆퐺푠푃표퐵푙푣 + ∆퐺
Eq. (2)
푆퐴
푠표푙푣
푠표푙푣
where
and
account for the polar and non-polar (proportional to changes in
∆퐺
∆퐺
푆퐴
푃퐵
solvent accessible surface areas) contributions, as described previously [34]. Meanwhile, an
estimate for the loss of ligand entropy on binding was calculated using MM with the
(∆푆푀푀
)
OPLS-AA(2005) forcefield [54] and the Rigid Rotor Harmonic Oscillator (RRHO)
approximation. Using this method, the change in vibrational, rotational and translational
entropy of the ligands on binding was considered. QM/MM-PBSA calculations were
performed using QSite 6.9 [49]; RRHO calculations using MacroModel 11.0 [49].
The
values of ligands from both protein models (model 1 and 2) were compared and
∆퐺푏푖푛푑
the best
푏푖푛푑 value taken, together with its receptor-ligand complex. Ligands were ranked
∆퐺
by
푏푖푛푑 values and those with a value < -11 kcal/mol considered further. A Prime 4.2 [49]
∆퐺
minimization was used to refine the corresponding protein-ligand complexes using default
settings that included the OPLS3 forcefield [57] and VSGB solvation model [58]. Sidechains
of residues surrounding the hydrophobic pocket (Asp283, Asn284, Glu382, Leu380, His571,
Glu572, Tyr573, Arg770 and Phe771) together with the ligands were refined during this
process, but flavonoid rings A and C constrained to maintain the orginal π-stacking distances
with Phe285 and Tyr613. Final QM/MM-PBSA binding free energies were determined using
the Prime output complexes as input, employing the same settings as before.
4.2 – Synthetic Methods
General procedure for the synthesis of 1, 9, 10, 11, 20, 43 and 52
2’-Hydroxy-4’,6’-dimethoxyacetophenone (1 eq.) was added to the required benzoyl chloride
(1.1 eq.) dissolved in dichloromethane (0.1M). 4-Dimethylaminopyridine (0.1 eq.) and
triethylamine (1.5 eq.) were added and the reaction was allowed to stir for 24 hours under a
nitrogen atmosphere. Upon completion (TLC – EtOAc, petroleum ether mixtures) the product
was acidified with HCl and the organic layer was washed with distilled water, brine and dried
(MgSO4). The solution was evaporated in vacuo leaving a crude solid. The crude solid was
dissolved in dry THF (0.1M) and NaH (60% in mineral oil, 1.0 eq.) was added and the
reaction was heated at reflux for 20 hours. Upon completion (TLC – EtOAc, petroleum ether
23

mixtures) the solution was washed with HCl and extracted with EtOAc. The organic layer
was washed with distilled water, brine and dried (MgSO4). The solution was evaporated in
vacuo leaving a crude solid. The crude solid was suspended in conc. H2SO4 and acetic acid
and this mixture was allowed to stir at reflux for 15 hours. After allowing to cool, the solution
was poured onto an ice/water slurry and the resulting crystalline solid was collected using
vacuum filtration. The crude solid was suspended in HBr (47-49%) and glacial acetic acid
and this mixture heated at reflux for 40 hours. After allowing to cool, the solution was poured
onto an ice/water slurry, and following isolation and purification by flash column
chromatography, yielded the title compound as a pure solid.
Procedure for the synthesis of 2
2,4,6-Trimethoxybenzoic acid (1.1 g, 5.2mmol) was dissolved in dichloromethane (65 mL),
the reaction flask was purged with nitrogen for 10 minutes and then oxalyl chloride (0.50 mL,
5.7 mmol) was added along with 2 drops of dimethylformamide. The mixture was allowed to
react and the progress was monitored by TLC (50% EtOAc in petroleum ether) over the
period of an hour until complete. The solvent was removed in vacuo to yield a viscous yellow
oil (1.4 g, 6 mmol). The crude product was dissolved in dichloromethane (50 mL) and 2’-
hydroxyacetophenone (0.66 mL, 5.5 mmol) was added along with triethylamine (1.1 mL) and
4-dimethylaminopyridine (70 mg) and the reaction was allowed to progress overnight at room
temperature. The organic layer was washed with NaOH (2 x 25 mL), water (2 x 25 mL) and
brine (30 mL), and the solvent was removed in vacuo to yield a viscous yellow oil (1.8 g, 5.4
mmol). The flask containing the crude material was purged with nitrogen for ten minutes
followed by the addition of anhydrous tetrahydrofuran (50 mL), followed by NaH (60% in
mineral oil, 0.22 g, 5.4 mmol) and the reaction heated at 66 °C for 20 hours. The solution was
acidified with HCl (250 mL) to yield a pale yellow crystalline precipitate by vacuum
filtration (1.12 g, 3.4 mmol). Flash column chromatography (SiO2; ethyl acetate:petroleum
ether) yielded the desired rearranged product (0.8 g, 2.5 mmol). H2SO4 (5 mL) was added
along with CH3COOH (60 mL) and this was heated to reflux for 6 hours until completion, as
determined by TLC. The solution was poured over an ice/water slurry to yield cream-like
crystals (0.2 g 0.64 mmol). Hydroiodic acid (5 mL) was added to these crystals along with
phenol (3.21 g) and this mixture was heated to 120 °C (bath temperature) for 2 hours. The
solution was then poured over an ice/water slurry (300 mL) to yield a pale-cream crystalline
solid (20 mg, 0.07 mmol).
24

4.3 – Kinetics Measurement
rmGPb was purified from rabbit skeletal muscles following the protocol developed by
Fischer and Krebs [59], while hlGPb was produced as described previously [60]. rmGPa and
hlGPa were prepared by phosphorylation of rmGPb and hlGPb, respectively, using a
truncated form of the γ (catalytic) subunit of rabbit skeletal muscle phosphorylase kinase
[61]. Kinetic studies were performed at 30 °C in the direction of glycogen synthesis using the
method by Saheki et al. [62] for the determination of the released orthophosphate anions. 3
µg/mL rmGPb, rmGPa, or 1 µg/mL hlGPa were assayed following an established protocol
described previously [63] and various concentrations of inhibitors. All inhibitors were treated
as allosteric inhibitors (do not obey Michaelis-Menten model) and hence Hill plots were
produced (Figures S1-S3). Inhibition constant (Ki) values were then calculated from the
intercept to horizontal axis of the plot of KM(app.) vs [inhibitor] using the non-linear regression
program GRAFIT [64].
4.4 – X-ray Crystallography
rmGPb crystals were grown by the batch method as described previously [63]. X-ray
crystallographic binding studies were performed by diffusion of 20 (10 mM; 38 h), 43 (10
mM; 24 h), and 52 (10 mM; 47 h) solution in the crystallization media supplemented with
25% (v/v) DMSO in preformed rmGPb crystals at room temperature prior to data collection.
X-ray diffraction data were collected using a Cu X-ray microfocus source (Oxford
Diffraction SuperNova) equipped with a 4-kappa goniometer and the ATLAS CCD (135 mm)
detector at room temperature. X-ray data processing was performed by the program
CrysalisPro (Agilent Technologies UK Ltd.) [65]. Scaling and merging of intensities were
performed by Aimless and the optimum resolution was selected based on the CC1/2 criterion
[66]. Crystallographic refinement of the complexes was performed by maximum-likelihood
methods using REFMAC [67]. The starting model employed for the refinement of the
complexes was the structure of T state rmGPb in complex with a C-β-D-glucopyranosyl
triazole inhibitor after the removal of the ligand [68]. Ligand molecule coordinates and
topologies were constructed using AceDRG [69] within Coot [70] and they were fitted to the
electron density maps after adjustment of their torsion angles. A summary of the data
processing and refinement statistics for the inhibitor complex structures is given in Table 3.
The validity of the refinement procedure was checked using the PDB_REDO server [71]. As
25

there were more than five reflections per atom available, both an isotropic and an anisotropic
B-factor model were considered, and the isotropic B-factor model was selected based on the
Hamilton R ratio test. A TLS model for grouped atom movement with one TLS group was
used. The stereochemistry of the protein residues was validated by MolProbity [72].
Accessibility surfaces were calculated by the NACCESS program [73]. Figures were
prepared with the CCP4 Molecular Graphics program [74]. The coordinates of the new
structures have been deposited with the RCSB Protein Data Bank [75] with codes presented
in Table 3.
Table 3. Summary of the diffraction data processing and refinement statistics for the rmGPb
complexes. Values in parentheses are for the outermost shell.
rmGPb complex
20
43
52
Data collection and processing statistics
Resolution (Å)
Outermost shell (Å)
Reflections measured
Unique reflections
Multiplicity
13.74 - 2.33
2.42 - 2.33
175524
13.67 - 2.38
2.47 - 2.38
200647
13.68 – 2.40
2.49 – 2.40
164680
38529 (3446)
4.6 (3.1)
34581 (3009)
5.8 (4.3)
34314 (3151)
4.8 (3.4)
Rmerge
0.085 (0.596)
92.6 (80.3)
10.7 (1.7)
0.095 (0.570)
90.1 (75.7)
11.3 (1.9)
0.101 (0.623)
91.2 (80.2)
10.7 (1.4)
Completeness (%)
<I/σI>
CC1/2
0.991 (0.768)
0.995 (0.840)
0.988 (0.754)
Refinement statistics
Rcryst
0.164 (0.256)
0.212 (0.294)
241
0.157 (0.238)
0.211 (0.244)
227
0.168 (0.264)
0.211 (0.301)
244
Rfree
No of solvent molecules
r.m.s. deviation from ideality in
bonds (Å)
0.005
1.3
0.004
1.2
0.003
0.9
angles (°)
Average B factor (Å2)
Protein atoms
28.8
29.1
31.5
29.2
27.1
24.2
Solvent molecules
Inhibitor atoms
PDB entry
50.7
49.7
41.4
6Y5C
6Y5O
6Y55
26

4.5 – Ex-vivo Experiments
Assessment of the efficacy of the potent GP inhibitors to suppress endogenous GP activity in
hepatocarcinoma HepG2 cells was performed as previously described [76]. Briefly, human
HepG2 hepatocarcinoma cells were maintained in Dulbecco’s modified Eagle medium
(DMEM), containing 25 mM glucose 10% FBS,
2
mM glutamine, and
penicillin/streptomycin, in a humidified atmosphere with 5% CO2, at 37 °C. For GP activity
assessment, cells were seeded in 60 mm culture dishes at a density of 1.5 X 106 cells per dish.
After cell attachment (16-18 h), the medium was replaced by serum-free medium containing
10 nM dexamethasone, 25 mM glucose, and 10 nΜ insulin, and cells were further incubated
for 16-18 h to replete glycogen stores. Cells were then incubated in the absence and presence
of 43 at a concentration range of 54.5 to 327 μΜ, in serum-free medium supplemented with 5
mM glucose and 10 nΜ dexamethasone (low-glucose growth medium) for 3 h. Subsequently,
cells were washed in phosphate buffer, harvested, centrifuged at 700 X g for 5 min. Cell
pellets were resuspended in 20 mM Tris pΗ 7.5, containing 0.05% Triton and supplemented
with 1 mM PMSF. Then, cell lysates were sonicated followed by centrifugation at 12,000 X g
at 40 C for 15 min. GP activity was assessed in cleared lysates as described above. Glycogen
(1% w/v) and 20mM Glc-1-P were added to the lysates for the assay while GP activity values
were normalized against total protein amount, for each condition. The IC50 value of inhibitor
43 was calculated by the % GP activity vs inhibitor concentration diagram using the
nonlinear regression program GRAFIT [64].
Acknowledgments
We would like to thank Dr S.E. Zographos and Ms P.G.V. Liggri for their help during X-ray
data collection. We acknowledge support of this work by the project “The National Research
Infrastructures on integrated biology, drug screening efforts and drug target functional
27

characterization – INSPIRED-Thessaly” (MIS 5002550) which is implemented under the
Action “Reinforcement of the Research and Innovation Infrastructure”, funded by the
Operational Programme "Competitiveness, Entrepreneurship and Innovation" (NSRF 2014-
2020) and co-financed by Greece and the European Union (European Regional Development
Fund). E.K. would like to acknowledge financial support by Greece and the European Union
(European Social Fund- ESF) through the Operational Programme «Human Resources
Development, Education and Lifelong Learning» in the context of the project “Strengthening
Human Resources Research Potential via Doctorate Research” (MIS-5000432), implemented
by the State Scholarships Foundation (ΙΚΥ). The work was also supported in part by the
General Secretariat for Research and Technology and the Hellenic Foundation for Research
and Innovation (HFRI) under the HFRI PhD fellowship grant to A.G.K. (code: 2246).
Supporting Information
Supplementary data related to this article can be found at XXXX and includes Figures from
1
13
multiple inhibition studies (Dixon plots); melting point, H-NMR and C-NMR, IR and MS
characterisation data (purities > 95%) for identified potent inhibitors (9, 11, 20, 43 and 52);
copies of 1H-NMR spectra and HPLC chromatograms.
References
[1] World Health Organisation. www.who.int (15/05/20).
[2] R.A. DeFronzo, R.C. Bonadonna, E. Ferrannini, Pathogenesis of NIDDM. A balanced
overview, Diabetes Care, 15 (1992) 318-368.
[3] M. Roden, E. Bernroider, Hepatic glucose metabolism in humans - its role in health and
disease, Best Practice & Research Clinical Endocrinology & Metabolism, 17 (2003) 365-383.
[4] N.G. Oikonomakos, Glycogen phosphorylase as a molecular target for type 2 diabetes
therapy, Curr. Protein. Pept. Sci., 3 (2002) 561-586.
[5] C.E. Zois, A.L. Harris, Glycogen metabolism has a key role in the cancer
microenvironment and provides new targets for cancer therapy, J. Molec. Med.-JMM, 94
(2016) 137-154.
[6] L. Xu, H. Sun, Pharmacological manipulation of brain glycogenolysis as a therapeutic
approach to cerebral ischemia, Mini Rev. Med. Chem., 10 (2010) 1188-1193.
28

[7] T. Guan, Y.S. Qian, X.Z. Tang, M.H. Huang, L.F. Huang, Y.M. Li, H.B. Sun, Maslinic
Acid, a Natural Inhibitor of Glycogen Phosphorylase, Reduces Cerebral Ischemic Injury in
Hyperglycemic Rats by GLT-1 Up-Regulation, J. Neurosci. Res., 89 (2011) 1829-1839.
[8] L. Somsák, K. Czifrák, M. Tóth, É. Bokor, E.D. Chrysina, K.M. Alexacou, J.M. Hayes,
C. Tiraidis, E. Lazoura, D.D. Leonidas, S.E. Zographos, N.G. Oikonomakos, New inhibitors
of glycogen phosphorylase as potential antidiabetic agents, Curr. Med. Chem., 15 (2008)
2933-2983.
[9] A.L. Kantsadi, A. Apostolou, S. Theofanous, G.A. Stravodimos, E. Kyriakis, V.A.
Gorgogietas, D.S. Chatzileontiadou, K. Pegiou, V.T. Skamnaki, D. Stagos, D. Kouretas,
A.M. Psarra, S.A. Haroutounian, D.D. Leonidas, Biochemical and biological assessment of
the inhibitory potency of extracts from vinification byproducts of Vitis vinifera extracts
against glycogen phosphorylase, Food Chem. Toxicol., 67 (2014) 35-43.
[10] E.D. Chrysina, M.N. Kosmopoulou, C. Tiraidis, R. Kardakaris, N. Bischler, D.D.
Leonidas, Z. Hadady, L. Somsak, T. Docsa, P. Gergely, N.G. Oikonomakos, Kinetic and
crystallographic studies on 2-(beta-D-glucopyranosyl)-5-methyl-1, 3, 4-oxadiazole, -
benzothiazole, and -benzimidazole, inhibitors of muscle glycogen phosphorylase b. Evidence
for a new binding site, Protein Sci., 14 (2005) 873-888.
[11] G.A. Stravodimos, B.A. Chetter, E. Kyriakis, A.L. Kantsadi, D.S. Chatzileontiadou,
V.T. Skamnaki, A. Kato, J.M. Hayes, D.D. Leonidas, Phytogenic Polyphenols as Glycogen
Phosphorylase Inhibitors: The Potential of Triterpenes and Flavonoids for Glycaemic Control
in Type 2 Diabetes, Curr. Med. Chem., 24 (2017) 384-403.
[12] J.M. Hayes, A.L. Kantsadi, D.D. Leonidas, Natural products and their derivatives as
inhibitors of glycogen phosphorylase: potential treatment for type 2 diabetes, Phytochem.
Rev., 13 (2014) 471-498.
[13] S. Jakobs, D. Fridrich, S. Hofem, G. Pahlke, G. Eisenbrand, Natural flavonoids are
potent inhibitors of glycogen phosphorylase, Mol. Nutr. Food Res., 50 (2006) 52-57.
[14] A. Kato, N. Nasu, K. Takebayashi, I. Adachi, Y. Minami, F. Sanae, N. Asano, A.A.
Watson, R.J. Nash, Structure-activity relationships of flavonoids as potential inhibitors of
glycogen phosphorylase, J. Agric. Food. Chem., 56 (2008) 4469-4473.
[15] O. Kamiyama, F. Sanae, K. Ikeda, Y. Higashi, Y. Minami, N. Asano, I. Adachi, A. Kato,
In vitro inhibition of alpha-glucosidases and glycogen phosphorylase by catechin gallates in
green tea, Food Chem., 122 (2010) 1061-1066.
[16] K.E. Tsitsanou, J.M. Hayes, M. Keramioti, M. Mamais, N.G. Oikonomakos, A. Kato,
D.D. Leonidas, S.E. Zographos, Sourcing the affinity of flavonoids for the glycogen
phosphorylase inhibitor site via crystallography, kinetics and QM/MM-PBSA binding
studies: Comparison of chrysin and flavopiridol, Food Chem. Toxicol., 61 (2013) 14-27.
[17] L. Anila, N.R. Vijayalakshmi, Beneficial effects of flavonoids from Sesamum indicum,
Emblica officinalis and Momordica charantia, Phytother. Res., 14 (2000) 592-595.
[18] M. Ahmad, M.S. Akhtar, T. Malik, A.H. Gilani, Hypoglycaemic action of the flavonoid
fraction of Cuminum nigrum seeds, Phytother. Res., 14 (2000) 103-106.
[19] K.C. Ong, H.E. Khoo, Effects of myricetin on glycemia and glycogen metabolism in
diabetic rats, Life Sci., 67 (2000) 1695-1705.
[20] O. Kamiyama, F. Sanae, K. Ikeda, Y. Higashi, Y. Minami, N. Asano, I. Adachi, A. Kato,
In vitro inhibition of α-glucosidases and glycogen phosphorylase by catechin gallates in
green tea, Food Chem., 122 (2010) 1061-1066.
[21] N.G. Oikonomakos, J.B. Schnier, S.E. Zographos, V.T. Skamnaki, K.E. Tsitsanou, L.N.
Johnson, Flavopiridol inhibits glycogen phosphorylase by binding at the inhibitor site, J.
Biol. Chem., 275 (2000) 34566-34573.
[22] L.J. Hampson, C. Arden, L. Agius, M. Ganotidis, M.N. Kosmopoulou, C. Tiraidis, Y.
Elemes, C. Sakarellos, D.D. Leonidas, N.G. Oikonomakos, Bioactivity of glycogen
29