Research Article
For reprint orders, please contact: reprints@future-science.com
Computational drug repositioning for
ischemic stroke: neuroprotective drug
discovery
Yunong Li‡ ,1 , Jingbo Yang‡ ,1, Yan Zhang1, Qingkang Meng1 , Andreas Bender**,2
& Xiujie Chen*,1
1Harbin Medical University, Harbin, China
2Department of Chemistry, University of Cambridge, Cambridge, UK
*Author for correspondence: Tel.: +86 045 186 352 422; chenxiujie@ems.hrbmu.edu.cn
**Author for correspondence: Tel.: +44 790 615 8533; ab454@cam.ac.uk
‡Authors contributed equally
Background: A comprehensive approach to drug repositioning will be required to overcome translational
hurdles and identify more neuroprotective drugs. Results & methods: Gene Set Enrichment Analysis was
applied to identify related pathways and enriched genes. Candidate genes were optimized using Topp-
Gene, ToppGenet and pBRIT. From the perspective of the local structures, gene–domain–substructure–
drug relationships were constructed. Using the MCODE algorithm and K-means clustering, 31 functional
subnetworks were obtained, and 252 drugs with proposed neuroprotective function were identified. Us-
ing computational analysis, 72 substructures with different scores were found to correspond to neuropro-
tective functions. The protective effects of benidipine and barnidipine were confirmed in vitro. Conclu-
sion: The authors’ research has great potential to discover more neuroprotective drugs and obtain more
information regarding mechanisms of action and functional substructures.
First draft submitted: 22 January 2021; Accepted for publication: 18 May 2021; Published online:
17 June 2021
Ischemic stroke (IS) is a leading cause of neurologic disability and mortality worldwide [1]. The disease itself has
a significant economic burden on society that will increase with the aging population. For decades, the main
approaches to therapy for IS have focused on reperfusion and neuroprotection [2]. In the acute stage of IS,
recombinant tissue plasminogen activator remains the only US FDA-approved drug that can accelerate reperfusion
via thrombolysis, but its use is unfortunately limited by a narrow time window, low recanalization rate and increased
risk of hemorrhagic transformation [3]. Apart from reperfusion, neuroprotective therapy has been proposed since
the 1980s, with the main aim of intervening in the events of the ischemic cascade, blocking the pathological process
and avoiding the death of nerve cells [4]. Several key factors in ischemic cell death within the penumbra have been
identified, including oxidative stress, excitotoxicity and inflammation. Targeting these mechanisms shows beneficial
neuroprotective effect [5]. However, translating these mechanistic theories into clinically meaningful drugs for stroke
is very challenging, and continuous efforts have been directed toward identifying new drugs [6].
Over the past few decades, de novo drug discovery has become increasingly expensive and time-consuming, and
the number of novel compounds transferred to therapeutic drugs has stagnated [7]. To overcome this burden, drug
repositioning (DR), which identifies new indications for existing drugs, has emerged as a significant strategy in
new drug discovery [8]. However, because of the large number of diseases and known drugs, it is still very costly to
completely screen new uses of known drugs via experiments. With the accumulation of omics and pharmaceutical
informatics data, DR has entered the stage of combining rational design and experimental screening. Computational
DR analysis strategies have become an important research direction in computational biology and systems biology.
In addition, accurately identifying the interactions between drugs and targets is a key stage in accelerating drug
development [9]. If we can accurately identify a significant correlation between the candidate drug studied and certain
target proteins, we can avoid aimlessly screening candidate targets from the massive protein data and accelerate the
entire drug development process [10]. At present, research on the prediction of drug–target interactions is mainly
FutureMed. Chem. (Epub ahead of print) ISSN 1756-891910.4155/fmc-2021-0022 C© 2021 Newlands Press
Research Article Li, Yang, Zhang, Meng, Bender & Chen
based on the overall structure of the drug. However, the binding of drug and target protein is the interaction of
the local structure of the drug (drug substructure) and the active pocket of the protein (protein domain). From
the perspective of local structure, the relationship between structure and function can be more comprehensively
expressed.
For these reasons, the authors’ study adopted a computational DR strategy based on drug–target interaction and
identified drugs with strong neuroprotective effects. In addition, the correlations between substructures and drug
functions were further explored. In contrast to previous studies, the strategy used in the authors’ study shows great
promise in identifying more neuroprotective drugs, which can, to an extent, make up for the lack of treatment
strategies.
Methods
Preparing candidate gene set
The authors searched the Gene Expression Omnibus database for gene expression profiling studies associated with
IS. Genome-wide RNA sequencing data (GSE58294) with normalized values were downloaded from the Gene
Expression Omnibus database [11]. In this dataset, blood samples were collected from 69 patients within 24 h of
stroke onset and 23 controls with no history of symptomatic vascular disease. Microarray expression profile was
determined by array using the [HG-U133 Plus 2] Affymetrix Human Genome U133 Plus 2.0 Array.
To assess gene expression signatures and activation status of pathways, Gene Set Enrichment Analysis (GSEA)
was performed using Java GSEA implementation with 186 Kyoto Encyclopedia of Genes and Genomes gene sets of
canonical pathways, a C2 subcollection in MSigDB [12]. Using a permutation test with 1000 repetitions, the cutoff
for p-value significance level for the significant pathways related to IS was determined to be 0.05. Accordingly,
GSEA gave the authors a subset of genes that better contributed to score enrichment and significant pathways
by comparing the samples with IS and the controls. The subset genes were used as the candidate gene set in the
subsequent process.
Process of gene prioritization
To further obtain the genes closely related to IS (or neuroprotection), three methods of gene prioritization –
ToppGene, which prioritizes or ranks candidate genes based on functional similarity to the training gene list;
ToppGenet, which identifies and prioritizes the neighboring genes of the seeds in the protein–protein interaction
network based on functional similarity to the ‘seed’ list (ToppGene); and pBRIT, which prioritizes genes based on
phenotypic concordance to the training genes – were used to select the most related genes among the gene candidate
set. The three gene prioritization algorithms were used to measure the similarity of genes and to select the most
related genes based on different complementary characteristics. The genes that had the greatest similarity with the
genes the authors collected as the training set and had been documented to be associated with neuroprotection
were selected [13]. Prioritized genes were screened (p < 0.05) and sorted using the three algorithms. The authors
denoted score1(Gj) as the prioritization score of the j-th gene by integrating the scores of the j-th gene in the three
tools using the formula
Score1(G j ) =
TG j + TNj + P B j
3T
where T represents the total number of genes, j∈[1,T], and TGj, TNj and PBj are the score rankings of the j-th genes
in ToppGene, ToppGenet and pBRIT, respectively. According to score1(Gj), the most promising neuroprotection-
related genes with score1(Gj) greater than the mean were identified.
Identification & quantification of gene–drug relationships
To find drugs targeting these neuroprotection-related genes, a gene (protein)–domain–substructure–drug rela-
tionship network was established based on domain-substructure associations identified in the authors’ previous
study [14]. Detailed information regarding the data used to establish the relationship network is shown in the
identification and quantification of gene–domain–substructure–drug relationships in the supplementary material.
To evaluate the intensity of the interaction in each drug–gene pair, score2(Zpq), which was determined by the
10.4155/fmc-2021-0022 FutureMed. Chem. (Epub ahead of print) future science group
Computational drug repositioning for ischemic stroke: neuroprotective drug discovery Research Article
intensity score of the interaction between the p-th drug and the q-th gene, was defined as
s core2(Zpq) =
Npq
Tp
× Npq
Gq
where Npq represents the number of domain–substructure relationships shared between the p-th drug and the q-th
gene, p∈[1,2436] and q∈[1,25]; Tp represents the number of all domain–substructure relationships corresponding
to the p-th drug; and Gq represents the number of all domain–substructure relationships corresponding to the
q-th gene. The higher the score2(Zpq), the stronger the association between a drug and a gene. Finally, drug–gene
relationships with a score2 above the average were obtained and used to optimize the gene–domain–substructure–
drug network.
Obtaining drugs with strong similarity
To evaluate the similarity of drugs, the authors used a clustering method based on the assumption that drugs in the
same class tend to be similar in function. Therefore, the MCODE algorithm [15] and K-means clustering [16] were
used to cluster drugs based on the corresponding substructure–domain relationships in the authors’ neuroprotective
network. The authors determined that the drugs that were clustered into one class by the two clustering algorithms
simultaneously had strong similarity at the substructure level. A detailed description of this can be found in the
supplementary material.
Identification & quantification of the correlations between substructures & functions
The authors next identified drug substructures related to neuroprotective effect and quantified the correlations
between substructures and neuroprotective effect. To this end, the authors attempted to find specified substructures
of drugs in each cluster that were considered to have similar mechanisms related to neuroprotection. In this method,
the 881-bit PubChem fingerprints were used as substructures to analyze the mechanism of drug neuroprotective
action from the perspective of the local structure of the drug. The authors used the following formula to identify
drug substructures that significantly correlated with neuroprotective function and elucidate the mechanism of drug
action from the perspective of the local structure
score3(Si ) =
D
Fi
score4(Si ) =
Mi
P
score5(Si ) = score3(Si ) × score4(Si )
where score3(Si) is defined as the specificity score of the i-th substructure to highlight the rare substructures
distributed in approved drugs; score4(Si) is defined as the commonality score of the i-th substructure in a specific
drug group to highlight the common substructures of the drug functional mechanism in a cluster; score5(Si) is
defined as the correlation intensity score of the i-th substructure and drug function, i∈[1,881]; the number of all
approved small-molecule drugs in the DrugBank database is represented by D; the number of drugs containing the
i-th substructure in all approved small-molecule drugs is represented by Fi; the number of drugs in one cluster is
represented by P; and the number of drugs containing the i-th substructure in the drugs of the cluster is represented
by Mi.
In vitro experiments
Cell culture, reagents & antibodies
Mouse brain microvascular endothelial cells (bEnd.3) were obtained from the Bena Culture Collection (Beijing,
China). Cells were cultured in DMEM supplemented with 10% fetal bovine serum, 100 IU/ml penicillin and
100 μg/ml streptomycin in a humidified incubator with 5%CO2 at 37◦C. Benidipine was purchased from Aladdin
future science group 10.4155/fmc-2021-0022
Research Article Li, Yang, Zhang, Meng, Bender & Chen
Table 1. Antiapoptotic, anti-inflammatory and antioxidative genes related to ischemic stroke.
Antiapoptotic gene Antiapoptopic gene (cont.) Anti-inflammatory gene Antioxidative gene
ACOX1 IDH1 ACSL4 GSTM3
ACSL1 ILK CD36 GSTO2
ACSL3 MGST1 CPT1A GSTT2
ACSL4 NR1H3 DBI GSTZ1
CD36 OLR1 FABP4 IDH1
CPT1A RRM2 ILK LAP3
CPT2 RXRB NR1H3 PGD
DBI SORBS1 RXRB RRM2
FABP4 SORBS1 RRM2B
GSTM3 TXNDC12
(Shanghai, China). Barnidipine was purchased fromMolbase (Shanghai, China). Antibodies used in this study were
anti-GPX-1, anti-SOD-1, anti-Bcl-2, anti-Bax and anti-cleaved caspase-3 (Wanleibio Co. Ltd., Shenyang, China).
Oxygen−glucose deprivation & reoxygenation
The bEnd.3 cells were subjected to ischemia-like injury via oxygen−glucose deprivation (OGD) for 8 h by replacing
the culture medium with serum- and glucose-free DMEM (Gibco, Shanghai, China) and placing cultures in a
hypoxic incubator containing 5% CO2 and 95%N2, as described previously [17]. After 8 h of OGD, reoxygenation
was induced by placing all cultures in normoxic conditions for a further 24 h and replacing the OGD medium
with the normal medium supplemented with 10% fetal bovine serum. Control cultures (no injury) were incubated
with DMEM containing glucose in a normoxic incubator.
Western blot analysis
Western blot analysis was used to determine levels of GPX-1, SOD-1, Bcl-2, Bax and cleaved caspase-3 in the
bEnd.3 cells given that they are associated with apoptosis and oxidation. Cell lysates were obtained from the bEnd.3
cells at reoxygenation 24 h following OGD. Protein samples (30 μg) were resolved on polyacrylamide sodium
gels and electrophoretically transferred to polyvinylidene difluoride membranes. The primary antibodies against
GPX-1, SOD-1, Bcl-2, Bax and cleaved caspase-3 were used, with GAPDH as an internal control. The western
blot bands were quantified using Odyssey (LI-COR Biosciences, NE, USA).
Statistical analysis
The in vitro experiment results were analyzed using Prism 7.0 (GraphPad Software, CA, USA). Statistical com-
parisons between experimental groups were assessed with one-way analysis of variance, and the level of statistical
significance was set at p < 0.05.
Results
Candidate gene selection & validation
Based on GSEA of the dataset, which was achieved by comparing IS and control samples, the authors identified
45 significantly upregulated and enriched gene sets (pathways) (p < 0.05) associated with IS. These included
the antiapoptosis/antioxidative-related (glutathione metabolism) and antiapoptosis/inflammation (PPAR signal-
ing) [18] pathways as well as the proinflammatory-related (cytosolic DNA-sensing) and apoptosis-related (apoptosis)
pathways (Figure 1). The authors obtained the gene sets that better contributed to score enrichment in the glu-
tathione metabolism and PPAR signaling pathways (Figure 2).
To select antiapoptotic, anti-inflammatory and antioxidative genes, the authors used core-enriched genes from
GSEA as the test set and genes that have known antiapoptotic, antioxidative and anti-inflammatory effects as the
training set. Prioritized genes were identified with three algorithms (ToppGene, ToppGenet and pBRIT) (p< 0.05)
and sorted by score1. The authors identified 18 antiapoptotic, nine anti-inflammatory and ten antioxidative genes
that might serve as the targets of therapy and a total of 25 neuroprotective genes (Table 1).
The neuroprotective functions of 25 genes were confirmed by literature search. Searching in the PubMed database
using ‘neuroprotective,’ ‘antiapoptotic,’ ‘anti-inflammatory,’ ‘antioxidative’ and ‘gene symbol’ as search terms for
10.4155/fmc-2021-0022 FutureMed. Chem. (Epub ahead of print) future science group
Computational drug repositioning for ischemic stroke: neuroprotective drug discovery Research Article
K
E
G
G
_S
ha
re
_i
nt
er
ac
tio
ns
_i
n_
ve
si
cu
la
r_
tr
an
sp
or
t_
si
gn
al
K
E
G
G
_C
om
pl
em
en
t_
an
d_
co
ag
ul
at
io
n_
ca
sc
ad
es
_s
ig
na
l
K
E
G
G
_G
lu
ta
th
io
ne
_m
et
ab
ol
is
m
_s
ig
na
l
K
E
G
G
_D
ru
g_
m
et
ab
ol
is
m
_o
th
er
_e
nz
ym
es
_s
ig
na
l
K
E
G
G
_R
eg
ul
at
io
n_
of
_a
ut
op
ha
gy
_s
ig
na
l
K
E
G
G
_U
bi
qu
iti
n_
m
ed
ia
te
d_
pr
ot
eo
ly
si
s_
si
gn
al
K
E
G
G
_P
er
ox
is
om
e_
si
gn
al
K
E
G
G
_L
ys
in
e_
de
gr
ad
at
io
n_
si
gn
al
K
E
G
G
_A
di
po
cy
to
ki
ne
_s
ig
na
lin
g_
pa
th
w
ay
_s
ig
na
l
K
E
G
G
_F
at
ty
_a
ci
d_
m
et
ab
ol
is
m
_s
ig
na
l
K
E
G
G
_P
P
A
R
_s
ig
na
lin
g_
pa
th
w
ay
_s
ig
na
l
K
E
G
G
_G
ly
co
sa
m
in
og
ly
ca
n_
de
gr
ad
at
io
n_
si
gn
al
K
E
G
G
_L
ys
os
om
e_
si
gn
al
K
E
G
G
_P
en
to
se
_p
ho
sp
ha
te
_p
at
hw
ay
_s
ig
na
l
K
E
G
G
_G
al
ac
to
se
_m
et
ab
ol
is
m
_s
ig
na
l
K
E
G
G
_S
ta
rc
h_
an
d_
su
cr
os
e_
m
et
ab
ol
is
m
_s
ig
na
l
K
E
G
G
_A
lz
he
im
er
s_
di
se
as
e_
si
gn
al
K
E
G
G
_P
ro
te
in
_e
xp
or
t_
si
gn
al
K
E
G
G
_P
53
_s
ig
na
lin
g_
pa
th
w
ay
_s
ig
na
l
K
E
G
G
_E
pi
th
el
ia
l_
ce
ll_
si
gn
al
in
g_
in
_h
el
ic
ob
ac
te
r_
py
lo
ri_
in
fe
ct
io
n_
si
gn
al
K
E
G
G
_V
ib
rio
_c
ho
le
ra
e_
in
fe
ct
io
n_
si
gn
al
K
E
G
G
_L
ei
sh
m
an
ia
_l
at
er
al
_s
ol
er
os
is
)a
ls
_s
ig
na
l
K
E
G
G
_A
m
yo
tr
op
hi
c_
la
te
ra
l_
sc
le
ro
si
s_
A
LS
_s
ig
na
l
K
E
G
G
_N
O
D
_l
ik
e_
re
ce
pt
or
_s
ig
na
lin
g_
pa
th
w
ay
_s
ig
na
l
K
E
G
G
_T
ol
l_
lik
e_
re
ce
pt
or
_s
ig
na
lin
g_
pa
th
w
ay
_s
ig
na
l
K
E
G
G
_R
IG
_I
_l
ik
e_
re
ce
pt
or
_s
ig
na
lin
g_
pa
th
w
ay
_s
ig
na
l
K
E
G
G
_C
yt
os
ol
ic
_D
N
A
_s
en
si
ng
_p
at
hw
ay
_s
ig
na
l
K
E
G
G
_A
po
pt
os
is
_s
ig
na
l
K
E
G
G
_M
A
P
K
_s
ig
na
lin
g_
pa
th
w
ay
_s
ig
na
l
K
E
G
G
_P
ro
ge
st
er
on
e_
m
ed
ia
te
d_
oo
cy
te
_m
at
ur
at
io
n_
si
gn
al
K
E
G
G
_O
oc
yt
e_
m
ei
os
is
_s
ig
na
l
K
E
G
G
_T
yp
e_
II_
di
ab
et
es
_m
el
lit
us
_s
ig
na
l
K
E
G
G
_I
ns
ul
in
_s
ig
na
lin
g_
pa
th
w
ay
_s
ig
na
l
K
E
G
G
_S
m
al
l_
ce
ll_
lu
ng
_c
an
ce
r_
si
gn
al
K
E
G
G
_P
at
hw
ay
s_
in
_c
an
ce
r_
si
gn
al
K
E
G
G
_V
E
G
F
_s
ig
na
lin
g_
pa
th
w
ay
_s
ig
na
l
K
E
G
G
_F
C
_e
ps
ilo
n_
R
I_
si
gn
al
in
g_
pa
th
w
ay
_s
ig
na
l
K
E
G
G
_G
N
R
H
_s
ig
na
lin
g_
pa
th
w
ay
_s
ig
na
l
K
E
G
G
_E
R
B
B
_s
ig
na
lin
g_
pa
th
w
ay
_s
ig
na
l
K
E
G
G
_P
an
cr
ea
tic
_c
an
ce
r_
si
gn
al
K
E
G
G
_C
hr
on
ic
_m
ye
lo
id
_l
eu
ke
m
ia
_s
ig
na
l
K
E
G
G
_A
cu
te
_m
ye
lo
id
_l
eu
ke
m
ia
_s
ig
na
l
K
E
G
G
_P
ro
st
at
e_
ca
nc
er
_s
ig
na
l
K
E
G
G
_E
nd
om
et
ria
l_
ca
nc
er
_s
ig
na
l
K
E
G
G
_N
on
_s
m
al
l_
ce
ll_
lu
ng
_c
an
ce
r_
si
gn
al
KEGG_Share_interactions_in_vesicular_transport_signal
KEGG_Complement_and_coagulation_cascades_signal
KEGG_Glutathione_metabolism_signal
KEGG_Drug_metabolism_other_enzymes_signal
KEGG_Regulation_of_autophagy_signal
KEGG_Ubiquitin_mediated_proteolysis_signal
KEGG_Peroxisome_signal
KEGG_Lysine_degradation_signal
KEGG_Adipocytokine_signaling_pathway_signal
KEGG_Fatty_acid_metabolism_signal
KEGG_PPAR_signaling_pathway_signal
KEGG_Glycosaminoglycan_degradation_signal
KEGG_Lysosome_signal
KEGG_Pentose_phosphate_pathway_signal
KEGG_Galactose_metabolism_signal
KEGG_Starch_and_sucrose_metabolism_signal
KEGG_Alzheimers_disease_signal
KEGG_Protein_export_signal
KEGG_P53_signaling_pathway_signal
KEGG_Epithelial_cell_signaling_in_helicobacter_pylori_infection_signal
KEGG_Vibrio_cholerae_infection_signal
KEGG_Leishmania_lateral_solerosis)als_signal
KEGG_Amyotrophic_lateral_sclerosis_ALS_signal
KEGG_NOD_like_receptor_signaling_pathway_signal
KEGG_Toll_like_receptor_signaling_pathway_signal
KEGG_RIG_I_like_receptor_signaling_pathway_signal
KEGG_Cytosolic_DNA_sensing_pathway_signal
KEGG_Apoptosis_signal
KEGG_MAPK_signaling_pathway_signal
KEGG_Progesterone_mediated_oocyte_maturation_signal
KEGG_Oocyte_meiosis_signal
KEGG_Type_II_diabetes_mellitus_signal
KEGG_Insulin_signaling_pathway_signal
KEGG_Small_cell_lung_cancer_signal
KEGG_Pathways_in_cancer_signal
KEGG_VEGF_signaling_pathway_signal
KEGG_FC_epsilon_RI_signaling_pathway_signal
KEGG_GNRH_signaling_pathway_signal
KEGG_ERBB_signaling_pathway_signal
KEGG_Pancreatic_cancer_signal
KEGG_Chronic_myeloid_leukemia_signal
KEGG_Acute_myeloid_leukemia_signal
KEGG_Prostate_cancer_signal
KEGG_Endometrial_cancer_signal
KEGG_Non_small_cell_lung_cancer_signal
Figure 1. Forty-five significantly upregulated and enriched gene sets associated with ischemic stroke. This is a set to
set graph, showing the overlap between gene sets, using green color to indicate the number of genes shared
between sets. The deeper the color, the more the number of shared genes.
the 25 genes, it was confirmed that ten genes were related to neuroprotective function and 13 genes were related to
protective function in other tissues. The overall verification rate was 92%. The PubMed identifiers of the related
literature can be found in Supplementary Table 1.
Network construction, similarity analysis & verification of drug function
The authors established the neuroprotective gene–domain–substructure–drug network (Supplementary Figure 1),
which contained 22 genes, 24 domains, 480 substructures and 2297 drugs with score2(Zpq) greater than the
mean (Supplementary Table 2). Using MCODE, the authors identified 37 modules and obtained scores for each
module. The authors then conducted further analysis of ten modules whose scores were greater than the average
score of 4.6 (Supplementary Table 3). Using the elbow method, the authors determined that the number of clusters
in K-means clustering was 8 (Supplementary Figure 2). Finally, the authors screened out the drugs classified into the
same class by MCODE and K-means clustering and formed 31 functional subnetworks, with a total of 252 drugs
future science group 10.4155/fmc-2021-0022
Research Article Li, Yang, Zhang, Meng, Bender & Chen
G
S
M
14
06
05
6
G
S
M
14
06
05
7
G
S
M
14
06
05
8
G
S
M
14
06
05
9
G
S
M
14
06
06
0
G
S
M
14
06
06
1
G
S
M
14
06
06
2
G
S
M
14
06
06
3
G
S
M
14
06
06
4
G
S
M
14
06
06
5
G
S
M
14
06
06
6
G
S
M
14
06
06
7
G
S
M
14
06
06
8
G
S
M
14
06
06
9
G
S
M
14
06
07
0
G
S
M
14
06
07
1
G
S
M
14
06
07
2
G
S
M
14
06
07
3
G
S
M
14
06
07
4
G
S
M
14
06
07
5
G
S
M
14
06
07
6
G
S
M
14
06
07
7
G
S
M
14
06
07
8
G
S
M
14
06
07
9
G
S
M
14
06
08
0
G
S
M
14
06
08
1
G
S
M
14
06
08
2
G
S
M
14
06
08
3
G
S
M
14
06
08
4
G
S
M
14
06
08
5
G
S
M
14
06
08
6
G
S
M
14
06
08
7
G
S
M
14
06
08
8
G
S
M
14
06
08
9
G
S
M
14
06
09
0
G
S
M
14
06
09
1
G
S
M
14
06
09
2
G
S
M
14
06
09
3
G
S
M
14
06
09
4
G
S
M
14
06
09
5
G
S
M
14
06
09
6
G
S
M
14
06
09
7
G
S
M
14
06
09
8
G
S
M
14
06
09
9
G
S
M
14
06
10
0
G
S
M
14
06
10
1
G
S
M
14
06
10
2
G
S
M
14
06
10
3
G
S
M
14
06
10
4
G
S
M
14
06
10
5
G
S
M
14
06
10
6
G
S
M
14
06
10
7
G
S
M
14
06
10
8
G
S
M
14
06
10
9
G
S
M
14
06
11
0
G
S
M
14
06
11
1
G
S
M
14
06
11
2
G
S
M
14
06
11
3
G
S
M
14
06
11
4
G
S
M
14
06
11
5
G
S
M
14
06
11
6
G
S
M
14
06
11
7
G
S
M
14
06
11
8
G
S
M
14
06
11
9
G
S
M
14
06
12
0
G
S
M
14
06
12
1
G
S
M
14
06
12
2
G
S
M
14
06
12
3
G
S
M
14
06
12
4
G
S
M
14
06
03
3
G
S
M
14
06
03
4
G
S
M
14
06
03
5
G
S
M
14
06
03
6
G
S
M
14
06
03
7
G
S
M
14
06
03
8
G
S
M
14
06
03
9
G
S
M
14
06
04
0
G
S
M
14
06
04
1
G
S
M
14
06
04
2
G
S
M
14
06
04
3
G
S
M
14
06
04
4
G
S
M
14
06
04
5
G
S
M
14
06
04
6
G
S
M
14
06
04
7
G
S
M
14
06
04
8
G
S
M
14
06
04
9
G
S
M
14
06
05
0
G
S
M
14
06
05
1
G
S
M
14
06
05
2
G
S
M
14
06
05
3
G
S
M
14
06
05
4
G
S
M
14
06
05
5
S
am
pl
e 
na
m
e
G
S
M
14
06
05
6
G
S
M
14
06
05
7
G
S
M
14
06
05
8
G
S
M
14
06
05
9
G
S
M
14
06
06
0
G
S
M
14
06
06
1
G
S
M
14
06
06
2
G
S
M
14
06
06
3
G
S
M
14
06
06
4
G
S
M
14
06
06
5
G
S
M
14
06
06
6
G
S
M
14
06
06
7
G
S
M
14
06
06
8
G
S
M
14
06
06
9
G
S
M
14
06
07
0
G
S
M
14
06
07
1
G
S
M
14
06
07
2
G
S
M
14
06
07
3
G
S
M
14
06
07
4
G
S
M
14
06
07
5
G
S
M
14
06
07
6
G
S
M
14
06
07
7
G
S
M
14
06
07
8
G
S
M
14
06
07
9
G
S
M
14
06
08
0
G
S
M
14
06
08
1
G
S
M
14
06
08
2
G
S
M
14
06
08
3
G
S
M
14
06
08
4
G
S
M
14
06
08
5
G
S
M
14
06
08
6
G
S
M
14
06
08
7
G
S
M
14
06
08
8
G
S
M
14
06
08
9
G
S
M
14
06
09
0
G
S
M
14
06
09
1
G
S
M
14
06
09
2
G
S
M
14
06
09
3
G
S
M
14
06
09
4
G
S
M
14
06
09
5
G
S
M
14
06
09
6
G
S
M
14
06
09
7
G
S
M
14
06
09
8
G
S
M
14
06
09
9
G
S
M
14
06
10
0
G
S
M
14
06
10
1
G
S
M
14
06
10
2
G
S
M
14
06
10
3
G
S
M
14
06
10
4
G
S
M
14
06
10
5
G
S
M
14
06
10
6
G
S
M
14
06
10
7
G
S
M
14
06
10
8
G
S
M
14
06
10
9
G
S
M
14
06
11
0
G
S
M
14
06
11
1
G
S
M
14
06
11
2
G
S
M
14
06
11
3
G
S
M
14
06
11
4
G
S
M
14
06
11
5
G
S
M
14
06
11
6
G
S
M
14
06
11
7
G
S
M
14
06
11
8
G
S
M
14
06
11
9
G
S
M
14
06
12
0
G
S
M
14
06
12
1
G
S
M
14
06
12
2
G
S
M
14
06
12
3
G
S
M
14
06
12
4
G
S
M
14
06
03
3
G
S
M
14
06
03
4
G
S
M
14
06
03
5
G
S
M
14
06
03
6
G
S
M
14
06
03
7
G
S
M
14
06
03
8
G
S
M
14
06
03
9
G
S
M
14
06
04
0
G
S
M
14
06
04
1
G
S
M
14
06
04
2
G
S
M
14
06
04
3
G
S
M
14
06
04
4
G
S
M
14
06
04
5
G
S
M
14
06
04
6
G
S
M
14
06
04
7
G
S
M
14
06
04
8
G
S
M
14
06
04
9
G
S
M
14
06
05
0
G
S
M
14
06
05
1
G
S
M
14
06
05
2
G
S
M
14
06
05
3
G
S
M
14
06
05
4
G
S
M
14
06
05
5
S
am
pl
e 
na
m
e
Gene symbol
ACSL3
GK
ACOX2
CPT2
ACSL4
CD36
RXBB
ACOX1
ACSL1
PPARG
CPT1A
TLK
DBT
ME1
ACOX3
CYP27A1
NR1H3
OLR1
SORRS1
CPT1C
FABP4
MGST2
MGST1
RRM2B
TDH1
GGT1
OPLAH
GSS
MGST3
RRM2
GCLM
GSTZ1
GCLC
ANPEP
TXNDC12
SMS
PGD
GSTO2
GSTM3
GSTT2
LAP3
Gene symbol
-2
Rank in ordered dataset
Enrichment plot: KEGG_glutathione_metabolism
R
an
ke
d
 li
st
 m
et
ri
c 
(S
ig
n
al
2N
o
is
e)
0
-1
0
1
2
0.0
E
n
ri
ch
m
en
t 
sc
o
re
 (
E
S
)
0.1
0.2
0.4
0.3
0.5
2500 5000 7500 10,000 12,500 15,000 17,500 20,000
‘IS’ (positively correlated)
‘Control’ (negatively correlated)
Zero cross at 12,950
-2
Rank in ordered dataset
Enrichment plot: KEGG_PPAR signaling_pathway
R
an
ke
d
 li
st
 m
et
ri
c 
(S
ig
n
al
2N
o
is
e)
0
-1
0
1
2
0.00
E
n
ri
ch
m
en
t 
sc
o
re
 (
E
S
)
0.10
0.05
0.15
0.20
0.35
0.25
0.30
0.40
2500 5000 7500 10,000 12,500 15,000 17,500 20,000
‘IS’ (positively correlated)
‘Control’ (negatively correlated)
Zero cross at 12,950
Enrichment profile Hits Ranking metric scoresEnrichment profile Hits Ranking metric scores
Figure 2. Gene sets that more contributes to the score enrichment from KEGG Glutathione metabolism pathway
and KEGG PPAR signaling pathway. (A) GSEA enrichment plot of KEGG glutathione metabolism pathway genes. (B)
The heat map of core enrichment genes in KEGG glutathione metabolism pathway. (C) GSEA enrichment plot of
KEGG PPAR signaling pathway genes. (D) The heat map of core enrichment genes in KEGG PPAR signaling pathway.
(Supplementary Table 4). The drugs in each network were considered to have strong neuroprotective function and
functional similarities between drugs.
To confirm the authors’ prediction, literature verification of 252 drugs was conducted. Searching in the PubMed
database using ‘neuroprotective,’ ‘antioxidative,’ ‘anti-inflammatory,’ ‘antiapoptotic’ and ‘drug name’ as search
terms, it was confirmed that 85 drugs were related to neuroprotective function. In each subnetwork, there was
at least one drug that was documented to be neuroprotective. Because of space limitations, the authors selected
compounds from only two subnetworks for detailed analysis.
The 18 drugs in subnetwork 1 were shared bymodule 1 (MCODE) and cluster 3 (K-means clustering) (Figure 3).
Using the method detailed in the current study, the authors predicted that these drugs had neuroprotective effects
after cerebral ischemia and confirmed the brain protection effects of some drugs via literature search.
10.4155/fmc-2021-0022 FutureMed. Chem. (Epub ahead of print) future science group
Computational drug repositioning for ischemic stroke: neuroprotective drug discovery Research Article
Digoxin Ouabain
SORBS1
RRM2 RRM2B
RXRB NR1H3
Ginseng
ILK
Figure 3. Sub-network 1. The red node represents the gene, the blue node represents the drug, and the green node
represents the domain-substructure. (A) The local network of digoxin and ouabain. (B) The local network of
steviolbioside, ivermectin and ginseng.
Digoxin is a drug used to control ventricular rate in atrial fibrillation and to aid in the management of congestive
heart failure with atrial fibrillation. Peng et al. showed that preconditioning treatment using digoxin contributed to
improved functional recovery and exerted a prominent neuroprotective effect, including reduction of apoptosis and
promotion of cell proliferation [19]. Ouabain is used to treat congestive heart failure and to aid in the treatment of
chronic atrial fibrillation. At present, studies have confirmed that ouabain has anti-inflammatory and antiapoptotic
effects and can protect human renal cells [20]. However, there is no literature confirming its protective effect on
the brain after ischemia. In the authors’ network, both drugs acted on the same target: RRM2, RRM2B, NR1H3,
RXRB and SORBS1 (Figure 3A). In the initial gene prioritization, the authors identified the protective effects of
these target genes. The authors also confirmed via the literature that these genes had antiapoptotic, antioxidative
or anti-inflammatory effects [21]. In addition, ouabain has a large number of substructures that are identical to
those found in digoxin. Among the 111 substructures of ouabain, 110 are shared with digoxin. Since the biological
functions of a drug depend on its structure, the authors believe that ouabain also has neuroprotective effects.
Furthermore, results of the MCODE algorithm and K-means clustering suggested that ouabain and digoxin play a
role in brain protection through a similar mechanism of action. A study by Winnicka et al. indicated that 30 nM of
digoxin and ouabain stimulated an antiapoptotic effect via promotion of the level of phosphorylated extracellular
signal-regulated kinases [22].
Ginseng is promoted as an adaptogen, a position that is, to a certain extent, supported with reference to its
anticarcinogenic and antioxidant properties. A study conducted by Liu et al. suggested that apoptosis induced by
cerebral ischemia/reperfusion was attenuated by ginseng via downregulation of the levels of cleaved caspase-3 and
future science group 10.4155/fmc-2021-0022
Research Article Li, Yang, Zhang, Meng, Bender & Chen
caspase-9 [23]. Sood et al. showed that ginseng ameliorated middle cerebral artery occlusion-induced oxidative stress,
apoptosis, mitochondrial dysfunction and cognitive impairment [24]. Steviolbioside, ivermectin and ginseng act on
the same protective targets – RRM2, RRM2B, NR1H3 and RXRB – and the substructures are similar (Figure 3B).
Therefore, the authors concluded that steviolbioside and ivermectin have neuroprotective effects, and ivermectin
has been shown to have antioxidant and anti-inflammatory effects. By the same token, the rest of the drugs in
subnetwork 1 all have neuroprotective effects.
The seven drugs in subnetwork 21 are shared by module 8 (MCODE) and cluster 2 (K-means clustering), and
six of these drugs are dihydropyridines, which are widely used in the treatment of hypertension and cerebrovascular
disease (Figure 4A). Azelnidipine demonstrates a neuroprotective effect in the ischemic brain [25]. Lercanidipine
possesses anti-inflammatory, antioxidant and antiapoptotic properties, which are cardinal mechanisms involved
in acute IS [26]. Nicardipine has anti-neuroinflammatory effects on microglial cells and exerts a protective effect
against rotenone-induced apoptosis [27]. The combination of manidipine and idebenone significantly ameliorates
neurological deficits and histological changes in the rat brain following stroke [28]. In addition, benidipine has
been shown to prevent inflammatory changes and oxidative stress in a rat model of myocardial infarction [29].
Barnidipine reduces the plasma levels of inflammatory and oxidative biomarkers in hypertensive rats [30]. Although
the targets and substructures of these drugs are similar, the neuroprotective effects of benidipine and barnidipine
against cerebral ischemia have not been confirmed. Therefore, the authors’ study verified the neuroprotective effects
of these two drugs, which are widely used in clinical practice in vitro.
The protein expression of GPX-1, SOD-1, Bax, Bcl-2 and cleaved caspase-3 was determined using western blot
assay. The authors’ results showed that OGD and reoxygenation (OGD/R)-induced apoptosis and oxidative stress
were inhibited by benidipine and barnidipine. As shown in Figures 4B & 5, the level of protein expression of Bax
and cleaved caspase-3 was upregulated by OGD/R compared with the control group. Benidipine and barnidipine
suppressed the expression of Bax and cleaved caspase-3 compared with the OGD/R group. The level of Bcl-2,
SOD-1 and GPX-1 was downregulated by OGD/R compared with the control group. Benidipine and barnidipine
promoted an increase in the level of these proteins compared with the OGD/R group.
Correlation analysis between substructure & drug function
With regard to the 252 drugs with strong neuroprotective function, the authors further explored the relationships
between drug substructures and drug functions to elucidate the structural basis of the neuroprotective effects
of these drugs. Based on score5(Zpq), the authors obtained correlation scores between drug substructures and
neuroprotective function, with a total of 72 substructures (Supplementary Table 5). The higher the score, the
stronger the correlation intensity. Since the drugs in each subnetwork had a common brain protection effect and
a similar mechanism of action, the authors concluded that these shared and specific substructures were likely to
be an important structural basis for the neuroprotective role played by these drugs. For example, in subnetwork
21 (Figure 4A), the six dihydropyridine drugs all had highly correlated substructures: SUB402(N(∼O)(∼O)) and
SUB456(N(-O)(=O)). These substructures may allow the six dihydropyridines to exert brain protection effects
through the interaction of similar mechanisms with the domains (PF00268, PF00104 and PF00105).
Discussion
The basic aim of neuroprotective therapies for IS is to interfere with the events of the ischemic cascade by focusing
on one or more of the mechanisms of damage, including excitotoxicity, oxidative stress and inflammation, blocking
the pathological processes and preventing the death of vulnerable nerve cells in the ischemic penumbra [4]. However,
clinically effective neuroprotectants have thus far remained elusive. Therefore, the authors’ study aimed to identify
new treatment targets and look for neuroprotective drugs through a screening strategy that targeted multiple
mechanisms of ischemic injury.
Since the interaction of drug and protein is essentially the interaction of the drug substructures and protein
domains, the relationship between structure and function can be better reflected from the local perspective. For
this reason, the authors used substructure–domain relationships as the basis for predicting drug–target protein
interaction relationships. The strategy used in the authors’ study can identify not only drugs with similar overall
structures and targets (e.g., nicardipine, benidipine, barnidipine and lercanidipine) but also drugs with different
overall structures but similar targets (e.g., clenbuterol, entacapone, verapamil and oxymetazoline).
The occurrence and development of complex diseases rarely stem from mutations of a single gene, but rather
involve comprehensive changes in the intracellular molecular network. From the perspective of a complex network,
10.4155/fmc-2021-0022 FutureMed. Chem. (Epub ahead of print) future science group
Computational drug repositioning for ischemic stroke: neuroprotective drug discovery Research Article
PF00268_SUB456
RRM2B RRM2
RXRB
NR1H3
Nicardipine Azelnidipine
Bcl-2
Bax
Cleaved caspase-3
SOD-1
GPX-1
GAPDH
OGD/R
Benidipine (µM)
Benidipine (µM)
OGD/R
0.0
0.5
1.0
1.5
2.0
B
cl
-2
/G
A
P
D
H
n
o
rm
al
iz
ed
 t
o
 c
o
n
tr
o
l 2.5
B
ax
/G
A
P
D
H
n
o
rm
al
iz
ed
 t
o
 c
o
n
tr
o
l
C
le
av
ed
 c
as
p
as
e-
3/
G
A
P
D
H
n
o
rm
al
iz
ed
 t
o
 c
o
n
tr
o
l
-
--
+ + + +
0.1 1 10
Benidipine (µM)
OGD/R
Benidipine (µM)
OGD/R
Benidipine (µM)
OGD/R
Benidipine (µM)
OGD/R
-
--
+ + + +
0.1 1 10
BcI-2
## **
**
**
-
--
+ + + +
0.1 1 10
-
--
+ + + +
0.1 1 10
-
--
+ + + +
0.1 1 10
-
--
+ + + +
0.1 1 10
0
2
4
6
8
Bax
##
**
**
**
0
2
3
1
Cleaved caspase-3
##
**
0.0
0.5
1.5
2.0
S
O
D
-1
/G
A
P
D
H
n
o
rm
al
iz
ed
 t
o
 c
o
n
tr
o
l 2.5
1.0
SOD-1
##
**
**
**
0.0
0.5
1.5
2.0
G
P
X
-1
/G
A
P
D
H
n
o
rm
al
iz
ed
 t
o
 c
o
n
tr
o
l 2.5
1.0
GPX-1
##
*
**
Figure 4. Sub-network 21, and benidipine inhibited OGD/R-induced apoptosis and oxidative stress in bEnd.3 cells.
(A) Sub-network 21. The red node represents the gene, the blue node represents the drug, and the green node
represents the domain-substructure. (B) The level of protein expression of Bax and Cleaved caspase-3 were
upregulated by OGD/R compared with the control group. Benidipine suppressed the expression of Bax and Cleaved
caspase-3 compared with the OGD/R group. The expression of Bcl-2, SOD-1 and GPX-1 were downregulated by
OGD/R compared with the control group. Benidipine promoted the increase in these proteins level compared with
the OGD/R group (n = 3).
*p < 0.05.
**p < 0.01 compared with the OGD/R group.
##p < 0.01 compared with the control group.
future science group 10.4155/fmc-2021-0022
Research Article Li, Yang, Zhang, Meng, Bender & Chen
Bax
Bcl-2
Cleaved caspase-3
GPX-1
SOD-1
GAPDH
Barnidipine (µM)
OGD/R
Barnidipine (µM)
OGD/R-
--
+ + + +
0.1 1 10
Barnidipine (µM)
OGD/R
Barnidipine (µM)
OGD/R -
--
+ + + +
0.1 1 10
Barnidipine (µM)
OGD/R -
--
+ + + +
0.1 1 10 Barnidipine (µM)
OGD/R
0.0
0.5
1.0
1.5
B
cl
-2
/G
A
P
D
H
n
o
rm
al
iz
ed
 t
o
 c
o
n
tr
o
l 2.0
Bcl-2
-
--
+ + + +
0.1 1 10
##
**
**
**
0
2
4
6
C
le
av
ed
 c
as
p
as
e-
3
/G
A
P
D
H
n
o
rm
al
iz
ed
 t
o
 c
o
n
tr
o
l
8
-
--
+ + + +
0.1 1 10
Cleaved caspase-3
##
**
**
0
1
2
3
B
ax
/G
A
P
D
H
n
o
rm
al
iz
ed
 t
o
 c
o
n
tr
o
l
4
Bax
##
**
**
0.0
0.5
1.0
S
O
D
-1
/G
A
P
D
H
n
o
rm
al
iz
ed
 t
o
 c
o
n
tr
o
l
1.5
SOD-1
##
* **
**
0.0
1.0
1.5
G
P
X
1
/G
A
P
D
H
n
o
rm
al
iz
ed
 t
o
 c
o
n
tr
o
l
2.5
2.0
0.5
-
--
+ + + +
0.1 1 10
GPX-1
##
**
**
**
Figure 5. Barnidipine inhibited OGD/R-induced apoptosis and oxidative stress in bEnd.3 cells. Barnidipine
suppressed the expression of Bax and Cleaved caspase-3 compared with the OGD/R group, and it promoted the
increase in these proteins level compared with the OGD/R group (n = 3).
*p < 0.05.
**p < 0.01 compared with the OGD/R group.
##p < 0.01 compared with the control group.
with the concept of a multilevel and multiangle interaction network (‘gene–domain–substructure–drug’), the
authors used the idea of network pharmacology to establish a large-scale association analysis of genes and drugs
and discover new combinations of drugs and targets. In addition, the authors’ study used two methods – MCODE
and K-means clustering – to identify drugs with the strongest neuroprotective effects according to the modularity
of the gene–drug network. Drugs clustered into the same subnetwork were determined to protect the ischemic
brain via the same mechanism. For example, in subnetwork 1, the authors predicted that ouabain and digoxin had
10.4155/fmc-2021-0022 FutureMed. Chem. (Epub ahead of print) future science group
Computational drug repositioning for ischemic stroke: neuroprotective drug discovery Research Article
similar targets and substructures and may exert neuroprotective effects through similar mechanisms of action. A
study by Winnicka et al. confirmed this inference [22]. Similarly, drugs in different modules were determined to
have different mechanisms of action. Therefore, researchers could try to combine the drugs in different modules
to exert the neuroprotective effects through a variety of mechanisms, a process that is expected to increase the
neuroprotective effects of drugs in future research.
It is well known that structure determines function, but previous research has mainly focused on the overall
structure of drugs. This study, for the first time, explored the mechanisms of action of drugs from the perspective
of the local structure and used computational analysis to quantify the correlations between drug functions and
substructures. Seventy-two substructures related to neuroprotective function were identified, and the score of each
substructure was obtained. The higher the score of the substructure, the stronger the correlation intensity between
the substructure and the neuroprotective effect of the drug. This theory has strong practical significance. First,
it allows us to better focus on the similarities of these specific substructures rather than the drugs themselves to
discover drugs that contain these substructures and have neuroprotective effects, further expanding the space for
drug discovery. Second, in future rational drug design, the specificity of drug substructures can be used to retain
the therapeutic effects, delete the substructures that produce side effects and optimize druggability.
However, the authors’ approach has several limitations with regard to predicting drugs with neuroprotective
effects in IS. The approach is limited by the number of protein domains and drug substructures and is not
applicable to predicting drugs without substructures or proteins without domains. These problems will be solved
with further developments in structural chemistry and pharmacology.
Conclusion
Our study proposed a strategy to discover more neuroprotective drug candidates and confirmed that benidipine
and barnidipine exerted neuroprotective effects in ischemic injury by inhibiting apoptosis and oxidative stress. In
addition, our study obtained more information regarding mechanisms of action and functional substructures.
Future perspective
Neuroprotection is important for reducing cerebral injury as much as possible in the context of stroke. It is very
urgent that we find more drugs that can be translated to clinical practice. Drug repositioning, a strategy for finding
new indications for old drugs, is an important strategy for drug research and development in both the present
and future because of its advantages of low risk of failure and short development time (e.g., the discovery of anti-
coronavirus drugs). Moreover, from the perspective of the local active structures of drugs and proteins, this study
demonstrates new indications for old drugs. This strategy not only can quantify the effects of drugs by measuring
the connection strength of the local structures between drugs and proteins but can also find more drugs candidates
with neuroprotective effects, which is consistent with the development trend in innovative drugs. It is also the trend
to extract structural features based on the relationships between local active structures and functions to combine
effective structures and generate new compounds. At the same time, based on the method used in this study,
a drug–substructure–domain–adverse drug reaction correlation network will be established in a future study to
discover new drug candidates with neuroprotective effects and less toxicity and improve the success rate of new drug
research and development. In addition, dihydropyridine calcium antagonists are widely used as antihypertensive
drugs in patients with IS. From the perspective of clinical treatment, the neuroprotective effect of these drugs is of
great significance and needs to be further confirmed in animal models and clinical trials.
Supplementary data
To view the supplementary data that accompany this paper, please visit the journal website at: www.futuremedicine.com/doi/sup
pl/10.2217/doi-fmc-2021-0022
Financial & competing interests disclosure
This work was supported by the National Natural Science Foundation of China (grant no. 61671191 and 61971166) and the China
Postdoctoral Science Foundation (grant no. 2019M661302). The authors have no other relevant affiliations or financial involvement
with any organization or entity with a financial interest or financial conflict with the subject matter or materials discussed in the
manuscript apart from those disclosed.
No writing assistance was utilized in the production of this manuscript.
future science group 10.4155/fmc-2021-0022
Research Article Li, Yang, Zhang, Meng, Bender & Chen
Summary points
• Using the bioinformatics tools Gene Set Enrichment Analysis, ToppGene, ToppGenet and pBRIT, the authors’ study
identified 25 neuroprotective target genes.
• The relationships between protein domains and drug substructures were predicted, and
gene–domain–substructure–drug relationships were constructed.
• Using computational analysis, the authors’ study quantified gene–drug relationships and constructed a
neuroprotective gene–domain–substructure–drug network.
• Using the MCODE algorithm and K-means clustering, the similarity of drug functions was analyzed, and
promising drugs were identified.
• The modularity of the network can better aggregate drugs with the same mechanism. Since the drugs and
corresponding target proteins are different between the modules, it can be inferred that the mechanisms of
action of the drugs are also different. In subsequent studies, the drugs between the various modules can be used
in combination to study their potential therapeutic effects.
• Using computational analysis, correlations between substructures and neuroprotective effect were identified and
quantified. Seventy-two substructures with different scores were identified as corresponding to neuroprotective
functions.
• In future rational drug design, the specificity of drug substructures can be used to retain the therapeutic effects,
delete the substructures that produce side effects and optimize druggability.
• In in vitro experiments, the authors’ study demonstrated that benidipine and barnidipine exerted
neuroprotective effects in ischemic injury by inhibiting apoptosis and oxidative stress.
References
Papers of special note have been highlighted as: • of interest; •• of considerable interest
1. Meschia JF, Cheryl B, Bernadette BA et al. Guidelines for the primary prevention of stroke: a statement for healthcare professionals from
the American Heart Association/American Stroke Association. Stroke 45(12), 3754–3832 (2014).
2. Yanishevsky SN, Tsygan NV, Golokhvastov SY et al. Modern strategies of protection of hypoxic-ischemic brain damage. Zh. Nevrol.
Psikhiatr. Im. S. S. Korsakova 117(12), 78 (2017).
3. Hawkins KE, Demars KM, Alexander JC et al. Targeting resolution of neuroinflammation after ischemic stroke with a lipoxin A4 analog:
protective mechanisms and long-term effects on neurological recovery. Brain Behav. 7(5), e00688 (2017).
4. O’Collins VE, MacLeod MR, Donnan GA, Horky LL, Van Der Worp BH, Howells DW. 1,026 experimental treatments in acute stroke.
Ann. Neurol. 59(3), 467–477 (2006).
5. Chamorro A´, Dirnagl U, Urra X, Planas AM. Neuroprotection in acute stroke: targeting excitotoxicity, oxidative and nitrosative stress,
and inflammation. Lancet Neurol. 15(8), 869–881 (2016).
•• Summarizes key factors in ischemic cell injury, including oxidative stress, excitotoxicity and inflammation.
6. Xing C, Arai K, Lo EH, Hommel M. Pathophysiologic cascades in ischemic stroke. Int. J. Stroke 7(5), 378–385 (2012).
7. Booth B, Zemmel R. Prospects for productivity. Nat. Rev. Drug Discov. 3(5), 451–456 (2004).
8. Shaughnessy AF. Old drugs, new tricks. BMJ 342(7793), 360–361 (2011).
9. Lu Y, Guo Y, Korhonen A. Link prediction in drug–target interactions network using similarity indices. BMC Bioinformatics 18(1), 39
(2017).
• Accurately identifies that the interaction between drugs and targets is a key stage in accelerating drug development.
10. Yuan Q, Gao J, Wu D, Zhang S, Hiroshi M, Zhu S. DrugE-Rank: improving drug–target interaction prediction of new candidate drugs
or targets by ensemble learning to rank. Bioinformatics 32(12), i18–i27 (2016).
11. Stamova B, Jickling GC, Ander BP et al. Gene expression in peripheral immune cells following cardioembolic stroke is sexually
dimorphic. PLoS One 9(7), e102550 (2014).
12. Keller A, Backes C, Lenhof H-P. Computation of significance scores of unweighted Gene Set Enrichment Analyses. BMC Bioinformatics
8(6), 8–290 (2007).
13. Kumar AA, Van Laer L, Alaerts M et al. pBRIT: gene prioritization by correlating functional and phenotypic annotations through
integrative data fusion. Bioinformatics 34(13), 2254–2262 (2018).
14. Li Y, Zhu H, Yang J et al. Discovering proangiogenic drugs in ischemic stroke based on the relationship between protein domain and
drug substructure. ACS Chem. Neurosci. 10(1), 507–517 (2019).
•• Establishes a gene (protein)–domain–substructure–drug relationship network based on the association of domain and
substructure.
10.4155/fmc-2021-0022 FutureMed. Chem. (Epub ahead of print) future science group
Computational drug repositioning for ischemic stroke: neuroprotective drug discovery Research Article
15. Bader GD, Hogue CW. An automated method for finding molecular complexes in large protein interaction networks. BMC
Bioinformatics 4(1), 4–2 (2003).
16. Kodinariya TM, Makwana PR. Review on determining number of cluster in K-means clustering. Int. J. Adv. Res. Comput. Sci. Manag.
Stud. 1(6), 90–95 (2013).
17. Yang J, Shi Q-D, Song T-B et al. Vasoactive intestinal peptide increases VEGF expression to promote proliferation of brain vascular
endothelial cells via the cAMP/PKA pathway after ischemic insult in vitro. Peptides 42, 105–111 (2013).
18. Song J, Park J, Oh Y, Lee JE. Glutathione suppresses cerebral infarct volume and cell death after ischemic injury: involvement of FOXO3
inactivation and Bcl2 expression. Oxid. Med. Cell. Longev. 2015(2015), 1–11 (2015).
• Confirms that the glutathione metabolism pathway is related to antiapoptotic and antioxidative functions.
19. Peng K, Tan D, He M et al. Studies on cerebral protection of digoxin against hypoxic–ischemic brain damage in neonatal rats.
Neuroreport 27(12), 906–915 (2016).
20. Amaral M, Girard M, A´lvarez R et al. Ouabain protects human renal cells against the cytotoxic effects of Shiga toxin type 2 and subtilase
cytotoxin. Toxins (Basel) 9(7), 226 (2017).
21. Becares N, Gage MC, Voisin M et al. Impaired LXRα phosphorylation attenuates progression of fatty liver disease. Cell Rep. 26(4),
984–995.e6 (2019).
22. Winnicka K, Bielawski K, Bielawska A, Miltyk W. Dual effects of ouabain, digoxin and proscillaridin A on the regulation of apoptosis in
human fibroblasts. Nat. Prod. Res. 24(3), 274–285 (2010).
23. Liu A, Zhu W, Sun L et al. Ginsenoside Rb1 administration attenuates focal cerebral ischemic reperfusion injury through inhibition of
HMGB1 and inflammation signals. Exp. Ther. Med. 16(4), 3020–3026 (2018).
24. Sood A, Mehrotra A, Dhawan DK, Sandhir R. Indian ginseng (Withania somnifera) supplementation ameliorates oxidative stress and
mitochondrial dysfunctions in experimental model of stroke. Metab. Brain Dis. 33(4), 1261–1274 (2018).
25. Lukic-Panin V, Kamiya T, Zhang H et al. Prevention of neuronal damage by calcium channel blockers with antioxidative effects after
transient focal ischemia in rats. Brain Res. 1176, 143–150 (2007).
26. Gupta S, Sharma U, Jagannathan NR, Gupta YK. Neuroprotective effect of lercanidipine in middle cerebral artery occlusion model of
stroke in rats. Exp. Neurol. 288, 25–37 (2017).
27. Huang BR, Chang PC, Wei-Lan Y et al. Anti-neuroinflammatory effects of the calcium channel blocker nicardipine on microglial cells:
implications for neuroprotection. PLoS One 9(3), e91167 (2014).
28. Nagisa Y, Mihara T et al. Beneficial effects of the combination of idebenone and manidipine 2HCl on neurological deficits and
histological changes following cerebrovascular lesions in stroke-prone spontaneously hypertensive rats. Nihon Yakurigaku Zasshi 106(5),
327–337 (1995).
29. Md Quamrul H, Md Sayeed A, Mohd A et al. Benidipine prevents oxidative stress, inflammatory changes and apoptosis related myofibril
damage in isoproterenol-induced myocardial infarction in rats. Toxicol. Mech. Methods 25(1), 26–33 (2015).
30. Alp Yildirim FI, Eker Kizilay D, Ergin B et al. Barnidipine ameliorates the vascular and renal injury in l-NAME-induced hypertensive
rats. Eur. J. Pharmacol. 764, 433–442 (2015).
future science group 10.4155/fmc-2021-0022