ArticlePDF Available

Computational drug repurposing against SARS-CoV-2 reveals plasma membrane cholesterol depletion as key factor of antiviral drug activity

Authors:

Abstract and Figures

Comparing SARS-CoV-2 infection-induced gene expression signatures to drug treatment-induced gene expression signatures is a promising bioinformatic tool to repurpose existing drugs against SARS-CoV-2. The general hypothesis of signature-based drug repurposing is that drugs with inverse similarity to a disease signature can reverse disease phenotype and thus be effective against it. However, in the case of viral infection diseases, like SARS-CoV-2, infected cells also activate adaptive, antiviral pathways, so that the relationship between effective drug and disease signature can be more ambiguous. To address this question, we analysed gene expression data from in vitro SARS-CoV-2 infected cell lines, and gene expression signatures of drugs showing anti-SARS-CoV-2 activity. Our extensive functional genomic analysis showed that both infection and treatment with in vitro effective drugs leads to activation of antiviral pathways like NFkB and JAK-STAT. Based on the similarity—and not inverse similarity—between drug and infection-induced gene expression signatures, we were able to predict the in vitro antiviral activity of drugs. We also identified SREBF1/2, key regulators of lipid metabolising enzymes, as the most activated transcription factors by several in vitro effective antiviral drugs. Using a fluorescently labeled cholesterol sensor, we showed that these drugs decrease the cholesterol levels of plasma-membrane. Supplementing drug-treated cells with cholesterol reversed the in vitro antiviral effect, suggesting the depleting plasma-membrane cholesterol plays a key role in virus inhibitory mechanism. Our results can help to more effectively repurpose approved drugs against SARS-CoV-2, and also highlights key mechanisms behind their antiviral effect.
Content may be subject to copyright.
RESEARCH ARTICLE
Computational drug repurposing against
SARS-CoV-2 reveals plasma membrane
cholesterol depletion as key factor of antiviral
drug activity
Szilvia BarsiID
1
, Henrietta PappID
2,3
, Alberto ValdeolivasID
4
, Da
´niel J. To
´thID
1
,
Anett Kuczmog
2,3
, Mo
´nika Madai
2,3
, La
´szlo
´Hunyady
1,5,6
, Pe
´ter Va
´rnaiID
1,5
, Julio Saez-
RodriguezID
4
, Ferenc Jakab
2,3
, Bence SzalaiID
1¤
*
1Semmelweis University, Faculty of Medicine, Department of Physiology, Budapest, Hungary, 2National
Laboratory of Virology, University of Pe
´cs, Pe
´cs, Hungary, 3Institute of Biology, Faculty of Sciences,
University of Pe
´cs, Pe
´cs, Hungary, 4Heidelberg University, Faculty of Medicine, and Heidelberg University
Hospital, Institute for Computational Biomedicine, Bioquant, Heidelberg, Germany, 5MTA-SE Laboratory of
Molecular Physiology, Budapest, Hungary, 6Institute of Enzymology, Research Centre for Natural Sciences,
Budapest, Hungary
¤Current address: Turbine Simulated Cell Technologies Ltd., Budapest, Hungary
*ben.szalai@gmail.com
Abstract
Comparing SARS-CoV-2 infection-induced gene expression signatures to drug treatment-
induced gene expression signatures is a promising bioinformatic tool to repurpose existing
drugs against SARS-CoV-2. The general hypothesis of signature-based drug repurposing is
that drugs with inverse similarity to a disease signature can reverse disease phenotype and
thus be effective against it. However, in the case of viral infection diseases, like SARS-CoV-
2, infected cells also activate adaptive, antiviral pathways, so that the relationship between
effective drug and disease signature can be more ambiguous. To address this question, we
analysed gene expression data from in vitro SARS-CoV-2 infected cell lines, and gene
expression signatures of drugs showing anti-SARS-CoV-2 activity. Our extensive functional
genomic analysis showed that both infection and treatment with in vitro effective drugs leads
to activation of antiviral pathways like NFkB and JAK-STAT. Based on the similarity—and
not inverse similarity—between drug and infection-induced gene expression signatures, we
were able to predict the in vitro antiviral activity of drugs. We also identified SREBF1/2, key
regulators of lipid metabolising enzymes, as the most activated transcription factors by sev-
eral in vitro effective antiviral drugs. Using a fluorescently labeled cholesterol sensor, we
showed that these drugs decrease the cholesterol levels of plasma-membrane. Supple-
menting drug-treated cells with cholesterol reversed the in vitro antiviral effect, suggesting
the depleting plasma-membrane cholesterol plays a key role in virus inhibitory mechanism.
Our results can help to more effectively repurpose approved drugs against SARS-CoV-2,
and also highlights key mechanisms behind their antiviral effect.
PLOS COMPUTATIONAL BIOLOGY
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 1 / 20
a1111111111
a1111111111
a1111111111
a1111111111
a1111111111
OPEN ACCESS
Citation: Barsi S, Papp H, Valdeolivas A, To
´th DJ,
Kuczmog A, Madai M, et al. (2022) Computational
drug repurposing against SARS-CoV-2 reveals
plasma membrane cholesterol depletion as key
factor of antiviral drug activity. PLoS Comput Biol
18(4): e1010021. https://doi.org/10.1371/journal.
pcbi.1010021
Editor: James Costello, University of Colorado
Denver, UNITED STATES
Received: November 19, 2021
Accepted: March 15, 2022
Published: April 11, 2022
Peer Review History: PLOS recognizes the
benefits of transparency in the peer review
process; therefore, we enable the publication of
all of the content of peer review and author
responses alongside final, published articles. The
editorial history of this article is available here:
https://doi.org/10.1371/journal.pcbi.1010021
Copyright: ©2022 Barsi et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: All relevant data are
within the manuscript and its Supporting
Information files. All analysis code to reproduce the
Author summary
Targeting the infected host cells is an effective strategy in infectious diseases, like COVID-
19. Better understanding the virus and drug induced cellular mechanisms can help to
identify new compounds with potential antiviral activity. We used computational meth-
ods to analyse gene expression data from in vitro SARS-CoV-2 infected cell lines, and
gene expression signatures of drugs showing anti-SARS-CoV-2 activity. With the help of
machine learning methods, we were able to predict in vitro effective antiviral drugs from
gene expression based features. We found that effective drugs activate antiviral pathways
like JAK-STAT and NFkB, and also the SREBF transcription factors, key regulators of
cholesterol synthesis. Using microscopic measurements we validated that several antiviral
drugs influence the cholesterol content of the plasma membrane. Finally, we showed that
cholesterol rescue inhibited the in vitro antiviral effect of amiodarone, demonstrating the
importance of drug induced cholesterol changes in the antiviral drug effect.
1. Introduction
The newly emerged Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), caus-
ing the coronavirus disease 2019 (COVID-19), has led to more than 420,000,000 reported
infections and 5,500,000 reported deaths worldwide [1] until February 2022. Identification of
new therapeutic compounds against SARS-CoV-2 / COVID-19 is an urgent need until effec-
tive vaccination is worldwide available and given the emergence of SARS-CoV-2 strains show-
ing immune evasion [2]. The main therapeutic strategies include A) inhibiting key viral
enzymes (like remdesivir [3]); B) modulating the infected cells to decrease viral replication
[4,5] and C) modulating the over-activation of the immune system to treat late complications
like “cytokine storm” [68]. Repurposing already approved drugs for these indications is espe-
cially important as it allows a shorter time of approval for anti-SARS-CoV-2 treatment.
Comparing gene expression signatures of drugs and diseases have been previously shown
to be an effective strategy to repurpose drugs for new therapeutic indications [9]. The general
principle of these studies is that a drug inducing an opposite gene expression signature to a dis-
ease signature can reverse the disease-related gene expression changes, thus the disease pheno-
type. This “signature reversal” principle has also been used to predict effective drugs against
SARS-CoV-2 infection [1012]. However, these predictions lack, in most cases, mechanistic
insight and experimental validation. Moreover, as infected cells activate adaptive antiviral
pathways (like interferon pathway), inhibiting these pathways does not necessarily decrease
viral replication.
In this study, we analyzed transcriptomics data from in vitro SARS-CoV-2 infected cell
lines (section 2.1) and from cell lines treated with drugs showing anti-SARS-CoV-2 activity
(effective drugs, section 2.2). Functional genomic analysis revealed shared transcription factor
and pathway activity changes (eg. increased activity NFkB and JAK-STAT pathways) in the
infected and effective drug-treated cell lines. Similarity between infection signature and drug-
induced signature was predictive for in vitro effective drugs, contradictory to the classical “sig-
nature reversal” principle (section 2.3). Machine learning-based prediction of effective drugs
identified SREBF1 and SREBF2 transcription factors, key regulators of lipid metabolism, as
important factors of antiviral drug effect. Using a fluorescently labeled cholesterol sensor, we
showed the decreased level of plasma-membrane cholesterol in cells treated with effective
drugs, like chlorpromazine, confirming the effect of these drugs on cholesterol metabolism
(section 2.4). We also identified amiodarone, a drug decreasing plasma-membrane cholesterol
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 2 / 20
results of this manuscript is available at https://
github.com/comp-sys-pharm/SARS-CoV-2-
cholesterol.
Funding: BS was supported by the Premium
Postdoctoral Fellowship Program of the Hungarian
Academy of Sciences (460044). DJT and PV were
supported by the Hungarian Scientific Research
Fund (OTKA K134357). On behalf of Project
DRUGSENSPRED we thank for the usage of ELKH
Cloud (https://science-cloud.hu/) that significantly
helped us achieve the results published in this
paper. The in vitro SARS-CoV-2 experiments were
funded by the Hungarian Scientific Research Fund
(OTKA KH129599), by the European Union and the
European Social Fund (EFOP-3.6.1.-16-2016-
00004), and by the Ministry for Innovation and
Technology of Hungary (TUDFO/47138/2019-ITM)
to FJ. Also, project no. TKP2021-NVA-07 has been
implemented with the support provided from the
National Research, Development and Innovation
Fund of Hungary, financed under the TKP2021-
NVA funding scheme to FJ. The funders had no
role in study design, data collection and analysis,
decision to publish, or preparation of the
manuscript.
Competing interests: I have read the journal’s
policy and the authors of this manuscript have the
following competing interests: JSR reports funding
from GSK and Sanofi and fees from Travere
Therapeutics and Astex. BS is a full time employee
of Turbine Ltd., Budapest, Hungary.
content, thus a potential in vitro effective drug. Using an in vitro SARS-CoV-2 infection assay,
we demonstrated that the antiviral effect of amiodarone can be reversed by cholesterol supple-
ment, underlying the relevance of decreased plasma-membrane cholesterol in the antiviral
drug effect (section 2.5).
2. Results
2.1 Analysis of host pathway and transcription factor activities reveals
adaptive response of SARS-CoV-2 infected cells
We analysed gene expression data from two recent studies (GSE147507 [13] and GSE148729
[14]), where lung epithelial cancer cell lines (Calu-3 and A549) were infected with SARS-CoV-
2. To identify infection-induced pathway and transcription factor (TF) changes, we used the
PROGENy [15,16] and DoRothEA [17,18] tools, respectively (more details in Methods).
PROGENy analysis showed increased activity of NFkB and TNFa pathways in both ana-
lysed cell lines, while the activity of JAK-STAT pathway increased more pronounced in
infected Calu-3 cell lines (Fig 1A). DoRothEA analysis (Fig 1B) revealed strong activation of
STAT, IRF and NFkB transcription factors, while cell growth-related transcription factors
(E2Fs, Myc) showed decreased activity. Also SREBF1/2, key transcriptional regulators of cho-
lesterol synthesis, showed decreased activity. STATs, IRFs and NFkB pathways / TFs play a key
role in antiviral innate immunity [19]. Decreased activity of E2Fs and Myc [20] and decreased
synthesis of cholesterol [21] are also part of the physiological antiviral / interferon response.
To further analyse which upstream signalling pathways regulate the inferred TF activity
changes, we used CARNIVAL [22], a signaling network contextualisation tool, which connects
transcription factor activities to perturbations in signaling networks via integer linear pro-
gramming (more details in Methods). We performed CARNIVAL analysis using inferred tran-
scription factor activities from a SARS-CoV-2 infected cell line (GSE147507, Calu-3), and used
RIG-I like receptors (DDX58 and IFIH1), key receptors for foreign RNA sensing [23], as main
perturbation target. CARNIVAL results showed (Fig 1C), that activation of RIG-I like recep-
tors by the dsRNA of SARS-CoV-2 can directly lead to the observed transcription factor activ-
ity changes, including activation of NFkB, IRFs and STATs and inhibition of SREBF2 and
E2F4. Key identified intermediate nodes AKT1 and MAPK1 were already connected to coro-
navirus infection [5,24] and other viral infections [25,26], also suggesting that the observed TF
changes are initiated by the RIG-I like receptors, thus corresponding to the antiviral response
of the host cell.
In summary, our functional analysis of the gene expression changes in SARS-CoV-2
infected cell lines suggests that a large part of the induced pathway / transcription factor activ-
ity changes are adaptive, i.e. part of the physiological antiviral response.
2.2 Analysis of in vitro anti-SARS-CoV-2 drug-induced pathway and
transcription factor activities reveals similar changes to virus infection
To compare infection and drug-induced signatures, we used a large compendium of drug-
induced gene expression signatures from the LINCS-L1000 project [27]. LINCS-L1000 con-
tains drug-induced gene expression signatures from different cell lines, concentrations and
time points. We calculated consensus gene signatures for each drug using our previous
approach ([28], Methods), ending up with gene expression signatures for 4671 drugs. To select
drugs effectively inhibiting SARS-CoV-2 replication in vitro, we used a curated database cre-
ated by ChEMBL (http://chembl.blogspot.com/2020/05/chembl27-sars-cov-2-release.html).
This dataset contains 133 drugs previously showing effective inhibition of viral replication in 8
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 3 / 20
studies [4,2935]. We found an intersection of 47 drugs between LINCS-L1000 (available gene
expression signatures) and ChEMBL dataset (in vitro effective drugs). To characterize drug-
induced pathway and transcription activity changes, we analysed consensus drug-induced sig-
natures using PROGENy and DoRothEA.
PROGENy analysis showed strong activation of NFkB and TNFa pathways by several
drugs, including niclosamide, perhexiline and digoxin (Fig 2A). Several drugs also strongly
activated the JAK-STAT pathway (RTK inhibitors osimertinib and regorafenib). In case of TF
analysis, we found similar patterns (Fig 2B) to the infection-induced signatures: increased
activity of NFkB and STAT transcription factors and decreased activity of Myc/E2Fs transcrip-
tion factors. Interestingly, SREBF1/2 showed strongly increased activity for a large cluster of
drugs, but (similar to the infection signatures) decreased in another cluster. To further analyse
the TF activity changes in the different clusters of drugs, we calculated average TF activities for
these clusters and plotted these values against the average TF activities of the 3 SARS-CoV-2
infection signatures (Fig 2C). One cluster (Fig 2C, upper left panel), showed high correlation
(Spearman’s rho = 0.64, p = 8.55e-35) across all TFs. Two other clusters (Fig 2C, upper middle
Fig 1. Functional genomic analysis of SARS-CoV-2 infected cell lines. (A) Inferred pathway and (B) TF activities of
SARS-CoV-2 infected samples from lung epithelial cell lines (Calu-3 and A549). Activities were calculated from
differential expression signatures (infected—control) using PROGENy and DoRothEA tools for pathway and TF
activities, respectively. Only TFs with high absolute level of activity changes (absolute normalized enrichment
score >4) are shown. (C) Causal signalling network in SARS-CoV-2 infected Calu-3 cells (GSE147507) identified by
CARNIVAL. RIG-I like receptors (DDX58 and IFIH1) as perturbation targets andDoRothEA inferred TF activities
were used as the input of the CARNIVAL pipeline. Color code represents inferred activity of protein nodes (blue:
inhibited, red: activated).
https://doi.org/10.1371/journal.pcbi.1010021.g001
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 4 / 20
and upper right panels) showed lower, but still significant correlation with infection TF activity
signature (Spearman’s rho = 0.14 and 0.18, p = 0.0122 and 0.00174, respectively), with promi-
nent increase of STATs and decrease of E2F4 transcription factor activity. For the remaining
two large clusters, we found either negligible (Fig 2C, lower right panel) or high (Fig 2C, lower
left panel) correlation with infection-induced TF activities (Spearman rho = 0.04 and 0.58,
p = 0.484 and 3.14e-27, respectively), but we found high drug-induced activity of SREBF1/2
transcription factors in these clusters, opposite to the inhibition of these TFs by SARS-CoV-2
infection.
Fig 2. Functional genomic analysis of effective drugs treated cell lines. (A) Inferred pathway and (B) TF activities of
anti-SARS-CoV-2 drug-treated cell lines. Activities were calculated from LINCS-L1000 consensus drug-induced
signatures, using PROGENy and DoRothEA tools for pathway and TF activities, respectively. Drug clusters in (B) are
color coded. Only selected transcription factors (corresponding to Fig 1B) are shown. (C) Relationship between
average TF activities induced by drug treatment and SARS-CoV-2 infection for 5 different drug clusters (colors of
clusters correspond to panel B). TFs with the highest/lowest average activities are text labeled. (D) Density plot of
similarities between SARS-CoV-2- and drug-induced signatures for all LINCS-L1000 drugs and known anti-
SARS-CoV-2 drugs (ChEMBL drugs).
https://doi.org/10.1371/journal.pcbi.1010021.g002
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 5 / 20
As we found that, for several drug clusters, drug-induced TF activities showed positive cor-
relation with SARS-CoV-2-induced TF activities, we were interested in the general similarity
of drug and infection-induced gene expression signatures. To achieve this we calculated the
signature similarity (Spearman’s correlation coefficient, which has been previously shown to
be an effective metric to analyse signature similarity for the LINCS-L1000 data [27,28])
between all the 4,671 drug-induced signatures from our LINCS-L1000 dataset and the infec-
tion signatures. We found that effective anti-SARS-CoV-2 drugs (ChEMBL dataset) have
higher similarity to infection signatures, than ineffective drugs / drugs with unknown efficacy
(Fig 2E, Mann-Whitney U test p-value = <1e-200).
In summary, we found that known in vitro effective anti-SARS-CoV-2 drugs induce similar
pathway and TF activity patterns, and appropriately similar gene expression signatures to
virus infection signatures. We also identified two large clusters of drugs inducing strong acti-
vation of SREBF1/2 transcription factors, key regulators of cholesterol / lipid metabolism.
2.3 Prediction of drugs with in vitro anti-SARS-CoV-2 activity
After identifying some general patterns in the gene expression signatures of in vitro effective
anti-SARS-CoV-2 drugs, we investigated how well we can predict drug effectiveness using
gene expression signatures.
As a first strategy, we simply used the previously calculated drug—infection signature simi-
larity to predict effective drugs. Using these similarity values (predicted score) and the known
in vitro effective drugs (ChEMBL dataset, true positive values) we performed ROC analysis
(Fig 3A). We found that similarity to infection signatures is predictive for effective drugs, i.e.
drugs with high similarity to infection signature are more frequently effective (ROC AUCs:
0.75, 0.74 and 0.64 for GSE147507 A549, GSE147507 Calu-3 and GSE148729 Calu-3, respec-
tively). To test the specificity of this signature similarity-based approach for SARS-CoV-2
infection signature, we included several other virus infection-induced gene expression signa-
tures for SARS-CoV (GSE33267 [36], GSE148729), MERS (GSE45042 [37], GSE56677 [38]),
respiratory syncytial virus (RSV, GSE147507), influenza (GSE28166 [39], GSE37571) and
human parainfluenza (HPIV, GSE147507) infected Calu-3 and/or A549 cell lines. Similarity to
these infection signatures showed lower predictive performance for anti-SARS-CoV-2 drugs
(ROC AUC values <0.7 except one SARS and RSV signature with ROC AUCs 0.70 and 0.71,
respectively, Fig 3B), suggesting the relative SARS-CoV-2 specificity of the similarity-based
methods.
Following this unsupervised prediction strategy, we also performed supervised, machine
learning-based predictions. We used the drug-induced TF activities as features, and effective
drugs from the ChEMBL dataset as positive examples, with Random Forest Classification as
prediction algorithm. We set up a random subsampling based cross-validation scheme and
evaluated the performance using ROC analysis (Methods). Our results showed a slightly
improved performance compared to the unsupervised, similarity-based approach (mean ROC
AUCs: 0.72 and 0.68, 0.66, 0.57, respectively for the machine learning and similarity-based
methods, paired t-test p-values between machine learning and similarity-based methods:
3.02e-07, 2.76e-15, 4.89e-15 for GSE147507 A549, Calu-3 and GSE148729 Calu-3 signatures
respectively, Fig 3C). To gain some more mechanistic insight from the prediction of machine
learning models, we analysed feature importances (Gini importance, Fig 3D) of the Random
Forest Regression models and found that SREBF1 and SREBF2 activity were the two most
important features, followed by TFAP2A, HNF4A and TP63 transcription factors.
In summary, our two different prediction approaches showed reasonable performance
(comparable to studies based on network medicine and chemical similarity [40,41]), to predict
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 6 / 20
drugs with in vitro anti-SARS-CoV-2 activity, and also highlighted the importance of previ-
ously discussed SREBF1/2 transcription factors. Drug—SARS-CoV-2 signature similarities,
and predicted probabilities of anti-SARS-CoV-2 activity is available in S1 Table.
2.4 Anti-SARS-CoV-2 drugs are increasing SREBF activity by depleting
plasma membrane cholesterol
While in most cases we found similarity between the activity of SARS-CoV-2 infection and in
vitro effective drug-induced transcription factor activities, in case of SREBF1/2 we found
opposite changes: SARS-CoV-2 infection inhibited SREBF1/2, while a large cluster of effective
drugs lead to increased activity of SREBFs. SREBFs are activated through the decreased choles-
terol content of plasma membrane and endoplasmic reticulum, and activated SREBFs induce
the expression of cholesterol, and other lipid synthesizing enzymes [42]. From this point of
view, decreased SREBF activity during viral infection can lead to decreased cholesterol synthe-
sis, which can inhibit the viral replication and/or viral entry [21], thus can be considered as an
adaptive response of the host cell (Fig 4A). Interestingly, we observed a strongly increased
SREBF activity in large clusters of effective drugs. To resolve this discordance, we hypothesized
that these in vitro effective drugs directly decrease plasma membrane cholesterol (Fig 4A). In
this case, drug-induced decrease of plasma membrane cholesterol can contribute to the
Fig 3. Evaluation of similarity-based and machine learning-based models in predicting in vitro effective drugs. (A,
B) ROC analysis of similarity-based predictions of effective drugs against SARS-CoV-2.Drug—SARS-CoV-2 (A) or
drug—other virus (B) infection signature similarity was used as prediction score, while known in vitro effective drugs
(ChEMBL dataset) were used as true positives. (FPR: false positive rate, TPR: true positive rate) (C) Comparison of
predictive performance (ROCAUCs) of similarity-based method (similarity to SARS-CoV-2 infection signature, x-
axis) and random forest-based (RF-based, x-axis) prediction. Results of 100 random subsampling cross-validations. In
case of similarity-based methods, ROC AUC curves were only calculated for the corresponding cross-validation sets.
Boxplots represent the median (central line), first and third quartile (box), minimum and maximum non-outlier values
(whiskers) and outliers (diamonds). (D) Feature importances (Gini importance) of the Random Forest model. Top and
bottom 10 features (TFs) are shown according to importance.
https://doi.org/10.1371/journal.pcbi.1010021.g003
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 7 / 20
antiviral effect, while decreased cholesterol levels can activate SREBFs, thus explaining the
observed increased activity of these TFs in our bioinformatic analysis.
To confirm this hypothesis, we performed high-throughput, automatic confocal micros-
copy imaging using a fluorescently labeled cholesterol sensor domain, D4H-mVenus [43,44].
Fig 4. Cholesterol depleting effect of SREBF activating drugs. (A) Schematic figure of the hypothesis that antiviral
drugs block virus entry into cells by cholesterol depletion from plasma membrane, and are leading to a compensatory
increased SREBF1/2 activity. Effects induced by viral infection are marked with black arrows (left side), while orange
arrows represent drug-induced changes (right side) The figure was created with BioRender.com. (B) Schematic
representation of the used fluorescent constructs. (C) Histogram of SREBF1 activation (left panel) and histogram of
predicted probabilities of in vitro antiviral activity of LINCS-L1000 drugs (right panel, according to the Random Forest
model). Drugs selected for in vitro experiments are text labeled. (D) Representative confocal microscopy images of
D4H-mVenus transfected HEK293A cells treated with DMSO, MβCD, chlorpromazine or amiodarone. White arrows
mark plasma membrane, while red arrows show intracellular localised cholesterol sensors. (E) Time-dependent change
of log
2
(PM/IC) ratio of average cholesterol sensor intensity in HEK293A cells treated with DMSO, MβCD,
chlorpromazine, amiodarone, loperamide or rosuvastatin. Red line marks drug treatment. : significant (p<0.001)
interaction between drug treatment and elapsed time in linear model.
https://doi.org/10.1371/journal.pcbi.1010021.g004
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 8 / 20
HEK293A cells were co-transfected with D4H-mVenus and cytoplasmic Cerulean as cytosolic
marker (Fig 4B), and treated with dimethyl sulfoxide (DMSO, negative control), MβCD
(methyl β-cyclodextrin, plasma membrane cholesterol depleting compound, as positive con-
trol) and 3 drugs from our computational drug repurposing pipeline, loperamide, amiodarone
and chlorpromazine (all drugs were used in 10 μM final concentration). All these three drugs
increased the activity of SREBF transcription factors (Fig 4C, left panel). Loperamide and
chlorpromazine have been previously shown to be in vitro effective against SARS-CoV-2
(ChEMBL dataset), while amiodarone was one of the top predicted drugs of the Random For-
est model (Fig 4C, right panel, ranked 36/4671 drugs, S1 Table). We also treated HEK293A
cells with rosuvastatin, an inhibitor of cholesterol synthesis. Rosuvastatin also alters cellular
cholesterol metabolism, however, it does not influence plasma membrane cholesterol directly,
but inhibits HMG-CoA reductase, the rate limiting enzyme of de novo cholesterol synthesis.
Rosuvastatin was not predicted as an effective anti-SARS-CoV-2 drug by the Random Forest
model (Fig 4C left panel, ranked 1821/4671 drugs).
Cells were treated with the different drugs and serial confocal microscopy images were
recorded for 4.5 hours. In untreated, or DMSO treated cells, we observed a predominantly
plasma membrane localisation of the fluorescent protein labeled cholesterol sensor (Fig 4D,
top left panel). Treatment with MβCD led to decreased plasma membrane cholesterol levels,
while cholesterol accumulated in intracellular vesicles (Fig 4D, top right panel). We observed
similar phenotypic changes in case of amiodarone and chlorpromazine (Fig 4D, bottom pan-
els), while the localisation of cholesterol sensor in loperamide and rosuvastatin treated cells
was more similar to control condition (S1 Fig).
For a more systematic and unbiased analysis of the changes in the localisation of cholesterol
sensors, we performed quantitative image analysis (S2 Fig). For each cell in each image, we cal-
culated the ratio of average plasma membrane (PM) and average intracellular (IC) D4H-mVe-
nus fluorescence (PM/IC ratio). To segment cells in confocal microscopy images, we used
Cellpose library ([45], Methods). Plotting the PM/IC ratio as a function of elapsed time after
drug treatment (Fig 4E) revealed that PM/IC ratio did not decrease in loperamide and rosuvas-
tatin treated samples, while MβCD, chlorpromazine and amiodarone treatment induced sig-
nificant decrease of the ratio (linear model coefficients values for interaction between drug
treatment and time: -0.002, -0.00086, -0.00017, -0.000032 and 0.000083 for MβCD, chlorprom-
azine, amiodarone, loperamide, rosuvastatin respectively, p values: <1e-200, <1e-200, 2.71e-
09, 0.25 and 0.0047), confirming the plasma membrane cholesterol depleting effect of chlor-
promazine and amiodarone, two SREBF activating drugs.
In summary, our high-throughput image acquisition and analysis pipeline confirmed that
chlorpromazine and amiodarone decreased plasma membrane cholesterol content, which
explains the increased activity of SREBF transcription factors in case of gene expression
readout.
2.5 Supplementing cholesterol reverses anti-SARS-CoV-2 activity of
amiodarone
As our experiments revealed that the selected drugs with in vitro anti-SARS-CoV-2 activity
decreased the cholesterol content of plasma membrane, we were interested in whether
decreased plasma membrane cholesterol levels could play a causal role in the antiviral effect,
according to our assumptions (Fig 4A). To test this hypothesis, we performed in vitro SARS-
CoV-2 viral infection assay with cholesterol rescue in Vero-E6 cells.
At first we tested whether the investigated drugs show anti-SARS-CoV-2 activity in our pre-
viously described experimental system [46]. Briefly, Vero-E6 cells were co-treated with
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 9 / 20
SARS-CoV-2 and the selected drugs (6 μM amiodarone, 12 μM chlorpromazine or 50 μM
loperamide, effective drug concentrations were selected based on preliminary experiments) for
30 minutes, then washed and incubated with the drugs for 48 hours. Infection efficacy was
evaluated by microscopic examination of infection-induced cytopathic effect (CPE, more
details in Methods). Untreated, SARS-CoV-2 infected cells showed strong cytopathy (Fig 5A,
top left panel), while amiodarone, chlorpromazine and loperamide markedly reduced the
infection-induced cytopathy, confirming the antiviral effect of these drugs (Fig 5A). The used
compounds did not lead to cellular toxicity in the used concentrations (S3 Fig).
To test the effect of plasma membrane cholesterol depletion on SARS-CoV-2 infectivity, we
performed cholesterol rescue experiments (Fig 5B). Vero-E6 cells were treated with drugs
overnight, then the media was replaced with cholesterol (80 μM) containing media. After 1
hour of cholesterol treatment, the cells were infected for 30 min with SARS-CoV-2. Infection
efficacy was evaluated 48 hours after infection by droplet digital PCR based viral RNA
Fig 5. Cholesterol replenishment inhibits antiviral effect of amiodarone. (A) Predicted drugs inhibit SARS-CoV-2
replication in infected Vero-E6 cells. Vero-E6 cells were infected with SARS-CoV-2 (top left) and co-treated either
with amiodarone (top right), chlorpromazine (bottom left) or loperamide (bottom right). Antiviral effect (reduced
cytopathy) was evaluated by microscopic imaging (10x objective) 48 hours after infection. (B) Schematic figure of
cholesterol rescue experiments. The figure was created with BioRender.com. (C) Effect of cholesterol rescue on
antiviral drug effect. Vero-E6 cells were pretreated with drugs (x-axis), cholesterol was replenished (color code) and
cells were infected with SARS-CoV-2. Antiviral effect of drugs was evaluated 48 hours after infection by droplet digital
PCR (viral copies, y-axis). , #: significant (p<0.05) effect of drug treatment and drug treatment-cholesterol interaction
in linear model, respectively.
https://doi.org/10.1371/journal.pcbi.1010021.g005
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 10 / 20
quantification (Fig 5C). Chlorpromazine and loperamide did not have antiviral effect in the
pretreatment setting (linear model p values: 0.62 and 0.18, respectively), while amiodarone
decreased viral particle number significantly (linear model p-value: 1.47e-05). Cholesterol
replenishment significantly increased viral particle number in amiodarone treated Vero-E6
cells (amiodarone: cholesterol interaction term p-value: 0.026), confirming the causal role of
drug-induced cholesterol depletion in the antiviral effect of amiodarone.
Our in vitro SARS-CoV-2 infection assay confirmed the antiviral effects of chlorpromazine,
loperamide and amiodarone, and cholesterol rescue experiments suggest that plasma mem-
brane cholesterol depletion plays an important role in the antiviral effect of amiodarone.
3. Discussion
In this study, we analysed the gene expression signatures of in vitro SARS-CoV-2 infected cells
and effective anti-SARS-CoV-2 drugs. Using functional genomic computational tools, we
showed that both virus infection and drug treatment leads to similar changes of pathway and
transcription factor activities, like activation of antiviral NFkB and JAK-STAT pathways. Sig-
nature similarity between infection and drug-induced signature was predictive for drugs with
in vitro anti-SARS-CoV-2 activity, contrary to the classical “signature reversal” hypothesis.
Using machine learning models we effectively predicted anti-SARS-CoV-2 drugs, and pre-
dicted amiodarone as an in vitro antiviral compound. More detailed functional genomic analy-
sis of TF activities revealed that SREBF1/2 TFs are strongly activated by large clusters of
effective drugs. Using a high-throughput confocal microscopy setup and quantitative image
analysis we showed that two of the three investigated effective drugs influence cellular distribu-
tion of cholesterol, leading to decreased plasma membrane cholesterol content. Viral infection
assay confirmed the already described in vitro antiviral activity of loperamide and chlorproma-
zine, and also the predicted antiviral activity of amiodarone. Cholesterol supplement reversed
the antiviral effect of amiodarone, suggesting the causal role of decreased membrane choles-
terol in the antiviral effect.
Gene expression-based computational drug repurposing is a promising field to find new
disease indications of existing drugs [47]. Despite its simplicity, it has been used successfully to
identify repurposable drugs for different diseases from cancer [48,49] through inflammatory
[50] to metabolic [51] diseases. While most of the related works rely on the “signature reversal”
hypothesis, in case of infection diseases, like COVID-19, it is less clear whether signature rever-
sal (inhibiting the virus-hijacked signalisation) or signature similarity (promoting the antiviral
response of infected host cells) can be more effective. While early studies at the beginning of
the COVID-19 pandemic applied mostly the original signature reversal hypothesis, more
recent works [52,53] also assumed that drugs with similarity to the SARS-CoV-2-induced gene
expression signature can be effective. In our work, we performed a more unbiased analysis of
signature-based drug repurposing against SARS-CoV-2. We compared the gene expression
signatures of known effective drugs against SARS-CoV-2 infection signatures, and found that
signature similarity, and not dissimilarity, is predictive for antiviral effect. These results suggest
that increasing the antiviral response of host cells can be a more effective strategy than inhibit-
ing viral infection-induced pathways. Whether this is specific for SARS-CoV-2, known for
evading several antiviral systems of the host cell [54], or a general mechanism for (viral) infec-
tions, needs further analysis with large scale in vitro drug screenings against other viruses. Nev-
ertheless, using our “signature similarity” principle instead of—or together with—the
“signature reversal” hypothesis can accelerate computational drug repurposing against existing
and emerging infectious diseases, complemented by network-based repurposing strategies
[40,55].
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 11 / 20
While signature (dis)similarity-based computational drug repurposing has promising pre-
dictive performance, it gives no real mechanistic insight. To overcome this problem, we per-
formed extensive functional genomic analysis of SARS-CoV-2 and drug-induced gene
expression signatures. We found that both viral infection and effective drugs stimulate known
antiviral pathways like NFkB and JAK-STAT. We observed lower induction of these pathways
in virus-infected A549 cells, compared to Calu-3 cell lines, probably based on the lower expres-
sion of virus receptor ACE2 in A549 cells. The activation of antiviral pathways in virus-
infected and effective drug-treated cells also supports the “signature similarity” principle.
Beside the activation of antiviral TFs and pathways, we also observed inhibition of
(inferred) SREBF1/2 transcription factors in SARS-CoV-2 infected samples, while an activa-
tion of these TFs in a large cluster of antiviral drug-treated cells. SREBF1/2 regulate the expres-
sion of key members of cholesterol synthesis. Cholesterol depletion of plasma membrane can
reduce SARS-CoV-2 infection [56,57], and decreasing SREBFs activity (and cholesterol syn-
thesis) can be also part of the physiological, interferon-induced antiviral response of the host
cell [21,58,59]. Recently, inhibitors of the SREBFs—DNA interaction were found to exert anti-
viral effects [60] and CRISPR based knock-out of the SREBF pathway members also led to
SARS-CoV-2 resistant phenotype [61], suggesting that inhibition of SREBFs could be benefi-
cial in case of SARS-CoV-2 infection. In contrast, we found increased activity of SREBFs in
case of several effective drug-induced gene expression signatures. Previous works also showed
the increased expression of lipid metabolic enzymes [12] in antiviral drug-treated cells, and a
recent large scale CRISPR screen [62] also found that increased cholesterol synthesis can
reduce SARS-CoV-2 infection. However, these two later conclusions were based on the analy-
sis of gene expression changes of the cholesterol synthetic pathway. Gene expression changes
are in several cases not the cause, but the (compensatory) consequence of perturbed cell states
[63,64]. Based on this, we hypothesized that increased SREBF1/2 activity (based on transcrip-
tional readout) can be a compensatory consequence of decreased plasma membrane choles-
terol levels in case of several antiviral drugs. Using a fluorescent cholesterol sensor, we found
that amiodarone and chlorpromazine, two effective in vitro antiviral drugs, indeed decreases
the cholesterol content of plasma membranes, which can explain the (compensatory) increased
SREBF1/2 activity. In an in vitro SARS-CoV-2 infection assay, coupled with cholesterol rescue,
we also showed that cholesterol replenishment reduced the antiviral activity of amiodarone,
thus confirmed the causal role of plasma membrane cholesterol decrease in the antiviral effect
of amiodarone. While our computational analysis also predicted that PM cholesterol depletion
plays a role in the antiviral effect of chlorpromazine and loperamide, we were not able to verify
these predictions experimentally. Noteworthy, these two drugs had antiviral effect in case of
co-treatment with virus infection, but not in the case of the pre-treatment setup used in choles-
terol rescue experiments (probably due to pharmacokinetic factors). It is thus hard to draw
conclusions about the role of cholesterol in the antiviral effect of these drugs.
While we showed that PM cholesterol depletion can be an important factor in the in vitro
antiviral effect of drugs, whether this can be translated to in vivo is still an open question. A
recent large scale study [65] showed that several in vitro repurposable drugs exert their antivi-
ral effect via altering the membrane composition of drug-treated cells, and this antiviral effect
has low translation potential based on concerns regarding drug concentration and adverse
effects. While the authors of this study concluded that phospholipidosis is the main drug-
induced membrane component change, our results argue that altered cholesterol content can
also be a causal factor in the antiviral effect of drugs. Whether altered lipid composition of cel-
lular membranes is only a factor confounding drug repurposing studies, or this effect can be
exploited towards effective therapy, needs further studies.
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 12 / 20
In summary, our study showed that in vitro SARS-CoV-2 infection and effective antiviral
drugs lead to similar pathway and transcription activity changes. We found that gene expres-
sion signature similarity, and not the dissimilarity, predicts in vitro effective antiviral com-
pounds, which can accelerate computational drug repurposing against infectious diseases, and
we made the results of our predictions available for the research community (S1 Table). We
also identified that plasma membrane cholesterol depletion plays an important role in the
mechanism of action of several antiviral drugs, and that cholesterol replenishment inhibits the
in vitro antiviral effect of amiodarone, thus our results also give mechanistic insight about the
antiviral effect of repurposable drugs.
4. Methods
Virus infection-induced gene expression signatures
Microarray gene expression profiles of different virus-infected cell lines were downloaded
from Gene Expression Omnibus (GEO) with accession numbers GSE28166 (H5N1),
GSE37571 (Influenza), GSE33267 (SARS-CoV-1), GSE56677 and GSE45042 (MERS-CoV).
Preprocessing and differential expression (DE) analysis was performed by using R package
limma [66].
Total RNA-Seq profiles of SARS-CoV-2 and other virus-infected human cell lines were
downloaded from GEO with accession numbers GSE147507 (SARS-CoV-2, RSV, IAV, HPIV)
and GSE148729 (SARS-CoV-1 and 2). Differential expression (DE) analysis was performed
using R library DESeq2 [67].
In all gene expression datasets, we used (virus-infected—control) contrasts for differential
expression calculation, where the control condition was mock infection. Where gene expres-
sion data after multiple time points were available, we used 24 h post-infection data. Shared
genes across all datasets were selected and further analyzed.
Drug treatment-induced signatures
We used Level 5 gene expression profiles from the LINCS-L1000 dataset [27]. We calculated
consensus expression signatures for each drug (across different cell lines, concentrations and
time points) using the MODZ method [27,28]. We matched LINCS-L1000 drugs with
ChEMBL effective drug dataset (http://chembl.blogspot.com/2020/05/chembl27-sars-cov-
2-release.html) using drug names and simplified molecular-input line-entry system (SMILES).
Only measured (landmark) genes were used in the further analysis.
Functional genomic analysis
From previously calculated SARS-CoV-2 infection and effective drug-induced signatures, we
inferred pathway activities using PROGENy (R package progeny [15,16]) and transcription fac-
tor activities using DoRothEA (R package dorothea [18]).
PROGENy was applied to infer activities of 14 different pathways from expression and
weight of their footprint gene sets. Z-scores of pathway activities were calculated using 10000
permutations of genes as background distribution. DoRothEA was applied to infer transcrip-
tion factor activities using the viper algorithm [68]. DoRothEA is a collected, curated resource
of signed TF-target interactions. Interactions are assigned a confidence level ranging from A
(highest) to E (lowest) based on the number of supporting evidence. In this study interactions
assigned A, B, C confidence levels were used. In transcription factor activity heatmaps (Figs 1B
and 2B), only selected transcription factors (absolute normalised enrichment score >4 in
SARS-CoV–2 infection signatures) are shown.
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 13 / 20
We used the CARNIVAL tool [22] to contextualize our transcriptomics-based results into a
mechanistic causal network. Briefly, CARNIVAL takes as input a prior knowledge network
and a set of constraints and infers the most likely causal interactions by solving an integer lin-
ear programming problem. We assembled a curated prior knowledge signaling network from
OmniPath resources [69]. As constraints, we selected the RIG-I like receptors (DDX58 and
IFIH1) as upstream signaling perturbation and the top 25 most deregulated TFs (according to
DoRothEA and viper results) upon SARS-CoV-2 infection as their downstream target. In addi-
tion, we used PROGENy pathway activity scores to weight the prior knowledge network and
assist CARNIVAL in the discovery of optimal networks connecting the upstream perturbation
(RIG-I like receptors) to the downstream targets (TFs).
Signature similarity and machine learning-based prediction
We calculated similarities using Spearman’s correlation between each virus infection-induced
and each drug treatment-induced signature after selecting shared genes.
TF activity scores from drug-treated cells were used to predict effective drugs against
SARS-CoV-2 using Random Forest Classifier from scikit-learn Python library [70]. The model
was trained using 300 trees, with default parameters otherwise and with 100 different training
sets. Training sets consisted of a 50% random sampling of effective drugs and non-effective
drugs as well. The average importance of features (TFs) was computed (sum of feature impor-
tances, divided by the number of models). Predicted probabilities of antiviral activity were also
computed in each prediction and the mean of them was calculated for each drug (probabilities
were summed for each drug and divided by the number of occurrences in validation sets).
We performed ROC analysis using scikit-learn Python library to evaluate similarity-based
and machine learning-based predictions. Effective drugs against SARS-CoV-2 curated by
ChEMBL and overlapping with drugs of the LINCS-L1000 dataset were used as the positive
class. The negative class consisted of the part of drugs from the LINCS-L1000 dataset not con-
sidered as effective by ChEMBL. To compare machine learning-based and similarity-based
methods ROC curves were computed for each different validation set (100) and signature simi-
larity scores of the corresponding drugs were considered.
Fluorescent cholesterol sensor experiments
The cellular cholesterol sensor used in this study was the D4H domain [43,44] fluorescently
labeled with monomer Venus (mVenus) on its N-terminus. To create the construct coding
this sensor, we used a plasmid coding the bioluminescent version of the sensor (described in
[71]), a kind gift from Tamas Balla (NICHD, NIH, Bethesda, USA). The D4H domain-coding
sequence from this plasmid was subcloned into the pEYFP-C1 plasmid containing mVenus in
place of EYFP, using BglII and BamHI restriction enzymes. Cytosolic Cerulean was expressed
from a pEYFP-N1 plasmid where EYFP had been replaced with Cerulean.
For fluorescent imaging, HEK293A cells (ATCC, USA) were maintained in Dulbecco’s
Modified Eagle Medium (DMEM—Lonza, Switzerland) complemented with 10% fetal bovine
serum (Biosera, France) and Penicillin/Streptomycin (100 U/ml and 100 μg/ml, respectively—
Lonza, Switzerland). Cells were seeded on poly-L-lysine pretreated (0.001%, 1h) 24-well imag-
ing plates (Eppendorf, Germany) at a density of 1e05 cells/well. On the next day, cells were co-
transfected with plasmids coding cytoplasmic Cerulean and D4H-mVenus (0.25 μg/well each)
using Lipofectamine 2000 (0.75 μl/well, Invitrogen, USA).
Image acquisition started 24h post-transfection, after the medium had been changed to
300 μl/well HEPES-buffered DMEM without phenol-red (Gibco, USA). Images were acquired
automatically using the ImageXpress Micro Confocal High-Content Imaging System
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 14 / 20
(Molecular Devices, USA), with a 40x Plan Fluor objective. CFP-2432C and fluorescein iso-
thiocyanate (FITC) filter sets were used for Cerulean and D4H-mVenus images, respectively,
both with an exposure time of 300 ms. After acquiring control images (30 min), cells were
treated with either DMSO (as control) or with the drugs indicated on Fig 4E in a volume of
100 μl/well (270 min). Measurements were performed at 30˚C. Three independent measure-
ments were made, with duplicate wells for each condition and 5 images/well taken for each
time point.
All chemicals used for treatment were purchased from Sigma-Aldrich Merck (Germany).
Amiodarone HCl, chlorpromazine HCl, loperamid HCl and rosuvastatin calcium were dis-
solved in DMSO, stored at -20˚C as 10 mM stock solutions and diluted in cell medium
promptly before cell treatment to a final concentration of 10 μM. MβCD was stored as powder
at 4˚C and freshly dissolved in cell medium before treatment to a final concentration of 10
mM.
Image analysis pipeline
Images were segmented with Cellpose Python library [45], which is a generalist, deep learning-
based segmentation method. To select high-quality images the cytoplasm marker channel was
used with Laplace filtering. We used high-quality images (filtered according to an appropriate
upper threshold of standard deviation of Laplace value in each experiment) as input of the
Cellpose model, with parameter channel set to greyscale and cell diameter greater than 200
pixels.
After identifying cell boundaries, we applied binary erosion (scipy Python library [72]) with
default structure and 10 iterations to determine cytoplasm boundary, or binary dilation with
default structure and 5 iterations to determine PM outer boundary. The boundary of PM was
determined by subtracting the cytoplasm boundary from the outer boundary. We calculated
the log
2
ratio of the mean PM and mean intracellular D4H fluorescence intensities for each
cell in the D4H channel to examine the changes of plasma membrane cholesterol distribution.
For statistical analysis, we used log2(PM/IC) ~ Time + Time: Drug + Exp linear model, where
Time corresponds to elapsed time after drug treatment, Drug factor represents the used drug,
using DMSO as reference level. Exp factor represents the (n = 3) individual experiments.
Viral infection and cholesterol rescue experiments
Amiodarone HCl (Sigma-Aldrich, Merck KGaA, Germany) was dissolved in DMSO (Sigma-
Aldrich, Merck KGaA, Germany) and kept at -20˚C. Chlorpromazine (in house synthesized
based on [73]) and loperamide HCl (Sigma-Aldrich, Merck KGaA, Germany) were freshly dis-
solved in water and filtered prior to the treatment. 10 mM stock solutions were made from the
drugs. Vero-E6 cells were seeded in a 96-well plate on the day before the experiments. On the
next day the cells were treated with 100 μl of 50 μM remdesivir or loperamide or 12 μM chlor-
promazine or 6 μM amiodarone solution overnight. 1 hour prior to the infection the cell cul-
ture media containing the different drugs was replaced with media containing 80 μM
cholesterol (Sigma-Aldrich, Merck KGaA, Germany). After the 1-hour-long cholesterol treat-
ment the cells were infected with SARS-CoV-2 (GISAID accession ID: EPI_ISL_483637) at
MOI:0.01 in a BSL-4 laboratory. Cells were incubated with the virus for 30 minutes then the
media was replaced with fresh cell culture media. During the investigation (except cell seeding)
DMEM (Lonza Group Ltd, Switzerland) supplemented with 1% Penicillin- Streptomycin
(Lonza Group Ltd, Switzerland) and 2% heat-inactivated fetal bovine serum (Gibco, Thermo
Fisher Scientific Inc., MA, USA) were used. 48 hours post infection (hpi) the cells were
inspected under microscope and RNA was extracted from the supernatant (Zybio EXM 3000
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 15 / 20
Nucleic Acid Isolation System, Nucleic Acid Extraction Kit B200-32). Viral copy number was
determined using droplet-digital PCR technology (Bio-Rad Laboratories Inc., CA, USA).
SARS-CoV-2 RdRp gene specific primers and probe were utilized (Forward: GTGARATGGT
CATGTGTGGCGG, reverse: CARATGTTAAASACACTATTAGCATA and the probe was:
FAM-CAGGTGGAACCTCATCAGGAGATGC-BBQ). For statistical analysis, measured viral
copy numbers were log
2
transformed, and we used a log2(CV) ~ Drug Cholesterol + Exp,
where Drug factor represents the used drug (untreated as reference level), Cholesterol factor
represents cholesterol replenishment treatment (no treatment as reference level). Exp factor
corresponds to the (n = 4) individual experiments.
Supporting information
S1 Fig. Representative confocal microscopy images of D4H-mVenus transfected HEK293A
cells treated with loperamide and rosuvastatin.
(TIFF)
S2 Fig. Determination of cytoplasm and membrane boundaries. Cells are detected on the
cytoplasm marker channel, then boundaries of cytoplasm and membrane are determined for
each cell. The D4H channel is used for the calculation of the PM/IC ratio.
(TIFF)
S3 Fig. Absence of marked toxic effects of used drugs in the tested concentrations.
(TIFF)
S1 Table. Drug signature similarities (Spearman’s rho) to SARS-CoV-2 infection signa-
tures (GSE147507 A549 SARS-CoV-2, GSE147507 Calu3 SARS-CoV-2 and GSE148729
Calu3 SARS-CoV-2 columns, respectively) and Random Forest Classifier based predicted
probability of antiviral effect.
(CSV)
Acknowledgments
We thank Aure
´lien Dugourd, GergőGulya
´s, Kinga Kova
´cs and Andra
´s D. To
´th for the helpful
discussions regarding the manuscript, and Pe
´ter Ma
´tyus for his help in organising the collabo-
rations between computational and experimental groups. We thank Kata Szabolcsi for techni-
cal assistance in the cholesterol sensor experiments, Katalin Gombos, Zso
´fia Lanszki and
Bala
´zs Somogyi for their help in the in vitro infection experiments, and Tama
´s Ka
´lai for com-
pound synthesis. On behalf of Project DRUGSENSPRED we thank for the usage of ELKH
Cloud (https://science-cloud.hu/) that significantly helped us achieve the results published in
this paper. Schematic figures were created with BioRender.com.
Author Contributions
Conceptualization: Szilvia Barsi, Bence Szalai.
Data curation: Szilvia Barsi, Bence Szalai.
Formal analysis: Szilvia Barsi, Alberto Valdeolivas, Bence Szalai.
Funding acquisition: Pe
´ter Va
´rnai, Ferenc Jakab, Bence Szalai.
Investigation: Szilvia Barsi, Henrietta Papp, Da
´niel J. To
´th, Anett Kuczmog, Mo
´nika Madai,
Pe
´ter Va
´rnai, Bence Szalai.
Methodology: Szilvia Barsi, Alberto Valdeolivas, Bence Szalai.
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 16 / 20
Project administration: La
´szlo
´Hunyady.
Resources: Da
´niel J. To
´th, Pe
´ter Va
´rnai, Ferenc Jakab, Bence Szalai.
Supervision: La
´szlo
´Hunyady, Pe
´ter Va
´rnai, Julio Saez-Rodriguez, Ferenc Jakab, Bence Szalai.
Validation: Henrietta Papp, Da
´niel J. To
´th, Anett Kuczmog, Mo
´nika Madai, Pe
´ter Va
´rnai.
Visualization: Szilvia Barsi, Alberto Valdeolivas.
Writing – original draft: Szilvia Barsi, Bence Szalai.
Writing – review & editing: Szilvia Barsi, Henrietta Papp, Alberto Valdeolivas, Da
´niel J. To
´th,
Anett Kuczmog, Mo
´nika Madai, La
´szlo
´Hunyady, Pe
´ter Va
´rnai, Julio Saez-Rodriguez, Fer-
enc Jakab, Bence Szalai.
References
1. Dong E, Du H, Gardner L. An interactive web-based dashboard to track COVID-19 in real time. Lancet
Infect Dis. 2020; 20: 533–534. https://doi.org/10.1016/S1473-3099(20)30120-1 PMID: 32087114
2. McCallum M, Bassi J, De Marco A, Chen A, Walls AC, Di Iulio J, et al. SARS-CoV-2 immune evasion by
the B.1.427/B.1.429 variant of concern. Science. 2021; 373: 648–654. https://doi.org/10.1126/science.
abi7994 PMID: 34210893
3. Wang M, Cao R, Zhang L, Yang X, Liu J, Xu M, et al. Remdesivir and chloroquine effectively inhibit the
recently emerged novel coronavirus (2019-nCoV) in vitro. Cell Res. 2020; 30: 269–271. https://doi.org/
10.1038/s41422-020-0282-0 PMID: 32020029
4. Gordon DE, Jang GM, Bouhaddou M, Xu J, Obernier K, White KM, et al. A SARS-CoV-2 protein interac-
tion map reveals targets for drug repurposing. Nature. 2020. https://doi.org/10.1038/s41586-020-2286-
9PMID: 32353859
5. Bouhaddou M, Memon D, Meyer B, White KM, Rezelj VV, Correa Marrero M, et al. The Global Phos-
phorylation Landscape of SARS-CoV-2 Infection. Cell. 2020. https://doi.org/10.1016/j.cell.2020.06.034
PMID: 32645325
6. Cuccarese MF, Earnshaw BA, Heiser K, Fogelson B, Davis CT, McLean PF, et al. Functional immune
mapping with deep-learning enabled phenomics applied to immunomodulatory and COVID-19 drug dis-
covery. 2020. p. 2020.08.02.233064. https://doi.org/10.1101/2020.08.02.233064
7. Ragab D, Salah Eldin H, Taeimah M, Khattab R, Salem R. The COVID-19 Cytokine Storm; What We
Know So Far. Front Immunol. 2020; 11: 1446. https://doi.org/10.3389/fimmu.2020.01446 PMID:
32612617
8. Olbei M, Hautefort I, Modos D, Treveil A, Poletti M, Gul L, et al. SARS-CoV-2 Causes a Different Cyto-
kine Response Compared to Other Cytokine Storm-Causing Respiratory Viruses in Severely Ill
Patients. Front Immunol. 2021; 12: 629193. https://doi.org/10.3389/fimmu.2021.629193 PMID:
33732251
9. Sirota M, Dudley JT, Kim J, Chiang AP, Morgan AA, Sweet-Cordero A, et al. Discovery and preclinical
validation of drug indications using compendia of public gene expression data. Sci Transl Med. 2011; 3:
96ra77. https://doi.org/10.1126/scitranslmed.3001318 PMID: 21849665
10. Napolitano F, Gambardella G, Carrella D, Gao X, di Bernardo D. Computational Drug Repositioning
and Elucidation of Mechanism of Action of Compounds against SARS-CoV-2. arXiv [q-bio.GN]. 2020.
Available: http://arxiv.org/abs/2004.07697
11. Zhou Y, Hou Y, Shen J, Huang Y, Martin W, Cheng F. Network-based drug repurposing for novel coro-
navirus 2019-nCoV/SARS-CoV-2. Cell Discov. 2020; 6: 14. https://doi.org/10.1038/s41421-020-0153-3
PMID: 32194980
12. Hoagland DA, Clarke DJB, Møller R, Han Y, Yang L, Wojciechowicz ML, et al. Modulating the transcrip-
tional landscape of SARS-CoV-2 as an effective method for developing antiviral compounds. bioRxiv.
2020. p. 2020.07.12.199687. https://doi.org/10.1101/2020.07.12.199687
13. Blanco-Melo D, Nilsson-Payant BE, Liu W-C, Uhl S, Hoagland D, Møller R, et al. Imbalanced Host
Response to SARS-CoV-2 Drives Development of COVID-19. Cell. 2020; 181: 1036–1045.e9. https://
doi.org/10.1016/j.cell.2020.04.026 PMID: 32416070
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 17 / 20
14. Wyler E, Mo
¨sbauer K, Franke V, Diag A, Gottula LT, ArsièR, et al. Transcriptomic profiling of SARS-
CoV-2 infected human cell lines identifies HSP90 as target for COVID-19 therapy. iScience. 2021; 24:
102151. https://doi.org/10.1016/j.isci.2021.102151 PMID: 33585804
15. Schubert M, Klinger B, Klu¨nemann M,Sieber A, Uhlitz F, Sauer S, et al. Perturbation-response genes
reveal signaling footprints in cancer gene expression. Nat Commun. 2018; 9: 20. https://doi.org/10.
1038/s41467-017-02391-6 PMID: 29295995
16. Holland CH, Szalai B, Saez-Rodriguez J. Transfer of regulatory knowledge from human to mouse for
functional genomics analysis. Biochim Biophys Acta Gene Regul Mech. 2019; 194431. https://doi.org/
10.1016/j.bbagrm.2019.194431 PMID: 31525460
17. Garcia-Alonso L, Iorio F, Matchan A, Fonseca N, Jaaks P, Peat G, et al. Transcription Factor Activities
Enhance Markers of Drug Sensitivity in Cancer. Cancer Res. 2018; 78: 769–780. https://doi.org/10.
1158/0008-5472.CAN-17-1679 PMID: 29229604
18. Garcia-Alonso L, Holland CH, Ibrahim MM, Turei D, Saez-Rodriguez J. Benchmark and integration of
resources for the estimation of human transcription factor activities. Genome Res. 2019; 29: 1363–
1375. https://doi.org/10.1101/gr.240663.118 PMID: 31340985
19. Seth RB, Sun L, Chen ZJ. Antiviral innate immunity pathways. Cell Res. 2006; 16: 141–147. https://doi.
org/10.1038/sj.cr.7310019 PMID: 16474426
20. Ramana CV, Grammatikakis N, Chernov M, Nguyen H, Goh KC, Williams BR, et al. Regulation of c-
myc expression by IFN-gamma through Stat1-dependent and -independent pathways. EMBO J. 2000;
19: 263–272. https://doi.org/10.1093/emboj/19.2.263 PMID: 10637230
21. York AG, Williams KJ, Argus JP, Zhou QD, Brar G, Vergnes L, et al. Limiting Cholesterol Biosynthetic
Flux Spontaneously Engages Type I IFN Signaling. Cell. 2015; 163: 1716–1729. https://doi.org/10.
1016/j.cell.2015.11.045 PMID: 26686653
22. Liu A, Trairatphisan P, Gjerga E, Didangelos A, Barratt J, Saez-Rodriguez J. From expression footprints
to causal pathways: contextualizing large signaling networks with CARNIVAL. NPJ Syst Biol Appl.
2019; 5: 40. https://doi.org/10.1038/s41540-019-0118-z PMID: 31728204
23. Rehwinkel J, Gack MU. RIG-I-like receptors: their regulation and roles in RNA sensing. Nat Rev Immu-
nol. 2020; 20: 537–551. https://doi.org/10.1038/s41577-020-0288-3 PMID: 32203325
24. Kindrachuk J, Ork B, Hart BJ, Mazur S, Holbrook MR, Frieman MB, et al. Antiviral potential of ERK/
MAPK and PI3K/AKT/mTOR signaling modulation for Middle East respiratory syndrome coronavirus
infection as identified by temporal kinome analysis. Antimicrob Agents Chemother. 2015; 59: 1088–
1099. https://doi.org/10.1128/AAC.03659-14 PMID: 25487801
25. Tian J, Zhang X, Wu H, Liu C, Li Z, Hu X, et al. Blocking the PI3K/AKT pathway enhances mammalian
reovirus replication by repressing IFN-stimulated genes. Front Microbiol. 2015; 6: 886. https://doi.org/
10.3389/fmicb.2015.00886 PMID: 26388843
26. Ehrhardt C, Wolff T, Pleschka S, Planz O, Beermann W, Bode JG, et al. Influenza A virus NS1 protein
activates the PI3K/Akt pathway to mediate antiapoptotic signaling responses. J Virol. 2007; 81: 3058–
3067. https://doi.org/10.1128/JVI.02082-06 PMID: 17229704
27. Subramanian A, Narayan R, Corsello SM, Peck DD, Natoli TE, Lu X, et al. A Next Generation Connec-
tivity Map: L1000 Platform and the First 1,000,000 Profiles. Cell. 2017; 171: 1437–1452.e17. https://doi.
org/10.1016/j.cell.2017.10.049 PMID: 29195078
28. Szalai B, Subramanian V, Holland CH, Alfo
¨ldi R, Puska
´s LG, Saez-Rodriguez J. Signatures of cell
death and proliferation in perturbation transcriptomics data-from confounding factor to effective predic-
tion. Nucleic Acids Res. 2019; 47: 10010–10026. https://doi.org/10.1093/nar/gkz805 PMID: 31552418
29. Riva L, Yuan S, Yin X, Martin-Sancho L, Matsunaga N, Burgstaller S, et al. A Large-scale Drug Reposi-
tioning Survey for SARS-CoV-2 Antivirals. bioRxiv. 2020. p. 2020.04.16.044016. https://doi.org/10.
1101/2020.04.16.044016 PMID: 32511357
30. Weston S, Coleman CM, Haupt R, Logue J, Matthews K, Frieman MB. Broad anti-coronaviral activity of
FDA approved drugs against SARS-CoV-2 in vitro and SARS-CoV in vivo. 2020. p.
2020.03.25.008482. https://doi.org/10.1101/2020.03.25.008482
31. Heiser K, McLean PF, Davis CT, Fogelson B, Gordon HB, Jacobson P, et al. Identification of potential
treatments for COVID-19 through artificial intelligence-enabled phenomic analysis of human cells
infected with SARS-CoV-2. bioRxiv. 2020. p. 2020.04.21.054387. https://doi.org/10.1101/2020.04.21.
054387
32. Si L, Bai H, Rodas M, Cao W, Oh CY, Jiang A, et al. Human organ chip-enabled pipeline to rapidly
repurpose therapeutics during viral pandemics. 2020. p. 2020.04.13.039917. https://doi.org/10.1101/
2020.04.13.039917
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 18 / 20
33. Touret F, Gilles M, Barral K, Nougairède A, van Helden J, Decroly E, et al. In vitro screening of a FDA
approved chemical library reveals potential inhibitors of SARS-CoV-2 replication. Sci Rep. 2020; 10:
13093. https://doi.org/10.1038/s41598-020-70143-6 PMID: 32753646
34. Ellinger B, Bojkova D, Zaliani A, Cinatl J, Claussen C, Westhaus S, et al. A SARS-CoV-2 cytopathicity
dataset generated by high-content screening of a large drug repurposing collection. Sci Data. 2021; 8:
70. https://doi.org/10.1038/s41597-021-00848-4 PMID: 33637768
35. Jeon S, Ko M, Lee J, Choi I, Byun SY, Park S, et al. Identification of antiviral drug candidates against
SARS-CoV-2 from FDA-approved drugs. https://doi.org/10.1101/2020.03.20.999730
36. Sims AC, Tilton SC, Menachery VD, Gralinski LE, Scha
¨fer A, Matzke MM, et al. Release of severe
acute respiratory syndrome coronavirus nuclear import block enhances host transcription in human
lung cells. J Virol. 2013; 87: 3885–3902. https://doi.org/10.1128/JVI.02520-12 PMID: 23365422
37. Josset L, Menachery VD, Gralinski LE, Agnihothram S, Sova P, Carter VS, et al. Cell host response to
infection with novel human coronavirus EMC predicts potential antivirals and important differences with
SARS coronavirus. MBio. 2013; 4: e00165–13. https://doi.org/10.1128/mBio.00165-13 PMID: 23631916
38. Selinger C, Tisoncik-Go J, Menachery VD, Agnihothram S, Law GL, Chang J, et al. Cytokine systems
approach demonstrates differences in innate and pro-inflammatory host responses between genetically
distinct MERS-CoV isolates. BMC Genomics. 2014; 15: 1161. https://doi.org/10.1186/1471-2164-15-
1161 PMID: 25534508
39. Li C, Bankhead A 3rd, Eisfeld AJ, Hatta Y, Jeng S, Chang JH, et al. Host regulatory network response
to infection with highly pathogenic H5N1 avian influenza virus. J Virol. 2011; 85: 10955–10967. https://
doi.org/10.1128/JVI.05792-11 PMID: 21865398
40. Morselli Gysi D, do Valle I
´, Zitnik M, Ameli A, Gan X, Varol O, et al. Network medicine framework for
identifying drug-repurposing opportunities for COVID-19. Proc Natl Acad Sci U S A. 2021;118. https://
doi.org/10.1073/pnas.2025581118 PMID: 33906951
41. Liang H, Zhang L, Wang L, Gao M, Meng X, Li M, et al. Repositioning Drugs on Human Influenza A
Viruses Based on a Novel Nuclear Norm Minimization Method. Front Physiol. 2020; 11: 597494. https://
doi.org/10.3389/fphys.2020.597494 PMID: 33536933
42. Horton JD, Goldstein JL, Brown MS. SREBPs: activators of the complete program of cholesterol and
fatty acid synthesis in the liver. J Clin Invest. 2002; 109: 1125–1131. https://doi.org/10.1172/JCI15593
PMID: 11994399
43. Maekawa M, Fairn GD. Complementary probes reveal that phosphatidylserine is required for the proper
transbilayer distribution of cholesterol. J Cell Sci. 2015; 128: 1422–1433. https://doi.org/10.1242/jcs.
164715 PMID: 25663704
44. Maekawa M. Domain 4 (D4) of Perfringolysin O to Visualize Cholesterol in Cellular Membranes-The
Update. Sensors. 2017;17. https://doi.org/10.3390/s17030504 PMID: 28273804
45. Stringer C, Wang T, Michaelos M, Pachitariu M. Cellpose: a generalist algorithm for cellular segmenta-
tion. Nat Methods. 2021; 18: 100–106. https://doi.org/10.1038/s41592-020-01018-x PMID: 33318659
46. Konrat R, Papp H, Szija
´rto
´V, Gesell T, Nagy G, Madai M, et al. The Anti-histamine Azelastine, Identified
by Computational Drug Repurposing, Inhibits SARS-CoV-2 Infection in Reconstituted Human Nasal
Tissue In Vitro. bioRxiv. 2020. p. 2020.09.15.296228. https://doi.org/10.1101/2020.09.15.296228
47. Pushpakom S, Iorio F, Eyers PA, Escott KJ, Hopper S, Wells A, et al. Drug repurposing: progress, chal-
lenges and recommendations. Nat Rev Drug Discov. 2019; 18: 41–58. https://doi.org/10.1038/nrd.
2018.168 PMID: 30310233
48. Chen B, Ma L, Paik H, Sirota M, Wei W, Chua M-S, et al. Reversal of cancer gene expression correlates
with drug efficacy and reveals therapeutic targets. Nat Commun. 2017; 8: 16022. https://doi.org/10.
1038/ncomms16022 PMID: 28699633
49. Stathias V, Jermakowicz AM, Maloof ME, Forlin M, Walters W, Suter RK, et al. Drug and disease signa-
ture integration identifies synergistic combinations in glioblastoma. Nat Commun. 2018; 9: 5315. https://
doi.org/10.1038/s41467-018-07659-z PMID: 30552330
50. Malcomson B, Wilson H, Veglia E, Thillaiyampalam G, Barsden R, Donegan S, et al. Connectivity map-
ping (ssCMap) to predict A20-inducing drugs and their antiinflammatory action in cystic fibrosis. Proc
Natl Acad Sci U S A. 2016; 113: E3725–34. https://doi.org/10.1073/pnas.1520289113 PMID: 27286825
51. Kunkel SD, Suneja M, Ebert SM, Bongers KS, Fox DK, Malmberg SE, et al. mRNA expression signa-
tures of human skeletal muscle atrophy identify a natural compound that increases muscle mass. Cell
Metab. 2011; 13: 627–638. https://doi.org/10.1016/j.cmet.2011.03.020 PMID: 21641545
52. Laise P, Bosker G, Sun X, Shen Y, Douglass EF, Karan C, et al. The Host Cell ViroCheckpoint: Identifi-
cation and Pharmacologic Targeting of Novel Mechanistic Determinants of Coronavirus-Mediated
Hijacked Cell States. bioRxiv. 2020. p. 2020.05.12.091256. https://doi.org/10.1101/2020.05.12.091256
PMID: 32511361
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 19 / 20
53. Chen F, Shi Q, Pei F, Vogt A, Porritt RA, Garcia G Jr, et al. A systems-level study revealshost-targeted
repurposable drugs against SARS-CoV-2 infection. Mol Syst Biol. 2021; 17: e10239. https://doi.org/10.
15252/msb.202110239 PMID: 34339582
54. Lei X, Dong X, Ma R, Wang W, Xiao X, Tian Z, et al. Activation and evasion of type I interferon
responses by SARS-CoV-2. Nat Commun. 2020; 11: 3810. https://doi.org/10.1038/s41467-020-17665-
9PMID: 32733001
55. Sadegh S, Matschinske J, Blumenthal DB, Galindez G, Kacprowski T, List M, et al. Exploring the
SARS-CoV-2 virus-host-drug interactome for drug repurposing. Nat Commun. 2020; 11: 3518. https://
doi.org/10.1038/s41467-020-17189-2 PMID: 32665542
56. Wang S, Li W, Hui H, Tiwari SK, Zhang Q, Croker BA, et al. Cholesterol 25-Hydroxylase inhibits SARS-
CoV-2 and other coronaviruses by depleting membrane cholesterol. EMBO J. 2020; 39: e106057.
https://doi.org/10.15252/embj.2020106057 PMID: 32944968
57. Sanders DW, Jumper CC, Ackerman PJ, Bracha D, Donlic A, Kim H, et al. SARS-CoV-2 requires cho-
lesterol for viral entry and pathological syncytia formation. Elife. 2021;10. https://doi.org/10.7554/eLife.
65962 PMID: 33890572
58. O’Neill LAJ. How Low Cholesterol Is Good for Anti-viral Immunity. Cell. 2015. pp. 1572–1574. https://
doi.org/10.1016/j.cell.2015.12.004 PMID: 26687349
59. Blanc M, Hsieh WY, Robertson KA, Watterson S, Shui G, Lacaze P, et al. Host defense against viral
infection involves interferon mediated down-regulation of sterol biosynthesis. PLoS Biol. 2011; 9:
e1000598. https://doi.org/10.1371/journal.pbio.1000598 PMID: 21408089
60. Yuan S, Chu H, Chan JF-W, Ye Z-W, Wen L, Yan B, et al. SREBP-dependent lipidomic reprogramming
as a broad-spectrum antiviral target. Nat Commun. 2019; 10: 120. https://doi.org/10.1038/s41467-018-
08015-x PMID: 30631056
61. Wang R, Simoneau CR, Kulsuptrakul J, Bouhaddou M, Travisano KA, Hayashi JM, et al. Genetic
Screens Identify Host Factors for SARS-CoV-2 and Common Cold Coronaviruses. Cell. 2021; 184:
106–119.e14. https://doi.org/10.1016/j.cell.2020.12.004 PMID: 33333024
62. Daniloski Z, Jordan TX, Wessels H-H, Hoagland DA, Kasela S, Legut M, et al. Identification of Required
Host Factors for SARS-CoV-2 Infection in Human Cells. Cell. 2021; 184: 92–105.e16. https://doi.org/
10.1016/j.cell.2020.10.030 PMID: 33147445
63. Dugourd A, Saez-Rodriguez J. Footprint-based functional analysis of multi-omic data. Current Opinion
in Systems Biology. 2019. Available: https://www.sciencedirect.com/science/article/pii/
S2452310019300149
64. Szalai B, Saez-Rodriguez J. Why do pathway methods work better than they should? FEBS Lett. 2020;
594: 4189–4200. https://doi.org/10.1002/1873-3468.14011 PMID: 33270910
65. Tummino TA, Rezelj VV, Fischer B, Fischer A, O’Meara MJ, Monel B, et al. Drug-induced phospholipi-
dosis confounds drug repurposing for SARS-CoV-2. Science. 2021; 373: 541–547. https://doi.org/10.
1126/science.abi4708 PMID: 34326236
66. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, et al. limma powers differential expression analy-
ses for RNA-sequencing and microarray studies. Nucleic Acids Res. 2015; 43: e47. https://doi.org/10.
1093/nar/gkv007 PMID: 25605792
67. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data
with DESeq2. Genome Biol. 2014; 15: 550. https://doi.org/10.1186/s13059-014-0550-8 PMID:
25516281
68. Alvarez MJ, Shen Y, Giorgi FM, Lachmann A, Ding BB, Ye BH, et al. Functional characterization of
somatic mutations in cancer using network-based inference of protein activity. Nat Genet. 2016; 48:
838–847. https://doi.org/10.1038/ng.3593 PMID: 27322546
69. Tu¨rei D, Valdeolivas A, Gul L, Palacio-Escat N, Klein M, Ivanova O, et al. Integrated intra- and intercellu-
lar signaling knowledge for multicellular omics analysis. Mol Syst Biol. 2021; 17: e9923. https://doi.org/
10.15252/msb.20209923 PMID: 33749993
70. Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, et al. Scikit-learn: Machine
Learning in Python. J Mach Learn Res. 2011; 12: 2825–2830.
71. Sohn M, Korzeniowski M, Zewe JP, Wills RC, Hammond GRV, Humpolickova J, et al. PI(4,5)P2 con-
trols plasma membrane PI4P and PS levels via ORP5/8 recruitment to ER-PM contact sites. J Cell Biol.
2018; 217: 1797–1813. https://doi.org/10.1083/jcb.201710095 PMID: 29472386
72. Virtanen P, Gommers R, Oliphant TE, Haberland M, Reddy T, Cournapeau D, et al. SciPy 1.0: funda-
mental algorithms for scientific computing in Python. Nat Methods. 2020; 17: 261–272. https://doi.org/
10.1038/s41592-019-0686-2 PMID: 32015543
73. Galons H, Miocque M, Combet-Farnoux C, Bensaid Y, Decodts G, Bram G. A Convenient Procedure
for the Synthesis of Phenothiazine Drugs. Chem Pharm Bull. 1985; 33: 5108–5109.
PLOS COMPUTATIONAL BIOLOGY
SARS-CoV-2 computational drug repurposing - role of cholesterol
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1010021 April 11, 2022 20 / 20
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Understanding the mechanism of SARS-CoV-2 infection and identifying potential therapeutics are global imperatives. Using a quantitative systems pharmacology approach, we identified a set of repurposable and investigational drugs as potential therapeutics against COVID-19. These were deduced from the gene expression signature of SARS-CoV-2-infected A549 cells screened against Connectivity Map and prioritized by network proximity analysis with respect to disease modules in the viral-host interactome. We also identified immuno-modulating compounds aiming at suppressing hyperinflammatory responses in severe COVID-19 patients, based on the transcriptome of ACE2-overexpressing A549 cells. Experiments with Vero-E6 cells infected by SARS-CoV-2, as well as independent syncytia formation assays for probing ACE2/SARS-CoV-2 spike protein-mediated cell fusion using HEK293T and Calu-3 cells, showed that several predicted compounds had inhibitory activities. Among them, salmeterol, rottlerin, and mTOR inhibitors exhibited antiviral activities in Vero-E6 cells; imipramine, linsitinib, hexylresorcinol, ezetimibe, and brompheniramine impaired viral entry. These novel findings provide new paths for broadening the repertoire of compounds pursued as therapeutics against COVID-19.
Article
Full-text available
A novel variant of concern (VOC) named CAL.20C (B.1.427/B.1.429), originally detected in California, carries spike glycoprotein mutations S13I in the signal peptide, W152C in the N-terminal domain (NTD), and L452R in the receptor-binding domain (RBD). Plasma from individuals vaccinated with a Wuhan-1 isolate-based mRNA vaccine or convalescent individuals exhibited neutralizing titers, which were reduced 2-3.5 fold against the B.1.427/B.1.429 variant relative to wildtype pseudoviruses. The L452R mutation reduced neutralizing activity of 14 out of 34 RBD-specific monoclonal antibodies (mAbs). The S13I and W152C mutations resulted in total loss of neutralization for 10 out of 10 NTD-specific mAbs since the NTD antigenic supersite was remodeled by a shift of the signal peptide cleavage site and formation of a new disulphide bond, as revealed by mass spectrometry and structural studies.
Article
Full-text available
Repurposing drugs as treatments for COVID-19 has drawn much attention. Beginning with sigma receptor ligands, and expanding to other drugs from screening in the field, we became concerned that phospholipidosis was a shared mechanism underlying the antiviral activity of many repurposed drugs. For all of the 23 cationic amphiphilic drugs tested, including hydroxychloroquine, azithromycin, amiodarone, and four others already in clinical trials, phospholipidosis was monotonically correlated with antiviral efficacy. Conversely, drugs active against the same targets that did not induce phospholipidosis were not antiviral. Phospholipidosis depends on the physicochemical properties of drugs, and does not reflect specific target-based activities, rather it may be considered a toxic confound in early drug discovery. Early detection of phospholipidosis could eliminate these artifacts, enabling a focus on molecules with therapeutic potential.
Article
Full-text available
Many enveloped viruses induce multinucleated cells (syncytia), reflective of membrane fusion events caused by the same machinery that underlies viral entry. These syncytia are thought to facilitate replication and evasion of the host immune response. Here, we report that co-culture of human cells expressing the receptor ACE2 with cells expressing SARS-CoV-2 spike, results in synapse-like intercellular contacts that initiate cell-cell fusion, producing syncytia resembling those we identify in lungs of COVID-19 patients. To assess the mechanism of spike/ACE2-driven membrane fusion, we developed a microscopy-based, cell-cell fusion assay to screen ~6000 drugs and >30 spike variants. Together with quantitative cell biology approaches, the screen reveals an essential role for biophysical aspects of the membrane, particularly cholesterol-rich regions, in spike-mediated fusion, which extends to replication-competent SARS-CoV-2 isolates. Our findings potentially provide a molecular basis for positive outcomes reported in COVID-19 patients taking statins, and suggest new strategies for therapeutics targeting the membrane of SARS-CoV-2 and other fusogenic viruses.
Article
Full-text available
Molecular knowledge of biological processes is a cornerstone in omics data analysis. Applied to single-cell data, such analyses provide mechanistic insights into individual cells and their interactions. However, knowledge of intercellular communication is scarce, scattered across resources, and not linked to intracellular processes. To address this gap, we combined over 100 resources covering interactions and roles of proteins in inter- and intracellular signaling, as well as transcriptional and post-transcriptional regulation. We added protein complex information and annotations on function, localization, and role in diseases for each protein. The resource is available for human, and via homology translation for mouse and rat. The data are accessible via OmniPath's web service (https://omnipathdb.org/), a Cytoscape plug-in, and packages in R/Bioconductor and Python, providing access options for computational and experimental scientists. We created workflows with tutorials to facilitate the analysis of cell-cell interactions and affected downstream intracellular signaling processes. OmniPath provides a single access point to knowledge spanning intra- and intercellular processes for data analysis, as we demonstrate in applications studying SARS-CoV-2 infection and ulcerative colitis.
Article
Full-text available
Hyper-induction of pro-inflammatory cytokines, also known as a cytokine storm or cytokine release syndrome (CRS), is one of the key aspects of the currently ongoing SARS-CoV-2 pandemic. This process occurs when a large number of innate and adaptive immune cells activate and start producing pro-inflammatory cytokines, establishing an exacerbated feedback loop of inflammation. It is one of the factors contributing to the mortality observed with coronavirus 2019 (COVID-19) for a subgroup of patients. CRS is not unique to the SARS-CoV-2 infection; it was prevalent in most of the major human coronavirus and influenza A subtype outbreaks of the past two decades (H5N1, SARS-CoV, MERS-CoV, and H7N9). With a comprehensive literature search, we collected changing the cytokine levels from patients upon infection with the viral pathogens mentioned above. We analyzed published patient data to highlight the conserved and unique cytokine responses caused by these viruses. Our curation indicates that the cytokine response induced by SARS-CoV-2 is different compared to other CRS-causing respiratory viruses, as SARS-CoV-2 does not always induce specific cytokines like other coronaviruses or influenza do, such as IL-2, IL-10, IL-4, or IL-5. Comparing the collated cytokine responses caused by the analyzed viruses highlights a SARS-CoV-2-specific dysregulation of the type-I interferon (IFN) response and its downstream cytokine signatures. The map of responses gathered in this study could help specialists identify interventions that alleviate CRS in different diseases and evaluate whether they could be used in the COVID-19 cases.
Article
Full-text available
SARS-CoV-2 is a novel coronavirus responsible for the COVID-19 pandemic, in which acute respiratory infections are associated with high socio-economic burden. We applied high-content screening to a well-defined collection of 5632 compounds including 3488 that have undergone previous clinical investigations across 600 indications. The compounds were screened by microscopy for their ability to inhibit SARS-CoV-2 cytopathicity in the human epithelial colorectal adenocarcinoma cell line, Caco-2. The primary screen identified 258 hits that inhibited cytopathicity by more than 75%, most of which were not previously known to be active against SARS-CoV-2 in vitro . These compounds were tested in an eight-point dose response screen using the same image-based cytopathicity readout. For the 67 most active molecules, cytotoxicity data were generated to confirm activity against SARS-CoV-2. We verified the ability of known inhibitors camostat, nafamostat, lopinavir, mefloquine, papaverine and cetylpyridinium to reduce the cytopathic effects of SARS-CoV-2, providing confidence in the validity of the assay. The high-content screening data are suitable for reanalysis across numerous drug classes and indications and may yield additional insights into SARS-CoV-2 mechanisms and potential therapeutic strategies.
Article
Full-text available
Detailed knowledge of the molecular biology of SARS-CoV-2 infection is crucial for understanding of viral replication, host responses and disease progression. Here, we report gene expression profiles of three SARS-CoV and SARS-CoV-2 infected human cell lines. SARS-CoV-2 elicited an approximately two-fold higher stimulation of the innate immune response compared to SARS-CoV in the human epithelial cell line Calu-3, including induction of miRNA-155. Single-cell RNA sequencing of infected cells showed that genes induced by virus infections were broadly upregulated, whereas interferon beta/lambda genes an pro-inflammatory cytokines such as IL-6 were expressed only in small subsets of infected cells. Temporal analysis suggested that transcriptional activities of interferon regulatory factors precede those of nuclear factor κB. Lastly, we identified heat shock protein 90 (HSP90) as a protein relevant for the infection. Inhibition of the HSP90 activity resulted in a reduction of viral replication and pro-inflammatory cytokine expression in primary human airway epithelial cells.
Article
Full-text available
Influenza A viruses, especially H3N2 and H1N1 subtypes, are viruses that often spread among humans and cause influenza pandemic. There have been several big influenza pandemics that have caused millions of human deaths in history, and the threat of influenza viruses to public health is still serious nowadays due to the frequent antigenic drift and antigenic shift events. However, only few effective anti-flu drugs have been developed to date. The high development cost, long research and development time, and drug side effects are the major bottlenecks, which could be relieved by drug repositioning. In this study, we proposed a novel antiviral Drug Repositioning method based on minimizing Matrix Nuclear Norm (DRMNN). Specifically, a virus-drug correlation database consisting of 34 viruses and 205 antiviral drugs was first curated from public databases and published literature. Together with drug similarity on chemical structure and virus sequence similarity, we formulated the drug repositioning problem as a low-rank matrix completion problem, which was solved by minimizing the nuclear norm of a matrix with a few regularization terms. DRMNN was compared with three recent association prediction algorithms. The AUC of DRMNN in the global fivefold cross-validation (fivefold CV) is 0.8661, and the AUC in the local leave-one-out cross-validation (LOOCV) is 0.6929. Experiments have shown that DRMNN is better than other algorithms in predicting which drugs are effective against influenza A virus. With H3N2 as an example, 10 drugs most likely to be effective against H3N2 viruses were listed, among which six drugs were reported, in other literature, to have some effect on the viruses. The protein docking experiments between the chemical structure of the prioritized drugs and viral hemagglutinin protein also provided evidence for the potential of the predicted drugs for the treatment of influenza.
Article
The COVID-19 pandemic has highlighted the need to quickly and reliably prioritize clinically approved compounds for their potential effectiveness for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infections. Here, we deployed algorithms relying on artificial intelligence, network diffusion, and network proximity, tasking each of them to rank 6,340 drugs for their expected efficacy against SARS-CoV-2. To test the predictions, we used as ground truth 918 drugs experimentally screened in VeroE6 cells, as well as the list of drugs in clinical trials that capture the medical community's assessment of drugs with potential COVID-19 efficacy. We find that no single predictive algorithm offers consistently reliable outcomes across all datasets and metrics. This outcome prompted us to develop a multimodal technology that fuses the predictions of all algorithms, finding that a consensus among the different predictive methods consistently exceeds the performance of the best individual pipelines. We screened in human cells the top-ranked drugs, obtaining a 62% success rate, in contrast to the 0.8% hit rate of nonguided screenings. Of the six drugs that reduced viral infection, four could be directly repurposed to treat COVID-19, proposing novel treatments for COVID-19. We also found that 76 of the 77 drugs that successfully reduced viral infection do not bind the proteins targeted by SARS-CoV-2, indicating that these network drugs rely on network-based mechanisms that cannot be identified using docking-based strategies. These advances offer a methodological pathway to identify repurposable drugs for future pathogens and neglected diseases underserved by the costs and extended timeline of de novo drug development.