Abstract
Background
Higher mortality of COVID-19 patients with lung disease is a formidable
challenge for the health care system. Genetic association between
COVID-19 and various lung disorders must be understood to comprehend
the molecular basis of comorbidity and accelerate drug development.
Methods
Lungs tissue-specific neighborhood network of human targets of
SARS-CoV-2 was constructed. This network was integrated with lung
diseases to build a disease–gene and disease-disease association
network. Network-based toolset was used to identify the overlapping
disease modules and drug targets. The functional protein modules were
identified using community detection algorithms and biological
processes, and pathway enrichment analysis.
Results
In total, 141 lung diseases were linked to a neighborhood network of
SARS-CoV-2 targets, and 59 lung diseases were found to be topologically
overlapped with the COVID-19 module. Topological overlap with various
lung disorders allows repurposing of drugs used for these disorders to
hit the closely associated COVID-19 module. Further analysis showed
that functional protein–protein interaction modules in the lungs,
substantially hijacked by SARS-CoV-2, are connected to several lung
disorders. FDA-approved targets in the hijacked protein modules were
identified and that can be hit by exiting drugs to rescue these modules
from virus possession.
Conclusion
Lung diseases are clustered with COVID-19 in the same network vicinity,
indicating the potential threat for patients with respiratory diseases
after SARS-CoV-2 infection. Pathobiological similarities between lung
diseases and COVID-19 and clinical evidence suggest that shared
molecular features are the probable reason for comorbidity.
Network-based drug repurposing approaches can be applied to improve the
clinical conditions of COVID-19 patients.
Supplementary Information
The online version contains supplementary material available at
10.1186/s12920-021-01079-7.
Keywords: COVID-19, SARS-CoV-2, Lung disease, Comorbidity, Disease
network
Background
The novel coronavirus disease 2019 (COVID-19) cases, caused by
SARS-CoV-2, crossed 189,000,000 globally as of July 16, 2021. Data show
that the most affected groups had two or more pre-existing medical
conditions such as hypertension, diabetes, and metabolic,
cardiovascular, and digestive disorders [[25]1–[26]3]. Moreover,
comorbidity (or existence of multiple disorders) in COVID-19 patients
is associated with a higher risk of severe illness, poor prognosis, and
high mortality [[27]4]. During viral infection, a virus hijacks the
host cell machinery for its replication. Virus–host interactions
perturb highly organized host cellular networks and reconstruct
different networks favouring virus replication. The topology of
molecular interactions is altered in a disease. Hence, the interaction
of SARS-CoV-2 with healthy human cells is different from that with
disease cells, which thus leads to various impacts on humans after
SARS-CoV-2infection. Human diseases are connected via defects in common
genes [[28]5, [29]6]. Moreover, the similarity in disease phenotypes
often indicates underlying genetic connections. Therefore, pre-existing
medical conditions can facilitate the development of another disease if
they share the same or functionally related genes [[30]7, [31]8].
SARS-CoV-2 has been associated with respiratory tract infections, and
in some cases, it severely damages lungs in adult patients. Here, we
investigated the underlying molecular link between COVID-19 and lung
diseases to understand the basis of comorbidity. In the present study,
we have considered a disease in the lung or symptoms in the lung or
diseases in other tissues or organs affecting the lungs as a "lung
disease." Gordon et al. [[32]9] recently identified 26 of the 29
SARS-CoV-2 proteins that bind to 332 human proteins and hijack the host
translational machinery. Here, we constructed a tissue (lungs)-specific
neighborhood network of the 332 human targets of SARS-CoV-2. This
network was integrated with lung diseases to build a disease–gene
network of the lung. Subsequently, we constructed a lung disease
network, which also includes COVID-19. In total, 141lung diseases were
found to be associated with COVID-19. Among them, 49 were directly
linked to COVID-19, apparently justifying the characteristics of a
complex disorder. Further, we observed that 59 lung diseases
topologically overlapped with COVID-19, indicating a higher risk of
comorbidity. This observation also presents the opportunity to
repurpose drugs used to treat lung diseases because these drugs can
simultaneously hit a lung disease and closely associated COVID-19
module. Moreover, we observed that genes in overlapping lung diseases
and COVID-19 are coexpressed and involved in a similar molecular
function and biological processes, representing pathobiological
similarities between various lung disorders and COVID-19.
Next, we identified functional protein modules that are maximally
perturbed by SARS-CoV-2 and involved in RNA processing, export, and
protein synthesis machinery of the cell. Moreover, these modules are
associated with various lung disorders, indicating the hotspots for
comorbidity. Hence, we employed a network-based proximity approach
[[33]10] and explored the DrugBank database [[34]11] to identified
approved targets in these protein modules that can be hit by existing
drugs and rescued from virus possession. Studies have reported that a
network-based toolset can be effectively used to identify drugs for
COVID-19 treatment [[35]12, [36]13].We identified 56 druggable human
proteins in proximity to the COVID-19 disease module and found that
these proteins can be targeted by FDA-approved or investigational
drugs. SARS-CoV-2 has a very high mutation rate, which allows it to
develop drug resistance [[37]14]. Therefore, identifying and targeting
host factors, rather than targeting viral proteins, will be an enduring
approach. In summary, this work presents the risk of different lung
disorders at COVID-19 onset and drug repurposing opportunities to treat
patients with lung disorders.
Materials and methods
Construction of a lung-specific PPI network of SARS-CoV-2 targets
Human lung tissue-specific interactome data were retrieved from the
TissueNet v.2 database. TissueNet v.2 synergizes between large-scale
data of human PPIs and tissue-specific expression profiles to generate
tissue-specific PPIs. This database also consolidates PPI data from
four major databases, BioGrid, IntAct, MINT, and DIP, and integrates
resulting PPIs with RNA-sequencing profiles of the Genotype-Tissue
Expression consortium (GTEx). We downloaded 168,296 lung-specific
interactions from TissueNet v.2 to construct a SARS-CoV-2 target
interactome. Next, we obtained a list of 332 human proteins targeted by
SARS-CoV-2, which were identified through affinity-purification mass
spectrometry [[38]9]. Using these 332 proteins, we built a subnetwork
from 168,296 lung-specific interactions, SARS-CoV-2 target network
(STN). Nine SARS-CoV-2 targets (AATF, CEP43, CISD3, MTARC1, NUP62,
SRP19, THTPA, TIMM10B, and TRIM59) showed no interaction in the lung.
Construction of a lung-specific disease–gene and disease–disease network
The disease-gene association data in the lungs were retrieved from the
Gene ORGANizer [[39]15], which is a phenotype-based curated database
that links human genes to the body parts they affect. Phenotypes
classified by Human Phenotype Ontology (HPO) were considered with
certain modifications. After disease-gene association data were
pre-processed, disease–gene pairs that were not included but matched
with the HPO phenotype were manually added. Aspirin-induced asthma and
asthma were both considered as asthma. Pulmonary emphysema,
sarcoidosis, and silicosis and their associated genes were also added
to the list. Finally, 6040 disease–gene pairs and 184 various lung
diseases were listed. If a gene is associated with a known lung
disorder, then the gene and lung disorder were connected via links.
Subsequently, nodes in the STN were linked to the lung disorder to
construct the disease–gene association map of the network. Of the 5050
nodes of the STN, 618 were linked to 145 lung diseases. Of the 618
genes, 36 were the direct targets of SARS-CoV-2 and were connected to
COVID-19 as a new disease–gene pair. Finally, a lung disease–gene
network (LDGN) consisting of 1815 disease–gene pairs, including that
for COVID-19, was constructed. The disease–disease association network
(DDAN) was derived from the lung disease–gene association network; two
diseases were connected if they shared one common gene. The disgenet2r
package [[40]16] was used to study the association between disease
classes and functional protein modules.
Network-based separation measure between diseases
To identify the overlapping disease modules, a "separation" measure,
S[ab] was calculated between COVID-19 (a) and lung disease (b) using
the following formula:
[MATH: Sab=<dab>
-<daa>
-<dbb>
2 :MATH]
[MATH: Sab :MATH]
compares the shortest distances between proteins connected to each
disease,
[MATH: <daa>
:MATH]
and
[MATH: <dbb>
:MATH]
, to those
[MATH: <dab>
:MATH]
between a–b protein pairs. Positive
[MATH: Sab :MATH]
shows that the two disease modules are separated on the lung
interactome, whereas a negative value indicates overlapping modules.
The statistical significance of module overlap between COVID-19 and the
lung disease was evaluated using the full randomization model. The same
number of proteins associated with two diseases was randomly sampled
1000 times, and the corresponding
[MATH: Sabran :MATH]
between the two gene sets was calculated. Next, z-score was calculated
as follows:
[MATH: z-score=S<
mrow>ab-mσ :MATH]
where m and σ indicates the mean value and standard deviation of 1000
[MATH: Sabran :MATH]
. Here, z-score < 0 indicates that the two diseases are closely
overlapped than expected by chance [[41]17].
Community detection
We applied fast-greedy, walktrap, louvain, leading eigenvector, and
spinglass on the STN as an undirected, unweighted network. These
community detection algorithms segregate the nodes into higher-density
modules and optimize an objective function, that is, modularity.
Communities separated by spinglass were selected for subsequent
analysis based on the modularity score and community size. Spinglass
uses a random number generator to find the communities. Therefore, we
ran Spinglass 10 times with different seed values. We compared the rand
statistics between each run, and results showed that the structures of
these communities are highly similar (> 0.7) [[42]18, [43]19].
Network-based proximity measure
Network proximity between drug targets (A) and SARS-CoV-2 targets in
the host (B) was measured using the closest method (d[c]).
[MATH:
dc=1‖A‖+‖B‖∑a∈A<
/munder>minb∈Bda,b+
∑b∈B<
mi
mathvariant="italic">minb∈Ad(a,b) :MATH]
where d(a, b) represents the shortest distance between genes a and b in
the lung interactome. The statistical significance of proximity was
evaluated using z-score (z[c]). z[c] was calculated by comparing the
observed distance to a reference distance distribution. To compute
reference distance distribution, the sets of proteins of size and
degree similar to those of the drug targets and disease proteins were
randomly selected for 1000 times from the lung interactome. The mean
and standard deviation of distance distribution was calculated to
compute z[c] [[44]10, [45]20].
Process and pathway enrichment analysis and gene ontology semantic similarity
Pathway and process enrichment analysis was performed using Metascape
[[46]21]. Gene ontology biological processes, KEGG Pathway, and
Reactome were used as ontology sources. GO semantic similarity between
genes was measured using Wang et al. [[47]22] method with the GOSemSim
package in R. Considering that two genes G1 and G2 are annotated by the
GO term sets GO1 = [go11, go12, …, go1m] and GO2 = [go21, go22,
…, go2n], respectively, their semantic similarity score, which is
determined using Wang's method, is defined as follows:
[MATH: SimG1,G2=<
msub>∑1≤i≤mSimgo1i,
GO2+∑1≤j≤nSimgo2j,
GO1m+n.
:MATH]
Correlation analysis
GTEx gene expression datasets of healthy human lung tissues were
downloaded from the UCSC Xena project [[48]23]. log[2](RSEM + 1) (RSEM:
RNA-Seq by Expectation Maximization) transformed gene expression data
(n = 288) were retrieved, and the Pearson correlation coefficient was
computed to measure coexpression levels using the Hmisc package in R.
Computation of topological parameters
The largest connected component (LCC), dyadicity, and Jaccard
similarity coefficient were measured using the igraph package in R.
Dyadicity (D) is the number of same label edges divided by the expected
number of same label edges, and D > 1 indicates higher connectedness
between the nodes with the same label. The Jaccard similarity
coefficient of two nodes was calculated as the number of common
neighbors divided by the number of nodes that are neighbors of at least
one of the two nodes.
Tools for data analysis, plotting, and statistical analysis
R packages tidyverse and stringr were used for data analysis, and
graphs were plotted using ggplot2. Networks were visualized using
Gephi. Statistical significance between the groups was analyzed using
the non-parametric Mann–Whitney test in R, and that of the overlap
between gene lists was analyzed using Fisher's exact test.
Results
Construction of SARS-CoV-2–host interactome in the lung
To depict the SARS-CoV-2–host interaction network, the protein–protein
interaction (PPI) network of the lungs (lung interactome) was obtained
from the TissueNet v.2 database [[49]24]. We referred to Gordon et al.
[[50]9] for the list of 332 human targets of SARS-CoV-2 and constructed
a subnetwork of these proteins from the PPI network. Of the 332 viral
targets, 323 proteins were present in the subnetwork. The resulting
subnetwork, named as the SARS-CoV-2 target network (STN), has 5050
nodes and 11,256 pairwise interactions (Fig. [51]1a, Additional file
[52]2: Table S1). Next, 181 of the 323 viral targets form the LCC
within the lung interactome. To determine the statistical significance
of the LCC, we randomly selected proteins with a matching degree and
calculated the size of the LCC. We repeated the random selection 1000
times and found that the size of the random LCC was 136.28 ± 16.05
(Fig. [53]1b), and z-score = 2.78 (p-value = 5.36 × 10^−3), indicating
that the SARS-CoV-2–host interaction network did not appear by chance
and the target proteins were located in the same network vicinity
[[54]10, [55]13]. To confirm this result, we computed dyadicity (D) (a
measure of the connectedness of the nodes with the same label, see
Materials and Methods) among the SARS-CoV-2 targets in the STN to
determine if they share more or fewer edges than expected in a random
configuration of the network. We found D = 7.664, indicating high
connectedness among SARS-CoV-2 targets. D > 1 signifies that the
SARS-CoV-2 targets form a community-like structure to hijack the host
cellular machinery. If implicated in diseases, proteins in a community
confer a higher chance of comorbidity than those not in the community
because proteins in a community frequently interact, coexpress, and are
functionally interconnected [[56]25]. Therefore, to understand the link
between COVID-19 and other lung diseases, we constructed and analyzed
the disease–gene and disease–disease association map linked to the STN.
Fig. 1.
[57]Fig. 1
[58]Open in a new tab
a Neighbourhood interaction network of SARS-CoV-2 targets (STN) in the
lung. The size of the node is proportional to its degree. b SARS-CoV-2
targets form a LCC of size 181 in the lung interactome. The size of the
LCC is significantly larger than the random expectation
Disease–gene and disease–disease association map of COVID-19 in lungs
To construct a disease association map of the STN, we obtained the
disease-gene association data from the ORGANizer database [[59]15]. In
total, 184 lung diseases, 1957 genes, and 6040 disease–gene pairs were
considered for further analysis (see Materials and Methods) (Additional
file [60]3: Table S2). However, 1442 of the 1957 genes are present in
the lung interactome. To create the disease–gene association map, we
screened the diseases associated with proteins (nodes) in the STN. A
disease and gene are then connected if the gene is associated with the
lung disorder. We observed that 618 proteins consisting of 36
SARS-CoV-2 targets were linked to 146 disorders, which includes
COVID-19 (Additional file [61]4: Table S3). The overlap between
SARS-CoV-2 targets and 1442 lung disease-associated genes was not
statistically significant (Fisher's exact test, value = 0.454). Gysi et
al. [[62]13] reported a similar observation with a group of genes
involved in various disease classes. However, the overlap between 5050
nodes in the STN and lung disease-associated genes was statistically
significant (Fisher's exact test, p-value = 2.93 × 10^−5).
Figure [63]2a shows the resulting disease–gene association map of the
STN, named as the lung disease–gene network (LDGN), consisting of1814
disease–gene pairs.
Fig. 2.
[64]Fig. 2
[65]Open in a new tab
Disease-gene association network. a Lung disease-gene network (LDGN),
including COVID19 (yellow node). The network shows the SARS-CoV-2
targets (red) and neighborhood genes (green). b, c Dot plot shows the
highly connected diseases (k > 20) and genes in LDGN, respectively
The LCC within the LDGN consists of 141 lung diseases and 610 genes,
indicating that many of the disorders share a common genotype. For
example, the SARS-CoV-2 targets, FBN1 (degree, k = 15), FBLN5 (k = 11),
and COMT (k = 9), and neighborhood nodes, OFD1 (k = 19), DNAAF2
(k = 16), and DNAAF5 (k = 16), are linked to multiple disorders
(Fig. [66]2b). Similarly, a disorder in the LDGN is also connected with
multiple genes [e.g., ventricular septal defect (k = 142), respiratory
insufficiency (k = 133), congestive heart failure (k = 95), apnea
(k = 63), and hypothyroidism (k = 60) (Fig. [67]2c, Additional file
[68]1: Fig. S1 and Fig. S2)].
The disease-gene association pattern in the LDGN indicates the presence
of a molecular connection between COVID-19 and a wide range of lung
disorders. To comprehend this connection, a disease-disease association
network, DDAN was constructed, where two diseases were linked if they
share one associated gene (Fig. [69]3a). DDAN consists of 141 diseases
(nodes) and 1326 links, indicating a higher clustering between
diseases. Further, the degree distribution of the DDAN did not follow
the scale-free property (Fig. [70]3b). To determine the exact
topological nature, we measured network transitivity (
[MATH: TDDAN :MATH]
=0.4264) and average path length (
[MATH: LDDAN :MATH]
=2.0585) of the DDAN and compared them with the equivalent 1000
Erdős − Rényi random graphs. The results showed that the average path
length is significantly lower (p-value < 0.0001), whereas transitivity
is significantly higher (p-value < 0.0001) than random graphs (
[MATH: Lrandom=2.44680 :MATH]
and
[MATH: Trandom=0 :MATH]
.0668) (Fig. [71]3c, d). Further, we calculated the small-worldness
scalar (S) for the DDAN as follows:
[MATH: γ=TDDANTrandom=6.3
83 :MATH]
[MATH: λ=LDDANLrandom=0.8
41 :MATH]
[MATH:
S=γλ=
7.589 :MATH]
Fig. 3.
[72]Fig. 3
[73]Open in a new tab
Disease-disease association network (DDAN). a DDAN, including COVID19,
red nodes represent the diseases that are directly direct linked to
COVID19. b Scatter plot shows the degree distribution of DDAN, which
does not follow the scale-free property. c The average path length
between the diseases in DDNA and distribution of average path length of
1000 random networks (green). d Transitivity of DDNA and distribution
of transitivity of 1000 random networks (pink)
A network is considered a small-world network if S > 1 [[74]26]. Hence,
the topology of the DDAN represents a small-world property, indicating
that any two diseases in this network have a high tendency to be
interconnected and may cause the overlapping disease pathogenesis.
Forty-nine diseases in the DDAN were directly connected to COVID-19.
Using the number of common genes, the Jaccard similarity coefficient
was computed to identify the extent of molecular overlap between the 49
lung diseases and COVID-19 (Additional file [75]1: Fig. S3). Several
diseases, such as respiratory insufficiency, congestive heart failure,
respiratory failure, ventricular septal defect, mitral regurgitation,
and hyperthyroidism, are closely associated with COVID-19. Of note,
although molecular connections exist between COVID-19 and various lung
diseases, these overlaps are not statistically significant (considering
only SARS-CoV-2 targets). Nevertheless, these molecular connections are
crucial for analyzing the effect of SARS-CoV-2 infection on lung
patients; however, opportunities to comprehend disease comorbidity are
limited with these molecular connections.
Topological overlap between disease modules, pathobiological similarities,
and opportunities for drug repurposing
For a greater understanding of comorbidity, we measured the
network-based separation between two disease modules to comprehend
their degree of overlap. The network-based separation measure is
primarily advantageous because it can predict disease–disease
association, even if two diseases share no genes. If two disease
modules overlap, then perturbations to one disease can cause
disturbance to another, indicating that they have similar clinical
characteristics. The magnitude of the overlap indicates the biological
and pathobiological similarities between the two disease modules
[[76]17]. Network-based separation (
[MATH: Sab :MATH]
) (see Materials and Methods) between COVID-19 and all lung diseases
was measured. Of the184 lung diseases, 59 demonstrated overlapping
modules (
[MATH: Sab<0
:MATH]
) with COVID-19 (Additional file [77]5: Table S4). The statistical
significance of
[MATH: Sab :MATH]
for each disease pair, that is, COVID-19 and each lung disease, was
evaluated using a full randomization model. We observed, for all 59
diseases, the z-score was < 0, indicating that these diseases are
closely overlapped with COVID-19 than expected by chance.
Figure [78]4a–j shows the top 10 closely overlapping lung disease
modules with COVID-19 (e.g., hemolytic-uremic syndrome (
[MATH: Sab=-0.214
2 :MATH]
), abnormal respiratory motile cilium morphology (ARMCM;
[MATH: Sab=-0.211
38 :MATH]
), obstructive lung disease (
[MATH: Sab=-0.210
22 :MATH]
), pleural effusion (
[MATH: Sab=-0.182
16 :MATH]
), patent foramen ovale (PFO,
[MATH: Sab=-0.161
9 :MATH]
), and pulmonary insufficiency (
[MATH: Sab=-0.156
94 :MATH]
). Thus, patients with these disorders are probably more vulnerable to
COVID-19 symptoms or vice versa because of overlapping disease modules.
The same set of genes induce ARMCM, absent respiratory ciliary axoneme
radial spokes, and respiratory insufficiency, which are caused because
of defective ciliary clearance; therefore, we considered only ARMCM in
the top 10 list. According to the network-based separation measure,
almost 32% of lung diseases have overlapping modules with COVID-19 and
the remaining 68% are topologically separated. To understand the
biological relationship and pathobiological similarities, the
expression correlation and semantic similarity (molecular functions and
biological processes) of genes involved in COVID-19 and overlapping
lung diseases (
[MATH: Sab<0
:MATH]
).) were measured. Gene coexpression and semantic similarity were
significantly (p-value < 0.0001) higher compared to those in the random
control (Fig. [79]4k–m), indicating the biological and pathobiological
similarities between COVID-19 and overlapping lung diseases. To further
investigate the similarities in clinical features, results from recent
publications were explored. Reports have raised concerns about lung
injuries linked to COVID-19 [[80]27, [81]28]. A higher percentage of
COVID-19 patients in severe conditions are more likely to develop
chronic obstructive pulmonary disease (COPD) and impairment of
diffusion capacity [[82]4, [83]29]. Many lung diseases (
[MATH: Sab<0) :MATH]
(Additional file [84]5: Table S4) with a overlapping module with
COVID-19 are linked to these aforementioned phenotypes. A disease
closely associated with COVID-19, haemolytic-uremic syndrome
[MATH: (Sab<0) :MATH]
(Fig. [85]4a), causes pulmonary hemorrhage, which is linked to kidney
failure [[86]30], and studies have recently reported that chronic
kidney diseases and chronic pulmonary disease cause adverse outcomes in
COVID-19 patients [[87]4, [88]31]. Abnormal respiratory motile cilium
(Fig. [89]4b) or ciliary dyskinesia (
[MATH: Sab=-0.211
38 :MATH]
) causes chronic respiratory tract infections because the improper
movement of mucus restricts the complete elimination of fluid,
bacteria, and particles from the lungs, leading to bronchitis (chronic
bronchitis,
[MATH: Sab=-0.146) :MATH]
([90]www.ghr.nlm.nih.gov). Lack of respiratory clearance in a patient
with ciliary dyskinesia could confer a higher risk of health hazard
after SARS-CoV-2 infection. Another study showed that patients with
obstructive lung disease (
[MATH: Sab=-0.21<
/mn> :MATH]
) (Fig. [91]4c) and pulmonary emphysema (
[MATH: Sab=-0.09) :MATH]
are at a higher risk of pneumothorax after SARS-CoV-2 infection
[[92]32]. Rajendram et al. [[93]33] predicted that PFO (Fig. [94]4e)
may be common in COVID-19 patients because PFO causes pulmonary
embolism [[95]34]. Even the disease module of pulmonary embolism
overlapped (
[MATH: Sab=-0.008) :MATH]
with COVID-19. A clinical study in Wuhan, China [[96]35] reported that
almost 5% COVID-19 patients had pleural effusion(
[MATH: Sab=-0.18) :MATH]
(Fig. [97]4d), which is often caused by congestive heart failure and
blood clots in lung arteries. Importantly, pleural effusion is commonly
associated with age-related respiratory problems and cancer [[98]36].
On the other hand, congestive heart failure, which causes many
lung-related diseases [[99]37], also overlapped with the COVID-19
disease module (Additional file [100]5: Table S4). A meta-analysis by
Alqahtani et al. [[101]38] demonstrated that the risk of more severe
COVID-19 was higher in patients with COPD (risk of severity = 63%) than
in those without (33.4%). Although these results suggest the clinical
similarities between COVID-19 and overlapping lung disorders, they are
limited and cannot be extrapolated for all overlapping lung disorders
without clinical evidence. Furthermore, a genome-wide association study
has presented the genetic susceptibility locus in the chromosome of
patients with COVID-19 and respiratory failure [[102]39], and genes
present in this locus (SLC6A20, LZTFL1, and CCR9) were also associated
with different lung disorders (
[MATH: Sab<0) :MATH]
such as pulmonary fibrosis, respiratory distress, asthma, and nephrotic
syndrome.
Fig. 4.
[103]Fig. 4
[104]Open in a new tab
Network-based separation (
[MATH: Sab :MATH]
) and pathobiological similarities. a–j shows observed
[MATH: Sab :MATH]
, z-score (red arrow) and distribution of
[MATH: Sabran :MATH]
of top 10 overlapping lung diseases with COVID-19 (here, ARMCM
indicates abnormal respiratory motile cilium morphology). k Box plot
represents the pairwise correlation between genes is significantly
(p-value < 0.0001) higher than the random gene sets. l, m Box plots
show the distribution of functional similarities (MF) and GO processes
(BP) between the genes involved in lung disease and COVID-19. The GO
processes and functional similarity between the genes are significantly
high (p < 0.0001) compared to the random gene sets (note, in figures k,
l, and m 1–10 indicates disease in a similar sequence as it is
mentioned in figures a to j). n The strategy of drug repurposing to
target the COVID19 module
The availability of efficient drugs for the treatment of clinically
characterized lung diseases having overlapping neighborhoods with
COVID-19 has shown the scope for repurposing these drugs for
COVID-19treatment. When two diseases are localized in the same network
vicinity and overlap with each other, then targeting one disease can
affect another disease module (Fig. [105]4n), leading to efficient
clinical outcomes for both as they have common network neighborhoods
[[106]20]. Clinical data from the Clinicaltrials.gov database show that
some drugs used for lung diseases, such as methylprednisolone for
tracheal stenosis (
[MATH: SAB=-0.149
15 :MATH]
) and ketamine and budesonide for COPD are in clinical trials for
COVID-19 treatment [[107]40]. Therefore, existing drugs that are
presently used for treating lung disorders on COVID-19 patients must be
tested for better clinical outcomes. Treating a comorbid patient is
challenging, but an accurate clinical picture of patients, molecular
signature of diseases, and drug target information can improve the
present crisis.
Functional protein modules preferentially hijacked by SARS-CoV-2 are linked
to a broad range of lung disorders
Modularity in the network refers to the pattern of connectedness in
which nodes are grouped into highly connected subsets [[108]41]. A key
feature in the PPI network is that tightly connected proteins within a
community are mostly involved in similar biological functions
[[109]42]. Similarly, genes involved in related diseases are highly
connected; moreover, diseases linked to common genes result in the
formation of disease modules and comorbidity [[110]43]. We compared
various community detection algorithms, that is, fast-greedy, walktrap,
louvain, leading eigenvector, and spinglass, to identify protein
modules in the STN [[111]19, [112]44]. Spinglass showed good
partitioning with a higher modularity score compared with other
algorithms (see “Materials and Methods” section and Additional file
[113]6: Table S5). Our findings are in agreement with those of the
study by Rahiminejad et al. [[114]18], where good partitioning of the
functional protein module was observed using spinglass in eukaryotes.
Of the 21 modules, the top 4 protein modules were selected according to
the presence of a large number of SARS-CoV-2 targets (> 20) and a gene
ontology semantic similarity score (> 0.2) of biological processes
(Additional file [115]6: Table S6). Numerous viral targets were
considered because these modules are largely hijacked and strongly
perturbed after infection compared with other functional modules in the
network. The modules were named as modules 1, 2, 3, and 4, and each
module contains 63, 50, 28, and 23 SARS-CoV-2 target proteins,
respectively (Fig. [116]5). The biological process and pathway
enrichment analysis showed that module1, the largest module, is mainly
enriched with RNA metabolism, including transcription, mRNA processing,
transport, mRNA de-adenylation, and surveillance. Presumably,
biological processes linked to module1 are hijacked by SARS-CoV-2 in
the early stage of infection for viral RNA production. Notably, the
components of module1 are linked to 64 disorders, among which the
highly connected are respiratory insufficiency, ventricular septal
defect, respiratory distress, pneumonia, and lung neoplasm
(Fig. [117]5a, 3rd column, Additional file [118]7: Table S7). Most
module 1-associated diseases are directly connected to COVID-19
(Fig. [119]3b, Additional file [120]1: Fig. S2). On the other hand,
hijacking module2 can predominantly affect protein degradation (ERAD
pathway, HRD1 complex, and regulation of the protein catabolic
process), transport, folding, and stability (retrograde protein
transport, regulation of protein stability,
VCP-VIMP-DERL1-DERL2-HRD1-SEL1L complex, regulation of intracellular
transport, regulation of vesicle-mediated transport, and protein
folding in the endoplasmic reticulum). Module3 and module4 involve
several processes, primarily cellular transport, localization,
organization, and cell cycle. Modules 2, 3, and 4 were linked to 79,
60, and 32 different disorders, respectively (Additional file [121]7:
Table S7). The disease association of all four protein modules was
significantly higher (p-value < 0.0001) than 1000 random gene sets.
Moreover, a broad spectrum of disorders of various classes, such as
neoplasms, neurological, and digestive systems, was associated with
these modules (Additional file [122]1: Fig. S4). Gysi et al. [[123]13]
predicted that the manifestation of SARS-CoV-2 in different human
tissues could cause various disorders. Therefore, not only lung-related
disorders, but diseases in other organs can also be a potential threat
for COVID-19 patients. To confirm this observation, gene coexpression
pattern in functional modules was analyzed. Genes in the same
functional module often show a high coexpression profile, which
indicates their involvement in similar biological processes. Therefore,
we calculated Pearson correlation coefficients of gene pairs using gene
expression data of healthy lung tissues from GTEx. The median value of
the positive correlation between the genes in all modules was
significantly higher (p-value < 0.0001) than that for the random gene
set (Fig. [124]5, [125]4th column). Therefore, these modules can be
identified as coexpression modules that share core transcriptional
programs in the lung, which indicates that their perturbation can lead
to a similar disease phenotype. Next, we used drug repurposing to find
the targets to hit functional modules.
Fig. 5.
[126]Fig. 5
[127]Open in a new tab
Community detection in STN and functional protein module. a–d show the
modules 1 to 4, pathway and process enrichment analysis of each module,
their disease associations, and positive correlation between genes in
each module in healthy lung tissue. The pairwise correlation between
genes in each module is significantly (p-value < 2.2 × e^−16) higher
than the random gene sets
Drug repurposing to target functional modules
We propose targeting functional protein modules hijacked by SARS-CoV-2,
by drug repositioning. There are two main reasons to target these
modules. First, the binding of a drug to its target in a module will
prevent virus replication. Second, as a module is linked to several
lung diseases, targeting a module can reduce severity in patients. From
Drug Bank, we identified 56 approved targets (red color nodes in
Fig. [128]6) that can be hit by 144 approved or investigational drugs
in the clinical trial (Additional file [129]8: Table S8) [[130]11]. The
list contains10 approved drugs at different stages of clinical trials
for COVID-19 treatment, including chloroquine targeting Glutathione
S-transferase Mu 1 in module3 (Additional file [131]1: Fig. S5).
However, the efficacy of chloroquine on COVID-19 patients is arguable.
Coagulation factor X (F10) was observed in module2, which has recently
been implicated as a target due to the potential role of coagulopathy
in COVID-19 [[132]45]. To determine the effectiveness of targets, we
applied network-based proximity measures to calculate the proximity
between the COVID-19 disease module (network among SARS-CoV-2 targets)
and FDA-approved targets in the functional module. We used the
"closest" (d[c]) measure, representing the closest path length between
a target and the nearest SARS-CoV-2 target protein. Then, we calculated
z-score (z[c]) to validate the proximity by comparing the observed
target–disease protein distance to the random expectation [[133]10].
All 56 targets were proximal (z[c] < − 2) to the COVID-19 disease
module (Additional file [134]9: Table S9) compared to the random
expectation. Next, approved drugs in clinical trials for COVID-19
treatment (Additional file [135]1: Fig. S5) were also significantly
closer to the COVID-19 disease module (z[c] < − 2) (Additional file
[136]10: Table S10). The highly connected (hub) targets in the
functional module, such asNTRK1 (k = 43) and IMPDH2 (k = 37) in
module1, as well as PLAT (k = 17) and COMT (k = 10) in module2, can be
considered as potential targets for COVID-19 treatment (Additional file
[137]1: Fig. S5). Considering the complexity of COVID-19, different
locations in the STN must be targeted as this may help efficiently
rewire the cellular network [[138]46] and rescue these functional
modules from the virus, thereby reducing virus growth [[139]20]. Many
of the target proteins suggested do not directly interact with
SARS-CoV-2; instead, they are neighborhood nodes. The binding of drugs
to these targets present in the same network vicinity may efficiently
perturb the network modules, including viral growth [[140]47]. Notably,
this article presents the computation analysis; therefore all
drug-target combinations should be tested on SARS-CoV-2-infected cell
lines and validated through clinical trials.
Fig. 6.
[141]Fig. 6
[142]Open in a new tab
Targetable protein in functional modules: The red nodes in each module
indicate the FDA-approved targets
Discussion
Currently, a speedy drug discovery is urgently required to stop the
infection and rapid transmission of SARS-CoV-2. Aged COVID-19 patients
with comorbidity are at severe health risks worldwide. The present
study evidenced the risk of COVID-19 at the onset of various
lung-related disorders and the molecular basis of comorbidity by
applying the principle of network biology. COVID-19 can be considered
as complex disease because of wide-ranging SARS-CoV-2 targets in the
host cell, which thus establishes the molecular connection with various
lung-related disorders. The disease-gene, disease–disease association
map, and network separation analysis have revealed molecular links and
clustering of diseases in the same network vicinity, indicating a
pathobiological similarity between COVID-19 and various lung disorders.
Some diseases closely associated with COVID-19 are haemolytic-uremic
syndrome, obstructive lung disease, pleural effusion, and chronic
bronchitis. Because of the close association, the pre-existence of
these diseases can lead to higher mortality of COVID-19 patients.
Severity of one of the common respiratory problems, asthma, which has
an overlapping disease module and is directly connected to COVID-19,
becomes moderate to high with SARS-CoV-2 infection ([143]www.cdc.gov).
These observations provide a detailed understanding of the molecular
basis of severe illness in COVID-19 patients with specific lung
disorders and help us decipher the patient-specific etiology of
COVID-19. Because of multiple molecular connections and overlapping
disease modules of COVID-19 with various lung disorders, finding
specific targets and potential drugs for COVID-19 patients with
pre-existing medical conditions is challenging. The present crisis
cannot wait until new drugs arrive; therefore, we proposed two
approaches for drug repositioning. The first approach is testing drugs
approved for lung diseases having modules overlapping with COVID-19.
These drugs can simultaneously affect two disease modules, leading to
much-improved treatment outcomes. The second approach is targeting host
functional protein modules that are linked with many lung disorders and
are primarily hijacked by SARS-CoV-2. Perturbing these modules by
repurposing FDA-approved (or investigational) drugs may rescue the host
cellular machinery utilized for virus replication. Considering the
complexity of SARS-CoV-2 infection, we suggest hitting multiple targets
in different functional modules to improve clinical outcomes. However,
systematic studies through clinical trials for identifying drug
combinations and their targets are highly recommended to increase
clinical efficacy and lower toxicity [[144]20]. Moreover,
patient-specific high-throughput transcriptomics data or construction
of a weighted gene expression networks from SARS-CoV-2-infected lung
tissues can further the possibility of target identification, like in
other human diseases [[145]48, [146]49]. In addition, Mendelian
randomization study can be performed to understand the causal
relationships between lung diseases and susceptibility and severity of
COVID-19 [[147]50, [148]51]. Lastly, the experimental validation of our
observation and in vitro or in vivo assays of drug combination and
study of pharmacokinetics are warranted to establish a proper treatment
strategy.
Conclusion
In summary, this study used the network biology framework to elucidate
the molecular link between lung disorders and COVID-19. The
network-based separation measure identified 59 lung diseases
topologically overlapped with the COVID-19 module. In addition, the
Disease-disease association network showed forty-nine diseases were
directly connected to COVID-19. This revealed the cause of severe
illness of patients with respiratory problems after SARS-CoV-2
infection. Genes in functional protein modules, hijacked by SARS-CoV-2,
are coexpressed and connected to several lung diseases. The
perturbation of these modules may block the virus growth in the host
cells. Therefore, existing FDA-approved drugs can target the hijacked
protein modules to avoid the life-threatening situation of COVID-19
patients with lung disorders.
Supplementary Information
[149]12920_2021_1079_MOESM1_ESM.docx^ (3.6MB, docx)
Additional file 1. Fig.S1. Dot plot shows the number of genes
associated with a lung disorder in LDGN. Fig. S2. Dot plot shows the
number of shared genes between COVID-19 other lung disorders. Fig.S3.
The network view of the Jaccard similarity coefficient between lung
diseases and COVID19. Fig.S4. Heat map shows functional protein modules
are associated with different disease classes. Fig.S5. Drug repurposing
to target functional protein modules.
[150]12920_2021_1079_MOESM2_ESM.pdf^ (703.2KB, pdf)
Additional file 2. Table S1. Edge list of SARS-CoV-2 target network
(STN) from TissueNet v.2 database.
[151]12920_2021_1079_MOESM3_ESM.pdf^ (570.4KB, pdf)
Additional file 3. Table S2. Disease-gene association data of lungs
from the ORGANizer database.
[152]12920_2021_1079_MOESM4_ESM.pdf^ (298.5KB, pdf)
Additional file 4. Table S3. Disease-gene association data of STN.
[153]12920_2021_1079_MOESM5_ESM.pdf^ (419.6KB, pdf)
Additional file 5. Table S4. Network-based separation measure between
disease modules and statistical significance.
[154]12920_2021_1079_MOESM6_ESM.pdf^ (408.7KB, pdf)
Additional file 6. Table S5 and S6. Comparison of different community
detection algorithms applied to SARS-CoV2 target network (STN) and
protein modules generated using Spinglass algorithem.
[155]12920_2021_1079_MOESM7_ESM.pdf^ (429.1KB, pdf)
Additional file 7. Table S7. Disease association with functional
protein modules.
[156]12920_2021_1079_MOESM8_ESM.pdf^ (448.7KB, pdf)
Additional file 8. Table S8. Targets in functional protein modules and
drugs from DrugBank database.
[157]12920_2021_1079_MOESM9_ESM.pdf^ (403.1KB, pdf)
Additional file 9. Table S9. Proximity between targets and COVID-19
disease module.
[158]12920_2021_1079_MOESM10_ESM.pdf^ (397.1KB, pdf)
Additional file 10. Table S10. Proximity between drugs in clinical
trial and COVID-19 disease module.
Acknowledgements