Computers in Biology and Medicine 158 (2023) 106740

Contents lists available at ScienceDirect

Computers in Biology and Medicine

journal homepage: www.elsevier.com/locate/compbiomed

Bioinformatics analysis of ferroptosis-related gene AKR1C3 as a potential

biomarker of asthma and its identification in BEAS-2B cells
Yufei Wang a, 1, Junwen Fan a, 1, Yu Tong a, Lei Wang a, Lingya Wang a, Cuiye Weng a, b,
Chuqiao Lai a, Jingjing Song a, **, Weixi Zhang a, *
a
Department of Pediatric Allergy and Immunology, The Second Affiliated Hospital and Yuying Children’s Hospital of Wenzhou Medical University, Wenzhou, 325027,
China
b
Department of Neonatology, The Second Affiliated Hospital and Yuying Children’s Hospital of Wenzhou Medical University, Wenzhou, 325027, China

A R T I C L E I N F O A B S T R A C T

Keywords: Ferroptosis is a newly discovered type of cell death and has recently been shown to be associated with asthma.
Asthma However, the relationship between them at the genetic level has not been elucidated via informatics analysis. In
Bioinformatics analysis this study, bioinformatics analyses are conducted using asthma and ferroptosis datasets to identify candidate
Diagnostic biomarker
ferroptosis-related genes using the R software. Weighted gene co-expression network analysis is performed to
AKR1C3
identify co-expressed genes. Protein–protein interaction networks, the Kyoto encyclopedia of genes and genomes,
BEAS-2B
Ferroptosis and gene ontology enrichment analysis are used to identify the potential functions of the candidate genes. We
experimentally validate the results of our analysis using small interfering RNAs and plasmids to silence and
upregulate the expression of the candidate gene in human bronchial epithelial cells (BEAS-2B). The ferroptosis
signature levels are examined. Bioinformatics analysis of the asthma dataset GDS4896 shows that the level of the
aldo-keto reductase family 1 member C3 (AKR1C3) gene in the peripheral blood of patients with severe therapy-
resistant asthma and controlled persistent mild asthma (MA) is significantly upregulated. The AUC values for
asthma diagnosis and MA are 0.823 and 0.915, respectively. The diagnostic value of AKR1C3 is verified using the
GSE64913 dataset. The gene module of AKR1C3 is evident in MA and functions through redox reactions and
metabolic processes. Ferroptosis indicators are downregulated by the overexpression of AKR1C3 and upregulated
by silencing AKR1C3. The ferroptosis-related gene AKR1C3 can be used as a diagnostic biomarker for asthma,
particularly for MA, and regulates ferroptosis in BEAS-2B cells.

Patients are typically classified as having severe therapy-resistant

asthma (SA) or controlled persistent mild asthma (MA) based on
1. Introduction
whether they have persistent symptoms despite treatment with corti­
costeroids [6,7]. Therefore, new targeted methods that can improve the
Asthma is a chronic and complex inflammatory disease of the airway
efficiency of asthma diagnosis and treatment are urgently required [8].
wall as well as the most typically encountered chronic lung disease in
Ferroptosis is a unique form of programmed cell death that depends
childhood [1], affecting more than 350 million people worldwide [2]. It
on intracellular iron and differs from autophagy and apoptosis. Recent
is a complex gene–environment interaction involving multiple signaling
studies have implicated the vital role of ferroptosis in various patho­
pathways that form its pathogenesis [3]. Asthma is characterized by
physiological processes and diseases [9]. Ferroptosis is typically char­
bronchial remodeling and airway hyperreactivity; therefore, the diag­
acterized by the regulation of iron metabolism, reactive oxygen species
nosis and prognosis of asthma are closely related to the degree of airway
(ROS) levels, and lipid peroxidation [10]. Many genes involved in these
epithelial cell injury. Airway epithelial cells are considered priority re­
processes are direct targets for ferroptosis, such as acyl-CoA synthetase
sponders to inflammatory, immune, and regenerative processes in
long-chain family member 4 (ACSL4, also known as FACL4), solute
asthma [4,5]. Owing to the complexity and heterogeneity of the disease,
carrier family 7 member 11 (SLC7A11, also known as xCT), and
patients typically encounter difficulties in their treatment courses.

** Corresponding author.
* Corresponding author.
E-mail addresses: songjj_immunity@wmu.edu.cn (J. Song), zhangweixi112@163.com (W. Zhang).
1
These authors contribute equally to this work.

https://doi.org/10.1016/j.compbiomed.2023.106740
Received 26 December 2022; Received in revised form 24 January 2023; Accepted 2 March 2023
Available online 15 March 2023
0010-4825/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Y. Wang et al.
List of abbreviations and acronyms
SA Severe therapy-resistant asthma
MA Controlled persistent mild asthma
ROS Reactive oxygen species
GPX4 Glutathione peroxidase 4
ACSL4 Acyl-CoA synthetase long-chain family member 4
SLC7A11 Cystine/glutamate antiporter
GSH Glutathione
WGCNA Weighted gene co-expression network analysis
AKR1C3 Aldo-keto reductase family 1 member C3
AKRs Aldo-keto reductases
NADPH Nicotinamide adenine dinucleotide phosphate
siRNAs Small interfering RNAs
GEO Gene Expression Omnibus
DEGs Differentially expressed genes
FC Fold change
ROC Receiver-operating-characteristic
AUC Area under the ROC curve
glutathione peroxidase 4 (GPX4), which regulate diseases by affecting
iron metabolism, intermediate metabolism, and glutathione (GSH)
synthesis/metabolism [11–13].
Asthma and ferroptosis are closely associated. A study demonstrated
the use of anti-ferroptosis drugs in ovalbumin-induced mouse asthma
models and IL-13 induction cells, in which inflammation and ROS levels
were reduced [14]. ROS is vital to airway epithelium-mediated airway
remodeling and hyper-responsiveness, which are integral to ferroptosis
[8]. However, the relationship between genes involved in asthma and
ferroptosis is yet to be elucidated.
Weighted gene co-expression network analysis (WGCNA) is a novel
computer algorithm that identifies the relationship between key genes
and clinical features by clustering genes with similar gene expression
patterns to enable deep learning analysis of data and accurate clinical
judgments [15]. It can effectively select potential target genes for
therapeutic use in diseases such as prostate cancer and schizophrenia
[16,17]. Currently, clinical features and biomarkers are increasing
required by asthma patients to enable precision medicine [18]. The
recent development in computer algorithms has resulted in increased
interest toward the use of algorithms, artificial intelligence, and other
modes of deep learning to aid the diagnosis and treatment of diseases
using existing data. For example, deep learning-based algorithms have
been implemented in oncology for clinical purposes such as diagnosis,
staging, and prognosis prediction [19]. Different image filtering
methods combined with sample features are used to classify Alzheimer’s
disease [20]; meanwhile, improved automatic segmentation methods
can improve the accuracy of magnetic resonance imaging data of the
prostate [21], among others. Therefore, the use of computers for the
comprehensive learning of data, especially based on indicators for which
large amounts of patient data have been identified by consensus as
important markers of disease [22], to develop automatic or
semi-automatic diagnostic and therapeutic models may be the future
solution [23].
Aldo-keto reductase family 1 member C3 (AKR1C3) is a member of
the aldo-keto reductase (AKR) family and functions in NAD(P)(H)-
dependent oxidation and reduction as an oxidoreductase [24]. AKRs
have been shown to regulate the metabolism of steroids and lipid al­
dehydes [25]. Hence, the AKR family was included in the list of
ferroptosis-related genes. In addition, AKR1C3 is a promising biomarker
and drug target for many diseases, including acute myocardial infarc­
tion, hepatocellular carcinoma, and metastatic melanoma [26–28].
Notably, no murine AKR1C3 protein or gene paralog exists [29]. The
role of AKR1C3 in airway-related diseases remains unclear, and its role
2
Computers in Biology and Medicine 158 (2023) 106740
KEGG Kyoto encyclopedia of genes and genomes
GO Gene ontology
PPI Protein-protein interaction
BEAS-2B Human bronchial epithelial cells
qRT-PCR Real-time quantitative PCR
GAPDH Glyceraldehyde-3-phosphate dehydrogenase
DCFH-DA 2′ , 7′ -dichlorofluorescin diacetate
GSSG Oxidized glutathione
HBSS Hank’s balanced salt solution
MAPK Mitogen-activated protein kinases
ns Nonsense
FDR False discovery rate
CC Cellular components
BP Biological process
MF Molecular function
CI Confidence interval
OE Overexpression
PLOOH Phospholipid hydroperoxides
PUFAs Polyunsaturated fatty acids
in the regulation of ferroptosis has not been investigated.
In this study, we screen for potential biomarkers associated with
ferroptosis in asthma, particularly MA. We analyze the gene association
patterns between different samples for this gene using WGCNA. We
overexpress the gene with plasmids and small interfering RNAs (siRNAs)
to suppress it and then investigate its effect on ferroptosis in airway
epithelial cells. Mechanistically, we explore its effects on lipid peroxi­
dation and redox reactions in the airway epithelium. Thus, in this study,
a new biomarker for the diagnosis, treatment, and typing of asthma is
provided, and an important regulator of ferroptosis is identified.
2. Materials and methods
2.1. Data acquisition
To investigate the relationship between asthma and ferroptosis at the
genetic level, we retrieved asthma sequencing datasets from the Gene
Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/).
GDS4896 is one of the few datasets of blood samples tested from patients
with varying degrees of asthma. Because we utilized airway epithelial
cells for validation, GSE64913, which is a gene set sequenced from
epithelial cells, was duplicated. The two datasets selected were those
with relatively large sample sizes retrieved from the GEO asthma data­
set. Microarray datasets GDS4896 and GSE64913 were downloaded
from the “GEOquery” package of the R software. Ferroptosis-related
genes were obtained from the FerrDb database (http://www.zhounan.
org/ferrdb).
2.2. Candidate gene discovery and assessment
The “limma” R package was used to identify differentially expressed
genes (DEGs) with an adjusted p-value of less than 0.05 and an absolute
fold change (FC) greater than 0.5 between any two groups. Venn dia­
grams of overlaps were plotted using an online Venn diagram generator.
Calculations were performed based on a receiver operating character­
istic (ROC) curve and its area under the ROC curve (AUC) to assess the
diagnostic ability of an online generator.
2.3. WGCNA
WGCNA was performed using the “WGCNA” R package to identify
the co-expression gene module associated with the candidate gene
discovered and the severity of the diseases. We constructed a sample tree
Y. Wang et al.
via hierarchical clustering and removed the outliers (cut height = 55,
min size = 10). An adjacency cutoff value of 0.85, a soft threshold power
of 10, and a minimum module size of 100 were used to construct a
weighted adjacency matrix and detect co-expressed gene modules.
2.4. Kyoto encyclopedia of genes and genomes (KEGG), gene ontology
(GO), and protein–protein interaction (PPI)
KEGG and GO enrichment analysis was performed using the
“enrichKEGG” and “enrichGo” functions in the “clusterprofiler” R
package, where the gene expression profiles described above were used.
An online service tool was used for visualization. The protein network
was output from the STRING database (https://string-db.org/) and
visualized using Cytoscape (v3.9.1).
2.5. Cell culture
Cells from human bronchial epithelial cell line BEAS-2B (FuHeng
Biology, Shanghai, China) were cultured in complete Dulbecco’s modi­
fied Eagle’s medium with 10% fetal bovine serum (Excell) and 1%
penicillin/streptomycin (Gibco) in a humidified incubator at 37 ◦ C with
5% CO2. In addition, BEAS-2B cells were passaged every 2 or 3 d after
digestion with pancreatin.
2.6. Plasmid transfection and RNA interference
Cells were seeded in a complete medium for 24 h and subsequently
transfected with siRNAs or plasmids. Full-length AKR1C3 cDNA was
synthesized and inserted into pcDNA3.1 vectors. The siRNAs and plas­
mids were constructed by Tsingke Biotechnology; the siRNAs were
transfected using Lipofectamine RNAiMAX. The resulting vector or
empty vector was transfected using Lipo3000 and P3000 (Thermo Fisher
Scientific, USA). Ferroptosis groups were exposed to 1 μM erastin
(MedChemExpress, USA) for 24 h after transfection. All experiments
were performed using the transfected cells 48 h after transfection. The
sequences of si-AKR1C3 are listed as follows (5′ –3′ ):
siAKR1C3-1: GGUAGAAUGUCAUCCGUAUTT
siAKR1C3-2: CCCUAAUUAUCCAUAUUCATT.
2.7. Real-time quantitative PCR (qRT-PCR)
The total cellular RNA was extracted using a TRIzol reagent
(GLPBIO, USA). The concentration of the RNA samples was measured at
260/280 nm using a NanoDrop microvolume spectrophotometer
(Thermo Fisher Scientific, Waltham, MA, USA) both qualitatively and
quantitatively. Reverse transcription to cDNA was performed using the
qRT Master (TOROIVD, China). Subsequently, qRT-PCR was performed
using ChamQ SYBR Qpcr Master Mix (Vazyme, China) and a fluores­
cence quantitative PCR instrument (Bio-Rad, USA). The primers were
synthesized by Sangon Biotech (Shanghai) and are listed as follows
(5′ –3′ ):
GAPDH-F, GTCTCCTCTGACTTCAACAGCG.
GAPDH-R, ACCACCCTGTTGCTGTAGCCAA.
AKR1C3–F, AAGCTTTGGTCCACTTTTCATC.
AKR1C3-R, GGTCAACATAGTCCAATTGAGC.
GPX4-F, GAGGCAAGACCGAAGTAAACTAC.
GPX4-R, CCGAACTGGTTACACGGGAA.
ACSL4-F, CATCCCTGGAGCAGATACTCT.
ACSL4-R, TCACTTAGGATTTCCCTGGTCC.
SLC7A11-F, TTACCAGCTTTTGTACGAGTCT.
SLC7A11-R, GTGAGCTTGCAAAAGGTTAAGA.
The relative expression of the isolated mRNA was calculated using
the 2-ΔΔCT method with GAPDH as an endogenous control.
3
Computers in Biology and Medicine 158 (2023) 106740
2.8. Western blot
The total protein was extracted using cell lysis buffer (RIPA buffer
with a protease inhibitor cocktail and PMSF (Beyotime Biotechnology,
Shanghai, China). After being centrifuged at 12,000×g for 10 min at
4 ◦ C, the protein concentration of the cell lysates was estimated using the
BCA method. Proteins were separated and electrotransferred using a
PAGE Gel Fast Preparation Kit (Epizyme, Shanghai, China) and PVDF
membrane (Millipore, Bedford, MA, USA). After being obstructed by a
Western blocking reagent (Solarbio Science & Technology, Beijing,
China), the membranes were incubated at 4 ◦ C overnight with specific
primary antibodies: anti-AKR1C3 antibody (1:20000, Abcam), anti-
ACSL4 (1:20000, Abcam), anti-SLC7A11 (1:1000, ZENBIO), anti-GPX4
(1:1000, ZENBIO), anti-tubulin (1:1000, Proteintech), and anti-GAPDH
(1:1000, Cell Signaling Technology). Subsequently, the membranes
were incubated with a secondary antibody goat anti-rabbit IgG
(1:10000, Biosharp) at room temperature for 1.5 h. All protein bands in
the membranes were visualized and photographed using an ECL detec­
tion kit (Affinity, China) and a multifunctional gel imaging system (Bio-
Rad, USA). The data were quantitated using the Image Lab software
(Bio-Rad, USA).
2.9. Total cellular ROS measurement
The total cellular ROS generation was measured using a ROS
detection assay kit (Thermo Fisher Scientific) with 2′ , 7′ -dichloro­
fluorescein diacetate (DCFH-DA) and a positive control. DCFH-DA reacts
with ROS to form the fluorescent product, dichlorofluorescein. Thus, the
mean fluorescence intensity indicates the intercellular ROS levels via
flow cytometry. The cells were co-cultured with 1 μM DCFH-DA for 2 h
after digestion by pancreatic enzymes and washed with PBS. We set the
number of cells to 50,000, used the FL1-A:FITC-A channel based on the
instructions, and then visualized th results using FlowJo (v10.6.2).
2.10. Cell proliferation assay
Cells transfected after 48 h were seeded into 96-well plates (5 × 103
cells/well) and incubated for 24 h. CCK-8 assay (Beyotime Biotech­
nology, Shanghai) was used to assess cell viability. CCK-8 solution was
added to each well and further co-cultured for 4 h. An automatic enzyme
label analyzer (Thermo Fisher Scientific, USA) was used to determine
the optical density at 450 nm.
2.11. GSH/GSSG ratio measurement
GSH and oxidized glutathione (GSSG) levels were measured using
GSSG/GSH Quantification Kit II (Dojindo, Shanghai, China) based on
the manufacturer’s instructions. We separated one supernatant into two
and measured the amounts of total GSH and GSSG, separately and
simultaneously. The wells of the sample and standard curves were
incubated for 10 min after the working enzyme was added rapidly.
Subsequently, based on the standard curve, the concentrations of total
GSH and GSSG were measured at 405 nm using an automated enzyme
analyzer (Thermo Fisher Scientific, USA). The GSH/GSSG ratio was
calculated using the following formula: GSH/GSSG = (Total GSH-
2GSSG)/(GSSG).
2.12. Intracellular Fe2+ and lipid peroxidation measurement
Fe2+ levels were measured using FerroOrange (Dojindo, Shanghai)
and lipid peroxidation was detected using Liperfluo (Dojindo, Shanghai)
via laser scanning confocal microscopy. For confocal imaging, the cells
were directly seeded in a 35 mm confocal dish, and the transfection
process was completed as described previously. After washing twice
using Hank’s balanced salt solution (HBSS), the cells were stained with
1 μmol/L of FerroOrange probe and 8 μmol/L of Liperfluo (diluted in
Y. Wang et al.
HBSS) and incubated at 5% CO2 and 37 ◦ C in an incubator for more than
30 min, based on the instructions. Living cells were observed, and im­
ages were captured using a confocal microscope (FerroOrange: Ex: 561
nm/Em: 570–620 nm; Liperfluo: Ex: 488 nm/Em: 500–550 nm). The
images were analyzed using the ImageJ software to quantify the fluo­
rescence intensity.
2.13. Statistical analysis
Bioinformatics analyses were performed using the R software
(v4.2.1). Data acquisition and collation were performed using Microsoft
Excel and GraphPad Prism 8.2.0, and statistical software was used to
perform data analysis and image production. An unpaired t-test was
used to examine the differences between the two groups. All data are
presented as mean ± standard deviation. P values of ≥0.05, <0.05, 0.01,
0.001, and 0.0001 are indicated by nonsense (ns), *, **, ***, and **** in
all figures, respectively. The statistical significance was set at P < 0.05.
3. Results
3.1. AKR1C3 is the only ferroptosis-related DEG and a prognostic
biomarker of asthma
We downloaded the original file of the human asthma gene dataset
GDS4896, replaced the probes with gene names, and obtained the
average value for the same gene. Gene set GDS4896 was used to analyze
white blood cells from children with SA and MA. GDS4896 contained 54
samples, including those of 18 healthy individuals, 19 patients with MA,
and 17 patients with SA. Using the presence of asthma for categoriza­
tion, the KEGG pathway analysis showed that the genes of GDS4896
were enriched in many pathways (P < 0.05, FDR < 0.25), including the
mitogen-activated protein kinases (MAPK) pathway, Th17 pathway,
fatty acid metabolism, and ferroptosis (Fig. 1a and Supplementary
Table S1).
Meanwhile, 56 genes differed be significantly between the asthma
and control groups (logFC absolute values > 0.5 and adjusted P-values <
0.05). Ferroptosis-related gene sets containing 259 driver, suppressor,
and marker genes were obtained from the previous iron death database
after assembling and deleting duplicate values. A Venn plot revealed
that only one ferroptosis-related gene, AKR1C3, was differentially
4
Computers in Biology and Medicine 158 (2023) 106740
expressed between the asthmatic and control groups. To further examine
the differential expression of AKR1C3, we constructed Venn plots of the
differential genes in every two groups using ferroptosis gene sets. Our
results show that AKR1C3 revealed significant differences between the
MA and control groups and between the SA and MA groups (Fig. 1b).
AKR1C3 was significantly upregulated in patients with asthma
compared with that in the healthy group, regardless of the severity of
asthma. Interestingly, the expression of AKR1C3 in the MA group was
substantially higher than that in the SA group (Fig. 1c).
We used the expression level of AKR1C3 as a predictor to construct
ROC curves and calculate the AUC values. AKR1C3 indicated good
predictive efficacy for asthma (AUC = 0.823; CI: 0.713–0.932). Between
the asthma groups, MA was more diagnostic (AUC = 0.915; CI:
0.826–1.000) than SA (AUC = 0.719; CI: 0.546–0.892) (Fig. 1d). Dif­
ferential comparisons and ROC curve construction were performed
using GSE64913 to validate the finding. This database was used to
analyze epithelial gene expression in the peripheral airways and
included SA and control groups. AKR1C3 differed significantly between
the asthma patients and the control (AUC = 0.640; CI: 0.504–0.777)
(Fig. 1e).
3.2. Identification of modules associated with AKR1C3 by WGCNA
To obtain the co-expression genes of AKR1C3, we further analyzed
the GDS4896 dataset by performing WGCNA using the R software. After
removing five outliers, we adopted 10 as the soft threshold to reach an
adjacency cutoff value of 0.85 (Fig. 2a). Subsequently, we set 100 as the
minimum gene module and constructed a clustering dendrogram, from
which 19 regulatory modules were identified (Fig. 2b).
The association between the modules and asthma severity is shown
in Fig. 2c. The black and green–yellow modules indicated the highest
correlation with the control (negative value equal to total asthma).
Moreover, the black module was critical for severe asthma (cor = 0.61;
P = 3 × 10− 6), whereas the green–yellow module was important for mild
asthma (cor = 0.5; P = 2 × 10− 4). The gene sets and asthma phenotypes
were significantly associated in the black and green–yellow modules
(cor = 0.51; P = 1.7 × 10− 37 and cor = 0.51; P = 2.1 × 10− 20), sepa­
rately (Fig. 2d). Furthermore, AKR1C3 appeared only in the green­
–yellow module, which contained 287 genes that were separated from
the black module (Fig. 2e). Among them, 11 genes were differentially
Fig. 1. Prognostic value of AKR1C3 in asthma. (a)
KEGG analysis based on genes differentially expressed
in GDS4896 asthma dataset. (b) Intersection of
ferroptosis-related genes and asthma differential
genes to obtain AKR1C3 in GDS4896 asthma dataset.
(c) AKR1C3 highly expressed in asthma and expressed
higher in MA compared with in SA. (d) ROC curves of
AKR1C3 in asthma, SA, and MA in GDS4896 asthma
dataset. (e) Expression and ROC curves of AKR1C3 in
GSE64913 asthma dataset.
Y. Wang et al.
expressed in asthma, including AKR1C3.
Genes within the green–yellow module were analyzed using the
STING database and visualized using Cytoscape to establish a PPI
network. The thickness and color of the lines between nodes represent
the combined scores of the genes. Twelve proteins in the green–yellow
module, combined with AKR1C3, are shown in Fig. 2f. Among them,
AKR1C4 and PTGDS were more strongly associated with AKR1C3.
KEGG and GO analyses were performed to identify the potential
pathways of the gene module in which AKR1C3 was located. KEGG was
significantly enriched in metabolism-related pathways, including purine
metabolism, metabolic pathways, fatty acid degradation, regulation of
lipolysis in adipocytes, and cholesterol metabolism (Fig. 2g and Sup­
plementary Table S2). Further analysis using the GO database revealed
that the top cellular components were the cytosol, endomembrane sys­
tem, vesicle, and extracellular activity. In the biological process, pro­
grammed cell death as well as lipid metabolic and oxidation–reduction
processes showed clear associations with this module. In the molecular
function, this module showed significant correlations with enzymes,
such as catalytic activity, enzyme regulator activity, lipid binding,
oxidoreductase activity, and metabolic and oxidative biological pro­
cesses (Fig. 2h and Supplementary Table S3).
3.3. Expression of AKR1C3-protected BEAS-2B cells from ferroptosis
To observe the effect of AKR1C3 on ferroptosis, we used siRNAs to
5
Computers in Biology and Medicine 158 (2023) 106740
Fig. 2. WGCNA of AKR1C3 using GDS4896 dataset.
(a) WGCNA with 0.85 and 10 as adjacency cutoff
value and soft threshold, respectively. (b) Formation
of 19 co-expression gene modules via clustering. (c)
Heatmap showing association between each module
and asthma severity. (d) Correlation analysis of genes
and phenotypes within black and green–yellow
modules. (e) Venn diagram of black and green–yellow
modules and GDS4896 DEGs. (f) PPI network of
AKR1C3 within green–yellow module. (g–h) KEGG
and GO analysis of genes in green–yellow module.
silence AKR1C3 and a plasmid to overexpress AKR1C3 in BEAS-2B cells.
Both si-AKR1C3-1 and si-AKR1C3-2 showed good silencing efficiency of
AKR1C3 at both the mRNA and protein levels. Therefore, we randomly
selected si-AKR1C3-1 to perform subsequent experiments. Results show
that pcDNA3.1-AKR1C3 significantly upregulated the expression of
AKR1C3 in BEAS-2B cells (Fig. 3a).
Results of the WGCNA suggested that AKR1C3 contributes signifi­
cantly to asthma and ferroptosis via its effect on lipid metabolism and
enzyme-related redox reactions. Fe2+ and lipid peroxides are vital to
ferroptosis. After co-staining the BEAS-2B cells using ferroOrange and
Liperfluo probes, they were photographed via confocal microscopy.
Fluorescence images showed that both Fe ions and lipid peroxides were
present in the cytoplasm at similar locations (Fig. 3b). Quantitative
analysis showed that the fluorescence levels of both Fe2+ and lipid
peroxides increased significantly in AKR1C3-silenced cells compared
with those in the control group, whereas the levels of both decreased
significantly in AKR1C3-overexpressed cells. This indicates that a high
AKR1C3 expression level can effectively reduce the intracytoplasmic
levels of Fe2+ and lipid peroxides, thus suggesting that the over­
expression of AKR1C3 can inhibit ferroptosis. The ferroptosis inducer
erastin significantly increased the levels of subferric ions and lipid
peroxides in the BEAS-2B cells. Meanwhile, the number of cells over­
expressing AKR1C3 increased significantly when erastin was added.
However, in cells with low AKR1C3 expression, erastin failed to provide
a similar effect and instead significantly inhibited the increase in Fe2+
Y. Wang et al.
Fig. 3. Expression of AKR1C3 protected BEAS-2B cells from ferroptosis. siRNAs and plasmid successfully silenced and overexpressed mRNA and protein expression of
AKR1C3 (Fig. 3a). Qualitative and quantitative characterization of intracellular Fe2+ and lipid peroxidation after AKR1C3 regulation using FerroOrange and Liperfluo
probes (Fig. 3b–c).
and lipid peroxides induced by AKR1C3 silencing (Fig. 3c–d).
The CCK-8 method was used to examine the activity of cells with
different levels of AKR1C3 expression. The data showed that the
viability of AKR1C3-silenced cells reduced significantly, whereas it
increased in cells overexpressing AKR1C3, indicating that AKR1C3
expression can significantly affect cell viability (Fig. 4a). GSH is
involved in biological processes, such as antioxidants and ferroptosis,
and is used as a substrate for GPX4 [30]. The total GSH includes reduced
GSH and GSSG, which are interconvertible, and the GSH/GSSG ratio is
often considered an important indicator of oxidative stress [31]. Our
results show that silencing AKR1C3 can significantly downregulate
reduced GSH, whereas AKR1C3 overexpression can upregulate GSH.
Furthermore, GSSG content was not significantly altered by AKR1C3
overexpression but was reduced by silencing AKR1C3. The GSH/GSSG
ratio, as an important ferroptosis intermediate that represents the
oxidative stress process, can decrease significantly under the inhibition
of AKR1C3 and can be upregulated when AKR1C3 is overexpressed
(Fig. 4b). Additionally, we investigated whether AKR1C3
6
Computers in Biology and Medicine 158 (2023) 106740
overexpression significantly reduced ROS production and whether the
downregulation of AKR1C3 induced ROS production (Fig. 4c).
During ferroptosis, AKR1C3 was significantly downregulated by the
addition of the ferroptosis inducer, erastin (Fig. 4d). We examined the
expression levels of several key genes of ferroptosis at different
expression levels of AKR1C3 via qRT-PCR and protein blotting. ACSL4
promotes ferroptosis, whereas SLC7A11 and GPX4 impose an inhibitory
effect on ferroptosis. Our data show that at both the gene and protein
levels, silencing AKR1C3 increased ACSL4 and decreased SLC7A11
expression, whereas the overexpression of AKR1C3 yielded the opposite
effect. This suggests that the high expression of AKR1C3 may protect the
cells from ferroptosis by affecting the ACSL4 and SLC7A11 pathways.
Interestingly, both the silencing and overexpression of AKR1C3
increased the expression of the key factor, GPX4. Further analysis
revealed that AKR1C3 overexpression significantly inhibited erastin-
induced GPX4 downregulation (Fig. 4e).
Finally, we provide a schematic diagram showing the role of AKR1C3
in regulating ferroptosis in BEAS-2B cells (see Fig. 5).
Y. Wang et al.
Fig. 4. Function of AKR1C3 toward ferroptosis in BEAS-2B cells. Expression of AKR1C3 promoted proliferation (Fig. 4a).GSH/GSSG ratio and ROS levels are affected
by AKR1C3 expression (Fig. 4b–c). Significant downregulation of mRNA and protein expression of AKR1C3 in ferroptosis (Fig. 4d). Expression of AKR1C3 affecting
ferroptosis marker genes in BEAS-2B cells (Fig. 4e).
4. Discussion
Asthma is a global health problem affecting all age groups, and its
diagnosis and treatment remain challenging [32]. In recent years, new
algorithms have been created and successfully solved some practical
problems such as improving the economic emission scheduling problem
[33], building forward-looking bankruptcy prediction models [34],
enhancing search and mining capabilities [35], and improving remote
pulse extraction [36], respectively. At the same time, many new
methods for disease diagnosis and prediction have emerged, such as
genetic diagnostic markers, machine learning [37], and image seg­
mentation [38], which are different from classical models. In the present
study, we re-analyzed an existing database GDS4896, which included
populations with different asthmatic severities, and discovered that the
ferroptosis pathway was strongly associated with asthma in terms of
gene relationship. This is consistent with the results of previous studies
performed on mice, i.e., the inhibition of ferroptosis improves asthma
[14].
However, our results show that only one ferroptosis-related gene,
AKR1C3, was both significantly and differentially expressed and upre­
gulated in asthma. Additionally, we discovered that AKR1C3 showed
excellent diagnostic efficacy for asthma, particularly MA. To further
determine the effect of AKR1C3 on the severity of asthma, we used a new
7
Computers in Biology and Medicine 158 (2023) 106740
bioinformatics tool, WGCNA, to obtain the signature module of MA and
SA by setting them as different phenotypes to infer the recognition role
of AKR1C3 and identify the genes co-expressed with AKR1C3 to further
investigate the pathway of AKR1C3. For the first time, we discovered
that the co-expressed gene module in which AKR1C3 was located
showed high sensitivity for the characterization of MA, which is in
contrast to the module associated with SA. In addition, KEGG and GO
enrichment analyses revealed that this module was closely related to
enzyme-related pathways and metabolism, particularly those related to
lipid and redox reactions. Ferroptosis is a newly discovered type of
programmed death that involves both lipid metabolism and oxidative
stress [39,40]. Although pathways such as the MAPK pathway have been
extensively demonstrated in previous studies, investigations into the
ferroptosis pathway in asthma have only begun. Based on all findings
thus far, AKR1C3 may contribute significantly to asthma, particularly
MA, via the ferroptosis pathway.
The protective role of AKR1C3 in many diseases, such as leukemia
and endometriosis has been demonstrated [41,42]; hence, one may
speculate that AKR1C3 also offers a protective role in asthma. The
Global Initiative for Asthma recommends classifying asthma into mild
and severe segments for treatment [43]. Although AKR1C3 differs
significantly between MA and SA, based on the specificity of the
co-expression module in which AKR1C3 is located for MA, we
Y. Wang et al.
Fig. 5. Diagram of hypothesis mechanism of this study.
hypothesized that the expression of AKR1C3 can delay the progress of
asthma by counteracting inflammation. Thus, AKR1C3 may be used as
an indicator of asthma progression and prognosis. In conclusion,
ferroptosis-related gene AKR1C3 was reported for the first time in this
study as a diagnostic gene for asthma, which can potentially be used to
differentiate the degree of asthma progression to facilitate asthma
treatment.
BEAS-2B cells were used as experimental objects owing to the
following reasons. First, the homologous genes of AKR1C3 in mice do
not exhibit one-to-one correspondence. The mouse Akr1c18 gene is
homologous to several human genes including AKR1C1, AKR1C2, and
AKR1C3. Second, obtaining bronchial epithelial cells from asthmatic
children is challenging. BEAS-2B cells are considered ideal cell models
for respiratory diseases and molecular mechanisms. Previous studies
have identified the transcription factor HOXB4, which can regulate the
expression of AKR1C3 and affect ferroptosis in H9C2 cells [26]. This is
comparable to the results of the present study; nonetheless, we are the
first to perform investigations based on BEAS-2B cells as well as directly
regulate AKR1C3 expression, thereby demonstrating the importance of
AKR1C3 in ferroptosis more directly and comprehensively. Our experi­
ments based on BEAS-2B cells demonstrated that the overexpression of
AKR1C3 significantly downregulated Fe ions and lipid peroxides and
substantially protected cells from ferroptosis, which supports the results
of the bioinformatics analysis above. Furthermore, the ability of erastin
to enhance Fe ions and lipid peroxides was significantly inhibited in
AKR1C3-deficient cells, whereas the effect of erastin remained in
AKR1C3-overexpressed cells. Combining th above with the down­
regulation of AKR1C3 expression after the addition of erastin, we
speculate that erastin may induce ferroptosis by reducing AKR1C3
expression.
To further demonstrate that AKR1C3 mediates biological processes
via the ferroptosis pathway, we investigated several indicators of fer­
roptosis. We discovered that high AKR1C3 expression reduced intra­
cellular ROS levels, increased GSH and thus the GSH/GSSG ratio,
decreased ACSL4, increased SLC7A11, and protected GPX4 from erastin
modulation. Meanwhile, AKR1C3 knockdown resulted in an increase in
GPX4 in our data; in our opinion, this occurred because the significantly
reduced GSH caused by AKR1C3 knockdown resulted in a compensatory
increase in GPX4, which indicates that AKR1C3 does not interact
directly with GPX4.
8
Computers in Biology and Medicine 158 (2023) 106740
Ferroptosis is triggered by the iron-dependent oxidation of phos­
pholipid hydroperoxides (PLOOH), which are produced via the meta­
bolism of polyunsaturated fatty acids using ACSL4 [11]. Fe ions can
directly generate significant amounts of PLOOH through the Fenton
reaction [44]. Cells rely on the antioxidant GSH to neutralize PLOOHs
with the contribution of the co-factor GPX4. High levels of GSH are
essential for clearing excess ROS [45]. ROS accumulation has been
proven to result in elevated levels of PLOOH. The cell use system x(C)
(− ) to import oxidized cysteine (cysteine) for GSH synthesis, SLC7A11,
is a key subunit of system x(C) (− ); in fact, system x(C) (− ) enables
erastin to act in ferroptosis [46]. NADPH biosynthesis is vital to the
ferroptosis process [9]. Combining this with the ability of AKR1C3 to
convert aldehydes and ketones into hydroxylated compounds via
NADPH [47], we hypothesize that AKR1C3 regulates related genes,
ROS, and GSH, which consequently affects lipid peroxidation and in­
fluences ferroptosis. The action mode of AKR1C3 is illustrated in Fig. 5.
In addition, researchers have suggested that AKR1C3 may decrease the
detoxification of toxic lipid metabolites such as 4-HNE [48]. Moreover,
the protective effects of AKR1C3 have been demonstrated in numerous
diseases. AKR1C3-specific prodrugs TH3424 and AST-3424/OBI-3424
have been demonstrated to exhibit effective antitumor activity [31,
32]. For example, AKR1C3 overexpression offers a notable protective
role in prostate cancer [33]. In addition, AKR1C3 inhibitors can restore
chemotherapy drug sensitivity in multiple tumors, such as hepatocel­
lular carcinoma and breast cancer [49,50]. Nevertheless, studies
regarding AKR1C3 in non-tumor diseases, particularly
respiratory-related diseases, remain insufficient. Our findings suggest
that AKR1C3 offers a prognostic role in asthma and balances iron ion
and redox levels in BEAS-2B cells. Thus, AKR1C3 is a promising mo­
lecular therapeutic target for asthma in the future.
However, our study remains inadequate, as we have only demon­
strated our findings based on bioinformatics and in vitro experiments, i.
e., the expression of AKR1C3 in samples with varying degrees of clinical
asthma was not analyzed. Although numerous signature genes for iron
death exist, the manner by which they interact is yet to be elucidated;
hence, this will be addressed in our subsequent study.
In conclusion, our results show that AKR1C3 offers high diagnostic
and prognostic value in asthma, particularly MA. Based on experimental
verification, the overexpression of AKR1C3 can protect BEAS-2B cells
from ferroptosis. Therefore, AKR1C3 may be a promising new diagnostic
Y. Wang et al.
biomarker and drug target for preventing ferroptosis in asthma.
Funding
This research was supported by Zhejiang Provincial Natural Science
Foundation, China under Grant No.LY23H010003.
CRediT authorship contribution statement
Yufei Wang: Formal analysis, Invstigation, Methodology, Writing-
original draft. Junwen Fan: Formal analysis, Methodology, Software.
Yu Tong, Visualization. Lei Wang, Software. Lingya Wang: Software,
Validation. Cuiye Weng: Investigation, Validation. Chuqiao Lai: Vali­
dation. Jingjing Song: Conceptualization, Founding acquisition,
Writing-review & editing. Weixi Zhang: Conceptualization, Founding
acquisition, Writing-review & editing.
Declaration of competing interest
All authors declared no competing interest.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.
org/10.1016/j.compbiomed.2023.106740.
References
[1] H.S. Zahran, et al., Vital signs: asthma in children - United States, 2001-2016,
MMWR Morb. Mortal. Wkly. Rep. 67 (5) (2018) 149–155.
[2] Global, regional, and national deaths, prevalence, disability-adjusted life years, and
years lived with disability for chronic obstructive pulmonary disease and asthma,
1990-2015: a systematic analysis for the Global Burden of Disease Study 2015,
Lancet Respir. Med. 5 (9) (2017) 691–706.
[3] A. Papi, et al., Asthma. Lancet 391 (10122) (2018) 783–800.
[4] Y. Gon, S. Hashimoto, Role of airway epithelial barrier dysfunction in pathogenesis
of asthma, Allergol. Int. 67 (1) (2018) 12–17.
[5] S.S. Athari, Targeting cell signaling in allergic asthma, Signal Transduct. Targeted
Ther. 4 (2019) 45.
[6] M.W. Pijnenburg, L. Fleming, Advances in understanding and reducing the burden
of severe asthma in children, Lancet Respir. Med. 8 (10) (2020) 1032–1044.
[7] T.W. Guilbert, L.B. Bacharier, A.M. Fitzpatrick, Severe asthma in children,
J. Allergy Clin. Immunol. Pract. 2 (5) (2014) 489–500.
[8] A. Banno, et al., Bidirectional interaction of airway epithelial remodeling and
inflammation in asthma, Clin. Sci. (Lond.) 134 (9) (2020) 1063–1079.
[9] B.R. Stockwell, et al., Ferroptosis: a regulated cell death nexus linking metabolism,
redox Biology, and disease, Cell 171 (2) (2017) 273–285.
[10] S.J. Dixon, et al., Ferroptosis: an iron-dependent form of nonapoptotic cell death,
Cell 149 (5) (2012) 1060–1072.
[11] S. Doll, et al., ACSL4 dictates ferroptosis sensitivity by shaping cellular lipid
composition, Nat. Chem. Biol. 13 (1) (2017) 91–98.
[12] K. Bersuker, et al., The CoQ oxidoreductase FSP1 acts parallel to GPX4 to inhibit
ferroptosis, Nature 575 (7784) (2019) 688–692.
[13] X. Lang, et al., Radiotherapy and immunotherapy promote tumoral lipid oxidation
and ferroptosis via synergistic repression of SLC7A11, Cancer Discov. 9 (12) (2019)
1673–1685.
[14] N. Yang, Y. Shang, Ferrostatin-1 and 3-methyladenine ameliorate ferroptosis in
OVA-induced asthma model and in IL-13-challenged BEAS-2B cells, Oxid. Med.
Cell. Longev. 2022 (2022), 9657933.
[15] P. Langfelder, S. Horvath, WGCNA: an R package for weighted correlation network
analysis, BMC Bioinf. 9 (2008) 559.
[16] G.P. Risbridger, et al., Patient-derived xenografts reveal that intraductal carcinoma
of the prostate is a prominent pathology in BRCA2 mutation carriers with prostate
cancer and correlates with poor prognosis, Eur. Urol. 67 (3) (2015) 496–503.
[17] E. Radulescu, et al., Identification and prioritization of gene sets associated with
schizophrenia risk by co-expression network analysis in human brain, Mol.
Psychiatr. 25 (4) (2020) 791–804.
[18] R. Kaur, G. Chupp, Phenotypes and endotypes of adult asthma: moving toward
precision medicine, J. Allergy Clin. Immunol. 144 (1) (2019) 1–12.
[19] Y. Jiang, et al., Emerging role of deep learning-based artificial intelligence in tumor
pathology, Cancer Commun. 40 (4) (2020) 154–166.
Computers in Biology and Medicine 158 (2023) 106740
[20] B. Yan, et al., Quantifying the impact of Pyramid Squeeze Attention mechanism
and filtering approaches on Alzheimer’s disease classification, Comput. Biol. Med.
148 (2022), 105944.
[21] Y. Jin, et al., 3D PBV-Net: an automated prostate MRI data segmentation method,
Comput. Biol. Med. 128 (2021), 104160.
[22] D. Sacks, et al., Multisociety consensus quality improvement revised consensus
statement for endovascular therapy of acute ischemic stroke, Int. J. Stroke 13 (6)
(2018) 612–632.
[23] H. Li, et al., Human treelike tubular structure segmentation: a comprehensive
review and future perspectives, Comput. Biol. Med. 151 (2022), 106241.
[24] Y. Jin, T.M. Penning, Aldo-keto reductases and bioactivation/detoxication, Annu.
Rev. Pharmacol. Toxicol. 47 (2007) 263–292.
[25] T.M. Penning, Aldo-keto reductase regulation by the Nrf2 system: implications for
stress response, chemotherapy drug resistance, and carcinogenesis, Chem. Res.
Toxicol. 30 (1) (2017) 162–176.
[26] J. Liang, et al., AKR1C3 and its transcription factor HOXB4 are promising
diagnostic biomarkers for acute myocardial infarction, Front Cardiovasc. Med. 8
(2021), 694238.
[27] Q. Zhou, et al., A positive feedback loop of AKR1C3-mediated activation of NF-κB
and STAT3 facilitates proliferation and metastasis in hepatocellular carcinoma,
Cancer Res. 81 (5) (2021) 1361–1374.
[28] M. Gagliardi, et al., Aldo-keto reductases protect metastatic melanoma from ER
stress-independent ferroptosis, Cell Death Dis. 10 (12) (2019) 902.
[29] J.M. Jez, T.G. Flynn, T.M. Penning, A new nomenclature for the aldo-keto
reductase superfamily, Biochem. Pharmacol. 54 (6) (1997) 639–647.
[30] M. Maiorino, M. Conrad, F. Ursini, GPx4, lipid peroxidation, and cell death:
discoveries, rediscoveries, and open issues, Antioxidants Redox Signal. 29 (1)
(2018) 61–74.
[31] D. Giustarini, et al., Analysis of GSH and GSSG after derivatization with N-
ethylmaleimide, Nat. Protoc. 8 (9) (2013) 1660–1669.
[32] A. Cepelis, et al., Associations of asthma and asthma control with atrial fibrillation
risk: results from the nord-trøndelag health study (HUNT), JAMA Cardiol 3 (8)
(2018) 721–728.
[33] R. Dong, et al., Boosted kernel search: framework, analysis and case studies on the
economic emission dispatch problem, Knowl. Base Syst. 233 (2021), 107529.
[34] Y. Zhang, et al., Towards augmented kernel extreme learning models for
bankruptcy prediction: algorithmic behavior and comprehensive analysis,
Neurocomputing 430 (2021) 185–212.
[35] Y. Liu, et al., Simulated annealing-based dynamic step shuffled frog leaping
algorithm: optimal performance design and feature selection, Neurocomputing 503
(2022) 325–362.
[36] C. Zhao, et al., JAMSNet: a remote pulse extraction network based on joint
attention and multi-scale fusion, IEEE Trans. Circ. Syst. Video Technol. (2022) 1, 1.
[37] R.C. Deo, Machine learning in medicine, Circulation 132 (20) (2015) 1920–1930.
[38] S. Zhao, et al., Boosted crow search algorithm for handling multi-threshold image
problems with application to X-ray images of COVID-19, Expert Syst. Appl. 213
(2023), 119095.
[39] D. Liang, A.M. Minikes, X. Jiang, Ferroptosis at the intersection of lipid metabolism
and cellular signaling, Mol. Cell. 82 (12) (2022) 2215–2227.
[40] J. Zheng, M. Conrad, The metabolic underpinnings of ferroptosis, Cell Metabol. 32
(6) (2020) 920–937.
[41] K. Verma, et al., Potent and highly selective aldo-keto reductase 1C3 (AKR1C3)
inhibitors act as chemotherapeutic potentiators in acute myeloid leukemia and T-
Cell acute lymphoblastic leukemia, J. Med. Chem. 62 (7) (2019) 3590–3616.
[42] T.L. Rižner, T.M. Penning, Aldo-keto reductase 1C3-Assessment as a new target for
the treatment of endometriosis, Pharmacol. Res. 152 (2020), 104446.
[43] H.K. Reddel, et al., Global initiative for asthma strategy 2021: executive summary
and rationale for key changes, Am. J. Respir. Crit. Care Med. 205 (1) (2022) 17–35.
[44] Y.J. He, et al., Fenton reaction-independent ferroptosis therapy via glutathione and
iron redox couple sequentially triggered lipid peroxide generator, Biomaterials 241
(2020), 119911.
[45] B. Niu, et al., Application of glutathione depletion in cancer therapy: enhanced
ROS-based therapy, ferroptosis, and chemotherapy, Biomaterials 277 (2021),
121110.
[46] M.A. Badgley, et al., Cysteine depletion induces pancreatic tumor ferroptosis in
mice, Science 368 (6486) (2020) 85–89.
[47] T.M. Penning, P. Wangtrakuldee, R.J. Auchus, Structural and functional Biology of
aldo-keto reductase steroid-transforming enzymes, Endocr. Rev. 40 (2) (2019)
447–475.
[48] M.E. Burczynski, et al., The reactive oxygen species–and Michael acceptor-
inducible human aldo-keto reductase AKR1C1 reduces the alpha,beta-unsaturated
aldehyde 4-hydroxy-2-nonenal to 1,4-dihydroxy-2-nonene, J. Biol. Chem. 276 (4)
(2001) 2890–2897.
[49] J. Zheng, et al., Knockdown of AKR1C3 promoted sorafenib sensitivity through
inhibiting the phosphorylation of AKT in hepatocellular carcinoma, Front. Oncol.
12 (2022), 823491.
[50] Y. Liu, et al., Development of highly potent and specific AKR1C3 inhibitors to
restore the chemosensitivity of drug-resistant breast cancer, Eur. J. Med. Chem.
247 (2023), 115013.