Construction of a potential microRNA and messenger RNA regulatory network of acute lung injury in mice

Acute lung injury (ALI) is a life-threatening clinical condition associated with critically ill patients, and the construction of potential microRNA (miRNA) and messenger RNA (mRNA) regulatory networks will help to fully elucidate its underlying molecular mechanisms. First, we screened fifteen upregulated differentially expressed miRNAs (DE-miRNAs) and six downregulated DE-miRNAs from the Gene Expression Omnibus (GEO) database. Then, the predicted target genes of the upregulated and downregulated DE-miRNAs were identified from the miRNet database. Subsequently, differentially expressed mRNAs (DE-mRNAs) were identified from the GEO database and subjected to combined analysis with the predicted DE-miRNA target genes. Eleven target genes of the upregulated DE-miRNAs and one target gene of the downregulated DE-miRNAs were screened out. To further validate the prediction results, we randomly selected a dataset for subsequent analysis and found some accurate potential miRNA-mRNA regulatory axes, including mmu-mir-7b-5p-Gria1, mmu-mir-486a-5p-Shc4 and mmu-mir-486b-5p-Shc4 pairs. Finally, mir-7b and its target gene Gria1 and mir-486b and its target gene Shc4 were further validated in a bleomycin-induced ALI mouse model. We established a potential miRNA-mRNA regulatory network of ALI in mice, which may provide a basis for basic and clinical research on ALI and advance the available treatment options.

www.nature.com/scientificreports/ However, no miRNA and mRNA regulatory network of bleomycin-induced ALI in mice has been constructed.
In this study, we searched datasets of bleomycin-induced ALI in mice by accessing the network database. We first screened differentially expressed miRNAs (DE-miRNAs) in bleomycin-treated lung tissues compared with normal lung tissues in mice. Then, we predicted the potential target genes of the DE-miRNAs using network database resources. Next, differentially expressed mRNAs (DE-mRNAs) between bleomycin-treated lung tissues and normal lung tissues were obtained by analyzing the mRNA dataset. Subsequently, candidate target genes were identified, a protein-protein interaction (PPI) network was constructed, and Gene Ontology (GO) functional enrichment and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed. Finally, a potential miRNA-mRNA regulatory network was established, another dataset was used to detect the candidate target gene expression levels, and two relatively meaningful miRNA-mRNA pairs were experimentally verified. In summary, our findings reveal the potential comprehensive mechanisms of miRNA-mRNA regulatory axes in the pathogenesis of bleomycin-induced ALI and a potential ALI-related miRNA-mRNA regulatory network. The flowchart of our study is depicted in Fig. 1.
Prediction of potential DE-miRNA target genes. We used the miRNet database to predict the potential target genes of the DE-miRNAs, as miRNAs exert their biological effects mainly by directly targeting the 3' untranslated regions of mRNAs. The potential target genes for the upregulated DE-miRNAs included 1068 genes associated with 13 miRNAs (see Supplementary File 1A), and the potential target genes for the downregulated DE-miRNAs included 76 genes associated with 4 miRNAs (see Supplementary File 1B). The upregulated DE-miRNA-target gene network was established and is presented in Fig. 3A, and the downregulated DE-miRNAtarget gene network was established and is presented in Fig. 3B. Additionally, the degrees of target genes for the DE-miRNAs are listed in Table 1.
Searching mRNA datasets to identify DE-mRNAs. To improve the reliability of our subsequent analysis of the target genes of the screened DE-miRNAs, we searched GEO datasets focusing on mRNA expression. One dataset (GSE123808) was selected to screen DE-mRNAs between bleomycin-treated samples and control   Fig. 4.

Figure 2.
The DE-miRNAs between bleomycin-treated samples and control samples. A |log2FC|> 2 and P < 0.05 were set as the thresholds for identifying DE-miRNAs. The red and green dots represent the upregulated and downregulated miRNAs in bleomycin-treated samples, respectively; the black dots represent miRNAs that were not differentially expressed between the bleomycin-treated samples and control samples.  www.nature.com/scientificreports/ Identification of candidate target genes. It is widely acknowledged an inverse relationship exists between miRNAs and mRNA target genes. We conducted a combined analysis of 261 downregulated DE-mRNAs and 1068 predicted target genes of the upregulated DE-miRNAs, and 11 candidate target genes of the upregulated DE-miRNAs were further screened out (Fig. 5A, Table 2). We conducted a combined analysis of 287 upregulated DE-mRNAs and 76 predicted target genes of the downregulated DE-miRNAs, and 1 candidate target gene was further screened out (Fig. 5B, Table 3).
Construction of the PPI network. We mapped these candidate target genes into the STRING database, setting the research species as "Mus musculus", to construct the PPI network. When the lowest interaction score was set to 0.15, 7 candidate target genes of the DE-miRNAs in the network were predicted to have protein inter-

Candidate target genes Upregulated DE-miRNAs
Ddx3y mmu-mir-871-5p Sema4g mmu-mir-7b-5p Cadm2 mmu-mir-3092-3p Table 3. Candidate target genes of the downregulated DE-miRNAs. KEGG pathway enrichment analysis of the candidate target genes was then conducted. The candidate target genes were significantly enriched for the breast cancer, gastric cancer, hepatocellular carcinoma, nicotine addiction, neurodegeneration-multiple diseases, long-term depression, basal cell carcinoma, long-term potentiation, amphetamine addiction, and prolactin signaling pathways (see Supplementary File 3D). The top 20 KEGG pathways enriched ranked by their adj P values are shown in Supplementary Figure S2D.

Candidate target genes Downregulated DE-miRNAs
Identification of a potential miRNA-mRNA regulatory network. According to the miRNA and candidate target gene pairs analyzed above (Tables 2, and 3), we found a link between miRNAs and target genes, and the potential miRNA-mRNA (target gene) regulatory network related to the development of bleomycininduced ALI in mice was constructed as shown more intuitively in Fig. 6.

Searching the datasets and validating the candidate target gene expression levels.
To make the validation more credible, we randomly selected a dataset that met the inclusion criteria. Finally, GSE109913 was selected for subsequent analysis. The expression levels of five candidate target genes from the GSE109913 dataset were determined and are shown in Fig. 7. In the GSE109913 dataset, the expression level of Gria1 was significantly lower in bleomycin-treated ALI samples than in normal control samples (P < 0.05), and the expression level of Shc4 was significantly higher in bleomycin-treated ALI samples than in normal control samples (P < 0.05).
Analysis of target gene expression demonstrated the inhibitory effect of Gria1 and the promotional effect of Shc4 on ALI. Based on this preliminary validation, more accurate potential miRNA-mRNA regulatory axes contributing to ALI were established, including the mmu-mir-7b-5p-Gria1, mmu-mir-486a-5p-Shc4 and mmumir-486b-5p-Shc4 regulatory pathways, which could first be further studied in clinical and basic experiments.
Experimental validation of a bleomycin-induced ALI model. To further validate the prediction results, we constructed a bleomycin-induced ALI mouse model by intratracheally administering 5 mg/kg bleomycin. Hematoxylin and eosin (HE) staining of lung sections from the bleomycin-treated groups showed comprehensive features of morphological damage, such as congestion, hemorrhaging, thickening of the alveolar walls and infiltration of inflammatory cells, especially neutrophils, while no histological defects were observed in the phosphate-buffered saline (PBS) treated lungs (Fig. 8A). Compared with that of the control group, the lung injury score of the bleomycin-treated group was higher (Fig. 8B). www.nature.com/scientificreports/ Then, we explored the expression of miR-7b and its target gene Gria1 and miR-486b and its target gene Shc4 in lung tissues using real-time polymerase chain reaction (PCR) (Fig. 9). Consistent with the predicted results, the experimental validation showed that the expression of miR-7b was significantly upregulated and that of the Gria1 gene was downregulated in the ALI groups (P < 0.05). In addition, the expression of miR-486b was significantly downregulated and that of Shc4 was upregulated in the ALI groups (P < 0.05).

Discussion
ALI/ARDS is a life-threatening clinical condition associated with multiple symptoms and influenced by numerous factors 1,2 . Functional genomics approaches have provided novel insights into ALI/ARDS, a complex trait that requires both a severe environmental insult and an individual predisposition 15 . To date, the only study showing a link between the miRNA-mRNA regulatory network and ARDS was a study that induced ARDS in rats by using saline lavage and mechanical ventilation 11 .
As an antitumor antibiotic, bleomycin can form complexes with oxygen and iron to break DNA strands, resulting in the secretion of oxygen free radicals and cell apoptosis 16 . During the process of cell damage and apoptosis, a number of cytotoxic factors, such as reactive oxygen species (ROS) and nitrogen (NO) inflammatory factors, are generated in the lungs and can directly damage cells through lipid and protein oxidation 17 . Therefore, bleomycin has been widely used in animal studies to model pulmonary fibrosis and ALI/ARDS 13,14,18 . The bleomycin-induced ALI mouse model is widely applied because it is characterized by an inflammatory response and alveolar epithelia leading to excessive matrix deposition [12][13][14] . However, no microRNA-mRNA regulatory network of bleomycin-induced ALI in mice has been constructed.
In this study, we searched the GEO database and conducted differential expression analysis using miRNA and mRNA datasets. Finally, six upregulated DE-miRNAs and two downregulated DE-miRNAs were identified. Some of the screened DE-miRNAs were consistent with previous results. MiR-344 was also identified as  www.nature.com/scientificreports/ an upregulated miRNA in a rat model of ARDS that inversely correlated with the expression of their predicted targets, such as Aco2, Mdh1 and Eif2ak1 11 . MiR-7b was previously shown to be upregulated in the ALI/ARDS model 19,20 . Silencing the lncRNA MEG3 augments the binding of miR-7b to NLRP3 and downregulates NLRP3 expression, which ultimately improves endotoxin-induced ALI 21 . Although there are no direct reports of miR-298 in ALI/ARDS, miR-298 was predicted to bind with high affinity to the 5'UTR of the SARS-CoV-2 genome, and SARS-CoV-2 can cause ARDS 22 . MiR-298 was also identified as a potential regulator of the NOD-dependent Tnf-α and Il-6 mRNA levels in pulmonary endothelial cells, which represents the vital pathogenesis of ARDS 23 .
In an LPS-induced ALI mouse model, the miR-486-5p level was significantly higher than that in the controls 24 . However, in bleomycin-induced ALI, miR-486-5p was shown to be downregulated 25 . Hence, our findings provide a basis for the use of miRNAs as biomarkers or targets for miRNA-based pharmacological therapies for ALI. After integrating the DE-mRNAs and target genes of the DE-miRNAs, multiple candidate genes were screened, including 11 candidate target genes of the upregulated DE-miRNAs and 1 candidate target gene of the downregulated DE-miRNAs. Then, a PPI network was constructed to analyze the protein interactions of these target genes. In STRING, each PPI is annotated with one or more 'scores' . These scores are indicators of confidence. All scores rank from 0 to 1, with 1 being the highest possible confidence. There are two types of scores: the "normal" score, and the "transferred" score. The latter is computed from data that is not originally observed in the organism of interest, but instead in some other organism and then transferred via homology/orthology. In this study, we mainly studied miRNA-mRNA interaction. There is indeed little evidence for interactions between potentially related target genes (proteins). Our PPI network results show interactions between possible target genes (proteins) for future in-depth studies. We selected with different thresholds to establish confidence of PPI. When we chose a higher threshold, there were fewer confidence of PPI. For a more comprehensive analysis, we chose a lower threshold score value 0.15.
GO BP functional enrichment analysis showed that the candidate target genes were significantly enriched for membranous septum morphogenesis, synaptic vesicle fusion to the presynaptic active zone membrane, planar cell polarity involved in neural tube closure, stem cell differentiation, and regulation of the establishment of planar polarity involved in neural tube closure. GO MF functional enrichment analysis showed that the candidate target genes were significantly enriched for PDZ domain binding, Wnt-activated receptor activity, ionotropic glutamate receptor activity, syntaxin-1 binding, and frizzled binding. GO CC functional enrichment analysis showed that the candidate target genes were significantly enriched for the dendritic spine membrane, Wnt signalosome, postsynaptic membrane, AMPA glutamate receptor complex, and dendrite membrane. KEGG pathway enrichment analysis showed that candidate target genes were significantly enriched for pathways related to breast cancer, gastric cancer, hepatocellular carcinoma, nicotine addiction and multiple neurodegeneration diseases. Some of these functional enrichment and pathways are closely related to ALI/ARDS, and some of these genes have been identified to act as pivotal modulators. For example, Fzd1 expression was decreased in the lungs of rats with endotoxic shock, and decreased Fzd1 expression may hinder the sensitivity of Wnt3a/β-catenin signaling to regulate inflammatory responses 26 . Shc4 was shown to enhance intracellular antioxidant defense via the nuclear Figure 9. Experimental validation of a bleomycin-induced ALI model by real-time PCR. The expression of miR-7b was significantly upregulated and that of the Gria1 gene was downregulated in the ALI groups. The expression of miR-486b was significantly downregulated and that of the Shc4 gene was upregulated in the ALI groups. n = 6 per group. *P < 0.05. www.nature.com/scientificreports/ factor erythroid 2-related factor 2 (Nrf2)/heme oxygenase-1 (HO-1) signaling pathway, which was associated with the oxidative stress response in ALI 27 . The upregulated miRNA and downregulated mRNA regulatory network constructed herein included mir-344e-3p-Ugt2b35/Fzd1, mir-7b-5p-Stk11ip/Gria1/Cplx1/Necab1/Sema4g, mir-298-5p-Ces1g/Senp5, mir-881-3p-Senp5, mir-871-5p-Ddx3y and mir-3092-3p-Cadm2, and the downregulated miRNA and upregulated mRNA regulatory network included mir-486a-5p-Shc4 and mir-486b-5p-Shc4. There are still relatively few reports on these regulatory networks, and ALI-related research has been particularly limited. As a result, these miRNAs and target genes can be combined to perform in-depth studies and thereby identify potential targets for the treatment of related diseases. Thus, further research on this potential ALI-related miRNA-mRNA regulatory network is warranted to verify the relevant mechanism. Almost none of these miRNA-mRNA pairs in the network potentially contributing to the pathogenesis of ALI have been studied, which is of importance for exploring and developing novel mechanisms and therapeutic targets.

Scientific Reports
To enhance the applicability of our data, we first used datasets including bleomycin-treated samples and control samples to further select suitable pathways to study. In the GSE109913 dataset, the expression levels of Gria1 were significantly lower in bleomycin-induced ALI tissues than in normal tissues, and the expression levels of Shc4 were significantly higher in bleomycin-induced ALI tissues than in normal tissues. Then, we constructed a bleomycin-induced ALI mouse model, which was confirmed by the HE staining of lung sections. Furthermore, we explored the expression levels of miR-7b and its target gene Gria1 and of miR-486b and its target gene Shc4 in lung tissues by real-time PCR. Fortunately, the experimental validation showed that the expression of miR-7b was significantly upregulated while that of the Gria1 gene was downregulated in the ALI groups; the expression of miR-486b was significantly downregulated and that of the Shc4 gene upregulated was in the ALI groups. Although miR-7b was upregulated in the ALI/ARDS model 19,20 , the predicted target genes of miR-7b are IRS2, OXR1, GSK3B, and NFAT5. Here, we identified a new miRNA-mRNA regulatory pathway (miR-7b/Gria1), which was preliminarily verified in a bleomycin-induced ALI mouse model. miR-486b and Shc4 have been shown to be related to oxidative stress, but the miR-486b/Shc4 pathway has not yet been confirmed. Therefore, these generelated miRNA-mRNA regulatory pathways should be further studied in basic experiments.
Although a potential miRNA-mRNA regulatory network was constructed in this study, there are still some limitations. First, we utilized only one miRNA dataset and one mRNA dataset, and the number and sample sizes of the datasets included in this study were small. Second, we screened out DE-miRNAs and DE-mRNAs from a web database with data from multiple sources to avoid the limitations of a single-center study as much as possible, but a single study to validate and screen the constructed regulatory network is still needed. It is best to verify both miRNAs and mRNAs in the same set of samples. We used only the GSE109913 dataset to preliminarily validate gene expression. On this basis, our next studies will further validate and further explore the underlying mechanisms to find effective interventions to target the established regulatory network. Finally, as we further validated the gene expression and regulatory network, we explored the expression of miR-7b and its target gene Gria1 and miR-486b and its target gene Shc4 in lung tissues using real-time PCR. Further studies on the other miRNAs and target genes are needed in the future.
In conclusion, we herein reveal a potential comprehensive mechanism of miRNA-mRNA regulatory axes in the pathogenesis of bleomycin-induced ALI and establish a potential ALI-related miRNA-mRNA regulatory network, which may provide a basis for basic and clinical research on ALI and advance its treatment.

Materials and methods
Searching and screening of datasets. We searched datasets focusing on miRNAs, mRNAs and genes in the GEO dataset (https:// www. ncbi. nlm. nih. gov/ gds/). Taking miRNA expression as an example, the retrieval strategy was as follows: (("micrornas"[MeSH Terms] OR microRNA [All Fields]) AND ("bleomycin"[MeSH Terms] OR bleomycin [All Fields])) AND "Mus musculus" [porgn]. We included datasets based on bleomycininduced mice and datasets containing bleomycin-treated lung tissue samples and control lung tissue samples. One dataset (GSE147138) met the inclusion criteria mentioned above and was selected for subsequent analysis. Prediction of potential target genes of DE-miRNAs. An integrated platform linking miRNAs, their targets and their functions named miRNet (https:// www. mirnet. ca/) was used to predict the downstream target genes of the screened DE-miRNAs [29][30][31] . The screened upregulated and downregulated DE-miRNAs were entered into the web platform, and the data of the potential target genes of the upregulated and downregulated DE-miRNAs were downloaded. Then, these data were input into Cytoscape 3.6.0 software to access the DE-miRNAtarget gene network 32 . Using the "Network Analyzer" tools of the software, the data were subjected to topology analysis, and the degrees of target genes for the DE-miRNAs were finally identified.
Identification of DE-mRNAs and candidate target genes. GSE123808  www.nature.com/scientificreports/ File 4. Setting |log2FC|> 2 and P value < 0.05 as the thresholds, DE-mRNAs were identified using the RGUI and limma packages 28 . Then, we analyzed the DE-mRNAs and predicted target genes of DE-miRNAs in combination, and candidate target genes were further screened.
Construction of the PPI network. The candidate target genes were introduced into the STRING database (https:// string-db. org/). STRING is an ELIXIR Core Data web server that retrieves and displays repeatedly occurring gene neighborhoods [33][34][35] . After adding the candidate target genes into the database, a PPI network was constructed. The research species was defined as "Mus musculus", the lowest interaction score was set to 0.15, and the remaining parameters were set to the default settings. Nodes represented target genes, and edges represented the interactions between the target genes in the PPI network.
GO function and KEGG pathway enrichment analyses. The RGUI 4.0.3 and org.Hs.eg.db packages were applied to obtain the entrezIDs of the candidate target genes. RGUI and the clusterProfiler package were used to perform GO function enrichment analysis, which included BP, MF and CC, as well as KEGG pathway enrichment analysis 36,37 .
Identification of a potential miRNA-mRNA regulatory network and validation of target gene expression levels. According to the miRNA and candidate target gene pairs analyzed, we established a link between miRNAs and candidate target genes to identify a potential miRNA-mRNA regulatory network. Subsequently, the GEO dataset was used to detect the candidate target gene expression levels. We searched GEO datasets focusing on gene expression and included those based on bleomycin-treated samples and control samples. To make the validation more credible, we randomly selected a dataset that met the inclusion criteria, and GSE109913, which was based on the GPL21103 Illumina HiSeq 4000 platform (Mus musculus), was selected for subsequent analysis. Basic information about GSE109913 is provided in Supplementary File 4. We downloaded gene expression data from the GEO dataset and accessed candidate target gene expression data to perform statistical analysis (Supplementary File 6). The expression levels of target genes in the regulatory network were further validated by analyzing the gene expression data downloaded from the GEO dataset. P < 0.05 was considered statistically significant.
Animal experiments. Six-to eight-week-old male C57BL/6 wild-type mice (Shanghai Laboratory Animal Center, Chinese Academy of Sciences, Shanghai, China) were maintained in a controlled environment and provided water and standard rodent food. The mice were anesthetized with sodium pentobarbital (60 mg/kg) and then administered bleomycin (BLM, Sigma-Aldrich Co. LLC., USA) dissolved in PBS via a single intratracheal instillation at a dose of 5 mg/kg body weight in a volume of 50 μl to induce ALI; mice in the control group received an equal volume of PBS. The mice were anesthetized and sacrificed on day 7 after the bleomycin or PBS injection. The animal experiments were approved by the Animal Ethics Committee of Nantong University on the Use and Care of Animals and were performed in accordance with the committee's guidelines (ethical approval number, S20210304-019). This study also adhered to the ARRIVE guidelines (https:// arriv eguid elines. org).
Histopathological examination. Histological analysis of the left lung was performed. Briefly, immediately after euthanasia, the left lung tissues were collected and fixed in 10% formalin for 24 h, embedded in paraffin, sliced into 5-μm thick sections, and stained with hematoxylin and eosin (HE) for the detection of pathological changes in the lung tissues. Lung injury scores were utilized to evaluate BLM-induced lung injury based on HE images and Matute-Bello's published criteria in a blinded manner 38 .
Statistical analysis. Some statistical analyses were automatically performed by the bioinformatic tools on the web platforms mentioned above. We used a series of matrix files downloaded from the GEO dataset analyzed with the RGUI 4.0.3 and the limma packages to identify DE-miRNAs and DE-mRNAs. Only miRNAs and mRNAs with a |log2FC|> 2 and P < 0.05 were considered statistically significant. We used the RGUI 4.0.3 and the org.Hs.eg.db packages to obtain the entrezIDs of the candidate target genes. We used the RGUI and clusterProfiler packages to perform GO functional enrichment analysis, and adjusted P < 0.05 was considered statistically significant. The data of target gene expression levels in GSE109913 dataset, the lung injury scores and miRNA-mRNA pairs expression levels in mouse models were analyzed using IBM SPSS Statistics 25 software and GraphPad Prism 8.0.2 software. Student's t test or Welch's t test was used to compare two groups. If the data were not normally distributed, the Mann-Whitney U test was used. P < 0.05 was considered significant. www.nature.com/scientificreports/

Data availability
The datasets generated and/or analyzed during the current study are available from the corresponding author upon reasonable request. All data generated or analyzed during this study are included in this published article (and its Supplementary Information files).