Analyses of the Competing Endogenous RNA Network in Rats with Middle Cerebral Artery Occlusion Based on lncRNA, miRNA, and mRNA Expression

Ischemic stroke, a prevalent neurological disorder, leads to permanent brain damage, long-term disability, or death^1,2,3,4. MicroRNAs (miRNAs or miR), long noncoding RNAs (lncRNAs), and messenger RNAs (mRNA) form RNA-mediated regulatory networks^5,6,7,8. These molecules regulate cellular functions through various complex mechanisms^9,10are increasingly recognized as contributors to the pathophysiology of ischemic stroke. However, despite their growing association with the condition, the precise roles of these noncoding RNAs remain unclear. Further investigation of novel noncoding RNAs is essential for clarifying the molecular mechanisms underlying ischemic stroke.

LncRNAs act as key regulators of gene expression during the initiation and progression of various pathological conditions. Serving as competing endogenous RNAs (ceRNAs), lncRNAs bind to miRNAs to exert specific regulatory effects¹¹. Wei et al. report that lncRNA AK038897 acts as a ceRNA by targeting miR-26a-5p, thereby modulating death-associated protein kinase 1 (DAPK1) to exacerbate cerebral ischemia-reperfusion injury¹². Studies show that long noncoding RNA SNHG1 (lncRNA SNHG1) functions as a ceRNA to regulate cerebrovascular diseases by modulating the HIF-1α/VEGF signaling pathway through interaction with miR-18a¹³. Additional studies show that the long noncoding RNA maternally expressed gene 3 (lncRNA MEG3) modulates neuronal apoptosis through the miR-21/PDCD4 signaling cascade¹⁴. While numerous lncRNAs, miRNAs, and mRNAs have been identified through high-throughput sequencing or microarray analysis, the functions of these molecules in stroke remain unclear, and the systematic establishment of RNA-mediated regulatory networks is urgently needed.

Therefore, this study aims to construct a comprehensive lncRNA-miRNA-mRNA network based on the ceRNA hypothesis using previously collected data, and to analyze selected ceRNA subnetworks through Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment to better understand lncRNA functions. The results could reveal that several lncRNAs function as ceRNAs, potentially regulating specific miRNAs and their target mRNAs. GO enrichment could highlight key biological processes, such as cellular development, signal transduction, and cell cycle regulation, while pathway enrichment could reveal their involvement in stroke-related pathways, immune responses, and metabolism. These findings could highlight the potential roles of lncRNAs in diverse cellular processes and disease mechanisms.

Several key improvements are offered over existing ceRNA network analysis methods in ischemic stroke research, as supported by comparisons with related studies: first, integrate multi-layered validation by combining network construction with functional pathway verification; unlike Li et al.¹⁵, which focuses primarily on bioinformatic profiling of immune-related ceRNA networks using public transcriptome data, construct a ceRNA network comprising 334 lncRNAs, miRNAs, and mRNAs and validate critical signaling pathways (calcium signaling, gap junction signaling, and neuroactive ligand-receptor interaction) through Western blot analysis, addressing the limitation of over-reliance on in silico predictions as seen in Fan et al.¹⁶, which constructs a circRNA-associated ceRNA network but lacks experimental validation of downstream pathways, and confirm the functional relevance of the predicted network by quantifying protein expression (e.g., CaMKII, CX36, GRIA3) in both model and sham group; focus on core lncRNAs with high topological importance, enhancing the specificity of findings; unlike Cheng et al.¹⁷, which constructs a lncRNA-miRNA-mRNA ceRNA network with 3 lncRNAs, 2 miRNAs, and 24 mRNAs but does not prioritize key regulators, identify 3 lncRNAs with high degree centrality to build a representative subnetwork, ensuring mechanistic insights are anchored to the most influential nodes and improving interpretability compared to broader, less focused networks like those described in Li et al.¹⁸, which includes 62 lncRNAs but lacks targeted subnetwork analysis; provide detailed experimental parameters to improve reproducibility: use 500 ng-1 µg of total RNA extracted from brain tissues (6 rats per group: 20 MCAO models and 20 sham-operated controls) for library preparation, with RNA integrity number (RIN) > 8.0 to ensure quality, a detail not specified in Wang et al.¹⁹, which validates circRNA biomarkers but omits RNA input specifications; employ male SPF SD rats (220 ± 30 g) with MCAO, a widely used model, but explicitly note its limitations, it does not fully replicate human ischemic stroke's vascular complexity or immune responses as highlighted in Li et al. -- and mitigate this by stratifying rats by neurological scores (1-3) to standardize infarct severity; acknowledge study limitations: the ceRNA network may not capture post-translational modifications, and the sample size for Western blot validation (n = 6 per group) could be expanded, with future work to include larger cohorts and proteomic analyses to address these gaps.

The animal experiments were approved by the Experimental Ethics Committee of Anhui University of Chinese Medicine (license number: AHUCM-rats-2024006) and conducted in accordance with institutional guidelines and JoVE's animal use standards. Forty male SPF-grade SD rats (220 g ± 30 g) were used in this study. The reagents and equipment used are listed in the Table of Materials.

1. Experimental animals

1. House the animals in the Animal Experimental Center under standard experimental conditions. Maintain a 12 h light/dark cycle, change bedding daily, and provide free access to food and water.
2. Euthanize the animals at the end of the experiment by administering 150 mg/kg sodium pentobarbital via intraperitoneal injection to induce a deep anesthetic state (following institutionally approved protocols). Confirm deep anesthesia by the absence of reflexive response to paw pinch, monitor respiration to ensure complete loss of vital signs, and then proceed with decapitation.

2. Induction of the middle cerebral artery occlusion (MCAO) model in rats

1. Induce anesthesia by administering an intraperitoneal injection of 1% pentobarbital sodium at a dose of 40 mg/kg²⁰(following institutionally approved protocols). Evaluate anesthetic depth using the toe pinch and corneal reflexes.
   1. Continuously monitor respiratory and heart rate throughout the procedure. Apply sterile mineral oil-based eye ointment immediately after anesthesia induction for eye protection²¹.
2. Once fully anesthetized, make a midline neck incision and carefully expose the right common carotid artery (CCA), internal carotid artery (ICA), and external carotid artery (ECA). Under direct visualization, use appropriate microsurgical instruments (e.g., forceps and scissors) to gently separate the surrounding connective tissue and fully expose each artery.
   1. After exposing the arteries, ligate the ECA and CCA using surgical sutures. Tie secure surgical knots to occlude blood flow and minimize bleeding during subsequent procedures. Ensure proper tension to create a stable field for suture insertion.
   2. Make a small incision at the bifurcation of the ECA and ICA using ophthalmic scissors to prepare for suture insertion. Select a nylon suture with a diameter of 0.25 mm and a length of 18 mm from the tip of the rounded (ball-shaped) end. Insert the suture through the incision site at the ECA-ICA junction and gently advance it along the ICA to a depth of approximately 18.5 mm ± 0.5 mm to occlude the middle cerebral artery and induce focal cerebral ischemia.
   3. Secure the inserted nylon suture to the surrounding tissue or blood vessels using surgical sutures to prevent displacement. After maintaining MCAO for 2 h, induce ischemia/reperfusion injury. Use fine surgical forceps to gently grasp the nylon suture and slowly withdraw approximately 5 mm to partially restore cerebral blood flow and induce reperfusion injury. In the sham group, ligate only the carotid artery without inserting a thread embolus²².
3. One week after the successful establishment of the MCAO model, deeply anesthetize the rats. Once there is no response to paw pinch, quickly decapitate and harvest the brains. Rinse the brain tissues with normal saline, then store them in a −80 °C refrigerator for cryopreservation²³.
4. Neurological function score
   1. Evaluate neurological function using the Longa scoring method to determine the success of the MCAO model²⁴. At 24 h post-reperfusion, evaluate neurological function in each rat and assign a score based on the following criteria: Score 0: no observable neurological deficit; Score 1: flexion of the left forelimb during tail suspension, forelimb not fully extended; Score 2: flexion of the left forelimb during tail suspension with markedly reduced resistance on the affected (left) side when placed on a flat surface; Score 3: same as score 2, with added circling or involuntary turning movements while crawling; Score 4: unconsciousness or death within 24 h.
      NOTE: A higher score indicates a more severe behavioral disorder and greater neurological impairment.
5. Calculate cerebral infarction volume using 2,3,5-triphenyltetrazolium chloride (TTC) staining
   1. Section each ex vivo rat brain along the coronal plane into five slices of equal thickness. Ensure each slice is precisely 2.0 mm ± 0.1 mm thick. Fully immerse the sections in freshly prepared 2% (w/v) TTC staining solution.
      1. Wrap the staining container in aluminium foil to protect the solution from light and incubate the samples in a constant-temperature incubator at 37 °C ± 0.5 °C for 30 min. After staining, observe rose red staining in normal brain tissue, while ischemic infarcted areas remain unstained (white) due to loss of dehydrogenase activity.
   2. Transfer TTC-stained brain slices into a 4% (w/v) paraformaldehyde solution and fix at 4 °C for 24 h to preserve tissue morphology. After fixation, capture digital images of the slices using a high-resolution imaging system. Record and archive the data 24 h post-fixation.
   3. Use ImageJ software to quantify infarct volume and total brain tissue volume for each rat group following TTC staining. Calculate the corrected infarct volume percentage using the following formula²⁵:
      ​Corrected infarct volume (%) = {[total lesion volume − (left hemisphere volume − right hemisphere volume)] / right hemisphere volume} × 100%.
   4. Use GraphPad Prism software to visualize and statistically analyze corrected infarct volumes via an unpaired t-test. Define statistical significance as P < 0.05.

3. RNA high-throughput sequencing

1. Select brain tissues from three rats in each group. Remove ribosomal RNA (rRNA) from the total RNA samples using the Ribo-Zero rRNA Removal Kit according to the manufacturer's instructions.
   1. Construct complementary DNA (cDNA) libraries using the TruSeq RNA Library Prep Kit following the standard protocol. Use purified cDNA libraries for cluster generation and sequencing.
2. Use the core data previously generated²⁶, which provided all sequencing libraries and datasets for this study, without reliance on newly constructed libraries. Focus the current work on the integrated fusion analysis of these existing datasets without generating new RNA-seq libraries or sequencing data.
   1. Confirm approval from relevant institutional ethics committees for the previous study. Conduct high-throughput sequencing to systematically assess the miRNA expression profile in rats with MCAO, adhering to standardized protocols as previously described²⁷.
      NOTE: Quality control data for RNA-sequencing library is provided in Supplementary File 1.

4. Differential expression analysis of long noncoding RNAs, microRNAs, and messenger RNAs

1. Obtain annotations for protein-coding genes, miRNAs, and lncRNAs from Ensembl (https://www.ensembl.org/index.html), miRBase (https://www.mirbase.org/), and NONCODE (http://www.noncode.org/), as appropriate^28,29,30. Quantify gene expression using StringTie³¹ (https://ccb.jhu.edu/software/stringtie/). 
   1. Assemble the aligned reads from each sample into transcriptomes using the StringTie assembler. Use the StringTie merge module to integrate the transcriptome assemblies from all samples, generating a unified set of transcripts consistent across the cohort. Use this merged transcriptome to estimate transcript abundance³².
   2. Perform quantification of protein-coding genes and lncRNAs by applying Trimmed Mean of M-values (TMM) normalization and calculating Fragments Per Kilobase of transcript per Million mapped reads (FPKM) for each transcript^33,34.
2. Perform differential expression analysis using the edgeR package (https://bioconductor.org/) to identify differentially expressed lncRNAs (DELs), miRNAs (DEMis), and mRNAs (DEMs) between the MCAO and control groups. Convert the filtered raw count matrix into an edgeR DGEList object. Estimate the common dispersion and assign the same dispersion value to each gene or miRNA. 
   1. Perform an exact test to compare gene expression between groups. Retrieve the log2 fold change (log2FC), log counts per million (logCPM), P-values, and false discovery rate (FDR)-adjusted P-values for each differentially expressed gene³⁵. Define statistical significance as P < 0.05 with a fold-change threshold of |log2(FC)| > 1.

5. Prediction of long noncoding RNA and messenger RNA targets of differentially expressed microRNAs

1. Retrieve the sequences of DEMis and DELs from the appropriate databases. Use the miRNA-lncRNA interaction prediction tools miRanda³⁶ and RNAhybrid³⁷ to identify potential interactions between DEMis and DELs. Obtain miRNA-mRNA interaction pairs from the miRWalk 3.0 database (http://mirwalk.umm.uni-heidelberg.de/)³⁸

6. Construction of the long noncoding RNA - microRNA - messenger RNA regulatory network

1. Construct the lncRNA-miRNA-mRNA regulatory network based on the ceRNA hypothesis^39,40.
   1. Calculate the Pearson correlation coefficient (PCC) between DELs and DEMs. Select lncRNA-mRNA pairs for further analysis if PCC > 0.99 and P < 0.01.
   2. For each co-expressed lncRNA-mRNA pair, identify shared miRNAs targeting both transcripts. Use these shared miRNAs to construct the miRNA-mRNA-lncRNA co-expression network.
   3. Use Cytoscape (https://cytoscape.org/) to construct and visualize the lncRNA-miRNA-mRNA interaction network^41,42. Follow these steps:
      1. Use the CytoNCA plugin to calculate betweenness centrality (BC), closeness centrality (CC), and degree centrality (DC) for each node.
      2. Select genes with centrality scores exceeding the median values of all three metrics to construct the final key subnetwork.
      3. Use the CytoHubba plugin to identify the top 10 hub genes in the protein-protein interaction network and extract key nodes identified in Step 2.
      4. Calculate the node degree for each component of the ceRNA network.

7. Functional enrichment analyses

1. Perform GO and KEGG pathway enrichment analyses to further investigate RNA function⁴³. Use the clusterProfiler package (https://bioconductor.org/) to visualize the enrichment results⁴⁴.

8. Western blot analysis to detect the expression levels of key proteins involved in lncRNA - ceRNA-related pathways

1. Under deep anesthesia, quickly remove the entire brain. After carefully removing the meninges and blood vessels, isolate the cortical tissue in the infarcted area of the rat⁴⁵. Use fine scissors to cut the tissue into small pieces, ensuring each piece weighs between 50 mg and 100 mg. Transfer the weighed tissue samples into pre-cooled homogenization tubes.
   1. Prepare the protein lysis buffer at a volume ratio of PMSF to RIPA lysis buffer of 1:100 (final concentration 1 mM PMSF). Add the lysis buffer at a ratio of 200 µL per 20 mg of tissue. Homogenize the samples thoroughly using a heat-sterilized glass homogenizer until the tissue is finely broken down.
   2. Incubate the homogenates on ice for 30 min, gently agitating every 10 min to facilitate complete lysis. Centrifuge the samples at 28,656.6 × g for 15 min at 4 °C. Carefully aspirate the supernatant containing total protein without disturbing the pellet.
2. Add 5× sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) loading buffer to the protein sample at a ratio of 1:4 (1 volume of 5× buffer to 4 volumes of protein sample). After dilution, the final working concentration of the SDS-PAGE loading buffer is 1×.
   1. Mix thoroughly and heat in a boiling water bath for 10 min to completely denature the proteins. Allow the samples to cool to room temperature. Load 30 µg of each sample into the SDS-PAGE gel wells.
   2. Perform electrophoresis in two stages: apply 80 V for 30 min to resolve proteins through the stacking gel, then increase to 120 V for 1 h to separate proteins in the resolving gel.
3. Prepare pre-cut filter paper and polyvinylidene difluoride (PVDF) membranes to match the size of the gel. Soak the PVDF membrane in methanol for 3 min to activate it. Immerse both the PVDF membrane and filter paper in the transfer buffer for 5 min.
   1. Assemble the transfer sandwich in the following order (top to bottom): anode plate, three layers of filter paper, PVDF membrane, gel, three layers of filter paper, cathode plate. Ensure proper alignment of the gel, membrane, and filter paper layers. Remove all air bubbles at each step to ensure even transfer.
4. Immediately after protein transfer, rinse the PVDF membrane in Western washing buffer for 5 min to remove residual transfer buffer. Block the membrane at room temperature for 2 h with gentle shaking in Western blocking buffer containing 5% skimmed milk powder, or use 5% BSA if detecting phosphorylated proteins.
5. After blocking, incubate the membrane with the primary antibody (1:1000) at 4 °C overnight with gentle shaking. The following day, incubate the membrane with the secondary antibody (1:10,000) for 2 h at room temperature.
6. Perform chemiluminescent detection in a dark room to prevent light interference. Place the protein side of the PVDF membrane flat at the center of the exposure plate in the automatic imaging system.
   1. Mix equal volumes of ECL reagent A and ECL reagent B in a clean centrifuge tube immediately before use. Apply the mixed ECL solution evenly to the membrane surface to initiate the luminescent reaction.
   2. After 1-2 min of incubation at room temperature, gently remove excess liquid without disturbing the membrane surface. Adjust the exposure parameters on the imaging system according to luminescence intensity to optimize signal capture.

9. Statistical analysis

1. Analyze Western blot results using ImageJ software to quantify band gray values. Use GraphPad Prism for statistical analysis of experimental data. Express all data as mean ± standard deviation (x ± SD). Evaluate statistical differences between groups using an unpaired t-test.
   1. Enter data from the two independent sample groups into the software's data table, with each group corresponding to a separate column and the samples independent of each other. Verify normality by selecting "Normality and symmetry tests" under the "Analyze" function and applying the Shapiro-Wilk test to both groups.
   2. If P > 0.05 for both groups, consider the data normally distributed. On this basis, select Unpaired t-test under t-tests in the Analyze function to perform the analysis. Define statistical significance as P < 0.05.

The success rate verification experimental results of each group of rat models are shown in Figure 1. After modeling, neurological function scoring was performed on the rats (Figure 1A). The model group exhibited severe neurological deficits, and those with scores ranging from 1 to 3 were included in subsequent experiments. Figure 1B and C show the TTC staining results, with red representing normal brain tissue and white indicating the infarcted area. The model group exhibited severe cerebral infarction. These findings confirm the successful establishment of the ischemic model and validate its use for subsequent experiments. Differential expression analysis using edgeR revealed 920 DEMs and 710 DELs in the MCAO group. Additionally, 16 miRNAs exhibited aberrant expression patterns.

LncRNAs can act as ceRNAs by competitively binding to miRNAs. Overall, 717 lncRNA-miR interaction pairs were identified, comprising 16 miRNAs and 121 lncRNAs (Figure 2). miRWalk (http://mirwalk.umm.uni-heidelberg.de/) was used to predict miRNA–mRNA interactions, resulting in 7,813 candidate pairs. Integrative analysis with DEMs revealed 587 negatively correlated pairs, comprising 334 mRNAs and 16 miRNAs (Figure 3).

To further clarify these interactions, Cytoscape was used to integrate and visualize miRNA-mRNA and miRNA-lncRNA pairs, resulting in the construction of an lncRNA-miRNA-mRNA network (Figure 4). This network represents the interactions among DELs, DEMIs, and DEMs, comprising 240 DELs, 16 DEMIs, 334 DEMs, and 1,304 edges. To gain deeper insights, the degree distribution of nodes in the ceRNA network was analyzed using Cytoscape, revealing key connectivity metrics for RNA species in rats exposed to MCAO. Table 1 and Figure 5 present these metrics, highlighting the central role of specific RNAs in the ceRNA network.

For enrichment analysis, 334 genes were analyzed using the clusterProfiler package. In the GO Biological Processes (BP) category, the top five significantly enriched terms included "leukocyte cell-cell adhesion" (GO:0007159), "regulation of integrin-mediated signaling pathway" (GO:0033628), "collagen fiber organization" (GO:0030199), "integrin-mediated cell adhesion" (GO:0033627), and "positive regulation of calcium-mediated signaling" (GO:0050850). In the Cellular Component (CC) category, the top five enriched terms were "integrin complex" (GO:0008305), "condensed chromosome, centromeric region" (GO:0000777), "condensed chromosome, centromere region" (GO:0000779), "centromere" (GO:0000776), and "transport vesicle" (GO:0030133). For GO molecular function (MF) enrichment, the top five functions included "oligosaccharide binding" (GO:0070492), "hyaluronic acid binding" (GO:0005540), "protein kinase activator activity" (GO:0070492), "signal transducer activity" (GO:0030295), and "kinase activator activity" (GO:0019209) (Figure 6A). Additionally, KEGG pathway enrichment analysis revealed several significantly enriched pathways, including "cell adhesion molecules" (rno04514), "long-term depression" (rno04730), "p53 signaling pathway" (rno04115), "hematopoietic cell lineage" (rno04640), and "complement and coagulation cascades" (rno04610) (Figure 6B).

Further computational analysis revealed primary interactions between lncRNAs and miRNAs, and secondary interactions between miRNAs and mRNAs (Table 2). Six lncRNAs demonstrated a higher degree of connectivity, resulting in an increased number of lncRNA-miRNA and miRNA-mRNA interaction pairs. This finding suggests that these lncRNAs potentially play critical roles in the initiation and progression of MCAO, making them promising candidates for further investigation. Among these, rno-miR-653-3p exhibited the highest nodal degree. Network analysis revealed three crucial lncRNAs with total node degrees >60, along with their associated mRNAs and miRNAs. Sub-networks were reconstructed to highlight these interactions. The NONRATT006800.2-miRNA-mRNA network includes 1 DEL, 5 DEMis, 40 DEMs, and 61 edges (Figure 7A). Similarly, the NONRATT018501.2-miRNA-mRNA network comprises 1 DEL, 5 DEMis, 43 DEMs, and 62 edges (Figure 8A). The lncRNA NONRATT023334.2 formed a complex interaction network comprising 1 DEL, 8 DEMis, 43 DEMs, and 75 edges (Figure 9A). Enrichment analysis revealed that the NONRATT023334.2-miRNA-mRNA subnetwork is associated with 86 enriched biological functions and 5 signaling pathways. Similarly, the NONRATT018501.2-miRNA-mRNA subnetwork is associated with 81 enriched functions and 4 signaling pathways. The NONRATT006800.2-miRNA-mRNA subnetwork is enriched in 104 biological functional and 4 signaling pathways. Figure 7B, Figure 8B, and Figure 9B show the top 30 GO enrichment results, while Table 3 summarizes the enriched signaling pathways.

To further validate these findings, Western blot analysis was performed to assess the involvement of the calcium signaling pathway, gap junction communication, and neuroactive ligand-receptor interaction pathways. The results confirmed the functional significance of these pathways (Figure 10). Compared to the sham group, the model group exhibited a significant decrease in the expression of key proteins CAMKII and calmodulin (P < 0.01), along with a significant increase in calcineurin (CaN) levels (P < 0.01). The expression levels of gap junction pathway-related proteins, including PKC, CX36, and CX43, were significantly reduced (P < 0.01). Similarly, proteins associated with the neuroactive ligand-receptor interaction pathway, including GRIA3, GABRA6, and NPY1R, exhibited a statistically significant decrease in expression (P < 0.01).

Figure 1: Model success rate verification experiment. (A) Neurological function score (n = 20); (B) TTC Staining (n = 3); (C) Quantification of infarct volume. Data are presented as mean ± SD, **P < 0.01, model group vs sham group. Please click here to view a larger version of this figure.

Figure 2: lncRNA-miRNA interaction network in brain tissues from rats following MCAO. Hexagons represent lncRNAs, squares represent miRNAs, and edges indicate predicted targeting interactions between lncRNAs and miRNAs, predicted using MiRanda and RNAhybrid (P < 0.05). Node color indicates statistical significance, with red indicating higher significance. Please click here to view a larger version of this figure.

Figure 3: miRNA-mRNA interaction network in brain tissues from rats following MCAO. Circles represent mRNAs, and squares represent miRNAs. Edges indicate predicted negative regulatory interactions between miRNAs and mRNAs, based on miRWalk. Node color indicates statistical significance, with red indicating higher significance. Please click here to view a larger version of this figure.

Figure 4: ceRNA network comprising lncRNAs, miRNAs, and mRNAs in brain tissue samples from rats following MCAO. Hexagons represent lncRNAs, circles represent mRNAs, and squares represent miRNAs. Edges indicate predicted lncRNA-miRNA and miRNA-mRNA interactions. Overall correlations among these elements were further examined. Please click here to view a larger version of this figure.

Figure 5: Node degree distribution analysis illustrating key characteristics of the lncRNA-miRNA-mRNA interaction network. The X-axis represents the degree (i.e., the number of connections per RNA node). The Y-axis represents the number of RNA molecules corresponding to each degree. This distribution highlights RNAs with connectivity within the ceRNA network. Please click here to view a larger version of this figure.

Figure 6: GO and KEGG enrichment analyses of genes involved in the ceRNA network. (A) Top 30 GO enrichment terms. (B) Top 30 KEGG enrichment pathways. Please click here to view a larger version of this figure.

Figure 7: Subnetwork and functional enrichment analysis of lncRNA NONRATT006800.2. (A) Network diagram showing lncRNA (hexagon), miRNAs (squares), and mRNAs (circles). (B) Top 30 GO enrichment terms associated with the NONRATT006800.2-centered network. Please click here to view a larger version of this figure.

Figure 8: Subnetwork and functional enrichment analysis of lncRNA NONRATT018501.2. (A) Network diagram showing interactions among lncRNA (hexagon), miRNAs (squares), and mRNAs (circles). (B) Top 30 GO enrichment terms associated with the NONRATT018501.2-centered network. Please click here to view a larger version of this figure.

Figure 9: Subnetwork and functional enrichment analysis of lncRNA NONRATT023334.2. (A) Network visualization showing lncRNA (hexagon), miRNAs (squares), and mRNAs (circles). (B) Top 30 GO enrichment terms associated with the NONRATT023334.2-centered network. Please click here to view a larger version of this figure.

Figure 10: Expression levels of key proteins involved in the relevant signaling pathways. (A) Representative Western blot images showing proteins associated with the calcium signaling pathway, gap signaling pathway, and neuroactive ligand-receptor interaction pathway. (B) Quantification of protein expression levels: CAMKII, calmodulin, and CaN (calcium signaling pathway); PKC, CX36, and CX43 (gap junction signaling pathway); and GRIA3, GABRA6, and NPY1R (neuroactive ligand-receptor interaction pathway). **P < 0.01, model group vs. sham group (n = 6). Please click here to view a larger version of this figure.

Table 1: Differentially expressed genes of the competing endogenous RNAs (node degree >5). Degree represents the number of directly connected edges in the network. Please click here to download this Table.

Table 2: Number of lncRNA-miRNA and miRNA-mRNA pairs. LncRNAs with high interaction logarithms may be key regulators. Please click here to download this Table.

Table 3: Key-lncRNA-ceRNA related pathway enrichment. Elaboration of enriched signaling pathways. Please click here to download this Table.

Supplementary File 1: Quality control data for RNA-sequencing library. Please click here to download this File.

To investigate the molecular mechanisms underlying stroke pathogenesis and identify potential diagnostic and therapeutic targets, RNA-Seq and small RNA-Seq (sRNA-Seq) analyses were conducted on brain tissues from rats following MCAO. This study primarily aims to investigate the complex regulatory networks involving lncRNAs, miRNAs, and mRNAs. A comprehensive analysis of the RNA-Seq and sRNA-Seq datasets from MCAO-induced rat brain tissues was conducted to elucidate the lncRNA-miRNA-mRNA regulatory network. An extensive network interaction was constructed to illustrate the regulatory relationships among lncRNAs, miRNAs, and mRNAs, thereby advancing understanding of the RNA-mediated mechanisms underlying stroke pathogenesis. Analysis of the ceRNA network revealed 344 protein-coding genes associated with biological processes such as "leukocyte-cell adhesion," "regulation of integrin-mediated cell adhesion," "collagen fiber organization," "integrin-mediated cell adhesion," "calcium-dependent positive regulation of signal transduction following ischemic stroke," "heterophil intercellular adhesion," and "monocyte chemotaxis." Consistent with our findings, several protein-encoding genes -- including matrix metalloproteinase 9 (MMP9)^46,47,48, and C5a anaphylatoxin chemotactic receptor 1 (C5ar1) -- play critical roles in stroke pathogenesis^49,50.

lncRNAs function as ceRNAs, interacting with miRNAs to regulate gene expression. Wei et al. demonstrate that lncRNA AK038897 binds to miR-26a-5p, leading to upregulation of DAPK1 and exacerbation of ischemia-reperfusion injury. Similarly, SNHG12 functions as a ceRNA for miR-150, regulating VEGF expression to promote angiogenesis following ischemic stroke⁵¹. LncRNARPL34-AS1 mitigates hypoxia/reoxygenation-induced neuronal injury by modulating the miR-223-3p/IGF-1 signaling axis⁵². Similarly, lncRNA NORAD is activated in response to DNA damage and influences ischemic injury through the miR-30a-5p/YWHAG signaling axis⁵³, while Peg13 regulates neuronal death via the miR-20a-5p/XIAP signaling pathway⁵⁴. These findings further highlight the involvement of specific lncRNAs in calcium signaling and neuroactive ligand-receptor interaction pathways following stroke.

In this study, the lncRNA-miRNA-mRNA-ceRNA network in brain tissue of MCAO model rats was successfully constructed by systematically integrating RNA sequencing and bioinformatics analysis. The key steps include: (1) establishing an ischemic stroke model using standardized MCAO surgery⁵⁵, ensuring the reproducibility of the ischemic region; (2) removing rRNA using the Ribo-Zero Kit to significantly improve the capture efficiency of noncoding RNAs⁵⁶; (3) identifying co-expressed lncRNA-mRNA pairs based on a Pearson correlation coefficient (PCC > 0.99), and predicting shared miRNAs using a combination of MiRanda and RNAhybrid algorithms⁵⁷, thereby enhancing the traditional ceRNA network construction workflow. Significant changes in lncRNA expression were observed following ischemic stroke, suggesting their potential involvement in the underlying pathophysiological mechanisms. Bioinformatics analyses revealed that specific lncRNA subsets contribute to stroke progression through the lncRNA-miRNA-mRNA regulatory network. This network offers insight into the complex regulatory functions of lncRNAs and mRNAs. To investigate potential mechanisms, Western blot analysis was conducted to assess the expression of key proteins involved in the calcium signaling pathway, specifically CAMKII, calmodulin, and CaN, and those related to the gap junction signaling pathway, including PKC, CX36, and CX43. Additionally, proteins associated with the neuroactive ligand-receptor interaction signaling pathway, including GRIA3, GABRA6, and NPY1R, were assessed to clarify their roles under experimental conditions. The results showed that, compared to the sham group, the model group exhibited a significant increase in the expression of CaN, a downstream effector in the calcium signaling pathway, while the expression levels of other proteins demonstrated a downward trend. The dysregulated expression of lncRNAs during cerebral ischemia-reperfusion injury may be associated with the downregulation of CAMKII, calmodulin, CX36, PKC, CX43, GRIA3, GABRA6, and NPY1R, along with the upregulation of CaN. These changes further support of the regulatory interactions identified in the lncRNA-miRNA-mRNA network. Future research should explore the potential of lncRNAs as diagnostic biomarkers and therapeutic targets.

Despite advancing the understanding of the ceRNA network, this study has some limitations: (1) Western blot validation focused primarily on selected pathways, such as calcium signaling and gap junctions, without encompassing all potential targets predicted by the network. Consequently, future studies may benefit from the application of multiplex fluorescent Western blot analysis for broader target validation⁵⁸; (2) Co-expression analysis based on Pearson correlation coefficients may not fully reflect the direct regulatory relationships among RNAs. This limitation can be addressed by integrating proteomics technologies in the future^59,60; (3) Sex represents a critical biological variable in biomedical research and was not explicitly addressed in this study. In this study, only male SD rats were used. This decision was based on two primary considerations: first, the success rate of establishing the MCAO model in rats and the rational use of animals; second, the physiological cycles and hormone levels of female animals, which may further interfere with the accuracy of experimental results⁶¹. Future studies should investigate the potential influence of sex as a biological variable on the regulatory mechanisms observed; (4) This study did not conduct an in-depth analysis on the functions of key gene pairs. We fully recognize the importance of functional validation in enhancing the robustness of research conclusions. Therefore, in subsequent studies, we will further improve the experimental content to verify these regulatory relationships by carrying out key detection methods such as luciferase reporter gene assays and RNA pull-down assays.

In terms of technical improvements and troubleshooting, the integration of multiple datasets ensures their consistency, while strict RNA quality control (using optimized extraction protocols to address degradation issues) and adjusted Western blot parameters (antibody dilution, protein loading) can minimize non-specific signals to the greatest extent. This method improves upon existing approaches by integrating multi-omics (lncRNAs, miRNAs, mRNAs) into a "prediction→ enrichment→ experimental validation" pipeline -- constructing a 334-node ceRNA network, identifying key pathways (e.g., calcium signaling) via enrichment, and validating via Western blot (e.g., downregulated CaMKII and calmodulin, P < 0.01) -- unlike conventional studies relying solely on bioinformatics or single-omics⁶². Focusing on 3 high-centrality lncRNAs enhances mechanistic specificity compared to large undirected networks. In usability, its modular workflow with detailed protocols aids adaptability; standardized parameters (RNA input: 500 ng-1 µg; animal stratification by neurological scores; technical replicates) ensure reproducibility; and targeting core lncRNAs boosts efficiency by reducing downstream workload, making it robust for studying lncRNA-mediated ceRNA regulation in ischemic stroke.

Future research may focus on developing clinical diagnostic markers, with hub lncRNAs or miRNAs serving as potential novel biomarkers for ischemic stroke. Monitoring the dynamic expression levels of these RNAs in the blood or cerebrospinal fluid of patients may facilitate early diagnosis or prognostic evaluation⁶³. Additionally, artificial intelligence-driven network modeling can further predict dynamic regulatory relationship⁶⁴, offering theoretical support for precision medicine. The ceRNA network constructed in this study provides a foundational basis for future investigations. Integration with single-cell sequencing or spatial transcriptomes could further elucidate the dynamic regulatory mechanisms underlying RNA molecules in specific brain regions or cell types⁶⁵.