The GSE58438, GSE27274, GSE3219, and GSE9943 datasets were downloaded from the Gene Expression Omnibus (GEO) database. These datasets contain information about gene expression in rat tissue; for further information, see Table 1. The data in GSE58438 explored the effects of valproic acid and dexamethasone on acute renal IR injury, while the original purpose of the GSE27274 dataset was to explore the protective effect of upregulated fibrinogen expression on renal IR injury, and samples from the renal cortex and medulla were sequenced separately. The GSE3219 data was from an investigation into early biomarkers of ischemic and nephrotoxic acute renal failure, and the GSE9943 dataset was originally used to explore differential gene expression in the context of the renal IR response in Brown Norway (BN) and Sprague Dawley (SD) rats. In this study, GSE58438 was used for hub gene screening, whereas GSE27274, GSE3219, and GSE9943 were used to confirm the hub gene expression patterns. Samples with additional intervention factors in the GSE58438 and GSE3219 datasets were not included in this study. To control for possible heterogeneity among rat strains, data from BN and SD rats were analyzed separately. For an overview of the study process, see Figure 1.

The GSE58438 dataset was initially normalized using the R “limma” package, after which differentially expressed genes (DEGs) between the negative control (Ctr) and IR samples collected at the 3 h (IR_3h) or 24 h (IR_24h) time points were identified, using the following criteria: adjusted p < 0.05 and | logFC | ≥1. Then the R “RobustRankAggreg” (RRA) package was utilized to filter the cDEGs between the Ctr vs IR_3h group and the Ctr vs IR_24h group. Volcano plots were constructed using the R “pheatmap” package.

The DAVID online tool was used to conduct Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses and the resultant bubble charts were constructed using the R “ggplot2” package.

The STRING database was used to construct a PPI network incorporating identified the cDEGs using the default parameters, after which the network was imported into Cytoscape software. The top five hub genes were identified using the Cytohubba plugin based on the Degree, Edge Percolated Component (EPC), Density of Maximum Neighborhood Component (DMNC), and Maximal Clique Centrality (MCC) topological measurements. HMOX1 was identified as the key gene based on the weights of these top five hub genes under each screening method.

The R “limma” package was used to normalize and process the GSE27274, GSE3219, and GSE9943 datasets. HMOX1 expression was confirmed in all three datasets and for the Ctr vs IR samples collected at the 120 h (IR_120 h) comparison in the GSE58438 dataset.

The TargetScan, miRWalk, and miRDB databases were utilized to identify putative miRNA regulators of HMOX1. Venn diagrams were used to identify miRNAs that were predicted by all three of these databases.

The TransDetect^® Double-Luciferase Reporter Assay Kit (TransGen Biotech, Beijing, China; catalog number FR201-01) was used according to the manufacturer’s instructions. The wild-type (Wt) and mutant (Mut) fragments in the 3′-UTR of HMOX1 related to the miR-3587 binding sites were cloned into the pcDNA3.1 vector. Then, 293T cells transfected with the Relina luciferase plasmid and HMOX1-Wt or HMOX1-Mut were co-transfected with the miR-3587 inhibitor or microRNA negative control (miR-nc). After 48 h, the luciferase activity was determined in a Varioskan^® Flash Full-wavelength scanning multi-function reader (Thermo Fisher, Waltham, MA, United States).

Rat renal tubular epithelial NRK-52E cells (Procell, CL-0174) were cultured in high-glucose Dulbecco’s modified Eagle medium (DMEM) containing 1% penicillin and streptomycin (Solarbio, 11965) and 10% fetal bovine serum (FBS, Gibco, 16000-044). Cells in the logarithmic phase of growth were added to culture plates and allowed to attach to the plate surface for 12 h, after which they were transferred to a hypoxic incubator (1% O₂, 95% N₂, and 5% CO₂) for 24 h. The media were replaced and the cells were transferred to a normoxic incubator (21% O₂ and 5% CO₂) for 3, 6, or 9 h to establish an in vitro HR model of IR injury. Changes in cell morphology were assessed with an Axio Observer 3 microscope (Carl Zeiss, Germany).

RNA was extracted from renal tubular epithelial cells after exposure to hypoxic conditions for 24 h followed by 3, 6, and 9 h of reoxygenation for measurement of HMOX1 and miR-3587 expression. TRIzol (RNAiso Plus, TaKaRa, Japan, 108-95-2) was used to extract the RNA from the samples, and a NanoDrop 2000 spectrophotometer (Thermo Fisher) was used to assess RNA concentrations and purity. The EasyScript One-Step gDNA Removal and cDNA Synthesis Supermix (TransGen Biotech, AE311-02) were used to prepare cDNA using a StepOnePlus analyzer Real-Time PCR system (Applied Biosystems, United States). All qPCR analyses were performed using a TB Green PreMix Ex Taq TM kit (TaKaRa, RR420A) and an Applied Biosystems 7500 Real-Time PCR instrument. The primer sequences were as follows: β-Actin, positive: 5′-CTA TGA GGG TTA CGC GCT CC-3′ and reverse: 5′-ATG TCA CGC ACG ATT TCC CT-3'; HMOX1, positive: 5′-CAG AAG AGG CTA AGA CCG CC-3′ and reverse: 5′-TTG GTG AGG GAAA ATG TGC CA-3′. U6 and miR-3587 primers were synthesized by RiboBio (Guangzhou, China).

The miR-3587 inhibitor, miR-nc, and a riboFECT™ CP Transfection Kit were purchased from RiboBio. NRK-52E cells in the logarithmic phase of growth were plated in 6-well plates and subjected to the following four treatment conditions: negative control (Ctr), HR, HR + miR-3587 inhibitor transfection, and HR + miR-nc transfection. Cells were transfected with miR-3587 inhibitor or miR-nc based on the provided kit directions. Briefly, after 12 h when cells were adherent and 50% confluent, the miR-3587 inhibitor or miR-nc transfection mixtures were applied and cells were incubated for 24 h. In this experiment, the HR model system was established as above, with the appropriate transfection mixture being added when the media was changed. For an overview of this experimental protocol, see Figure 4A.

RIPA buffer (Applygen, C1053) containing 1% phenylmethanesulfonyl fluoride (PMSF) was used to harvest protein from the treated cells and protein levels were quantified using a BCA kit (Applygen, P1511). Equal amounts of protein (20–30 μg/lane) were then separated on 10% SDS-PAGE and transferred to polyvinylidene fluoride (PVDF) membranes (Millipore, IPVH00010 Immobilon-P Transfer Membrane) with a wet transfer system using 300 mA. Blots were blocked with 5% non-fat milk for 1 h at room temperature, followed by overnight incubation with primary antibodies (HO-1, Abcam, ab68477; GPX4, Abcam, ab125066) at 4°C. According to the antibody instructions, the concentration of the HO-1 and GPX4 antibodies diluted at the ratio of 1:10000 and 1:1,000 were 5 ng/ml and 420 ng/ml, respectively. After three washes with Tris-buffered saline containing Tween 20 (TBST), the membranes were incubated with HRP-conjugated secondary antibodies (ProteinTech, sa00001-2) for 1 h at room temperature. After three additional washes in TBST, Super ECL Prime (US EVERBRIGHT, S6008) was used to detect protein bands together with a Bio-Rad (United States) imaging system. ImageJ v1.8.0 (National Institutes of Health) was used for the densitometric analysis of the protein bands.

Following successful cellular transfection and establishment of the HR model as detailed above, the cell viability, MDA content, and Fe²⁺ level were analyzed using a cell counting kit-8 kit (CCK-8, Abmole, M4839), an MDA determination kit (Nanjing Jiancheng, A003-1–2), and an Iron Colorimetric Assay Kit (Applygen, E1042), respectively, according to the kit instruction manuals.

The NRK-52E cells were placed in a 96-well plate at a density of 4,000 cells per well. Cells were transfected with the miR-3587 inhibitor or miR-nc 24 h before HR induction. Cell viability was measured as described above and absorbances at 450 nm were determined using a Varioskan^® LUX Full-wavelength scanning multi-function reader (ThermoFisher).

The MDA concentration was measured by the thiobarbituric acid (TBA) method to evaluate lipid peroxidation. Briefly, treated cells were washed three times with phosphate-buffered saline (PBS), lysed with RIPA buffer containing 1% PMSF for 30 min, and the 0.15 ml lysate was transferred into the sample tube. The blank tube comprised 0.15 ml of distilled water. After addition of 2.5 ml of TBA to each tube and thorough mixing, the tubes were incubated at 100°C for 40 min, cooled under running water, and centrifuged at 4,000 rpm for 10 min. The supernatants were placed in 96-well plates and the absorbances at 532 nm were measured using the multi-function reader. The total protein concentrations were determined with a BCA kit and the MDA level was normalized to that of a sample protein.

The levels of Fe²⁺ were measured as follows: after modeling, the cells were washed three times with PBS, lysed using the supplied lysis solution, and homogenized in a 60 rpm shaker for 2 h. The 3 mM standard supplied with the kit was diluted to concentrations of 300, 150, 75, 37.5, 18.75, 9.38, and 4.69 µM. The buffer was mixed with 4.5% potassium permanganate solution at a 1:1 ratio to prepare mixture A. The standard sample, sample, and diluent of 100 µl were added to the standard, sample, and blank control tubes, respectively, followed by addition of mixture A and thorough mixing. The tubes were incubated at 60°C for 1 h, cooled to room temperature, and immediately centrifuged. Thirty microliters of the Fe²⁺ were added to each tube, mixed well, and incubated for 30 min at room temperature. Two hundred microliter aliquots were then placed in 96-well plates and the absorbance read at 550 nm in the multi-function reader.

Changes in MMP were assessed using an Olympus IX71 fluorescence microscope (Olympus, Japan) using the directions provided with the JC-1 MMP assay kit (Abmole, ab113850). In brief, NRK-52E cells were placed in 24-well plate built-in climbing tablets with a density of 1.5 × 10⁵ cells per well. The cells were transfected with the miR-3587 inhibitor and miR-nc 24 h before HR induction. After modeling, the culture medium was removed and each well was washed once with 500 µl of buffer. One hundred microliters of the supplied kit solution were then added to each well and incubated at 3°C for 10 min. The cells were then washed three times with 500 µl of buffer for each wash for 5 min per wash. A drop of the anti-quenching agent (about 10 µl) was added to the slide, and the round coverslip was inversely covered on the slide, followed by examination and imaging with a fluorescence microscope.

Cells were fixed in 2.5% electron microscope-grade glutaraldehyde phosphate buffer for 4 h, rinsed three times with 0.1 M phosphoric acid rinse solution (15 min per wash), fixed in 1% osmic acid solution for 1–1.5 h, and again washed three times with the phosphoric acid rinse solution. The specimens were then dehydrated in step-by-step acetone (50–100%), embedded in Epon812 resin, and cured at 37°C for 24 h, 45°C for 24 h, and 60°C for 24 h. Sections (70 nm) were prepared using a Reichert-Jung ULTRACUT E ultra-thin slicer (Austria), covered copper net, and stained with uranium acetate for 15 min and lead citrate for 15 min. The sections were examined and images using a JEM1200 electron microscope (JEOL, Japan).

Three biological replicates were used for all experiments. Data are expressed as means ± standard deviations. GraphPad Prism 8.3.0 was used for all analyses. Student’s t-test was used to compare differences between two groups, and ANOVA was used to compare differences between multiple groups. Dunn’s multiple comparisons test was used for one-way ANOVA and Fisher’s least significant difference (LSD) was used for two-way ANOVA. p < 0.05 was considered to be significant.

Samples included in the GSE58438 datasets were normalized with the R “limma” package (Supplementary Figure S1).

When analyzing the GSE58438 datasets, 638 DEGs were identified in the Ctr vs IR_3h comparison, including 413 and 225 up- and down-regulated genes, respectively. Additionally, 1,467 DEGs were identified for the Ctr vs IR_24h comparison, including 671 and 796 up- and down-regulated genes, respectively. The DEGs were arranged into volcano plots using the R “ggplot2” package (Figures 2A,B), while clustering heatmaps were drawn using the R “pheatmap” package (Supplementary Figures S2A,B). These heatmaps revealed that one sample in the IR_24h dataset (sample number: GSM1411067) clustered with the Ctr group, suggesting either that the modeling was unsuccessful or that this sample was atypical. To improve the specificity of our overall results, this sample was omitted from subsequent analyses as it had the potential to introduce false-negative experimental results. Following re-screening, a total of 2746 DEGs were identified for the Ctr vs IR_24h, of which 1,304 and 1,442 were up- and down-regulated, respectively.

A total of 80 cDEGs were identified using the R “RRA” package when comparing the Ctr vs IR_3h and Ctr vs IR_24h datasets. Of these cDEGs, 55 and 25 were up- and down-regulated, respectively (Supplementary Figure S2C).

GO and KEGG functional enrichment analyses of the 80 cDEGs were performed using the DAVID database. The top 10 biological process (BP), cellular component (CC), and molecular function (MF) GO terms are shown in Figures 2C–E. The top five enriched BP terms were the “response to cytokine”, “positive regulation of cell proliferation”, “cellular response to interleukin-1”, “negative regulation of protein kinase activity”, and “positive regulation of gene expression” categories, while the top five enriched CC terms were “extracellular space”, “extracellular matrix”, “extracellular region”, “proteinaceous extracellular matrix”, and “cell surface”, and the top five enriched MF terms were “growth factor activity”, “cytokine activity”, “receptor binding”, “protein kinase inhibitor activity”, and “protease binding”. These analyses suggested that IR-related genes were primarily associated with responses to the release of inflammatory mediators and the regulation of cell proliferation.

The KEGG pathway enrichment results are shown in Figure 2F, with the top five enriched pathways including the “JAK-STAT”, “TNF”, “cytokine-cytokine receptor interaction”, “rheumatoid arthritis”, and “HIF-1 signaling” pathways. JAK-STAT pathways play critical roles in diverse processes including proliferation and blood production. HIF-1 signal pathway activation regulates cellular responses to hypoxia through the engagement of various compensatory pathways.

A PPI network was constructed for the cDEGs using the STRING database. The network included 78 nodes and 124 edges and was visualized using Cytoscape (Figure 2G). The PPI network was analyzed with the MCODE plugin, leading to the identification of three functional clusters (module 1, MCODE score = 7.429; module 2, MCODE score = 3.6; and module 3, MCODE score = 3) (Figures 2H–J). Degree, EPC, DMNC, and MCC values were used to identify the top five hub genes in functional cluster 1 (Table 2). The top five hub genes identified based upon the Degree, MCC, and EPC metrics were identical, and scores for the top five hub genes as screened based upon the Degree and MCC values were identical. Only the DMNC values yielded distinct hub genes. HMOX1 was a key gene that exhibited the highest EPC score, and it was thus chosen as a target for further study.

Samples in the GSE58438 and GSE27274 datasets were from Wistar rats, while samples in the GSE3219 datasets were from SD rats, and samples from the GSE9943 dataset were from both SD and BN rats. HMOX1 expression in the GSE58438, GSE27274, GSE3219, and GSE9943 datasets is shown in Table 1 and Figures 3A–D. These analyses revealed that HMOX1 expression rose significantly within 3–8 h after reperfusion, before declining significantly at 24 h post-reperfusion.

Using the TargetScan, miRWalk, and miRDB databases, 56, 48, and 11 miRNAs putatively targeting HMOX1 were identified, respectively. The only miRNA predicted by all three databases was miR-3587 (Figure 3E). Sequence complementarity between HMOX1 and miR-3587 is shown in Figure 3F. The direct targeting of HMOX1 by miR-3687 was confirmed by the dual-luciferase reporter assay. As shown in Figure 3G, compared with the mutant HMOX1 luciferase reporter, co-transfection of the miR-3587 inhibitor and the wild-type HMOX1 luciferase reporter resulted in a significant increase in luciferase activity, indicating that miR-3587 binds directly to the 3′UTR of HMOX1.

To confirm the above results, an in vitro renal IR model system was established by incubating NRK-52E under hypoxic conditions followed by reoxygenation for 3, 6, or 9 h qPCR showed that HMOX1 expression decreased gradually as the reoxygenation time increased, in contrast to miR-3587 expression which was steadily upregulated, with significant differences among groups (Figure 3H). Light microscopic analyses for these cells are shown in Figure 3I. The NRK-52E cells in the Ctr group resembled a cobbled surface. HR3h treatment led to the morphological contraction of the cells, with protrusions from the flattened cobblestone appearance visible. HR6h treatment led to cell membrane rupture and blistering with nuclei of normal size and no evidence of chromatin condensation, the typical morphological changes seen in ferroptosis. HR9h treatment caused the nucleus to deviate to one side and the rest showed vacuolar changes.

Details of the experimental conditions and sample collection methods are outlined in Figure 4A. The results of Western blotting indicated significant increases in HO-1 expression in the HR group, while treatment with the miR-3587 inhibitor resulted in further increases in HO-1 expression (Figure 4B). GPX4 is an important regulator of ferroptosis, and transfection with the miR-3587 inhibitor was associated with increased GPX4 expression (Figure 4C). The effects of miR-3587 on NRK-52E cell viability were assessed using the CCK-8 assay, showing that miR-3587 inhibition was associated with increased viability (Figure 4D). MDA is a lipid peroxidation by-product that reflects the degree of lipid peroxidation in cells, thereby offering indirect insight into ferroptosis induction. The MDA levels in HR-induced NRK-52E cells declined significantly following miR-3587- transfection (Figure 4E), as did the level of Fe²⁺ (Figure 4F). Ferroptosis is associated with significant reductions in MMP. When the MMP is high, JC-1 aggregates in the mitochondrial matrix to form a polymer that fluoresces red. When the MMP is low, JC-1 cannot accumulate in the mitochondrial matrix and is seen as a green-fluorescing monomer. The Ctr group showed red JC-1 fluorescence, whereas green fluorescence was observed following HR treatment, and miR-3587-inhibitor treatment normalized MMP in these HR-treated cells, which exhibited red fluorescent signals (Figure 4G). As shown in Figure 4H, the mitochondrial ridge was reduced or had disappeared, while the mitochondrial outer membrane was damaged in the HR and HR + miR-3587-nc groups, the typical mitochondrial changes associated with ferroptosis.