Identification of Ferroptotic Genes in Spinal Cord Injury at Different Time Points: Bioinformatics and Experimental Validation

Programmed cell death (PCD) is an important pathologic process after spinal cord injury (SCI). As a new type of PCD, ferroptosis is involved in the secondary SCI. However, the underlying molecular mechanism remains unclear. In this study, we validated ferroptotic phenotype in an animal model of SCI. Then, the bioinformatic analyses performed on a microarray data of SCI (GSE45006). KEGG analysis suggested that the pathways of mTOR, HIF-1, VEGF, and protein process in endoplasmic reticulum were involved in SCI-induced ferroptosis. GO analysis revealed that oxidative stress, amide metabolic process, cation transport, and cytokine production were essential biological processes in ferroptosis after SCI. We highlighted five genes including ATF-3, XBP-1, HMOX-1, DDIT-3, and CHAC-1 as ferroptotic key gene in SCI. These results contribute to exploring the ferroptotic mechanism underlying the secondary SCI and providing potential targets for clinical treatment.


Introduction
Spinal cord injury (SCI) is a severe central nervous system injury that results in irretrievable loss of sensory and motor function below the site of injury [1]. According to pathophysiology of SCI, it can be divided into primary and secondary SCI [2]. Primary SCI is the initial trauma caused by mechanical contusion or extrusion, while secondary SCI refers to the multifaceted pathological processes after primary SCI that includes tissue edema, inflammatory reaction, necrosis, and programmed cell death (PCD) [3]. As the hurdles of neural regeneration in SCI, PCD has attracted increasing attention in recent years [4][5][6]. Nevertheless, the molecular mechanism remains inconclusive.
Ferroptosis was firstly proposed by Dixon et al. in 2012 [7]. Differing from apoptosis, necrosis, autophagy etc., it was a new form of PCD mediated by ferric ion [8]. Symbols of ferroptosis, including overloaded iron, shrunken mitochondria, accumulated lipid peroxidation, and upregulated ROS, have been observed in animal models of SCI, and the SCI-induced ferroptosis can be reversed by ferroptosis inhibitors such as deferoxamine (DFO) and SRS 16-86 [9][10][11]. The previous studies revealed that ferroptosis plays an important role in SCI. Therefore, a comprehensive analysis of the ferroptotic genes in SCI might contribute to deeper understanding of secondary SCI and seeking new therapeutic targets.
GEO (gene expression omnibus) is an international public repository that archives and freely distributes microarray, next-generation sequencing, and other forms of highthroughput functional genomics data submitted by the researchers [12]. FerrDb, the first database of ferroptosis, was constructed by Nan Zhou et al. in 2019 [13]. Based on Yu Kang and Qiangwei Li contributed equally to this work and share first authorship. GEO and FerrDb, the interaction of SCI and ferroptosis can be analyzed by bioinformatic methods.
In this study, we verified the ferroptotic phenotype by the levels of tissue iron, malondialdehyde (MDA) and reactive oxygen species (ROS), and mitochondrial morphology. Subsequently, the datasets derived from GEO and FerrDb were used to capture differentially expressed genes (DEGs) related to ferroptosis at different time points of SCI. The DEGs further went through the Gene Ontology functional annotation analysis (GO) and Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis (KEGG) and gene set enrichment analysis (GSEA). Finally, we established the protein-protein interaction (PPI) networks and determined the ferroptotic key genes in SCI (Fig. 1).

Animal Groups and SCI Model
Female Sprague-Dawley (SD) rats (Anhui Medical University, license number: SYXK(wan)2017-001) were chosen as the experimental animal in the present study (8-10 weeks old, weighing 230-260 g). All rats were housed under a 12-h light/dark cycle pathogen-free condition with a controlled temperature of 24 ± 2 °C and 60 ± 5% humidity and free access to food and water. All experiments and procedures were performed in accordance with the National Institutes of Health (NIH) Guide for the Care and Use of Laboratory Animals. The study protocols regarding the animals were approved by the Ethics Committee of Anhui Medical University of China (No. LLSC 20,201,135). For behavioral assessment, animals were divided into two groups randomly: Sham and SCI, 5 rats in each group. For ferroptotic phenotype validation, animals were divided into three groups randomly: sham, SCI_Day 3, and SCI_Day 14, 5 rats in each group.
The SCI model was constructed via modified version of Allen's method as described previously [14]. Animals were anesthetized by 10% chloral hydrate (3 ml/kg, intraperitoneal injection) [15,16]. After skin preparation, the surgical area was disinfected and T10 spinous process was located. Skin over T10-12 was incised, bilateral perispinous muscles were separated. Then, T10 spinous process and vertebral plate were exposed. Spinal cord was exposed by T10 laminectomy. Subsequently, spinal cord was contused by an Fig. 1 Schematic of the experimental strategy used to validate ferroptotic phenotype, identify hub genes, and verify expression level of key genes impactor: 10 g weight falling from 5 cm height. The rats in sham group were accepted the T10 laminectomy only. The bladder was evacuated manually twice a day until bladder function was recovered.

Behavioral Assessment for SCI
The classical Basso, Beattie, and Bresnahan (BBB) scale was used here for the behavioral assessment [17]. BBB scale ranges from 0 to 21 points. Score 0 means complete hind limb disability and score 21 means normal locomotor function. Behavioral assessments were conducted before surgery and at days 1, 3, 7, 14 after SCI for two groups (n = 5). To avoid experimental errors, the rats should be allowed to move freely for 5 min in the open filed before every evaluation. To avoid bias results, assessments were executed by three researchers who were blinded to animal groups and time points.

Transmission Electron Microscope
Rats were perfused with transmission electron microscope (TEM) fixation liquid (Paraformaldehyde 2%-glytaraldehyde 2.5% in PBS, P885738, Macklin). Then, samples were cut to a size of 1mm 3 and immersed in fixation liquid for 24-48 h. Fixed samples were collected and cut into Sects. (70-90 nm) by an ultramicrotome (LEICA, EM UC7, Germany). After post-fixed in 2% osmium tetroxide, dehydrated in ethanol and then embedded in eponate, sections were placed on a copper mesh and stained with 2% uranyl acetate and 0.04% lead citrate. These sections were observed and imaged with a transmission electron microscope (TEM, Talos L120C G2, Thermo Scientific).

Iron and MDA Concentration Test
Iron content within injured spinal cord was tested using the tissue iron assay kit (A039-2-1, Nanjing Jiancheng Bioengineering Institute). Malondialdehyde (MDA), as a natural product of lipid peroxidation, was determined using lipid peroxidation MDA assay kit (S0131S, Beyotime). Briefly, spinal cord was removed and washed free of blood, and then samples were assayed for iron and MDA concentration tests according to the manufacturer's instructions. For iron content, the absorbance was measured at 532 nm. For MDA concentration, the absorbance was measured at 520 nm.

ROS Assay
Reactive oxygen species was determined by the fluorescent probe 2′, 7′-dichlorodihydrofluorescein diacetate (DCFH-DA, WanleiBio). Rats were anesthetized and perfused transcardially with pre-cooled PBS. The spinal cords were removed and digested by pancreatin to obtain homogenate. Homogenate was filtered with 200 mesh and then cell suspension was incubated with DCFH-DA (10 μM) at 37℃ for 0.5 h. The cells were observed and pictured by an inverted fluorescence microscope (Axio Observer 3, Germany). For every single spinal cord cell suspension, three fields of view (FOV) were pictured for quantitative analysis. The ROS fluorescence intensity was calculated by software ImageJ version 1.53c (National Institutes of Health, Bethesda, MD, USA).

Data Source
The microarray expression profiling dataset GSE45006 was download from the GEO database (https:// www. ncbi. nlm. nih. gov/ geo/). The gene sets of ferroptosis markers and regulators were download from FerrDb (http:// www. zhoun an. org/ ferrdb/). GSE45006 was based on GPL1355 [Rat230_2] Affymetrix Rat Genome 230 2.0 Array. GSE45006 contained 24 samples consisting of 4 intact spinal cords and 20 injured spinal cords that were from different time points. We grouped the injured spinal cords by 1 day, 3 days, 7 days, 14 days, and 56 days and the 4 intact spinal cords were defined as control group.

Analysis of DEGs
GEO2R (https:// www. ncbi. nlm. nih. gov/ geo/ geo2r/) was used for differentially expression analysis. We analyzed the gene expression of different time point after SCI. The genes which satisfied the following criteria: (1) adjusted P-value < 0.05; (2) |logFC|≥ 0.5 were screened out. Intersection of these screened genes and ferroptotic marker genes was taken by a online tool named Venn diagram Sham SCI (http:// bioin forma tics. psb. ugent. be/ webto ols/ Venn/). We defined the genes within the intersection as the differentially expressed genes (DEGs).

GO and KEGG Enrichment Analysis of DEGs
The Database for Annotation, Visualization and Integrated Discovery (DAVID v6.8, https:// david. ncifc rf. gov/ home. jsp) was used for the Gene Ontology functional enrichment analysis (GO) [18]. The GO annotation analysis was focused on biological process (BP), molecular function (MF), and cellular component (CC) [19]. Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis (KEGG) was performed using the KOBAS online tool (http:// kobas. cbi. pku. edu. cn/) [20]. Adjusted P-value < 0.05 was considered as the cutoff criteria. Bar charts of GO enrichment analysis were visualized by "ggplot2" package of R software, GOChord plot, and the bubble plot for KEGG enrichment analysis was produced by "bioinformatics," a free online platform for data analysis and visualization (http:// www. bioin forma tics. com. cn).

Gene Set Enrichment Analysis
Gene set enrichment analysis (GSEA) was performed to avoid missing the ferroptotic genes that did not meet the cutoff criteria but might be crucial to SCI. All the ferroptosis marker genes of each time point were defined as the target gene set. Background gene set was combination of KEGG and GO (BP, MF, CC) gene sets. An absolute value of the standardized enrichment score (NES) > 1 and p-value < 0.05 was taken as the cutoff criteria of statistical significance [21][22][23]. Packages "org. Rn.eg.db," "Clus-terProfiler," "enrichplot," "gseaplot2" of R software were used for GSEA analysis and visualization.

PPI Network Analysis and Hub Gene Identification
STRING (https:// string-db. org/) is a database of known and predicted protein-protein interactions that include direct (physical) and indirect (functional) associations [24]. We used the STRING to establish the PPI network of DEGs. The minimum combined score of PPI pairs was set as 0.4. PPI network was visualized by Cytoscape software v3.7.2 (www. cytos cape. org) [25]. CytoHubba, a plugin in Cytoscape software, can predict and explore important nodes and subnetworks in a given network by several topological algorithms. We used CytoHubba to calculate the MCC score of each protein node and top six genes were identified as hub genes.

Selection of Key Genes for Ferroptosis Regulation in SCI
We focused on the links between differentially expression and ferroptosis regulation of hub genes obtained from 1 day, 3 days, 7 days, 14 days, and 56 days. The hub genes that exist in acute phase (days 1, 3, 7), sub-acute phase (day 14), and chronic phase (day 56) of SCI were identified as the key genes [26]. The regulating effect of the key genes in the process of ferroptosis were checked in the Ferroptosis driver, suppressor, and marker dataset downloaded from FerrDb. The function of key genes were annotated by the GeneCard database (https:// www. genec ards. org/).

Protein Extraction and Western Blot Test
Western blot was used for the validation of key genes.

Statistical Analysis
Statistical analysis were performed by SPSS version 26.0 (SPSS Inc., Chicago, IL, USA). Graphs were plotted by GraphPad 8.0.2 version (GraphPad Inc., San Diego, CA, USA). Comparisons between two groups (BBB score, quantitation of WB) were analyzed using Mann-Whitney U test. Comparisons between three groups (iron content, quantitation of ROS, and MDA content) were analyzed using median test, followed by Bonferronis post hoc test. P < 0.05 indicated that the difference was statistically significant.

Validation of Ferroptosis Following SCI
Ferroptosis is characterized by iron accumulation, shrunken mitochondria, and lipid peroxidation. Here, we constructed SCI animal model by modified Allen's method, and then the model was evaluated by BBB scale and histological staining. BBB scores of SCI group were significantly lower than sham group ( Fig. 2A). HE staining showed the apparently cavity in both SCI_3 days and SCI_14 days (Fig. 2B). Nissl staining depicted that the number of motor neuron was decreased after SCI (Fig. 2C). Results of Perls-blue staining and tissue iron assay showed that iron content was significantly increased within the spinal cord after injured (Fig. 2D, E). Meanwhile, the shrunken mitochondria were observed in the SCI groups by TEM (Fig. 3A). As shown in Fig. 3B-D, MDA and ROS were upregulated in these two SCI groups, suggesting a higher oxidant level.

Fig. 3
Observation of shrunken mitochondria and higher level of oxidation in SCI. A TEM was used to examine the ultrastructure of tissues after SCI. Shrunken mitochondria in SCI_Day 3 and SCI_Day 14 groups were more than sham group. B, C Fluorescence of ROS and quantitative analysis of sham, SCI_Day 3, and SCI_Day 14 groups (n = 5). D MDA, one of the products of lipid peroxidation, was augmented at SCI_Day 3 and SCI_Day 14 groups compared with sham group (n = 5). Bar represents median, limits of box represent first and third quartile, and whiskers represent minimum and maximum. **p < 0.01  The boxplot showed that the medians of each sample data were almost at the same level, which means that the data met the standard for further analysis (Fig. 4B). Among the upregulated and downregulated genes obtained previously, we screened out the ones satisfied with the criteria of |logFC| ≥ 0.5 and Venn analysis was performed to get the intersection of these genes and ferroptosis marker dataset (Fig. 4C). Ferroptosis marker dataset was downloaded from FerrDb and contained 123 annotated genes with 137 genes symbols (14 genes with two different symbols). Finally, ferroptosisrelated DEGs of 5 time points after SCI was obtained for subsequent analysis (Table 1).

GO Functional Enrichment Analysis
GO functional enrichment analysis was performed using the DAVID 6.8 and visualized by using R software and the bioinformatic online tool. The eligible terms for GO analysis of each time point were shown in Fig. 5. We listed the GO terms in Supplementary Table 1 according to the different phases of SCI (days1, 3, and 7 were acute phase, day 14 was sub-acute phase and day 56 was chronic phase). We picked out 7 GO terms related to ferroptosis (Supplementary  Table 2): cellular response to oxidative stress, positive regulation of transcription from RNA polymerase II promoter in response to oxidative stress, response to oxidative stress, cellular oxidant detoxification, cellular response to hydrogen peroxide, positive regulation of cell death, mitochondrion, the genes enriched in these terms were shown in Fig. 6.

KEGG Pathway Enrichment Analysis
KEGG pathway enrichment analysis was performed using the KOBAS database and visualized by using the bioinformatic online tool. We arranged KEGG terms in ascending order of adjusted P values at each time point, top ten terms were remained and visualized in Fig. 7. Afterwards, we filtered these terms further and the terms that throughout the three phases of SCI were mTOR signaling pathway, HIF-1 signaling pathway, ferroptosis, VEGF signaling pathway, and protein processing in endoplasmic reticulum. Details of these KEGG terms were shown in Table 2. A total of 69 different pathways with adjusted P value less than 0.05 were enriched in 5 time points (Supplementary Fig. 1).

Establishment of PPI Network
PPI network analysis was performed by the STRING database. The software Cytoscape (version 3.7.2) was used for visualization of PPI network. The PPI network of day 1 contained 24 nodes (18 upregulated genes, 6 downregulated genes) and 49 linkages. The PPI network of day 3 contained 33 nodes (26 upregulated genes, 7 downregulated genes) and 97 linkages. The PPI network of day 7 contained 17 nodes (15 upregulated genes, 2 downregulated genes) and 25 linkages. The PPI network of day 14 contained 19 nodes (12 upregulated genes, 7 downregulated genes) and 39 linkages. The PPI network of day 56 contained 22 nodes (16 upregulated genes, 6 downregulated genes) and 53 linkages (Fig. 10A).

Discussion
A previous bioinformatic study found that ANXA1, SNAP25, and SPP1 were closely related to spinal cord repair and regeneration after injury [19]. A sequence analysis revealed the effect of noncoding RNA in SCI and a ceRNA regulation network with miR-21 as the center was constructed [27]. Recently, a research combined proteomics with bioinformatics indicates that the protein module continuously upregulated at the acute and sub-acute time points after SCI was associated with the lipid metabolism [28]. Moreover, an increasing number of studies focused on the ferroptosis following SCI in recent years. It was reported that proanthocyanidin could inhibit ferroptosis and promote functional recovery of SCI [29]. Zinc was found to improve the recovery of SCI via upregulating the expression of NRF2 and GPX4 that could attenuate ferroptosis [30]. Although it has been demonstrated that inhibition of ferroptosis could improve the recovery of SCI, the specific pathways and genes were still unclear. Therefore, we performed a bioinformatic analyses here to investigate the core genes of ferroptosis in SCI. This study combined animal experiment and bioinformatic analysis to identify the ferroptotic genes in SCI. Vascular damage caused by contusion resulted in the iron accumulation at the epicenter lesion. Meanwhile, injury stress accelerated production of ROS and enhanced excitatory toxicity of glutamate [3]. In our study, we primarily examined the symbols of ferroptosis following SCI, including iron accumulation, ROS, MDA, and morphological structure of mitochondria. Iron accumulation in SCI was verified by the tissue iron assay kit whose results were consist with the Perls-blue staining. Increased levels of ROS and MDA suggested the peroxidation status of cells in SCI. Shrunken mitochondria were observed via TEM at 3 days and 14 days after SCI. For bioinformatic analysis, two datasets (GSE45006 and ferroptosis marker dataset) were downloaded to screen DEGs-related ferroptosis at different time point after SCI. Among the genes that met the cutoff criteria, we found 37, 43, 28, 28, 32 ferroptosis markers respectively at days 1, 3, 7, 14, 56. These markers considered as DEGs underwent a series of analyses subsequently. Eventually, ferroptosis-related hub genes, signaling pathways and key genes were yielded from the analyses. Discovery of these genes and pathways provide insight into the pathophysiology of SCI.
GO enrichment analysis showed that the DEGs were mainly involved in cell response to hypoxia and oxidative stress, the regulation of programmed cell death (PCD), mitochondrion function, transcription factor activity, DNA/protein binding, protein homodimerization activity, and blood vessel endothelial cell migration. Among these GO terms, there were 7 ferroptosis-related terms enriched with 25 genes. KEGG pathway enrichment analysis suggested that the DEGs were primarily involved in mTOR signaling pathway, VEGF signaling pathway, mitophagy, IL-17 signaling pathway, metabolic pathways, HIF-1 signaling pathway, TNF signaling pathway, ferroptosis, protein processing in endoplasmic reticulum, cytosolic DNA-sensing pathway, and MAPK signaling pathway. Among these KEGG terms, 5 signaling pathways were enriched in acute, sub-acute, and chronic phases of SCI: mTOR signaling pathway, HIF-1 signaling pathway, ferroptosis, VEGF signaling pathway, and protein processing in endoplasmic reticulum. Results of GSEA supplemented that cation transport, cytokine-mediated signaling pathway, and amide biosynthetic process were also worthy to be focused in SCI.
The mammalian target of rapamycin (mTOR) signaling pathway plays an important role in trauma and various diseases in the central nervous system (CNS) [31]. Modulation of mTOR signaling pathway can improve motor recovery via restraint of inflammation and apoptosis after SCI [32]. Furthermore, as a negative autophagy regulator, mTOR signaling pathway is implicated in regulating autophagydependent ferroptosis [33].
Hypoxia-inducible factor-1 (HIF-1) signaling pathway is involved in the cellular response to hypoxia. Elevation expression of HIF-1 benefits neurological recovery in SCI rats [34,35]. And the negative effect of HIF-1 on ferroptosis was implemented via inducing transcription of fatty acid-binding protein 3 and fatty acid-binding protein 7 [36].
The primary function of vascular endothelial growth factor (VEGF) signaling pathway is associated with angiogenesis. VEGF may indirectly protect the nerve from ischemic and anoxic injury and activation of HIF-1/VEGF signaling pathway contributes to recovery from SCI [37]. Evidence of the cross-talk between VEGF and ferroptosis signaling pathway was that overexpression of VEGF was detected in retinal pigment epithelium (RPE) cells treated with H 2 0 2 , and the production of VEGF was inhibited by the ferroptosis suppressor SLC7A11 [38]. Endoplasmic reticulum (ER) is vital for redox balance of cells. Persistent exposure to excessive stress drives the ERassociated cell death pathway [39]. Recently, a new research presented that ER stress was crucial for blood-spinal cord barrier (BSCB) disruption after SCI and inhibition of ER stress contributed to the integrity of BSCB [40]. As a regulator of lipid homeostasis, ER is closely related with ferroptosis [41]. Unfolded protein response (UPR), the adaptive mechanism for ER homeostasis, is affected by ferroptotic regulators such as erastin [42], xCT [43], ATF4 [44], and NRF2 [45].
PPI networks of each time point were constructed and the hub genes were identified via CytoHubba based on MCC algorithm. There were total 12 hub genes at 5 time points: ATF3, XBP1, HMOX1, VEGFA, ATF4, NFE2L2, PTGS2, DDIT3, ASNS, CHAC1, RELA, BNIP3. Among these genes, ATF3, XBP1, HMOX1, ATF4, NFE2L2, PTGS2, DDIT3, ASNS, RELA were upregulated,; VEGFA and BNIP3 were downregulated; and CHAC1 was upregulated at day 3 while downregulated at day 14 and day 56. Furthermore, 5 genes were found differentially expressed in all three phases following SCI: ATF3, XBP1, HMOX1, DDIT3, and CHAC1. Activation transcription factor 3 (ATF3) is a member of the ATF/CREB family of transcription factors, and it is induced under a wide range of stress condition, including cell injury and oxidative stress [46]. A previous study showed that ATF3 was upregulated after SCI in rats and expression of NeuN was declined with AFT3 expressing in cells [47]. On the other hand, ATF3 was reported to suppress the system Xc-via binding to the SLC7A11 promoter and inhibiting the expression of SLC7A11 so that AFT3 was a promoter of ferroptosis induced by erastin [48]. In current study, we found ATF3 was upregulated at days 1, 3, 7, 14, and 56 after SCI, and it was annotated as a driver of ferroptosis in FerrDb.
X-box-binding protein 1 (XBP1) was first described in 1990 [49]. It is downstream of inositol-requiring enzyme 1 (IRE1α) and is a key transcription factor in the unfolded protein response (UPR) [50]. According to the literature, IRE1-XBP1 axis affected homeostasis of ER in oligodendrocyte and functional recovery after SCI [51]. Although XBP1 has a close link with ER stress and its accumulation also be seen at the injury epicenter after SCI [51], effects of XBP1 in ferroptosis is rare reported. Here, expression of XBP1 was increased at days 1, 3, 7, 14,and 56 after SCI, the annotation of XBP1 in FerrDb was the marker of ferroptosis and may promote ferroptosis. Hence, we reckoned that ER stress response might have the effect of bridge between XBP1 and ferroptosis, which provided a novel insight for deeper exploration of ferroptosis-related mechanism in SCI. Heme oxygenase 1 (HMOX1, HO-1) is critical to heme metabolism [52]. The function of HO-1 in ferroptosis is controversial. It can either drive or suppress ferroptosis as it described in FerrDb [53]. It was reported that augmented level of HO-1 contributed to defense against ferroptosis in SCI [29,30], which indicated that HO-1 might play the ferroptotic suppressor role in SCI. Moreover, HO-1 is present downstream of NFE2L2 (NRF2) and HO-1 expression is closely regulated by NFE2L2 [54], suggesting a upregulated hub gene in our research.
DNA damage inducible transcript 3 (DDIT 3), or named C/EBP homologous protein (CHOP), belongs to the family of CCAAT/enhancer-binding proteins (C/EBPs) and regulates genes that encode proteins involved in proliferation, differentiation and expression, and energy metabolism [55]. Restraint of DDIT3 can promote functional recover of SCI via diminishing neuronal apoptosis preserving oligodendrocytes and axons [56,57]. DDIT3 is also involved in regulation of ferroptosis by affecting synthesis of intracellular GSH [58].
Gamma-GCT acting on glutathione homolog 1 (CHAC1) is a newly discovered ER stress inducible gene, which is involved in glutathione metabolism and cell apoptosis or ferroptosis [59]. CHAC1 can be triggered by depletion of glutathione and activation of ATF4. Expression of CHAC1 and ATF4 enhance the glutathione depletion in turn [60], functioning in a feedforward mechanism to strengthen ferroptosis. CHAC1 was reported to be upregulated at 1 day and 3 days post-SCI. Attenuation of CHAC1 via CD36 knockout improved the vascularity within the injury region [61]. Notably, the improvements were observed at both 1 and 3 days post-injury but not by 7 weeks post-injury [61].

Conclusion
In this study, we established a contusive SCI model and verified ferroptotic phenotype. Then, a comprehensive bioinformatic analysis was conducted based on GEO and FerrDb     database. Ferroptotic DEGs were obtained at 5 time points (days 1, 3, 7, 14, 56). The mTOR, VEGF, HIF-1 signaling pathways, and protein processing were essential in ferroptosis after SCI. Eventually, total 12 hub genes in five time points were identified by PPI network, among which ATF3, XBP1, HMOX1, DDIT3, and CHAC1 were differentially expressed in acute, sub-acute, and chronic phases of SCI. Thus, they were considered as the key genes of ferroptosis following SCI. The differentially expressions of key genes in SCI were verified by western blot. Collectively, the current study contributes to a deeper understanding of SCI-induced ferroptosis and provides novel insights for clinical therapeutic strategy of SCI.
Acknowledgements Schematic was generated using BioRender (www. biore nder. com). We appreciate Dr. Haijian Cai (Center for Scientific Research of Anhui Medical University) for his guidance in performing TEM images.  Fig. 11 The expression validation of key genes. ATF3, XBP1, and HMOX1 were upregulated at day 1; ATF3, XBP1, DDIT3, and CHAC1 were upregulated at day 3; ATF3, XBP1, and HMOX1 were upregulated at day 7 (n = 5). A, B, C Results of WB for key genes.

Author Contribution
E-F Quantitative analysis for WB. Bar represents median, limits of box represent first and third quartile, and whiskers represent minimum and maximum. *p < 0.05, **p < 0.01