Identification of new hub- ferroptosis-related genes in Lupus Nephritis

Abstract

Background: Lupus Nephritis (LN) is the primary causation of kidney injury in systemic lupus erythematosus (SLE). Ferroptosis is a programmed cell death. Therefore, understanding the crosstalk between LN and ferroptosis is still a significant challenge. Methods: We obtained the expression profile of LN kidney biopsy samples from the Gene Expression Omnibus database and utilised the R-project software to identify differentially expressed genes (DEGs). Then, we conducted a functional correlation analysis. Ferroptosis-related genes (FRGs) and differentially expressed genes (DEGs) crossover to select FRGs with LN. Afterwards, we used CIBERSORT to assess the infiltration of immune cells in both LN tissues and healthy control samples. Finally, we performed immunohistochemistry on LN human renal tissue. Results: 10619 DEGs screened from the LN biopsy tissue were identified. 22 hub-ferroptosis-related genes with LN (FRGs-LN) were screened out. The CIBERSORT findings revealed that there were significant statistical differences in immune cells between healthy control samples and LN tissues. Immunohistochemistry further demonstrated a significant difference in HRAS, TFRC, ATM, and SRC expression in renal tissue between normal and control groups. Conclusion: We developed a signature that allowed us to identify 22 new biomarkers associated with FRGs-LN. These findings suggest new insights into the pathology and therapeutic potential of LN ferroptosis inhibitors and iron chelators.

Keywords:

• hub-ferroptosis-related genes
• ferroptosis
• lupus nephritis
• systemic lupus erythematosus
• machine learning

1. Introduction

Systemic lupus erythematosus (SLE) is a systemic autoimmune disease with multiorgan involvement and multiple autoantibodies [1], it is a worldwide disease that mainly impacts women in their reproductive years [2]. The aetiology of SLE is exceedingly complicated and has not been clear, which may be related to genetic susceptibility [3,4], environmental triggers [5,6], apoptosis [7,8], complement cascade [9], DNAse1 [10], and sex hormones [11]. Lupus nephritis (LN) commonly arises as a serious complication of SLE. Because LN has high morbidity and mortality, early diagnosis and intervention can significantly reduce chronic kidney disease (CKD) and end-stage renal disease (ESRD). Within 5 years of diagnosis, approximately 10–20% of patients of LN will develop to ESRD [12]. Studies try to make sure of the pathogenesis and molecular mechanisms of LN, but it is still uncertain. So, further exploring the mechanisms and characteristics of the disease is imperative.

With machine learning (ML) development, people can implement medical frameworks or disease activities. ML has shown capability and affected clinical decision-making in multiple fields, predicting mortality [13], response to biological agents [14], and disease activity [15].

Ferroptosis is a type of iron-dependent cell death that occurs when there is a harmful accumulation of lipid peroxides on cellular membranes [16,17]. Two major biochemical mechanisms define its characteristics: iron accumulation and lipid peroxidation [18]. At present, mounting evidence has demonstrated that multiple signalling pathways can mediate ferroptosis, including the canonical cystine/GSH/GPX4 regulated ferroptosis pathway, GPX4-independent pathway, lipid metabolism, iron metabolism, cell metabolism, AMPK signalling pathways, and so on [19]. It has been shown that various organelles drive iron death, including mitochondria, lysosomes, endoplasmic reticulum (ER), lipid droplets (LDs), peroxisomes, golgi apparatus, and nucleus [20]. Ferroptosis is characterised by small mitochondrial size, concentrated mitochondrial density, reduction or disappearance of mitochondria, and rupture of outer mitochondrial membrane (OMM) [21].

Ferroptosis is closely linked to multiple physiological and pathological processes, including tumour development [22], nervous disease [23], and cardiovascular disease [24], suggesting its potential as a therapeutic target.

Until now, few articles have identically studied the relationship between LN and Ferroptosis. Li et al. found that neutrophil ferroptosis results in neutropenia and disease symptoms in SLE [25]. Marks [26] reported that LN appears renal iron accumulation, and iron chelate delays the appearance of proteinuria. So, studies on ferroptosis and LN remain limited. Here, we try to use ML to discover the significant genes playing essential roles in the pathogenesis and anticipate the curative potential of ferroptosis for LN to provide insights into ferroptosis and LN.

2. Materials and Methods

2.1. Data download

We used the “GEOquery” package in R (Version 4.10) to download the searched glomerulus biopsy samples expression profile datasets GSE32591, GSE127797, GSE69438, GSE99967, and GSE113342 from the Gene Expression Omnibus (GEO) database https://www.ncbi.nlm.nih.gov/geo/.GSE200306 (which includes 79, health control 15) was extracted for confirmatory studies. We downloaded FRGs from the public FerrDb database http://www.zhounan.org/ferrdb [27]. Our data have been processed by SVA package to remove the batch effect.

2.2. Data Processing and DEG Screening

The expression matrices of GSE32591, GSE127797, GSE69438, GSE99967, and GSE113342 were downloaded and combined. Glomerulus biopsy samples from 210 patients with LN and 46 healthy controls were evaluated and selected according to the data. The “limma" package and "ggplot2" software were used to scan for the DEGs. Then, the overlapping genes between DEGs and FRGs were identified. Finally, 273 FRGs were obtained for further study.

2.3. Functional analysis

Gene ontology (GO) and Disease Ontology (DO) enrichment analyses were conducted on the DEGs using the "clusterProfiler" package. The same package was also employed to perform KEGG pathway enrichment analyses on the gene expression matrix. Significance was determined on the basis of a false discovery rate < 0.25 and p < 0.05.

2.4. Screening and verification of target FRGs in LN

To develop a consensus FRGs-LN model with high accuracy and stability, we integrated 12 ML algorithms and 113 algorithm combinations. The integrated algorithms used were Lasso, Ridge, Enet, Stepglm, SVM, glmBoost, LDA, plsRglm, RandomForest, GBM, XGBoost, and NaiveBayes. The signature generation procedure consisted of the following steps: (a) Univariate logistic regression identified binary classification behaviour mRNAs in the training data (including GSE32591, GSE127797, GSE69438, GSE99967, GSE113342); (b) Next, 113 algorithm combinations were applied to the prognostic mRNAs to create prediction models using the leave-one-out cross-validation (LOOCV) framework in the training data; (c) All models were evaluated using the validation dataset (GSE200306); (d) For each model, the C-index was calculated across all validations.

2.5. Construction of protein–protein interaction (PPI) network and module analysis

We utilised the STRING database to analyse and visualise interactions among various FRGs http://string-db.org/ [28].

2.6. Correlation between the selected FRGs-LN with 24h-proteinuria and serum creatinine

We used the GSE200306 dataset to apply Spearman correlation matrix analysis, analysing the markers in relation to 24h-proteinuria and serum creatinine. Afterwards, we visualised the results using the "ggplot2" package.

2.7. Assessment of immune cell infiltration

We run the algorithm locally using the CIBERSORT R-packages and the LM22 document [15]. Samples with a p-value less than 0.05 were filtered out, resulting in the generation of an immune cell infiltration matrix. The “corrplot” package was employed to create a correlation heatmap, which visualised the relationships among the infiltrated immune cells. Additionally, the "ggplot2" package generated diagrams to visualise correlations in immune cell infiltration.

2.8. Correlation between FRGs-LN and immune-infiltrated cells in LN

To analyse the correlation between markers and infiltrating immune cells, we conducted Spearman correlation matrix analysis. The results were visualised using the "glue" package.

2.9. Immunohistochemistry

We selected HRAS, TFRC, ATM, and SRC for further Immunohistochemistry in renal tissue between normal and control groups.

3 Results

3.1. Data processing and DEGs screening

Initially, we combined the expression profile datasets GSE32591, GSE127797, GSE69438, GSE99967, and GSE113342, resulting in a total of 10,619 genes. Following data normalisation, we utilised R-project to extract 466 differentially expressed genes (DEGs) from the gene expression matrix. The volcano map (Figure 1A) visually represents these DEGs. Additionally, the heatmap of the top 50 DEGs illustrates the differential expression patterns between average and LN samples (Figure 1B).

Figure 1. (A) volcano plot, |Log2FC|≥0.107 and p < 0.05 were chosen as filtering conditions, scanned 264 upregulated genes and 202 downregulated genes. (B) Heatmap of top 50 different expressed genes.

3.2. Functional correlation analysis set enrichment analysis was employed to identify different functional phenotypes

The results revealed that DEGs were mainly related to associated with biological processes (BP), including defense response to virus, defense response to symbiont, response to virus, type 1 interferon signalling pathway, cellular response to type 1 interferon, negative regulation of viral process, negative regulation of viral genome replication, response to type 1 interferon (Figure 2A). The GO analysis of DEGs revealed several cellular components (CC), including collagen-containing extracellular matrix, external side of plasma membrane, external encapsulating structure, extracellular matrix, blood microparticle, specific granule, secretory granule lumen, cytoplasmic vesicle lumen (Figure 2B). The GO analysis results for molecular function (MF) encompassed Gene included pattern recognition receptor activity, lipopeptide binding, extracellular matrix structural constituent conferring tensile strength, immunoglobulin binding, Toll-like receptor binding, extracellular matrix structural constituent, glycosaminoglycan binding, and tetrapyrrole binding (Figure 2C). DO analysis revealed that DEGs were mainly associated with lung disease, kidney disease, urinary system disease, bacterial infectious disease, primary bacterial infectious disease, obstructive lung disease, coronary artery disease, and myocardial infarction (Figure 2D).

Figure 2. GO and DO enrichment analysis of DEGs. (A) BP enrichment analysis. (B) CC enrichment analysis. (C) MF analysis. (D) DO enrichment analysis.

KEGG pathway enrichment (Figure 3) showed that the top 10 pathways of up DEGs are coronavirus disease COVID-19, influenza A, Epstein–Barr virus infection, NOD-like receptor signalling pathway, chemokine signalling pathway, lipid and atherosclerosis, tuberculosis, pertussis, complement and coagulation cascades, staphylococcus aureus infection and top 10 down DEGs were metabolism of xenobiotics by cytochrome P450, chemical carcinogenesis receptor activation, steroid hormone biosynthesis, chemical carcinogenesis DNA adducts, drug metabolism cytochrome P450, bile secretion, drug metabolism other enzymes, protein digestion and absorption, carbon metabolism, and cholesterol metabolism.

Figure 3. KEGG analysis of DEGs.

3.3. Screening of hub LN-FRGs for LN

Models obtained from 113 algorithms on the training set, evaluation results on the test set (or including the training set), and heat map display of evaluation of each model. Then, using the results from integrated ML, we examined each gene’s diagnostic effect in LN samples. The prediction model constructed by stepglm method has the best performance in both the training set and test set, and this model has a leading C-index in all validation datasets. The highest average C-index reaches 0.927 (Figure 4A)

Figure 4. A. prediction performance of each machine learning model in training data and testing data, the models were arranged by average AUC in each dataset. B. PPI network of targeting genes.

3.4. PPI network construction

We utilised the STRING database to analyse and visualise the interactions between the 22 FRGs-LN (Figure 4B).

3.5. Correlation between the selected FRGs-LN with 24h proteinuria and serum creatinine

Using the data of the GSE200306, the correlation between the selected FRGs-LN with 24h-proteinuria and serum creatinine was calculated. CD44, PML, CYBB, PTPN6, STAT3, and TGFB1 were positively related to 24h-proteinuria, IFNA2 and ZEB1 were negatively related to 24h-proteinuria. CD44, CYBB, SRC, and TGFB1 were positively related to serum creatinine (Figure 5).

Figure 5. Correlation between targeting genes and clinical factors (proteinuria and serum creatinine (SCr).

3.6. Differential analysis of immune-infiltrated cells in LN and correlation between FRGs-LN and immune-infiltrated cells in LN

We use the CIBERSORT tool to estimate the proportion of immune cells in LN and normal tissues. The results, depicting the estimated proportion of immune cells for each sample, are illustrated in (Figure 6A, B). Upon calculating the immune cell composition in both the LN and control groups, we observed significant statistical differences in several immune cell types. These include Macrophages M1, T cells regulatory Tregs, B cells naive, monocytes, Macrophages M2, B cells memory, Macrophages M0, and T cells CD4 memory resting, as shown in (Figure 6C).

Figure 6. (A) Immune infiltrated cells composition in each sample. (B) immune infiltrated cells heatmap in different samples. (C) immune infiltered cell in different groups (lupus vs. control).

Further, we study the 22 different FRGs with immune cells. For example, ATM was positively related to mast cells resting, T cells CD4 memory resting, B cells memory, plasma cells, Macrophages M2 and negatively related to neutrophils, monocytes, NK cells resting, Macrophages M0, and B cells naive. HRAS was positively related to monocytes, NK cells resting, neutrophils, Macrophages M1, B cells naive and negatively related to plasma cells, B cells memory, mast cells resting, T cells CD4 memory resting, NK cells activated. SRC was positively related to plasma cells, B cells memory, T cells CD8, Macrophages M0, Mast cells resting, and negatively related to monocytes, NK cells resting, MacrophagesM1, neutrophils, Macrophages M2, B cells naive. TFRC was positively related to T cells CD4 memory resting, Macrophages M1, Macrophages M0, neutrophils and negatively related to T cells regulatory Tregs.

3.7. Immunohistochemistry

Further, immunohistochemistry demonstrated a significant difference in HRAS, TFRC, ATM, and SRC expression in renal tissue between normal and control groups (Figure 7B).

Figure 7. (A) HE, PAS, MASSON, PSAM staining of Lupus Nephritis and normal renal tissue. (B) Targeting gene expression in Lupus nephritis and normal renal tissue through immunohistochemical staining.

4. Discussion

LN is one of the manifestations of SLE, which is regarded as the most severe one. Despite its significance, the exact mechanisms underlying the pathogenesis of LN remain poorly understood. Studies have revealed that the pathogenesis of LN includes humoral and cellular immunity [29], including various autoantibodies that form immune complexes [30] activated B-cell, infiltration T cells, macrophages released [31,32], neutrophil dysregulation [33]. However, only autoimmunity cannot fully clarify the pathogenesis of the disease. SLE can be caused by the accumulation of cell remnants resulting from various pathways of cell death, for example, necrosis, apoptosis, and necrosis [34].

Ferroptosis is a programmed cell death that occurs independently of apoptosis. It is characterised by the accumulation of iron and lipid peroxidation, leading to the disruption of intracellular redox balance [35,36], iron-dependent ROS is the primary cause of necroptotic cell death [37]. Excessive oxidative stress in SLE contributes to various detrimental effects including abnormal cell death signals, immune dysfunction, autoantibody production, and fatal comorbidities [38,39].

Because of the complex occurrence and regulatory mechanism of ferroptosis, we postulate that there should be a mysterious link between ferroptosis and the pathogenic process in inflammation, immune dysfunction, and clinical manifestations in lupus.

The study identified 10619 DEGs screened from the LN biopsy tissue in the microarray datasets. Using R-project, a total of 466 DEGs were extracted from the gene expression matrix. This is illustrated in the volcano map, which highlights 264 upregulated genes and 202 downregulated genes. First, GO enrichment analysis revealed that DEGs were mainly correlative to defense response to virus, defense response to symbiont, response to virus, type 1 interferon signalling pathway, and cellular response to type1interferon. These results reveal that immune activity plays a vital role in LN processes. KEGG enrichment analysis revealed that up-regulated DEGs mainly belonged to Coronavirus disease-COVID-19, Influenza A, Epstein–Barr virus infection, NOD-like receptor signalling pathway, chemokine signalling pathway, lipid and atherosclerosis, tuberculosis, pertussis, complement and coagulation cascades, Staphylococcus aureus infection. During the progression of LN, it has been observed that there are alterations in immune infiltration patterns and changes in immunomodulatory genes. These findings highlight the dynamic nature of the immune response in LN pathology. So, these all follow the upper pathogenesis of LN we have discussed.

Next, the crossover genes between DEGs and FRGs were identified. Finally, 273 FRGs-LN were obtained. In our study, we conducted an integration of 12 different ML algorithms and explored a total of 113 algorithm combinations. By evaluating the performance of each model, we identified the one with the highest average C-index as the optimal model for our analysis. Finally, about 22 hub-FRGs-LN were identified. They are CYBB, TP53, HRAS, ATG5, ATG7, MAPK1, ZEB1, SOCS1, TGFBR1, ATM, CD82, IL1B, TFRC, IFNA2, PTPN6, MUC1, SRC, STAT3, PML, CDKN1A, CD44, TGFB1. We analysed the interactions among various hub-FRGs-LN through the use of STRING database.

CIBERSORT was implemented to assess immune cells in LN tissues and health control. The 22 FRGs-LN was also studied for infiltration of immune cells. Firstly, we found that the immune cell composition in both LN and control groups was statistically different. LN is positive with Macrophages M1, monocytes, Macrophages M2, memory B cells and negatively related with cells regulatory Tregs, B cells naive, T cells memory CD4 resting, and Macrophages M0. Iron is essential in B cell maturation, germinal centre formation, and immune responses [40]. Using elastin to cause Lipid peroxidation can promote the proliferation and differentiation of B cells and natural killer (NK) cells [41]. Therefore, ferroptosis can regulate the immune system and promote lupus progression.

However, these differences in immune cell composition ratios are only one aspect of the common pathogenesis in LN. We still need to confirm whether these 22 shared hubs FRGs-LN are associated with immune infiltration. These expressed proteins by these FRGs-LN are also distributed in various immune cells, including T cells CD4 memory resting, B cells memory, plasma cells, Macrophages M1, Neutrophils, Monocytes, NK cells resting, and B cells naive.

Ferroptosis, a form of regulated cell death, has been found to release a diverse array of cytokines and chemokines. This release of inflammatory mediators ultimately contributes to tissue injury and the initiation or perpetuation of inflammation [42], such as ROS via pro-inflammatory change can facilitate inflammatory disease [43], and iron accumulation induces a direct polarisation of macrophages towards a pro-inflammatory profile [44]. So ferroptosis in LN may result in organ damage and clinical manifestations through excessive inflammation, exerting its pathogenic effect.

Meanwhile, the correlation between the selected FRGs-LN with 24h-proteinuria and serum creatinine was calculated. CD44, PML, CYBB, PTPN6, TGFB1 and STAT3 were positively related to 24h-proteinuria. IFNA2 and ZEB1 were negatively related to 24h-proteinuria. CD44, CYBB, SRC, and TGFB1 were positively related to serum creatinine.

Among the 22 hub-FRGs-LN, we selected HARS, TFRC, ATM, and SRC for further identification, because they are in the center of the PPI figure and have more relationships with others. From the existing articles, there are relatively few studies on the relationship between these genes and lupus nephritis, however, other genes have been extensively studied. We selected LN kidney tissue and standard control for immunohistochemistry. The results suggest these four genes are upper expression in the LN group.

TFRC, which encodes the transferrin receptor, plays a crucial role in cellular iron uptake. When there is an increase in unstable iron pools, it can lead to the oxidative stress. This oxidative stress can then trigger the initiation of ferroptosis [45]. For example, increased TFRC will cause reactive oxygen species (ROS), decreased glutathione peroxidase 4 (GPX4), increased lipid peroxidation, and the presence of increased ferrous iron (Fe2+) can lead to higher levels of cell ferroptosis [46].

In NZB/WF1 mice, the PTEC showed a lower expression of transferrin receptor one and increased ferritin expression. Treating with deferiprone delayed the onset of albuminuria, markers of tubular injury, and renal function without change of anti-dsDNA IgG levels [26]. Studies in MRL/lpr mice, which are a spontaneous model of SLE and LN, have shown that exogenous administration of hepcidin can reduce renal iron accumulation, labile iron content, and injury parameters [47]. Mice with myeloid-specific Gpx4 haploid deficiency showed manifestations resembling lupus, which was relieved by treatment with a targeted ferroptosis inhibitor [25].

In humans, some iron homeostasis-related proteins have also been identified to be urinary SLE biomarkers, such as ferroxidase enzyme ceruloplasmin [48], NGAL [48], and transferrin [49]. Some end products of lipid peroxidation were also increased and positively correlated with SLE activity [50,51]. TFRC should be essential during the ferroptosis process, but the detailed and precise mechanism should be studied further.

ATM (Ataxia-telangiectasia-mutated proteins) is a critical protein kinase in DNA damage repair and can participate in downstream pathways that activate the cell cycle. Chen et al. found cystine deprivation, or ATM-elastin inhibitor, rescued multiple cancer cells from ferroptosis. ATM inhibition regulates the mRNA levels of iron regulators and reduces cellular levels of labile iron [52]. Numerous ATM inhibitors have been formulated as radiosensitizers to boost the effectiveness of radiation therapies [53], because the study found that ionising radiation can induce ferroptosis [54]. The presence of ATM is essential for cellular inflammatory toxicity and cell death induced by oxidative stress [53]. Researchers have discovered that inhibiting ferroptosis and promoting cell survival through the PI3K/AKT pathway and ATM could be beneficial in the context of osteoporosis (OP) [55].

DNA double-strand break (DSB) is the most severe and threatening damage. ATM remains inactive as a homodimer until DNA damage occurs [56]. Upon DSB, ATM becomes activated at DSB lesion sites, leading to activation of downstream effectors [57] to mediate the p53 phosphorylation, epigenetic DNA repair, chromatin remodeling, and transcription inhibition, respectively [58,59]. Meanwhile, one important mechanism by which ATM safeguards the genome from DSB damage is through the activation of p53 [60].

The Src family kinases (SFKs) are a group of non-receptor protein tyrosine kinases that play important roles in intracellular signal transduction. The SFKs include 11 members, with Src being the prototypical member. When activated, SFKs phosphorylate specific tyrosine residues in other substrate proteins, leading to the initiation of various signalling pathways. Some of the substrate proteins that are commonly phosphorylated by SFKs include NF-κB, STAT3, AKT (protein kinase B), and MAPK [61–63], and result in inflammation, oxidative stress, apoptosis, autophagy, and ER stress et al. [64].

SFKs have been implicated in some kidney diseases, including LN [65]. Inhibition of Src family kinases ameliorates LPS-induced acute kidney injury [66]. Also, in LPS-induced septic AKI mice, fisetin alleviated kidney inflammation and apoptosis via inhibiting Src-mediated NF-κB and MAPK signalling pathways [67].

Inhibition of Src reduced cell viability, which ferrostatin-1 and other ferroptosis inhibitors rescued. The α6β4-mediated activation of Src suppresses lipid peroxidation [68]. All these previous studies give us more think about the ferroptosis function of SRC.

HRAS, NRAS, and KRAS, collectively referred to as oncogenes, are the most common mutation-driven proto-oncogenes in cancer. RAS coordinates reprogramming lipid metabolism and promotes the production of reactive oxygen species (ROS) in cells. In addition to increased cellular iron intake, cancer cells with mutant RAS can increase cellular iron intake and reduce iron storage capacity [69]. Since these metabolic changes are critical for ferroptosis, there should be a correlation between HRAS and ferroptosis.

Altogether, we discuss some of the above candidates FRGs-LN in the function of ferroptosis, others are also famous during the course, such as p53 being involved in polyamine metabolism, which affects cellular growth, development, and programmed cell death [70]. STAT3 may jointly regulate FRG expression with NRF2 [71]. So, the dysregulation or inadequate availability of iron can contribute to ferroptosis. In the case of patients with LN, ferroptosis has been implicated in the development of certain characteristic features such as proteinuria, neutropenia and the production of anti-ds DNA antibodies. Since the above data comes from previous studies, it should be valid but limited, new and intensive studies should be given further for identification.

5. Conclusion

In conclusion, we identified higher expression levels of 22 hub-FRGs-LN through ML in LN kidney biopsy tissue. Traditionally, LN treatment includes glucocorticoids, immunosuppressants, and biological agents that act on T and B cells, which may cause substantial side effects to patients if used for a long time. Our findings may become a therapeutic target with great promise, and the potential of ferroptosis inhibitors and iron chelators may be a new therapeutic selection in LN. Further research is required to validate these results in the future.

Author contributions

X.-J.Z. performed software operations and wrote the manuscript. Y.C. participated in statistical analysis. L.Y. and X.-L.L. provided clinical data of L.N. D.S. and Y.-Q.L. participated in research design and revised the manuscript.

Ethics approval and consent to participate

This study was approved by the Ethics Committee in The First Affiliated Hospital of China Medical University (approve number: 2022-306-2), and the patients involved signed the informed consent form.

Disclosure statement

The authors confirm that the study was conducted independently, without any commercial or financial relationships.

Data availability statement

We can acquire the data from GEO Database (GSE32591, GSE127797, GSE69438, GSE99967, GSE113342 and GSE200306) and FerrDb database.

Additional information

Funding

This work was supported by National Natural Science Foundation of China (grant no. 81273297), Shenyang Science and Technology Plan. Public Health R&D Special Project [21-173-9-67].