Potential novel diagnostic biomarkers of atrial fibrillation: four ferroptosis-related genes linking immune infiltration

Abstract

Background

Atrial fibrillation (AF) is a common atrial arrhythmia in clinic, regulated by the immune system and associated with ferroptosis. We hypothesized that combining the analysis of ferroptosis and immune infiltration in AF will help identify more precise diagnostic biomarkers.

Methods

We analyzed two gene expression omnibus (GEO) data sets (GSE41177 and GSE122188) and extracted characteristic ferroptosis-related genes related to sinus rhythm and AF via bioinformatic analysis. CIBERSORT was used to identify ferroptosis/immune-related genes (FIRGs) in AF. LASSO model analysis was used to identify novel FIRGs. The GSE79768 data set and qRT-PCR were used to validate the FIRGs. ROC curves were then drawn to evaluate the diagnostic power of the FIRGs, and GSEA was used to detect the pathways enriched with the validated FIRGs.

Results

A total of eight FIRGs were identified between the healthy and AF groups through LASSO model analysis. Four FIRGs (ALOX15, SNX5, CA9, and PROK2) were subsequently validated as novel FIRGs with high diagnostic power for AF (AUC = 0.851–0.911). They were enriched mainly in cytokine–cytokine receptor interactions, ascorbate and aldarate metabolism, the nod-like receptor signaling pathway, and the intestinal immune network for IgA production. In addition, ceRNA networks (mRNA–miRNA–lncRNA) such as SNX5-hsa-miR-185-3p-LINC01165 and PROK2-hsa-miR-125b-2-3p-RP11-333E1.2 were constructed. Candidate drugs, such as linoleic acid, which is targeted by ALOX15, and sulfamide, targeted by CA9, were also identified.

Conclusions

Our findings reveal the significant ferroptosis/immune-related genes and the potential pathways and biofunctions enriched with these genes in AF and provide new insights for the diagnosis and interference of AF.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40001-025-03092-3.

Background

Atrial fibrillation (AF), an age-related cardiac arrhythmia, poses significant clinical challenges as it remains incurable and is associated with serious complications, such as tachycardia, cerebral infarction, heart failure (HF), and amaurosis/syncope. Epidemiological data reveal a rising global prevalence, with an estimated 50 million cases in 2020, and a substantial economic burden, costing AF patients up to $63,031 annually in the United States, significantly higher than non-AF individuals. Despite identified risk factors, such as age, alcohol, diabetes, hyperthyroidism, and obstructive sleep apnea, the pathogenesis of AF is not fully clear [1–4].

Recent research has highlighted the roles of immune dysregulation and ferroptosis in AF. Chronic inflammation can induce atrial structural remodeling, including fibrosis and collagen deposition, thereby promoting AF development. Senescent CD8 + T cells have been identified as an immune senescence marker in AF, potentially serving as a progression biomarker [5]. Interleukin-6 (IL-6) is implicated in cardiac hypertrophy and fibrosis, with its trans-signaling activation contributing to AF development. Selective blockade of IL-6 signaling may offer a novel therapeutic strategy. In addition, inflammasomes like NLRP3 are involved in AF, suggesting immune-mediated pathways drive atrial remodeling and arrhythmogenesis [6–8].

Ferroptosis, characterized by iron-dependent lipid peroxidation and reactive oxygen species (ROS) accumulation, plays a critical role in various cardiovascular diseases, including cardiomyopathy, ischemia/reperfusion injury, and heart failure [9, 10]. Myocardial iron overload can lead to oxidative stress and mitochondrial dysfunction, key markers of ferroptosis. Inhibiting ferroptosis may reduce AF susceptibility by balancing iron levels and decreasing ROS production. Moreover, immune cells such as macrophages and neutrophils can release iron-containing compounds and ROS, promoting ferroptosis, which in turn affects immune cell function and cytokine release, creating a vicious cycle of inflammation and tissue damage [11, 12].

Although a close relationship between the occurrence of atrial fibrillation, immunity, and ferroptosis has been continuously reported, the core genes involved in regulating the occurrence of AF and related to immunity and ferroptosis are not clear, and the potential regulatory mechanisms between them are also unclear. By focusing on immune-related genes and ferroptosis-related genes, this study intends to integrate bioinformatics analysis to dig out the core genes involved in the occurrence of atrial fibrillation and explore the potential biological functions, so as to fill this gap and provide ideas and targets for future research.

Methods

Data collection

The whole flow chart of this study is shown in Fig. 1. First, two gene expression data sets (GSE41177 and GSE122188) were downloaded from the Gene Expression Omnibus (GEO) (http://www.ncbi.nlm.nih.gov/geo), combined as training data sets. The GSE79768 data set was used as the validation data set. These data sets were generated via the GPL570-Affymetrix Human Genome U133 plus 2.0 Array [HG-U133_Plus_2]. Among them, GSE41177 contained 16 left atrial appendage (LAA) samples with persistent AF and 3 LAA samples with sinus rhythm (SR). The GSE128188 data set included paired left and right atrial appendages from 5 SR patients and 5 AF patients. The GSE79768 data set consisted of paired LAA and RAA data from 13 patients with persistent AF. The raw GSE41177 and GSE122188 data sets were transformed into an expression value matrix named the training data set via the “limma” package in R software (version 4.3.1) [13]. The batch effects were removed via the “sva” package after the data sets GSE41177 and GSE122188 were merged (Fig. 2a–f) [14]. In addition, ferroptosis-related genes (FRGs) were obtained from the FerrDb website (version 2, http://www.zhounan.org/ferrdb/) for further analysis [15].

Fig. 1.

Flow chart of the current study. miRDB, Targetscan, and miRWalk are three predictive sites for miRNA and gene-targeted binding. Spongescan is a website for predicting the binding of lncRNA and target miRNA. DEGs: differentially expressed genes; DEFRGs: differentially expressed ferroptosis-related genes; GO: gene ontology; KEGG: Kyoto encyclopedia of genes and genomes; LASSO: least absolute shrinkage and selection operator; ssGSEA: single-sample gene set enrichment analysis

Fig. 2.

Removal of batch effect. A, B Boxplot of the expression before and after balance, representing the gene expression of each sample. C, D Density map of the expression before and after balance, representing the overall gene expression of each group. E, F Principal component analysis of the GSE data sets before and after balance, representing the distribution of individuals in each group

Identification of differentially expressed FRGs (DEFRGs)

On the basis of the FRGs and the expression matrix of the training data set, the expression matrix of the ferroptosis-related genes was extracted via the “limma” package. Subsequently, the differentially expressed FRGs (DEFRGs) were identified. The results were visualized as volcano plots and heatmaps via the online tools “Sangerbox” and “pheatmap (version 1.0.12)” R packages (Fig. 3a, d) [16]. DEFRG screening and functional enrichment analysis were performed with cutoff values of P < 0.05 and log₂|fold-change (FC)|> 0.2 [17].

Fig. 3.

Differentially expressed genes (DEGs) analysis and enrichment analysis. A Volcano plot of the DEGs. B GO analysis of the DEGs. The outermost circle represents ontology, the second circle represents the number of genes, and the color depth represents the value of -log₁₀(P value). The third layer represents the number of enriched genes screened in this study. C KEGG pathway analysis of the DEGs. D Heatmap of the DEGs. The top grid is sample clustering; the one on the right is genes; the middle of each is the amount of expression, and the color corresponds to the value. *P < 0.05,**P < 0.01,***P < 0.001

Functional annotation analysis and correlation analysis

After the functional enrichment analysis including gene ontology (GO) and Kyoto encyclopedia of genes and genomes (KEGG) was performed, the results were visualized via the “clusterProfiler (version 3.18.0)” “colorspace (version 2.1–1)” “stringi(version 1.8.4)” “ggplot2 (version 3.5.1)” “circlize (version 0.4.16)” “RColorbrewer (version 1.1–3)” “enrichplot (version 1.24.4)” “ComplexHeatmap (version 2.20.0)” R package. Pathways with P < 0.05 were considered significant. The correlation coefficients of the DEFRG expression levels were subsequently calculated and visualized as a matrix graph via the “corrplot (version 0.94)” package in R.

Identification of characteristic genes via LASSO regression

The LASSO model, a compressed estimation method that obtains a more refined model by constructing a penalty function was utilized to identify novel DEFRGs with high diagnostic value for AF. To screen valuable genes, “glmnet (version 4.1–8)” package in R was used for LASSO regression analysis. It has several advancements: (1) LASSO automatically performs feature selection by shrinking some coefficients to zero, effectively eliminating irrelevant variables from the model, which results in simpler, and more interpretable models. (2) The regularization aspect of LASSO helps prevent overfitting by constraining the size of the coefficients, leading to models that generalize better to new, unseen data. (3) It can manage multicollinearity by selecting one variable from a group of highly correlated variables, reducing redundancy and improving model stability. (4) By focusing on the most relevant variables and reducing noise, LASSO often enhances prediction accuracy compared to traditional regression models. (5) It promotes sparsity by driving many coefficients to zero, resulting in models that are computationally efficient and easier to interpret. (6) We selected the optimal penalty term (lambda) through tenfold cross-validation. By evaluating the cross-validation error for different lambda values, we chose the lambda value that minimized the error, thereby ensuring the stability and predictive power of the model. [18]. Then the intersection of the selected DEFRGs and the immune-related genes for LASSO logistic regression analysis was utilized to identify the FIRGs of AF.

Immune cell subtype distribution in AF

The CIBERSORT algorithm (https://cibersort.stanford.edu/), a widely used technique for calculating the relative abundance of immune infiltrates, was used to determine the relative proportions of the 22 immune cells in the training data set [19]. The distribution ratio and immune score of each immune cell subtype in each sample were automatically calculated via the above online tool. The distribution of each immune cell subtype and the comparison of the distribution ratios between the AF and SR groups were drawn in histograms and boxplots using “ggplot2” package, respectively. We only the samples with a deconvolution p value less than 0.05 for analysis. This was done to ensure the reliability of the results and to avoid errors introduced by poor sample quality.

Gene set enrichment analysis (GSEA) of the FIRGs

GSEA was performed to further analyze the potential mechanisms and pathways of the FIRGs using “limma” “org.Hs.eg.db (version 3.19.1)” “clusterProfiler (version 4.12.6)”, and “enrichplot (version 1.24.4)” packages. Significance thresholds were set as absolute normalized enrichment score (NES) values > 1 and p values < 0.05 [20].

Validation of FIRGs and correlations between hub genes and immune cells

The GSE79768 data set was used to validate the differential expression of the candidate FIRGs. The correlation between the validated novel genes and immune cells, as well as the correlation among the immune cells, were calculated via the Spearman test.

Diagnostic efficacy of novel FIRGs

On the basis of the expression data of the training set, a ROC curve was generated, and the area under the curve (AUC) of the ROC curve was used to estimate the diagnostic efficacy of each novel gene using “glmnet (version 4.1–8)” “pROC (version 1.18.5)” packages. Medcalc version 22.001), a free tool statistical software (MedCalc Software bvba, Ostend, Belgium; https://www.medcalc.org/), was used to estimate the sample size of the AUC [21]. With the following parameters: alpha = 0.05, beta = 0.10, hypothesis AUC = 0.85, null hypothesis value = 0.5, and ratio of sample sizes in negative/positive groups = 1:3, it was estimated that the number of positive cases and negative cases was 7:20. Thus, the sample size of the current study was deemed sufficient.

Quantitative real-time PCR (qRT-PCR) validation of FIRGs in normal subjects and AF patients

Peripheral blood samples (5 ml) were collected from 5 normal controls and 5 AF patients in our hospital from Sep 2023 to November 2023, respectively. Real-time PCR was performed to validate the expression of FIRGs validated by GEO data set GSE79768. The study was approved by the institutional ethics committee, and informed consent was obtained from all participants.

Total RNA was extracted from cells using the TRIZOL reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. The Maxima Reverse Transcriptase was used to reverse-transcribe 550 ng of total RNA into cDNA for use in the qRT-PCR assay (EP0743, Thermo Scientific, USA). Gene expressions were determined by utilized fast real-time PCR set (LightCycler480 II, Rotkreuz, Switzerland). PCR reactions were performed in a total volume of 10 µL containing 5 µL of 2X SGExcel FastSYBR Mixture, 0.2 µL of 10 µM F-primer and 0.2 µL of 10 µM R-primer, 3.6 µL of RNase-Free ddH₂O, and 1 µL of Template (cDNA). The PCR protocol included an initial denaturation at 95 °C for 3 min, followed by 45 cycles of 95 °C for 15 s and 60 °C for 45 s. GAPDH was used as an internal control, and the relative quantification values for CKS2 were calculated by the 2^−△△Ct method. The primers are shown in Table 1.

Table 1.

Primer sequences of the genes

Gene names	Forward primer (5′–3′)	Reverse primer (5′–3′)
SNX5	ACGCACCAAAGCCCTCAT	CAAATTTCTGGCAGCACTCC
PROK2	CTGTGCTGTCAGTATCTGGGTC	CTCTTTCTTCTTTCCTGCCTTC
CA9	GCTGCTTCTGGTGCCTGTC	GGGAGCCCTCTTCTTCTGATT
ALOK15	GGAGCCTTCCTAACCTACAGC	ACCGATAGATGATTTCCCAGAG
GADPH	TGGGTGTGAACCATGAGAAGT	TGAGTCCTTCCACGATACCAA

Detection of drugs targeted by FIRGs

The online Dgibd database (https://dgibd.genome.wustl.edu/) offers user-friendly browsing, searching, and filtering of information on drug–gene interactions and the druggable genome, mined from over thirty trusted sources [22]. The drugs targeted by the candidate FIRGs were identified via this database, and their targeted relationships were visualized as a network via Cytoscape (version 3.6.1) [23].

Construction of the competitive endogenous RNA (ceRNA) network

The FIRG-targeted miRNAs were identified through three online databases: miRDB [24], miRWalk [25], and TargetScan [26]. Only the miRNAs that were targeted in all the above three databases were considered reliable. Moreover, the miRNA-targeted lncRNAs were confirmed via the Spongescan database and Starbase database [27, 28]. The ceRNA networks were then constructed and visualized via Cytoscape.

Results

Screening and functional enrichment analysis of DEFRGs

The training data set was screened for DEFRGs (Fig. 3). The intersection of the up- and down-regulated DEFRGs of the training data set was then calculated, the overlapping DEFRGs were functionally enriched, and the significant GO and KEGG pathways are presented in Fig. 3. Notably, the up-regulated DEFRGs were enriched in pathways related to oxidative stress and immune response, while the down-regulated DEFRGs were associated with cell cycle regulation and apoptosis.

Further screening of characteristic genes with LASSO regression

A total of 138 genes intersecting ferroptosis-related genes and 782 immune-related genes were utilized for LASSO regression analysis (Fig. 4a). Figure 4b, c shows the identified genes and partial likelihood deviance of LASSO regression analysis. The LASSO regression model identified eight genes, SLC38A1, ALOX15, SNX5, CPEB1, CA9, ACOT1, PARP15, and PROK2, were subsequently identified as ferroptosis/immune-related hub genes of AF. According to the heatmap in Fig. 5, these hub genes could clearly separate the samples of the AF group and the SR group.

Fig. 4.

LASSO model of the DEFRGs. A Path plot of lasso coefficients. The e horizontal coordinate above is the number of non-zero coefficients in the model, the horizontal coordinate below is the normalized coefficient vector, the vertical coordinate is the value of the coefficient, and the different colors of the lines represent different variables. B lasso regression analysis cross-validation curve. The top abscissa is the number of variables corresponding to different λ, and the bottom abscissa is log(λ). In general, the horizontal coordinate corresponding to the dotted line on the left is taken as the number of included variables. The smaller the vertical coordinate is, the better the fitting effect of lasso is. According to the above figure, eight genes were selected

Fig. 5.

Immune cells distribution and correlation among the immune cells. A Distribution of the immune cells. The tests on the left is the name of the samples, the balls and tests on the right are the name of the immune cells, and the bands in the middle are the proportions of immune cells. B Correlation among the immune cells. The text on the left is the name of the immune cell, the ball and color represents the correlation coefficient. *P < 0.05,**P < 0.01,***P < 0.001

Profile of the immune cell subtype distribution pattern

The differential expression of immune fractions between the AF group and the SR group was calculated via the CIBERSORT algorithm. The distributions of various immune cell subtypes are shown in Fig. 6a. Figure 6b shows the correlations among the immune cells. The strongest negative correlation was found between M2 macrophages and regulatory T cells (r = − 0.64), dendritic cells activated with M1 macrophages (r = − 0.67), and mast cells activated with resting mast cells (r = − 0.72). A positive correlation was detected between regulatory T cells and CD8 T cells (r = 0.58), memory B cells and naive CD4 T cells (r = 0.44), NK cells and monocytes (r = 0.46), neutrophils and activated memory CD4 T cells, and dendritic cells and activated mast cells (r = 0.40).

Fig. 6.

Violin plot of the expressions of the immune cells in SR individuals and in AF patients. The red violin presents the AF patients, and the blue violin presents the SR individual. The difference with two-side P < 0.05 is defined as significant

As shown in Fig. 6c, the expression level of neutrophils was greater in the AF group than in the SR group, whereas the expression of T-cell regulation was greater in the SR group than in the AF group (P < 0.05 for all). No significant differences were detected in B native cells, CD8 + T cells, NK resting cells, or other cells between the AF group and the SR group (P > 0.05).

Correlations between novel genes and immune cells

The correlations between the hub FRGs and immune cells are presented clearly in Fig. 7. Positive correlations were found between PROK2 (P < 0.001) and neutrophils and memory activated CD4 T cells (P < 0.05), between SLC38A1 and resting mast cells (P < 0.05) and neutrophils (P < 0.05), between PARP15 and resting mast cells and neutrophils (P < 0.05), between CPEB1 and monocyte and activated NK cells (P < 0.05), and between ACOT1 and M1 macrophages (P < 0.001) and CD8⁺ T cells (P < 0.05).

Fig. 7.

Correlation between the FIRGs and immune cells, and among the FIRGs. A Correlation between the FIRGs and immune cells. B Correlations among the FIRGs. Red presents positive correlation, and blue presents negative correlation. The texts on the left and on the top are the name of the immune cells, the ball and color represents the correlation coefficient

In contrast, negative correlations were found between SNX5 and naive B cells (P < 0.001), between SLC38A1 and naive B cells, M0 macrophages, and monocytes (P < 0.05 for all), between CPEB1 and activated NK cells (P < 0.001), between ALOX15 and resting memory CD4 T cells (P < 0.05), and between ACOT1 and active dendritic cells (P < 0.001). CA9 showed no significant correlation with any immune cells (P > 0.05).

Functional annotation of GSEA

GSEA was performed to elucidate the potential mechanism of the validated FIRGs (Fig. 8). ACOT1 gene was associated with terpenoid backbone biosynthesis, the PPAR signaling pathway, and the NOD-like receptor signaling pathway. ALOX15 gene was associated with the chemokine signaling pathway, natural killer cell-mediated cytotoxicity, and primary immunodeficiency. CA9 gene was associated with the chemokine signaling pathway, the intestinal immune network for iga production, and cytokine–cytokine receptor interactions. CPEB1 gene was associated with oxidative phosphorylation, autoimmune thyroid disease, and allograft rejection. PARP15 gene was associated with complement and coagulation cascades, the PPAR signaling pathway, the citrate cycle, and the TCA cycle. The PRKK2 gene was associated with arachidonic acid metabolism, primary immunodeficiency, and the chemokine signaling pathway. SNX5 gene was associated with the chemokine signaling pathway, adipocytokine signaling pathway, PPAR signaling pathway, and cytokine–cytokine receptor interaction. SCL38A1 gene was associated with valine, leucine and isoleucine degradation, peroxisome interactions, and cm receptor interactions.

Fig. 8.

GSEA analysis of the identified FIRGs. Each line presents different pathways the FIRG enriched in. The coordinates on the top left represent enriched fractions; the coordinates of the seat represent the ranking of the corresponding index; the short colored strip in the middle represents the sequence and location of genes; the gray bands below represent the degree of correlation between genes and phenotypes

Validation of the novel gene in the testing set

The identified hub genes were validated in the validation set GSE79768. Figure 9 shows that the expression levels of SNX5 and PROK2 were significantly greater in the AF group than in the SR group (P < 0.05). Conversely, the expression of ALOX15 and CA9 were lower in the AF group than in the SR group (P < 0.05). Unfortunately, the expression of other genes, including ACOT1, CPEB1, PARP15, and SLC38A1, did not significantly differ between the AF group and the SR group (P > 0.05 for all).

Fig. 9.

Validation of the identified FIRGs. A–H Expression of SNX5, PROK2, CA9, ALOK15, RARP15, SLC38A1, CPE81, and ACOT1 gene in SR and in AF patients. The red box presents the expression of FIRG in AF patients, and blue box presents the expression of FIRG in SR individuals. The difference with two-side P < 0.05 is defined as significant

qRT-PCR analysis

The relative FIRG expression levels in peripheral blood cells of AF patients were significantly higher than those in the SR group (SNX5, P = 0.014; PROK2, P = 0.016; CA9, P = 0.005). The expression levels of ALOX15 gene were not significantly different between the two groups (P > 0.05) (Fig. 10).

Fig. 10.

qRT-PCR results of the identified FIRGs. A–D Expression of SNX5, PROK2, CA9, and ALOK15 gene in SR and in AF patients. The blue band presents the expression of FIRG in AF patients, and red band presents the expression of FIRG in SR individuals. The difference with two-side P < 0.05 is defined as significant

Diagnostic efficiency

According to the ROC curve (Fig. 11), the AUC values of SLC38A1, ALOX15, SNX5, CPEB1, CA9, ACOT1, PARP15, and PROK2 were 0.952, 0.851, 0.857, 0.815, 0.911, 0.786, 0.851, and 0.964, respectively (P < 0.05). Given that the sample size was not large enough for ROC curve analysis, we did not validate the diagnostic efficiency of the hub genes in the validation set.

Fig. 11.

ROC curve analysis. ROC curve analysis of SNX5, PROK2, CA9, ALOK15, RARP15, SLC38A1, CPE81, and ACOT1 gene in the training set. Different colored curves represent different genes. The ordinate represents sensitivity and the ordinate represents specificity

Functional annotation of ssGSEA for the FIRGs

SsGSEA was performed to further elucidate the mechanism of the down-regulated FIRGs (Fig. 12a, b). The ALOX15 gene was associated with steroid hormone biosynthesis, ascorbate and aldarate metabolism, cytokine–cytokine receptor interactions, and sulfur metabolism. The CA9 gene was also associated with cytokine–cytokine receptor interactions, systemic lupus erythematosus, the nod-like receptor signaling pathway, the intestinal immune network for iga production, and sulfur metabolism.

Fig. 12.

ssGSEA analysis of the validated FIRGs. A–D Pathways the ALOX15, CA9, SNX5, and PROK2 gene enriched in. The green band presents the down regulated enrichment, and the red band presents the up regulated enrichment. The gray bands represent not significant DEGs

Among the up-regulated FIRGs (Fig. 12c, d), the SNX5 gene was associated with retinol metabolism, steroid hormone biosynthesis, the renin angiotensin system, metabolism of xenobiotics by cytochrome P450, ascorbate and aldarate metabolism, taurine and hypotaurine metabolism, sphingolipid metabolism, and the ppar signaling pathway. The PRKK2 gene was associated with glycosphingolipid biosynthesis globo series, drug metabolism cytochrome P450, and the metabolism of xenobiotics by cytochrome P450.

FIRG-targeted drug network and ceRNA network

According to the Didg database, a total of 11 drugs are targeted by the CA9 gene, while a total of 149 drugs were targeted by the ALOX15 gene. The drug–gene network is shown in Fig. 13. A total of 51 DelncRNAs and 14 DemiRNAs were paired into 58 DelncRNA–DEmiRNA interactions, whereas 108 DemiRNAs and 4 DemRNAs were subsequently matched to form 110 pairs of DemiRNA–DemRNA interactions. Finally, the lncRNA‒miRNA–mRNA ceRNA regulatory network was constructed (Fig. 14). The DemiRNAs with their matching DemRNAs are provided in Table S1. The DelncRNAs and their matching DemiRNAs in the ceRNA network are provided in Table S2.

Fig. 13.

Targeted drugs–FIRGs network. The downward green arrow represents the FIRG, and the red frontal ball represents the drug

Fig. 14.

LncRNA–miRNA–mRNA (ceRNA) network. The red hexagon represents the FIRG, the green triangle represents the miRNA, and the blue rhombus represents the lncRNA

Discussion

In the present study, we employed machine learning to identify ferroptosis-related genes associated with immune infiltration in AF. Among them, SNX5, CA9, SNX5, and PROK2 were validated as potential biomarkers of AF with high diagnostic power, and the mechanism, targeted drugs, and ceRNA networks of these genes were identified. These findings help us understand the associations of ferroptosis and immune infiltration with the development of AF.

Ferroptosis is a crucial mechanism of several diseases, and it is also reported to be associated with the development of AF [29]. A previous study revealed that phosphatidylethanolamine can increase oxidation products and upregulate the expression of ferroptosis-related proteins, which leads to cell death and cardiac fibrosis and promotes the process of AF [30]. In a sepsis-induced AF rat model, ferroptosis increased the intracellular iron concentration and oxidative stress and increased AF vulnerability, possibly because silencing ferroptosis may worsen alterations in the expression of calcium-handling proteins [31]. Yu et al. revealed that icariin can protect against atrial damage by inhibiting ferroptosis via SIRT1 signaling, subsequently inhibiting the development of AF [32]. Thus, ferroptosis can have different effects on the development of AF through several pathways.

In addition, accumulating evidence indicates that inflammatory processes resulting from innate immune responses constitute a cornerstone of AF pathogenesis [33]. Currently, the concept of the “gut–immune–heart” axis has been revealed, and the intricate causal relationship between the gut microbiome and AF has become known [34]. IL-33 recombinant protein treatment increases the expression of Nav1.5, Kv1.5, NCX, and NLRP3 and then promotes atrial remodeling through ST2 signaling [35]. Recently, Maguy et al. reported an autoimmune pathogenesis of AF with direct evidence of Kir3.4 autoantibody-mediated AF, emphasizing the close association of autoimmunity with the occurrence of AF [36]. In addition, a Mendelian randomization study suggested that circulating levels of the cytokines MCP-3, selectin, and FGFBasic are associated with AF [37]. According to these findings, bypassing the important role of the immune system in the pathogenesis of AF is difficult.

In a mouse model, ischemia and reperfusion injury could be mitigated, and cardiac function could be restored after myocardial-specific knockout of ALOX15. 15-Hydroperoxyeicosatetraenoic acid, an intermediate metabolite derived from arachidonic acid by Alox15, was identified as a trigger for cardiomyocyte ferroptosis [34]. In vivo and in vitro, ALOX15-induced lipid peroxidation can increase susceptibility to ferroptosis in asthmatic epithelial cells [38]. Genipin treatment protects against CCl4-triggered acute liver injury by abrogating hepatocyte ferroptosis, and the pharmacological modification of dysregulated GPX4 and ALOX15-mediated lipid peroxidation is responsible for the underlying medicinal effects on a molecular basis [39]. In an early study, interleukin-1 receptor antagonistand ALOX15, whose expression is selectively potentiated in macrophages upon concomitant exposure to apoptotic cells or the liver X receptor agonist T0901317 and Th2 cytokines, were considered anti-inflammatory and pro-resolving genes [40]. Thus, ALOX15 plays different roles in anti-inflammatory, apoptotic and other immune responses. In this study, ALOX15 was positively enriched in cytokine–cytokine receptor interactions and ascorbate and aldarate metabolism, and its expression was down-regulated in AF patients. According to previous study, ALOX15, as a maker of ferroptosis, regulates the infiltrater of neutrophils and monocytes/macrophages in vascular endothelial, resulting in vascular remodeling. In addition, ALOX15 was regulated by miR-774 levels, and then mediated inflammation and iron ferroptosis [41–43]. Thus, the down-regulated expression of ALOX15 may lead to immune activation and weak antioxidant effects, which may promote the cardiac remodeling and development of AF [44].

This study revealed that the FIRG gene CA9 was down-regulated in AF patients. A previous study revealed that the Na⁺/HCO³⁻ cotransporter catalyzes the electrogenic movement of 1 Na⁺:2 HCO³⁻ into the cardiomyocyte cytosol and that CA9 forms a complex that activates NBCe1-mediated HCO³⁻ influx in the myocardium, maintaining the homeostasis of cardiomyocytes [45]. In addition, CA9 knockdown reduced the protective effects of NF-κB inhibition on JTC801-induced cell death and intracellular alkalinization in the PANC1 and MiaPaCa2 cell lines, revealing the role of CA9 in cell death and ferroptosis [46], and CA9 has also been shown to play a role in equilibrating hypoxia, iron metabolism and redox regulation in tumor cells [47]. Moreover, CA9 can inhibit the excessive death of intestinal epithelial cells caused by ferroptosis, preserve the integrity of the intestinal barrier and prevent the progression of unresolved inflammation [48]. Here, CA9 participates in cytokine–cytokine receptor interactions, the nod-like receptor signaling pathway, and the intestinal immune network for IgA production. Though there was no significant association between CA9 and immune cells, it is still widely reported to be involved in the activation of inflammation or immune responses, which may be performed through other ways like above that do not directly regulate immune cells, and then affects the process of AF.

Macroautophagy/autophagy is an important form of immune response. The inner membrane of the autophagosome and the enclosed substrates are degraded and transported out of the lysosome for recycling. Autophagosome outer membrane components are not degraded but are recycled through an unidentified process, which is composed of sorting nexin 4 (SNX4), SNX5, and SNX17, which cooperate with the dynein–dynactin complex to mediate autophagosomal component recycling, regulating the immune response [49–51]. In addition, SNX4 and SNX5 form a heterodimer that recognizes autophagosomal membrane proteins and is required for generating membrane curvature on autolysosomes, both via their BAR domains, to mediate the cargo sorting process [51]. Caveolin-1 (CAV1) and macropinocytosis-related protein sorting nexin 5 (SNX5) are associated with endocytosis, and CAV1- and SNX5-knockout experiments revealed that both caveolae-mediated endocytosis and macropinocytosis occur [52]. In this study, SNX5 was negatively enriched in the renin–angiotensin system (RAAS), ascorbate and aldarate metabolismin AF patients. RAAS was involved in blood pressure regulation, which is an important regulatory pathway of cardiac remodeling. RAAS inhibitor has been widely used in clinic, significantly improved the prognosis of patients with heart failure and cardiac remodeling [53, 54]. In addition, it was negatively correlated with naive B cells. In mice research, the level of naïve B cells was strongly associated with glucose intolerance in Type 1 diabetes mice, which increased the humoural immunity. In single-cell sequencing analysis, naive B cell was notably a significant subgroup of pathological myocardial remodeling, associated with up-regulation of oxidative phosphorylation, and leukocyte extravasation [55, 56]. Thus, up-regulated SNX5 expression may affect the process of AF through regulating atrial tension via the renin–angiotensin system or oxidative stress.

The last validated FIRG is PROK2. Under pathological conditions, prokineticin 2 can induce the proliferation, migration, and angiogenesis of endothelial cells, suggesting that this molecule plays a role in tumor growth, angiogenesis, and metastasis [57]. Notably, researchers confirmed the differential expression of the cytokines IL32 and PROK2 in an independent cohort and provided functional validation of the opposing effects of these two cytokines on collagen-induced platelet aggregation [58]. In addition, PROK2 prevents neuronal cell death by suppressing the biosynthesis of lipid peroxidation substrates, arachidonic acid–phospholipids, via accelerated F-box only protein 10 (Fbxo10)-driven ubiquitination, degradation of long-chain-fatty-acid-CoA ligase 4 (Acsl4), and inhibition of lipid peroxidation [59]. Pressure overload-mediated PKOR2 signaling in cardiomyocytes contributes to cardiac hypertrophy through autocrine signaling and vascular rarefaction via cardiac cytokine-mediated cardiomyocyte–endothelial cell communication [60]. Moreover, TBX20 can be considered a novel transcription factor that regulates angiogenesis through the PROK2–PROKR1 pathway to relay and sustain the proangiogenic effect of vascular endothelial growth factor [61]. Ceramides and sphingomyelins with palmitic acid are associated with increased AF risk (ceramide with a palmitic acid hazard ratio of 1.31; 95% CI 1.03–1.66; sphingomyelin with a palmitic acid hazard ratio of 1.73; 95% CI 1.18–2.55) [62]. In addition, PROK2 is down-regulated in idiopathic pulmonary fibrosis, involved in ferroptosis, neutrophil recruitment, and fibrosis formation as a neutrophil-related gene. In bronchopulmonary dysplasia tissue, PROK2 was up-regulated, negatively associated with neutrophils and resting memory CD4 T cells and related to ferroptosis, involved in endothelial dysplasia [63, 64]. These findings suggest the important role of PROK2 in the immune system. In this study, it was significantly more highly expressed in AF patients, positively associated with neutrophils and resting memory CD4 T cells. AF cannot be maintained without atrial fibrosis, and endometrial injury or hypoplasia can lead to myocardial fibrosis and remodeling, increasing susceptibility of AF. What’s more, PROK2 was enriched in metabolism of xenobiotics by cytochrome p450, drug metabolism cytochrome p450, and glycosphingolipid biosynthesis globo series. It was observed that the use of cytochrome P450 inhibitors was positively correlated with the concentration of anticoagulants in AF patients. In animal studies, nuclear factor–erythroid-2-related factor 2 in combination with warfarin reduced atrial fibrosis in AF, which may partly explain our findings [65, 66].

Immune cells are crucial participants in the immune system and play strong immune modulatory functions. An early study suggested that fibrosis-independent atrial fibrillation in older patients is driven by substrate leukocyte infiltration and is associated with the neutrophil-to-lymphocyte ratio [67]. Another study revealed that multiple systemic inflammatory indicators, including neutrophils, are strongly associated with the onset of AF [68]. Moreover, neutrophil extracellular traps (NETs) are DNA fragments with cytoplasmic proteins released from neutrophils that are involved in various cardiovascular diseases. He et al. demonstrated the lethal effects of NETs on cardiomyocytes through the induction of mitochondrial injury and autophagic cell death, which comprehensively described the positive feedback comprising NETs and stimuli secreted by cardiomyocytes that sustain the progression of AF and atrial fibrosis. Moreover, an enlarged LA diameter and a reduced LVEF are associated with increased neutrophil extracellular traps in AF patients [69, 70].

In addition to neutrophils, regulatory T cells are also considered to be closely related to AF, and Bacteroides fragilis can prevent aging-related AF via regulatory T-cell-mediated inflammation in rat models [71, 72]. In periodontitis rat models, AF susceptibility and atrial fibrosis are significantly increased, associated with up-regulated RGS1 expression, which is positively correlated with regulatory T cells [73]. In addition, single-cell transcriptomes from human atria revealed inflammatory monocyte and SPP1⁺ macrophage expansion in AF, which suggests that SPP1 + macrophages may be targets for immunotherapy in AF [74]. In this study, T-cell regulation was greater in the SR group than in the AF group, whereas neutrophil expression was greater in the AF group than in the SR group and was positively correlated with the PROK2 gene. Our findings suggest that neutrophils and regulatory T cells play essential roles in the development of AF and that interference with AF should focus on the interaction between PROK2 and neutrophils. Our analysis revealed that ALOX15, SNX5, CA9, and PROK2 were significantly differentially expressed in AF and control groups, positively or negatively correlated with different types of immune cells, involved in atrial remodeling, development of AF by regulating ferroptosis, which has been implicated in myocardial fibrosis, metabolism, and other immune pathways. They have high power for the diagnosis of AF, which suggests that they are novel biomarkers for the diagnosis of AF and potential targets for the interference of AF through ferroptosis/immune-related pathways or biofunctions.

There are several limitations of this study. First, the sample size is not large enough, even though the test power has been confirmed by statistical software. Given the limited sample size of the test set and the lack of extensive clinical validation, the diagnostic conclusions drawn from the identified FIRGs should be interpreted with caution. Further validation in larger and more diverse clinical cohorts is necessary to confirm their diagnostic potential. Second, the mechanism of AF is complex, and we analyzed only the current data; several unknown but potential factors were not included in this study. Finally, further experiments are needed to verify our findings in vivo and in vitro.

Conclusion

We identified a correlation between eight ferroptosis-related genes whose expression differed between the AF group and the SR group, suggesting an inter-relationship between ferroptosis and immune pathways in the development of atrial fibrillation. In addition, via bioinformatics analysis, we screened 4 ferroptosis/immune-related genes as potential biomarkers for the diagnosis of AF. On the basis of the ceRNA network, we also identified potential targeted drugs and targeted ceRNA networks for the selected FIRGs. The ROC curve demonstrated the significant predictive value of the novel FIRGs. Furthermore, ALOX15, SNX5, CA9, and PROK2 may be potential biomarkers for AF diagnosis, which provides a promising foundation for further validation studies. However, as sample size of the test set is relatively small, our findings still need to be verified through larger samples or multi-level in-depth studies.

Abbreviations

AF

Atrial fibrillation

SR

Sinus rhythm

FRG

Ferroptosis-related genes

FIRGs

Ferroptosis/immune-related genes

DEFRGs

Differentially expressed FRGs

HF

Heart failure

PV

Pulmonary vein

LA

Left atrium

IL-6

Interleukin-6

ROS

Reactive oxygen species

LAA

Left atrial appendage

WGCNA

Weighted correlation network analysis

DEG

Differential expression gene

GSEA

Gene set enrichment analysis

LASSO

Least absolute shrinkage and selection operator

GEO

Gene expression omnibus

PCA

Principal component analysis

BP

Biological process

CC

Cytological components

MF

Molecular functions

GO

Gene ontology

KEGG

Kyoto encyclopedia of genes and genomes

NES

Normalized enrichment score

ceRNA

Competitive endogenous RNA

FC

Fold-change

ROC

Receiver operating characteristic

AUC

Area under curve

Author contributions

Y.-Z. G. and H. L. conceived the study, participated in the design, performed the statistical analyses, and drafted the manuscript. H. W., H. -J. H., and D.-Y. L. conceived the study, participated in the design and helped to draft the manuscript. X.-Y. L. and D.-S. L. verified the results and revised the manuscript. All authors gave final approval and agree to be accountable for all aspects of work ensuring integrity and accuracy.

Funding

This study was supported by the Natural Science Foundation of Guangxi Zhuang Autonomous Region (2019GXNSFAA245099).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Informed consent for patient information to be published in this article was not obtained, because the data sets presented in this study can be found in online platform. As this study is based on an analysis of publicly available data sets, there is no need for the approved by the ethics committee.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.