Identification of Ferroptosis-Related Gene in Age-Related Macular Degeneration Using Machine Learning

Abstract

Background: Age-related macular degeneration (AMD) is a major cause of irreversible visual impairment, with dry AMD being the most prevalent form. Programmed cell death of retinal pigment epithelium (RPE) cells is a central mechanism in the pathogenesis of dry AMD. Ferroptosis, a recently identified form of programmed cell death, is characterized by iron accumulation-induced lipid peroxidation. This study aimed to investigate the involvement of ferroptosis in the progression of AMD.

Methods: A total of 41 samples of AMD and 50 normal samples were obtained from the data set GSE29801 for differential gene expression analysis and functional enrichment. Differentially expressed genes (DEGs) were selected and intersected with genes from the ferroptosis database to obtain differentially expressed ferroptosis-associated genes (DEFGs). Machine learning algorithms were employed to screen diagnostic genes. The diagnostic genes were subjected to Gene Set Enrichment Analysis (GSEA). Expression differences of diagnostic genes were validated in in vivo and in vitro models.

Results: We identified 462 DEGs when comparing normal and AMD samples. The GO enrichment analysis indicated significant involvement in key biological processes like collagen-containing extracellular matrix composition, positive cell adhesion regulation, and extracellular matrix organization. Through the intersection with ferroptosis gene sets, we pinpointed 10 DEFGs. Leveraging machine learning algorithms, we pinpointed five ferroptosis feature diagnostic genes: VEGFA, SLC2A1, HAMP, HSPB1, and FADS2. The subsequent experiments validated the increased expression of SLC2A1 and FADS2 in the AMD ferroptosis model.

Conclusion: The occurrence of ferroptosis could potentially contribute to the advancement of AMD. SLC2A1 and FADS2 have demonstrated promise as emerging diagnostic biomarkers and plausible therapeutic targets for AMD.

Keywords: age‐related macular degeneration; ferroptosis; machine learning.

Figure 1

Identification and enrichment analysis of DEGs. (A) The volcano plot illustrating the 462 DEGs. (B) The GO barplot enrichment analysis of DEGs. (C) The GO network enrichment analysis of DEGs.

Figure 2

Identification of DEFGs. (A) The Venn diagram illustrates the genes that are common between DEGs and ferroptosis‐related genes. (B) Box plots provide an overview of the expression levels of DEFGs in AMD patients.

Figure 3

Machine learning in the identification of ferroptosis‐signatures diagnostic genes. (A–C) identify ferroptosis feature genes using LASSO regression, SVM, and RF algorithm. (D) The Venn diagram shows the overlap of candidate genes between the above three algorithms. (E) ROC curve of ferroptosis‐signatures in AMD diagnosis. (F, G) Box plots of RiskScores in the Normal and AMD groups. (H) Clustered heatmap of ferroptosis‐signatures diagnostic genes expression levels.

Figure 4

GSEA and small molecule drug prediction analysis. (A–G) GSEA investigation of FADS2, HAMP, HSPB1, SLC2A1, and VEGFA.

Figure 5

Expression levels of ferroptosis‐signature diagnostic genes in the ARPE‐19 cell line (A) mRNA expression of FADS2, HAMP, HSPB1, SLC2A1, and VEGFA. (B) Protein expression of FADS2 and SLC2A1.ns, no significance; *p < 0.05; **p < 0.01; ***p < 0.001.

Figure 6

Validation of FADS2 and SLC2A1 expression in an AMD animal model. (A) HE staining shows disruption of the outer nuclear layer and loss of the RPE layer. (B) ZO‐1 staining indicates damage to the tight junctions of RPE cells. (C, D) IF staining shows high expression of FADS2 and SLC2A1 in the RPE area (indicated by arrowheads).