We downloaded the GSE5281 dataset from the Gene Expression Omnibus (GEO) as the analysis object. This dataset, based on the GPL570 platform, includes gene expression profiles of brain tissue samples from 87 AD patients and 73 healthy controls, with sample regions covering AD-related brain areas such as the hippocampus and temporal cortex. The ferroptosis-related gene list was downloaded from the FerrDb database (http://www.zhounan.org/ferrdb/), and the immune-related gene list was downloaded from the ImmPort database (https://www.immport.org/).

The relevant gene sets were integrated from the following sources: 1) The FerrDb database (version V2.0, downloaded on April 20, 2026), yielding a total of 621 protein-coding ferroptosis-related genes (including 264 drivers, 238 suppressors, 9 markers, and 110 unclassified genes); 2) Ferroptosis-related non-coding RNAs were integrated from miRBase (v22.1), CircBank (v3.0), and LncBase (v2.0); 3) Supplementary literature-reported ferroptosis-associated genes (e.g., RPL8, MIOX) were added. A final ferroptosis-related gene set comprising protein-coding genes, miRNAs, circRNAs, and lncRNAs was compiled for subsequent intersection analysis. The immune-related gene list was downloaded from the ImmPort database (Release 61, downloaded on April 20, 2026), yielding a total of 3,118 immune-related genes. This gene set was obtained by integrating core genes from 17 immune-related pathways in ImmPort and supplementing with literature-reported immune regulatory genes to ensure coverage of key nodes in AD-related immune-inflammatory regulation.

Preprocessing of the GSE5281 dataset was performed using R software (version 4.3.1), with the following specific steps:

Data Reading: The raw expression matrix of the dataset was read using the affy and GEOquery packages in R, extracting sample information and gene expression data.

Background Correction: The robust multi-array average (RMA) algorithm was used for background correction of raw data to eliminate background noise interference during chip hybridization.

Quantile Normalization: The corrected expression matrix was quantile-normalized using the limma package to ensure consistent gene expression distribution across all samples in both datasets, eliminating systematic errors between samples and ensuring comparability.

Probe Annotation: Chip probe IDs were converted to corresponding gene symbols, and probes that could not be matched to gene names were removed. Additionally, low-expression genes in the lowest 25% across all samples were removed to avoid interference in subsequent analyses.

Sample Cleaning: Principal component analysis (PCA) was performed to identify outliers. After PCA analysis, four outlier samples were identified and removed (including sample GSM119676), and a final set of 83 AD samples and 69 control samples (totaling 152 samples) was retained for subsequent analysis.

We used the “limma” package to screen for differentially expressed genes (DEGs) in the dataset, with thresholds set at adjusted p < 0.05 and |log₂ fold change (FC)| ≥ 1. The analytical model included disease status (AD vs. control) as the main effect, without incorporating brain region as a covariate. After obtaining significantly differentially expressed genes, we used the “pheatmap” and “ggplot2” R packages to create volcano plots to visualize the distribution of DEGs.

The ferroptosis-related gene list from FerrDb and the immunity-related gene list from ImmPort were intersected with the DEGs to obtain ferroptosis- and immunity-related intersection DEGs, designated as Immunity- and Ferroptosis-Related Differentially Expressed Genes (IF-DEGs), which served as the candidate gene set for subsequent analyses.

We used the “clusterProfiler” R package for Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis of the candidate genes. GO analysis covered three categories: biological process (BP), cellular component (CC), and molecular function (MF). Adjusted P < 0.05 was used as the significance threshold to filter statistically significant enrichment terms and signaling pathways.

We used the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database to construct a PPI network of IM-DEGs, with a confidence threshold set at ≥0.700 (high confidence), and unconnected isolated nodes were hidden. Subsequently, we used Cytoscape 3.10.1 software for network visualization. Furthermore, using the CytoHubba plugin in Cytoscape, we employed the Maximal Clique Centrality (MCC) algorithm to identify the top 5 core hub genes.

We performed Gene Set Enrichment Analysis (GSEA) based on the whole-genome expression profile, using KEGG and Reactome gene sets as references. Based on the normalized enrichment score (NES) and adjusted P value, we identified signaling pathways significantly activated or suppressed in the AD group, verifying the regulatory characteristics of immune-related pathways in AD at a global level.

We used the CIBERSORT algorithm to estimate the relative infiltration abundance of 22 immune cell types in each sample, and drew stacked bar plots of immune cells to visually display differences in immune cell composition between samples. Intergroup comparisons of immune cell infiltration levels were conducted using the Wilcoxon rank-sum test, followed by false discovery rate (FDR) multiple-testing correction, with differences displayed via boxplots. An FDR-adjusted P < 0.05 was considered statistically significant.

All data analyses were performed in the R software environment (version 4.3.1). Measurement data were expressed as mean ± standard deviation. Comparisons between groups were performed using t-tests or Wilcoxon rank-sum tests, depending on data distribution. The significance criteria for differential gene screening were |log₂FC| ≥ 1 and adjusted P < 0.05; the significance criterion for enrichment analysis was adjusted P < 0.05. All tests were two-sided, and P < 0.05 was considered statistically significant.

After RMA background correction, quantile normalization, and PCA-based removal of four outlier samples, 83 AD samples and 69 control samples were ultimately included, and the quality of the analyzed samples was good. Principal component analysis results showed that the AD group samples were concentrated with high intra-group homogeneity; outlier samples in the control group had been removed, and there was a clear transcriptomic separation trend between the two groups, suggesting systematic differences in gene expression profiles of brain tissue between AD and healthy controls (Figure 1(a)).

Using adjusted P < 0.05 and |log₂FC| ≥ 1 as screening thresholds, a total of 1191 DEGs were identified between the AD group and the control group. The number of differentially expressed genes under this threshold ensured both the biological significance of expression changes and avoided the loss of valid genes due to overly strict thresholds. The volcano plot clearly displayed the distribution of expression fold changes and statistical significance for all genes: red dots represent significantly upregulated genes, blue dots represent significantly downregulated genes, and gray dots represent genes without statistical differences, with slightly more upregulated genes than downregulated ones, indicating widespread disease-specific transcriptomic disruption in AD brain tissue (Figure 1(b)).

To focus on key genes intersecting ferroptosis and immune regulation in AD, the above DEGs were intersected with ferroptosis-related genes from the FerrDb database and immune-related genes from the ImmPort database. Venn diagram results showed that a total of 54 co-associated differentially expressed genes belonging to all three gene sets were obtained (Immunity- and Ferroptosis-Related Differentially Expressed Genes, IF-DEGs) (Figure 2). These genes were defined as candidate gene sets for subsequent analysis to explore the cross-mechanism of ferroptosis and immune regulation in AD.

(a) (b)

Figure 1. Identification of differentially expressed genes. (a) Principal component analysis of the expression microarray dataset from AD patients; (b) Volcano plot of differentially expressed genes.

Figure 2. Identification of intersection genes between ferroptosis-related and immunity-related DEGs. Venn diagram of ferroptosis-immunity co-associated differentially expressed genes.

Bidirectional clustering was performed on the 54 IF-DEGs, and an expression heatmap was generated (Figure 3). The results showed that the expression patterns of these genes could clearly distinguish the AD group from the control group, with convergent expression within groups and significant differences between groups. Some genes were upregulated in the AD group, while others were downregulated, consistent with the volcano plot results, indicating that this gene set has good discriminative ability between groups. The clustering heatmap did not detect obvious batch effects, suggesting that the preprocessing steps effectively removed systematic errors.

Figure 3. Validation of expression patterns of candidate genes. Expression pattern heatmap of candidate genes.

To elucidate the biological functions of IF-DEGs, GO and KEGG enrichment analyses were conducted using clusterProfiler (adjusted P < 0.05).

The results for the three GO subcategories are as follows: In the biological process (BP) category, genes were enriched in terms such as negative regulation of immune system process, monocyte differentiation, and lymphocyte proliferation, suggesting that candidate genes are involved in immune-inflammatory regulation in AD (Figure 4(a)); In the cellular component (CC) category, enrichment was found in vesicle lumen and secretory granule lumen, indicating their functions are related to vesicle transport and secretion processes (Figure 4(b)); In the molecular function (MF) category, genes were concentrated in DNA-binding transcription factor binding, RNA polymerase II-specific transcription factor binding, and transcription co-regulator binding, suggesting that candidate genes primarily play transcriptional regulatory roles (Figure 4(c)).

KEGG pathway enrichment showed: the outer ring represents significantly enriched KEGG pathway IDs, the inner ring color indicates the overall regulatory direction of the pathway (Z-score, red for upregulation, purple for downregulation); outer ring scatter points represent enriched genes within the pathway, with purple dots for upregulated genes and red dots for downregulated genes. Results indicated that IF-DEGs were mainly enriched in the TNF signaling pathway, lipid metabolism-related pathways, and immune-inflammatory pathways, suggesting that candidate genes participate in the pathological process of AD by regulating these pathways (Figure 4(d)).

(a)

(b)

(c)

(d)

Figure 4. Functional enrichment analysis of candidate genes. (a) GO enrichment biological process diagram of candidate genes; (b) GO enrichment cellular component diagram of candidate genes; (c) GO enrichment molecular function diagram of candidate genes; (d) KEGG enrichment circle diagram of candidate genes.

The 54 IF-DEGs were submitted to the STRING database to construct a PPI network with a confidence score ≥ 0.7, containing 54 nodes and 217 interaction edges, with no isolated nodes (Figure 5). Using Cytoscape and the CytoHubba plugin, combined with the MCC algorithm and node connectivity, the top 5 core Hub genes were screened: BCL2, CD44, FOXO1, PPARG, and SOX2. These genes had the highest connectivity in the network and extensive interactions with other candidate genes, suggesting their pivotal role in the ferroptosis-immunity cross-regulation in AD.

Figure 5. PPI network construction and core gene identification. PPI network and core genes of candidate genes.

To compensate for the limitation of traditional differential gene enrichment analysis, which only focuses on screened genes, GSEA was performed based on the whole gene expression profile using KEGG and Reactome gene sets, with adjusted P < 0.05 as the threshold to identify significantly activated or inhibited pathways in the AD group.

Among significantly inhibited pathways, DNA replication, DNA synthesis, G2/M cell cycle checkpoint, removal of ORC1 from chromatin, SCF-SKP2-mediated degradation of P27/P21, Dectin-1-mediated non-canonical NF-κB signaling, and host interaction of HIV factors from the Reactome database were all negatively enriched. Pathway interaction networks showed close functional associations among these pathways, collectively involved in cell cycle regulation, DNA replication, and innate immune signaling (Figure 6(a)). The NES bar chart further quantified the inhibition degree of each pathway (NES range −2.431 to −2.39) (Figure 6(d)). Among them, the inhibition of neurotransmitter receptors and postsynaptic signal transduction pathways was the most significant (NES = −2.523), with a continuously declining enrichment curve, consistent with the pathological features of impaired synaptic function and neurotransmitter transmission disorders in AD (Figure 6(b), Figure 6(c)).

Among significantly activated pathways, the cytokine-cytokine receptor interaction pathway (NES = 2.484) and the SUMOylation pathway (NES = 2.404) had the highest enrichment. The enrichment curve of the former continuously rose, indicating overall activation of this pathway in the AD group, corroborating the increased proportion of pro-inflammatory immune cells in immune infiltration analysis; the latter is involved in protein post-translational modification and homeostasis regulation, and its abnormal activation may be related to the accumulation of misfolded proteins in AD (Figure 6(b), Figure 6(c)).

The ridge plot and bar chart, respectively, displayed the distribution characteristics of NES values and enrichment degrees of each pathway, with consistent results, collectively revealing the overall transcriptomic features of AD, including activation of immune-inflammatory pathways and inhibition of synaptic and cell cycle-related pathways (Figure 6(c), Figure 6(d)).

(a)

(b)

(c)

(d)

Figure 6. Genome-wide pathway enrichment analysis. (a) Interaction network diagram of significantly inhibited pathways; (b) GSEA enrichment curves of key pathways; (c) Ridge plot of NES distribution for significant pathways; (d) Bar chart of NES for significant pathways.

The CIBERSORT algorithm was used to estimate the relative infiltration abundance of 22 immune cell types in the AD group and the control group. Stacked bar charts showed significant differences in immune cell composition between the two groups, with an overall increase in the proportion of pro-inflammatory activated immune cells in the AD group (Figure 7(a)). Box plots further indicated that compared to the control group, the proportions of memory B cells, CD8⁺ cytotoxic T cells, eosinophils, and neutrophils were significantly increased in the AD group (FDR-adjusted P < 0.05, P < 0.001); while the proportions of follicular helper T cells, resting NK cells, and M0/M1/M2 macrophages were significantly decreased (FDR-adjusted P < 0.05, *P < 0.01, **P < 0.001) (Figure 7(b)). These results suggest remodeling of the immune microenvironment in AD brain tissue, with increased infiltration of pro-inflammatory immune cells coexisting with a decrease in homeostatic immune cells, constituting the cellular basis of central immune imbalance.

(a)

(b)

Figure 7. Analysis of immune cell infiltration characteristics. (a) Stacked bar chart of immune cell infiltration composition in AD and control group samples; (b) Box plot of inter-group differences in immune cell infiltration levels between AD and control groups.