Role of ferroptosis in neuroimmunity and neurodegeneration in multiple sclerosis revealed by multi-omics data

Abstract

Previous studies have found that ferroptosis plays an important role in a variety of neurological diseases. However, the precise role of ferroptosis in the multiple sclerosis patients remains uncertain. We defined and validated a computational metric of ferroptosis levels. The ferroptosis scores were computed using the AUCell method, which reflects the enrichment scores of ferroptosis-related genes through gene ranking. The reliability of the ferroptosis score was assessed using various methods, involving cells induced to undergo ferroptosis by six different ferroptosis inducers. Through a comprehensive approach integrating snRNA-seq, spatial transcriptomics, and spatial proteomics data, we explored the role of ferroptosis in multiple sclerosis. Our findings revealed that among seven sampling regions of different white matter lesions, the edges of active lesions exhibited the highest ferroptosis score, which was associated with activation of the phagocyte system. Remyelination lesions exhibit the lowest ferroptosis score. In the cortex, ferroptosis score were elevated in neurons, relevant to a variety of neurodegenerative disease-related pathways. Spatial transcriptomics demonstrated a significant co-localization among ferroptosis score, neurodegeneration and microglia, which was verified by spatial proteomics. Furthermore, we established a diagnostic model of multiple sclerosis based on 24 ferroptosis-related genes in the peripheral blood. Ferroptosis might exhibits a dual role in the context of multiple sclerosis, relevant to both neuroimmunity and neurodegeneration, thereby presenting a promising and novel therapeutic target. Ferroptosis-related genes in the blood that could potentially serve as diagnostic and prognostic markers for multiple sclerosis.

Keywords: ferroptosis; multiple sclerosis; multi‐omics; neurodegeneration; neuroimmunity.

FIGURE 1

SnRNA‐seq data reveals ferroptosis scores in white matter lesions of MS. (A) Schematic diagram of the seven sampling regions in MS brain tissues. (B) The tSNE projections of cell clusters (n = 84,222 nuclei from eight controls and nine MS patients). (C) The tSNE projection of ferroptosis score on control (middle), MS lesions (upper left, including acute lesion, CI and CI), MS lesion edges (upper right, including CIE and CAE), CIE (lower left) and CAE (lower right). Brightness represents ferroptosis score. Immune cells are circled. (D) Density plot of ferroptosis score distribution in different groups: control versus MS (left), control, MS lesions, and MS lesion edges (middle), control, CIE, and CAE (right). Dashed lines represent means, the t‐test was employed. (E) Comparison of ferroptosis score between seven sampling regions and the control (the dashed line represents the median ferroptosis score of the control, the Wilcoxon rank‐sum test was employed). (F) Line graph of the ferroptosis score (green), and enrichment scores of ferroptosis driver (lavender), and ferroptosis suppressor (red) at different sampling regions. Points represents the means, and the width of the ribbon represents the 95% confidence interval (CI). (G) Faceted violin plot comparing ferroptosis scores in control (green) and MS (brown) in different cell types. Compared with the control, the t‐test was employed. Boxplot centred on median, bounds defined between the 25th and 75th percentile. *p < 0.05, **p < 0.01, ***p < 0.001. A, acute lesions; CA, chronic active lesion; CAE, chronic active lesion edge; CI, chronic inactive lesion; CIE, chronic inactive lesion edge; MS, multiple sclerosis; NAWM, normal appearing white matter; OPCs, oligodendrocyte progenitor cells; RM, remyelinated lesions; tSNE, t‐distributed Stochastic Neighbour Embedding.

FIGURE 2

Higher ferroptosis scores in MS white matter lesions are associated with phagocyte activation. (A) tSNE projections of immune cell clusters (n = 7048 nuclei). (B) Immune cell distribution between controls and MS patients, and the projection plot of ferroptosis score on tSNE. (C) Stacked column chart of the distribution of immune cell clusters in control and MS. (D) Comparison of the ferroptosis scores of immune cells at different sampling regions (the dashed line represents the median ferroptosis score of the control, the Wilcoxon rank‐sum test was employed). (E) The comparison of ferroptosis scores among different immune cell clusters (dashed line represents median ferroptosis score of homeostatic microglia). (F) Pseudotime trajectory of phagocytes (left). The locations of the three pseudotime branches on the trajectory (right). Arrows represent pseudotime directions. (G) Changes of phagocytic homeostatic‐state genes (P2RY12) and active‐state genes (TREM2) with pseudotime. (H) Comparison of ferroptosis scores in three pseudotime branches (compared with branch 1, the Wilcoxon rank‐sum test was employed). *p < 0.05, **p < 0.01, ***p < 0.001. A, acute lesions; CA, chronic active lesion; CAE, chronic active lesion edge; CI, chronic inactive lesion; CIE, chronic inactive lesion edge; MIMS, microglia inflamed in MS; mono/moDC, monocytes/dendritic cells; NAWM, normal appearing white matter; RM, remyelinated lesions; tSNE, t‐distributed Stochastic Neighbour Embedding.

FIGURE 3

Phagocytes with higher ferroptosis scores may contribute to neuroimmune inflammation. (A) Density plot of the ferroptosis scores of immune cells in controls and MS patients. (B) The tSNE projection of immune cells in high‐ferroptosis score group and low‐ferroptosis score group. (C) Pie chart of the distribution of immune cells in the high‐ferroptosis score group and the low‐ferroptosis score group. (D) Volcano plot of the DEGs of phagocytes in the high‐ferroptosis score group compared to the low‐ferroptosis score group. (E) GO enrichment analysis of DEGs in phagocytes of high‐ferroptosis score group and low‐ferroptosis score group. (F) KEGG functional enrichment analysis of DEGs in phagocytes of high‐ferroptosis score group and low‐ferroptosis score group. (G) Diagram of the cell communication network between immune cell clusters. Circle sizes are proportional to the number of cells in each cell cluster. Edge colours represent cell clusters that signalled and edge width represents the communication strength. (H) Cell communication map of T cells as signal recipients. (I) Cell communication map of T cells as signal senders. (J) Dot plot showing the receptor‐ligand pairs that immune cell clusters communicate with T cells. Dot colour reflects communication probabilities and dot size represents computed p‐values. Empty space means the communication probability is zero. *p < 0.05, **p < 0.01, ***p < 0.001. DEGs, differentially expressed genes; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; MIMS, microglia inflamed in MS; mono/moDC, monocytes/dendritic cells; tSNE, t‐distributed Stochastic Neighbour Embedding.

FIGURE 4

Cortical snRNA‐seq data revealed the relationship between ferroptosis score and neurodegeneration. (A) tSNE projections of cortical cell clusters (n = 48,919 nuclei from 16 controls and 19 MS patients). (B) Violin plot of marker genes expression in different cell clusters. (C) Projection of ferroptosis scores on tSNE. (D) Comparison of the control (left) and MS (right) ferroptosis scores on tSNE projection maps. Inside the red circle are neurons. (E) Comparison of ferroptosis scores in control cortex and MS cortex, the t‐test was employed. (F) Faceted violin plot comparing ferroptosis scores between control (blue) and MS (red) in different cell clusters, the t‐test was employed. (G) tSNE projection map showing the distribution of the high‐ferroptosis score group and the low‐ferroptosis score group in cortical cell clusters. (H) Stacked column chart of the distribution of cortical cell clusters in the high‐ferroptosis score group and the low‐ferroptosis score group. (I) KEGG functional enrichment analysis of DEGs between high‐ferroptosis score group neurons and low‐ferroptosis score group neurons. (J) GO functional enrichment analysis of DEGs between high‐ferroptosis score group neurons and low‐ferroptosis score group neurons. *p < 0.05, **p < 0.01, ***p < 0.001. tSNE, t‐distributed Stochastic Neighbour Embedding.

FIGURE 5

The spatial colocalization of ferroptosis score and neurodegeneration in the spatial transcriptomics. (A) GSEA of ferroptosis pathway in MS patient compared to control samples. (B) The fitting curves of ferroptosis markers and ferroptosis scores with intact neuronal signature enrichment (each point represents a sampling spot, a total of 83,256 spots). Note that the x‐axis is reversed. The lower the intact neuronal signature enrichment score, the more severe the neurodegeneration. The left shows the overall analysis, and the right panel shows the control and MS patients, respectively. (C) Mapping of ferroptosis markers onto grey matter slides (up). Mapping of levels of neurodegeneration (intact neuronal signature enrichment) onto grey matter slides (down). Representative regions are circled by red circles. (D) Changes of the components of cell types with the ferroptosis score, displayed separately for controls and MS patients (each point represents a sampling spot). (E) Projection of immune cells, ferroptosis scores, and intact neuronal signature enrichment onto grey matter slides. (F) GO enrichment modules of the DEGs between high‐ferroptosis score tissue blocks and low‐ferroptosis score tissue blocks (Visualized the top 100 items). DEGs, differentially expressed genes; GM, grey matter; GO, Gene Ontology; GSEA, gene set enrichment analysis.

FIGURE 6

Spatial proteomics verification of the association between ferroptosis score and neurodegeneration. (A) Correlation of log2FC of ferroptosis genes between control and MS in the spatial proteome and the spatial transcriptome. (B) Correlation between ferroptosis scores and intact neuronal signature enrichment in the spatial proteome and the spatial transcriptome. (C–E) Correlation between microglia, oligodendrocytes, and astrocytes with ferroptosis scores in the spatial proteome and the spatial transcriptome. Green dots/line represent the transcriptome; red dots/line represent the proteome. n = 4343 jointly detected genes/proteins are plotted. Correlation coefficients were based on Pearson correlations. log2FC, log2FoldChange; SP, spatial proteome; ST, spatial transcriptome.

FIGURE 7

Ferroptosis gene serve as a marker for the diagnosis and prognosis of MS. (A) Schematic diagram of CSF and PB samples (n = 130,015 cells from 12 RRMS and 3 Controls). (B) UMAP of cell clusters in control and RRMS. (C) Faceted violin plot comparing ferroptosis scores between RRMS and control in different cell clusters. (D) Correlation of ferroptosis scores and ferroptosis‐related scores in PB with those in CSF of the same individual (n = 15). (E) DEGs in 9 bulk RNA datasets of PB. Black dots: all DEGs, red dots: ferroptosis‐related genes. (F) UpSet plot visualizing the properties of intersecting and unique sets of upregulated ferroptosis‐related genes between MS and control, ordered from left to right by amount of evidence. The top 10 upregulated ferroptosis‐related genes with the most evidence are marked in red. The vertical bar chart in the UpSet plot shows the number of ferroptosis‐related genes containing single dataset and co‐occurring datasets, which are indicated by single dots and connected dots in the matrix, respectively. The horizontal bar chart in the UpSet plot shows the total number of ferroptosis‐related genes containing each dataset. (G) UpSet plot visualizing the properties of intersecting and unique sets of downregulated ferroptosis‐related genes between MS and control, ordered from left to right by amount of evidence. The top 10 downregulated ferroptosis‐related genes with the most evidence are marked in red. (H) Machine learning flowchart. (I) Comparison of decision values in training set (n = 1017) and validation set (n = 436). Control: blue, MS: red. (J) AUC curves of training set and validation set. (K) Comparison of decision values in the test set (n = 70). Control: blue, MS: red. (L) The AUC curve of test set. (M) Follow‐up changes of decision values in controls (n = 125) and MS paitents (n = 690), GSE41850. Compared with the Baseline, the Wilcoxon rank‐sum test was employed. (N) Survival analysis of high‐score group and low‐score group (GSE15245, n = 65). *p < 0.05, **p < 0.01, ***p < 0.001. Correlation coefficients were based on Pearson correlations. CSF, cerebrospinal fluid; PB, peripheral blood; RRMS, relapsing–remitting multiple sclerosis; UMAP, Uniform Manifold Approximation and Projection.