Single-cell analysis of cerebrospinal fluid reveals common features of neuroinflammation

Abstract

Neuroinflammation is often characterized by immune cell infiltrates in the cerebrospinal fluid (CSF). Here, we apply single-cell RNA sequencing to explore the functional characteristics of these cells in patients with various inflammatory, infectious, and non-inflammatory neurological disorders. We show that CSF is distinct from the peripheral blood in terms of both cellular composition and gene expression. We report that the cellular and transcriptional landscape of CSF is altered in neuroinflammation but is strikingly similar across different neuroinflammatory disorders. We find clonal expansion of CSF lymphocytes in all disorders but most pronounced in inflammatory diseases, and we functionally characterize the transcriptional features of these cells. Finally, we explore the genetic control of gene expression in CSF lymphocytes. Our results highlight the common features of immune cells in the CSF compartment across diverse neurological diseases and may help to identify new targets for drug development or repurposing in multiple sclerosis (MS).

Keywords: cerebrospinal fluid; clonal expansion; multiple sclerosis; neuroinflammation; oligoclonal bands; omics; single-cell sequencing.

Figure 1

The cellular composition of the CSF is notably different from that seen in PBMC (A) Uniform manifold approximation and projection (UMAP) plot displaying individual cells in the single-cell dataset colored according to cell type. (B) Volcano plot showing differential abundance comparing the cell type proportions in CSF vs. PBMC (pooled across all disease cohorts). The x axis shows the log-fold change (logFC) in cell type proportion, with positive values indicating a higher proportion in CSF compared with PBMC. The y axis shows the −log10 of the p value, with values above the horizontal gray line achieving statistical significance at a Bonferroni-adjusted p value threshold (alpha 5%). (C) Bar plot showing cell type proportions in CSF and PBMC in each cohort separately. Abbreviations: NIND, non-inflammatory neurological diseases; OIND, other inflammatory neurological diseases; ID, infectious neurological diseases; MS, multiple sclerosis; CSF, cerebrospinal fluid; PBMC, peripheral blood mononuclear cell.

Figure 2

Transcriptional profiling of CSF leukocytes reveals tissue-specific gene expression (A) Volcano plot displaying results of differential expression testing comparing gene expression in CSF and PBMC for four selected cell types of interest (pooling data across disease cohorts). Each dot represents a gene tested, the y axis shows the −log10(p value), and the x axis shows the log2-fold change in transcript abundance. Genes colored in red are upregulated in CSF while genes colored in blue are downregulated. Tests with a Bonferroni-corrected p value greater than 0.01 are shown in gray. Note that the y axes are on different scales for clarity. (B) Scatterplot comparing differential expression results in MS and ID CSF. Each dot is a gene. The x axis shows the log-fold change from MS CSF to MS PBMC (i.e., positive values indicate upregulation in MS CSF compared with PBMC), and the y axis depicts the log-fold change in ID CSF versus ID PBMC. Dots are colored according to cell type in which they were tested. Only genes achieving statistical significance in MS are shown. The dotted line represents the null hypothesis that the change in gene expression in CSF is identical between MS and ID. (C) Gene set enrichment analysis (GSEA) results comparing the expression of genes involved in Hallmark canonical pathways in CSF vs. PBMC across multiple cell types. The tiles are colored by the direction of their normalized enrichment score (NES), with red (positive) indicating upregulation in CSF and blue (negative) indicating downregulation. ∗, FDR < 0.05; ∗∗, FDR < 0.005; ∗∗∗, FDR < 0.0005. Again, these results reflect the pooled analysis, i.e., comparing all CSF samples with all PBMC samples. Cohort-specific results are presented in Table S6.

Figure 3

The nature of clonally expanded B cells (A) Barplots showing the proportion of B cells and ASCs in each disease cohort and each of CSF/PBMC which were part of an identified expanded clonal group. The numbers at the top of the plot indicate the p values for the comparison of clonal % between CSF and PBMC in each disease cohort. ∗, p < 0.01; ∗∗∗, p < 0.0001. (B) Barplot showing the overall cellular composition of the expanded vs. non-expanded B cell pool, highlighting the observation that the expanded pool is primarily composed of ASCs. (C) Volcano plot showing differential expression results contrasting clonally expanded vs. non-expanded CSF memory B cells in a pooled analysis of all disease cohorts. (D) As per (B), but showing the isotypes expressed by expanded vs. non-expanded cells, showing the marked shift toward IgG isotypes, particularly IgG1 among expanded cells.

Figure 4

The TCR repertoire in neuroinflammation (A) Boxplots showing the proportion of clonally expanded T cells in the CSF and PBMC of each disease group. Numbers at the top of the plot show the p values for the comparison of CSF vs. PBMC. ∗, p < 0.01; ∗∗, p < 0.001; ∗∗∗, p < 0.0001. (B) Barplots showing the cellular composition of the clonally expanded vs. non-expanded T cell pools in CSF and PBMC. (C) DE volcano plot contrasting gene expression in MS CSF T resident memory T (Trm) cells vs. non-expanded cells, highlighting upregulation of cytotoxicity markers and HLA molecules. (D) Boxplot showing the proportion of T cells within the expanded and non-expanded subsets with a TCR beta chain CDR3 predicted to bind various epitopes in each cohort divided by clonal status (data are shown for CSF TCRs only). Clonally expanded TCRs were more likely to recognize Epstein-Barr virus (EBV) antigens in both MS and controls.

Figure 5

CSF cell eQTL analysis (A) Regional association plot for a previously unknown eQTL on chromosome 11 for ETS1 expression CSF CD4⁺ T cells. Each dot represents one tested single nucleotide polymorphism (SNP), colored by the degree of linkage disequilibrium (LD, r²) to the lead SNP. (B) Correlation of eQTL p values and p values for MS risk (IMSGC 2019 susceptibility GWAS) for the same locus on chromosome 11. (C) Forest plot showing the eQTL effect estimates +95% confidence intervals of the lead SNP rs61909096 on ETS1 expression in different cell types and compartments, suggesting a specific effect for CSF CD4⁺ T cells. (D) Regional association plot for the locus around a known eQTL on chromosome 8 associated with ZC2HC1A expression in CSF B cells that colocalizes with an MS risk signal. (E) Correlation of eQTL p values and p values for MS risk (IMSGC 2019 susceptibility GWAS) for the same locus on chromosome 8. (F) Forest plot showing the eQTL effect estimates +95% confidence intervals of rs1466526 on ZC2HC1A expression in 8 cell types.