Endothelial TDP-43 depletion disrupts core blood-brain barrier pathways in neurodegeneration

Abstract

Endothelial cells (ECs) help maintain the blood-brain barrier but deteriorate in many neurodegenerative disorders. Here we show, using a specialized method to isolate EC and microglial nuclei from postmortem human cortex (92 donors, 50 male and 42 female, aged 20-98 years), that intranuclear cellular indexing of transcriptomes and epitopes enables simultaneous profiling of nuclear proteins and RNA transcripts at a single-nucleus resolution. We identify a disease-associated subset of capillary ECs in Alzheimer's disease, amyotrophic lateral sclerosis and frontotemporal degeneration. These capillaries exhibit reduced nuclear β-catenin and β-catenin-downstream genes, along with elevated TNF/NF-κB markers. Notably, these transcriptional changes correlate with the loss of nuclear TDP-43, an RNA-binding protein also depleted in neuronal nuclei. TDP-43 disruption in human and mouse ECs replicates these alterations, suggesting that TDP-43 deficiency in ECs is an important factor contributing to blood-brain barrier breakdown in neurodegenerative diseases.

Fig. 1. Single-nuclei analysis of human tissue samples with endothelial and microglial enrichment.

a, Schematic representation of the methodology employed for endothelial enrichment, leveraging ERG and intranuclear CITE-seq antibodies. FACS, fluorescence-activated cell sorting. b, UMAP visualizing 132,859 nuclei from 92 human frontal cortex samples, color-coded by cell type. OPC, oligodendrocyte precursor cell; PVMP, perivascular macrophage. c, A dual representation featuring a dot plot of cell type-specific gene markers and a bar chart illustrating the count of captured nuclei for each cell type.

Fig. 2. Principal-component analysis reveals shared transcriptional processes across neurodegeneration.

a, Correlation matrix showing Pearson correlations of PCA-adjusted capillary EC profiles from 88 donors, having at least ten nuclei processed via Harmony for integration based on batches. The PCA, based on 50 components, is followed by hierarchical clustering and dendrogram visualization, categorizing data by sex and disease state. b, Similar correlation matrix for microglial population derived from 82 donors. Unaff., unaffected. c, Heatmaps showing top pathways enriched among the genes differentially expressed (pseudobulk) between each of the indicated donor populations compared to unaffected aged donors. Dots indicate differences in pathways and specific cell types where Padjusted < 0.05.

Fig. 3. Microglial p65/NF-κB associates with disease-associated microglia.

a, UMAP clustering representation of cortical microglial cells, derived from 22,588 cells from 90 donors (of which 9,479 cells from 45 donors belonging to inCITE dataset), with disease state regressed out. b, In silico isolation of microglia nuclei and UMAP clustering based on batch-corrected gene expression patterns, without regression of disease state. c, Heatmap illustrating differential gene expression patterns across the identified microglial subclusters. d, Gene counts for DAM, homeostatic and dystrophic microglial markers across cells. e, Comprehensive analysis of inferred pathways prevalent in the microglial clusters by PROGENY and mapping of inferred pathway scores onto cells in UMAP; red signal indicates higher expression of pathway genes. f, GSEA of the TNF signaling pathway within DAM and homeostatic clusters in contrast to other clusters from MSigDB. FDR, false discovery rate; NES, normalized enrichment score. g, Scatter-plot showing correlations between donor age (x axis), disease status (color) and their respective contributions to individual microglial clusters (y axis). h, Scatter-plot showing the correlation between NF-κB levels in the DAM cluster and pTau levels determined by western blot in the same brain tissue, with 95% confidence intervals (P = 0.014; R = 0.71). Two-sided Spearman correlation was used to assess the relationships. i, Violin plot showing the mean level of p65/NF-κB protein (normalized to levels of histone H3 protein) in cells of microglial DAM and homeostatic clusters and informatic isolation of nuclei with highest (top 10th percentile), high (top 10–25th percentile) and low (bottom 25th percentile) levels of NF-κB:H3 shown as a density plot. Areas of the UMAP enriched for cells falling into this range of NF-κB:H3 are red; lower levels are blue. j, Scatter-plot showing gene weight (relative association in the NF-κB pathway as predicted by the Progeny database, x axis) and log fold change between DAM cluster and other clusters (y axis). In red are transcripts increased or decreased in a manner consistent with NF-κB activation, and in blue are transcripts following a pattern of reduced NF-κB activity. k, Similar plot for homeostatic cluster. Stat, statistics.

Fig. 4. Distinct brain capillary endothelial states associate with healthy aging versus neurodegenerative diseases.

a, UMAP clustering representation of brain EC subtypes, encompassing capillaries, veins and arteries, derived from 70,006 EC nuclei from 92 donors, 38,218 of these contain inCITE-seq antibodies, with disease state regressed out. b, Heatmap showing identification markers for capillary EC subtypes. Green, artery; orange, vein; blue, capillary; red, capillary.2. c, In silico isolation of capillary nuclei with inCITE antibody labeling and UMAP clustering based on batch-corrected gene expression patterns and without regression of disease state. Data represent 23,152 nuclei from 47 donors. d, UMAP feature plots showing markers of the REV1 population. e,f, Scatter-plot showing correlation (Spearman, two sided) between donor age and contribution to each capillary cluster (e) and box plot showing the proportions of capillary ECs for each donor falling into each cluster (f). g, Plot showing pathway enrichment from MSigDB in disease-associated REV1 cluster versus HC cluster, by GSEA. Color bar is normalized enrichment score. Analysis of variance (ANOVA) with a post hoc Tukey test was used to compare the proportion of donor’s cell per cluster (f). P values represent comparison to healthy aged donors. REV1 (AD P = 0.011; ALS P = 0.022; FTD P = 0.012). REV2 (NS). CapA (AD P = 0.002; ALS P = 0.008; FTD NS). HC (AD P = 0.0001; ALS P = 0.0001; FTD 0.0002). CapV (NS). ***P < 0.001; **P < 0.01; *P < 0.05; NS, not significant.

Fig. 5. inCITE analysis reveals a loss of Wnt/β-catenin signaling in disease-associated nuclei.

a, Dot plots showing relative levels of protein across all cell clusters and within the endothelial and microglial clusters. b, Violin plot showing histone-normalized protein levels for β-catenin, TDP-43 and p65/NF-κB in REV1 cluster compared to HC. REV1 cluster represents 6,297 nuclei and HC cluster represents 4,507. The distribution of those protein levels displayed in box plot shows interquartile range (IQR) (25th–75th percentiles) with the median (50th percentile) as the center line, whiskers extending to the minimum and maximum values within 1.5 × IQR, and individual points representing outliers beyond these bounds. Unpaired t-test used to evaluate significance. β-catenin (t = 12.91, P = 7.36 × 10⁻³⁸, Cohen’s d of 0.252); TDP-43 (t = 31.96, P < 2.71 × 10⁻²¹⁴, Cohen’s d of 0.624); p65/NF-κB (t = 4.43, P < 9.10 × 10⁻⁰⁶, Cohen’s d of 0.087). c, Density plots illustrating the top 10th percentile of protein levels for β-catenin (relative to H3), and predicted positive regulation of Wnt signaling pathway by gene expression. d, GSEA plot of Wnt/β-catenin genes enriched in disease-associated REV1 cluster (versus all other clusters). Gene sets are either genes positively associated with Wnt activation in human ECs or negatively associated with Wnt activation. e, Feature plots showing example genes in the β-catenin pathway, projected onto the capillary UMAP. f, Violin plots with inner box-and-whisker plot with default parameters showing Wnt target gene expression in REV1 and HC clusters. REV1 cluster represents 6,297 nuclei and HC cluster represents 4,507. The rank_genes_groups function was used to derive statistics between cluster REV1 and HC using a t-test. APCDD1 (logFC = 1.96, Padj = 4.14 × 10⁻¹³⁹), ABCG2 (logFC = 1.85, Padj = 0.00), LEF1 (logFC = 0.94, Padj = 1.58 × 10⁻¹³⁵). FC, fold change.

Fig. 6. inCITE-seq analysis reveals a specific loss of nuclear TDP-43 and increased NF-κB transcriptional targets in disease-associated nuclei.

a, Density plots illustrating the top 10th percentiles of protein levels for p65/NF-κB and TDP-43. b, Transcripts positively associated with nuclear p65/NF-κB in each cluster. P values reported on the y axis are derived from a t-test on differential gene expression between the two groups. c, Pathway enrichment, by Enrichr, shows MSigDB pathways enriched in nuclei in top 10th percentile of nuclear p65/NF-κB compared to the bottom in REV1 and HC clusters. Enrichment of the pathways shown for each cluster is also shown in gray for the other cluster. P values (FDR q-values) are indicated as color scales, and derived from hypergeometric analysis using Enrichr. d, A smoothed line plot comparing NF-κB protein levels (x axis) to TDP-43 (y axis). e, Violin plot showing histone-normalized protein levels for p65/NF-κB and TDP-43 in the bottom tenth percentile (2,208 nuclei) and top 25th percentile (5,780 nuclei) of p65/NF-κB levels within the capillary endothelial cell cluster. Box plots display the IQR (25th–75th percentiles) with the median (50th percentile) as the center line and whiskers extending to the minimum and maximum values within 1.5 × IQR. ANOVA with Tukey’s HSD post hoc tests was used to assess significance. Significance of TDP-43 protein levels in the bottom tenth percentile (bottom left) and top 25th percentile (bottom right) of NF-κB/p65 nuclear protein in capillaries relative to healthy old donors is shown. TDP-43 levels showed a significant difference between aged and young donors (P < 0.001) (bottom left). TDP-43 levels were significantly different in AD (P < 0.001), ALS (P < 0.001) and FTD (P < 0.001), but not in young controls (P > 0.05) (bottom right).

Fig. 7. Correlation of transcriptional effects of TDP-43 loss with dementia capillary signature.

a, Correlation between the effect of TDP-43 depletion on mRNA transcript level in human brain ECs (y axis) and the same mRNA affected in the comparison of REV1 and HC endothelial clusters (x axis). Human brain ECs were treated with no siRNA, nontargeting siRNA and two different TDP-43 siRNA across four biological perturbations (static culture, TNF stimulation and laminar or disturbed flow; n = 4 per treatment, n = 8 without TDP-43 knockdown and n = 8 with knockdown). b,c, Enriched Hallmark (dark green and dark blue) and Gene Ontology (GO) terms (light green and light blue) for transcripts increased in REV cluster and by siTDP-43 (b) or decreased in REV cluster and by siTDP-43 (c). Enrichr in GSEApy was used to assess values from hypergeometric test of overlap between regulated transcripts and the indicated msigDB databases. d, Heatmap showing the results of GSEA of Hallmark and Kegg pathways, and a set of custom EC response profiles (for example, disturbed flow and β-catenin activation) from the literature. NES derived from ranked and weighted (=1) GSEA analysis of all expressed transcripts (DESeq2 base mean 200). DESeq2 ranked lists and GSEA sets (KEGG, HALLMARK and custom) are reported in the supplementary tables. Dots indicate FDR q-value significance from GSEA. e–h, Focused plots showing the correlated response in specific pathways, between disturbed flow ECs with and without siTDP-43. Pathway genes shown are Hallmank TNF via NF-κB (e), Hallmark G2-M transition (f), top transcripts suppressed by activation of Wnt/β-catenin in PMID34755601 (g) and top transcripts induced by exposure of carotid endothelium to low and disturbed flow in PMID29293084 (h). Scatter-plots show data points and trend line with 95% confidence intervals (a,e–h). Histograms (e–h) show the fraction of points within the indicated interval on each axis. Two-sided Pearson correlations and P values are shown.