Repeated head trauma causes neuron loss and inflammation in young athletes

Abstract

Repetitive head impacts (RHIs) sustained from contact sports are the largest risk factor for chronic traumatic encephalopathy (CTE)^1-4. Currently, CTE can only be diagnosed after death and the events that trigger initial hyperphosphorylated tau (p-tau) deposition remain unclear². Furthermore, the symptoms endorsed by young individuals are not fully explained by the extent of p-tau deposition², severely hampering therapeutic interventions. Here we observed a multicellular response prior to the onset of CTE p-tau pathology that correlates with number of years of RHI exposure in young people (less than 51 years of age) with RHI exposure, the majority of whom played American football. Leveraging single-nucleus RNA sequencing of tissue from 8 control individuals, 9 RHI-exposed individuals and 11 individuals with low-stage CTE, we identify SPP1-expressing inflammatory microglia, angiogenic and inflamed endothelial cells, astrocytosis and altered synaptic gene expression in those exposed to RHI. We also observe a significant loss of cortical sulcus layer 2/3 neurons independent of p-tau pathology. Finally, we identify TGFβ1 as a potential signal that mediates microglia-endothelial cell cross talk. These results provide robust evidence that multiple years of RHI is sufficient to induce lasting cellular alterations that may underlie p-tau deposition and help explain the early pathogenesis in young former contact sport athletes. Furthermore, these data identify specific cellular responses to RHI that may direct future identification of diagnostic and therapeutic strategies for CTE.

Fig. 1. Cell-type identification and cell proportion analysis across pathological groups.

a, Experimental workflow. Images created in BioRender. Cherry, J. (2025) https://BioRender.com/5kj3gsd. The workflow was run once for each sample. FACS, fluorescence-activated cell sorting; GEM, gel bead in emulsion. b, AT8 immunohistochemistry of dorsolateral frontal cortex depth of sulci. The dashed line represents the grey matter–white matter (WM) interface. Scale bars, 100 μm. c, Uniform manifold approximation and projection (UMAP) analysis of nuclei from all donors labelled for cell type based on cell-type marker expression. OPCs, oligocendrocyte precursor cells. d, Expression of cell-type markers across cell-type clusters in c. Astro, astrocytes; Endo, endothelial cells; Exc, excitatory neurons; Inh, inhibitory neurons; Micro, microglia; Oligo, oligodendrocytes. e, Stacked bar plot of pathological group fractions within cell-type clusters. f, Stacked bar plot of cell-type counts coloured by pathological group.

Fig. 2. RHI exposure induces distinct microglial phenotypes.

a, UMAP of microglia coloured by 11 Seurat clusters determined by unsupervised clustering. Mφ, macrophage; PVM, perivascular macrophage. b, Heat map of selected cluster DEGs annotated by function. Mono, monocyters; Ribo. bio., ribosome biogenesis; norm., normalized. c, Proportion of microglial subtypes per pathological group. Statistical analysis was performed using a chi-squared test. n = 28 individuals. Tests were two-tailed. Homeo, homeostasis. d, UMAP of each pathologic group. The dotted line depicts RHIM2/3 subtypes. e, hdWGCNA module analysis showing the Turquoise module localization to the RHIM2/3 subtype. f, GO analysis of the hdWGCNA Turquoise module. Statistics generated using gene set enrichment analyses (GSEA) and single-tailed hypergeometric test with Benjamini–Hochberg multiple hypothesis correction. ER, endoplasmic reticulum; miRNA, micro RNA; PID, Pathway Interaction Database. g–l, Violin plots of the Celda gene modules homeostasis (g), complement response (h), inflammation (i), hypoxia response (j), hypoxia response (k) and metabolic process (l). Colour represents the cellular subtype associated with the module. The black line represents the median. Statistical analysis performed by linear mixed effects modelling, correcting for patient-specific effects. Tests were two-tailed. n = 28 individuals. m, Quantification of grey matter sulcal homeostatic microglia (P2RY12^hiIBA1⁺) with years of football play, coloured by pathological group. Statistical analysis performed by linear regression with age as a covariate. Each dot represents an individual donor. The black line represents linear model regression; the grey region shows the 95% confidence interval. The test was two-tailed. n = 37 individuals. n, Representative image of P2RY12 immunofluorescent labelling. Open arrowheads depict cells exhibiting high P2RY12 expression. Solid arrowheads depict cells exhibiting low P2RY12 expression. Scale bar, 50 μm. o, Representative image of SPP1+HIF1A+P2RY12+ microglia. Solid arrowheads indicate triple-positive cells. The white box indicates the inset displayed on the right. Scale bars: left, 50 μm; right, 5 μm. p,q, Quantification of the SPP1+HIF1A+ microglial fraction (p) and microglial SPP1 expression (q) in the grey matter sulcus with years of football play. Each dot represents an individual donor. n = 22 individuals. Coloured by pathological group status. Statistical analysis performed by linear regression. The test was two-tailed. The black line represents linear model regression; the grey region shows the 95% confidence interval.

Fig. 3. Endothelial angiogenic responses to RHI.

a, UMAP analysis of endothelial cells coloured by endothelial cell subcluster. Solid arrowheads indicate clusters that are enriched in RHI and CTE. Art, arterial; Ven, venous. b, Stacked bar plots of capillary subtype abundance across pathological groups. Statistical analysis performed using a chi-squared test. Tests were two-tailed. c, Dot plot of selected DEGs that are upregulated in RHI and CTE across endothelial subtypes, annotated for function. d, Violin plots of Celda module expression across capillary subtypes. Black bars indicate the median statistic from ggsignif. Statistical analysis performed with linear mixed effects model, accounting for sample variability and comparing Cap2 and Cap3 with Cap1. Tests were two-tailed. n = 28 individuals. e, Violin plot of ITGAV expression across pathological groups. Each dot represents a cell. Statistical analysis performed by Wilcoxon test from ggsignif. Test was two-tailed. n = 28 individuals. f, Scatter plot of the ITGAV+ vessel fraction in the grey matter sulcus with years of football play, coloured by pathological group status using in situ hybridization. Each dot represents an individual donor. Statistical analysis performed by linear regression. n = 19. The line represents linear model regression; the grey region shows the 95% confidence interval. g, Representative image of an ITGAV+ vessel. Solid arrowheads indicate the ITGAV+ vessel. The white box indicates the region in the inset. Scale bars: left, 5 μm; right, 50 μm. Each individual sample was stained and analysed once.

Fig. 4. Synaptic transcriptomic changes and loss of sulcal excitatory layer 2/3 neurons.

a, UMAP analysis of excitatory neurons coloured and labelled by layer subtype. b, Venn diagram depicting the overlap between DEGs from RHI versus control, CTE versus control and RHI versus CTE comparisons. n = 28 individuals. c, Heat map of GO terms identified in comparisons listed in b. Statistics generated using GSEA and single-tailed hypergeometric test with Benjamini–Hochberg multiple hypothesis correction. GPCR, G-protein-coupled receptor. d, Bar plot representing cell counts per pathological group for each excitatory neuron layer subtype. Statistical analysis performed by ordinary one-way analysis of variance (ANOVA) with Bonferroni correction. The test was two-tailed. n = 28 individuals. e, Representative density heat map of CUX2+LAMP5+ cells from in situ hybridization. Solid arrowheads indicate depth of the cortical sulcus. Red indicates high cellular density; blue indicates low cellular density. Scale bars, 1 mm. f, Scatter plot showing the fraction of CUX2+LAMP5+ neurons among total excitatory neurons in snRNA-seq data with total years of football play, coloured by pathological group identity. Dots depict individual donors, the black line represents general linear model regression; the grey region shows the 95% confidence interval. The test was two-tailed. n = 28 individuals. g, Scatter plot showing cell density of CUX2+LAMP5+ neurons in sulcal layer 2/3 from in situ hybridization coloured by pathological group identity, with years of football play. Dots depict individual donors; the black line represents general linear model regression; the grey region shows the 95% confidence interval. n = 28 individuals. Statistics performed by general linear regression. The test was two-tailed. h, Representative images of Nissl-stained neurons in superficial cortical layer 2/3. Scale bar, 50 μm. i, Scatter plot showing Nissl-stained neuronal densities across football exposure groups. Dots depict individual donors; the line represents general linear model regression; the grey region shows the 95% confidence interval. The test was two-tailed. n = 28 individuals.

Fig. 5. Ligand–receptor pair analysis in RHI exposure and CTE.

a, Circos plots from MultiNicheNet analysis depicting microglia as sender cells. RHI comparison with control is labelled RHI; CTE comparison with RHI is labelled CTE. n = 28 individuals. b, RNAScope in situ hybridization depicting a TGFB1+ microglia (P2RY12; solid arrowheads) contacting a ITGAV+TGFBR2+ vessel (GLUT1; open arrowheads). Scale bars, 10 μm. c,d, Quantification of in situ hybridization of TGFB1+ microglia (c) and ITGAV+TGFBR2+ vessels (d) in the grey matter sulcus with years of football play, colour-coded by pathological group. Each dot represents an individual donor. Statistical analysis performed by simple linear regression. The black line represents general linear model regression; the grey region shows the 95% confidence interval. The test was two-tailed. n = 19 individuals. e, Bar plot representing ITGAV+TGFBR2+ vessels with CTE status. Statistical analysis performed using a two-tailed t-test. Data are mean ± s.e.m. The test was two-tailed. n = 19 individuals. f, The proportion of TGFB1+ microglia within 25 μm of a ITGAV+TGFBR2+ vessel with CTE status. Statistical analysis performed using a two-tailed t-test. n = 19 individuals. Data are mean ± s.e.m. g, Scatter plots depicting ITGAV+TGFBR2+ vessels in the grey matter sulcus with the fraction of CUX2+LAMP5+ neurons colour-coded by pathological group. Each dot represents an individual donor. Statistical analysis performed by simple linear regression. n = 17 individuals. The black line represents general linear model regression; the grey region shows the 95% confidence interval. The test was two-tailed.

Extended Data Fig. 1. Dataset quality control and cell type marker validation.

a. Fluorescence activated cell sorting gating strategy of DAPI-positive nuclei. b. Stacked bar plot representing the proportion of cell type per donor. c. Stacked bar plot representing the cell type counts per donor. d-e. Violin plots for each donor of (d) total gene counts per cell, (e) unique genes detected per cell, (f) percent of mitochondrial genes detected per cell, and (g) percent ribosomal genes detected per cell. Line represents median. h, i UMAP of full dataset before cleaning colored by (h) doublet or singlets or (i) mitochondrial contamination. j, k. UMAP of full dataset after cleaning colored by (j) doublets or singlets or (k) mitochondrial contamination. l. UMAP of full dataset colored by Seurat clusters. m. Dot plot of cell type marker expression across Seurat clusters depicted in (l). n. Dot plot of cell type marker expression in annotated cell type clusters.

Extended Data Fig. 2. Cell type proportions, OPCs, Oligodendrocytes, and T-Cells.

a. Bar plots of overall cell type proportions across pathological groups with each dot representing a sample, bars represent the mean, error bars represent standard error of the mean. Statistical analysis performed by ANOVA with Bonferroni correction. Test was two-tailed. n = 28 individuals. b. UMAP depicting OPCs colored by Seurat clustering, solid arrow indicating RHI/CTE depleted cluster. c. Stacked bar plot showing OPC Seurat cluster distribution across pathological groups. d. Bar plots showing OPC cluster distribution across control and pathological group or control and RHI-exposed samples, bar represents mean, error bars show standard error of the mean. Statistical analysis performed by ANOVA with Bonferroni correction (left) and two-tailed Mann-Whitney U test. n = 28 individuals. e. Heatmap showing GO analysis of OPC cluster DEGs. Statistics generated with GSEA and single-tailed hypergeometric test with Benjamini-Hochberg multiple hypothesis correction. f. UMAP showing oligodendrocytes colored by Seurat cluster, solid arrow indicates RHI and CTE depleted cluster. g. Stacked bar plot showing oligodendrocyte pathological group distribution per Seurat cluster. h. Bar plots representing cluster distribution across pathological groups or control and RHI-exposed samples. Bar represents mean, error bar represents standard error of the mean. Statistical analysis performed by ANOVA with Bonferroni correction (left) or two-tailed t-test (right). n = 28 individuals. i. Heatmap showing GO analysis of oligodendrocyte cluster DEGs. Statistics generated with GSEA and single-tailed hypergeometric test with Benjamini-Hochberg multiple hypothesis correction. j. UMAP showing T cells colored by Seurat cluster. k. Heatmap of GO analysis of T cell cluster DEGs. Statistics generated with GSEA and single-tailed hypergeometric test with Benjamini-Hochberg multiple hypothesis correction.

Extended Data Fig. 3. Microglial cluster GO analysis, histology, and validation.

a,b. snRNAseq fraction of (a) homeostatic and (b) RHI microglia. Statistics performed by ANOVA with Bonferroni correction. Bar represents mean, error bars show SEM. Tests were two-tailed. n = 28 individuals. c. Microglial subtypes across control compared to RHI-exposed individuals. Statistical analysis performed by two-tailed t-test or Mann Whitney U test with Welch correction. Bar represents mean, error bars show SEM. n = 28 individuals. d. UMAPs of SPP1 and HIF1A microglial expression. Dashed lines indicate RHIM2/3. e,f. Volcano plots showing DEGs between RHIM2/3 and homeostatic microglia (e) and RHIM2 and RHIM3 (f). n = 28 individuals g,h. Heatmap of (g) GO analysis of RHI microglia and (h) transcriptional regulatory network analysis of microglial subtype DEGs. Statistics generated with GSEA and single-tailed hypergeometric test with Benjamini-Hochberg multiple hypothesis correction. i. UMAPs depicting microglia module scores of microglial subtypes from Sun et al. j. Heatmap depicting Jaccard score similarity analysis between Sun et al. and current study microglial DEGs. Significance denoted by **p < 0.01, ***p < 0.001. Statistical analysis performed using GeneOverlap package and Jaccard analysis settings. k. Representative images of P2RY12, Iba1. Scale bar, 50 μm. l, m, n. SPP1 + /HIF1A+ microglial fraction in the grey matter (l) crest, (m) L2/3 Sulcus (n) layers 4–6 sulcus colored by pathological group compared to years of football play. Statistical analysis performed by linear regression. Black line represents linear regression, grey shows 95% confidence interval. Test was two-tailed. n = 22 individuals. o. P2RY12 low/Iba1+ microglial densities in the grey matter sulcus compared to years of football play. Statistics with linear regression with age included as a covariate. Test was two-tailed. n = 37 individuals. p. Homeostatic microglial densities compared to Nissl+ neuronal densities in layers 2/3 (left) and layers 4–6 (right). Each dot represents a single individual. Statistical analysis performed by linear regression with age as a covariate. Tests were two-tailed. n = 37. Black line represents linear regression, grey shows 95% confidence interval.

Extended Data Fig. 4. Celda module workflow and cell type expression.

a. Celda module workflow diagram. b. Examples of Celda module expression in microglia. UMAPs show module expression, heatmaps show per-cell expression with genes listed on the right. c. Examples of Celda module expression in endothelial cells. UMAPs show module expression, heatmaps show per-cell expression with genes listed on the right. Genes can be viewed in Supplementary Tables 17–19.

Extended Data Fig. 5. Microglia comparison with Sun et al. dataset.

a. UMAP depicting microglia from Sun et al. dataset with original labels. b. Microglia from Butler et al. dataset projected onto Sun dataset UMAP space with original labels. n = 28 individuals. c. Microglia from Butler et al. dataset projected onto Sun dataset UMAP space with predicted Sun labels. n = 28 individuals. d. Bar plot depicting result of boostrapping, with number of predicted microglial labels and error bars depicting bootstrap confidence. e. UMAP depicting Butler microglia projected onto Sun dataset UMAP space colored by label consistency throughout bootstrapping. f. Stacked barplot depicting proportion of Butler microglia with Sun dataset labels across pathological groups. g. Stacked barplot depicting proportion of predicted labels across original labels within Butler dataset microglia showing fidelity across original labels and predicted labels.

Extended Data Fig. 6. Astrocytic responses to head trauma.

a. UMAP representing 4 astrocytic subtypes. b. UMAP from (a) colored by pathological group. c. Stacked bar plots showing astrocyte subtype distribution across pathological groups, statistics performed by chi-squared test. Tests were two tailed. n = 28 individuals. d. Bar plots showing astrocyte subcluster distribution in control and RHI-exposed samples, dots represent individual donors colored by pathological group identity. Bars represent mean, error bars represent standard error of the mean. Statistical analysis was performed using two-tailed Mann Whitney U-test. n = 28 individuals. e. Stacked bar plots showing pathological distribution across astrocyte subtypes. f. Violin plots showing Celda module expression across astrocyte subtypes. Black bar showing median statistic. Colored by astrocyte subtype most associated with specific module expression. Statistical analysis performed by linear mixed effects model. Tests were two-tailed. n = 28 individuals. g. Gene ontology analysis of astrocytic subtypes performed by Metascape. Statistics generated with GSEA and single-tailed hypergeometric test with Benjamini-Hochberg multiple hypothesis correction. n = 28 individuals. h. Dot plot representing expression of selected DEGs across astrocytic subtype and annotated by function. i. Projection of current astrocytic modules on to Visium spatial transcriptomic data. Top row of heatmaps show expression of white matter (PLP1, MBP) and grey matter (SLC27A7, SNAP25) genes. Dotted line indicates separation of grey and white matter. Heatmaps on the bottom row show expression of astrocyte subtype modules based on significantly upregulated genes in each subtype.

Extended Data Fig. 7. Vascular cell subtype identification and proportion analysis.

a. UMAP showing all vascular cells colored by Seurat clustering. n = 28 individuals. b. Heatmap depicting vascular cell marker expression. c. Heatmap depicting Jaccard scoring of vascular cell Seurat cluster DEGs compared to Sun and Akay et al. vascular subtype DEGs. Significance denoted by **p < 0.01, ***p < 0.001. d. Bar plots depicting pathological group proportions of vascular subtypes, bar represents mean, error bar represents SEM, dots represent individual samples. Statistical analysis performed by ANOVA with Bonferroni correction. Tests were two-tailed. n = 28 individuals. e. Heatmap depicting Jaccard scoring of vascular cell subtype DEGs compared to Sun and Akay et al. vascular subtype DEGs. Significance denoted by ** p < 0.01, *** p < 0.001. f. Scatter plot of fibroblast proportion or Cap2 proportion compared to years of football play from snRNAseq dataset, colored by pathological group status. n = 28 individuals. Statistical analysis performed by linear regression with age as a covariate. Black line represents linear model regression, grey shows 95% confidence interval. Tests were two-tailed. g, h. Bar plots of total capillary and relative endothelial cell subtype distribution across control and RHI-exposed samples, dots represent individual donors and are colored by pathological group identity. n = 28 individuals. Bar indicates mean, error bars indicate standard error of the mean. Statistical analysis was performed by two-tailed Mann-Whitney U test. i. GO enrichment analysis of DEGs from depicted comparisons. Statistics generated with GSEA and single-tailed hypergeometric test with Benjamini-Hochberg multiple hypothesis correction. n = 28 individuals. j. Proportion of cap2 cells in each individual compared to years of football. Statistics generated with linear regression correcting for age. Black line represents linear model regression, grey shows 95% confidence interval. Each dot represents a single individual. Tests were two-tailed. n = 28 individuals.

Extended Data Fig. 8. Neuronal layer subtype identification.

a. UMAP depicting all neurons clustered together colored by Seurat cluster. n = 28 individuals. b. Dot plot of gene expression of inhibitory and excitatory neuron and astrocyte marker genes Seurat clusters from (a). c. UMAP from (a) colored by cell type determination. d. Stacked bar plot of sequencing batch distribution of Seurat clusters from (a). e. UMAP showing excitatory neurons colored by Seurat cluster. f. UMAP showing excitatory neurons colored by later subtype. g. Dot plot showing expression of excitatory neuron layer subtype genes in excitatory neuron Seurat clusters from (e). h. UMAP showing inhibitory neurons colored by Seurat cluster. i. UMAP showing inhibitory neurons colored by layer subtype. j. Dot plot showing expression of inhibitory neuron layer subtype genes across inhibitory neuron Seurat clusters from (h).

Extended Data Fig. 9. Neuron layer GO analysis, pathological group enrichment and RNAScope validation.

a. UMAP depicting excitatory neurons colored by layer subtype. b,c. Heatmap showing GO analysis of (b) excitatory layer and (c) inhibitory layer up and downregulated DEGs. n = 28 individuals. Statistics generated with GSEA and single-tailed hypergeometric test with Benjamini-Hochberg multiple hypothesis correction. d. Bar plots of excitatory neuron layer proportions by pathological group. Bar represents mean, dots represent individual samples, error bars show standard error of the mean. Statistical analysis performed by ANOVA with Bonferroni correction. Tests were two tailed. n = 28 individuals. e. UMAP showing inhibitory neurons colored by layer subtype. f. Bar plots of inhibitory neuron layer proportions by pathological group. Bar represents mean, dots represent individual samples, error bars show standard error of the mean. Statistical analysis performed by ANOVA with Bonferroni correction. Tests were two tailed. n = 28 individuals. g. Representative image showing RNAScope in situ hybridization of CUX2/LAMP5 image analysis with correct anatomical layer-wise distribution. White squares showing HALO identification of double-positive cells. Scale bar = 1 mm. h. Representative image of Nissl+ staining and neuronal masking using HALO AI. Top box is the raw image and bottom box is AI generated mask over neurons. Scale bar = 50 µm.

Extended Data Fig. 10. Layer 2/3 neurons are selectively lost in the grey matter sulcus and do not associate with tau pathology.

a. CUX2/LAMP5 proportion from snRNAseq against total years of football play colored by pathological group. Statistical analysis performed by linear regression, depicted as black line. Test was two-tailed. n = 28 individuals. b. CUX2/LAMP5 density from in situ hybridization compared to highest level of football played. Statistical analysis performed by linear regression. Dots represent single individuals; line shows linear regression. Test was two-tailed. n = 23 individuals. c. CUX2/LAMP5 cells identified by in situ experiment compared to proportion of CUX2/LAMP5 neurons from snRNAseq experiment. Statistical analysis performed by linear regression, depicted as black line with 95% confidence intervals in grey. Test was two-tailed. n = 7 individuals. d. Representative image showing the annotation strategy used to identify the sulcus (yellow line) and crest (red line) layer 2/3. Scale bar = 1 mm. e. (left) Layer 2/3 CUX2+/LAMP5+ neuronal density in the sulcus and crest. Statistical analysis performed by paired t-test. Tests were two tailed. n = 18 individuals. (right) Layer 2/3 CUX2+/LAMP5+ neuronal density in the crest compared to years of football play, colored by pathological group. Statistical analysis performed by linear regression. Black line represents general linear model regression, grey shows 95% confidence interval. Tests were two tailed. n = 18 individuals. f. Total CUX2 populations and subpopulations in snRNAseq (right panel) and in situ hybridization (left panel) experiments compared to years of football play. Statistical analysis performed by linear regression and represented by lines. Tests were two-tailed. left panel n = 22 individuals, right panel n = 28 individuals. g, h, i. Scatter plots depicting (g) Sulcus layers 4–6 Nissl+ density compared to binned years of football play, (h) Crest layer 2/3 Nissl+ density compared to binned years of football play, (i) L2/3 Nissl+ density compared to log tau+ density, Statistical analysis performed by linear regression, (g, h) corrected for age and staining batch. n = 86 individuals. j. CUX2+/LAMP5+ density in situ compared to log AT8+ tau+ density, Colored by pathological group. Statistical analysis performed by linear regression. Test were two-tailed. Black line represents general linear model regression, grey shows 95% confidence interval. n = 22 individuals. k, l. Representative images of Nissl staining across cortical layers depicting neuronal loss in superficial layers in RHI and CTE individuals. Scale bars, 100 μm.