C9orf72 hexanucleotide repeat expansions impair microglial response in ALS

Abstract

Microglia and neuroinflammation are involved in amyotrophic lateral sclerosis (ALS), but the precise underlying molecular mechanisms remain elusive. We generated single-nuclei transcriptomes from the spinal cord and motor cortex of patients with sporadic ALS (sALS) and C9orf72 ALS (C9-ALS). Here we confirmed that C9orf72 is highly expressed in microglia and observed that the hexanucleotide repeat expansion (HRE) results in haploinsufficiency. Whereas sALS microglia transitioned toward disease-associated cell states, C9orf72 HRE microglia exhibited a diminished response, with alterations in endolysosomal pathways. We confirmed these observations using a human microglia xenograft model, in which C9orf72 mutations led to a reduced activation. We also confirmed the endolysosomal alterations in C9orf72 HRE and C9orf72-deficient induced pluripotent stem cell (iPSC)-derived microglia. We also found a diminished response of C9orf72 HRE astrocytes and provided a map of dysregulated ligand-receptor pairs in microglia and astrocytes. Our data highlight variations in the cellular substrate of sporadic and inherited forms of ALS, which have implications for patient stratification and selection of appropriate treatments.

Fig. 1. snRNA-seq of human postmortem motor cortex and spinal cord of patients with ALS.

a, UMAP plot visualizing 93,176 nuclei isolated from postmortem cortical samples of patients with ALS (C9-ALS and sALS) and healthy controls. Cells are annotated and colored according to their assigned cell identity into astrocytes, endothelial cells, neurons, microglia, oligodendrocytes and oligodendrocyte precursor cells. b, UMAP plot visualizing 63,076 nuclei isolated from postmortem spinal cord samples of patients with ALS (C9-ALS and sALS) and healthy controls. Cells are annotated and colored according to their assigned cell identity into astrocytes, endothelial cells, microglia, natural killer cells, neurons, oligodendrocytes and oligodendrocyte precursor cells. c, Heatmaps showing the top 10 most expressed genes per cluster. d, Heatmap depicting the z-scored average expression of C9orf72 across different cell types in both motor cortex and spinal cord. NK, natural killer.

Fig. 2. C9orf72 HRE changes the microglial transcriptomic profile compared to sALS.

a, UMAP plot visualizing 16,763 microglia nuclei from ALS (C9 and sALS) and healthy postmortem samples. Microglia subsets are colored based on their cluster into MG1, MG2, MG3, MG4, MG5 and MG6. b, UMAP plot displaying 149,172 nuclei after integration with the dataset by Gerrits et al.. The correlation between the clusters in a and b is shown in Extended Data Fig. 2. c, Heatmap showing the top 20 most expressed genes per cluster from b. d, Graphical representation of the neighborhood analysis using Milo. e, Relative enrichment of the neighborhood groups (left) in brain and spinal cord samples. The pie charts (right) depict the contribution of cells in patient group and sampling locations in the neighborhood groups. f, Overlap of markers from MG5 and differential abundance (DA) group 6. g, Volcano plot depicting the pairwise comparisons of C9-ALS versus sALS for the combined microglial dataset (DESeq2 negative binomial distribution; P values adjusted with Bonferroni correction based on the total number of genes in the dataset). Vertical lines correspond to the log₂FC threshold of 0.2, and the horizontal line corresponds to the adjusted P value threshold at <0.05. Colors indicate marker genes of microglia clusters in b. Positive log₂FC values are enriched in C9-ALS; negative log₂FC values are enriched in sALS. NA, not assigned to any cluster. h, Seurat module scores of microglia markers as described by Mancuso et al. visualized in the single-nuclei dataset, displayed per patient group and location. Black dots indicate the mean, and lines indicate the mean of healthy controls per location. i,j, Scatter plot depicting two pairwise comparisons of microglia from C9-ALS and healthy controls in the spinal cord versus GRN-FTD across cortical regions (R = 0.15, P < 2.2 × 10⁻¹⁶, DESeq2 negative binomial distribution) (i) and sALS and healthy controls in the spinal cord versus GRN-FTD and healthy controls across all cortical regions (R = −0.47, P < 2.2 ×10⁻¹⁶) (j). Lines represent the log₂FC threshold of ±0.2. Positive log₂FC values display enrichments in disease states; negative log₂FC values display enrichments in healthy controls. P values are included in Supplementary Tables 7, 8 and 16. FC, fold change.

Fig. 3. The C9orf72 HRE alters microglia in an HBCX model.

a, Schematic overview of the experimental design. MPs, microglial precursors; hMG, human microglia. Created with BioRender.com. b, UMAP plot visualizing microglia cell states colored based on their assigned cluster: HM, RM, tCRM, CRM1, CRM2, HLA, DAM and IRM. c, Heatmap showing the most expressed genes per microglial cluster. Volcano plots comparing C9-HRE versus control microglia (d) and C9KO versus control microglia (e). Top differentially downregulated and upregulated genes are reported and color coded based on the corresponding microglial cluster. NA, not assigned to any cluster. Wilcoxon test; P values adjusted with Bonferroni correction based on the total number of genes in the dataset (adjusted P < 0.05). P values can be found in Supplementary Tables 19 and 20. Number of mice per experimental group: C9-ISO n = 4, C9-HRE n = 2, CTRL n = 7, C9KO n = 10. Number of cells analyzed per group: C9-ISO = 1,184, C9-HRE = 554, CTRL = 11,104, C9KO = 10,386. Timepoint analyzed: C9-HRE and C9-ISO 3 months, CTRL and C9KO 6 months. FC, fold change.

Fig. 4. C9orf72 deficiency alters lysosomal fitness in iPSC-derived microglia.

a, CTSD protein levels in cortex (top) and spinal cord (bottom) of C9-ALS and sALS. Western blot quantification of cortex (b) and spinal cord (c). Data were normalized to GAPDH. ANOVA with Tukeyʼs post hoc multiple correction testing (n = 2 healthy controls, n = 5 C9-ALS and n = 5 sALS). Western blot was repeated two independent times, and the technical replicates were averaged. d, Quantification showing increased abundance of lysosomal (CTSD) clusters in C9-HRE iPSC microglia versus C9-ISO. Unpaired t-test (C9-ISO n = 4 independent replicates; C9-HRE n = 5 independent replicates). e, High-resolution expansion microscopy images and quantification showing increased abundance of lysosomal clusters in C9KO microglia. Scale bar, corrected for a 4× expansion of the sample, 2.5 µm. f, Quantification showing increased abundance of CTSD clusters in C9KO microglia versus controls. Unpaired t-test (CTRL n = 5 independent replicates; C9KO n = 5 independent replicates). g,h, TEM images and quantification showing increased abundance of late endosomes/lysosomes in C9KO compared to control microglia. Unpaired t-test (CTRL n = 5 independent replicates; C9KO n = 5 independent replicates). Scale bar, upper images, 10 µm; lower images, 20 µm. i, Microscopy images showing RAB7-positive and LAMP1-positive structures (that is, late endosomes) in C9KO and control iPSC microglia. Scale bar, 5 µm. Box plots in b−d, f and h show the median, interquartile range and full data range. Live-cell imaging experiment showing the kinetics of pHrodo-conjugated E. coli particles accumulation in C9KO versus control microglia over 15 hours (j). Data points represent log₁₀-transformed median pHrodo intensities, aggregated via a two-step median process (Methods). LOESS curves with 95% confidence intervals (gray area) were plotted to illustrate temporal trends (n = 6 independent replicates per genotype). k, Quantification of AUC and pHrodo median intensity at 30 minutes (l) and 900 minutes (m) as a proxy for uptake and cargo degradation processes, respectively. Ratio paired t-test (n = 6 independent replicates per group). Source data

Fig. 5. Astrocytes display transcriptomic alterations in C9-ALS compared to sALS.

a, UMAP plot visualizing 22,635 astrocytes from postmortem samples of patients with ALS (C9-ALS and sALS) and healthy controls. Astrocyte subsets are colored based on assigned cluster: AS1, AS2, AS3, AS4, AS5, AS6, AS7, AS8. b, UMAP plot displaying 216,574 nuclei after integration with the dataset from Gerrits et al.. New astrocyte subsets are colored based on their assigned cluster: AS1, AS2, AS3, AS4, AS5, AS6. The correlation between clusters in a and b is shown in Extended Data Fig. 5. c, Top 10 most expressed genes per cluster in b. Graphical representation of Milo’s differential abundance (DA) testing (d) and neighborhood group (e). f, Pie charts displaying the ratio of cells sampled in each neighborhood from patient groups and sampling locations. g, Changes in the distribution of neighborhood groups across brain and spinal cord (left) and average expression of differentially expressed genes (right) from the C9-ALS and sALS versus controls. Note that neighborhoods 5 and 6 are enriched in genes differentially expressed in C9-ALS and sALS astrocytes, respectively. Pathway analysis of the neighborhood markers of group 5 (h) and group 6 (i), analyzed using one-sided hypergeometric test. j, Volcano plot depicting pairwise comparisons between C9-ALS and healthy controls and sALS and healthy controls for the combined astrocytes dataset (DESeq2 negative binomial distribution; P values adjusted with Bonferroni correction on the total number of genes in the dataset). Vertical lines correspond to a log₂FC = 0.2 threshold, and horizontal line corresponds to an adjusted P < 0.05 threshold. Positive log₂FC values are enriched in C9-ALS (purple) or sALS (yellow); negative log₂FC values are enriched in healthy controls (yellow and purple). k, Volcano plot showing a pairwise comparison of C9-ALS versus sALS for the combined astrocyte dataset (DESeq2 negative binomial distribution; P values adjusted with Bonferroni correction on the total number of genes in the dataset). Vertical lines depict a log₂FC = 0.2 threshold, and horizontal line depicts the adjusted P < 0.05 threshold. Genes are colored based on the clusters in b. Positive log₂FC values are enriched in C9-ALS, and negative log₂FC values are enriched in sALS. P values are presented in Supplementary Tables 22 and 23. FC, fold change.

Fig. 6. Map of microglia−microglia and microglia−astrocyte communication is predicted to be altered in the spinal cord of patients with ALS.

Dot plots depicting scaled differential communication probabilities of microglia and astrocytes in healthy, C9-ALS and sALS spinal cords (a) and scaled average expression of both the source and the targets across the transcriptional clusters of microglia (Fig. 2b) and astrocytes (Fig. 5b) (b).

Extended Data Fig. 1. Overview of the full dataset and quality control.

(a) UMAP plot visualization of the single nuclei from motor cortex and spinal cord prior and after batch correction. (b) Quality control of the three isolation protocols (Habib, Soma-seq, TST): percentage of mitochondrial RNA, number of molecules detected per cell, number of genes detected per cell. (c) Contribution of each patient (y-axis) to each cluster (x-axis) of the motor cortex and spinal cord. (d) Contribution of each patient (y-axis) to each annotated cell type of the motor cortex and spinal cord.

Extended Data Fig. 2. Extended clustering and quality control in the microglial subset.

(a) UMAP plot visualizing 4,845 nuclei from the motor cortex and 13,980 nuclei from the spinal cord of the ALS dataset. Cell populations are coloured based on their assigned into the following clusters: microglia, peripheral macrophages, doublets (neurons), doublets (oligodendrocytes) and cycling cells. (b) Heatmap showing the top 20 most expressed genes per cluster. (c, d) Enrichment of transcriptomic signatures of different human microglial cell states from Mancuso et al.. in (c) the clusters from the ALS dataset and (d) the final clusters after integration with Gerrits et al. (e) Fraction of cells (y-axis) from each cluster in each subject, divided by anatomical regions (x-axis). (f) Heatmaps showing differentially expressed genes (log2FC, p-value adjusted <0.05) in both C9-ALS or sALS patients compared to healthy controls in the spinal cord, the motor cortex, or both locations combined. (g) Log fold change of the neighbourhood groups distribution between brain and spinal cord and (h) expression values of differentially expressed genes in C9-ALS and sALS, vs. healthy controls.

Extended Data Fig. 3. WGCNA reveals dysregulated gene modules in C9 dWGCNA reveals dysregulated gene modules in C9.

(a) Dendrogram showing the result of unsupervised hierarchical clustering of modules identified by WGCNA. (b) Bar plots showing the number of genes per module. (c) Boxplots visualizing the module eigengenes (y-axis) per genotype (x-axis) for the blue, turquoise, yellow, black, brown, red and green cluster in the spinal cord. (d) Pathway enrichment analysis of the WGCNA clusters blue, turquoise and yellow, with the associated GO term (y-axis) and –log10 p-value (colour heatmap), coloured per direction of expression changes across the patient groups.

Extended Data Fig. 4. Microglia from C9orf72 HRE patients show transcriptional alterations in functional pathways.

Heatmaps depicting differentially expressed genes (log2FC < +/-0,20, p-value adjusted <0,05) from selected gene sets associated with lysosomal, inflammatory, phagocytic and oxidative phosphorylation functional pathways.

Extended Data Fig. 5. Extended data on the HBDX model.

(a) Gating strategy for the isolation of human microglia from the HBDX model (CD11b + hCD45 + ). (b) Scatter plot showing the number of unique genes (nFeature_RNA) and total RNA counts (nCount_RNA) per cell, across five single-cell RNA sequencing libraries (LIB14, LIB17, LIB28, LIB59, and LIB60). Each dot represents an individual cell, coloured according to the percentage of mitochondrial gene expression. Marginal density plots display the distribution of nFeature_RNA (top) and nCount_RNA (right) for each library (c) UMAP plot showing cell populations coloured based on their assignment into the following clusters: microglia, macrophages, doublets (neurons), cycling cells, and doublets. (d) Heatmap showing the top 15 most expressed genes per cluster. (e) Fraction of cells per genotype in each cluster. (f) Percentage of cells in HLA cluster, showing reduced percentage of cells in C9KO versus controls. Mann-Whitney test (CTRL n = 7, C9KO n = 10 mice). Box plots show the median, interquartile range, and full data range.

Extended Data Fig. 6. Extended data related to Fig.4.

(a) Schematic representation of the microglial differentiation protocol and plating for in vitro experiments, based on Fattorelli et al.. Created in BioRender. Polanco, P. (2025) https://BioRender.com/83a97ah (b) Representative images of IBA1 immunostaining for control and C9KO iPSC-derived microglia. Scale bar, 20 µm. (c) Heatmap showing enrichment of microglial-related genes versus macrophages after iPSC differentiation in vitro (n = 7 independent replicates). (d) Example of low-magnification confocal images used for quantification of CTSD clusters in Fig. 4d and f. Upper panel: C9-HRE and C9-ISO; lower panel: C9KO and control microglia. Cells containing at least one CTSD puncta were considered for quantification in ImageJ. The percentage of cells containing CTSD positive puncta was calculated on total DAPI positive cells. (e) Representative low-magnification TEM images for C9KO and control iPSC-derived microglia that were used for quantification of endolysosomes/lysosomes clusters in Fig. 4h. Scale bar, 20 µm. (f) Quantification of percentage of cells containing enlarged endolysosomes/lysosomes. Unpaired T-test (CTRL n = 5 independent replicates, C9KO n = 5 independent replicates). Box plots show the median, interquartile range, and full data range. (g) Microscopy images showing EEA1 (that is, early endosomes) and Lamp1 positive structures in control and C9KO microglial cells. Scale bar, 5 µm. Flow cytometry analysis of uptake of (h, i) amyloid beta and (j, k) pHrodo E. coli particles, showing no difference in the median fluorescence intensity (MFI) between C9KO and controls, at 3 h and 1 h upon treatment, respectively. Ratio paired T-test (amyloid beta, CTRL n = 3 independent replicates, C9KO n = 3 independent replicates; pHrodo E. coli CTRL n = 4 independent replicates, C9KO n = 4 independent replicates).

Extended Data Fig. 7. Extended clustering and quality control in the astrocytic subset.

(a) UMAP plot visualizing 10,119 nuclei from the motor cortex and 13,364 nuclei from the spinal cord. Cell populations are coloured based on their assigned cluster into astrocytes, oligodendrocytes, doublets/low quality. (b) Heatmap showing the top 20 most expressed genes per cluster. (c) Heatmap showing average expression of the top 100 markers from the original 8 astrocyte clusters (Fig. 5a) onto the novel clustering after integration with Gerrits et al.. (Fig. 5b). (d) Proportion of cells per cluster (y-axis) for each individual, across all the different anatomical regions. (e) Heatmaps showing differentially expressed genes (log2FC, p-value adjusted <0.05) between C9-ALS or sALS and healthy controls, and displayed for the spinal cord, motor cortex or both locations combined.

Extended Data Fig. 8. Extended CellChat analysis to astrocyte-astrocyte communication pathways.

Dot Plots depicting (a) scaled communication probabilities of differential pathways in healthy, C9-ALS and sALS patients’ spinal cord, and (b) scaled average expression of the source and target in the astrocyte cell states from Fig. 5b.