Rod-shaped microglia interact with neuronal dendrites to attenuate cortical excitability during TDP-43 related neurodegeneration

SUMMARY

Microglia, the principal immune cells of the central nervous system, have emerged as important players in sensing and regulating neuronal activity. While microglial activation is a hallmark in neurodegeneration, the specific role of microglia in disease-related cortical excitability remains unknown. Utilizing multichannel probe recording and longitudinal in vivo calcium imaging, we observed neuronal hyperactivity at the initial stage of disease progression in a mouse model of TDP-43 neurodegeneration (rNLS8). Spatial and single-cell RNA sequencing revealed a specific subpopulation of microglia, rod-shaped microglia, with a distinct morphology and direct response to cortical hyperactivity. Rod-shaped microglia predominantly interacted with neuronal dendrites and remodeled excitatory synaptic inputs to attenuate motor cortical hyperactivity. TREM2 deficiency led to a marked reduction of rod-shaped microglia accompanied by increased neuronal activity in rNLS8 mice. Together, our results suggest that rod-shaped microglia play a neuroprotective role by attenuating cortical hyperexcitability in TDP-43-related neurodegeneration.

Graphical Abstract

eTOC Blurb

While microglial activation is a hallmark in neurodegeneration, the specific role of microglia in disease-related cortical excitability remains unknown. Xie et al. reveal that rod-shaped microglia form in response to early cortical hyperactivity in a mouse model of TDP-43 neurodegeneration. These specialized microglia interact with neuronal dendrites and remodel excitatory input, providing a neuroprotective effect.

INTRODUCTION

Amyotrophic lateral sclerosis (ALS) is a progressive neurodegenerative disease that defined by its effect on motor neurons in brain and spinal cord. The motor cortical circuit is intricately involved in ALS pathology ¹. Clinical observations highlight cortical hyperexcitability as an early characteristic in individuals with ALS, which could contribute to early neuronal toxicity and motor neuron degeneration ². Cortical hyperexcitability has also been observed in ALS mouse models ³ and other neurodegenerative diseases such as AD ^4,5 and frontotemporal dementia ⁶. Identifying the intrinsic mechanisms of early aberrations within the motor cortical circuit may reveal novel therapeutic targets capable of arresting disease progression in its early stages.

Microglia play a critical role in maintaining brain homeostasis ^7,8. Emerging evidence has highlighted microglial function in monitoring and regulating neuronal activity under both physiological and pathological conditions ^9,10. Notably, either neuronal hyperactivity (e.g., seizures) or hypoactivity (e.g., anesthesia) increases microglial process dynamics and interactions with neurons ^11-13. While microglial activation is acknowledged as a hallmark in the ALS ¹⁴, their specific role in motor cortical excitability remains unknown. Understanding the role of microglia in motor cortical excitability in ALS could potentially lead to novel therapeutics.

In this study, we utilized a mouse model of TDP-43 neurodegeneration, the rNLS8 mice, in which human TDP-43 without a nuclear localization sequence is expressed mostly in neurons in a DOX-dependent manner ¹⁵. Upon initiating TDP-43 expression by removing the DOX diet, the mice exhibited characteristic motor deficits ^15,16. This feature allowed us to precisely monitor neuronal activity during disease progression in real-time fashion. We observed neuronal hyperactivity and a distinct microglia subpopulation, rod-shape microglia, in the cortex at the initial stages of disease progression in the rNLS8 mice. Rod-shaped microglia formed tight complexes with neuronal apical dendrites that may remodel synaptic inputs to attenuate cortical excitability. Our findings highlight the neuroprotective role of rod-shaped microglia in dampening neuronal excitability within the motor cortical circuits in TDP-43 neurodegeneration.

RESULTS

Cortical hyperactivity occurs early in the motor cortex of rNLS8 mice

Hyperexcitability has been previously observed in the early stages of individuals with ALS ². We recorded extracellular neuronal activity across cortical layers in rNLS8 mice with a chronically implanted silicon probe (32-channels over 750 μm in primary motor cortex; Figure 1A), starting with the baseline (DOX ON: no TDP-43 overexpression) and then a 4-week period of disease progression (DOX OFF: TDP-43 overexpression; Figure 1B). Excitatory and inhibitory neurons sorted by spike duration and interspike interval (ISI) histograms (Figures S1A-S1F). The probe of 32 channels successfully covered neuronal activities from all cortical layers (Figure S1G). As expected, inhibitory neurons showed a higher firing frequency compared to excitatory neurons at baseline (Figure S1H) and neuronal activity was stable in control mice over consecutive recoding days (Figures S1I and S1J). In rNLS8 mice, we observed that the firing frequency of excitatory neurons increased during the first two weeks following DOX diet removal in both layer 2/3 and layer 5 (Figure 1C). Next, we employed longitudinal in vivo two-photon calcium imaging of excitatory neurons in the primary motor cortex (layer 2/3) through a chronic cranial window (Figures 1D and 1E). In alignment with silicon probe findings, neuronal calcium transient frequency gradually increased following DOX diet removal, reaching peak elevations 2-3 weeks post-DOX diet removal in rNLS8 mice (Figures 1F-1I; Video S1). Correspondingly, calcium signal area also showed a sustained increase and these neurons in rNLS8 mice exhibited a 228 ± 49% increase in the average ΔF/F·s calcium signal area at the peak (Figure 1J), while control mice exhibited consistent calcium activity over the 4-week recoding period (Figures 1G-1J; Video S1). These observations indicate that rNLS8 mice experience cortical hyperexcitability during the early stage of TDP-43 overexpression.

Figure 1. Motor cortical hyperactivity occurs early in the rNLS8 mice. See also Figure S1 and Figure S2.

(A) Schematic illustration of in vivo recording using a silicon probe in the motor cortex of the control and rNLS8 mice. M1, Primary motor cortex; IT, intratelencephalic neuron; PT, pyramidal tract neuron. (B) Experimental paradigm outlining probe implantation surgery and timeline for in vivo recording. After baseline recording, mice were switched to DOX-free chow to induce hTDP-43ΔNLS expression. (C) Firing frequency of pyramidal neurons during baseline and disease progression (n = 7 per group). Data are from 2 independent experiments. (D) Schematic of in vivo two-photon Ca imaging with a representative average intensity projection image of neurons expressing CamkIIα-GCaMP6s. Scale bar, 20 μm. (E) Experimental paradigm for virus injection and cranial window implantation surgery, with a timeline for chronic window imaging. (F) Representative ΔF/F calcium traces from the soma of a neuron expressing CamkIIα-GCaMP6s, with the threshold line, single transient, and signal area of Ca activity indicated by pink arrows separately. (G) Representative images of layer 2/3 neuronal calcium activity. Scale bars, 100 μm. (H) Heatmap of calcium activity from 60 representative somatic ROIs in one animal under baseline and DOX off day 16. (I) Quantification of transients (ratio) of neuronal calcium activity (n = 5 per group). Data are from 3 independent experiments. (J) Quantification of ΔF/F·s signal area (ratio) of neuronal calcium activity (n = 5 per group). Data are from 3 independent experiments. Statistical analysis, one-way ANOVA followed by Fisher’s post hoc test (C, dot: one cell), two-way ANOVA followed by Sidak’s post hoc test (I, J, dotted lines represent longitudinal imaging regions). Error bars, mean ± s.e.m. NS = not significant; *P < 0.05; **P < 0.01; *** P < 0.001; **** P < 0.0001.

We next explored the potential mechanisms underlying hyperactivity in pyramidal neurons in rNLS8 mice. NeuN staining revealed no significant changes in neuron density up to 3 weeks post-DOX diet removal (Figures S2A and S2B). However, there was a significant reduction in neuron soma size and increased cleaved caspase-3 staining as the disease progressed, suggesting neuronal stress despite stable neuronal numbers (Figures S2C-S2H). Neuronal morphology analysis through vial labeling and in vivo two-photon imaging revealed pronounced dendritic fragmentation and loss of dendritic spines at 2 weeks post-DOX removal in rNLS8 mice, further supporting the presence of neuronal stress (Figures S2I and S2J). Next, we conducted electrophysiological recordings of pyramidal neurons at 16 days post-DOX removal, a time point coinciding with the peak of neuronal hyperactivity. These recordings indicated that pyramidal neurons in rNLS8 mice exhibited increased excitability compared to control mice (Figures S2K-S2M). Notably, multi-channel recordings showed the firing of inhibitory neurons slightly increased at 2 weeks following DOX diet removal (Figures S2N and S2O), which may reflect a compensatory response to the early rise in excitatory neuron activity or TDP-43 overexpression in inhibitory neurons. Altogether, these data demonstrate cortical hyperactivity within the motor cortex of rNLS8 mice at early stages of disease progression, which is associated with increased neuronal stress and excitability of pyramidal neurons.

Spatial transcriptomics analysis reveals cortical microglia as the major responder

To explore the broad picture of cellular changes in the brain of rNLS8 mice in response to neuronal hyperactivity, we performed spatial RNA sequencing (spRNA-seq, 10x Visium) to create transcriptomic maps across various brain regions at three weeks post-DOX diet removal, a time point selected to capture peak neuronal hyperactivity (Figure 2A). Each section provided over 1,700 transcriptomic profiles from individual spots, totaling 8,474 profiles with a median depth of 8,619 unique molecular identifiers (UMIs) per spot and an average of 3,754 genes per spot (Figures S3A-S3E). By aligning spatial gene expressions with immunofluorescence staining (NeuN/DAPI), we confirmed spatial concordance for the transcriptomic data (Figure 2B). Using a clustering-based approach, we classified them into 6 major regions, including the cerebral cortex (CTX, layers I to V), cortical subplate (CTXsub), olfactory areas (OLF), white matter (WM), striatum (STR) and pallidum (PAL), based on the Allen Mouse Brain Atlas ¹⁷ (Figures 2C-2E; Figure S3F; Table S1A-S1M). Notably, we identified a distinct cortical cluster in the rNLS8 mice, termed the 'sp-rNLS8-specific cluster' (Figures 2C-2E). All the clusters were consistently captured across each sample (Figures S4A-S4D).

Figure 2. Spatial transcriptomics analysis of the cortical area reveals microglia as the major responder. See also Figure S3, Figure S4 and Table S1.

(A) Schematic workflow for spatial RNA sequencing (spRNA-seq). Mice were euthanized at 3 weeks post-DOX diet removal. Two coronal brain tissue sections from control or rNLS8 mice were processed for 10X Visium spatial transcriptomics. Data are from one independent experiment. (B) Immunostaining of NeuN and DAPI in the 10X Visium slices. Scale bar, 300 μm. Inset shows NeuN/DAPI expression with the barcoded spots at higher magnification as indicated by the area in the dotted white box in the motor cortex. Scale bar, 100 μm. (C-D) Representative spatial transcriptome data, colored by cluster-level annotation and represented as spatial transcriptome (C) and UMAP (D). Cluster-level data determined by Seurat clustering. (E) Data description and abbreviations of ontology and nomenclature. (F) Differential gene expression analysis across all brain regions. (G) Visualization of the number of differentially expressed genes (DEGs) in the corresponding brain regions. (H) Volcano plot of DEGs in 'rNLS8-specific cluster' from spRNA-seq data. The cortex (CTX) cluster (combining all cortical layers) in the control group was served as baseline. Significance is indicated at adjusted P value ≤ 0.05 and ∣avg_log2FC∣ ≥ 0.5 (log2FC, log2-fold change). (I) Heatmap of DEGs in 'rNLS8-specific cluster' from spRNA-seq data. (J) Spatially resolved expression of select genes across tissue sections from control and rNLS8 mice. (K) Gene Ontology (GO) enrichment analysis was performed for the 'rNLS8-specific cluster' from spRNA-seq data. Dot plot of the selected GO terms in order of gene ratio. The size of each dot indicates the number of genes in the significant DEG list that are associated with each GO term, and the color of the dots corresponds to the adjusted P-values.

We then analyzed differentially expressed genes (DEGs) between the rNLS8 group and the control group. The CTX cluster (combining all cortical layers) in the control group served as baseline and a total of 470 DEGs across all clusters were identified, with the 'sp-rNLS8-specific cluster' and CTX cluster showing more pronounced changes (Figures 2F and 2G; Table S1N). Despite its small size, the 'sp-rNLS8-specific cluster' exhibited the most significant number of DEGs (Figures 2F-2H; Table S1N). Notably, the majority of the top 100 upregulated DEGs in this cluster were highly expressed in microglia (https://brainrnaseq.org), including members of the complement C1q family, Cst7, Hexb, Trem2, Tyrobp, Cd68, Csf1r, Lgals3bp, H2-k1, Axl, and Ifitm3, among others (Figure 2I; Table S1N). These microglia-specific genes were further evaluated anatomically through spRNA-seq (Figure 2J). Additionally, Gene Ontology (GO) enrichment analysis confirmed that 'microglia activation' was one of the most upregulated functional terms, along with other terms related to 'immune responses', 'phagocytosis' and 'neuroinflammatory response' (Figure 2K).

To assess whether microglial changes spatially correlate with the ‘sp-rNLS8-specific cluster’, we performed immunostaining on the same tissue sections used for spRNA-seq and confirmed that activated microglia are spatially enriched within the ‘sp-rNLS8-specific cluster’ (Figures S4E). DEGs and GO enrichment analysis from the CTX cluster were consistent with those from 'sp-rNLS8-specific cluster' (Figures S4F-S4H). Additionally, we observed a subset of differentially expressed neuronal-related genes in the CTX cluster enriched in GO functional terms such as 'synaptic transmission', 'synapse organization' and 'synapse pruning' (Figures S4F-S4H), which align with increased neuronal excitability in rNLS8 mice. Overall, our spatial RNA-seq data reveal transcriptional reprogramming in the cortical regions in rNLS8 mice, with microglia emerging as primary responders.

Rod-shaped microglia identified in rNLS8 mice and the brains of individuals with ALS

To further assess the transcriptional profile changes, we conducted single cell RNA sequencing (scRNA-seq) (Figure 3A). Through unsupervised clustering, we identified 13 distinct clusters, including microglia, vascular endothelium, astrocytes, neurons, ependymal cells, endothelial cells, choroid plexus cells, smooth muscle cells, pericytes, monocytes/macrophages, mitotic microglia, lymphocytes, granulocytes, and fibroblasts (Figures 3A-3C; Figure S5A; Table S2A). The clustering was not biased by individual samples or genders (Figures S5B-S5D). UMAPs from the scRNA-seq data again corroborated microglial transcriptional reprogramming in rNLS8 mice (Figure 3D). Additionally, the mitotic microglia cluster was observed in the rNLS8 group (Figures 3B-3D; Figures S5E-S5F), indicating proliferative microglia at this stage of disease progression.

Figure 3. A specific microglia subpopulation identified in rNLS8 mice: rod-shaped microglia. See also Figure S5 and Table S2.

(A) Schematic workflow for single cell RNA sequencing (scRNA-seq). Mice were euthanized at 3 weeks post-DOX diet removal. Four half brains from control or rNLS8 mice were processed for scRNA-seq. Data are from one independent experiment. (B) UMAP visualization of all cells identified by scRNAseq. Cells are color-coded by their identities (number of cells = 101,162 from 8 samples). Pie charts show the cell proportions of each cluster. (C) Stacked violin plot of known marker genes for each identified cell type. (D) UMAP visualization of all cells identified by scRNA-seq. Pie charts show the cell proportions of each cluster. (E) Layer structure of the motor cortex of a control mouse brain is shown by NeuN and IBA1 expression. M1, supplementary motor area (M2), and lateral ventricle (LV) are separated by a dashed line. Scale bar, 300 μm. (F) IBA1 expression in M1, as indicated by the area in dotted white box in e. Scale bar, 100 μm. (G-H) Quantification of microglial density (G, n = 14 per group) and soma area (H, n = 9 per group). Data are from 3 independent experiments. (I) Representative images of microglia with different morphologies, including ramified microglia identified at baseline, and rod-shaped microglia, bushy microglia, and amoeboid microglia identified at 3 weeks post-DOX diet removal. Scale bar, 10 μm. Bushy microglia are characterized by an enlarged soma (1 to 2 times the diameter of the soma in ramified cells) with short and poorly ramified processes. Ameboid microglia have a larger cell body (more than twice the diameter of the soma in ramified cells) with either no processes or only a single process. Rod-shaped microglia are characterized by a slender, elongated cell body and few planar processes, which are distinctly polarized. (J) Percentage of microglia with different morphologies. Data are from 3 independent experiments. (K) Representative images of IBA1 staining in motor cortex tissue of individuals with ALS and age-matched controls. Scale bar, 100 μm. High-magnification images as indicated by the areas in the dotted black boxes are shown on the right. scale bar, 50 μm. (L-M) Quantification of microglia density (L) and rod-shaped microglia percentage (M) in the motor cortex of individuals with ALS (n = 31) and age-matched controls (n = 25). Data are from one independent experiment. (N) Correlation of rod-shaped microglia percentage change fold with cortical neuronal Ca signal area change fold. Statistical analysis, one-way ANOVA followed by Fisher’s post hoc test (G-H), or two-tailed unpaired Student’s t-test (L-M). Error bars, mean ± s.e.m. NS = not significant, *P < 0.05; **P < 0.01; *** P < 0.001; **** P < 0.0001.

To validate the observations from both spRNA-seq and scRNA-seq, we focused on microglia using IBA1 staining. In control mice, microglia were uniformly distributed throughout the motor cortical layers (Figures 3E-3F). However, in rNLS8 mice, we observed a significant increase in microglial density and soma size as the disease progressed, with a peak at three weeks post-DOX diet removal (Figures 3F-3H) but no significant sex differences (Figures S5G-S5H). These activated microglia exhibited a reactive phenotype with diverse morphologies, including amoeboid microglia, bushy microglia, and distinct rod-shaped microglia (Figure 3I-3J). Rod-shaped microglia are characterized by a slender, elongated cell body and few planar processes (Figure 3I). They were mainly distributed in layer 2/3 and layer 4 (Figure 3F). Notably, we also observed an increased microglial density and the abundance of rod-shaped microglia in the motor cortex of individuals with ALS compared to controls (Figures 3K-3M; Table S3A).

Our results showed that the proportion of rod-shaped microglia peaked at three weeks post-DOX diet removal, right after the peak of neuronal hyperactivity (Figure 3N). This temporal correlation suggests that rod-shaped microglia may be the primary responders to neuronal hyperactivity. To test this hypothesis, we used an inhibitory Gi-coupled DREADD (Designer Receptors Exclusively Activated by Designer Drugs), hM4Di, targeted to excitatory neurons in the motor cortex (Figure 4A). These neurons were virally co-expressing GCaMP6s to enable real-time neuronal activity recording by fiber photometry through an implanted optical fiber (Figure 4B-4D). We found that chronic administration of CNO led to an approximately 60% reduction in calcium signaling following DOX diet removal (Figure 4E). Consistently, there was decreased c-Fos expression in the Gi-DREADD group without any change in neuronal density, (Figures 4F-4H). Notably, we found that Gi-DREADD-mediated suppression of excitatory neuronal activity significantly reduced the percentage of rod-shaped microglia in rNLS8 mice (Figure 4I-4J). These findings provide direct evidence that neuronal activity is necessary for the formation of rod-shaped microglia.

Figure 4. Inhibition of neuronal activity reduced the formation of rod-shaped microglia.

(A) GCaMP6s and hM4D(Gi)-mCherry were co-expressed in M1 of 2-month-old rNLS8 mice via stereotactic injection. EMPTY-mCherry was used as a control. Schematic illustration of stereotactic viral injection targeting the M1 region, and in vivo fiber photometry recording. (B) Experimental timeline. Following baseline imaging, mice were switched to a DOX-free diet and treated with CNO. (C-D) Representative low-magnification (C, Scale bar, 200 μm) and high-magnification (D, Scale bar, 20 μm) image of neurons (NeuN, cyan), expressing hM4D(Gi)-mCherry in M1. (E) Quantification of ΔF/F·s signal area from in vivo calcium activity of neurons in the M1 of rNLS8 mice (n = 7 per group). Data are from one independent experiment. (F) Representative images showing NeuN (cyan) and c-Fos (magenta) expression in the M1 of rNLS8 mice at 3 weeks post-DOX removal and CNO treatment. Scale bar, 20 μm. (G) Quantification of NeuN (n = 6-8 per group). Data are from 2 independent experiments. (H) Quantification of c-Fos positive neurons (n = 6-8 per group). Data are from 2 independent experiments. (I) Representative images of IBA1-positive microglia (green) in the M1 of rNLS8 mice injected with Gi or control virus at 3 weeks post-DOX removal and CNO treatment. Scale bar, 50 μm. (J) Quantification of rod-shaped microglial percentage (n = 6-8 per group). Data are from 2 independent experiments. Statistical analysis, two-way ANOVA followed by Sidak’s post hoc test (E) and two-tailed unpaired Student’s t-test (G, H and J). Error bars, mean ± s.e.m. NS = not significant, *P < 0.05; **P < 0.01; *** P < 0.001; **** P < 0.0001.

Rod-shaped microglia exhibit distinct transcriptional and functional profiles

To explore the transcriptomic signature of microglia in rNLS8 mice, we re-clustered microglia across all groups, identifying 7 distinct microglia subpopulations from a total of 39,645 cells (Figure 5A). The clustering was unbiased by individual samples or genders (Figures S6A-S6D), and were characterized using lineage-specific markers (Figure S6E; Table S2B). Clusters 0, 1, and 3 (MG0, MG1, and MG3), predominantly found in the control group, were identified as 'homeostatic microglia', with high expression genes such as P2ry12, Tmem119, Cx3cr1, and Hexb (Figures 5B-5C; Figure S6E; Table S2B). Cluster MG6, with a slight increase in rNLS8 mice, was enriched with genes related to cell cycle (Figures 5B-5C; Figure S6E; Table S2B). Clusters 2, 4, and 5 (MG2, MG4, and MG5) were significantly enriched in the rNLS8 group and classified as 'sc-rNLS8-specific microglia clusters' (Figures 5B-5C). Notably, MG2 exhibited similarities with 'disease-associated microglia' (DAM) identified in AD mouse models ^18,19, characterized by the downregulation of homeostatic genes and upregulation of DAM marker genes (such as Apoe, Axl, Lpl, Spp1, Igf1), as well as the Lgals family genes and cytokine/chemokine genes (Figures 5B-5C, Figures S6E and S6F; Table S2B). Indeed, gene homology analysis revealed 85 of the top 100 genes and 380 of the top 500 genes in the microglia cluster shared with the DAM signature ¹⁸ (Figures S6G; Table S2C). MG4 resembled homeostatic clusters and exhibited very limited marker genes, representing a transitional state (Figure S6E; Table S2B). MG5, identified as 'interferon-responsive microglia', showed high expression of Ifit family genes (Figures 5B-5C; Figures S6E and S6F; Table S2B). To further integrate spRNA-seq with scRNA-seq data, we compared microglial DEGs from both datasets and identified 46 genes from the 'sp-rNLS8-specific cluster' that overlapped with the scRNA-seq microglial cluster, and 26 genes that overlapped specifically with the MG2 subcluster (Figures S6H; Table S2D).

Figure 5. Rod-shaped microglia exhibit a distinct transcriptional and functional profile. See also Figure S6 and Table S2.

(A) UMAP visualization of microglia subclusters identified by scRNA-seq. Cells are color-coded by their identities (number of cells = 39,645 from 8 samples). Pie charts show the cell proportions of each cluster. Data are from one independent experiment. (B) UMAP visualization of microglia subclusters in control and rNLS8 mice. Pie charts show the cell proportions of each cluster. (C) Quantification of cell proportions of each microglial subcluster. (D) Feature plot of Lgals3 expression in control and rNLS8 mice from scRNA-seq analysis. (E-F) Representative images (E) and quantification (F) of Galactin 3 (Gal3, red) positive microglia (IBA1, green) in M1 at 3 weeks post-DOX diet removal. scale bar, 50 μm. Data are from 3 independent experiments. (G) Volcano plot of DEGs in microglia subcluster 2 (MG2). Homeostatic microglia cluster (combination of MG0, MG1 and MG3 from the control group) served as baseline. Significance at adjusted P value ≤ 0.05 and ∣avg_log2FC∣ ≥ 1 (log2FC, log2-fold change). (H) Heatmap of top 100 upregulated and downregulated DEGs in the MG2 cluster compared with the homeostatic microglia cluster. (I) Dot plot of the selected GO terms in order of gene ratio in MG2. The size of each dot indicates the number of genes in the significant DEG list that are associated with each GO term, and the color of the dots corresponds to the adjusted P-values. Statistical analysis, two-tailed unpaired Student’s t-test (C). Error bars, mean ± s.e.m. NS = not significant, *P < 0.05; **P < 0.01; *** P < 0.001; **** P < 0.0001.

To determine which cluster rod-shaped microglia belonged to, we screened marker genes from 'sc-rNLS8-specific microglia clusters' using immunostaining. Our results showed that Galectin-3 (encoded by Lgals3) specifically labeled approximately 60% of rod-shaped microglia, but not other markers, like LPL, MHC II, CD11c or AXL (Figures 5D-5F; Figures S6I-S6P), suggesting that rod-shaped microglia are part MG2. Thus, we conducted a DEG analysis for MG2 clulster. Using the homeostatic clusters (MG0, MG1, and MG3 combined) as a control, we identified a total of 2,015 significant DEGs in MG2 (Figure 5G). Among them, the top 100 upregulated DEGs are involved in cell adhesion (Madcam1, Lgals3, Siglec1), phagocytosis (Spp1, Axl, Cybb), and cell remodeling (Baiap2l2, Actr3b, Gas2l3) (Figure 5H; Table S2E). GO analysis further revealed that the top enriched processes related to 'glial cell activation', 'regulation of cell morphology', 'cell-cell adhesion', and 'phagocytosis & endocytosis' (Figure 5I). Thus, microglia subcluster analysis from scRNA-seq extends spRNA-seq by revealing the distinct transcriptional and functional profiles of rod-shaped microglia in rNLS8 mice.

Rod-shaped microglia interact with neuronal dendrites

To evaluate the function of rod-shaped microglia, we performed co-immunostaining of IBA1 with markers for blood vessels (CD31), axons (ANKG), and dendrites (MAP2). Notably, we found that about 70% of interactions by rod-shaped microglia were with dendrites, rather than with blood vessels or axons (Figures 6A-6F). Additionally, we crossed rNLS8 mice with Thy1-YFP mice to specifically label layer 5 pyramidal neurons and observed that nearly 50% of these rod-shaped microglia are engaged with apical dendrites (Figures 6G-6J). This contrasted with only about 10% interaction observed in the ramified microglial soma of control mice (Figures 6G-6J). We identified three distinct interaction patterns: Type 1, where the microglia processes connect to both apical dendrites and spines; Type 2, involving microglial body contacting apical dendrites; and Type 3, where microglia completely encase apical dendrites (Figures 6H and 6I; Video S2). We further found that Galectin-3 specifically labeled over 60% of the rod-shaped microglia associated with dendrites (Figures 6K-6L).

Figure 6. Rod-shaped microglia interact with neuronal dendrites.

(A-B) Representative images of co-staining of microglia (IBA1, green) with axon initial segment (AIS) (ANKG, red) (A) and quantification (B) of their interaction in M1 of control and rNLS8 mice at 3 weeks post-DOX diet removal. Scale bar, 50 μm. (C-D) Representative images of co-staining of microglia (IBA1, green) with blood vessel (CD31, red) (C) and quantification (D) of their interaction in M1 at 3 weeks post-DOX diet removal. The interactions are indicated by arrowheads. Scale bar, 50 μm. (E-F) Representative images of co-staining of microglia (IBA1, green) with neuronal dendrites (MAP2, red) (E) and quantification (F) of their interaction in M1 of indicated groups at 3 weeks post-DOX diet removal. The interactions are indicated by arrowheads. Scale bar, 50 μm. (G) Representative images of YFP positive pyramidal neurons in M1 of thy1-YFP mice. Apical dendrites are indicated by the area in dotted white box. Scale bar, 100 μm. (H) Representative images of microglia (IBA1, green) interaction with apical dendrites (red) in M1 of control and rNLS8:Thy1-YFP mice at 3 weeks post-DOX diet removal. Scale bar, 20 μm. (I) Representative 3D-image of microglial (white) interaction with apical dendrites (red). Scale bar, 10 μm. (J) Quantification of ramified microglia (control) and rod-shaped microglia (rNLS8) interaction with apical dendrites in M1 at 3 weeks post-DOX diet removal (n = 10 per group). (K-L) Representative images (K) and quantification (L) of Gal3 (Gray) expression in dendrites (MAP2, red) associated rod-shaped microglia (IBA1, green) at 3 weeks post-DOX diet removal. Scale bar, 50 μm. (M-N) Representative images (M) and quantification (N, n = 8 per group) of CD68 (red) expression in ramified (control) and rod-shaped microglia (rNLS8) at 3 weeks post-DOX diet removal. Scale bar, 20 μm. (O) Spatially resolved expression of Cd68 across tissue sections from control and rNLS8 mice at 3 weeks post-DOX diet removal. (P) Representative 3D-image of microglial (IBA1, white) phagocytosis of excitatory synaptic materials (VLUT1/PSD95, red/green). Scale bar, 20 μm. Insets show images at higher magnification as indicated by the area in dotted yellow boxes. Scale bar, 5 μm. (Q) Quantification of ramified (control) and rod-shaped microglial (rNLS8) phagocytosis of PSD95 in M1 at 3 weeks post-DOX diet removal (n = 6 per group). (R) NicheNet’s ligand–target matrix denotes the regulatory potential between neuron-ligands and target receptors from microglia cluster. Statistical analysis, two-tailed unpaired Student’s t-test and data are from 2-4 independent experiments. (B, D, F, J, L, N and Q). Error bars, mean ± s.e.m. NS = not significant, *P < 0.05; **P < 0.01; *** P < 0.001; **** P < 0.0001.

We next explored whether the proximity of rod-shaped microglia to dendrites influences excitatory synaptic inputs. Consistent with the highlighted GO terms, upregulation of CD68 and downregulation of homeostatic marker of P2Y12 and TMEM119 confirmed phagocytic nature of rod-shaped microglia (Figures 6M-6O; Figures S6Q and S6R). Specifically, rod-shaped microglia contain abundant synaptic components, suggesting their function in synaptic remodeling (Figures 6P and 6Q; Video S3). We then conducted ligand-receptor interactome analyses of microglia and neuron clusters from scRNA-seq data using NicheNet ²⁰. The results showed that the highly expressed receptors on microglia were specifically associated with cell adhesion (Notch family, Integrin family, Nptn, Jam2/3, F11r) and phagocytosis (Trem2, Tlr2, Lrp1/5) (Figure 6R). Apoe from neurons is predicted to be a potential ligand for the phagocytosis receptor (Figure 6R; Figures S6S and S6T). The potential ligands for the cell adhesion receptors included Ncam1, Lrfn4, Mpdz and others (Figure 6R; Figures S6T). Altogether, our findings suggest that rod-shaped microglia have a close physical relationship to the dendritic compartment and could regulate neuronal activity by synaptic remodeling.

TREM2/DAP12 axis regulates the formation of rod-shaped microglia

Next, to identify the potential regulators mediating the transformation of rod-shaped microglia, we performed RNA velocity analysis for dynamic transcriptional transitions among microglial clusters, with a specific focus on MG2 and its adjacent counterparts. This technique uses the ratio of unspliced pre-mRNA to spliced mRNA to infer the dynamic state of gene expression, characterizing the direction and rate of changes in cellular transcriptional states ^21,22. Our results showed that RNA velocity vectors originated from homeostatic clusters (MG0, MG1, MG3, and MG4) and pointed toward MG5 and MG6, eventually converging on MG2 (Figure 7A). An increased latent time and pseudo-time, combined with the highest proportion of spliced RNA (74%) observed in MG2, confirmed that this cluster represents a terminal differentiation state (Figures 7B and 7C; Figures S7A and S7B). We further explored the critical genes driving these microglial transitions and found the most dynamically expressed genes related to transcriptional factors (e.g., Zfhx3, Atf3) and ribosome function (e.g., Rps21, Rpl13a, Rps3) (Figure 7D; Figures S7C and S7D; Table S2F). Notably, Tyrobp, encoding the DNAX-activating protein of 12 kDa (DAP12), was also among the highly dynamic genes (Figures 7D and 7E; Table S2F). As the major adapter protein for TREM2, DAP12 acts as the principal regulator mediating the conversion of homeostatic microglia into DAM ¹⁶. Both Tyrobp and Trem2 exhibited significant upregulation at later stages of pseudo-time, highlighting the crucial role in the microglial transition (Figures 7E and 7F). Consistently, qRT-PCR analysis confirmed a marked reduction in MG2-associated transcripts in the motor cortex of rNLS8:TREM2KO mice (Figure 7G). The morphological analyses confirmed that TREM2 deficiency attenuated the microglial response at three weeks following DOX diet removal (Figures 7H-7J). Moreover, TREM2 deficiency exerts a broad effect on the entire microglial population in rNLS8 mice, including a significant reduction in the proportion of rod-shaped microglia (Figure 7K).

Figure 7. TREM2/DAP12 axis regulates the transition of rod-shaped microglia and neuronal hyperactivity. See also Figure S7, Table S2 and Table S3.

(A) RNA velocity derived from the dynamical model for microglia subclusters is visualized as streamlines in a UMAP-based embedding with scVelo. The dynamic model accurately delineates the trajectory of mRNA transcription states among microglia clusters. (B-C) The scVelo’s latent time (B) and pseudo time (C) are based on the transcriptional dynamics and visualized in UMAP plots. (D) Heatmap of the 100 selected highly dynamic genes along latent time shows a clear cascade of transcription. (E-F) The expression plots (top) and referred expression levels along latent time (bottom) of Tyrobp (E) and Trem2 (F). (G) qRT-PCR analysis of microglial homeostatic marker genes and the top upregulated DEGs in the MG2 cluster across the indicated groups. P values for each gene are provided in the Table S3B. (H) Representative images of microglia (IBA1, green) in M1 at 3 weeks post-DOX diet removal. Scale bar, 100 μm. Inset shows microglia at higher magnification as indicated by the area in the dotted white box. Scale bar, 20 μm. (I-J) Quantification of microglial density (I) and soma size (J) (n =14 per group). Data are from 3 independent experiments. (K) Percentage of microglia with different morphologies at 3 weeks post-DOX diet removal. (L) Representative images of layer 2/3 neuronal calcium activity across experimental phases in rNLS8:TREM2 KO mice during baseline and disease progression. Scale bars, 100 μm. Insets show neuronal soma at higher magnification as indicated by the area in dotted yellow boxes. Scale bar, 10 μm. Data are from 2 independent experiments. (M) ΔF/F calcium traces from the soma of a representative healthy neuron at baseline and neurons with Ca overload with the threshold line (black dotted line). Duration of single transient and the maximum amplitude are indicated by red arrows. (N) Quantification of the number of Ca-overloaded neurons. (O) Quantification of ΔF/F signal area (ratio) of neuronal Ca activity. (P) Average latency to fall during rotarod tests (n = 25 per group). Data are from 3 independent experiments. (Q) Kaplan–Meier survival curves showing the percentage of mice alive up to 30 days post-DOX diet removal (n = 25 per group). Data are from 3 independent experiments. Statistical analysis, one-way ANOVA followed by Tukey’s post hoc test (G, I and J), two-tailed paired t-test within each genotype (N and O), unpaired t-test for comparing DOX off D16 between genotypes (N and O), or two-way ANOVA followed by Sidak’s post hoc test (P). Survival curves were analyzed using a log-rank (Mantel–Cox) test (Q). Error bars, mean ± s.e.m. NS = not significant, *P < 0.05; **P < 0.01; *** P < 0.001; **** P < 0.0001.

We next conducted longitudinal calcium imaging experiments and observed a significant increase in calcium overload at 16 days post-DOX diet removal in rNLS8:TREM2 KO mice compared with rNLS8 mice (Figures 7L-7N; Video S4). The analysis of the ΔF/F calcium traces indicated that neurons exhibited prolonged duration and elevated amplitude of calcium events (Figure 7M; Figures S7E-S7F). Consequently, the area of the calcium signal showed a significant increase in rNLS8:TREM2 KO mice compared with rNLS8 mice (Figure 7O). In addition, more severe motor deficits and a lower survival rate in the rNLS8:TREM2 KO mice compared to rNLS8 mice, suggesting a neuroprotective effect of the TREM2-mediated microglial response during disease progression (Figures 7P and 7Q). No significant sex difference was observed in the rotarod test (Figure S7G). Altogether, these data indicate that TREM2/DAP12 axis plays a crucial role in mediating the transformation of rod-shaped microglia.

DISCUSSION

Microglial activation is one of the hallmarks of ALS pathology ^16,23-26. However, the exact role of microglia in ALS is not fully understood. Previous studies suggest that microglia exhibit a dual function in ALS, encompassing both neuroprotective and neurotoxic effects ^16,27-29. This may be attributed to differences in disease stages and models used between studies. Growing evidence indicates that microglia play a crucial role in monitoring and regulating neuronal activity ^9,10,30. Yet, the response of microglia to neuronal hyperexcitability in ALS is understudied. Using silicon probes and chronic in vivo two-photon calcium imaging, we recapitulated neuronal hyperactivity in the motor cortex of an ALS mouse model in the early stages of disease. Notably, single-cell and spatial RNA-seq identified microglia as the primary responders to cortical hyperactivity. We discovered a distinct subpopulation of rod-shaped microglia whose emergence temporally coincides with the peak of cortical hyperactivity. Using chemogenetic approach, we demonstrated that neuronal activity is necessary for the formation of rod-shaped microglia. Further, these rod-shaped microglia may play a key neuroprotective role by engaging with dendrites and synaptic remodeling. TREM2 deficiency led to a marked reduction in rod-shaped microglia, accompanied by increased neuronal calcium overload, exacerbated motor deficits, and decreased survival in rNLS8 mice.

Cortical hyperexcitability in the early stages of ALS is believed to be accompanied by progressive degeneration of the motor cortex. Our study provides a characterization of motor cortical hyperexcitability in the rNLS8 mouse model. While previous studies have used electrophysiological and neuroimaging techniques to explore changes in motor neuron excitability in ALS mouse models ^{3,31,32 33}, none have conducted longitudinal studies to monitor neuronal activity throughout the disease progression with such detailed spatial and temporal resolution. Our findings reveal that early hyperactivity is most closely related to changes in the firing properties of pyramidal neurons, rather than reductions of inhibitory neuron output at this stage.

One of the major strengths of our study is the detailed profiling of transcriptional changes in rNLS8 mice through the integration of spRNA-seq and scRNA-seq. Specifically, our study revealed distinct rod-shaped microglia during the short time window of initial disease progression. Rod-shaped microglia were reported in pathological conditions such as neurodegenerative diseases. For example, pathology study from individuals with AD revealed rod-shaped microglia in close proximity to senile plaques ³⁴ and in hippocampal CA1 regions ³⁵. In individuals with PD, rod-shaped microglia have been observed in the substantia nigra area and were close proximity to degenerating dopaminergic neurons ³⁶. In individuals with HD, rod-shaped microglia were aligned along the dendrites and soma of pyramidal neurons ³⁷. However, rod-shaped microglia have not been reported in the context of ALS/FTD, and the formation and the function of rod-shaped microglia in neurodegeneration are largely unknown. Here we identified this distinct type of microglia in the motor cortex of rNLS8 mice and individuals with ALS, suggesting their clinical relevance in ALS pathophysiology. Considering that rod-shaped microglia are also present in various neurodegenerative and neurological disorders, additional roles beyond responding to hyperactivity may be revealed with further investigation. Thus, we chose to use a morphology-based term that provides a descriptive reference point until their full spectrum of functions is appreciated. Notably, we showed that 60% of rod-shaped microglia could be identified by Galectin-3, consistent with prior findings from individuals with ALS and mouse models with increased Galectin-3 in activated microglia ^38,39, as well as in plasma and cerebrospinal fluid ^40-43. The function of microglial Galectine-3 in AD was reported to act as an endogenous ligand for TREM2 ⁴⁴. In addition, Galectine-3 might play a potential role in microglial activation and phagocytosis in ALS ^38,45. Thus, further research is needed to fully understand whether and how Galectin-3 regulates the formation and function of rod-shaped microglia in the context of ALS.

We were intrigued to observe that the majority of rod-shaped microglia align along neuronal dendrites in the rNLS8 mice. This is consistent with previous studies indicating that neuronal processes guide rod-shaped microglia positioning ^46,47. We propose that rod-shaped microglia sense and regulate neuronal function via direct interaction with neuronal dendrites, considering that microglia frequently engage in neuronal circuits ^11,48-50. Established mechanisms through which microglia sense neuronal activity include ATP/purinergic signaling ^11,48-50, microglia Ca²⁺ signaling ⁵¹, fractalkine signaling ⁵², neuromodulators ^53,54, complement signals ⁵⁵, and others. Using chemogenetic manipulation of neuronal activity in rNLS8 mice, we further showed that neuronal hyperactivity is likely the driver for the formation of rod-shaped microglia. Future studies will identify the signaling pathways that enable rod-shaped microglia formation in response to neuronal hyperactivity. Microglia-mediated synaptic pruning, a well-documented phenomenon across various neurodegenerative disease models, involves mechanisms such as completement ⁵⁶, CX3CR1 ⁵⁷, TREM2 ^58,59, and others. In the context of ALS, it has been reported that C9orf72-depleted microglia exhibit enhanced cortical synaptic pruning ⁶⁰. Here we found that rod-shaped microglia have a high capacity for phagocytosing excitatory synaptic materials, suggesting their potential role in regulating neuronal activity in rNLS8 mouse model. Our receptor-ligand–based communication analysis uncovered putative molecules involved cell adhesion and phagocytosis that may facilitate their interaction. Future research is necessary to validate whether these receptor-ligand pairs madidate the interaction between rod-shaped microglia and neuronal dendrite.

In this study, we found that the TREM2/DAP12 axis plays a role in the emergence of rod-shaped microglia, as determined by trajectory analysis. The role of TREM2 in modulating microglial status in AD is well-documented, facilitating the transition to an activated state and the induction of lipid metabolism and phagocytic pathways ¹⁸. Our previous studies found that TREM2 mediates microglial phagocytosis of TDP-43 that is important for disease recovery¹⁶. Here we observed that TREM2 deficiency reduced the formation of rod-shaped microglia while increased the neuronal activity. Known TREM2 ligands, such as phosphatidylserine (PS), TDP-43, and damaged lipids ^16,61,62, may be produced or released by hyperactive neurons and serve as cues for TREM2. Since Galectin-3 is highly expressed in rod-shaped microglia and could also serve as a ligand for TREM2 ⁴⁴, an autocrine mechanism may be involved. Future research will ascertain how TREM2 detects signals from surrounding environment to regulate the transformation and function of rod-shaped microglia. In sum, our study reported neuronal hyperactivity and associated spatiotemporal formation of rod-shaped microglia in rNLS8 mice. These findings suggest that rod-shaped microglia may exert neuroprotective function through elimination of the excitatory synaptic inputs, highlighting the therapeutic potential of microglia in early ALS treatment.

Limitations of the study

Although we showed rod-shaped microglia remodel excitatory synapses in rNLS8 mice, our study lacks the direct evidence that the interactions dampen neuronal activity. The loss of rod-shaped microglia associated with heightened cortical excitability after TREM2 deficiency suggests their protective role in attenuation of cortical excitability. Thus, future studies employing chronic in vivo imaging would be a powerful way to directly monitor rod-shaped microglia formation, their interactions with neurons, and the functional consequences of these interactions. Another limitation is that TREM2 deficiency broadly affected multiple microglial subtypes, making it difficult to determine whether the worsened phenotype is solely attributable to the loss of rod-shaped microglia; identifying specific regulators of this microglial subset will be important for future work. Additionally, our current mouse model overexpresses hTDP-43 under the control of the NEFH promoter, leading to expression in both excitatory and inhibitory neurons ¹⁵. This makes it challenging to discern whether the observed alterations in inhibitory neuronal activity are secondary to changes in excitatory neurons or are a direct consequence of hTDP-43–induced stress within inhibitory neurons. Further research will be needed to clarify these mechanisms and to elucidate the progression of motor circuit alterations in later stages of ALS.

STAR METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Mice.

The rNLS8 mouse line was crossed by the NEFH-tTA line 8 (JAX, No. 025397) and tetO-hTDP-43-ΔNLS line 4 (JAX, No. 014650). Mice hemizygous for both NEFH-tTA and tetO-hTDP-43-ΔNLS were used as diseased mice. Mice hemizygous for only NEFH-tTA or tetO-hTDP-43-ΔNLS were used as control mice. Thy1-YFP-H (JAX, No. 003782) and CX₃CR-1^GFP knock-in/knock-out (JAX, No. 005582) mice were purchased from Jackson Labs. The TREM2 knockout (KO) mouse strain was generously provided by Dr. Marco Colonna at the Washington University School of Medicine, St. Louis. Thy1-YFP-H and TREM2 KO mice were crossed with rNLS8 mice. rNLS8 mice were maintained on Dox chow (200 mg/kg, Bio-Serv #3888) to prevent hTDP-43ΔNLS expression. Experimental mice were switched to Dox-free chow to induce hTDP-43ΔNLS expression. Mouse lines were housed with littermates with free access to food and water on a 12-hour light/dark cycle, and the room temperature is 20-22°C with 55% humidity. All animal procedures, including husbandry, were performed under the guidelines set forth by Mayo Clinic Institutional Animal Care and Use Committee. Both sexes were used in the experiments.

Human samples.

Post-mortem brain samples were dissected from the frozen brains of 31 ALS cases (age 61.32 ± 9.70 years, mean ± s.d.) and 25 controls (age 59.09 ± 12.73 years) from the Brain Bank for Neurodegenerative Disorders at Mayo Clinic Jacksonville (Table S3A). Brainstem-type Lewy body disease (BLBD) cases with minimal p62 immunoreactivity were identified as controls for motor cortex pathology. The study was approved by Mayo Biospecimen Committee. All donors or their next of kin provided informed consent. For clinic pathological studies, cases were included only if they had good quality medical documentation and there was diagnostic concurrence of at least minimum of two neurologists.

METHOD DETAILS

IN VIVO MULTICHANNEL SILICON PROBES RECORDING

Silicon probe implantation surgery.

All recordings were performed using A1x32-Poly2-3mm-50 s-177-CM32 silicon probes (177 μm² site surface area, 2-column honeycomb site geometry with 50 μm center-to-center span, 15 μm shank thickness) with a CM32 connector (NeuroNexus Technologies). Probe implantation surgery is performed based on the protocol provided by the NeuroNexus website. Specifically, animals were placed on a stereotaxic frame and under isoflurane anesthesia (4% induction, 1.5–2.5% maintenance). A circular craniotomy (1 mm diameter) was drilled to accommodate the probe shanks in the mouse primary motor cortex at the following stereotactic coordinates: AP, + 1.4 mm from bregma; ML, +1.5 mm; DV, − 0.8 mm from the brain surface. Additionally, three small holes (1 mm diameter) were then drilled for stainless steel screws: two over both sensory cortices and one over the contralateral site of the implant, which were later used as a ground reference. Screws (4 mm long, 0.86 mm diameter) were carefully advanced into the drilled holes. After removal of the dura in the craniotomy, probes were lowered into the surface of the cortex and inserted using an automatic or manual micromanipulator to the desired depth (0.8 mm) at a rate of 1 mm/min. The ground and reference wires were wrapped tightly around the bone screws before the probe and wires were secured using dental cement and a secure head cap was created. Mice were kept warm and accelerated recovery from anesthesia, then single-housed to protect the electrode.

Data collection and analysis.

Ten days following the procedure, mice were acclimated to the recording environment by placing them in a recording box with their electrodes connected to the recording adaptor for one hour daily. This habituation period lasted approximately 3 days, during which the mice were allowed to move freely within the recording box. During the recording sessions, mice were given 30 minutes to acclimate to the environment with their electrodes connected to the adapter before the recording commenced. Recordings were carried out weekly for a duration of 30 minutes each, covering both the baseline and disease progression periods over four weeks. The head stage was connected to a Zeus recording system (Zeus, Bio-Signal Technologies, McKinney, TX, USA) to capture peak potentials, using a sampling frequency of 3 kHz. Spike signals were band-pass filtered online at 300-7,000 Hz. Spike waveforms and timestamps were saved in Plexon data (*.plx) files. Spikes were sorted using the valley seek method with Offline Sorter software (Plexon), and any indistinct waveforms were manually removed in a three-dimensional feature space. The sorted units were analyzed with Neuro Explorer 5.0 to generate graphs and charts detailing the firing rates and timestamps of the neurons, which were then exported for further analysis in Microsoft Excel.

Data analysis was performed as previously described ^63-65. Units were classified as either excitatory or inhibitory neurons based on their spike width (trough-to-peak duration): units with a spike width greater than 0.4 ms were classified as excitatory neurons, while those with a spike width less than 0.4 ms were classified as inhibitory neurons. Additionally, the mean interspike interval (ISI) was used to further distinguish between excitatory and inhibitory neurons. Inhibitory neurons exhibited a shorter ISI compared to excitatory neurons, with putative interneurons having a mean ISI of 2 ms or less. The firing frequency of neuronal activity was analyzed and normalized to the baseline.

EX VIVO ELECTROPHYSIOLOGICAL RECORDING

Brain slices preparation.

Brain slices containing the motor cortex were prepared for electrophysiological recordings as previously described ⁶⁶. Mice aged P70 were deeply anesthetized using isoflurane. Following decapitation, the brain was quickly removed and submerged in an ice-cold slice solution, which was oxygenated with 95% O₂ and 5% CO₂. This solution contained the following components (in mM): 2.41 KCl, 1.22 NaH₂PO₄·2H₂O, 25 NaHCO₃, 0.4 ascorbic acid, 2 sodium pyruvate, 0.5 CaCl2·2H2O, 3.49 MgCl2·6H2O, and 240 sucrose; it was adjusted to a pH of 7.4 ± 0.5 using HCl. Coronal slices of 300 μm thickness were sectioned in the cold slice solution using a Leica 2000 vibratome. These slices were then allowed to recover in an oxygenated slice solution at 34°C for 13 minutes, followed by maintenance in an incubation chamber filled with oxygenated artificial cerebrospinal fluid (ACSF) at room temperature for 30 minutes before the recordings. The ACSF composition was as follows (in mM): 117 NaCl, 3.6 KCl, 1.2 NaH₂PO₄·2H₂O, 25 NaHCO₃, 0.4 ascorbic acid, 2 sodium pyruvate, 2.5 CaCl₂·2H₂O, 1.2 MgCl₂·6H₂O, and 11 glucose, adjusted to a pH of 7.4 ± 0.5 with HCl. All chemicals were sourced from Sigma.

Recording.

Whole-cell recordings were performed at room temperature in the oxygenated ACSF solution. The brain slices were observed under a fixed upright microscope (Scientifica) equipped with a water immersion lens (Olympus, 40x/0.8 W). Patch pipettes were fashioned from borosilicate glass capillary tubes using a horizontal pipette puller (P-97, Sutter Instruments). These pipettes were filled with a potassium-based internal solution for recording active potentials in current-clamp modes, with the neuronal membrane potential held at −70 mV. The composition of the internal solution was (in mM): 135 potassium gluconate, 0.5 CaCl₂·2H₂O, 2 MgCl₂·6H₂O, 5 KCl, 5 EGTA, and 5 HEPES; the pH was adjusted to 7.3 ± 0.5 using KOH, achieving an osmolarity of 290-300 mOsm. The resistance of the patch pipettes was between 5-7 MΩ.

Data collection and analysis.

Data collection was facilitated by a Multiclamp 700B amplifier (Molecular Devices), with the signal filtered at 2 kHz and digitized at 4 kHz using a Digidata 1440A converter (Molecular Devices). Analysis of the recorded data was conducted using Clampfit 10.3 (Molecular Devices).

CHRONIC IN VIVO TWO-PHOTON IMAGING

Stereotaxic AAV Delivery and Cranial Window Surgery.

Adult rNLS8 or control mice (2 months old) were anesthetized with isoflurane (4% for induction, 1.5-2.5% for maintenance). For in vivo calcium imaging, 250 nL of pENN.AAV9.CaMKII.GCaMP6s.WPRE.SV40 (1.0 x 10¹² vg/mL; Addgene #107790) was stereotaxically injected into the right primary motor cortex to target layer 2/3 neurons. Injections were performed at the following coordinates relative to bregma: anterior-posterior (AP) +1.4 mm, medial-lateral (ML) +1.5 mm, and dorsal-ventral (DV) −0.3 mm. Microinjections were delivered at a rate of 35 nL/min using an automated stereotaxic injector (Model UMC4, World Precision Instruments). The microsyringe was held in place for 10 minutes following injection to ensure proper diffusion. To enable long-term monitoring of neuronal morphology, 300 nL of a 1:1 mixture of pENN.AAV.CamKII 0.4.Cre.SV40 (7 × 10⁹ vg/mL; Addgene #105558-AAV1) and pAAV-FLEX-tdTomato (1 × 10¹² vg/mL; Addgene #28306-AAV9) was injected into the same site to selectively label pyramidal neurons (AP +1.4 mm; ML +1.5 mm; DV −0.5 mm).

Following viral delivery, a 4-mm diameter craniotomy was performed over the injection site using a high-speed dental drill. A 4-mm circular glass coverslip (Warner Instruments) was then positioned over the craniotomy and sealed in place. A four-point headbar (NeuroTar) was affixed above the cranial window using dental cement. To minimize post-operative discomfort, mice were provided ibuprofen (0.2 mg/mL) in their drinking water for 72 hours before and after surgery.

Data collection.

Two-photon imaging in awake animals was performed using a multiphoton microscope equipped with galvanometer scanning mirrors (Scientifica). For GCaMP6s fluorescence, excitation was achieved using a Mai-Tai DeepSee laser (Spectra-Physics) tuned to 920 nm, with the laser power kept below 55 mW for imaging layer 2/3 neurons. For tdTomato-labeled neurons, the excitation wavelength was adjusted to 950 nm. Emitted signals were filtered using a 520/15 nm bandpass filter for GCaMP6s and a 620/60 nm bandpass filter for tdTomato (Chroma Technology). Images were acquired at a frame rate of 1 Hz, with a resolution of 512 × 512 pixels and a field of view of 450 × 450 μm, using a 16× water-immersion objective (Nikon, NA 0.8). Prior to imaging, mice were habituated to head fixation on an air-lifted mobile platform (NeuroTar) for 30 minutes daily over the course of 7 consecutive days. Chronic imaging sessions were initiated 3–4 weeks post-surgery in mice that exhibited a clear cranial window. Before each session, mice were allowed a 10-minute acclimation period under head restraint. For each time point, a 15-minute imaging video was acquired. Mouse locomotion during calcium imaging was simultaneously recorded using the NeuroTar Mobile HomeCage magnetic tracking system.

Calcium image processing and analysis.

Time series data of neuronal calcium activity were subjected to motion correction using the TurboReg plugin in ImageJ. Subsequently, an average intensity image was generated for the selection of Regions of Interest (ROIs). For identifying neuronal somata, visually discernible cell bodies were manually selected as ROIs using ImageJ. Once the ROIs were established, the multi-measure tool was employed to acquire mean intensity values (F) for each ROI. The relative percentage change in fluorescence of the calcium signals was calculated using the formula ΔF/F = (F – F0)/F0. Here, the baseline F0 was defined as the lower 25th percentile value of the initial 200-ms period of the recording. In instances of heightened activity, the baseline was alternatively determined by identifying the lower 25th percentile value in a user-selected, 200-frame period exhibiting minimal wave activity. A calcium transient was considered to have occurred when the ΔF/F ratio exceeded a threshold threefold greater than the standard deviation of the baseline. The signal area was quantified as the sum of all ΔF/F values surpassing this threshold over the entire duration of the recording. The maximum amplitude was calculated as the highest peak value among all the transients.

FIBER PHOTOMETRY RECORDING

Stereotaxic AAV delivery and optical fiber implantation.

For the Gi-DREADD manipulation group, 800 nL of a 1:1 mixture of AAV2/9-CaMKIIα-GCaMP6s (titer: 1.18 × 10¹³ vg/mL; BrainVTA, Cat#: PT-0110) and AAV2/9-CaMKIIα-hM4D(Gi)-mCherry (titer: 5.32 × 10¹² vg/mL; BrainVTA, Cat#: PT-0017) was stereotaxically injected into the unilateral motor cortex (coordinates: AP = +1.4 mm; ML = +1.5 mm; DV = −0.8 mm) at a rate of 35 nL/min. For the control group, the same volume (800 nL) of a 1:1 mixture of AAV2/9-CaMKIIα-GCaMP6s and AAV2/9-CaMKIIα-mCherry (titer: 4.5 × 10¹² vg/mL; BrainVTA, Cat#: PT-0108) was injected into the same brain region. Following viral delivery, optical fibers (core: 100/0.22; length: 1 m; connector: FC-1.25 mm white ferrule; Inper) were implanted unilaterally in adult mice (7-8 weeks old), positioned 0.2 mm above the injection site. Dental cement was applied to secure the fiber and seal the exposed skull. Mice were allowed to recover for at least 2–3 weeks before the initiation of behavioral or recording experiments.

In vivo fiber photometry recording and data analysis.

Calcium signals from layer V of the motor cortex were recorded using a fiber photometry system (Inper Ltd., China). Neurons expressing GCaMP6s were stimulated with a 470 nm LED (laser intensity adjusted to 40 μW at the fiber tip) to detect calcium-dependent fluorescence signals, while a 410 nm LED (20 μW at the fiber tip) was used to capture calcium-independent signals (e.g., motion artifacts). Fluorescence signals were acquired at a sampling rate of 20 Hz with an exposure time of 10 ms. Both the Gi-DREADD and control groups underwent baseline recordings three weeks after surgery, followed by treatment with Dox-free chow and CNO-containing water (0.25 mg/mL). Calcium activity was recorded weekly during the 1st to 4th weeks of Dox-off treatment. Each recording session lasted 30 minutes, preceded by a 30-minute acclimation period in the recording cage to allow for environmental adaptation. Data were analyzed using the Inper plot (Inper Ltd., China). To minimize autofluorescence of the optical fiber, the recording fiber was photobleached using a high-power LED before recording. The fluorescence signal obtained after 410 nm light stimulation was used to correct for motion artifacts and background autofluorescence. Subtract the fitted 410 nm signal from the 470 nm signal to obtain the motion- and bleaching-corrected fluorescence signal. The relative change in calcium signal fluorescence was calculated using the formula ΔF/F = (F – F0)/F0, where F0 was defined as the 25th percentile of the baseline recording.

SPATIAL RNA SEQUENCING

10X Visium preparation and sequencing.

Mice were anesthetized with isoflurane (5% in O₂) and then intracardially perfused with 40 ml cold PBS, followed by 40 ml cold 4% paraformaldehyde (PFA). Brains were removed and post-fixed in 4% PFA for an additional 6-8 hours or overnight in RT. Samples were then raised by 1X PBS and then processed by the following steps using the automatic tissue processing machine: 70% Alc (1 hr), 80% Alc (1 hr), 95% Alc (30 min), 95% Alc (30 min), 95% Alc (30 min), 100% Alc (30 min), 100% Alc (30 min), 100% Alc (45 min), Xylene (60 min), Xylene (60 min), Paraffin (45 min 60° C), Paraffin (45 min 60° C), Paraffin (60 min 60° C), Paraffin (60 min 60° C). The tissues were then embedded in paraffin to create a paraffin block. The paraffin blocks were then cut using a microtome to generate thin sections of tissue for RNA quality assessment and 10X Visium preparation. After RNA quality assessment (DV200 > 50%), the FFPE tissue block was then sectioned by a microtome at 5μm to generate appropriately sized sections for Visium slides (app. Bregma 1.54 mm; Allen brain reference atlas coronal section 18). The slides were then placed in a slide drying rack and incubate for 3 h in an oven at 42°C (10X Genomics, CG000408). After overnight drying at room temperature, the slices were proceeded to deparaffinization, decrosslinking and immunofluorescence staining protocols according to the manufacturer’s protocol with recommended reagents (10X Genomics, 1000339 and 1000251; CG000410). After tissue imaging, the slides were proceeded immediately to Visium Spatial Gene Expression based on User Guide (10X Genomics, CG000407), including probe hybridization, probe ligation, probe release & extension. Library generation commenced with the amplification of eluted probes using real-time qPCR to ascertain the optimal number of amplification cycles. The qPCR protocol commenced with an initial denaturation step lasting 3 minutes at 98°C, succeeded by 25 cycles, each comprising a 5-second duration at 98°C and a 30-second duration at 63°C. The requisite number of PCR cycles for library amplification was established based on the qPCR outcomes, employing 16-19 cycles for the amplification of the entire set of eluted probes, utilizing a 10X dual index kit. Subsequently, the amplified libraries were purified with SPRI select beads (Beckman Coulter). Quantification of the final library was conducted using a TapeStation 4200 D1000 Screen Tape (Agilent) and Qubit (Invitrogen). Sequencing was performed employing paired-end reads of 101 bp, utilizing either the NextSeq2000 P2 or NovaSeq6000 SP.

spRNA-seq quality control, integration, clustering, and differential expression analysis.

Quality control was performed on each sample to remove low-quality spots and lowly expressed genes. Spots with low total UMI counts, low numbers of expressed genes, high mitochondria concentration, or high hemoglobin concentration were removed. Genes detected in fewer than 10 spots were also removed. The data were then normalized using SCTransform v2 and integrated using the Harmony algorithm ⁶⁷. Gene-expression-based clusters were generated for each sample, with dimension reduction and clustering implemented using the Seurat R package ⁶⁸. These clusters were then annotated based on pathology review of the corresponding slides within Loupe Browser. Spots that passed the initial quality control but were found to be outside the tissue area during pathology review were removed. Differential gene expression analysis was performed to identify differentially expressed genes between each cluster and to identify differences in the disease phenotype within each cluster. The FindMarkers function from the Seurat R package was used to perform the DESeq2 test. Genes with an absolute value of log2(fold change) greater than 0.5 and a Bonferroni-adjusted p-value less than 0.05 were considered significant. Gene Ontology (GO) analysis was performed with the clusterProfiler R package ⁶⁹. Upregulated pathways were determined by analyzing sets of genes significantly upregulated in the disease samples, and additional pathways were determined by analyzing the full set of significantly differentially expressed genes. The Benjamini-Hochberg method was used to control the false discovery rate.

SINGLE CELL RNA SEQUENCING

Single-cell preparation and sequencing.

Single cell preparation was performed as previously described ⁷⁰. Mice were transcardially perfused with cold PBS and cortical regions were quickly dissected and enzymatically digested using the Neural Tissue Dissociation Kit P (Miltenyi Biotec, 130-092-628) in the gentleMACS Octo Dissociator with heaters (Miltenyi Biotec) using program 37C_NTDK_1. The myelin was then removed by magnetic bead separation using Myelin Removal Beads II (Miltenyi Biotec, 130-096-733) followed by the red blood cell lysis (Miltenyi Biotec ,130-094-183). Cell suspension was then followed by dead cell removal (Miltenyi Biotec, 130-090-101;). The final cell pellet was suspended in Ca/Mg-free PBS with 0.5% BSA, and immediately submitted to the Genome Analysis Core for Single Cell partitioning. The cells were counted and measured for viability using the Vi-Cell XR Cell Viability Analyzer (Beckman-Coulter). The barcoded Gel Beads were thawed from −80 °C and the cDNA master mix was prepared according to the manufacture’s instruction for Chromium Next GEM Single Cell 3’ Kit v3.1 (10x Genomics). Based on the desired number of cells to be captured for each sample, a volume of live cells was mixed with the cDNA master mix. A per sample concentration of 500,000 cells per milliliter or better is required for the standard targeted cell recovery of up to 10,000 cells. The stock concentration requirements would not change for higher cell recovery numbers. The cell suspension and master mix, thawed Gel Beads and partitioning oil were added to a Chromium Next GEM G chip. The filled chip was loaded into the Chromium Controller, where each sample was processed and the individual cells within the sample were partitioned into uniquely labeled GEMs (Gel Beads-In-Emulsion). The GEMs were collected from the chip and taken to the bench for reverse transcription, GEM dissolution, and cDNA clean-up. The resulting cDNA contains a pool of uniquely barcoded molecules. A portion of the cleaned and measured pooled cDNA continues to library construction, where standard Illumina sequencing primers and a specific sample index (Dual Index Kit TT-Set A; 10x Genomics) were added to each cDNA pool, creating gene expression libraries. All cDNA pools and resulting libraries are measured using Qubit High Sensitivity assays (Thermo Fisher Scientific) and Agilent Bioanalyzer High Sensitivity chips (Agilent). Libraries are sequenced at 50,000 fragment reads per cell following Illumina’s standard protocol using the Illumina NovaSeq^™ 6000 S4 flow cell. S4 flow cells are be sequenced as 100 X 2 paired end reads using NovaSeq S4 sequencing kit and NovaSeq Control Software v1.8.0. Base-calling is performed using Illumina’s RTA version 3.4.4.

scRNA-seq quality control, integration, clustering, ligand-receptor interaction and differential expression analysis.

Single-cell sequence preprocessing was performed using the standard 10x Genomics Cell Ranger Single Cell Software Suite. Raw reads were aligned to the hg38 reference genome, and UMI (unique molecular identifier) counting was performed using Cell Ranger v7.1.0 with default parameters. Seurat v5 was used for all subsequent analyses ⁷¹. Genes not detected in at least three single cells were excluded. Cells with fewer than 500 specific genes detected were also excluded to remove potential low-quality cells. Mitochondrial quality control metrics were calculated using the ‘PercentageFeatureSet’ function to filter out cells with >20% mitochondrial counts, thereby avoiding low-quality and dying cells. Normalization was performed with the ‘LogNormalize’ function in Seurat, followed by log transformation for downstream analysis. The ‘FindVariableFeatures’ function was then used to identify a subset of highly variable features for each sample, highlighting biological signals for future analyses. For integration analysis and batch effect removal, the Harmony R package (harmony_1.1.0) was utilized ⁶⁷. Principal component analysis (PCA), a dimensionality reduction technique, was conducted to determine the dataset's dimensionality, and UMAP was chosen to visualize the combined dataset. The ‘FindNeighbors’ and ‘FindClusters’ functions were used to cluster cells. Cluster-specific markers conserved across conditions were identified using the ‘FindConservedMarkers’ function, and clusters were assigned to known cell types based on these markers.

After manual annotation of all cell clusters as specific cell types, comparative analysis was performed using the MAST ⁷² model to identify differentially expressed genes (DEGs) induced by different conditions. GO analysis was performed using the clusterProfiler package ⁶⁹. DEGs identified from the comparative analysis were used as input for the GO enrichment analysis. The ‘enrichGO’ function was applied to categorize DEGs into biological processes, molecular functions, and cellular components. The results were visualized using the ‘dotplot’ function to highlight significant GO terms.

The ligand-receptor interaction analysis was performed using NicheNet v2 ²⁰. We utilized the Seurat v5 object provided by the Department of Quantitative Health Sciences at Mayo Clinic. Our methodology closely followed the vignettes available on GitHub from saeyslab/nichenetr, specifically “Perform NicheNet analysis starting from a Seurat object” and “Seurat Wrapper + Circos visualization.” First, we loaded the ligand-receptor network, ligand-target matrix, and weighted networks for mice from https://zenodo.org/records/7074291. From our original Seurat object, we created a subset Seurat object that included only cells classified as “Microglia” and “Neuron,” with “Microglia” also encompassing the subset labeled as “Mitotic microglia.” After re-normalizing the raw counts of the subsetted object, we ran `nichenet_seuratobj_aggregate` with the sender population set to “Neuron” and the receiver population set to “Microglia.” We configured this function to return the top 30 ligands and top 200 gene targets. Finally, we used the output from this function to create a Circos plot, which facilitated better visualization of the identified interactions.

For RNA Velocity analysis, the Velocyto pipeline was used to quantify spliced and unspliced RNA counts from the single-cell RNA sequencing data ²¹ and generate loom files. Gene-specific velocities were then computed using the scVelo package ²². The ‘scvelo.pp.filter_and_normalize’ function was used for pre-processing, including filtering and normalization of the data. The function ‘scvelo.pp.moments’ is applied to capture transcriptional dynamics and RNA velocity vectors were estimated using the ‘scvelo.tl.velocity’ function. Finally, the velocities were projected onto the existing UMAP embedding using the ‘scvelo.tl.velocity_graph’ and ‘scvelo.pl.velocity_embedding’ functions to visualize the dynamic changes in gene expression over time. The latent time was calculated using the ‘scv.tl.latent_time’ function, which helps in identifying the progression of cells through different states. The results were visualized with ‘scv.pl.heatmap’, which displays the expression dynamics of top genes over the inferred latent time, and ‘scv.pl.scatter’, which plots the latent time on the UMAP embedding, allowing for visualization of temporal progression across the cell population.

IMMUNOHISTOCHEMISTRY

Immunofluorescence staining.

Mice were anesthetized with isoflurane (5% in O₂) and then intracardially perfused with 40 ml cold PBS, followed by 40 ml cold 4% paraformaldehyde (PFA). Brains were removed and post-fixed in 4% PFA for an additional 6 hours, then transferred to 30% sucrose in PBS for three days. Tissue was then embedded in OCT and frozen before being cryosectioned using a Leica Cryostat at 20 μm. For immunofluorescence staining, the sections were blocked for 60 mins with 10% goat or donkey serum in TBS buffer containing 0.4% Triton X-100 (Sigma), and then incubated overnight at 4 °C with a primary IgG antibody. After three washes with TBST, sections were exposed to appropriate secondary antibody for 60 mins at room temperature, washed and mounted and coverslipped with Fluoromount-G (SouthernBiotech). Fluorescent images were captured with a confocal microscope (LSM980, Zen software, Zeiss) in the primary motor cortex. Images were taken from a single Z-plane (1024 x 1024 Pixels). Cells counts, fluorescence signal intensity and area of each cell were quantified using the Analyze Particles function in ImageJ (National Institutes of Health, Bethesda, MD). Volume rendering and visualization were performed using Imaris v.9.2 (Oxford Instruments). The antibodies are listed in the key resources table.

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit anti-IBA1	Abcam	Cat# ab178847; RRID: AB_2832244
Goat anti-IBA1	Wako	Cat# 011-27991; RRID: AB_2935833
Rabbit anti-IBA1	Wako	Cat# 013-27691; RRID: AB_2934095
Rabbit anti-NueN	Abcam	Cat# ab104225; RRID: AB_10711153
Mouse anti-NueN	Abcam	Cat# ab104224; RRID: AB_10711040
Mouse anti-LPL	Abcam	Cat# ab21356; RRID: AB_446221
Rat anti-MHCII	Invitrogen	Cat# 14-5321-82; RRID: AB_467561
Hamster anti-CD11c	Thermofisher	Cat# 14011482; RRID: AB_467115
Mouse anti-AXL	R&D	Cat# AF854; RRID: AB_355663
Goat anti-Gal3	R&D	Cat# AF1197; RRID: AB_2234687
Mouse anti-ANKG	NeuroMab	Cat# 75-146; RRID: AB_10673030
Rat anti-CD31	BD	Cat# 550274; RRID: AB_393571
Rabbit anti-MAP2	Cell Signaling	Cat# 4542S; RRID: UniProt ID: P11137
Rabbit anti-P2Y12	Anaspec	Cat# AS-55043A; UNSPSC: 12352203
Rabbit anti-TMEM119	Abcam	Cat# ab209064; RRID: AB_2800343
Rabbit anti-CD68	Abcam	Cat# ab125212; RRID: AB_10975465
Guinea pig anti-VGLUT1	Synaptic Systems	Cat# 135-304; RRID: AB_887878
Mouse anti-PSD95	Millipore	Cat# MAB1596; RRID: AB_2092365
Rabbit anti-Cleaved Caspase 3	Cell signaling	Cat# 9661; RRID: AB_2341188
Chicken anti-mCherry	Antibodiesinc	Cat# MCHERRY-0020; UniProt ID: X5DSL3
Rabbit anti-cFOS	Cell Signaling	Cat# 2250; RRID: AB_2247211
Goat anti-mouse 488	Invitrogen	Cat# A11001
Goat anti-mouse 546	Invitrogen	Cat# A11030
Goat anti-mouse 647	Invitrogen	Cat# A21235
Goat anti-rabbit 488	Invitrogen	Cat# A11008
Goat anti-rabbit 555	Invitrogen	Cat# A21428
Goat anti-rabbit 647	Invitrogen	Cat# A21244
Goat anti-rat 594	Invitrogen	Cat# A11007
Goat anti-hamster 647	Abcam	Cat# ab173004
Goat anti-guinea pig 594	Invitrogen	Cat# A11076
Donkey anti-rabbit 488	Invitrogen	Cat# A21206
Donkey anti-rabbit 594	Invitrogen	Cat# A21207
Donkey anti-goat 488	Invitrogen	Cat# A11055
Donkey anti-goat 594	Invitrogen	Cat# A11058
Donkey anti-chicken 555	Invitrogen	Cat# A78949
Bacterial and virus strains
pENN.AAV9.CaMKII.GCaMP6s.WP RE.SV40	Addgene	Cat# 107790
pENN.AAV.CamKII 0.4.Cre.SV40	Addgene	Cat# 105558-AAV1
pAAV-FLEX-tdTomato	Addgene	Cat# 28306-AAV9
AAV2/9-CaMKIIα-GCaMP6s	BrainVTA	Cat# PT-0110
AAV2/9-CaMKIIα-hM4D(Gi)-mCherry	BrainVTA	Cat# PT-0017
AAV2/9-CaMKIIα-mCherry	BrainVTA	Cat# PT-0108
Biological samples
Post-mortem brain samples of ALS cases	Brain Bank for Neurodegenerative Disorders at Mayo Clinic Jacksonville	N/A
Post-mortem brain samples of Brainstem-type Lewy body disease (BLBD) cases	Brain Bank for Neurodegenerative Disorders at Mayo Clinic Jacksonville	N/A
Chemicals, peptides, and recombinant proteins
KCl	Sigma-Aldrich	Cat# P9541; CAS: 7447-40-7
NaH2PO4·2H2O	Sigma-Aldrich	Cat# 71505; CAS: 13472-35-0
NaHCO3	Sigma-Aldrich	Cat# S5761; CAS: 144-55-8
Ascorbic acid	Sigma-Aldrich	Cat# 1043003; CAS: 50-81-7
Sodium pyruvate	Sigma-Aldrich	Cat# P5280; CAS: 113-24-6
CaCl2·2H2O	Sigma-Aldrich	Cat# C3306; CAS: 10035-04-8
MgCl2·6H2O	Sigma-Aldrich	Cat# M9272; CAS: 7791-18-6
Sucrose	Sigma-Aldrich	Cat# S0389; CAS: 57-50-1
HCl	Sigma-Aldrich	Cat# 258148; CAS: 7647-01-0
NaCl	Sigma-Aldrich	Cat# S9888; CAS: 7647-14-5
EGTA	Sigma-Aldrich	Cat# 324626; CAS: 67-42-5
HEPES	Sigma-Aldrich	Cat# 391340; CAS: 7365-45-9
KOH	Sigma-Aldrich	Cat# 417661; CAS: 1310-58-3
CNO	Caymanchem chemical	Cat# 16882; CAS: 34233-69-7
Formaldehyde	Grainger	Cat# LC146705
Xylene	Sigma-Aldrich	Cat# 1.94600; CAS: 1330-20-7
O.C.T. Compound	Fisher Scientific	Cat# 23-730-571
Triton X-100	Sigma-Aldrich	Cat# X100; CAS: 9036-19-5
Cresyl violet acetate	Sigma-Aldrich	Cat# C5042; 10510-54-0
Acetic acid	Sigma-Aldrich	Cat# 695092; CAS: 64-19-7
Critical commercial assays
10X Visium kits	10X Genomics	Cat# 1000339; Cat# 1000251;
RNeasy FFPE Kit for RNA Extraction	Qiagen	Cat# 73504
Neural Tissue Dissociation Kit P	Miltenyi Biotec	Cat# 130-092-628
Red blood cell lysis	Miltenyi Biotec	Cat# 130-094-183
Dead cell removal	Miltenyi Biotec	Cat# 130-090-101
Myelin Removal Beads II	Miltenyi Biotec	Cat# 130-096-733
RNeasy Mini Kit	Qiagen	Cat# 74104
SensiFAST cDNA Synthesis Kit	Thomas Scientific	Cat# BIO-65053
SensiFAST SYBR No-ROX Kit	Thomas Scientific	Cat# C755J00
Deposited data
Sp-RNA-seq	This study	GEO: GSE302449
Sc-RNA-seq	This study	GEO: GSE295514
Experimental models: Organisms/strains
NEFH-tTA line 8	JAX	Cat# No. 025397
tetO-hTDP-43-ΔNLS line 4	JAX	Cat# No. 014650
Thy1-YFP-H	JAX	Cat# No. 003782
CX3CR-1GFP knock-in/knock-out	JAX	Cat# No. 005582
TREM2 Knock Out	Kind gift from Dr. Marco Colonna	N/A
Oligonucleotides
Primers, See Table S2	This paper	N/A
Software and algorithms
Adobe Illustrator CC2025	Adobe	https://adobe.com; RRID: SCR_010279
GraphPad Prism v10.3.0	Graphpad	https://www.graphpad.com; RRID: SCR_002798
ImageJ (Fiji)	Schindelin et al.	https://imagej.net; RRID: SCR_002285
Imaris v.9.2	Oxford Instruments	https://imaris.oxinst.com; RRID:SCR_007370
Matlab	MathWorks	https://www.mathworks.com; RRID:SCR_001622
Zeus High Throughput Recording software	Bio-Signaling Technologies	https://www.bio-signal.com
Offline Sorter software	Plexon	https://plexon.com
Neuro Explorer 5.0	Plexon	https://plexon.com
Fiber photometry system	Inper Ltd.,	https://www.inper.com
Inper plot	Inper Ltd.,	https://www.inper.com
Vi-Cell XR Cell Viability Analyzer	Beckman-Coulter	https://www.beckman.com
Multiclamp 700B amplifier	Molecular Devices	https://www.moleculardevices.com
Clampfit 10.3	Molecular Devices	https://www.moleculardevices.com
QuantStudio 6 system	ThermoFisher	https://www.thermofisher.com
Other
A1x32-Poly2-3mm-50 s-177-CM32 silicon probes	NeuroNexus Technologies	Customed designed item
Optical fibers	Inper	Customed designed item
Stereotaxic injector	World Precision Instruments	Model UMC4
Dox chow	Bio-Serv	Cat# 3888
Depex medium	Electron Microscopy Sciences	Cat# 13514
Fluoromount-G	SouthernBiotech	Cat# 0100-20

IMARIS Rendering.

Three-dimensional renderings of rod-shaped microglial interaction with apical dendrites (IBA1 staining in rNLS8:Thy1-YFP-H mouse brain slices) and microglial engulfment of synaptic markers (costaining of IBA1/ VGLUT1/PSD95) were performed in Imaris using representative Z stack images. Z-stack images of full microglia were collected by a 63x objective (oil, NA:1.4), with 2048x2048 pixel rendering and a 0.3 μm step size (LSM980, Zen software, Zeiss). Channels were respectively converted into surfaces using Imaris to preserve structural and spatial integrity. Individual PSD95 and VGLUT1 puncta that were not interacting or engulfed by the IBA1⁺ surface was manually removed. Representative images were taken with the Snapshot function and videos taken with the Animation function.

Immunohistochemistry.

Formalin-fixed brains underwent systematic and standardized sampling with neuropathologic examination by a single board-certified neuropathologist (D.W.D.). Specific brain regions were dissected from the fixed hemibrain and studied for gross and microscopic pathology using the Dickson sampling scheme ⁷³. 5 μm thick paraffin-embedded sections were cut. Immunohistochemistry for microglia (IBA-1, rabbit polyclonal, 1:3000, Wako Chemicals, 019-19741) was performed on the motor cortex of 25 ALS and 25 BLBD cases.

Nissl staining.

Tissue sections were first incubated in 100% ethanol for 6 minutes, followed by a defatting process in xylene for 15 minutes, and then placed back into 100% ethanol for another 10 minutes. After being rinsed with distilled water, the slides were stained with a 0.5% solution of cresyl violet acetate for 15 minutes, followed by another rinse in distilled water. Subsequently, the sections were treated in a differentiation buffer (comprising 0.2% acetic acid in 95% ethanol) for 2 minutes, dehydrated with ethanol and Xylene, and finally mounted with Depex medium. Pyramidal neurons were distinguished by their characteristic triangular shape and the presence of a single, large apical dendrite extending vertically toward the pial surface.

QUANTITATIVE REAL-TIME PCR

Total RNA was isolated from mouse motor cortex using the RNeasy Mini Kit (Qiagen, 74104) according to the manufacturer’s instructions. cDNA was synthesized from 300 ng of RNA using the SensiFAST cDNA Synthesis Kit (Thomas Scientific, BIO-65053). Quantitative real-time PCR (qRT-PCR) was performed on a QuantStudio 6 system using the SensiFAST SYBR No-ROX Kit (Thomas Scientific, C755J00). Relative gene expression levels were calculated using the ΔΔCt method, with GAPDH as the internal control. The list of primers used is provided in Table S3B.

BEHAVIOR TEST

Rotarod.

The Rotarod performance test was performed to assess the balance and motor coordination of mice. Before the tests, the mice were allowed to acclimate to the testing environment for 1 hour. Following this, the test was conducted on a five-lane Rotarod apparatus (Med Associates Inc.), which began at a speed of 4 revolutions per minute (rpm) and gradually increased to 40 rpm over a span of 5 minutes. Each mouse underwent the test three times, with a rest period of 10 minutes between each trial.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical details of the experiments, including sample sizes and statistical tests are described in figure legends. Mean values of multiple groups were compared using a one-way ANOVA, followed by a Fisher’s post-hoc test (comparison to baseline), or Tukey’s post-hoc test (comparisons between all groups). Mean values of two groups across time were compared using a two-way ANOVA, followed by either a Sidak’s post-hoc test (comparison between genotype/group at each time point), or a Dunnett’s post-hoc test (comparison of a time point against the baseline). All post-hoc testing accounted for multiple comparisons. Comparison of two groups at a single timepoint was performed using Student’s t-test. Survival curves were analyzed using a log-rank (Mantel–Cox) test. P values < 0.05 were considered as statistically significant. All statistical analysis was performed with GraphPad Prism software (version 10). No statistical methods were used to pre-determine sample sizes, but our samples sizes are comparable to similar studies ^54,74. Data distribution was assumed to be normal, but this was not formally tested. Mice were grouped according to genotype before they were randomly assigned to the experimental groups. The investigators were blinded to group allocation during data collection and analysis. Animals or samples were excluded from analysis only in the instance of technical failure. No data were excluded for other reasons.

Supplementary Material

2

Table S1. Gene lists for spatial RNA-seq dataset, related to Figure 2. A-M.) Complete gene lists for each cluster identified from the spatial RNA-seq dataset. N.) Complete gene lists from the differential gene expression analysis of each cluster identified from the spatial RNA-seq dataset.

3

Table S2. Gene lists for single cell RNA-seq dataset, related to Figure 3, Figure 5 and Figure 7. A.) Complete gene lists for each cell type identified from single cell RNA-seq (scRNA-seq) dataset. B.) Complete gene lists for each microglial subcluster identified from scRNA-seq dataset. C.) List of 500 overlapping genes between the microglial cluster identified in our scRNA-seq dataset and those reported in a foundational study. D.) List of overlapping genes between the rNLS8-specific cluster from the spatial RNA-seq dataset and either microglial cluster (List 1) or microglial subcluster 2 (List 2) from the scRNA-seq dataset. E.) Lists of the top 100 upregulated and downregulated genes from differential expression analysis of microglial subcluster 2 identified by scRNA-seq. F.) List of 100 selected highly dynamic genes along the latent time trajectory from RNA velocity analysis of microglial subclusters.

4

Table S3. Demographic characteristics of ALS and control cases and list of primers used in this study, related to STAR Methods. A.) Demographic characteristics of ALS and control cases. B.) The list of primers.

5

Video S1. Neuronal calcium activity in rNLS8 and control mice at baseline and DOX off day 16, related to Figure 1.

6

Video S2. Three-dimensional renderings of rod-shaped microglial interaction with apical dendrites, related to Figure 6.

7

Video S3. Three-dimensional renderings of ramified and rod-shaped microglial engulfment of excitatory synapses, related to Figure 6.

8

Video S4. Neuronal calcium activity in rNLS8:TREM2 KO mice at baseline and DOX off day 16, related to Figure 7.

Highlights.

Neuronal hyperactivity emerges early in TDP-43 neurodegeneration

Neuronal hyperactivity induces the formation of Rod-shaped microglia

Rod-shaped microglia contact dendrites and remodel excitatory inputs

TREM2 signaling promotes rod-shaped microglia and neuroprotection