Abstract

   Amyotrophic lateral sclerosis (ALS) involves motor neuron death due to
   mislocalized TDP-43. Pathologic TDP-43 associates with stress granules
   (SGs), and lowering the SG-associated protein ataxin-2 (ATXN2) using
   Atxn2-targeting antisense oligonucleotides prolongs survival in TAR4/4
   sporadic ALS mice but failed in clinical trials likely due to poor
   target engagement. Here we show that an AAV with potent motor neuron
   transduction delivering Atxn2-targeting miRNAs reduces Atxn2 throughout
   the central nervous system at doses 40x lower than published work. In
   TAR4/4 mice, miAtxn2 increased survival (50%) and strength, and reduced
   motor neuron death, inflammation, and phosphorylated TDP-43. TAR4/4
   transcriptomic dysregulation recapitulated ALS gene signatures that
   were rescued by miAtxn2, identifying potential therapeutic mechanisms
   and biomarkers. In slow progressing hemizygous mice, miAtxn2 slowed
   disease progression, and in ALS patient-derived lower motor neurons,
   our AAV vector transduced >95% of cells and potently reduced ATXN2 at
   MOI 4 logs lower than previously reported. These data support AAV-RNAi
   targeting ATXN2 as a translatable therapy for sporadic ALS.

   Subject terms: Amyotrophic lateral sclerosis, Amyotrophic lateral
   sclerosis, RNAi
     __________________________________________________________________

   Amado et al. develop a gene therapy for sporadic ALS using motor
   neuron-targeting AAVs to deliver RNAi targeting ataxin-2. In a mouse
   model, survival, strength, and disease-related pathology are improved;
   and human motor neurons are strongly transduced.

Introduction

   Amyotrophic lateral sclerosis (ALS) is a uniformly fatal disease
   characterized by progressive degeneration of upper and lower motor
   neurons in the cerebral motor cortex and anterior horn cells of the
   spinal cord, respectively. Although 10% of cases are familial and an
   additional 10–15% of patients have an identifiable causal
   mutation^[53]1,[54]2, 97% of all ALS cases have a shared downstream
   pathology of cytoplasmic mislocalization and aggregation of a key
   nuclear protein, Tar DNA binding protein of 43 kD (TDP-43). A
   contributory event to TDP-43 mislocalization is its aggregation in
   cytoplasmic stress granules, resulting in neuronal death through
   cytoplasmic gain-of-function^[55]3,[56]4 and nuclear
   loss-of-function^[57]5–[58]7. Inhibiting stress granule formation is
   therefore a promising therapeutic strategy. One approach to achieve
   this is downregulating the stress granule-associated protein ataxin-2
   (ATXN2). Indeed, recent work showed that antisense oligonucleotides
   (ASOs) could prolong survival by 35% in a mouse model of sporadic
   ALS^[59]8. Disappointingly, this strategy was not successful in a
   recent clinical trial ([60]NCT04494256), in which no clinical
   improvement was seen, and neurofilament light chain (NfL), a biomarker
   of neurodegeneration, was not reduced in the cerebrospinal fluid (CSF).
   However, only 21% reduction of CSF ATXN2 was achieved, and only 12%
   more reduction than was seen in placebo^[61]9, an amount far lower than
   the anticipated reductions predicted from preclinical studies. As the
   maximal safe dose had been administered, further development of this
   strategy was halted. Therefore, the effects of ATXN2 knockdown in ALS
   patients remain unknown. An alternative approach is to achieve strong
   and lasting knockdown throughout motor neurons of the brain and spinal
   cord using AAV-mediated RNAi delivery. If effective, this could be an
   attractive alternative to ASOs, reducing the morbidity and logistical
   challenges associated with repeated intrathecal injection.

   Here, we designed and packaged microRNAs (miRNA) targeting Atxn2 into a
   peptide-modified AAV9 variant, PM-AAV9, engineered for superior CNS
   targeting, and showed that a single intracerebroventricular (ICV)
   injection early in life provides sustained knockdown throughout the
   mouse central nervous system. In the same ALS mouse model used in ASO
   studies^[62]8,[63]10, we demonstrate that miAtxn2 treatment markedly
   improves survival, strength, and motor coordination. miAtxn2 treatment
   also decreases TDP-43 histopathology, reduces inflammation, and
   improves motor neuron phenotypes. We further demonstrate correction of
   key ALS-associated transcripts dysregulated in this model and
   demonstrate the ability of our vectors to both transduce and reduce
   human ATXN2 levels in human lower motor neurons, supporting the
   potential of AAV-mediated RNAi against ATXN2 as a promising treatment
   strategy for the 97% of ALS cases characterized by TDP-43 pathology.

Results

Generation of miRNA, capsid selection, and ATXN2 knockdown efficacy

   MiRNAs targeting mouse Atxn2 were designed and screened for knockdown
   efficacy in murine N2A cells (Fig. [64]S1A) as well as optimized for
   strand loading bias. From this screen, miRNA V1, which targets both
   mouse and human transcripts, was used for further studies, hereby
   referred to as miAtxn2 (Fig. [65]S1B).

   In parallel, three different AAV9 capsid variants were assessed for
   transduction efficiency of upper motor neurons in the cerebral cortex
   and lower motor neurons in the anterior horn of the mouse spinal cord
   to determine the optimal serotype for miAtxn2 delivery. Pups were
   generated by crossing wildtype C57Bl/6 mice with wildtype SJL mice to
   match the mixed background of the ALS mouse model to ensure
   strain-relevance of the transduction results. Pups were injected at
   postnatal day 1 (P1) in the lateral ventricles with identical doses of
   either wildtype AAV9, AAV-PHP.eB, or a peptide-modified AAV9 that we
   developed in earlier work^[66]11, AAV9.KGG, which we here refer to as
   PM-AAV9. Each capsid was used to deliver enhanced GFP (eGFP) driven by
   a universal promoter (Fig. [67]1A). Native fluorescence visualization
   of brains (Fig. [68]1B, C) and lumbar spinal cord sections
   (Fig. [69]1D) showed that PM-AAV9 provided superior transduction, with
   strong expression in Layer V of the motor cortex (Fig. [70]1C) and
   anterior horn cells of the spinal cord (Fig. [71]1D).

Fig. 1. Capsid selection and AAV-miAtxn2-mediated reduction of ataxin-2 in
target cells.

   Fig. 1
   [72]Open in a new tab

   A Schematic of vector and delivery route used in capsid comparison
   study. Enhanced GFP driven by a universal promoter was packaged into
   AAV9, AAV-PHP.eB, or PM-AAV9 and delivered by bilateral ICV injection
   into wildtype mice at P1. B Comparison of cortical transduction
   achieved by each capsid 3 weeks post-injection, with images taken at
   the same exposure for transduction comparison. C PM-AAV9 was re-imaged
   at lower exposure for more specific visualization of transduction
   including Layer V cortical cells. D Comparison of lumbar spine
   transduction by each capsid, with 10X image showing anterior horn cell
   transduction. Scale bars for (B) and (D) represent 250 μm. E Schematic
   of vector used in therapeutic studies, consisting of miAtxn2 driven by
   the mU6 promoter and packaged in PM-AAV9, herein referred to as
   miAtxn2. F ATXN2 protein levels in wildtype, TAR4/4, and
   miAtxn2-treated TAR4/4 mouse lumbar spine 3 weeks after treatment with
   miAtxn2, visualized by ATXN2 (green) and ChAT (red) immunofluorescence.
   Scale bar represents 250 μm. G Quantification of ATXN2 in ChAT+
   anterior horn cells, normalized to untreated TAR4/4 mice. Values for
   individual mice are shown along with mean and standard error for each
   group. N = 5 mice/group for Wildtype and miAtxn2-treated mutant and 4
   mice/group for Mutant, with 2-10 images analyzed per mouse. Data were
   analyzed by ordinary one-way ANOVA followed by Holm-Sidak post-hoc.
   *p < 0.05, ****p < 0.0001. Specific p-values: Wildtype vs Mutant,
   0.0137; Mutant vs miAtxn2-treated, 0.0137; Wildtype vs miAtxn2-treated,
   <0.0001. Source data are provided as a Source Data file. Schematics in
   (A) and (E) were created in BioRender: Amado, D. (2025),
   [73]https://BioRender.com/r71o037 and Amado, D. (2025)
   [74]https://BioRender.com/n53n743, respectively.

   MiAtxn2 was packaged in PM-AAV9 and injected into the lateral
   ventricles at P1 for all further studies (Fig. [75]1E). In C57Bl/6/SJL
   wildtype mice, Atxn2 mRNA levels were measured 3 weeks after injection
   in ALS-relevant regions of the central nervous system (Fig. S[76]1Ci),
   demonstrating 54% knockdown in the frontal cortex (Fig. S[77]1Cii) and
   23-25% knockdown in the brainstem (Fig. S[78]1Ciii), cervical spine
   (Fig. S[79]1Civ), and lumbar spine (Fig. S[80]1Cv) in mice treated with
   PM-AAV9.miAtxn2 (herein referred to as miAtxn2) vs mice treated with a
   non-Atxn2-targeting miRNA (miCtrl). Cortical ATXN2 protein levels were
   correspondingly reduced 32% in homogenized tissue by western blot
   (Fig. [81]S1D). There was no significant astrogliosis (Fig. [82]S1E) or
   microgliosis (Fig. [83]S1F) in response to treatment. In a cohort
   followed to 12 weeks, no dorsal root ganglia pathology was observed
   (Figure S[84]2A), with no morphologic differences or inflammatory
   infiltrates seen in the soma (Fig. [85]S2B) or nerve roots
   (Fig. [86]S2C) and no change in sensory neuron cell body size
   (Fig. [87]S2D, E).

Effect of ATXN2 knockdown on survival, weight, and strength in TAR4/4 mice

   The efficacy of miAtxn2 was assessed in the TAR4/4 mouse model of
   sporadic ALS^[88]10,[89]12, a model with wildtype human TDP-43
   pan-neuronally expressed starting at P7. Homozygous TAR4/4 mice develop
   an aggressive course of progressive weakness and weight loss, reaching
   endpoint in the fourth week of life. As disease advances, they lose
   motor neurons and display phosphorylated TDP-43 inclusions. Because of
   the rapid disease course and the delayed peak of AAV-mediated gene
   expression, TAR4/4 mice were treated at P1 by bilateral ICV injection
   (Fig. [90]1A). Rotarod performance and weight were measured at P17, and
   gait-related behaviors at P18, when disease is advanced but not
   end-stage. To mirror pre-IND-enabling studies, miAtxn2-treated mice
   were compared to buffer-treated mice.

   We first confirmed ATXN2 protein reduction in the lumbar spine of
   TAR4/4 mice. Western blot was felt to have low sensitivity for ATXN2
   reduction in lower motor neurons, which are preferentially targeted by
   PM-AAV9 (Fig. [91]1D), as these cells constitute a small percentage of
   the total cellular constituents of the spinal cord. We therefore used
   immunohistochemistry on lumbar spine sections of TAR4/4 mice treated
   with miAtxn2 or buffer and compared these to wildtype levels
   (Fig. [92]1F). ATXN2 was lower in untreated TAR4/4 motor neurons than
   in those of untreated wildtype mice. Treatment with miAtxn2 resulted in
   a mean ATXN2 protein reduction of 65% in ChAT+ motor neurons with a
   maximum reduction of 100% (Fig. [93]1G).

   In TAR4/4 mice treated with miAtxn2, there was a 45.5% increase in
   median survival over buffer-treated mice (Fig. [94]2A). Mean survival
   was increased 54% (from 22 days ± 1 day to 33.5 days ± 3 days), with
   the longest-lived treated mouse surviving 58 days. A composite gait
   score was calculated for each mouse by scoring and summing parameters
   of abdominal droop, limping, foot angling, kyphosis, tremor, and
   clasping (Fig. [95]S3A). For the composite gait score (Fig. [96]2B) as
   well as for each of the sub-scores of limping (Fig. [97]S3B), foot
   angling (Fig. [98]S3C), kyphosis (Fig. [99]S3D), tremor
   (Fig. [100]S3E), and clasping (Fig. [101]S3F), the TAR4/4 phenotype was
   improved by miAtxn2 treatment. Rotarod duration, which was markedly
   decreased in TAR4/4 mice compared to wildtype, was improved with
   miAtxn2 treatment, with treated mice performing 2.6 times better than
   buffer-treated mice (Fig. [102]2C).

Fig. 2. Effects of ATXN2 reduction on survival and strength in TAR4/4 mice.

   [103]Fig. 2
   [104]Open in a new tab

   A Survival curve of miAtxn2-treated TAR4/4 mice compared to
   buffer-treated mice. N = 14 TAR4/4 and 15 miAtxn2-treated mice/group.
   Data were analyzed by both Log-rank [Mantel-Cox] test and
   Gehan-Breslow-Wilcoxon test. B Comparison of composite gait score at
   P18. N = 17 WT, 12 TAR4/4, and 14 miAtxn2-treated. C Comparison of
   endurance on accelerating rotarod at P17. N = 17 WT, 10 TAR4/4 (with 2
   additional mice unable to perform test due to weakness), and 14
   miAtxn2-treated. D Duration and frequency of vertical activity at P18.
   N = 17 WT, 11 TAR4/4, 12 miAtxn2-treated WT, 12 miAtxn2-treated TAR4/4.
   Values for individual mice are shown along with mean and standard error
   for each group. *p < 0.05, **p < 0.005, ***p < 0.0005, ****p < 0.0001,
   n.s. not significant. Data were analyzed by ordinary one-way ANOVA
   followed by Holm-Sidak post-hoc. WT, wildtype; MT, TAR4/4. Specific
   p-values: (A) 0.0003; (B) All comparisons <0.0001; (C) WT vs MT,
   <0.0001; MT vs MT+miAtxn2, 0.0025; WT vs MT+miAtxn2, <0.0001; (D)
   Duration: WT vs MT, 0.9001; WT vs WT+miAtxn2, 0.8091; MT vs MT+miAtxn2,
   0.0051; WT+miAtxn2 vs MT+miAtxn2, 0.0016. Frequency: WT vs MT, 0.5283;
   WT vs WT+miAtxn2, 0.5267; MT vs MT+miAtxn2, <0.0001; WT+miAtxn2 vs
   MT+miAtxn2, <0.0001. Source data are provided as a Source Data file.

   The duration and frequency of vertical activity (Fig. [105]2D), which
   consists of both climbing and rearing behaviors and is considered a
   measure of hindlimb strength, were also assessed. While a TAR4/4
   phenotype was not observed, there was an unexpected increase in
   vertical activity specifically in miAtxn2-treated TAR4/4 mice over
   wildtype levels, both in duration (2.4-fold over wildtype) and
   frequency (2.2-fold over wildtype). To assess the genotype specificity
   of this effect, we measured vertical activity in age-matched wildtype
   littermates. We found that there was no change in vertical activity
   frequency or duration in miAtxn2-treated wildtype mice, suggesting that
   this effect is specific to the TDP-43 overexpressing model.

Effect of ATXN2 knockdown on cellular phenotypes

   Tissues from a subset of mice at P21-P23 were analyzed histologically.
   There was widespread apoptosis throughout the lumbar spine, as measured
   by the apoptotic marker cleaved caspase 3 (CC3); this was improved by
   treatment with miAtxn2 and fully corrected at the L3 and L4 levels
   (Fig. [106]3A, B). In the posterior lumbar spine there was loss of
   lower motor neurons, identified by ChAT positivity, that was rescued in
   miAtxn2-treated mice (Fig. [107]3C, D).

Fig. 3. Lower motor neuron survival in TAR4/4 mice after miAtxn2 treatment.

   [108]Fig. 3
   [109]Open in a new tab

   A Lumbar spinal sections from wildtype, TAR4/4, and miAtxn2-treated
   mice were stained with the apoptotic marker cleaved caspase 3 (CC3).
   Scale bar represents 250 μm. B CC3-positive cells were quantified at
   each of lumbar levels L3-L6 (L3 shown in A). N = 3–7 mice/group at each
   level (specifics below), 2–7 sections/mouse per level. C Lower motor
   neurons were identified by ChAT staining. Scale bar represents 250 μm.
   D ChAT-positive cells were quantified at each of lumbar levels L3-L6
   (L6 shown in C). N = 4–6 mice/group at each level (specifics below),
   4–10 sections/mouse per level. For (B) and (D), values for individual
   mice are shown along with mean and standard error for each group.
   Values from TAR4/4 mice treated with buffer or with miAtxn2 were
   normalized to wildtype levels from the same batch. *p < 0.05,
   **p < 0.005, ***p < 0.0005, ****p < 0.0001, n.s. not significant. Data
   were analyzed by two-way ANOVA split by mouse followed by Holm-Sidak
   post-hoc. LMN lower motor neuron; Mutant, TAR4/4. Specific N, listed in
   order of Wildtype, Mutant, and miAtxn2-treated: (B) L3: 6, 5, 3; L4: 7,
   4, 4; L5: 4, 6, 3; L6: 5, 5, 4; (D) L3: 4, 3, 4; L4: 4, 3, 4; L5: 5, 3,
   4; L6: 4, 3, 4. Specific p-values, listed in order of Wildtype vs
   Mutant, Mutant vs miAtxn2-treated, and Wildtype vs miAtxn2-treated: (B)
   L3: 0.0014, 0.0462, 0.1697; L4: <0.0001, 0.0001, 0.2163; L5: 0.0742,
   0.9990, 0.0466; L6: 0.0808, 0.1263, 0.0316; (D) L3: 0.8722, 0.7787,
   0.7366; L4: 0.0697, 0.7455, 0.0532; L5: 0.8611, 0.8611, 0.8611; L6:
   0.0397, 0.0482, 0.3224. Source data are provided as a Source Data file.

   TAR4/4 mice showed a band of astroglial and microglial inflammation
   throughout cortical layer 5 that was reduced with miAtxn2 treatment
   (Fig. [110]4A–D), although there was no impact on Layer V neuron counts
   (Fig. [111]S3G). In the lumbar spine, we focused on spinal level L6,
   where there was motor neuron loss in TAR4/4 mice, to assess whether
   this was accompanied by gliosis. A similar gliosis phenotype to that
   seen in Layer 5 of the cortex was seen (Fig. [112]4E, H), with
   significant improvement in microgliosis following miAtxn2 treatment
   (Fig. [113]4F, H).

Fig. 4. Effects of miAtxn2 on brain and spinal cord inflammation in TAR4/4
mice.

   [114]Fig. 4
   [115]Open in a new tab

   Markers of neuroinflammation were stained and quantified in
   miAtxn2-treated and buffer-treated TAR4/4 mice and buffer-treated
   wildtype littermates. A, C Representative images of motor cortex at 5X
   (top row) magnification, with arrows indicating the gliosis in Layer 5
   depicted below at 40X (bottom) magnification, stained for the astrocyte
   marker GFAP (A) or the microglial marker Iba1 (C). Scale bars represent
   500 μm and 50 μm, respectively. B, D Quantification of the percent area
   of Layer 5 staining positive for GFAP (B) and Iba1 (D). For (B), N = 3
   WT, 4 MT, 5 miAtxn2-treated, while for (D), N = 4 WT, 3 MT, 3
   miAtxn2-treated; 3-6 sections/mouse. E, G Representative images of
   anterior horn of lumbar spine at 10X (top row) and 20X (bottom row)
   magnification, stained for the astrocyte marker GFAP (E) and the
   microglial marker Iba1 (G). Scale bars represent 250 μm and 100 μm,
   respectively. F, H Quantification of the percent area of the anterior
   horn staining positive for GFAP (F) and Iba1 (H). For (F), N = 7 WT, 6
   MT, 5 miAtxn2-treated, while for (H), N = 6 WT, 5 MT, 6
   miAtxn2-treated; 1-4 sections/mouse. For all graphs, values for
   individual mice are shown along with mean and standard error for each
   group, with data normalized to WT. *p < 0.05, **p < 0.01,
   ***p < 0.0005, n.s. not significant. Data were analyzed by ordinary
   one-way ANOVA followed by Holm-Sidak post-hoc. WT, Wildtype; MT or
   Mutant, TAR4/4. Specific p-values, listed in order of WT vs MT, MT vs
   miAtxn2-treated, and WT vs miAtxn2-treated: (B) 0.0071, 0.0211, 0.1887;
   (D) 0.0015, 0.0142, 0.0739; (F) 0.0005, 0.0776, 0.0275; (H) 0.0003,
   0.0308, 0.0135. Source data are provided as a Source Data file.

   Pathologic TDP-43 is phosphorylated in TDP-43 protein-opathies^[116]13.
   TAR4/4 mice showed a strong increase in pTDP staining that normalized
   to WT levels after miAtxn2 treatment (Fig. [117]5A, B). This signal was
   prominent in motor neurons when lumbar spine sections were co-stained
   for pTDP, ChAT, and DAPI (Fig. [118]5C). To ensure this was not due to
   changes in total TDP-43 expression, we analyzed tissue from treated
   wildtype and hemizygous littermates of study mice, and found that in
   both the frontal cortex (Fig. [119]5D) and lumbar spine (Fig. [120]5E),
   total TDP-43 levels were not affected by treatment with miAtxn2.

Fig. 5. Effects of miAtxn2 on TDP-43 levels and pathology.

   [121]Fig. 5
   [122]Open in a new tab

   A Representative staining of phosphorylated TDP-43 (pTDP) in miAtxn2-
   or buffer-treated TAR4/4 mice and buffer-treated wildtype littermates
   at the L6 lumbar level. B Quantification of pTDP staining in the L6
   anterior horn. N = 4 mice/group for WT and MT and 5 mice/group for
   miAtxn2-treated, 1-5 slides/mouse. Values for individual mice are shown
   along with mean and standard error for each group, with data normalized
   to WT. **p < 0.01, n.s. not significant. Data were analyzed by ordinary
   one-way ANOVA followed by Holm-Sidak post-hoc. Specific p-values: WT vs
   MT, 0.0043; MT vs miAtxn2-treated, 0.0068; WT vs miAtxn2-treated,
   0.4572. C Confocal imaging of anterior horn cells stained with DAPI
   (blue), ChAT (red), and pTDP (green) at 40X magnification, with overlay
   shown in right panels; representative images from N = 3 mice/group are
   shown. Scale bars represent 50 μm. Total TDP-43 levels in treated and
   untreated wildtype and hemizygous TAR4 mouse frontal cortex (D) and
   lumbar spine (E) as measured by western blot analysis. N = 4 mice/group
   except for miAtxn2-treated WT for which N = 3. Values for individual
   mice are shown along with mean and standard error for each group, with
   data normalized to WT. *p < 0.05, n.s. not significant. Data were
   analyzed by unpaired two-tailed T-tests with comparisons shown on
   graph; a separate analysis by ordinary one-way ANOVA followed by
   Holm-Sidak post-hoc showed no significant differences between any
   comparison groups. WT Wildtype, MT or Mutant, TAR4/4; Hemi, TAR4
   hemizygous. Specific p-values: (D) WT vs WT+miAtxn2, 0.6703; Hemi vs
   Hemi+miAtxn2, 0.2678; WT vs Hemi, 0.0116; (E) WT vs WT+miAtxn2, 0.3149;
   Hemi vs Hemi+miAtxn2, 0.7654; WT vs Hemi, 0.2918. Source data including
   uncropped blots are provided as a Source Data file.

Effect of ATXN2 knockdown on transcriptional phenotypes

   To determine the disease-associated transcriptional signature of TAR4/4
   mice, we dissected the lumbar spine when there is pronounced disease at
   P19 (Fig. [123]6A) and determined differential gene expression (DGE)
   between mutant and wildtype mice using bulk-RNA sequencing. Six WT,
   five TAR4/4, and five miAtxn2-treated mice were included in analysis,
   balanced across litters such that each litter contained at least one
   mouse in each group. In total, 1288 genes were dysregulated as a result
   of TDP-43 overexpression (adjusted p-value < 0.05; | log[2](fold
   change) | > 0.5); Figs. [124]6B, [125]S4A). Gene enrichment analysis
   identified amyotrophic lateral sclerosis as the pathway term with the
   greatest gene overlap, as well as other processes known to be
   dysregulated in ALS such as calcium signaling, apoptosis, and steroid
   biosynthesis (Figs. [126]6C, [127]S4B).

Fig. 6. miAtxn2 treatment effect on transcriptomic dysregulation in TAR4/4
mice.

   [128]Fig. 6
   [129]Open in a new tab

   A Schematic of experimental design for bulk transcriptomics of gene
   rescue in the lumbar spine of miAtxn2-treated TAR4/4 mice. B Volcano
   plot of differential gene expression between buffer-treated TAR4/4
   (mutant) and buffer-treated wildtype (wildtype) mice. Significant
   differentially expressed genes are denoted in red (DESeq2 two-sided
   Wald test with Benjamini-Hochberg post-hoc, adjusted p-value < 0.05;
   | log[2](fold change) | > 0.5). C Lollipop plot of the top KEGG
   pathways enriched in genes dysregulated in mutant mice, identified in
   (A). richFactor and log[10](adjusted p-value) were calculated with the
   enrichR package using Fisher’s exact test. D Venn diagram of DEGs
   between mutant and wildtype mice and miAtxn2 and wildtype mice. 412
   genes were identified as “normalized” or no longer meeting our criteria
   for differential expression (DESeq2 two-sided Wald test with
   Benjamini-Hochberg post-hoc, adjusted p-value > 0.05; |log[2](fold
   change) | <0.5) between mutant and wildtype mice upon miAtxn2
   treatment. E Volcano plot of the log[2](fold change) and adjusted
   p-value for mutant vs. wildtype DEGs identified in (A) in response to
   miAtxn2 treatment. Blue denotes genes that were previously
   significantly expressed, but no longer meet these criteria upon miAtxn2
   treatment, as shown by the Venn diagram in (D). F Lollipop plot of the
   top KEGG pathways enriched in genes normalized upon miAtxn2 treatment,
   identified in (E). richFactor and log[10](adjusted p-value) were
   calculated with the enrichR package, which uses Benjamini-Hochberg
   post-hoc. G Tukey box plots showing the gene expression of selected top
   gene hits between mutant vs. wildtype and miAtxn2 vs. wildtype, across
   all experimental groups. H Tukey box plots of selected top hits from
   miAtxn2 vs. mutant mice. I Tukey box plots of Tardbp gene expression
   across experimental groups. For each boxplot in (G–I), each individual
   point corresponds to a single mouse; the central line corresponds to
   the median (50th percentile); the lower and upper bounds correspond to
   the first and third quartiles (25th and 75th percentile) respectively;
   and the upper and lower whisker lines extend to the largest and
   smallest values within 1.5x interquartile range, respectively. Mutant,
   TAR4/4; DEG differentially expressed genes. Schematic in (A) was
   created in BioRender: Robbins, A. (2025),
   [130]https://BioRender.com/6t575wq.

   Of the ~1300 genes differentially expressed in TDP-43 mice relative to
   wildtype mice, 452 genes were corrected with miAtxn2 treatment
   (Figs. [131]6D, [132]S4C). To visually illustrate the treatment-induced
   shift of the TAR4/4 disease signature (Fig. [133]6B) towards wildtype
   levels, the results of the DGE analysis between miAtxn2 and wildtype
   littermates were plotted for the subset of 1288 genes (Figs. [134]6E,
   [135]S4D). Pathway enrichment analysis of the normalized genes
   demonstrated biosynthetic processes, tissue homeostasis, and cell
   proliferation (Fig. [136]6F), and among these, 48 had expression levels
   in miAtxn2-treated mice that were no longer statistically significant
   from wildtype (adjusted p-value > 0.05; Fig. [137]S4E). We note rescued
   expression of genes involved in extracellular matrix organization
   (e.g., Fbln5 and Itga10)^[138]14–[139]16, the TCA cycle (e.g.,
   Idh3a)^[140]17, and DNA repair (e.g., Hist2h2aa1)^[141]18,[142]19, all
   of which were downregulated in TAR4/4 mice^[143]14–[144]16. We also
   note correction of signal transduction pathway genes that were
   up-regulated (e.g., Trpc6, Hif3a) or down-regulated (e.g., Vip, Vdr) in
   TAR4/4 compared to wildtype littermates^[145]20–[146]24 (Figs. [147]6G,
   [148]S4E).

   Additionally, 221 genes differentially expressed between
   miAtxn2-treated TAR4/4 mice and wildtype littermates were not in the
   original disease signature (Fig. [149]6D). To better assess the effect
   of miAtxn2 treatment, while controlling for genotype, we compared
   miAtxn2-treated mice to mutant littermates (Fig. [150]S4F). 93
   differentially expressed genes (DEGs) were identified (adjusted
   p-value < 0.05; | log[2](fold change) | > 0). Gene ontology analysis of
   this miAtxn2 gene signature identified pathways involved in
   neurogenesis, myelination, axonogenesis and sphingolipid binding
   (Fig. [151]S4G). 23 of the 93 DEGs overlapped with our TAR4/4 disease
   signature, a literature review of which showed them to be implicated in
   ALS and/or neurodegenerative pathogenesis. For example, Dbr1, a gene
   previously identified as a potential therapeutic target for
   ALS^[152]24, was upregulated in TAR4/4 mice and normalized toward
   wildtype with miAtxn2 treatment (Fig. [153]6H; Table [154]S1).

   Similar to prior reports^[155]8 and corroborating our western blot
   analysis, we found an expected increase in total TDP-43 transcript
   levels in TAR4/4 mice that was not affected by Atxn2 reduction
   (Fig. [156]6I).

Long-term effects of treatment in hemizygous and wildtype mice

   The hemizygous TAR4 mice in our colony show clasping onset at 3 months
   with emerging gait dysfunction shortly thereafter (Fig. [157]7A), which
   is an earlier onset than previously reported^[158]10. Gait dysfunction
   steadily progressed in buffer-treated TAR4 mice, but did not progress
   from 3 to 12 months in miAtxn2-treated TAR4 mice after which it
   progressed slowly. miAtxn2 treatment of wildtype littermates resulted
   in no phenotypic changes over 21 months of observation (Fig. [159]7B).

Fig. 7. Effect of miAtxn2 on gait function in TAR4 hemizygous and wildtype
mice.

   [160]Fig. 7
   [161]Open in a new tab

   TAR4 hemizygous mice and their wildtype littermates were treated with
   miAtxn2 or buffer at P1 and composite gait score was measured at 3
   month intervals. A Effect of miAtxn2 on TAR4 gait over time. B Effect
   of miAtxn2 on WT gait over time. N = 11–21 mice/group at 3–6 months,
   18–37 mice/group at 9–18 months, and 9–17 mice/group at 21 months
   (specifics below). Values for individual mice are shown along with mean
   and standard error for each group. *p < 0.05, **p < 0.01, ***p < 0.005,
   ****p < 0.001, n.s. not significant. Data were analyzed by two-way
   ANOVA followed by Holm-Sidak post-hoc. WT wildtype, Hemi, TAR4
   hemizygous. Specific N: For (A), listed in order of Wildtype, Mutant,
   and miAtxn2-treated mutant: 3 months: 11, 21, 21; 6 months: 11, 20, 21;
   9 months: 19, 19, 31; 12 months: 18, 35, 24; 15 months: 23, 37, 33; 18
   months: 17, 31, 23; 21 months: 9, 9, 17. For (B), listed in order of
   Wildtype and miAtxn2-treated wildtype: 3 months: 11, 14; 6 months: 11,
   9; 9 months: 19, 14; 12 months: 18, 12; 15 months: 23, 15; 18 months:
   17, 9; 21 months: 9, 6. Specific p-values: For (A), listed in order of
   Wildtype vs Mutant, Mutant vs miAtxn2-treated, and Wildtype vs
   miAtxn2-treated: 3 months, 0.0008, 0.4374, 0.0065; 6 months, <0.0001,
   0.2464, <0.0001; 9 months, <0.0001, 0.7395, <0.0001; 12 months,
   <0.0001, 0.0184, 0.0008; 15 months, <0.0001, 0.0102, <0.0001; 18
   months, <0.0001, 0.1635, <0.0001; 21 months, 0.0030, 0.2302, 0.0030.
   For (B), none significant (range: 0.7264-0.9935). Source data are
   provided as a Source Data file.

Translatability to human targets

   To determine the ability of miAtxn2 to target human ATXN2, we used the
   127Q transgenic mouse model of spinocerebellar ataxia type 2, in which
   mice overexpress mutant human ATXN2 in cerebellar Purkinje cells. After
   confirming targeting of affected cells by PM-AAV9.eGFP after ICV
   delivery at P1 (Fig. [162]8A), PM-AAV9.miAtxn2 delivery potently
   reduced human ATXN2 levels in the cerebellum (Fig. [163]8B).

Fig. 8. miAtxn2 targeting of human ATXN2 and transduction of ALS
patient-derived lower motor neurons.

   [164]Fig. 8
   [165]Open in a new tab

   A Mouse cerebellar Purkinje cell transduction after delivery of
   PM-AAV9.eGFP, representative image from N = 4 mice. Scale bar
   represents 250 μm. B Reduction of human ATXN2 mRNA in cerebellum of
   127Q transgenic mice, which express ATXN2 exclusively in Purkinje
   cells, after treatment with miAtxn2 compared to a non-ATXN2-targeting
   control (miCtrl). N = 5 mice/group. Values for individual mice are
   shown along with mean and standard error for each group.
   ****p < 0.0001. Data were analyzed by unpaired two-tailed T-test. C
   Confirmation of lower motor neuron identity of a differentiated
   C9orf72-ALS patient cell line using NeuN and ChAT markers;
   representative images from N = 3 replicates per marker are shown. Scale
   bar represents 100um. D Representative image and quantification of GFP
   positivity in patient-derived lower motor neurons after transduction
   with PM-AAV.miAtxn2. Scale bar represents 250 μm. N = 3 replicates per
   dose, and mean and standard error are shown. ****p < 0.0001. Data were
   analyzed by two-way ANOVA followed by Holm-Sidak post-hoc. E
   Dose-dependent reduction in ATXN2 levels in patient-derived lower motor
   neurons, without affecting total TDP-43 transcript levels (F), N = 2
   replicates per experiment. MOI, multiplicity of infection. For (D–F),
   each dot represents the averaged value of five images from an
   individual well. Source data are provided as a Source Data file.

   We next assessed the ability of PM-AAV9 to transduce human lower motor
   neurons, using a cell line derived from a C9orf72-ALS patient with 800
   repeats that was engineered to contain an integrated doxycycline
   (DOX)-inducible transposon expressing lower motor
   neuron-differentiating factors^[166]25. Upon DOX induction, lower motor
   neuron identity was confirmed by NeuN and ChAT staining (Fig. [167]8C).
   Prior studies using AAV have required MOI of up to 2 × 10^8 to
   transduce patient-derived lower motor neurons^[168]26. PM-AAV9 was used
   to deliver GFP and resulted in 70% neuronal transduction at MOI of
   5 × 10^4 that increased to >90% at MOI of 5 × 10^5 (Fig. [169]8C, D).
   PM-AAV9.miAtxn2 delivery resulted in dose-dependent ATXN2 mRNA
   reduction (Fig. [170]8E), achieving 70% reduction and 83% reduction at
   MOI of 5 × 10^4 and 5 × 10^5, respectively, without affecting total
   TDP-43 transcript levels (Fig. [171]8F).

Discussion

   In this study, we developed an RNAi-based method to reduce ATXN2 in a
   mouse model of sporadic ALS. We achieved potent reduction of target
   transcripts in ALS-relevant regions including the motor cortex,
   brainstem, and spinal cord without impacting total TDP-43 transcript
   levels, which in turn improved motor function and survival.
   Interestingly, we also observed increased vertical activity above
   wildtype levels. While Atxn2-null mice have locomotor hyperactivity, in
   our study only the TAR4/4 mice (and not their wildtype littermates)
   showed this effect of ATXN2 reduction, which therefore may represent an
   unmasking of impulsivity (i.e., an FTD-like phenotype) in the setting
   of improved motor function. Histopathologically, we observed markedly
   reduced inflammation in cortex and spinal cord, reduced apoptosis in
   proximal levels of the lumbar spine, and distal-predominant rescue of
   lower motor neurons with normalization of phosphorylated TDP-43 levels.
   This was accompanied by improvement in many genes dysregulated across
   the transcriptome, including several described in human ALS literature
   (see “Supplementary Information” bibliography). In the
   slower-progressing hemizygous littermates, there was significant
   slowing of progression in gait dysfunction with no adverse effect on
   gait in long-term study of wildtype littermates. The potential for
   translatability to human studies was shown by highly efficient
   transduction and ATXN2 reduction in lower motor neurons at MOIs nearly
   4 logs lower than published literature.

   These findings corroborate and elaborate on the functional and survival
   benefits seen in this model with ASO-based ATXN2 knockdown^[172]8, as
   well as findings described using a CRISPR/Cas13 system^[173]27.
   Notably, a recent ATXN2-targeting ASO was tested in human subjects with
   either sporadic ALS or ALS caused by trinucleotide repeat expansions in
   the ATXN2 gene. While the trial did not meet its target endpoints for
   either population of patients, only 21% reduction of ATXN2 protein from
   baseline levels was seen in the CSF and only a 12% reduction when
   compared to placebo (in whom ATXN2 levels also declined from baseline),
   a level that fell far short of the expected ~80% based on preclinical
   studies. We are not convinced that this is a sufficient degree of
   knockdown to expect a therapeutic effect, particularly given that
   unlike Tofersen for SOD1-ALS, the trial drug targets an indirect
   modifier of pathology in sporadic patients. The 12% ATXN2 reduction
   corresponded to the maximal safe dose of ASO, leading to complete
   cessation of the trial as the dose could not be increased. This
   highlights the need for therapies that can achieve strong target
   engagement in affected cells at safe doses. Modified AAV vectors like
   the one used in this study offer a means of improving CNS transduction
   of disease-relevant regions and cell types at relatively low
   therapeutic doses.

   From a translational standpoint, there are additional advantages of our
   approach. The first is long-term benefit after a single treatment,
   obviating the need for repeated CNS access. In the ALS patient
   population, physical limitations and the need for access to a tertiary
   care center are limiting factors in the broad application of ASO-based
   therapies that are overcome by AAV-based treatments. The second is that
   the PM-AAV9 doses used here achieved efficacy at ~1/40th the dosage of
   AAV9 used in the Cas13 mouse study, and at ~1/3600th the dosage of AAV5
   used in an AAV-mediated RNAi study in patient-derived lower motor
   neurons. This is important in translating these therapies by reducing
   both production costs and known AAV complications such as
   dose-dependent dorsal root ganglia (DRG) toxicity^[174]28. We did not
   see evidence of DRG toxicity in our study at 12 weeks either in numbers
   of DRG neurons or presence of inflammatory cells in the DRGs
   (Fig. [175]S2), nor did we see DRG-associated clasping
   phenotypes^[176]28 in PM-AAV9-treated wildtype mice even as late as 21
   months (Fig. [177]6). The third is that RNAi-based approaches have a
   strong safety and efficacy record in humans, with lipid
   nanoparticle-based delivery in current clinical use (e.g., patisiran
   for TTR amyloidosis) and AAV-based RNAi approaches in clinical trial
   (e.g., [178]NCT06100276, AAV-RNAi for SOD1-ALS; [179]NCT04120493,
   AAV-RNAi for Huntington’s disease, with recently reported positive
   Phase II data^[180]29). By contrast, CRISPR-based treatments have been
   used safely in humans only for transient delivery systems, and at this
   time RNAi may be more amenable to AAV-based expression from both an
   off-targeting and immunogenicity standpoint with permanently-on
   expression. From a further safety standpoint, a recent study
   demonstrated long-term safety of the miR30 miRNA cassette^[181]30, a
   modified version of which was used in our work^[182]31.

   A limitation of our approach and the prior Atxn2 knockdown
   studies^[183]8,[184]27 is that all three were tested in the TAR4/4
   mouse model. This mouse overexpresses wildtype human TDP-43, and was
   chosen because it is one of few models that replicate important human
   ALS findings at the pathologic and behavior level, including pTDP
   aggregates and progressive weakness respectively. Our study also
   provides the first glimpse at therapeutic benefit in the
   slower-progressing hemizygous mice, which overexpress half as much
   TDP-43 as the TAR4/4 mice. However, because patient neurons do not
   overexpress TDP-43 but rather mislocalize it, it remains unknown
   whether the Atxn2 knockdown benefit seen in our and other studies is
   unique to TDP-43 overexpression. An additional limitation is that
   TAR4/4 mice model ALS-relevant toxic gain-of-function but not loss of
   function^[185]5,[186]6. The fact that rescue is noted suggests that
   addressing toxic gain-of-function may be sufficient for therapeutic
   effect in patients, but further studies are needed to determine if
   there is efficacy of Atxn2 knockdown in non-TDP-overexpressing model
   systems.

   Transcriptomic analysis revealed several mechanisms by which ATXN2
   reduction may be exerting neuroprotection. Lipid dysmetabolism is a
   well-documented feature of ALS pathology^[187]32–[188]34, with elevated
   levels of sphingolipids in the CSF and plasma of ALS patients
   negatively correlating with prognosis^[189]35,[190]36. Atxn2 knockout
   mice present with dyslipidemia^[191]37—which has been suggested to be
   protective in ALS^[192]38—and specifically with reduced levels of
   sphingomyelin and intermediate metabolites^[193]37,[194]39. Our
   miAtxn2-responsive TAR4/4 gene signature was enriched in fatty acid and
   sphingolipid biosynthetic pathways^[195]39, suggesting a role of ATXN2
   in the regulation of sphingomyelin metabolism and a potential mechanism
   by which its reduction may be neuroprotective.

   ATXN2 reduction also normalized neuroinflammatory pathway transcripts,
   including via the upregulation of vascular endothelial growth factor A
   (Vegfa) and its modulator vasoactive intestinal peptide (Vip) towards
   wildtype levels. These are both known regulators of neuroinflammatory
   processes^[196]40,[197]41 whose protein levels are reduced in the
   cerebrospinal fluid of ALS patients^[198]22,[199]42, and multiple
   studies have demonstrated their therapeutic potential through rescue of
   ALS-associated gliosis and prolonged survival in rodent
   models^[200]43–[201]45. This mirrors the rescue of gliosis in the
   cortex and lumbar spine demonstrated by immunohistochemistry. We also
   note normalization of genes involved in oxidative stress, including the
   PI3K-Akt signaling pathway^[202]46 (Fig. [203]S3B), and rescue of
   immediate early gene transcription factor early growth response 3
   (Egr3) expression, which is also downregulated in motor neurons in
   sporadic ALS^[204]47,[205]48 and is a key regulator of neuronal
   activity-dependent transcription and DNA damage response genes
   pathways^[206]49,[207]50. Finally, miAtxn2 treatment rescued TDP-43
   induced overexpression of RNA lariat debranching enzyme Dbr1, whose
   reduction was recently shown to modify TDP-43 toxicity by decreasing
   TDP-43 cytoplasmic aggregation and increasing neuronal
   viability^[208]24. Future transcriptomic studies in other model systems
   and in TAR4 hemizygous mice will help determine the extent to which
   these results are generalizable to human disease and whether they are
   dependent on the degree of TDP pathology, respectively. An important
   limitation of this analysis is that bulk-RNA sequencing includes all
   cell types in the tissue, of which motor neurons comprise a small
   percentage. This approach enables analysis of global changes across
   cell types but precludes inference about which changes are specific to
   motor neurons. Therefore, future studies using single-cell or spatial
   transcriptomics approaches will be of benefit for parsing
   cell-autonomous changes.

   In sum, we developed an RNAi-mediated ATXN2 knockdown approach that
   shows benefit in rapid and slow progressing TDP-43 mouse models of ALS,
   which capture a range of typical human disease time-courses. We further
   identified potential mechanisms by which ATXN2 reduction leads to
   neuroprotection and demonstrated translatability in human motor
   neurons. Vectors with increased potency and specificity have the
   potential in humans to achieve the target engagement that ASOs did not,
   and further may reduce AAV dosage and off-targets, which in combination
   with RNAi can provide a translatable therapeutic approach for people
   living with sporadic ALS.

Methods

miRNA development and validation

   Artificial miRNAs were generated by polymerase extension of overlapping
   DNA oligonucleotides (IDT, Newark, NJ), purified using Qiaquick PCR
   purification kit, digested with XhoI/SpeI, and cloned into a XhoI/XbaI
   site in a miR30-like modified Pol-III expression cassette containing
   the mouse U6 promoter, a multiple cloning site, and the Pol-III
   terminator (6Ts)^[209]31. miRNA sequences were designed using siSPOTR,
   an algorithm developed to identify potent RNAi sequences with low
   unintended off-target profiles^[210]51. Additionally, the top RNAi
   sequence was further engineered for optimal strand loading of the guide
   strand and for reduced profile silencing of the passenger strand.
   miRNAs were screened for Atxn2 knockdown by transfection into mouse
   neuroblast (N2A) cell lines. For in vivo studies, the miRNA expression
   cassettes were cloned into an AAV shuttle plasmid upstream of a DNA
   stuffer sequence. The stuffer sequence was obtained by amplification
   and assembly of intronic sequences of human HTT and was designed to be
   devoid of enhancer or repressor sequences, splice activators or
   repressors, and antisense or other non-coding RNAs^[211]52. The miRNA
   expression cassette and stuffer sequence were flanked at each end by
   AAV2 145-bp inverted terminal repeat (ITR) sequences.

AAV vector selection

   AAV9, AAV-PHP.eB, and a novel peptide-modified AAV9 capsid variant,
   PM-AAV9, developed in our lab^[212]11 were each used to package eGFP
   under the universal CMV immediate enhancer/beta-actin (CAG) promoter,
   with vectors prepared by the Children’s Hospital of Philadelphia
   Research Vector Core. These were delivered through bilateral
   intracerebroventricular (ICV) injection into the brain of P1 mice at a
   dose of 2E10 VG/pup in 4uL total. Brains and spinal cords were
   harvested and sectioned 3 weeks post-injection and imaged on a Leica
   (Wetzlar, Germany) DM6000B fluorescence microscope for native
   fluorescence using LAS X software. The top performing capsid, PM-AAV9,
   was used to package the top-performing miRNA variant to generate the
   therapeutic used throughout subsequent studies. A non-targeting control
   miRNA, miCtrl, was prepared as previously described^[213]53 and was
   similarly packaged and used for comparison studies of knockdown
   efficacy in vivo.

Animals

   This study was conducted in accordance with the Guide for the Care and
   Use of Laboratory Animals (National Research Council), and all
   procedures were approved by the Children’s Hospital of Philadelphia
   Animal Care and Use Committee under protocol number 23-001358. Mice
   were housed at the CHOP Research Institute in a controlled temperature
   and humidity environment on a 12 h light/dark cycle. Food and water
   were provided ad libitum. TDP-43 transgenic mice generated by Samir
   Kumar-Singh (TAR4 strain) were purchased from JAX (strain 012836,
   B6;SJL-Tg(Thy1-TARDBP)4Singh/J). Mice were maintained on a B6/SJL
   background by crossing hemizygous mice with F1 hybrid mice from JAX
   (strain 100012, B6SJLF1/J). Hemizygous mice were crossed with one
   another to generate the TAR4/4, TAR4, and WT offspring used in our
   studies. Male and female mice were used for all studies based on
   published reports of a lack of sex-based phenotypic differences in this
   model^[214]8,[215]10. Pups were weaned at P21-P23. All litters were
   provided with DietGel Boost High Calorie Dietary Supplement
   (ClearH[2]O, Westbrook, ME) starting at P12 that was placed on the
   floor of the cage for easy access by impaired mice.

Genotyping

   Mice were genotyped at P6 via tail snip, using the common TDP primer
   5’-TGAAATCCGGGTGGTATTGG-3’ (13790, JAX, Bar Harbor, ME), wildtype TDP
   primer 5’-GGTGAGTTTAACCTTCAAGGGCT-3’ (13791, JAX), and transgene TDP
   primer 5’-AGCTTGCTAGCGGATCCAGAC-3’ (13792, JAX). Prior to weaning, mice
   were individually identified by skin-safe colored markers (Stoelting,
   Wood Dale, IL) and by ear-snips after weaning.

Injections

   Pups were anesthetized at P1 by hypothermia and their heads disinfected
   with 70% ethanol, and injections performed on a cold plate. Mice were
   injected in the bilateral cerebral ventricles with a total of 2E10
   VG/pup (2uL per hemisphere of 5E12 VG/mL), at coordinates 1/3 the
   distance between the lambda suture and the eyeball (~2 mm anterior to
   lambda and 1 mm lateral to midline) and at a depth of 2 mm, using a
   33-gauge 10 uL Hamilton syringe (Hamilton Company, Reno, NV). After
   each injection the needle was held in place for ten seconds prior to
   withdrawal. Mice were re-warmed prior to returning them to their cage.
   Mice were treated by litter to avoid administering tattoos so early in
   life, out of concern that related complications (e.g., paw damage)
   could interfere with downstream behavior assays. Vectors were prepared
   by a third party, and the person administering the injection was
   blinded to treatment.

Behavioral testing

   Behavioral tests were conducted by an investigator blinded to the
   genotype and treatment of mice. All testing was performed during the
   light period. Mice were habituated to the test room for 1 h and were
   given Clear H[2]O HydroGel if testing lasted more than 4 h. Work
   surfaces and equipment were cleaned with 70% ethanol between trials and
   then wiped with water to remove the ethanol scent. N = 10–40 mice/group
   for TAR4/4 studies in Fig. [216]2, and 11–31 mice/group for the TAR4
   studies in Fig. [217]7.

Rotarod analysis

   Mice were tested on accelerating rotarod (47600; Ugo Basile, Gemonio,
   Italy) at P17. Mice were trained for a single trial at a constant speed
   of 2 rpm for 5 min. Mice were tested for 3 trials with a starting speed
   of 2 rpm and acceleration to 20 rpm over 4 min, followed by a hold at
   20 rpm for 1 min. Latency to fall (or two consecutive rotations without
   running) was recorded for each trial, and mice were given 30 min of
   rest between trials. Mice were weighed at the end of the session.

Gait analysis

   Treated and untreated TAR4/4 mice and untreated WT littermates
   underwent gait testing on P18. Additionally, treated and untreated TAR4
   mice and treated and untreated WT littermates underwent gait testing
   every 3 months for 21 months. Mice were recorded using a Panasonic HD
   Camcorder HC-V180. Gait videos were analyzed for composite gait score
   (adapted from ref.^[218]54) consisting of hindlimb clasping, abdomen
   height, limping, foot angling, kyphosis, and tremor. Composite gait was
   analyzed by 2 independent scorers blinded to treatment and genotype.

   Hindlimb position was recorded for 30 s with mice held 1 cm from the
   tip of the tail. Mice received a score of 0 if the hindlimbs were
   consistently splayed outward. If a single hindlimb was retracted toward
   the abdomen for more than 50% of the time suspended, the mouse received
   a score of 1. If both hindlimbs were partially retracted for more than
   50% of the time suspended, the mouse received a score of 2. If both
   hindlimbs were completely retracted for more than 50% of the time
   suspended, the mouse received a score of 3. If both hindlimbs were
   completely retracted for more than 90% of the time suspended or if the
   hindlimbs were both retracted with toes clasped together, the mouse
   received a score of 4.

   Gait was tested immediately following hindlimb clasping. Mice were
   recorded walking for 150 s on a flat plastic gait walkway. If a mouse
   did not move for longer than 20 s, the tester tapped them lightly on
   the back to encourage walking. Abdomen height was scored as a 0 if the
   abdomen was raised, or a 1 if the abdomen was lowered or touching the
   ground. Limping was scored as a 0 if the mouse was not limping or a 1
   if the mouse exhibited a limp. Foot angling was scored as a 0 if the
   feet were aligned during walking or a 1 if the feet were pointed away
   from the body while walking. Mice received a kyphosis score of 0–3
   based on the criteria outlined by Guyenet et al.^[219]54. Mice received
   a tremor score between 0 and 3, with 0, 1, 2, and 3 representing no,
   slight, moderate, or severe tremor, respectively. Composite gait score
   between 0 and 13 was calculated by adding individual scores for
   hindlimb clasping, abdomen height, limping, foot angling, kyphosis, and
   tremor.

Vertical activity

   Mice underwent vertical activity testing following gait testing on P18,
   with at least 30 min of rest between tests. Mice were placed underneath
   wire mesh pencil holders (11 cm diam. x 17 cm height) and filmed for
   5 min. Videos were scored for frequency and duration of rearing and
   climbing by an independent scorer who was blinded to genotype and
   treatment. Rearing was defined as the mouse leaning against the side of
   the cup on its forelimbs, and climbing was defined as all four limbs
   off the ground.

Survival and humane endpoint determination

   All mice were observed daily for signs of weakness. Once severely
   impaired, they underwent daily righting trials consisting of laying
   them supine and allowing 30 s to right themselves, repeated up to three
   times. Euthanasia endpoint was defined as inability to right themselves
   in any of the 3 trials. Mice were also weighed 3 times a week
   throughout the study and were considered to have reached humane
   endpoint if they lost more than 25% of their maximum body weight.
   Examiners were blinded to genotype and treatment.

Tissue collection and preparation

   Mice were euthanized at P21 for RNA analyses and vector distribution
   studies, and at P22-24 for immunofluorescence (IF) and
   immunohistochemistry (IHC) studies. Mice were deeply anesthetized with
   isoflurane and transcardially perfused with ice-cold phosphate-buffered
   saline (PBS), followed by 4% paraformaldehyde perfusion for
   immunofluorescence (IF) or immunohistochemistry (IHC) studies. Brain
   was dissected out, and spinal cord was removed by hydraulic extrusion,
   and both washed in chilled PBS. For RNA or protein analyses, this was
   followed by microdissection and flash-freezing in liquid nitrogen. For
   IF and IHC studies, tissues were immersed in 4% paraformaldehyde in PBS
   at 4 °C overnight, then transferred to a 30% sucrose solution in 0.05%
   azide/PBS for 48 h prior to embedding in Tissue-Tek O.C.T. compound and
   storing at −80 °C.

RNA extraction and quantitative PCR

   RNA was extracted from relevant tissues or cells using Trizol
   (15596018; Thermo Fisher Scientific, Waltham, MA), with RNA quantity
   and quality measured using a NanoDrop 2000 (Thermo Fisher Scientific).
   Random-primer first-strand cDNA synthesis was performed using 1 μg of
   total RNA and a High Capacity cDNA Reverse Transcription Kit (Applied
   Biosystems, Waltham, MA) per manufacturer’s instructions. Assays were
   performed on a Bio-Rad (Hercules, CA) CFX384 Real Time System using
   TaqMan (Thermo Fisher Scientific) primer/probe sets specific for mouse
   Actb, mouse Atxn2, mouse Gfap, or mouse Aif1 in the case of mouse
   tissue, and human ATXN2, human TARDBP, or human GAPDH in the case of
   iPSCs (Thermo Fisher Scientific FAM TaqMan 2X Universal Master Mix,
   Life Technologies, Waltham, MA).

Protein extraction and western blot

   Frozen frontal cortex tissue samples were weighed and homogenized with
   a pestle in 10 μL/mg of extraction buffer. For ATXN2 quantification,
   RIPA buffer was used (50 mM Tris buffer [pH 8.0], 150 mM NaCl, 1%
   Triton X-100, 0.1% SDS, 0.5% sodium deoxycholate), and samples were
   incubated on ice for 30 min with vortexing every 10 min, then
   centrifuged at 16,000 x g for 20 min at 4 °C. Total protein
   concentration was quantified with the Peirce BCA Protein Assay kit
   (23225; ThermoScientific, Waltham, MA), and 30 μg of total protein was
   diluted with 7.5 μL of 4X Laemmli buffer (1610747; BioRad, Hercules,
   CA) in RIPA. For TDP-43 quantification, urea buffer was used (7 M urea,
   2 M thiourea, 4% CHAPS, 50 mM tris buffer [pH 8.5]) and samples were
   incubated at RT for 20 min with vortexing every 5 min, then centrifuged
   at 21,100 x g for 30 min at 4 °C. 10uL of supernatant was diluted with
   7.5 μL of 4X Laemmli buffer in 12.5uL of RIPA. Samples were run on
   10–20% Criterion Tris-HCl gels (3450042; BioRad) and transferred to
   PVDF membranes. Membranes were blocked in 5% milk in TBS-T (0.1%
   Tween-20) at RT for 1.5 h, followed by incubation for 2 h at RT with
   primary antibodies in 5% milk in TBS-T at the following dilutions:
   anti-TDP-43 rabbit polyclonal antibody (1:1500; 10782-2-AP;
   Proteintech, Rosemont, IL), anti-Gapdh mouse monoclonal antibody
   (1:12,000; sc-32233; Santa Cruz Biotechnology, Dallas, TX), anti-Ataxin
   2 rabbit polyclonal antibody (1:4000; 21776-1-AP; Proteintech),
   anti-Lamin B1 rabbit polyclonal antibody (1:4000; ab16048; Abcam,
   Cambridge, UK) or anti-beta Catenin rabbit polyclonal antibody (1:4000;
   ab2365; Abcam). Membranes were washed x3 with TBS-T for 5 min and
   treated with goat anti-mouse IgG (H + L) secondary antibody HRP
   (1:10,000; 31,430; Invitrogen, Waltham, MA) or goat anti-rabbit IgG
   (H + L) secondary antibody HRP (1:10,000; 31,460; Invitrogen) in 2%
   milk in TBS-T at 4 °C for 1.5 h. Membranes were washed x3 with TBS-T
   for 5 min and developed with Amersham ECL reagent (RPN2236, Cytiva,
   Marlborough, MA), and imaged on a BioRad ChemiDoc MP Imaging System.
   N = 3–4 mice/group, all from the same litter.

Histology

   Embedded tissues were cut at 16μm thickness at −20 °C using a Leica CM
   3050S Cryostat and stored at −80 °C. Primary antibodies used for tissue
   immunostaining included the following: anti-ATXN2 rabbit polyclonal
   antibody (1:500; 21776-1-AP; Proteintech); anti-TDP-43
   phospho-Ser409/410 rabbit polyclonal antibodies (1:200; 22309-1-AP;
   Proteintech | 1:2000; 80007-1-RR; Proteintech); anti-ChAT goat
   polyclonal antibody (1:400; AB144P; Millipore, Burlington, MA);
   anti-NeuN rabbit monoclonal antibody (1:1000; EPR 12763; Abcam);
   anti-Gfap rabbit polyclonal antibody (1:2000 [brain] or 1:4000 [spinal
   cord]; Z0334; Dako/Agilent, Santa Clara, CA); anti-Iba1 rabbit
   polyclonal antibody (1:2000 [brain] or 1:4000 [spinal cord]; 019-19741;
   Wako, Richmond, VA); and anti-Cleaved Caspase-3 polyclonal antibody
   (1:400; 9661; Cell Signaling, Danvers, MA). All primary antibodies were
   diluted in 2% serum (of the secondary antibody species) in 0.1%
   Triton-X100/PBS. Secondary antibodies used for immunofluorescence
   included Donkey anti-rabbit AlexaFluor 488 (A-21206), Donkey anti-mouse
   Alexafluor 488 (A-21202), and Donkey anti-goat AlexaFluor 568 (A-11057)
   polyclonal IgG antibodies at a concentration of 1:1000
   (ThermoScientific, Waltham, MA). Secondary antibodies used for
   immunohistochemistry included biotinylated goat anti-rabbit (BA-1000)
   and biotinylated horse anti-goat (BA-9500) at a concentration of 1:200
   (Vector Laboratories, Burlingame, CA).

   Embedded dorsal root ganglia were cut at 12 μm thickness and washed in
   PBS to remove O.C.T. medium. Sections were stained with Harris’
   Hematoxylin for 1 min (26041-05; Electron Microscopy Sciences),
   differentiated with 1% acid ethanol, blued in 1% sodium acetate, and
   counterstained with 0.5% Eosin Y (HT110216; Sigma-Aldrich, Millipore
   Sigma) for 1 min. Slides were cover-slipped using DPX (06522;
   Sigma-Aldrich).

   For immunofluorescence, slides were washed in PBS to remove O.C.T.
   medium and then blocked in 10% donkey serum in 0.1% Triton-X100/PBS at
   RT for 1 h. Sections were incubated in primary antibody (diluted in 2%
   donkey serum in 0.1% Triton-X100/PBS) overnight at 4 °C. Slides were
   washed in PBS and then incubated in secondary antibody (in 2% donkey
   serum in 0.1% Triton-X100/PBS). Slides were counterstained in Hoechst
   at 1:5,000 in PBS. Slides were coverslipped using Fluoromount-G
   (0100-01; SouthernBiotech, Birmingham, AL).

   For immunohistochemistry, slides were washed in PBS to remove OCT, then
   transferred to 5:1 methanol:30% hydrogen peroxide for 30 min to block
   endogenous peroxidase. Slides were triple-washed for 5 min at RT in
   dilution media (0.1% Tris-base, 0.0002% NaCl, and 0.2% Triton-X) and
   then blocked in 10% goat or donkey serum in 0.1% Triton-X100/PBS.
   Sections were then incubated in primary antibody (diluted in 2% serum
   in 0.1% Triton-X100/PBS) overnight at 4 °C in humidified chambers,
   PBS-washed, and incubated in similarly diluted secondary antibody for
   1 h at RT. Sections were then treated for one hour at RT with the
   VECTASTAIN ABC kit (PK-6100; Vector laboratories) for avidin binding
   and peroxidation, then treated with Vector ImmPACT
   3,3′-Diaminobenzidine (DAB) peroxidase substrate solution for detection
   (SK-4105; Vector laboratories). Slides were counterstained with Harris’
   hematoxylin (HHS16; Sigma-Aldrich, St Louis, MO) and dehydrated using
   an ethanol gradient and xylene prior to cover-slipping with DPX (06522;
   Sigma-Aldrich).

   Images were taken using a Leica (Wetzlar, Germany) DM6000B fluorescence
   episcope or a Leica SP8 confocal microscope and analyzed with LAS X
   software.

Quantification of DRG soma size

   Images of H&E stained DRG were processed in Fiji ImageJ for soma size
   quantification. The freehand selection tool was used to select only the
   image area containing neural soma. The image was converted to binary,
   inverted, and auto-thresholded using the minimum method, and soma were
   separated using watershed. Soma were measured using the particle
   analyzer with size 150–1000 μm^2 and circularity between 0.4 and 1.0.

Quantification of ATXN2 protein knock-down in lumbar spine

   Sections from L3 and L4 lumbar spine regions were co-stained with
   antibodies to both ChAT and ATXN2. Mean fluorescence of ATXN2 was
   quantified exclusively in ChAT-positive cells by a blinded investigator
   using ImageJ software. To remove background noise, mean fluorescence of
   extracellular background signal was subtracted from mean fluorescence
   in ChAT-positive cells. N = 3–6 mice/group, 2–10 images/mouse; graph
   points depict individual mice normalized to untreated mutant mice.

Quantification of motor neurons

   For upper motor neurons, NeuN-positive cells in the motor cortex were
   counted using ImageJ, using the Allen mouse brain atlas to identify the
   M1 primary motor cortex. N = 4–6 mice/group, 1–6 sections/mouse; graph
   points depict individual brain hemisphere images, which were normalized
   to the wildtype control from that batch of staining and analyzed using
   linear mixed models as previously described^[220]8. Images were
   converted to 8-bit.tif files and triangle threshold was applied to all
   images. The analyze particles function was used to measure cells larger
   than 50 microns and with a circularity between 0.2 and 1.0. For lower
   motor neuron quantification, lumbar regions L3-L6 were counted by two
   blinded observers and LMNs were identified by their ChAT-positive
   staining, location in the ventral horn and strict size and
   morphological criteria (with only diameter >20 µm and polygonal shape
   counted). Images shown are from the L6 level. N = 4–6 mice/group, 4–10
   sections/mouse; graph points depict individual mice normalized to the
   wildtype control from that batch of staining.

Quantification of apoptosis

   Sections were taken from L3-L6 lumbar spine regions and stained with
   antibodies to cleaved caspase 3 (CC3). Images shown are from the L3
   level where both pathology and rescue were strong. N = 3–7 mice/group
   at each lumbar level, 2–7 sections/mouse; graph points depict
   individual mice normalized to the untreated mutant mouse from each
   batch. CC3-positive cells were hand-counted by a blinded researcher and
   for each section, the number of positive cells was divided by the area
   of the section.

Quantification of neuroinflammation

   For brain, sections were selected from the M1 region as outlined above,
   while for spinal cord, sections were taken from the L6 lumbar region.
   N = 3–5 mice/group, 3-6 sections/mouse for brain, and N = 5–7
   mice/group, 1–4 sections/mouse for spinal cord; graph points depict
   individual mice normalized to the wildtype control from that batch of
   staining. Gfap- and Iba1-stained images were analyzed using ImageJ.
   Images were converted into 8-bit.tif files and a grayness threshold was
   determined for each batch based on the average wildtype stain presence.
   The “analyze particles” function was then used to quantify the total
   area of positive staining by setting the size limit to 0 and using a
   circularity of 0–1.0.

Quantification of phosphorylated TDP-43

   Sections were taken from the L6 lumbar region. Color balance threshold
   was applied to each image, with the threshold value determined by the
   average of the wildtype group. To determine the total level of
   phosphorylated TDP-43 in the ventral horn, mean fluorescence was
   quantified using ImageJ software. Variables selected for measurement
   included area, integrated intensity, and mean gray value. Multiple
   areas of the image with no fluorescence were measured as background.
   The corrected total cell fluorescence (CTCF) was calculated in excel
   using the following formula: CTCF = Integrated Density − (area of
   selected cell × mean fluorescence of background readings). N = 4–5
   mice/group, 1–5 slides/mouse; graph points depict individual mice
   normalized to the wildtype control from that batch of staining.

Whole transcriptome library preparation and sequencing

   Total RNA (1ug) was extracted from tissue using Trizol. RNA was
   quantified by Qubit^TM and RNA integrity (RIN) was assessed by RNA
   ScreenTape Analysis on an Agilent (Santa Clara, CA) 4200 TapeStation
   per manufacturer’s protocol. Samples with RIN values < 8 were excluded.
   cDNA sequencing libraries were enriched for mRNA with the NEBNext®
   Poly(A) mRNA Magnetic Isolation Module (S7590S; New England Biolabs,
   Ipswitch, MA) and prepared using NEBNext Ultra II Directional RNA
   Library Prep Kit with Sample Purification Beads (E7765; New England
   Biolabs). cDNA libraries were then indexed using the NEBNext Dual Index
   Kit (E7600S; New England Biolabs). Final cDNA libraries were analyzed
   by DNA ScreenTape Analysis on an Agilent 4200 TapeStation per the
   manufacturer’s protocol to determine library size, molar concentration,
   and purity. Libraries were indexed and pooled at concentrations of
   1.5 nM, then run on a NovaSeq 6000 S1 flow cell (Illumina) to a target
   sequencing depth of 60 million reads per sample using NovaSeq Control
   Software v1.5. The resulting sequencing reads, in fastq format, were
   aligned to the Mus musculus genome (GRCm39.104) obtained from
   ensembl.org. Alignment was performed with the STAR (STAR_2.6.0c)
   aligner.39. Read counts-per-gene values generated by STAR were used as
   the basis for differential expression analysis performed using DESeq2
   version 1.20.1 (R version 3.6.1). clusterProfiler (version 4.1.4) was
   used for all functional enrichment analyses. Data visualization was
   also performed in R and associated modalities within EnhancedVolcano
   (version 1.9.11), tidyverse (version 1.3.1), dbplyr (version 2.1.1),
   ggplot2 (version 3.3.5), pheatmap (version 1.0.12), and RColorBrewer
   (version 1.1.2). Raw and processed next-generation sequencing data have
   been deposited in the NCBI Gene Expression Omnibus (GEO) database under
   accession code [221]GSE283631.

Generation of Table [222]S1

   A literature search was conducted for each of the 23 genes listed in
   Table [223]S1, using that gene name and any relevant alternative names
   as search terms and PubMed.gov as the database. Genes relevant to human
   or other species ALS literature are included in the Supplemental
   bibliography.

iPSC maintenance

   A male C9orf72-ALS patient iPSC line (CS7VCZiALS-nxx) available through
   the Answer-ALS Project was obtained via the Cedars-Sinai iPSC
   repository ([224]https://biomanufacturing.cedars-sinai.org). The cell
   line is reported to contain >800 repeats in the C9orf72 gene; we
   confirmed presence of a large repeat expansion using the AmplideX-C9
   assay. A dox-inducible transcription factor cassette for lower motor
   neuron differentiation^[225]25 was inserted by Dr. Esteban Mazzoni at
   NYU. Karyotyping and CNV analysis were conducted prior to initiation of
   experiments and cells tested negative for mycoplasma. Cells were
   maintained in Matrigel (35460; Corning, Corning, NY)-coated plastic
   dishes in mTeSR Plus Basal Medium (100-0276; StemCell Technologies,
   Vancouver, BC) at 37 °C and 5% CO[2]. Media was supplemented with ROCK
   inhibitor Y-27632 (10 uM; 125410; Tocris, Bristol, UK) for the first
   24 h after thawing.

Neuronal differentiation

   On day 0, iPSCs were dissociated into single cells with Accutase
   (A1110501; Gibco, Grand Island, NY) and seeded at 1 × 106 cells/well in
   a 6-well plate in pre-differentiation media (PDM) (1 x KnockOut
   DMEM/F-12 [12660012; Gibco], 1 x Glutamax [35050061; Gibco], 1 x MEM
   NEAA [11140050; Gibco], 1 x N-2 Supplement [17502048; Gibco]). Cell
   media was supplemented with Y-27632 (10 uM) and doxycycline (2 μg/mL;
   D3072; Sigma-Aldrich) for 24 h after plating. On day 2, media was
   changed to PDM with retinoic acid (RA, 0.01uM) and doxycycline
   (2 μg/mL). On day 3, media was changed to PDM with RA (0.01uM),
   doxycycline (2 μg/mL), and 5-fluoro-2’-deoxyuridine/Uridine
   (FdU/Uridine, 10uM each). On day 4, cells were dissociated with
   Accutase and plated at 5 × 104 cells/well in HCl-treated glass-bottom
   24-well plates coated with poly-L-ornithine (PLO) or 2 × 105 cells/well
   in plastic 6-well plates coated with PLO. Cells were plated in
   maturation media (MM) (1 x Neurobasal Plus [A3582901; Gibco], 1 x B-27
   Plus [A3582801; Gibco], 1 x CultureOne [A3320201; Gibco], 1 x MEM NEAA,
   1 x Glutamax) supplemented with doxycycline (2 μg/mL), Y-27632 (10uM),
   brain-derived neurotrophic factor (BDNF, 10 ng/mL; 450-02; PeproTech,
   Cranbury, NJ), ciliary neurotrophic factor (CNTF, 10 ng/mL; 450-13;
   PeproTech), glial cell line-derived neurotrophic factor (GDNF,
   10 ng/mL; 450-10; PeproTech), and L-ascorbic acid (200 uM; A8960;
   Sigma-Aldrich). After 30 min, additional MM was added containing
   laminin (1 μg/mL). After replating, half media changes were performed
   on day 7 and day 10 using MM with laminin (1 μg/mL), BDNF (10 ng/mL),
   CNTF (10 ng/mL), GDNF (10 ng/mL), and ascorbic acid (200 μM).

iPSC transduction

   On day 14, neurons were transduced with PM-AAV9.eGFP or PM-AAV9.miAtxn2
   at MOIs of 1 × 104, 5 × 104, 1 × 105, or 5 × 105. Virus was diluted in
   MM with BDNF (10 ng/mL), CNTF (10 ng/mL), GDNF (10 ng/mL), and ascorbic
   acid (200 μM) and introduced in a half media change. Media was changed
   1x per week following transduction.

iPSC immunofluorescence

   On day 28, iPSC-MNs were washed with DPBS (+/+) (SH30264; Cytiva) and
   fixed in 4% paraformaldehyde (15711; Electron Microscopy Sciences,
   Hatfield, PA) and 0.25% glutaraldehyde (16019; Electron Microscopy
   Sciences) in PHEM buffer (11162; Electron Microscopy Sciences) at pH
   7.4 for 10 min. Cells were washed with DPBS (−/−) (SH30028; Cytiva)
   with Hoechst (1:2500; H3570; Invitrogen) added during the second wash.
   Cells were permeabilized and blocked in 0.25% Triton-X 100 and 3% goat
   or donkey serum in DPBS (−/−) for 60 min at RT. Anti-NeuN rabbit
   monoclonal antibody (1:100; EPR 12763; Abcam) and anti-ChAT goat
   polyclonal antibody (1:100; AB144P; Millipore) were diluted in blocking
   buffer and incubated for 60 min at RT. Goat anti-rabbit Alexafluor 568
   (A-11011) and Donkey anti-goat AlexaFluor 568 (A-11057) polyclonal IgG
   antibodies were diluted 1:1000 in blocking buffer and incubated for 1 h
   at RT. Cells were washed and stored in DPBS (−/−) until imaging.

Imaging and GFP quantification

   Brightfield and GFP images were taken on day 17, 21, 24, and 28 (3, 7,
   10, and 14 days post-transduction, respectively). No antibody-based
   amplification was used, i.e., images are of native GFP fluorescence.
   Images were taken with a Nikon Eclipse Ti2-E Microscope (Nikon,
   Melville, NY) at 10x or 40x magnification. Processing of image files
   was done in FIJI 2 (Version 2.14.0/1.54 f). The Cellpose3 model^[226]55
   was used to segment 10x brightfield images for GFP quantification. Cell
   masks were imported as ROIs into FIJI. Mean gray value was measured for
   each cell. Un-transduced wells were used to establish a threshold for
   GFP positivity. Transduction efficiency was calculated as number of GFP
   positive cells as a percent of total cells. Images from 3 separate
   wells were quantified for each MOI.

Statistics and reproducibility

   The objective of this study was to determine whether treatment with
   PM-AAV9.miAtxn2 rescues abnormalities seen in the TAR4/4 mouse model of
   sporadic ALS, including survival, strength-related behaviors,
   histopathology, and transcriptomics. Mice studied included male and
   female TAR4/4 homozygous mice, TAR4 hemizygous mice, and wildtype
   littermate controls in a series of controlled laboratory experiments.
   Mice were randomized by litter to receive PM-AAV9.miAtxn2,
   PM-AAV9.miCtrl, or virus buffer as outlined below and in the
   manuscript. Treatments were prepared by an independent investigator not
   involved in the execution of the studies. Injections, all experiments,
   and all analyses were conducted in a blinded fashion, including the
   investigators performing the injections or experiments, the
   investigators analyzing them, and all animal caretakers. Unblinding
   occurred at the end of the study.

   Sample sizes for these studies were determined based on predicted
   effect size and resultant power calculations from a previously
   published study using the same model and target transcript^[227]8.
   Treatment endpoints and a detailed analysis plan for each study were
   determined prior to initiation of the overall study, with the exception
   that in the case of lumbar neuron quantification, when total lumbar
   neuron counts were not significantly reduced in the model, additional
   analysis was done by level. These results are shown in Fig. [228]3.
   Samples or mice were added to the study if technical errors (e.g.,
   tissue damage during embedding) prevented inclusion of a sample. In
   some studies, if the magnitude of effect exceeded what was predicted
   (e.g., in cortical inflammation studies) and statistical significance
   was achieved early, collection was stopped at that point.

   All experiments described in this paper used numbers of animals and
   sections per animal as outlined in each section above and in figure
   legends. For histology, because the large number of slides required
   staining in batches, each batch contained representative animals from
   each treatment group, and results were normalized to wildtype controls
   contained in that batch. Histology data were excluded only in
   circumstances where outliers were identified by Tukey’s fences (1.5
   IQR, ~0.7%); behavior data were never excluded. Data were plotted and
   statistical tests were performed using GraphPad Prism. In all cases
   except for upper motor neuron counts (Fig. [229]S3G), dots on graphs
   represent individual mice, while for Fig. [230]S3G, graph points depict
   individual brain hemisphere images which were normalized to the
   wildtype control from that batch of staining and analyzed using linear
   mixed models as previously described^[231]8. Except for Fig. [232]S3G,
   for histology studies the value per mouse was determined by averaging
   the values of all slides for that mouse. Statistical significance was
   determined using unpaired two-tailed T-tests when comparing two groups,
   one-way ANOVA when comparing multiple groups, and two-way ANOVA split
   by mouse when comparing treatment groups over time or by spinal cord
   level, using GraphPad Prism. P-values were adjusted for multiple
   comparisons using the Holm-Sidak method. For neuronal cell counts
   (Figure [233]S3G), data was analyzed with general linear mixed modeling
   following the approach outlined by Becker et al.^[234]8 and pairwise
   multiple comparisons were made using the Holm-Sidak method. A p-value
   less than 0.05 was considered significant in all studies. Data are
   represented as mean +/− SEM for all graphs. For bulk RNA-seq analysis,
   pairwise comparisons were performed using the Wald test (DESeq2 v1.44)
   and p-values were adjusted for multiple comparisons using the
   Benjamini-Hochberg method. An adjusted p-value of <0.05 was considered
   significant, with log2 fold change cut-offs for individual comparisons
   listed within the results. Principal components analysis identified
   litter as a covariate, which was included in our model design (~Litter
   + Genotype_Treatment). For iPSC studies in Fig. [235]8E, F, no
   statistical method was used as the N was less than 3 per condition. All
   code related to the statistical analysis of differential gene
   expression is made available (see Data and materials availability
   section).

Reporting summary

   Further information on research design is available in the [236]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [237]Supplementary Information^ (5MB, pdf)
   [238]Reporting Summary^ (94.7KB, pdf)
   [239]Transparent Peer Review file^ (267.6KB, pdf)

Source data

   [240]Source Data^ (37.5MB, xlsx)

Acknowledgements