Allele-Specific RNA Silencing of Mutant Ataxin-3 Mediates Neuroprotection in a Rat Model of Machado-Joseph Disease

Recent studies have demonstrated that RNAi is a promising approach for treating autosomal dominant disorders. However, discrimination between wild-type and mutant transcripts is essential, to preserve wild-type expression and function. A single nucleotide polymorphism (SNP) is present in more than 70% of patients with Machado-Joseph disease (MJD). We investigated whether this SNP could be used to inactivate mutant ataxin-3 selectively. Lentiviral-mediated silencing of mutant human ataxin-3 was demonstrated in vitro and in a rat model of MJD in vivo. The allele-specific silencing of ataxin-3 significantly decreased the severity of the neuropathological abnormalities associated with MJD. These data demonstrate that RNAi has potential for use in MJD treatment and constitute the first proof-of-principle for allele-specific silencing in the central nervous system.


Introduction
Machado-Joseph disease (MJD) or spinocerebellar ataxia type 3 (SCA3) is a dominantly inherited disorder of the central nervous system (CNS) characterized by a wide range of clinical symptoms, including gait and limb ataxia, peripheral neuropathy, bulging eyes, ophthalmoplegia, postural instability, dystonia, amyotrophy, dysarthria, nystagmus, lingual fasciculations, facial myokymia [1] and, in some cases, parkinsonism [2]. The neuropathological features of the disease involve the afferent and efferent cerebellar systems, substantia nigra, and cranial nerve motor nuclei [1] but recent evidence suggests that the striatum is also affected [3,4]. Although initially identified in subjects of Portuguese Azorean descent [1], MJD is now the most common ataxia worldwide [5]. It is caused by an unstable CAG expansion in the coding region of the MJD1 gene encoding the ataxin-3 (ATX3) protein [6]. The expansion in the mutant allele ranges from 55 to 86 CAG repeats and the length of the polyglutamine tract is inversely correlated with age at onset of the disease [7,8]. The CAG expansion confers a toxic gain-of-function on the mutant protein, leading to the formation of neuronal intranuclear inclusions (NIIs) [9]. There is currently no available treatment.
Gene silencing by RNA interference (RNAi) has been successfully used to downregulate the expression of mutant genes and rescue phenotypes in various autosomal dominant neurodegenerative diseases, including Huntington's disease (HD) [10][11][12], familial forms of amyotrophic lateral sclerosis (ALS) [13,14] and spinocerebellar ataxia type 1 (SCA1) [15]. However, in vivo studies to date have been performed with siRNAs that do not discriminate between the wild-type and mutant alleles. The loss of function of wild-type ataxin-3, which has been shown to play a role in ubiquitin-mediated proteolysis [16], might be deleterious. Strategies based on the presence of a single nucleotide polymorphism (SNP) have been proposed to ensure discrimination between wildtype and mutant transcripts [17]. An SNP has been identified at the 39 end of the CAG tract of the ataxin-3 gene. This SNP is in linkage disequilibrium with the disease-causing expansion [18,19]. In most MJD patients, the mutant allele carries the C variant [20]. This feature provided us with an opportunity to develop and validate an allele-specific siRNA silencing strategy for the treatment of 70% of MJD patients. Similar approaches could potentially be applied to other neurodegenerative diseases.
In the present study, we used lentiviral vectors (LV) encoding short-hairpin RNAs (shRNAs) targeting this SNP, to downregulate mutant human ATX3 (MUT-ATX3) in vivo in a selective manner. We demonstrate the therapeutic efficacy and selectivity of this approach in a rat model of MJD [4].

Results
We developed two lentiviral vectors encoding siRNAs targeting an SNP (G 987 GGRC 987 GG) located at the 39 end of the CAG expansion of the ataxin-3 gene for the specific silencing of mutant human ataxin-3 mRNA but not of the wild-type ataxin-3 mRNA. The wild-type ataxin-3 gene has a G at position 987, whereas the mutant ataxin-3 has a C at this position. We therefore designed two siRNAs corresponding to this SNP. The sequences are identical with the exception of a single nucleotide (G/C polymorphism) at the center of the shRNA molecule: shAtax-MUT(C) and shAtaxWT(G) (Fig. 1A). These shRNAs were inserted in the 3'LTR of a lentiviral vector containing the H1 promoter. The lacZ reporter gene was inserted into these constructs downstream from the internal mouse phosphoglycerate kinase 1 (PGK) promoter, to facilitate the identification of transduced cells (Fig. 1B).

Selectivity and efficacy of mutant and wild-type ataxin-3 silencing in vitro
Quantitative RT-PCR analysis of 293T cells co-transfected with the shAtaxMUT(C) vector and a vector expressing the human mutant ataxin-3(C) gene (Fig. 1B) demonstrated robust mRNA degradation (Fig 2A). By contrast, ataxin-3 mRNA levels were slightly decreased by co-transfection with shAtaxMUT(C) and the wild-type ataxin-3(G) gene or with shAtaxWT(G) and the mutant ataxin-3(C) ( Fig. 2A-B), demonstrating the selectivity of the silencing. As a control, we included an siRNA (shGFP) targeting the green fluorescent protein ( Fig. 2A). RT-PCR with oligomers targeting lacZ confirmed that the decrease in ataxin-3 mRNA levels reflected siRNA efficiency rather than variations in transfection efficiency (data not shown).
We then investigated whether this silencing was associated with a decrease in wild-type and mutant ataxin-3 protein levels. Cotransfection with the mutant ataxin-3 plasmid and the specific shAtaxMUT led to significant (,70%) downregulation, as observed on western blots, whereas co-transfection with the non specific shAtaxWT had limited effect on mutant ataxin-3 protein levels ( Fig. 2C and E). Conversely, as expected, strong downregulation of wild-type ataxin-3(G) expression was observed in the presence of shAtaxWT(G), whereas shAtaxMUT(C) had a limited effect on wild-type ataxin-3(G) expression ( Fig. 2D and F).
We further assessed the functionality and efficacy of our siRNA in a more physiological situation, by analyzing the silencing of endogenous ataxin-3 in human embryonic kidney 293T cells. We transfected human 293T cells with the shAtaxWT(G), shAtax-MUT(C) or shGFP plasmid. The level of endogenous wild-type ataxin-3 was determined by western blotting and found to be lowered by transfection with shAtaxWT(G), but not with shAtaxMUT(C) or shGFP (Fig. 3). These data demonstrate the efficacy and selectivity of allele-specific ataxin-3 silencing.

Selectivity and efficacy of mutant ataxin-3 silencing in rat brain
No data are available concerning the allele-specific silencing of polyglutamine disorders in vivo. We assessed the efficacy and selectivity of this approach in the recently developed rat model of MJD [4]. The injection of lentiviral vectors encoding mutant ataxin-3 into the brain of adult rats leads to the appearance of disease reproducing the key characteristics of MJD. No such disease was induced by the injection of wild-type full-length ataxin-3. Misfolded mutant ataxin-3 proteins are detected within one week of infection and these ubiquitin-positive inclusions progressively accumulate in the nucleus of infected cells. Neuronal dysfunction, with the presence of pycnotic nuclei and a loss of expression of neuronal markers, is observed in the substantia nigra (TH, VMAT2) and striatum (DARPP-32, NeuN, TH) two months after injection, in this model.
We assessed the efficacy and selectivity of allele-specific silencing, by injecting lentiviral vectors encoding human mutant ataxin-3(C), together with the corresponding shRNA (shAtax-MUT(C)), into rat striatum. We recently demonstrated that the striatum is affected in MJD, whether in our rat model, transgenic mice or patients [4]. As a control, animals were injected with human mutant ataxin-3 and shAtaxWT(G) or shGFP. Three weeks after injection, two animals per group were killed to evaluate expression of the mutant ataxin-3 gene and of the lacZ reporter gene present in the vector encoding the small hairpin RNA.
In animals infected with lentiviral vectors encoding mutant ataxin-3(C) and the non specific shAtaxWT(G) or the shGFP control, immunohistochemical analysis of coronal rat brain sections with anti-ataxin-3 (1H9) and b-galactosidase antibodies showed that many neurons expressed both transgenes ( Fig. 4A-C, Fig 5A-L). In animals expressing mutant ataxin-3 and shAtax-MUT(C), far fewer neurons produced the pathogenic protein (Fig. 4E, Fig. 5M). The merged images (Fig. 4F, Fig. 5O) indicate that only a few mutant ataxin-3-positive-cells did not express the lacZ reporter gene present in the shAtaxMUT(C) vector (high magnification Fig. 5R). These cells corresponded to neurons not co-infected with both vectors, which were therefore not treated with the siRNA.
We assessed the efficacy of this approach further, by carrying out a, histological evaluation at two months, at which time MJD pathology was severe [4]. b-galactosidase staining indicated a similar transduction efficiency for all groups eight weeks postinjection ( Fig. 6A-C). In animals injected with mutant ataxin-3, we  Thus, allele-specific silencing efficiently and selectively inhibits human mutant ataxin-3 production, greatly decreasing the formation of disease-associated inclusions and neuronal dysfunction in vivo.

Discussion
We show here that allele-specific silencing of human mutant ataxin-3 is effective and selective in vivo. Autosomal dominant CNS diseases are good candidates for RNAi therapy, and Machado-Joseph disease is of particular interest because the presence of an SNP is linked to the disease. Other potential disorders suitable for allele-specific treatment include dystonia, in which a three-base pair deletion has been identified in the mutant torsinA gene product [21] and point mutations in the superoxide dismutase gene have been implicated in familial forms of amyotrophic lateral sclerosis (ALS) [22]. These diseases provide unique opportunities to silence the mutant gene product while preserving the wild-type protein.
It has been shown in vitro that targeting the CAG tract of ataxin-3 abolishes the expression not only of the mutant allele, but also of the wild-type allele [17]. Alternatively, the mutant MJD allele could be specifically targeted by making use of three single nucleotide polymorphisms present in the human population: C 987 GG/G 987 GG (Arg to Gly), A 669 TG/G 669 TG (Met to Val) and TAA 1118 /TAC 1118 (Stop to Tyr). Analysis of these polymorphic alleles showed there to be eight different haplotypes. However, the A 669 -C 987 -A 1118 haplotype was found in 72% of MJD patients from 249 families from different countries, but in only 2% of the normal population [20]. This allele is present in about half of all Portuguese patients [20] and is associated exclusively (100%) with mutant polyQ expansions in Japanese MJD patients [23]. It may therefore be possible to treat a large proportion of MJD patients with a single shRNA specific for the mutant allele.
We developed two lentiviral vectors encoding previously described shRNA sequences [17] specific for the C 987 GG (MUT) or G 987 GG (WT) alleles. These shRNAs were not produced according to available rules for the design of optimal siRNAs [24]. Instead, their design was based on the presence of the polymorphism, this polymorphic sequence being placed at the center of the shRNA. These shRNAs had been shown to function correctly in vitro, but not within the context of lentiviral vectors (24). It was therefore difficult to predict the results, particularly in vivo. Our data clearly demonstrate that shAtaxMUT selectively and efficiently silences the human mutant ataxin-3 gene. In vivo, we observed a significant decrease in the formation and accumulation of ataxin-3-positive aggregates (,50%) and preservation of DARPP-32 neuronal marker expression (,70%).
The selectivity of the approach was demonstrated in vitro, by the preservation of wild-type ataxin-3 expression upon shAtaxMUT co-expression, and in vivo, by the limited effect of shAtaxWT on mutant ataxin-3 gene expression. It should be noted that shAtaxMUT and shAtaxWT recognize only human ataxin-3 mRNAs, as the targeted region displays limited similarity of sequence to the endogenous rat ataxin-3 messenger RNA. These experiments therefore replicate allele-specific silencing as it would occur in human patients, with the preservation of wild-type ataxin-3 gene expression, an important safety measure given the role of ataxin-3 in proteolysis.
ATAX3 has been shown to act as a polyubiquitin-binding protein, linking its normal function to protein surveillance pathways [25]. Ataxin-3 may recruit polyubiquitinated substrates through its UIM domains, promoting cleavage via its Josephin domain, leading to proteasomal degradation [26][27][28] The potential role of ATAX3 in proteolysis might, upon silencing, impair neuronal functions and viability. No particular phenotype of neuronal degeneration was observed in MJD knockout (KO) animals, but an increase in protein ubiquitination was reported [29]. Significant dysregulation of genes involved in the ubiquitinproteasome pathway, structure/motility, and signal transduction have been reported in KO C. elegans [30]. Furthermore, ataxin-3 silencing may be well tolerated in wild-type animals, whereas similar silencing in the context of MJD may have unfavorable effects on disease progression. Studies in Drosophila have shown that wild-type ataxin-3 decreases the neurotoxicity of mutant ataxin-3 via a mechanism involving ubiquitin and the proteasome [31]. This protective role of wild-type ataxin-3 in MJD would not be perceptible in a non pathological context. Consistent with these findings, homozygous SCA3 patients with two mutant alleles present a more severe disease phenotype than patients with a single mutant allele [32,33]. It remains unclear whether adult  shAtaxMUT specifically silences MUT ATX3, promoting the clearance of MUT ATX3-positive aggregates (E and J), whereas shAtaxWT has almost no effect on MUT ATX3 expression (B and G). A considerable loss of DARPP-32-immunoreactivity is observed in rat striatum co-infected with MUT ATX3 and shAtaxWT (H and the merged image I), whereas no DARPP-32 downregulation is observed in rat striatum co-infected with MUT ATX3 and shAtaxMUT (K and the merged image L), suggesting neuroprotection. The adult rats were co-injected bilaterally in the striatum with MUT ATX3 and the shAtaxWT or shAtaxMUT vectors (n = 2) and were killed three weeks later. All the pictures were taken around the injection site. doi:10.1371/journal.pone.0003341.g004 neurons can tolerate partial non selective endogenous ataxin-3 downregulation. As a precautionary measure, it is therefore advisable to try to preserve wild-type ataxin-3 expression in gene silencing approaches.
Our data suggest that the use of an allele-specific shRNA efficiently decreases both mutant ataxin-3 production and MJDassociated neuropathological symptoms. A recent report implicated RNA toxicity in MJD pathology, providing further support for treatment at RNA level [34]. However, the clinical development of siRNA-based treatments for MJD still faces several hurdles. Studies are currently underway to demonstrate the safety and long-term efficacy of the approach, to determine which brain areas should be targeted for therapeutic benefit, and to validate a delivery system for administration to large areas of the brain. The recent incorporation of lentiviral vectors into clinical practice for the treatment of Parkinson's disease has, however, opened up new opportunities for the potential treatment of fatal and incurable diseases such as MJD.
In conclusion, we have generated lentiviral vectors markedly improving MJD-associated neuropathological signs in vivo. These data suggest that it is feasible to treat MJD patients by the selective knockdown of mutant ataxin-3.

Engineering of shRNAs plasmids
We used two shRNAs discriminating between wild-type and mutant human ataxin-3 mRNA molecules. We engineered a small hairpin RNA specifically targeting the single nucleotide polymorphism (G 987 GGRC 987 GG). A shRNA targeting the Green  shGFP(CTAGTTTCCAAAAAAAGCTGACCCTGAAGTTCA-TCTCTTGAATGAACTTCAGGGTCAGCTTGGGGATCTG-TGGTCTCATACAGAAC). Each of these oligomers and the primer H1-3F (CACCGAACGCTGACGTCATCAACCCG) were used for PCR with the pBC-H1 plasmid (pBC plasmid; Stratagene, Amsterdam, The Netherlands) containing the H1 promoter (Genbank: X16612, nucleotides 146-366) as a template. The PCR product was inserted into the pENTR/D-TOPO vector (Invitrogen, Cergy Pontoise, France). The H1-shRNA cassette was then transferred, with the LR clonase recombination system, into the SIN-cPPT-PGK-nls-LacZ-LTR-TRE gateway vector (SIN-CWP-nlsLacZ-TRE-Gateway), which contains a tetracyclineregulated operator upstream from the H1 promoter in the 39LTR. A LacZ reporter gene was also inserted downstream from the PGK promoter, facilitating the identification of infected neurons.
Cell culture and transfection. HEK 293T cells were cultured in DMEM (Gibco, Paisley, Scotland, UK) supplemented with 10% fetal bovine serum (FBS, Gibco, Paisley, Scotland, UK), 2 mM L-glutamine, 4500 mg/l glucose, 25 mM HEPES, 100 U/ml penicillin, and 100 U/ml streptomycin (Gibco, Paisley, Scotland, UK) at 37uC in a 5% CO 2 /95% air atmosphere. On the day before transfection, 293T cells were plated in six-well tissue culture dishes (Costar, NY, USA) at a density of 700,000 cells per well. The cells were co-transfected by the calcium phosphate method, with SIN-W-PGK-ATX3 72Q (mutant human ataxin-3, 1 mg) or SIN-PGK-W-ATX3 27Q (wildtype ataxin-3, 1 mg) and shATX3 or shGFP (5 mg). Six hours later, the medium was removed and replaced with fresh medium. Fortyeight hours after transfection, the cell cultures were washed with cold PBS, treated with trypsin and the cells were collected by centrifugation.
Western blotting. Cells were collected by centrifugation (10006g, 10 min). The pellets were incubated on ice in lysis buffer (150 mM NaCl, 50 mM Tris-base, pH 7.4, 5 mM EDTA, 1% Triton and 0.5% protease inhibitor cocktail; Sigma) for 30 minutes, with vortexing every 10 minutes. Homogenates were centrifuged at 13,0006g for 30 minutes at 4uC and supernatants were stored at 280uC. Protein concentration was determined with the Bradford protein assay (BioRad, Munich, Germany). Protein extract (20 mg) was resolved by electrophoresis in a 7.5 % or 12% SDS-polyacrylamide gels. The proteins were transferred onto nitrocellulose membranes (Schleicher & Schuell Bioscience, Germany) in TG 106liquid concentrate buffer (Amresco, Ohio, USA) containing 192 mM glycine, 25 mM Tris-HCl, and 20% methanol. The membranes were blocked by incubation in 5% nonfat milk powder in 0.1% Tween 20 in Trisbuffered saline (T-tris-buffered saline) for 1 h at room temperature, and were then incubated overnight with the following primary antibodies diluted in blocking buffer: anti-Myc tag antibody clone 4A6 (1:1000, Upstate, Cell Signaling Solutions, NY, USA), anti-b-galactosidase antibody (1:4000; Chemicon, Temecula, CA,USA) and anti-tubulin antibody (1:4000, Sigma, loss of DARPP-32 immunoreactivity is observed in the striatum infected with MUT ATX3 and shAtaxWT (L and O) or shGFP (J and M), whereas minor DARPP-32 is observed in the striatum infected with shAtaxMUT (K and N), this downregulation being limited to the needle track area. P-R) Quantification of the effect of the different shRNAs on the absolute number (P) and mean size/surface* (Q) of MUT ATX3-positive cells (*p,0,05). R) Quantitative analysis of the DARPP-32-depleted region in the brains of rats in which the striatum was injected with MUT ATX3 and various shRNAs. The lesion volume in brains infected with shAtaxMUT and MUT ATX3 is much smaller than that in brains infected with shAtaxWT or shGFP, indicative of a neuroprotective effect conferred by the selective shAtaxMUT. Statistical significance was evaluated with Fisher's test. (* the mean size of the objects was estimated taking into account the pixels with a gray-scale level for intensity below the mean value, used as a threshold). doi:10.1371/journal.pone.0003341.g006 Saint Louis, MO, USA). Blots were washed three times in Trisbuffered saline containing 0.1% Tween 20 (TBS-T), for 20 min each, and incubated with a horseradish peroxidase (HRP)-coupled secondary biotinylated goat anti-mouse or anti-rabbit IgG antibody (1:5000; Vector Laboratories, Burlingame, CA) for 1 h at room temperature. Blots were washed three times in TBS-T, for 20 minutes each, and binding to antigens was detected by enhanced chemiluminescence (ECL+, Amersham Pharmacia Biotech, Les Ulis, France). Semi-quantitative analysis was carried out based on the optical density (OD) of scanned films (Quantity One 1-D image analysis software version 4.4; Biorad, Hercules, CA, USA). Specific ODs were normalized with respect to those for tubulin for experiments on total homogenates. The specific OD was then normalized with respect to the amount of tubulin loaded in the corresponding lane of the same gel. A partition ratio was calculated and expressed as a percentage.

In vivo injection of lentiviral vectors
Concentrated viral stocks were thawed on ice and resuspended by repeated pipetting. The rats were anesthetized as previously described [36]. Lentiviral vectors encoding human mutant ataxin-3(C) and shRNA (shAtaxMUT(C), shAtaxWT(G), or shGFP) were co-injected into the rat striatum. Viral particle content was adjusted to 200,000 ng of p24/ml. The animals received a single 5 ml injection containing both lentiviruses into each side, at the following coordinates: 0.5 mm rostral to bregma, 63 mm lateral to midline, and 5 mm ventral to the skull surface, with the mouth bar set at 0. The viral suspensions were injected at a rate of 0.2 ml/min through an automatic injector (Stoelting Co., Wood Dale, USA), the needle being left in place for an additional 5 minutes. The skin was closed using wound clips (Phymep, Paris, France). Figure 9. Allele-specific silencing of mutant human ataxin-3 prevents neurodegeneration in the adult rat striatum. Coalescence of the internal capsule of the striatum is observed after co-infection with MUT ATX3 and shGFP (A, n = 4) or shAtaxWT (D, n = 8), at 2 months, on a brightfield photomicrograph, whereas no signs of striatal shrinkage are observed following co-infection with MUT ATX3 and the specific shAtaxMUT (G, n = 7) (left column). Neurodegeneration in rats co-infected with MUT ATX3 and shGFP (B) or shAtaxWT (E) is observed two months after injection on Fluorojade B-stained sections, but not in rats co-infected with MUT ATX3 and the selective shAtaxMUT (H) (middle column). Pycnotic nuclei are visible on cresyl violet-stained sections, suggesting cell injury and striatal degeneration after brain co-infection with MUT ATX3 and the control shGFP (C) or the non specific shAtaxWT (F), in adult rats at 2 months after injection. No such nuclei are observed on sections from rats co-infected with MUT ATX3 and the specific shAtaxMUT (I) (right column). All the pictures were taken around the injection site area and show representative immunohistochemical stainings that were reproducible among the different groups. doi:10.1371/journal.pone.0003341.g009 Histological processing Tissue preparation. Three (n = 2 for shAtaxMUT and shAtaxWT) or 8 weeks (n = 7 for shAtaxMUT, n = 8 for shAtaxWT and n = 4 for shGFP) after lentivirus injection, the animals were perfused with a phosphate solution followed by fixation with 4% paraphormaldehide (PAF, Fluka, Sigma, Buchs, Switzerland) and 10% picric acid, their brains removed and the entire striatum was sliced, the sections then being stored in 48-well trays (Corning Inc., NY, USA), as previously described [37].

Evaluation of the volume of the DARPP-32-depleted region
The extent of ataxin-3 lesions in the striatum was analyzed by digitizing 10 DARPP-32-stained sections per animal (25 mm sections at 200 mm intervals), selected so as to obtain complete rostrocaudal sampling of the striatum with a slide scanner and by quantifying the area of the lesion with a semiautomated imageanalysis program (Image J, USA). Sections from throughout the entire striatum were analyzed. The area of the striatum showing a loss of DARPP-32 staining was measured for each animal, with an operator-independent macro. The volume was then estimated with the following formula: volume = d(a 1 +a 2 +a 3 …), where d is distance between serial sections (200 mm or 300 mm), and a 1 , a 2 , a 3 .are DARPP-32-depleted areas for individual serial sections. The depleted area corresponds to area with a gray-scale value lower than the mean gray-scale value of all pixels measured in the lesioned area. Data are expressed as the area of the evaluated DARPP-32-depleted region.

Cell counts and morphometric analysis of ataxin-3 inclusions
Cell counts and morphometric analysis of ataxin-3 inclusions were performed as previously described, but with minor modifications [38,39]. Coronal sections showing complete rostrocaudal sampling (1 of 12 sections or 1 of 8 sections) of the striatum were scanned with a 610 objective, using a Zeiss (Oberkochen, Germany) Axioplan 2 imaging microscope motorized for X, Y, and Z displacements and an image acquisition and analysis system (Morphostar V 6.0; IMSTAR, Paris, France). The analyzed areas of the striatum encompassed the entire region containing mutant ATX3 aggregates, as revealed by staining with the anti-ataxin-3 antibody. This area corresponded, on average, to 100 contiguous digitized images per section. Image pixels were 0.8 mm60.8 mm in size. Section lighting was similar for all acquisitions and was automatically corrected using blank images. Images were automatically segmented for the quantification of dark objects (aggregates/inclusions), using the same parameters defining light intensity threshold, size and shape object filters. With this procedure, all inclusions with an apparent cross-sectional area greater than 3 mm 2 were reliably detected. None of the objects touching one of the X or Y borders of the fields of view were counted, in any of the images. For each animal, the estimated total number of inclusions (Ne) was calculated as Ne = (Ns)/Sf, where Ns is the number of inclusions detected in all sections and Sf is the rostrocaudal sampling fraction (1/8). Inclusions in striatal neurons tended to be round (mean rotundity index.0.90), with an isotropic orientation in the striatum. We therefore corrected the number of raw cell counts using the Abercrombie factor [40]. This factor was calculated using the formula, A = N/N+h where N is section thickness and h is mean object height, estimated for each experimental group based on morphometric analysis of segmented objects (shGFP, 6.61 mm; shAtaxWT(G), 6.51 mm; shAtax-MUT(C), 6.18 mm). The corrected total number of inclusions (Nc) was calculated as Nc = A6Ne.
Flurojade B staining. We stained striatal sections with FluoroJade-B (Chemicon, Temecula, CA), an anionic fluorescein derivative that stains neurons undergoing degeneration. The sections were first washed in water and then mounted on silanecoated glass slides, dehydrated, and stained according to the supplier's manual.. Bright-field and fluorescent images were acquired digitally on an Axioskop 2 Plus microscope (Zeiss) with Axiovision software version 4.2. All photographs for comparison were taken under identical image acquisition conditions and uniform adjustments of brightness and contrast were made to all images.

Data analysis
The intensity of immunostaining is expressed as mean6SEM. Statistical analysis was performed by one-way analysis of variance (ANOVA) followed by Fisher's PLSD post hoc test (StatView 4.0, version 3.2.6; Aladdin Systems). Values of p,0.05 were considered statistically significant.