Allele-Specific Silencing of Mutant Huntingtin in Rodent Brain and Human Stem Cells

Huntington's disease (HD) is an autosomal dominant neurodegenerative disorder resulting from polyglutamine expansion in the huntingtin (HTT) protein and for which there is no cure. Although suppression of both wild type and mutant HTT expression by RNA interference is a promising therapeutic strategy, a selective silencing of mutant HTT represents the safest approach preserving WT HTT expression and functions. We developed small hairpin RNAs (shRNAs) targeting single nucleotide polymorphisms (SNP) present in the HTT gene to selectively target the disease HTT isoform. Most of these shRNAs silenced, efficiently and selectively, mutant HTT in vitro. Lentiviral-mediated infection with the shRNAs led to selective degradation of mutant HTT mRNA and prevented the apparition of neuropathology in HD rat's striatum expressing mutant HTT containing the various SNPs. In transgenic BACHD mice, the mutant HTT allele was also silenced by this approach, further demonstrating the potential for allele-specific silencing. Finally, the allele-specific silencing of mutant HTT in human embryonic stem cells was accompanied by functional recovery of the vesicular transport of BDNF along microtubules. These findings provide evidence of the therapeutic potential of allele-specific RNA interference for HD.


Introduction
HD is an autosomal dominant neurodegenerative disease that affects 1 in 10,000 adults [1]. The symptoms, which include progressive motor, psychiatric and cognitive dysfunctions, are associated with the degeneration of the major population of striatal neurons, the GABAergic spiny projection neurons. The mutation underlying HD is an expansion of a trinucleotide CAG repeat which encodes a polyglutamine (polyQ) tract in the N-terminal region of the HTT protein [2]. This mutation confers a new toxic function on the protein, in part through the production of short Nterminal fragments carrying the polyglutamine and the accumulation of misfolded HTT [3].
Gene silencing techniques, aiming to reduce intracellular levels of polyglutamine-encoding mRNA, have the potential to halt, or at least delay, the process of neuronal death at its source and are therefore promising for the treatment of polyglutamine (polyQ) diseases [4]. However, most in vivo studies have been performed with small interfering RNAs (siRNAs) that do not discriminate between the WT and mutant alleles. Several groups including ours demonstrated that non allele-specific silencing of HTT with shRNA is well tolerated up to 9 months in a lentiviral-based HD model [5] and in transgenic N171-82Q mice [6] but leads to transcriptomic changes; the functional consequences of these changes are currently unknown. Thus, an allele-specific silencing of mutant HTT is potentially the optimal solution for blocking polyQ pathogenesis.
Recently, strategies based on chemically modified singlestranded RNAs targeting the CAG expansion were developed to selectively target mutant HTT [7][8][9]. As an alternative strategy, single nucleotide polymorphisms (SNP) have been used to discriminate between WT and mutant transcripts associated with polyglutamine disorders [10][11][12]. Studies with HD patients indicate that the level of heterozygosity of a limited number of SNP is sufficient to select most HD patients, confirming the feasibility of the approach [13][14][15]. Two studies showed that targeting three SNP would be sufficient to treat 75% of their respective HD cohort [13]. Note that a large proportion of patients of European origin clusters in a single haplotype with a specific set of SNPs [16][17][18].
One challenge is that available HD animal models are not entirely appropriate for assessing the therapeutic efficacy and selectivity of shRNAs targeting SNPs. Knock-in mice do not express the human HTT gene [19] and transgenic models expressing short N-terminal fragments of human HTT (R6/2, N171-82Q) or viral-based HD models do not contain the corresponding SNPs [20,21]. Transgenic mice expressing fulllength HTT possess only one human allele [22,23], allowing analysis of efficacy but not selectivity, which is an important aspect of this strategy. Only lately, one fully humanized transgenic mouse model of HD was created, containing two human HTT alleles, but it was not available at the time we started the study [24]. Therefore, we developed new HD models based on lentiviral expression of a chimeric mutant HTT reporter system. Using these new tools and neural derivatives of HD human embryonic stem cells (hESCs), we tested the efficacy and selectivity of shRNAs targeting SNPs in exons 39, 50, 60 and 67 of the human HTT gene and functional recovery associated with this silencing.

Plasmids and lentiviral vector production
We selected four SNPs within the HTT gene: rs363125 (exon 39, A or C), rs362331 (exon 50, C or T), rs2276881 (exon 60, A or G) and rs362307 (exon 67 C or T). We designed eight smallhairpin RNA (shRNA) targeting these HTT isoforms and cloned them in lentiviral vectors (Materials and Methods S1, Table S1). To assess the efficacy and the selectivity of these shRNAs, we produced constructs encoding chimeric mutant HTT, consisting of the sequence of the first 171 amino acids of the human HTT with 82 CAG repeats and fused to a part of HTT exons containing the SNP (SIN-W-PGK-htt171-82Q-exon; Materials and Methods S1, Table S2). Each chimeric mutant HTT construct contains one SNP of interest.
Lentiviral vectors encoding the various shRNAs and the eight chimeric HTT were produced in 293T cells using a 4-plasmid system as previously described [25]. The particle content of viral batches was determined by p24 antigen ELISA (RETROtek, Gentaur, France). The stocks were stored at 280uC until use.

In vivo experimental design and animals
Two sets of in vivo experiments were carried out. First, we validated the chimeric targets and tested shRNAs in 200 g adult male Wistar rats (Iffa Credo/Charles River, Les Oncins, France). Second, BACHD mice expressing full-length mutant HTT with 97Q [23] were used to validate the shRNA therapeutic strategy. BACHD mice were genotyped to identify HTT isoform present at SNPs studied. Stereotaxic injections of the lentiviral vectors are described in the Materials and Methods S1.
The animals were housed in a temperature-controlled room and maintained on a 12 h day/night cycle. Food and water were available ad libitum. All experiments were carried out in accordance with the European Community directive (86/609/EEC) for the care and use of laboratory animals as well as the Swiss animal welfare laws under the authorization nu VD 2486 and 2487 from the Service de la consommation et des affaires vétérinaires du Canton de Vaud, Switzerland.

Laser Capture Microdissection and punches
Four weeks after co-infection of chimeric HTT construct with fully matched or mismatched shRNA, or shCtrl, the animals were sacrificed by administration of an overdose of sodium pentobarbital. Processing of the brains for laser capture microdissection and RT-qPCR were conducted as previously described with human specific primers [5] (platform profileXpert: www.profilexpert.fr). BACHD mice injected with matched shRNA and shCtrl were sacrificed. One millimeter-thick fresh brain slices were obtained using a mouse brain matrix and punches were sampled from the GFP-positive area under a fluorescent microscope and were quickly lyzed in Trizol and stored at 280uC until RNA extraction.

Quantitative real-time PCR from 293T cells or BACHD striata
Total RNA was extracted with Trizol reagent 48 hrs posttransfection of 293T cells or 4 weeks post-injection of BACHD mice. RT-qPCR was performed in triplicate with random-primed (Invitrogen, Cergy Pontoise, France) cDNAs generated from 0.3-1 mg total RNA. Quantitative PCR was carried out in a 20 ml reaction volume containing Platinum SYBR Green qPCR super Mix-UDG (Invitrogen Cergy Pontoise, France), and 10 mM of both forward and reverse primers recognizing a sequence of human HTT present in the first part of all chimeric constructs and in the BACHD human allele (sequences in Table S3) using a Realplex thermal cycler (Eppendorf). Three to 5 samples from cells and 4-10 samples from the striatum punches were subjected to RT-qPCR analyses. Values for HTT mRNA were normalized to a reference: b-actin (ACTB) or peptidyl propyl isomerase A (PPIA) and are expressed as mean percentages 6 SEM relative to the control condition.

Histological processing
Four to 8 weeks post-lentiviral injection, the animals were given an overdose of sodium pentobarbital and were transcardially perfused with a phosphate solution followed by 4% paraformaldehyde (PFA, Sigma-Aldrich, St. Louis, Missouri, United States) and 10% picric acid for fixation. The brains were processed as previously described [5] for immunohistochemistry for dopamine and cAMP-regulated phosphoprotein of a molecular mass of 32 kDa (DARPP-32), ubiquitin (Ubi) and mutant HTT (EM-48). Pictures were taken using 4x, 20x or 63x objectives on an Olympus AX70 microscope.

Quantification of DARPP-32 lesions and formation of inclusions
The loss of DARPP-32 expression was analyzed by collecting digitized images of twelve sections per animal (separated by 300 mm) with an optic bench and by quantifying the lesion areas in mm 2 with image analysis software (MCID Core 7.0, InterFocus Imaging, GE Healthcare Niagara Inc.). Lesion areas in each section were determined as regions poor in DARPP-32 staining relative to the surrounding tissue. The lesion volume for each animal is expressed in mm 3 , calculated as the sum of the total lesion area in mm 2 of all sections multiplied by the inter-section distance (300 mm). For estimation of the number of ubiquitinpositive HTT inclusions, 12 serial coronal sections of the striatum (separated by 300 mm) were scanned with a 10x objective using a Zeiss Axioplan2 imaging microscope equipped with an automated motorized stage and acquisition system (Mercator Pro V6.50, ExploraNova). All ubiquitin-positive objects with an apparent cross-sectional area comprised between 1 and 50 mm 2 were measured as previously reported [5]. The same parameters were applied for EM-48 aggregates quantification. Bravais-Pearson correlation was performed to validate the use of Ubiquitin staining as a measure of HTT aggregation (Statistica, Statsoft, Maisons-Alfort, France).
Genotypes for SNP rs362331 of Huez2.3, VUB05 and SIVF018 NSC lines were analyzed by sequencing the PCR product encompassing this SNP generated using the following primers: 59-CCCAAACGAAGGTACACGA and 59-CCTGTTGGCC-ATCTCTCACC.
For videomicroscopy analysis, NSCs were electroporated with 1 mg BDNF-mCherry, a gift from G. Banker (Oregon Health and Science University) and 4 mg SIN-CWP-GFP-TRE-H1-shCtrl (controls: shLUC or shUNIV), -sh50T or -sh50C (see Table S2). NSCs were plated on glass coverslips previously coated with Poly-L-orninthine and Laminin (Sigma, St. Louis, Missouri, United States). Live videomicroscopy was carried out seven to ten days after transfection using an imaging system as previously described [30] with little modifications (Materials and Methods S1). After videomicroscopy experiments, the neuronal cultures were processed for immunofluorescence for anti-HTT-2166 (Materials and Methods S1).

Development of SNP-specific shRNA for the treatment of HD
To distinguish the normal and disease HTT alleles in HD patients, we developed shRNAs targeting the disease isoform of heterozygous single-nucleotide polymorphisms (SNP). We produced eight shRNAs corresponding to four SNP present within the HTT gene: in exons 39, 50, 60 and 67 (Fig. 1A). The selection was based on their presence in HTT exons and a high frequency of heterozygosity in the human population (http://www.ncbi.nlm. nih.gov/snp and [13,14,16]). DNAs encoding the shRNAs targeting each SNP (shSNPs: sh39A, sh39C, sh50C, sh50T, sh60A, sh60G and sh67C, sh67T) were inserted into a lentiviral vector [5] (Fig. 1B). The SNP was localized at 10 (p10, exons 39, 60 and 67) or 11 (p11, exon 50) bases from the 59 end of the guide strand of the shRNA to ensure selective cleavage of the target sequence [31]. All the exons selected correspond to the 39 part of the HTT mRNA. As the transcript (.10 kb) is too long for the entire sequence to be inserted into lentiviral vectors, we created a chimeric HTT gene to test the efficacy and selectivity of our shSNPs. The chimeric HTT contains a N-terminal fragment of mutant HTT corresponding to the first 171 amino acids with 82 glutamines (Htt171-82Q) fused in frame to the part of the protein encoded by the exons carrying the SNP of interest ( Fig. 1A and Materials and Methods S1). A sequence encoding the HA tag was added to 39 the end to facilitate detection of the resulting fusion proteins.
We tested the ability of the lentiviral vectors encoding the shSNPs to silence their corresponding chimeric constructs. 293T cells were co-transfected with the plasmids expressing the chimeric HTT and the shRNA, and mutant HTT levels were analyzed by quantitative RT-PCR (Fig. 2a) and western blotting (Fig. 2B). The presence of fully matched shSNP was associated with robust degradation of the chimeric HTT mRNA ( Fig. 2A), with the shSNP targeting exons 39 and 67 being the most potent (around 80% of degradation of the targeted mRNA). In contrast, HTT mRNA levels were only modestly reduced (10% of silencing in most cases) following co-transfection with mismatched shSNP ( Fig. 2A). This demonstrates the capacity of this approach to discriminate between the two HTT alleles. Selectivity was however poorer for sh50T and sh67C/T, with 40-50% degradation of HTT mRNA with the mismatched shRNAs. Additional experiments (n = 3) were performed to demonstrate the correlation between the HTT mRNA and protein levels ( Figure S1). HTT protein levels were decreased in all matched and, to a lower extent, in most mismatched conditions. A strong correlation was observed between the HTT mRNA and protein levels, in agreement with published studies [32,33].
Effective allele-specific silencing of mutant HTT in the HD rat model To evaluate the therapeutic efficacy of this approach in vivo, we needed an appropriate model expressing the SNPs of interest. We injected the various chimeric Htt171-82Q-exon vectors and the pathogenic control Htt171-82Q vector into the striatum of adult rats to induce a local expression of the disease protein, and compared the severity of HD neuropathology. Misfolded HTT labeled with the ubiquitin antibody was detected and the typical lesion with a down-regulation of DARPP-32 expression was observed for all the chimeric constructs ( Fig. 3A), consistent with previous findings [34]. This ubiquitin staining is faithfully reflecting the accumulation of misfolded HTT protein revealed with the EM48 antibody ( Figure S2). After 8 weeks, there was no apparent difference between animals injected with the Htt171-82Q or the other Htt171-82Q-exon vectors (Fig. 3A).
To assess the efficacy of allele-specific silencing, we co-injected lentiviral vectors encoding human Htt171-82Q-exon with the corresponding shSNP into the striatum of adult rats. Four weeks post-injection, using the GFP-reporter gene present on the vectors expressing the shSNP to identify the infected cells (mostly neurons as previously reported [35] [36]), the GFP-positive area was laser microdissected and the amount of HTT mRNA was determined by RT-qPCR. All shSNPs efficiently silenced HTT gene expression (Fig. 3B). In the same experiment, we then tested the ability of the shRNAs to selectively discriminate SNPs in vivo. The striatum of adult rats was co-injected with Htt171-82Q-exon and the corresponding mismatched shRNAs. The shSNPs targeting exons 50 or 60 efficiently discriminated between the two SNP variants (0 to less than 35% degradation of mismatched HTT mRNA), whereas the shSNPs targeting exon 39 or 67 did not discriminate (around 80% of degradation) (Fig. 3B). These experiments demonstrate the efficiency of all shSNPs is accompanied by good specificity for five of the eight shSNPs.

Inhibition of HD neuropathology in matched shSNPtreated lentiviral HD rat model
To further confirm the therapeutic potential of the shSNPs, we investigated the formation of HTT aggregates and striatal neuronal dysfunction. HTT inclusions are a hallmark of the pathology and a reliable marker of the severity of HD pathology. We co-injected the chimeric constructs with either the fully matched or mismatched shSNPs or a shRNA control (shCtrl). Eight weeks later, we quantified the down-regulation of DARPP-32 expression and the number of ubiquitin-positive aggregates.
DARPP-32 immunostaining (Fig. 4A) showed that the lesion size was reduced in the matched conditions, except for the sh39, but not in control and some mismatch injected areas (Fig. 4C). This staining indicated that most of the shSNPs in the matched condition inhibited the appearance of an HD neuropathology with very high efficacy. Large numbers of ubiquitin immunostained aggregates were observed in the control and most mismatch injected areas, whereas smaller numbers were found in the matched conditions (Figs. 4B and 4D). In the mismatch condition, the SNP discrimination power of sh39 and sh67 was low (not significant; Figs. 4C and 4D). Injection of the shSNP only didn't lead to striatal dysfunction based on DARPP32 staining, 8 weeks after injection (data not shown).
Pfister and collaborators have shown that for the SNP in exon 39, placing the mismatch at position 16 rather than 10 improved the discrimination ratio in a luciferase reporter system [13]. We therefore developed new vectors encoding sh39p16 and compared them with the sh39p10 constructs by evaluating in vitro mutant HTT mRNA levels in 293T cells co-transfected with the chimeric HTT and the shRNA ( Figure S3A) and in vivo lesion volumes and the numbers of ubiquitin-positive aggregates in co-injected rats ( Figure S3B and S3C). Sh39p16 efficiently silenced mutant HTT, however the selectivity was not improved by placing the mismatched base at position 16 of the shRNA.
Efficient silencing of mutant HTT in BACHD mice using matched shSNP BACHD mice express full-length human mutant HTT at physiological levels. Human mutant HTT mRNA in these mice is approximately 1.36 fold the level of endogenous WT HTT [37]; whereas mutant HTT mRNA level in animals injected with lentiviral vectors is about 25 times higher than the endogenous one [5]. We therefore used these animals to measure silencing efficacy on endogenous full-length HTT mRNA in the disease context. Genotyping showed that these mice express exon39C, exon60G and exon67C, and exon50T in agreement with Carroll et al. [15]. The four matched shSNPs and the shCtrl were injected bilaterally into the striatum of these adult and pre-symptomatic transgenic mice (we chose to inject sh39Cp16 because it is more efficacious than sh39Cp10, Figure S3). Four weeks post-injection, the mice were sacrificed and the GFP-positive area of the striatum was grossly punched out from 1 mm-thick striatal slices. HTT mRNA was measured by RT-qPCR from these punches to measure the average silencing obtained in the striatum (i.e. infected/treated and non-infected cells) and evaluate treatment efficiency, which can be expected with this gene transfer approach. The data gathered from this in vivo experiment and those presented in Figures 3-4 are therefore complementary and provide two distinct types of information on the allele-specific silencing. In striatal punches from BACHD mice, the amounts of mutant HTT mRNA (Fig. 5A) were lower (around 50%) in mice treated with each of the four shSNPs than shCtrl. There was no significant difference in the mRNA levels of the neuronal marker Neuronal Nuclei (NeuN) between any of the samples (Fig. 5B) confirming that they were comparable. To further demonstrate the selectivity of the silencing on full-length human mutant HTT, we performed one additional experiment in BACHD mice injected with sh50T, sh50C and shCtrl. As expected, a statistically significant silencing was observed with the sh50T while HTT mRNA level in the sh50C group was similar to the control group (Fig. 5C). Therefore, the capacity to discriminate the two alleles was confirmed.
Functional recovery after allele-specific silencing of mutant HTT in neurons derived from HD-hESCs To assess the silencing of endogenous HTT and the functional recovery, we took advantage of the recent isolation of human neural stem cells (NSCs) from HD-hESCs [26]. We transfected the HD-NSC line Huez2.3 homozygous for the SNP in exon 50 (T/T) and demonstrated that the silencing obtained with si50T was similar to that with the previously described siHtt6, which targets both alleles [5]. Transfecting the cells with the mismatched siRNA (si50C) had little effect on HTT expression, confirming the selectivity of the approach (Fig. 6A). Because the CAG expansion is located far from the SNP position, we then developed a new method based on SNP-specific reverse transcription and amplifi-cation of CAG repeats by PCR, coupled with size detection on a chip, to link the SNP and the mutant allele in the heterozygous HD-hESC. Using this strategy, we found that the mutant CAG expansion (46 or 44 CAG) was associated with the SNP50T in SIVF018 and VUB05 cells. Treatment of these cells with si50T and si50C significantly decreased mutant and WT HTT mRNA levels, respectively, and as expected the silencing was more pronounced with siHtt6 (Fig. 6B). We tested whether the allele-specific silencing of mutant HTT was associated with a functional recovery. We took advantage of the defect in the vesicular trafficking of the brain-derived neurotrophic factor, BDNF, previously described in primary cultures expressing mutant HTT [30]. We measured BDNF anterograde and retrograde velocity in neurons derived from WT (lines H9 and SA-01) and HD lines (Huez2.3 and SIVF018) (Fig. 7A). Direct fluorescence allowed us to select cells cotransfected with BDNF-mCherry and shRNA (GFP) (Fig. 7B).
Kymographs showing the spatial positions of BDNF vesicles through time were used to quantify the dynamics in different conditions (Figs. 7C and 7D). When differentiated into neurons, both HD-lines (Huez2.3 and SIVF018) displayed vesicular trafficking defects relative to the WT lines (H9 and SA-01) (Fig. 7A). This confirms that an expansion as limited as 44 CAG is sufficient to impair transport function [30]. Immunofluorescence staining confirmed that these cells expressed HTT (Fig. 7B). We then analyzed mCherry-BDNF transport in the SIVF018 line and Four weeks after injection, the GFP-positive area was laser-capture microdissected and HTT mRNA was assayed by RT-qPCR. Values for HTT mRNA were normalized to PPIA and are expressed as mean percentages of the value for the control condition 6 SEM (n = 6). One-way ANOVAs were conducted for each SNP. 39A: ***P,0.001; 39C: ***P,0.001; 50C: ***P,0.001; 50T: ***P,0.001; 60A: ***P,0.001; 60G: ***P,0.001; 67T: ***P,0.001; 67C: P. 0.05. Newman-Keuls Post-hoc comparison between the shCtrl groups and matched shRNA groups demonstrates significant silencing of all targeted HTT mRNAs, ***P,0.001, except for exon 67C. Post-hoc comparison between the shCtrl groups and mismatched shRNA groups revealed no significant differences for sh50T, sh60A or sh60G, a significant difference for sh67C, and highly significant differences for sh39A, sh39C, sh50C and sh67T. This comparison as the matched/mismatched post-hoc provides evidence of the different selectivities of the shSNPs. doi:10.1371/journal.pone.0099341.g003 a second HD heterozygous line, VUB05. Both anterograde and retrograde trafficking of BDNF was significantly greater in SIVF018 and VUB05 neurons when the mutant HTT was silenced with the sh50T, whereas targeting the WT HTT allele with sh50C had no effect (Figs. 7C and 7D). Thus, allele-specific silencing of mutant HTT in HD cells is associated with a functional recovery of BDNF transport. Also, silencing of the mutant allele is sufficient to alleviate the dominant effect of the mutant HTT over the WT HTT.

Discussion
Targeting heterozygous SNPs to degrade mutant HTT selectively while preserving the WT transcript is an extremely promising approach to HD treatment. However, developing and validating appropriate protocols is challenging. The frequencies and maps of heterozygous SNPs in the population have been established [13][14][15][16][17][18], and the number of SNP candidates is limited (approximately 40-50) compared to the countless sequences that could be targeted with a non allele-specific strategy. Also, the sequences flanking an SNP are not necessarily optimal for shRNA interaction, and the mismatches in SNPs have different discrimination powers that affect the selectivity of the silencing further reducing the number of possibilities. Furthermore, clinical implementation will undoubtedly require the development and validation of several products to treat the most patients.
Here, we addressed some of these issues by targeting four SNP among the most frequent in the population. We showed that the eight shRNAs covering these SNPs provide efficient and, in most cases, selective silencing of HTT in vitro and in vivo and hence are good candidates for the treatment of many HD patients.
These findings were obtained with novel models developed specifically for screening shSNPs. The chimeric reporter plasmids expressing the construct encoding human mutant HTT fused to the exons containing the various SNP allow large-scale and quick quantitative analysis of the silencing efficacy and selectivity. We reported that the allele-specific silencing demonstrated in 293T cells was reproduced in human NSC derived from HD-hESC lines involving a mutated HTT corresponding to that responsible for Figure 5. Effective and selective SNP-specific silencing in BACHD mice. (A) Four weeks after injection of constructs encoding fully matched shSNP in BACHD mice striatum, the GFP-positive area was dissected and HTT mRNA was assayed by RT-qPCR. Values for HTT mRNA were normalized to a reference peptidyl propyl isomerase A (PPIA) and are expressed as mean percentages relative to the control condition (set at 100%) 6 SEM (n = 10). One-way ANOVAs were conducted for each matched SNP for HTT 39C, 50T, 60G, and 67C: ***P,0.001. Newman-Keuls Post-hoc comparison between the shCtrl and matched shRNA groups demonstrate significant silencing of targeted human HTT mRNA. (B) Samples all showed similar NeuN gene expression (One-way ANOVAs and Newman-Keuls Post-hoc comparison test). (c) Three weeks after injection of constructs encoding shCtrl, sh50T and sh50C in BACHD mice striatum, 1 mm 3 punches of the GFP infected area were dissected and HTT mRNA was assayed by RT-qPCR with primers specific for the human HTT. Values for HTT mRNA were normalized to a reference PPIA and are expressed as mean percentages relative to the control condition (set at 100%) 6 SEM (n = 4-6). One-way ANOVAs were conducted for each groups: *P,0.05. Newman-Keuls Post-hoc comparison between the shCtrl and the sh50T confirm the silencing of human mutant HTT mRNA, whereas no statistically significant difference was observed between shCtrl and sh50C groups, demonstrating the selectivity of HTT silencing. doi:10.1371/journal.pone.0099341.g005 adult onset HD. These experiments showed that the in vitro platform based on 293T cells and HD-specific human pluripotent stem cells is suitable for evaluating the effects of allele-specific shRNA. We reported that neurons derived from HD-hESCs reproduce the defect in anterograde and retrograde BDNF vesicular transport associated with HD [30]. Allele-specific silencing of mutant HTT restored the mean velocity of BDNF vesicles to control values. Silencing of WT HTT in neurons heterozygous for SNP exon 50 did not alter trafficking along microtubules in neurites, as previously reported [38]. The different effects of treatment with sh50T and sh50C suggest that these shSNP are sufficiently potent and selective to modulate axonal vesicular transport in neurons. Importantly, HTT mRNA levels in NSC derived from hESC are closer of the values observed in patient brain samples than in HD-derived fibroblasts or lymphoblasts (which are 100x lower), increasing the potential value of this cellular model [26].
Silencing levels were similar to those obtained with shRNA targeting both alleles [5]. Even though the sequences of these shSNP were not selected according to established algorithms and criteria for optimal silencing, all of them efficiently promoted degradation of the target mRNA. In agreement with Lombardi et al. [14], we observed that the sh50C/T led to efficient degradation of the targeted mRNA (74%; fully matched sequence), while a central mismatch in the shRNA sequence abolished this silencing (21%; mismatched sequence). In contrast, antisense oligonucleotides (ASO) targeting the exon 50 were associated with poor discrimination power in human HD fibroblasts [15]. This discrepancy might be due to differences in the mode of action between shRNA and ASO, the experimental paradigm, method of administration modalities or different concentrations.
HD transgenic mice available when we initiated this study, express only one allele of the human HTT gene and only fulllength HTT genes are suitable for validating the chosen SNP, most of which map in the 39 part of the gene. We therefore used two approaches for in vivo validation. As a first screen to assess the efficacy and selectivity of HTT allele-specific silencing, we injected lentiviral vectors encoding the first 171 amino acids of the human mutant HTT fused to the exon containing the SNP into the striatum of adult rats [5,34]. Our shSNP recognized only the human HTT transcript, because the targeted regions display only limited sequence identity with the rodent HTT mRNA. These experiments therefore replicate allele-specific silencing as it would occur in human patients. Four weeks after the injection, the majority of the animals treated with fully matched shRNA showed HTT mRNA levels that were 80% lower than control values, and a concomitant reduction of the number of HTT aggregates and lesion size at 8 weeks. These experiments confirmed the efficacy  and selectivity of the shSNPs targeting Exon50C/T and Exon60G/A. The shSNPs specific for exon 39A/C and exon 67C/T were also extremely effective, although their discriminatory power was not completely preserved. Additionally, besides a strong transgene knockdown by targeting exon 39, as shown by reduction of target RNA and ubiquitin inclusions, there was a limited rescue of the lesion volume. These particular shRNAs could be partially complementary to one or several rat's transcripts and lead to their expression alteration, ultimately causing toxicity. Indeed, Jackson and collaborators [39] reported that 11 consecutive nucleotides could lead to unspecific degradation of off-target mRNA. However, in a human context, the panel of potential targets may be different because of sequence dissimilarities between rodents and human, therefore the shSNP39 could be better tolerated. Unfortunately, we are unable to verify this hypothesis since none of our human neural stem cell lines are heterozygous for SNP39; treating with either shSNP39 would degrade the WT HTT allele, which would provoke transcriptomic changes [6]. Work in Aronin's lab suggests that placing the SNP exon 39 at p16 of the shRNA might be more favorable than at p10 to discriminate the two alleles [13]. We tested this in vivo and found no improvement in selectivity but, interestingly, the DARPP-32 lesion volume was significantly smaller indicating a better tolerance of this sequence. These experiments demonstrate that most of the shSNP can selectively reduce the expression of the disease-causing HTT allele without significantly inhibiting expression of the WT allele in vitro or in vivo. To assess silencing of the endogenous full-length HTT transcript, we injected the lentiviral vectors into the striatum of BACHD mice. BACHD mice have alleles which are present in most HD patients, whereas this is not the case for YAC128 [15]. SNP in exons 39 and 50 are in linkage disequilibrium with each other and with the CAG expansion, and there are no such associations for the SNPs in exon 60 and 67 [14]. RT-qPCR analysis confirmed that all shRNAs targeting the SNPs present in these animals led to efficient degradation of the fulllength mutant human HTT. It is important to mention that the SNPs 67C/T were in their original context here (i.e. 39UTR) whereas they were in the coding sequence in the lentiviral HD rat model. Anyhow, it didn't alter its ability to trigger RNA cleavage as showed by decreased levels of HTT mRNA in both models. Nevertheless, we can't exclude that part of this effect is mediated by translational repression or mRNA decay mediated by Ago1, 3 and 4 as previously demonstrated [40,41].
The clinical implementation of this type of strategy will require the genotyping of HD patients for the selected SNPs, and, to treat a large proportion of HD patients, the development of eight therapeutic products (shSNP). A growing number of pre-clinical and clinical trials have been launched to assess the use of RNAi for the treatment of autosomal dominant diseases [42]. Both exogenous oligonucleotides and shRNA produced by the target cells are being investigated. Chemically modified siRNAs and ASO have an excellent safety profile and half-life, but the issues of crossing the blood-brain barrier to reach target cells in deep brain areas and long-term administration in general still represent major challenges for these therapeutic agents. Intraventricular infusion allows broad diffusion throughout the brain, but the dose of oligonucleotides needed to achieve efficient silencing are very high and could lead to an oversaturation of the RNAi machinery at the level of the RISC complex [9,43]. An alternative route of entry into the CNS is intraparenchymal administration of appropriate viral vectors. The design, production, and efficiency of gene transfer vectors, especially for transduction of the central nervous system (CNS), has improved remarkably, and protocols for safe transduction and long-term and robust transgene expression in the brain are now available [21,[44][45][46]. These developments have led to the initiation of several phase I/II clinical trials with adenoassociated vectors (AAV) [47]; more recently trials have started with lentiviral vectors for the treatment of adrenoleukodystrophy [48] and Parkinson's disease [49]. The issue of the safety of shRNA (off target effects) and lentiviral vectors (genotoxicity profile) needs further evaluation. Nevertheless, our study and the rapidly accumulating data evidencing the potential value of this type of approach argue for continued efforts to develop clinical applications of these techniques.  Fig. 4). The results are expressed as mean numbers of EM-48-positive aggregates 6 SEM (n = 6-7). One-way ANOVAs were conducted for each SNP; 50C and 50T: ***P,0.001. Newman-Keuls Post-hoc comparison between control and matched conditions showed highly significant prevention of aggregates formation for the matched shSNPs. For mismatched conditions, the post-hoc test was not significant compared to control condition. Matched and mismatched conditions were significantly different for the 2 SNP studied (50C: ***p,0.001; 50T: **p,0.01). (C) Correlation between the numbers of Ubiquitin-and EM-48-positive aggregates quantified from the same animals. EM-48-positive aggregates are always more numerous but the number is highly correlated to the number of ubiquitin-positive aggregates (Bravais-Pearson correlation coefficient r = 0.93; p,0.001). (TIF) Figure S3 Efficacy and selectivity of the sh39A and sh39C with SNP in p10 or p16. (A) Quantitative real-time PCR analyses showing the silencing of HTT mRNA in transfected P = 0.077; 44 to 144 events were recorded for each group. There was no significant difference between control group and treated groups for retrograde transport. doi:10.1371/journal.pone.0099341.g007

Supporting Information
293T cells co-expressing chimeric HTT and shSNP or the control shRNA. Levels of the chimeric HTT mRNA were normalized to a reference peptidyl propyl isomerase A (PPIA) and are expressed as mean percentages relative to the control condition (set at 100%) 6 SEM (n = 3). One-way ANOVAs were performed for each SNP. Newman-Keuls Post-hoc comparison between the shCtrl and shSNP groups indicated significant efficacy for the two SNP in the two positions: sh39Ap10, sh39Ap16 (*P,0.05), sh39Cp10 and sh39Cp16 (**P,0.01). The mismatched sh39p16 conditions were significantly different from the control condition (sh39Ap16 (*P, 0.05) and sh39Cp16 (**P,0.01)). (B) and (C) Eight weeks postinjection, DARPP32 and ubiquitin immunoreactivity were used to quantify the lesion volume and the number of ubiquitin-positive aggregates after sh39p10 or sh39p16 co-injection with htt171-82-exon39. The results are expressed as means 6 SEM (n = 8). Oneway ANOVAs were conducted for each SNP (**P,0.01; ***P, 0.001). Newman-Keuls post-hoc comparison between control and matched shRNA groups concerning the size of the lesion showed significant efficacy for the shSNP only in position 16 (B). Regarding the number of aggregates (C), the four constructs were effective. However, for the mismatched conditions, the post-hoc test also revealed significant differences for both stainings revealing that these shSNP are not selective.

(TIF)
Table S1 Sequences of the primers used to generate fragments of HTT gene containing each SNP. The sequences targeted by the shSNPs are in bold. The position of the SNP is in red. (DOC)