RNA Interference Applications for Machado-Joseph Disease

Abstract

Machado-Joseph disease (MJD), also named spinocerebellar ataxia type 3 (SCA3), is a dominantly inherited neurodegenerative disease caused by abnormal CAG expansions in MJD1 gene, which translate to an overexpanded tract of glutamines in the ataxin-3 (ATXN3) protein. Since the identification of the causative gene, a huge effort was made toward the development of animal models for MJD/SCA3, to increase the understanding of the molecular mechanisms underpinning disease pathogenesis, and to develop therapeutic strategies for the disease. Nevertheless, until now there are no therapies available capable of stopping or delaying the disease progression, which culminates with the death of the patients. Therefore, there is an urgent unmet need for therapeutic solutions, for which gene therapy stands out. The RNA interference (RNAi) mechanism discovery allowed the identification of small RNA molecules with the ability to regulate gene expression. For gene therapy, RNAi provided a way to silence mutant genes, which are particularly useful in dominantly inherited diseases. In the last years, several studies have focused on using RNAi molecules to target mutant ATXN3. The results showed that this could be an efficient and safe strategy for modifying MJD/SCA3 progression. Now, an additional effort must be done to translate these results into clinical trials.

1. Introduction

1.1 Machado-Joseph disease

Machado-Joseph disease (MJD), also named spinocerebellar ataxia type 3 (SCA3), is an inherited and rare neurodegenerative disease, usually with adult-onset, and is considered the most common autosomal dominant ataxia worldwide. It is part of the group of polyglutamine (polyQ) disorders. The group currently includes nine disorders — Huntington’s disease (HD), dentatorubral pallidoluysian atrophy (DRPLA), spinal and bulbar muscular atrophy (SBMA), and six different spinocerebellar ataxias (SCA1/2/3/6/7/17). These disorders are caused by abnormal expansions of the CAG trinucleotide in the coding region of the causative genes, which are translated into abnormally long polyQ tracts in the respective proteins [1, 2, 3, 4]. MJD/SCA3 was initially described in Portuguese descendants in the United States. Among SCAs, its relative frequency is higher in countries such as Brazil [5], Portugal [6, 7], China [8], Germany [9], and Japan [10, 11]. In the Azores islands, the disease has the highest prevalence registered worldwide (1:140 on the small island of Flores) [12].

The main clinical manifestation of MJD/SCA3 is progressive cerebellar ataxia – motor incoordination that can affect balance, gait, and speech [13]. Other symptoms of the disease include a pyramidal syndrome with brisk deep tendon reflexes and spasticity; peripheral neuropathy with amyotrophy; oculomotor abnormalities with nystagmus, eyelid retraction and progressive external ophthalmoplegia; facial and lingual fasciculation; and extrapyramidal signs like dystonia and rigidity [14, 15, 16, 17, 18]. The neurodegenerative process in MJD/SCA3 affects multiple neuronal systems, particularly cerebellum, brainstem, basal ganglia, spinal cord, and some cranial nerves [19, 20].

MJD/SCA3 is caused by the abnormal expansion of the CAG trinucleotide in the coding region of the MJD1 gene located on chromosome 14q32.1 [21]. The number of repeats is about 10–51 in healthy individuals and 55–87 in MJD/SCA3 patients, and there is a positive correlation between the CAG repeat number and both the severity and precocity of the symptoms, a neuropathological feature common to other polyQ disorders [22, 23, 24]. The MJD1 gene encodes ataxin-3 (ATXN3), a protein whose biochemical function seems to be associated with the UPS [25, 26]. Some studies also suggest that ATXN3 is involved in the regulation of transcription and in DNA repair mechanisms [27, 28, 29, 30]. Upon translation, the mutation results in an abnormally long polyQ tract at the carboxylic terminus of ATXN3. The mutant protein then acquires toxic properties and initiates a cascade of molecular mechanisms that culminate in neurodegeneration. An important neuropathological hallmark of MJD/SCA3 is the accumulation of neuronal insoluble aggregates containing the mutant ATXN3, predominantly in the nucleus, both inside and outside of the areas affected by neurodegeneration. That is a key feature of all polyQ diseases [31, 32, 33].

There is no cure for MJD/SCA3. However, several therapeutic strategies (Figure 1) have been developed to counteract the disease pathogenesis at different stages, namely RNA interference (RNAi)-based approaches. Considering the pathological features, the strategies involve targeting (i) mutant mRNA, (ii) mutant protein aggregation, (iii) toxic proteolytic cleavage of mutant protein, (iv) protein clearance pathways (autophagy and ubiquitin-proteasome system), (v) posttranslational modifications, (vi) transcriptional dysregulation, (vii) mitochondrial dysfunction, (viii) calcium homeostasis, and (ix) neuroprotective pathways. In quite general terms, the RNAi technology constitutes a powerful tool that allows targeting the mutant ATXN3 mRNA, thereby controlling the mutant ATXN3 protein expression [34].

Figure 1.

Schematic representation of possible MJD/SCA3 therapeutic strategies, especially detailing the non-pharmacological approaches [34].

1.2 RNA interference mechanism

All scientific discoveries have precedents, and the RNAi mechanism was no exception. In the late 1980s and early 1990s, plant biologists, trying to genetically increase the purple pigmentation of petunias’ flowers, were surprised when they noticed, not as expected, that introducing multiple extra copies of a gene that codes for “purple flowers,” via Agrobacterium, led to plants with white or variegated flowers. Somehow, the introduced extra copies of the gene had silenced both themselves and the plants’ own “purple-flower” gene [35, 36]. An explanation for these observations remained elusive until 1998 when Fire, Mello, and colleagues discovered the RNAi mechanism. The authors reported a selective and efficient silencing of a target gene using an exogenous and naked double-stranded RNA (dsRNA), in a sequence-specific manner, in Caenorhabditis elegans. Additionally, they also observed that the dsRNA was substantially more effective at silencing the gene than was the corresponding single-stranded (ssRNA) antisense strand individually [37]. Regarding the petunias’ experiments, the multiple copies of the gene introduced in the plant genome led to the generation of a homologous dsRNA, which, subsequently, mediated the silencing of both introduced and endogenous genes [38, 39].

Since then, the discovery of the RNAi mechanism, numerous studies have furthered our understanding of the RNAi mechanism, and how RNAi could be an extremely useful experimental tool for learning what genes do and for the development of potential therapeutic strategies.

1.2.1 Endogenous RNA interference mechanism

The endogenous RNAi mechanism is an evolutionarily conserved process used by cells to regulate gene expression. In general terms, the naturally occurring key molecules of the endogenous RNAi mechanism are categorized into three classes: microRNAs (miRNAs), endogenous small interfering RNAs (endo-siRNAs), and PIWI-interacting RNAs (piRNAs). The most extensively studied class is the class of the miRNAs. All those regulatory RNAs are small noncoding RNAs that have a particular homology for specific genes, and a wide variety of expression patterns, especially in a time and a cell or tissue-dependent manner [40, 41, 42].

The endogenous RNAi mechanism (Figure 2) has been deeply implicated in several aspects of animal and plant development and their regular physiological functioning, namely cell differentiation, cell proliferation, and cell death. It has been involved in the pathophysiological processes of numerous diseases, as well. Additionally, the endogenous RNAi mechanism also provides antiviral “molecular defense” response and restricts “genomic parasites,” such as transposable elements. It is known that RNAi can effectively protect hosts against viruses, by intercepting and inhibiting viral transcripts through miRNAs. RNAi can also protect cells against transposable elements, both by degrading the transcripts of transposable elements and by preventing the expression of transposable elements through heterochromatin formation [43, 44, 45, 46].

Figure 2.

Overview of the RNAi pathway depicting the two principal sub-pathways: the miRNA pathway and the siRNA pathway.

In animals, the miRNA pathway, the most notorious pathway, can be divided into multiple steps. Initially, in the nucleus, the miRNA genes are transcribed into long primary transcripts, the primary miRNAs (pri-miRNAs), that have a stem-loop structure flanked by single-stranded regions corresponding to the 5′ end (with 7-methylguanosine) and 3′ end (with the poly-A tail). The transcription is generally processed by RNA polymerase II [47, 48]. Then, the pri-miRNAs are cleaved at the opposite extremity of the loop by Drosha, a ribonuclease type III (RNase III), generating miRNA precursors, the precursor miRNAs (pre-miRNAs), which maintain the stem-loop structure but with a 2-nucleotide 3′ overhang. During the process, Drosha forms an enzymatic complex with another protein, the dsRNA-binding protein DiGeorge syndrome critical region gene 8 (DGCR8), that stabilizes the pri-miRNAs. The complex mentioned above is known as a microprocessor [49, 50, 51, 52]. Additional proteins, such as the enhancer of the rudimentary homolog (ERH), can also interact with the microprocessor, modulating its catalytic activity [53].

Still in the nucleus, the pre-miRNAs associate with the dsRNA-binding protein exportin-5 that transfers them to the cytoplasm, in the presence of the Ras-related nuclear- GTP-binding protein (Ran-GTP) [54, 55]. After the hydrolysis of GTP, the pre-miRNAs are released and intercepted by a cluster of proteins containing the RNase III Dicer and the dsRNA-binding proteins HIV-1 transactivating response (TAR)- RNA-binding protein (TRBP) and protein kinase R (PKR) activator (PACT). Dicer recognizes the 2-nucleotide 3′ overhang of the pre-miRNAs (through its Piwi-Argonaut-Zwille (PAZ) domain, an RNA-binding domain) and cleaves the loop extremity. That originates miRNAs duplexes of approximately 21–23 nucleotides in length (on each strand) harboring 2-nucleotide 3′ overhangs at both extremities. The proteins TRBP and PACT stabilize the pre-miRNAs during the process [56, 57, 58, 59].

After the previous processing step catalyzed by Dicer, the cluster of proteins containing Dicer, TRBP, and PACT provides a structural landing platform for the recruitment of another protein, argonaute (habitually argonaute-2), which associates with the miRNA duplexes, recognizing the 2-nucleotide 3′ overhangs (like Dicer, argonaute has a PAZ domain). Altogether, the proteins above are the major members of the RNA-induced silencing complex (RISC), which mediates later the messenger RNA (mRNA) silencing. The RISC becomes active when only one of the strands of the miRNA duplexes (guide strand, antisense strand, mature miRNA, or simply miRNA or miR) remains associated with argonaute. The other strand (the passenger strand, the sense strand) is removed and rapidly degraded [60, 61, 62]. If the nucleotide-pairing between the two strands of the miRNA duplexes is imperfect, it seems to be argonaute itself, through its endonucleolytic strand-dissociating activity, which dissociates the guide strand and the passenger strand (the most likely pathway since the miRNA duplexes frequently have mismatches). If the nucleotide-pairing is perfect or near-perfect, it is suggested that RNA helicases, through their strand-unwinding activity, separate the two strands [63, 64, 65]. Experimental evidence suggests that miRNA duplexes dissociation starts at the extremity with the lowest thermodynamic stability. The strand that has its 5′ end at this extremity is the one that preferentially remains associated with argonaute, and acts later as a guide strand in the mRNA silencing [66].

The final step of the miRNA pathway occurs when the miRNAs bound to argonaute selectively recognize and bind to the target mRNAs, and then the miRNAs-mRNAs macrostructures are degraded. The specificity of the mRNA recognition derives primarily from the high-level complementarity of the nucleotide sequence comprising the nucleotides 2–8 of the 5′ end of the miRNAs, known as seed region, with the 3′ untranslated region (UTR) of the mRNAs (in general, but the miRNAs can also target coding regions). Moreover, some experimental data revealed an additional nucleotide sequence in the miRNAs, termed supplementary region, comprising at least the 13–16 nucleotides, that seems to be equally important in the specificity of the mRNA recognition, given the high-level complementarity with the 3′ UTR of the mRNAs, as well. The importance increases when the complementarity of the seed region is suboptimal. As a rule, except for the localized nucleotide sequences mentioned above, the complementarity of the remaining nucleotides of the miRNAs with the 3’ UTR of the mRNAs is partial, and the occurrence of mismatches and bulges is common and naturally tolerated [67, 68, 69, 70]. After the miRNAs selectively recognize and bind to the target mRNAs, the miRNAs-mRNAs macrostructures, along with argonaute, are transported to cytoplasmic compartments called processing bodies (P-bodies) that promote its degradation. In this way, the miRNAs lead to the silencing of the target mRNAs, inhibiting the translation (or, in fact, leading to the silencing of the corresponding genes) [71, 72].

Considering the endo-siRNA pathway, in animals, there are several dsRNAs sources, which constitute endo-siRNAs precursors and ultimately originate endo-siRNAs. These sources include i) pairs of transposable element transcripts, which are formed by two transcripts from a single transposable element bi-directionally transcribed [73, 74]; ii) pairs of cis-natural antisense transcripts (cis-NATs), which are formed by two overlapping transcripts from the same genomic locus) [75]; iii) pairs of trans-NATs (also known as gene-pseudogene pairs), which are formed by two overlapping transcripts from distinct genomic loci, usually a gene mRNA and a pseudogene transcript [76]; and iv) hairpin RNA transcripts with stem-loop structure, which result from the transcription of long inverted repeats [77]. The endo-siRNAs precursors are long dsRNAs and, once in the cytoplasm, they are processed by Dicer into smaller endo-siRNAs duplexes of around 20–23 nucleotides in length (on each strand). After the RISC assembly and the subsequent strand selection, the endo-siRNAs bound to argonaute selectively recognize and bind to the target RNAs, such as transposon transcripts or endogenous mRNAs. Then, the endo-siRNAs-RNAs macrostructures are directly cleaved by argonaute in the RISC. This differs from the miRNA pathway due to the high-level complementarity of the entire nucleotide sequence of the endo-siRNAs with the target RNAs [78, 79, 80, 81].

The piRNA pathway was originally described in the Drosophila germline, and it has several features that distinguish it from miRNA and endo-siRNA pathways. First, in Drosophila, specific genomic loci, such as piRNA clusters, are transcribed into long antisense single-stranded piRNA precursors. After being transported to the cytoplasm, the endonuclease Zucchini (Zuc) (or mitochondrial phospholipase D6-MitoPLD — in mice and humans) processes the piRNA precursors into mature antisense piRNAs of approximately 25–33 nucleotides in length [82, 83, 84, 85]. Then, the antisense piRNAs are loaded into PIWI proteins, a subgroup of argonaute proteins, and depending on the PIWI protein involved, the piRNAs have different fates. piRNAs bound to aubergine (Miwi in mice and Hiwi in humans) participate in a posttranscriptional gene silencing of target RNAs in the cytoplasm, such as transposon transcripts. In contrast, piRNAs bound to PIWI (Miwi2 in mice and Hiwi2 in humans) translocate to the nucleus and, there, promote transcriptional gene silencing. As a rule, the posttranscriptional gene silencing mediated by piRNAs is a slicer-dependent mechanism that depends on catalytically active aubergine. By contrast, the transcriptional gene silencing mediated by piRNAs does not involve the cleavage of the target. Instead, it leads to a target shutdown through chromatin modifications, such as repressive histone marks and DNA methylation [82, 83, 86, 87, 88, 89].

During posttranscriptional gene silencing, additional antisense piRNAs are generated through an amplification mechanism termed the ping-pong cycle. In the cytoplasm, aubergine, through its normal slicer activity, naturally generate transposon transcript fragments that are used in the ping-pong cycle as sense piRNAs intermediates. Following a maturation process, which includes trimming to the appropriate length, the sense piRNAs intermediates originate mature sense piRNAs. Then, the sense piRNAs are loaded into the PIWI protein argonaute-3 (Mili in mice and Hili in humans) that subsequently cleaves the piRNA precursors, producing more antisense piRNAs with sequences identical, or near-identical to the original triggers. The ping-pong cycle continues with the aubergine loading once again [82, 83, 90, 91, 92].

1.2.2 RNA interference mechanism mediated by artificial RNA effector molecules

The endogenous RNAi mechanism can be artificially triggered to induce specific gene silencing by different RNA effector molecules: exogenous (exo)-siRNAs, short hairpin RNAs (shRNAs), artificial miRNAs, and miRNA mimics [93]. The exo-siRNAs are usually double-stranded molecules of around 21–23 nucleotides in length, chemically synthesized, and their guide strand has full complementarity with the target mRNAs. The delivery occurs using non-expression-based systems, including nanoparticles, such as lipid-based nanoparticles or polymer-derived nanoparticles. The exo-siRNAs are designed to mimic miRNA duplexes and enter the endogenous RNAi mechanism upon being loaded into the RISC in the cytoplasm. Following the strand selection, the passenger strand is degraded, whereas the guide strand bound to argonaute associates with specific complementary mRNAs, targeting them for direct cleavage by argonaute [94, 95]. The exo-siRNAs can be incorporated into stem-loop structures and originate shRNAs, when integrated into an artificial stem-loop, mimicking pre-miRNAs, or artificial miRNAs, when integrated into a backbone that derives from a natural pri-miRNA backbone, mimicking a pri-miRNA. Like exo-siRNAs, the guide strand of shRNAs and artificial miRNAs typically has full complementarity with the target mRNAs. Both molecules are delivered by expression-based systems, such as plasmids or viral vectors. Following viral-mediated transduction and subsequent transcription within the nucleus, the shRNAs are translocated to the cytoplasm by exportin-5. Then occurs Dicer processing and incorporation into the RISC. The next steps are like those of exo-siRNAs. In turn, the artificial miRNA genesis is most upstream and requires an additional step — Drosha processing within the nucleus [96, 97]. The miRNA mimics are synthetic double-stranded molecules comprising a guide strand that is designed to recognize and bind to a target mRNA with partial complementarity, as a mature miRNA. In fact, the guide strand typically corresponds to a naturally occurring mature miRNA, for a proper miRNA replacement and a natural mode of action silencing target mRNAs. The miRNAs mimics can be directly delivered by non-expression-based vectors or delivered by expression-based vectors. Depending on the delivery method, the miRNA mimics enter the endogenous RNAi mechanism as mentioned for the previous effector molecules [98, 99].

A possible categorization of the delivery systems is to divide them into expression-based vectors and non-expression-based vectors. The expression-based vectors are considered much more efficient, especially the viral vectors, by allowing the effector molecules to permanently silence a target gene upon one single administration. In contrast, the non-expression-based vectors are generally safer and easier to produce. Their transient nature allows an interruption of the administration [100].

Although RNAi technology has a widely recognized potential as a therapeutic strategy, its efficiency has been questioned due to unintended effects that culminate in cell dysfunction or even animal death [101]. The cytotoxic effects include i) the saturation of the endogenous RNAi processing machinery, which derives from the overexpression of the effector molecules [101, 102]; ii) the induction of the immune response, due to the activation of cellular sensors that typically recognize foreign RNA and DNA, which then leads to the production of proinflammatory cytokines and interferons [103, 104]; and iii) potential off-target effects, which in general derive from unintended interactions between the guide strand of the effector molecules and other transcripts containing complementary sequences [105, 106].

2. Therapeutic strategies for Machado-Joseph disease based on RNA interference

MJD/SCA3 is caused by a specific genetic mutation — a CAG repeat expansion — in the coding region of the MJD1 gene, similar to other polyQ disorders on their causative genes, which trigger various pathogenic mechanisms. Due to their dominant monogenic nature, RNAi technology provides a great opportunity to inhibit the mutant gene expression, at the earliest steps, over the mRNA, which could prevent the disease onset or progression. RNAi technology establishes not only a way to inhibit the toxic effects of the mutant protein, but also a way to inhibit the probable toxic effects of the mutant RNA [107]. Indeed, RNA toxicity has emerged as a crucial factor in the pathogenesis of polyQ disorders [108]. In MJD/SCA3, some studies have reported a mutant ATXN3 RNA-derived toxicity in Drosophila, Caenorhabditis elegans, and different mouse models [109, 110, 111].

Considering all this, several therapeutic strategies for MJD/SCA3 based on RNAi have been conceived, involving gene silencing mediated by exo-siRNAs and shRNAs (Table 1) and gene silencing mediated by artificial miRNAs and miRNA mimics (Table 2).

Effector molecule	Allele specificity	Expression system	Delivery system	Reference
shRNA	Non-allele-specific	HEK 293T1 cell model and lentiviral mouse model	Transfection and lentivirus-mediated transduction	[112]
exo-siRNA targeting CAG repeat expansion	Non-allele-specific	Patient-derived fibroblasts	Transfection	[113]
exo-siRNAs and shRNAs targeting G/C SNP	Allele-specific	COS-72 cell model	Transfection and adenovirus-mediated transduction	[114]
exo-siRNA targeting G/C SNP	Allele-specific	HEK 293T cell model	Transfection	[115]
shRNA targeting G/C SNP	Allele-specific	HEK 293T cell model, lentiviral rat and mouse models, and transgenic mouse model	Transfection and lentivirus-mediated transduction	[116, 117, 118]
exo-siRNA targeting G/C SNP	Allele-specific	Neuro2a cell model, lentiviral and transgenic mouse models	SNALP-mediated transfection	[119]
exo-siRNAs targeting CAG repeat expansion	Allele-specific	Patient-derived fibroblasts	Transfection	[120]
shRNA targeting CAG repeat expansion	Allele-specific	Patient-derived fibroblasts	Lentivirus-mediated transduction	[121]
ss-exo-siRNAs targeting CAG repeat expansion	Allele-specific	Patient-derived fibroblasts	Transfection	[122]

Table 1.

exo-siRNA and shRNA-mediated gene silencing approaches for MJD/SCA3.

¹

Human embryonic kidney 293T (HEK293T)

²

CV-1 simian cells transformed by an origin-defective mutant of SV40 (COS-7)

Effector molecule	Allele specificity	Expression system	Delivery system	Reference
Artificial miRNA targeting 3′ UTR of ATXN3	Non-allele-specific	Transgenic mouse model	AAV-mediated transduction	[123, 124]
Artificial miRNAs targeting exons within ATXN3	Non-allele-specific	Heterozygous knock-in mouse model	AAV-mediated transduction	[125]
miR-25 mimic	(Naturally) non-specific	HEK 293T1 cell model	Transfection	[126]
mir-9, mir-181a and mir-494 mimics	(Naturally) non-specific	HEK 293T cell model and lentiviral mouse model	Transfection and lentivirus-mediated transduction	[127]

Table 2.

Artificial miRNA and miRNA mimic-mediated gene silencing approaches for MJD/SCA3.

¹

Human embryonic kidney 293T (HEK293T)

2.1 Exogenous small interfering RNA and short hairpin RNA-mediated gene silencing

The gene silencing mediated by exo-siRNAs and shRNAs, applied to MJD/SCA3, can be divided into two distinct categories: (i) non-allele-specific gene silencing and (ii) allele-specific gene silencing. The non-allele-specific silencing constitutes the most straightforward methodology, and unselectively silences both wild-type and mutant genes. On the other hand, allele-specific silencing is a more accurate methodology that allows the selective silencing of the mutant gene. The allele distinction is particularly important when the wild-type protein is essential for cellular function. To accomplish the allele distinction, differences between both transcripts of the genes, such as single-nucleotide polymorphisms (SNPs) and the CAG repeat itself, are used to design the effector molecules [128, 129].

2.1.1 Non-allele-specific gene silencing

In a non-allele-specific gene silencing approach on MJD/SCA3, following in vitro validation of its efficacy, a shRNA designed to target both human wild-type and mutant ATXN3 proved to be safe and efficient in a lentiviral rat model. This lentiviral model was generated through the stereotaxic injection of lentivirus encoding the human mutant ATXN3 in the striatum of wild-type animals. The administration of the shRNA encoded by lentivirus led to a reduction of the human mutant ATXN3 levels and to a significant decrease of the neuropathological inclusions [112].

In another study, an exo-siRNA targeting the CAG repeat expansion and delivered by a liposome-based vector strongly reduced both mutant ATXN3 and wild-type ATXN3 protein levels, in MJD/SCA3 patient-derived fibroblasts. Furthermore, similar results were obtained for huntingtin, in a HD context, with the same exo-siRNA. It shows that a gene silencing approach targeting the CAG repeat expansion can be beneficial for different polyQ diseases [113].

Mouse and Caenorhabditis elegans knockout models for ATXN3, created to evaluate the physiological functions of this protein, showed to be viable and to have no major abnormalities [130, 131]. Nevertheless, a cellular experiment with a similar intent revealed that the absence of ATXN3 impacts the expression of a large set of genes involved in multiple signaling transduction pathways, and that may result in detrimental consequences [132]. Altogether, the experimental data above suggest that the optimal and safest gene silencing approach for MJD/SCA3 may be an allele-specific silencing of the mutant ATXN3, whenever possible, maintaining the endogenous ATXN3 functional.

2.1.2 Allele-specific gene silencing

Extensive efforts on MJD/SCA3 have been made toward allele-specific silencing of the mutant ATXN3. Several allele-specific approaches have been focused on a SNP (G⁹⁸⁷GG → C⁹⁸⁷GG) located at the 3′ end of the gene, in linkage disequilibrium and immediately following the CAG repeat expansion. The wild-type ATXN3 gene has a G at position 987, whereas the mutant ATXN3 gene has a C at that position. All the remaining sequence is identical in both genes. The G/C SNP is present in approximately 70% of MJD/SCA3 patients [133, 134, 135]. Taking into consideration the G/C SNP, Miller and colleagues designed exo-siRNAs and shRNAs encoded by plasmids or adenovirus, and then accomplished an allele-specific silencing of the mutant ATXN3 in cell cultures, with the three experimental systems. The mutant ATXN3 levels were effectively reduced, the accumulation of aggregated protein decreased and only slight effects on the wild-type ATXN3 levels were detected [114]. Similarly, Li and colleagues also created an exo-siRNA targeting the G/C SNP that led to a reduction of the mutant ATXN3 levels, with minimal impact on the wild-type ATXN3 levels, in a cellular model [115].

Later it was demonstrated in rodent models of MJD/SCA3 that it is possible to selectively and efficiently silence the mutant ATXN3 in vivo, using a shRNA delivered by lentivirus targeting the G/C SNP. Lentiviral rat and mouse models, were generated through the stereotaxic injection of lentivirus in the striatum and in the cerebellum, respectively, allowing to evaluate neuropathological features before the onset of the symptoms. In these models, a significant improvement in the associated neuropathological deficits upon silencing of the mutant ATXN3 was observed, namely less intranuclear inclusions, preservation of neuronal markers, and less neurodegeneration [116, 117]. The study using the rat lentiviral model established the first proof-of-concept for allele-specific gene silencing in the central nervous system (CNS) [116]. The allele-specific gene silencing in the cerebellum of the lentiviral mouse model also prevented the appearance of balance and motor coordination abnormalities and reduced the hyperactivity in the animals [117]. Additionally, in a severely impaired transgenic mouse model, especially useful for an evaluation after the disease onset, it was observed a rescue of the disease-associated motor disabilities and mitigation of the neuropathological deficits [118]. Moreover, considering the numerous reports of cytotoxic effects associated with the RNAi technology, the safety profile of the previously developed and tested shRNA, delivered by a lentivirus, was assessed. Upon brain injection, the stable and long-term expression of the shRNA in the striatum of wild-type mice did not lead to toxic effects. Indeed, no abnormal neuronal dysfunction, astrocytic activation, microglial activity and proinflammatory cytokines release, off-target effects or saturation of the endogenous RNAi processing machinery was detected five months after the injection of the lentiviral vectors. Similar results were obtained in human cell cultures for potential off-target effects and saturation of the endogenous RNAi processing machinery. This well-structured and complete study constitutes an important step in a future translation of gene silencing as therapy for MJD/SCA3 [136].

In a less invasive approach, following the validation of its efficacy in neuronal cells, an exo-siRNA, encapsulated in SNALPs, targeting the G/C SNP was administered intravenously in two different mouse models of MJD/SCA3 (lentiviral and transgenic mouse models). The SNALPs had covalently attached to the surface a small peptide derived from rabies virus glycoprotein (RVG-9r) that confers brain-targeting capability (ability to cross the blood-brain barrier (BBB); -RVG counterpart), as well as improves the cellular uptake and the cytosolic release (-9r, nine arginines counterpart). The administration of the exo-siRNA encapsulated in SNALPs resulted in a selective and efficient silencing of mutant ATXN3, a reduction of the neuropathological inclusions, and an improvement of the motor behavior deficits [119].

Differently, and in a G/C SNP-independent manner, some allele-specific approaches have been focused on the CAG repeat expansion. Several mismatch-containing exo-siRNAs delivered by a liposome-based vector and targeting the CAG repeat expansion successfully decreased the mutant ATXN3 protein levels, in MJD/SCA3 patient-derived fibroblasts, with minor effects on the wild-type ATXN3 levels [120]. Another study, also targeting the CAG repeat expansion, tried to develop an allele-specific approach for four polyQ diseases — MJD/SCA3, SCA7, HD, and DRPLA. The strategy demonstrated the efficacy and allele selectivity of a shRNA delivered by lentivirus in the silencing of all four mutant proteins, including the mutant ATXN3, using patient-derived fibroblasts. Additionally, an evaluation of potential off-target effects revealed that the shRNA does not induce a significant degradation of other complementary transcripts [121].

An alternative approach reported a potent and allele-selective inhibition of the mutant ATXN3 expression using chemically modified single-stranded exo-siRNAs (ss-exo-siRNAs) targeting the CAG repeat expansion, in MJD/SCA3 patient-derived fibroblasts. It was also observed that the ss-exo-siRNAs, which were delivered by a liposome-based vector, bind to argonaute (argonaute-2) and promote its recruitment to the ATXN3 mRNA, validating the involvement of the RNAi pathway in the gene silencing mediated by ss-exo-siRNAs. Besides the RNAi mechanism, a non-RNAi-related process was found to affect the gene expression after the addition of the ss-exo-siRNAs, the alternative splicing, which is a typical mode of action of the antisense oligonucleotides (ASOs). Altogether, this approach shows that chemically modified ss-exo-siRNAs have properties of conventional exo-siRNAs and ASOs. Like exo-siRNAs, the ss-exo-siRNAs can operate through the RNAi pathway, and like ASOs, the ss-exo-siRNAs are single-stranded, simplifying their synthesis and chemical modification, and can trigger the alternative splicing [122].

2.2 Artificial microRNA and microRNA mimic-mediated gene silencing

As previously mentioned, an artificial miRNA consists of an exo-siRNA and a scaffold based on a natural pri-miRNA [97]. Considering that, similarly to exo-siRNAs, the gene silencing mediated by artificial miRNAs can be divided into non-allele-specific or allele-specific gene silencing. In a non-allele-specific approach, an artificial miRNA targeting the 3′ UTR of ATXN3 mRNA and delivered by adeno-associated virus (AAVs) was able to decrease efficiently the human mutant ATXN3 expression in the cerebellum of a transgenic mouse model of MJD/SCA3 [123, 124]. It was also observed less neuronal nuclear accumulation of the mutant ATXN3. In addition, the silencing of the mutant ATXN3 resulted in a partial normalization of the endogenous miRNA steady-state levels in mice. Although the mouse wild-type ATXN3 expression has not been affected in vivo, the artificial miRNA led to a significant reduction of the human wild-type ATXN3 expression in human cell cultures [123]. Furthermore, the administration of the artificial miRNA encoded by AAVs was not neurotoxic and did not lead to signs of astrogliosis or microgliosis [124]. In another non-allele-specific gene silencing approach, artificial miRNAs were engineered to target several exons within the ATXN3 mRNA. Based on their silencing efficacy in cell cultures, the three most promising candidates encoded by AAVs were further tested in human induced pluripotent stem cell (iPSC)-derived neurons and in a heterozygous knock-in mouse model of MJD/SCA3. It was observed an efficient reduction of the mutant ATXN3 expression, as well as a reduction of the wild-type ATXN3 expression, in vivo and in vitro, respectively. No evidence for off-target effects or saturation of the endogenous RNAi processing machinery was found in human iPSC-derived neurons. In addition, the authors demonstrated in a large mammal, the minipig, that an intrathecal administration of AAVs (AAV serotype 5) can simultaneously transduce the cerebellum and brain stem, the main areas affected in MJD/SCA3 patients [125].

Some miRNA screening studies on MJD/SCA3 have shown that the expression of several miRNAs is dysregulated and closely associated with the neuropathology in SCA3/MJD [137, 138]. Since the miRNA mimics are generated to behave as endogenous miRNAs, and its guide strand is designed to correspond to a naturally occurring mature miRNA, the miRNA mimics are particularly useful to restore the function of a miRNA downregulated in a disease condition [98, 99]. That was accomplished on MJD/SCA3 using miR-25, which was found to be significantly downregulated in the serum of patients. Following transfection, the upregulation of miR-25 strongly reduced both the mutant ATXN3 and wild-type ATXN3 levels, by interacting with the 3′ UTR of the mRNA, in human cell cultures. miR-25 also decreased protein aggregation, suppressed early apoptosis, and increased cell viability [126]. A different study identified three miRNAs — mir-9, mir-181a, and mir-494 — whose expression is downregulated in human MJD/SCA3 iPSC-derived neurons and other MJD/SCA3 cellular and animal models. All of them interact with the 3′ UTR of the mRNA and are highly expressed in the brain. The reestablishment of the three miRNAs, encoded by plasmids or lentivirus, led to an efficient reduction of human mutant ATXN3 levels in human cell cultures and a lentiviral mouse model, and a decrease of the protein aggregation and neuronal dysfunction in the lentiviral mouse model. The upregulation of mir-9 and mir-181a also affected the mouse wild-type ATXN3 levels in vivo. Additionally, the authors verified that the absence of the 3’ UTR of the ATXN3, as a binding site for endogenous miRNAs, and the genetic and pharmacologic blockage of the miRNA biogenesis exacerbate the pathologic features of MJD/SCA3, in vitro and/or in vivo [127].

Another interesting but different study, based on a miRNA overexpression but not using miRNA mimics, performing an enhancer-promoter (EP) screen for modifiers through overexpression, showed that the miRNA bantam is a potent modulator in the neuropathology of MJD/SCA3 in a Drosophila model. The upregulation of the miRNA bantam suppressed the degenerative eye phenotype induced by the mutant ATXN3 toxicity. Surprisingly, miRNA bantam had no effect on the mutant ATXN3 protein levels. It was also verified that compromising the miRNA pathway/miRNA processing dramatically enhances the degeneration in the eyes of the Drosophila model and cell death in a human cell model of MJD/SCA3 [139]. Even though the miRNA bantam is not conserved between Drosophila and mammals, this study, together with the previous ones, suggests that the miRNA pathway/miRNAs have an important role in the neuropathology of MJD/SCA3.

3. Conclusions

An enormous effort was made by researchers to develop several gene silencing strategies based on RNAi molecules for MJD/SCA3. The results obtained decisively point to a huge therapeutic potential of these molecules. Overall, most of the studies showed both using allele or non-allele-specific strategies that various pathological features are mitigated, including in rodent models. Additionally, in several of these studies, the safety profile of the RNAi molecules was also assessed, corroborating their safety and increasing their therapeutic value. Nevertheless, an additional effort must be made to translate these preclinical results to human clinics, starting with their testing in clinical trials (searching on clinicalstrials.gov, there are no RNAi-based clinical trials yet). The approval in Europe and the US of an RNAi-based gene therapy for hereditary amyloid transthyretin (hATTR) amyloidosis, ONPATTRO® (patisiran), opens the way for these therapies and provides new hope for the RNAi-based gene therapies for MJD/SCA3 and other polyglutamine diseases.

Acknowledgments

We want to thank the projects and institutions funding Clévio Nóbrega laboratory, including the Portuguese Science and Technology Foundation (FCT) project (ALG-01-0145-FEDER-29480), “SeGrPolyQ” with CRESC ALGARVE 2020 cofunding, the French Muscular Dystrophy Association (AFM-Téléthon), ATAXIA UK, and the CureCSB project. JMC is supported by a Ph.D. fellowship from FCT (SFRH/BD/148760/2019).

Conflict of interest

The authors declare no conflict of interest.