Extracellular vesicle-based delivery of silencing sequences for the treatment of Machado-Joseph disease/spinocerebellar ataxia type 3

Abstract

Machado-Joseph disease (MJD)/spinocerebellar ataxia type 3 (SCA3) is the most common autosomal dominantly inherited ataxia worldwide. It is caused by an over-repetition of the trinucleotide CAG within the ATXN3 gene, which confers toxic properties to ataxin-3 (ATXN3) species. RNA interference technology has shown promising therapeutic outcomes but still lacks a non-invasive delivery method to the brain. Extracellular vesicles (EVs) emerged as promising delivery vehicles due to their capacity to deliver small nucleic acids, such as microRNAs (miRNAs). miRNAs were found to be enriched into EVs due to specific signal motifs designated as ExoMotifs. In this study, we aimed at investigating whether ExoMotifs would promote the packaging of artificial miRNAs into EVs to be used as non-invasive therapeutic delivery vehicles to treat MJD/SCA3. We found that miRNA-based silencing sequences, associated with ExoMotif GGAG and ribonucleoprotein A2B1 (hnRNPA2B1), retained the capacity to silence mutant ATXN3 (mutATXN3) and were 3-fold enriched into EVs. Bioengineered EVs containing the neuronal targeting peptide RVG on the surface significantly decreased mutATXN3 mRNA in primary cerebellar neurons from MJD YAC 84.2 and in a novel dual-luciferase MJD mouse model upon daily intranasal administration. Altogether, these findings indicate that bioengineered EVs carrying miRNA-based silencing sequences are a promising delivery vehicle for brain therapy.

Graphical abstract

De Almeida and colleagues engineered extracellular vesicles (EVs) with neuronal targeting properties carrying artificial miRNAs targeting mutant ataxin-3 (mutATXN3) mRNA. Therapeutic EVs were shown to silence mutATXN3 in vitro and downregulate mutant ataxin-3 species in an MJD/SCA3 animal model upon daily intranasal administration.

Introduction

Machado-Joseph disease (MJD), or spinocerebellar ataxia type 3 (SCA3), is the most common autosomal dominantly inherited ataxia worldwide. MJD is a polyglutamine (polyQ) disease characterized by a mutation in chromosome 14q32.1 that leads to an over-repetition of the trinucleotide CAG in the ATXN3 gene.^1,2,3 It is transcribed into a mutant mRNA and translated to a mutant ataxin-3 (mutATXN3) protein with an expanded polyQ tract. mutATXN3 protein is associated with gain of toxicity that leads to severe neuronal dysfunction over the disease course. Neurodegeneration occurs primarily in the cerebellum, pons, substantia nigra, and striatum, resulting in progressive neuronal loss.^4,5 Clinical symptoms have an adult onset and include gait and limb ataxia, ocular impairments, dystonia, and dysarthria, along with a progressive impairment of motor coordination.^6,7 MJD is an extremely debilitating disorder with no disease-modifying treatments available to cure it or delay its progression.

Our group and others have shown promising results in alleviating MJD in animal models upon direct intracranial injection of lentiviral vectors (LVs) encoding engineered short hairpin RNAs (shRNAs), microRNAs (miRNAs), and artificial miRNAs targeting mutATXN3 mRNA.^{8,9,10,11,12,13,14,15,16} These experiments showed that in vivo viral delivery of silencing sequences is able to downregulate the mutATXN3 mRNA and protein levels, ameliorating MJD phenotypical features. Nevertheless, direct intracranial injection of viral vectors into the brain parenchyma is an extremely invasive procedure, resulting in a circumscribed tissue transduction to some millimeters around the injection site.¹⁷ Additionally, insertional mutagenesis and immunogenicity associated with lentivirus delivery may lead to efficacy and safety concerns for clinical use.¹⁸ To address this, our group already developed less invasive strategies that rely on stable nucleic acid lipid particles (SNALPs) carrying small interfering RNAs (siRNAs) targeting mutATXN3 mRNA in the brain.¹⁹ Other non-viral strategies using extracellular vesicles (EVs) have also been shown to be a promising non-invasive method to deliver silencing sequences toward specific body tissues.^20,21,22,23

EVs are a heterogeneous group of membrane vesicles with a lipid bilayer, secreted by all cell types as a way of communicating at close and long distances and typically categorized by size and biogenesis process in exosomes, microvesicles, and apoptotic bodies.^24,25,26 Recently, some studies have described exomeres^27,28 and supermeres²⁹ as two distinct types of non-membranous extracellular nanoparticles with less than 50 nm playing biological functions.³⁰

EVs mediate the functional transfer of lipids, luminal and membrane proteins, and nucleic acids among cells, both in physiological and pathological conditions, thus playing a major role in intercellular communication.^31,32,33 Their cargo usually reflects the state of their donor cell and can be exploited both as biomarkers of disease as well as a platform to deliver targeted therapies.²⁶

EVs were found to carry DNA fragments, mRNAs, and particularly small RNAs due to their small size.³³ Among the small RNAs, it has been described that EVs are enriched in miRNAs,^32,33 small non-coding RNAs around 21 nucleotides in size that mediate post-transcriptional gene regulation of their mRNA targets, controlling translation or causing mRNA degradation.^34,35 Intriguingly, EVs carry specific subsets of miRNAs, suggesting a selective and active packaging of miRNAs during EV biogenesis. The sorting mechanism on which specific miRNAs get highly enriched into EVs is thought to be a multifactorial process depending on the presence of short sorting motifs that drive miRNAs into EVs. These sequences are called ExoMotifs, as is the case of GGAG, CCCU, GGCU, and CGGGAG sequences.^36,37,38 In the opposite direction, other sequences called CellMotifs restrain miRNAs in cells preventing their packaging into EVs.^36,38

The packaging process into EVs seems to be cell type-specific since different proteins were described to promote the incorporation of miRNAs into EVs in different cell types, such as heterogeneous nuclear ribonucleoprotein A2B1 (hnRNPA2B1) in EVs derived from T lymphocytes³⁶ and synaptotagmin-binding cytoplasmic RNA-interacting protein (SYNCRIP) in hepatocytes-derived EVs.³⁷ Recently, the study of the miRNA profile of five metabolic cell lines has suggested Alyref and Fus as RNA-binding proteins that efficiently load miRNAs into EVs.³⁸ Post-translational modifications of RNA-binding proteins were also described as a packaging trigger, acting as facilitators of the miRNA-protein binding.³⁶ Moreover, miRNA secondary and tertiary structures influence their packaging into EVs and can be stabilized by RNA-binding proteins such as Y-box protein 1.^39,40

These findings allow the design of artificial miRNAs with intrinsic EV-enrichment properties that can be exploited for therapeutic delivery of silencing sequences. Typically, silencing sequences are incorporated into EVs by electroporation.^20,41 However, it was described that this process compromises EVs integrity and leads to aggregation of siRNAs on EV surfaces, thus reducing the therapeutic efficiency of siRNA-loaded EVs in delivering their cargo to recipient cells.⁴² To overcome this limitation, endogenous machinery for the incorporation of miRNAs into EVs has already been considered as an alternative strategy to load small RNAs for therapeutic applications.⁴³

Nevertheless, using EVs for in vivo therapeutic approaches remains extremely challenging due to the lack of organ-specific targeting efficiency. In this regard, several studies have demonstrated good targeting efficiency levels in specific organs and distinct disease contexts, such as when targeting oncogenic Kras in pancreatic cancer,⁴¹ increasing dystrophin protein in muscles,²² and targeting the brain upon expression of the rabies virus glycoprotein (RVG) on EV surfaces.^20,21,22 Targeting the brain in a minimal/non-invasive way remains very demanding when considering gene therapy, primarily due to the difficulty in finding a suitable vehicle to carry the genetic material to the brain without targeting peripheral organs, such as the liver. In fact, intravenous administration has allowed the delivery of EVs to the brain upon modulation of their surface with brain-targeting peptides.^21,44,45 Similarly, intracerebrospinal fluid (intra-CSF) injections also demonstrated great potential for brain targeting since EVs are able to diffuse from CSF to the brain.⁴⁶ Another non-invasive way to direct EVs to the brain is through the intranasal route, which allows EVs to bypass the blood-brain barrier (BBB) after traversing the olfactory bulb.^47,48

In this study, we aimed at investigating whether ExoMotif GGAG would promote packaging into EVs of an engineered miRNA-based silencing sequence targeting mutATXN3 mRNA.⁴⁹ Engineered miRNAs were loaded into EVs in order to silence mutATXN3 mRNA for the treatment of MJD/SCA3. Additionally, to specifically target neurons and enable BBB crossing, the RVG peptide was inserted on EV surfaces. Finally, a packaging cell line stably producing RVG-EVs loaded with silencing sequences was generated. Engineered EVs were shown to efficiently downregulate mutATXN3 mRNA in primary cerebellar cultures of MJD YAC84.2 pups and in a new dual-luminescent MJD mouse model upon daily intranasal administrations of EVs. This system holds great promise for brain delivery and therapy.

Results

Endogenous miRNAs are extensively loaded into EVs due to the presence of ExoMotifs

To evaluate whether miRNAs previously described as containing ExoMotifs³⁶ were preferentially enriched into EVs, we tested their intrinsic loading efficacy in three different cell lines: human embryonic kidney 293 (HEK293T), KUM10, and SH-SY5Y. For that purpose, we started by optimizing a differential ultracentrifugation (dUC) protocol (Figure 1A) for EV isolation. Briefly, conditioned media were collected from cells at 80% confluence, and increasing UC forces were sequentially applied to remove cells in suspension, cell debris, and large vesicles. The supernatant was then filtered through a 0.22 μm syringe filter and ultracentrifuged at 100,000g for 2 h to pellet EVs. The pellet was then washed in cold PBS at 100,000g for 2 h to remove co-pelleted free protein and protein aggregates. EVs were characterized by western blotting to identify typical EV markers according to MISEV2018 guidelines.²⁵ The obtained EV population was shown to be enriched for Lamp-2, Alix, HSC70, and Flotilin-1, whereas the Golgi marker calnexin was absent (Figure 1B).

Figure 1.

Endogenous miRNAs with ExoMotifs are extensively loaded into EVs

(A) EVs isolation through differential ultracentrifugation (dUC). Conditioned media were centrifuged at 300g for 10 min to eliminate cells in suspension, followed by a centrifugation at 2,000g for 10 min to discard cell debris. Supernatant was then centrifuged at 16,500g for 1 h to remove large vesicles, filtered through a 0.22 μm syringe filter, and ultracentrifuged at 100,000g for 2 h to pellet EVs. The pellet was then washed in cold 1×PBS and centrifuged at 100,000g for 2 h to remove free protein and protein aggregates. (B) Characterization of HEK293T-derived EV protein markers. Western blotting of equimolar amounts of protein from cells and their derived EVs show the presence of the EV-positive markers Lamp-2 (110 kDa), Alix (100 kDa), HSC70 (70 kDa), and Flotilin-1 (48 kDa) and the absence of the negative marker calnexin (100 kDa). (C) Characterization of EVs by nanoparticle tracking analysis (NTA) and transmission electron microscopy (TEM). NTA shows a prominent peak at 110 ± 3.9 nm, corresponding to the typical EV size range. TEM shows the cupped-shaped morphology of EVs. Scale bar is 100 nm. (D) Number of particles produced per cell type during 48 h. KUM10 cells produce around 1,666 particles per cell, SH-SY5Y cells produce around 360 particles per cell, and HEK293T cells produce around 216 particles per cell during 48 h. Data are presented as means ± SEM throughout five and six independent isolations (n = 5/6). Ordinary one-way ANOVA followed by Tukey’s multiple comparisons test. (E) Levels of endogenous miRNA content in SH-SY5Y cells (black bars) and derived EVs (blue bars). miR-451 and miR-601 are enriched in EVs compared with their cells. miR-575, miR-125a-3p, miR-198, miR-887, and miR-181a are restrained in cells compared with EVs. (F) KUM10 miRNA sorting profile between cells and their derived EVs. miR-575, miR-451, and miR-601 are enriched in EVs compared with their cells. miR-125a-3p and miR-181a are restrained in cells compared with EVs. miR-198 is not detected in cells or EVs. (H) HEK293T miRNA sorting profile between cells and their derived EVs. miR-575, miR451, miR-198, miR-601, and miR-887 are enriched in EVs compared with their cells. miR-125a-3p and miR-181a are restrained in cells compared with EVs. Data are presented as means ± SEM throughout four independent experiments (n = 4). Data were normalized against U6 (SH-SY5Y) and SNO202 (KUM10 and HEK293T) housekeeping RNAs. Data were compared with multiple unpaired t tests. Statistical significance: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001 and N.D., not determined.

We evaluated EV size distribution and morphology by performing nanoparticle tracking analysis (NTA) and transmission electron microscopy (TEM) (Figures 1C and S1B). TEM allowed us to validate particle size, to sample purity (no protein aggregates), and to confirm the typical cup-shaped morphology of EVs. Further NTA evaluation for EVs produced by SH-SY5Y, KUM10, and HEK293T cell lines showed that each KUM10 cell produces around 1,666 particles, each SH-SY5Y cell produces around 360 particles, and each HEK293T cell produces around 216 particles over 48 h (Figure 1D). The isolated EVs were then evaluated concerning miRNA abundance. The levels of mir-155, a miRNA with the potential to be used as a scaffold, were evaluated between cells and EVs. mir-155 was found to be enriched more then 10-fold in EVs derived from SH-SY5Y, KUM10 and HEK293T cells, suggesting that this miRNA is naturally enriched in EVs (Figure S2A). Such enrichment has been previously associated with the ExoMotif GGAG, which was previously described as a driver of specific set of miRNAs into EVs.³⁶ To further investigate this activity, we evaluated the following ExoMotif-carrying miRNAs: miR-575, miR-451, miR-198, miR-601, miR-887, and miR-125a-3p. As control, we selected miR-181a, which was described as being restrained in cells due to the presence of a CellMotif.³⁶ miRNA levels were compared between parental cells and their derived EVs. From the set of evaluated miRNAs, EVs obtained from SH-SY5Y cells were significantly enriched in miR-451 and miR-601, while miR-575, miR-125a-3p, miR-198, miR-887, and miR-181a were more abundant in the producer SH-SY5Y cells (Figure 1E). In KUM10 cells, three miRNAs were found to be significantly enriched into EVs, namely miR-575, miR-451, and miR-601. Despite showing the same tendency, miR-887 did not reach statistical significance. miR-125a-3p and miR-181a were found to be restrained in cells compared with EVs. miR-198 was not detected in either cells or in EVs (Figure 1F). In HEK293T cells, four miRNAs were found to be significantly enriched into EVs compared with their origin cells, namely miR-575, miR-451, miR-601, and miR-887, while miR-198 did not reach statistical significance. In contrast, miR-125a-3p and miR-181a were more abundant in cells compared with EVs (Figure 1G). Interestingly, miR-451 was found to be significantly enriched in EVs derived from all the three cell lines with at least a more than 1,000-fold increase compared with their parental cells. Moreover, among the different cell lines, HEK293T cells exhibited more ExoMotif-containing miRNAs packaged into EVs from the set of miRNAs analyzed.

Overall, we found that ExoMotif signals drive miRNAs into EVs with different loading efficiencies depending on the cell line and the considered miRNA.

The GGAG ExoMotif and hnRNPA2B1 ribonucleoprotein increase EV loading efficiency of mutATXN3 miRNA-based silencer

To investigate whether the ExoMotif sequence GGAG, which is present in the set of miRNAs above described, would promote the packaging of an artificial miRNA embedded in a miR-155 scaffold into EVs, constructs encoding a mirSilencer targeting mutATXN3⁴⁹ were generated. The H1 promoter controls the expression of an allele-specific artificial miRNA targeting mutATXN3 mRNA (mirSilencer) or a scramble sequence that does not bind to mutATXN3 mRNA (mirScramble). Then, the ExoMotif GGAG was associated with mirSilencer and mirScramble (Figure 2A). First, we investigated whether the association of the GGAG ExoMotif would impact the silencing activity of the mirSilencer by transfecting Neuro2A cells stably expressing mutATXN3 with the described plasmids (Figure 2B). A significant reduction of 45% in mutATXN3 protein levels was confirmed by western blotting for both mirSilencer sequences (with and without the ExoMotif) when compared with the control condition (mirScramble), suggesting that ExoMotif incorporation does not affect the silencing efficiency of the mirSilencer. A similar experiment was done in HEK293T cells where mirScramble and mirSilencer plasmids associated with the ExoMotif were transfected in HEK293T cells expressing mutATXN3 plasmid. mirSilencer associated with ExoMotif shows allele specificity by significantly reducing the levels of mutATXN3 protein and not changing the levels of endogenous ATXN3 protein. Importantly, ExoMotif association to mirScramble does not promote silencing activity (Figure S3A).

Figure 2.

ExoMotif and hnRNPA2B1 association with miRNA mutant ataxin-3 (mutATXN3) silencer drives its packaging into EVs and reduces mutATXN3 mRNA levels

(A) Schematic representation of the experimental setup. All constructs express GFP and puromycin. H1 promoter is controlling the expression of an allele-specific artificial miRNA targeting mutATXN3 (mirSilencer) or a scramble sequence that does not bind to mutATXN3 mRNA (mirScramble). Then, the ExoMotif GGAG was associated with mirSilencer and mirScramble. (B) Levels of mutATXN3 protein in Neuro2A cells. mirSilencer plasmids (with and without ExoMotif) were transfected in Neuro2A cells encoding mutATXN3. Both plasmids led to a significant silencing of mutATXN3 protein level relative to mirScramble. (C) Levels of endogenous and mutant ATXN3 protein in HEK293T cells. mirScramble and mirSilencer plasmids associated with the ExoMotif were transfected in HEK293T cells expressing mutATXN3 plasmid. mirSilencer associated with ExoMotif shows allele specificity by significantly reducing the levels of mutATXN3 protein and not changing the levels of endogenous ATXN3 protein. Data are expressed as mean ± SEM. Unpaired t test. (D) Lentiviral vectors (LVs) were used to generate HEK293T stable cell lines encoding mirSilencers. mirSilencer levels in cells and their derived EVs were assessed by RT-PCR. (E) MirSilencer enrichment in EVs. Stable cell lines encoding mirSilencer with and without ExoMotif were used to compare the mirSilencer levels between progenitor cells and their derived EVs. mirSilencer associated with the ExoMotif and hnRNPA2B1 sequence is 3-fold significantly enriched in EVs when compared with their progenitor cells. A relative qPCR quantification was performed, and RNU1A1 was used as endogenous control. Results are expressed as mean ± SEM of arbitrary units (n = 4). Data were compared performing one-way ANOVA followed by Sidak’s multiple comparisons test (F = 4.336). (F) mutATXN3 mRNA levels upon EV incubation. EVs carrying mirSilencer with and without ExoMotif were incubated in Neuro2A cells encoding mutATXN3 mRNA. After 48 h, mutATXN3 mRNA levels were significantly decreased in cells incubated with EVs carrying mirSilencer with ExoMotif (n = 4). One-sample t test. Statistical significance: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001 and ns, not significant.

We then generated HEK293T stable cell lines using LVs encoding the mirSilencer either with or without the ExoMotif (Figure 2D). Moreover, an additional condition overexpressing the hnRNPA2B1 protein was added, aiming at promoting further enrichment of mirSilencer in EVs, as described in Villarroya-Beltri et al.³⁶ (Figure 2D). mirSilencer levels were evaluated in cells and their derived EVs by qRT-PCR (Figure S3A). The enrichment ratio between mirSilencer in EVs and cells was performed; remarkably, the mirSilencer was shown to be 3-fold enriched in EVs when compared with progenitor cells in the condition comprising both the ExoMotif and hnRNPA2B1 (Figure 2E). Then, we evaluated whether the enrichment of mirSilencer into EVs would enhance silencing of mutATXN3 mRNA in Neuro2A cells. For that, we incubated the enriched EVs with cells for 48 h. In fact, EVs carrying the mirSilencer associated with the ExoMotif were able to significantly downregulate mutATXN3 mRNA by 18.5%. In contrast, EVs carrying the mirSilencer without the ExoMotif showed no statistically significant downregulation of mutATXN3 mRNA (Figure 2F).

Overall, the co-expression of mirSilencer in association with the ExoMotif and the hnRNPA2B1 protein allowed increased packaging of mirSilencer sequences into EVs, enabling functional silencing of mutATXN3 mRNA in vitro.

Promotion of EVs internalization in neurons by incorporation of the PDGFR transmembrane protein fused with the RVG

In the previous sections, we demonstrated that we were able to improve the packaging of mirSilencer into EVs, enabling its functional delivery to cells. However, promoting gene silencing in the target neurons also depends on the subsequent step of EV internalization, which could be further optimized. We¹⁹ and other authors^21,44,50,51 previously showed that peptides derived from RVG can be displayed on the surface of particles to enhance neuronal targeting.⁵² In fact, it has been described that RVG confers neuronal targeting capacity due to the binding to acetylcholine receptors (AChRs) in neurons.⁵² Therefore, in this study, we took advantage of a CD63-NanoLuc reporter cell line to generate EVs decorated with the RVG peptide on their surface, which can be used to evaluate internalization efficiency by bioluminescence imaging. The RVG peptide fused with the platelet-derived growth factor receptor (PDGFR), a transmembrane protein that allows the anchoring of proteins on the surface of EVs through its fusion with a ligand of interest.⁵¹ The vesicles were incubated with bEnd.3 (endothelial cells from mouse brain tissue) or Neuro2A cells for 12 h. As a control, CD63-NanoLuc vesicles not displaying RVG on their surface were used (Figure 3A).

Figure 3.

RVG peptide expression on the surface of EVs promotes internalization into neurons

(A) Schematic representation of EVs expressing CD63-NanoLuc and PDGFR-RVG followed by their incubation in distinct cell lines. (B) RVG-EVs are internalized by multiple cell lines. EVs expressing PDGFR-RVG on their surface internalize 6× more into Neuro2A cells and 5× more in bEnd.3 cells when compared with CD63-Nanoluc EVs without RVG (n = 3). (C) Internalization profile of RVG-EVs in Neuro2A (red) and bEnd.3 (pink) cells at 2, 6, 12, 18, 24, and 36 h incubation time points. CD63-NanoLuc EVs without RVG expression were used as controls (black lines). Neuro2A cells internalize more EVs at early time points (6 h), while bEnd.3 cells showed a stable internalization profile over 36 h. Data are expressed as mean ± SEM (n = 3). (D) RVG-EVs internalize more into neuronal cells. Neuro2A cells showed an increase in the bioluminescence uptake signal when compared with bEnd.3 cells upon 6 h of incubation. The data were compared by unpaired t test. (E) Dose response of RVG-EVs. Internalization of RVG-EVs is proportional to the amount of EVs incubated. Higher dose of EVs (6 × 10⁸ particles) leads to higher internalization (37,873 RLU) when compared with the lower dose of EVs (1.5 × 10⁸ particles) that corresponds to 9,230 RLU in cells upon 6 h of incubation. The data are compared by ordinary one-way ANOVA followed by Tukey’s multiple comparisons test (F = 7,364). (F) Schematic representation of incubation of CD63-GFP EVs expressing PDGFR-RVG on their surface in primary rat cortical neurons. (G) Internalization of RVG-EVs by primary rat cortical neurons. Confocal images showing EVs (green) being internalized in neurons (red, β3-tubulin). The analysis was performed using laser confocal microscopy equipped with Plan-Apochromat 40×/1.40 Oil DIC M27 (420782-9900) (scale bar: 5 μm). CD63-GFP EVs expressing PDFGR-RVG on their surface internalize a high number of primary neurons when compared with control (CD63-GFP EVs) at 4 and 12 h incubation periods. The data are compared by one-way ANOVA followed by Sidak’s multiple comparisons test (F = 8.840). Statistical significance: ∗p < 0.05 and ∗∗∗∗p < 0.0001 (n = 3).

For every cell line, we found that the internalization was five to six times significantly higher for EVs expressing PDGFR-RVG when compared with the control condition (without the RVG-targeting moiety) (Figure 3B), suggesting that the RVG peptide allows a more efficient internalization both in neuronal and non-neuronal cells. Then, to investigate the internalization profile of the engineered vesicles overtime, we incubated vesicles for 2, 6, 12, 18, 24, and 36 h in bEnd.3 and Neuro2A cells. Neuro2A cells internalized more EVs at early time points (2, 6, and 12 h) compared with bEnd.3 cells. Neuro2A cells showed a peak of internalization corresponding to approximately 10,000 RLU after 6 h of incubation, while bEnd.3 showed 2,000 RLU at the same time point (Figure 3C), suggesting that PDGFR-RVG EVs are more efficiently internalized by Neuro2A cells. After 6 h of incubation, PDGFR-RVG EVs mediated 4 times more luminescence in Neuro2A cells (ΔRLU = 8,000) when compared with bEnd.3 cells (ΔRLU = 2,000), indicating that an increased number of vesicles were internalized (Figure 3D). To further understand whether the internalization is dependent on the number of RVG-containing vesicles, we then incubated RVG-EVs with two different doses of vesicles: 1.5 × 108 particles and 6 × 108 particles, respectively. The higher dose led to almost 4 times more luminescence in Neuro2A cells, suggesting that the internalization of RVG-EVs is proportional to the dose of incubated EVs (Figure 3E).

An additional experiment was performed to evaluate internalization of EVs in primary neurons. For that purpose, fluorescent EVs expressing CD63-GFP were used as a sensor for internalization. EVs exhibiting CD63-GFP and PDGFR-RVG on the surface were incubated with primary cortical rat neurons, and CD63-GFP EVs without PDGFR-RVG were used as control (Figure 3F). Laser confocal microscopy images displayed CD63-GFP EVs being internalized by β3-tubulin-positive neurons. Interestingly, CD63-GFP EVs expressing PDFGR-RVG on their surface were internalized by 63% of neurons, compared with 37% in the control condition, at 4 h post-incubation (Figure 3G). After 12 h of incubation, 82% of neurons displayed GFP expression compared with 57% GFP-positive neurons when PDGFR-RVG was absent from EVs (control), suggesting that PDFGR-RVG expression promotes internalization of EVs in primary neurons.

Engineered EVs significantly reduce mutATXN3 mRNA in vitro

Taking into consideration the enhanced delivery of the RVG-EV payload to neuronal cells, a packaging cell line encoding PDGFR-RVG (co-expressing mCherry) and hnRNPA2B1 (co-expressing turboGFP) was generated (Figure 4A). A stable expression of both transgenes was achieved after lentiviral transduction, followed by fluorescence-activated cell sorting (FACS) for double-positive cells co-expressing mCherry and turboGFP (Figure S4A). Afterward, cells were then split to separately overexpress each miRNA condition (mirScramble, mirSilencer, and mirSilencer with ExoMotif). The engineered vesicles continuously produced by these packaging cell lines were used to exploit the therapeutic potential and further in vitro applications (Figure 4A). A dual-luciferase reporter system encoding Firefly Luciferase (FLuc) associated with mutATXN3 and Renilla Luciferase (RLuc) under control of PGK and CMV promoters, respectively, was used to monitor the levels of mutATXN3 mRNA (Figure S5A). Engineered EVs were incubated in Neuro2A cells overexpressing the dual-luciferase construct (Figure 4B). After 48 h of incubation, the condition with EVs carrying mirSilencer with the ExoMotif significantly decreased the luminescence levels of FLuc-mutATXN3 by 34% compared with control (EVs carrying mirScramble), while EVs with mirSilencer without ExoMotif did not significantly silence FLuc-mutATXN3 mRNA. Moreover, a dose-dependent effect on reducing FLuc-mutATXN3 luminescence activity was observed upon EV incubation. Therapeutic EVs carrying the mirSilencer with the ExoMotif downregulated mutATXN3 mRNA levels from 57% to 72% upon doubling the dose of incubated EVs compared with control EVs carrying the mirScramble (Figure 4D). A similar experiment was performed in primary cortical neurons overexpressing the dual-luciferase reporter FLuc-mutATXN3 (Figure S5B). EVs carrying mirSilencer with the ExoMotif significantly decreased the luminescence levels of FLuc-mutATXN3 by 60% when compared with control (EVs carrying mirScramble) (Figure S5B).

Figure 4.

Engineered EVs significantly reduce mutATXN3 mRNA in vitro

(A) Representation of the packaging cell line overexpressing PDGFR-RVG, hnRNPA2B1, and mirSilencer to produce therapeutic EVs. Lentivirus encoding PDGFR-RVG and hnRNPA2B1 were used to generate a stable cell line (scale bar: 20 μm). The same cells were then split to overexpress mirSilencer with and without ExoMotif and mirScramble. Conditioned media were collected to isolate therapeutic EVs by dUC and incubated with various cell models to evaluate in vitro efficacy. (B) Incubation of therapeutic EVs in dual-luciferase reporter cells. Neuro2A cells overexpressing a dual-luciferase construct encoding Firefly Luciferase (FLuc) associated with mutATXN3 under control of PGK promoter and Renilla Luciferase (RLuc) under control of CMV promoter. (C) Luminescence of FLuc mutATXN3 upon EV incubation: therapeutic EVs carrying mirSilencer with ExoMotif significantly reduce luciferase activity by 34% compared with the control condition (EVs carrying mirScramble), while EVs with mirSilencer without ExoMotif do not significantly silence FLuc-mutATXN3 mRNA. Results are expressed in arbitrary units and mean ± SEM (n = 5). One-sample t test, column means significantly different than a hypothetical value of 1. (D) Dose-dependent effect of therapeutic EVs: therapeutic EVs carrying the mirSilencer showed a dose-dependent effect on reducing FLuc-mutATXN3 luciferase activity. Results are expressed as mean ± SEM of arbitrary units (n = 3). Data were compared by one-way ANOVA followed by Tukey’s multiple comparisons test (F = 21.01). (E) Establishment of cerebellar cultures from MJD YAC84.2 pups (P6–P7). Immunostaining at day 15 of primary cerebellar cultures showed positive staining for the neuronal marker MAP2, ATXN3 1H9, and deep cerebellar marker PCP4 under microscopy analysis (scale bar: 10 μm). (F) Therapeutic EVs downregulate endogenous mutATXN3 mRNA in cerebellar cultures. The first dose of EVs was incubated at day 10 and the second dose at day 12. At day 14, cells were collected, and mRNA was analyzed. A significant downregulation of endogenous mutATXN3 of approximately 49% was observed upon incubation with EVs containing the mirSilencer with ExoMotif. EVs carrying mirSilencer without ExoMotif showed a non-significant tendency to downregulate the levels of mutATXN3 mRNA (approximately 29%). Results are expressed as mean ± SEM of arbitrary units (n = 3/6). Data were compared by ordinary one-way ANOVA followed by Tukey’s multiple comparisons test (F = 4.75). Statistical significance: ns, nonsignificant, ∗p < 0.05, and ∗∗p < 0.01.

Therapeutic efficacy was then evaluated in cerebellar cultures from an MJD transgenic mouse model, hemizygous MJD YAC84.2 pups (P6-P7), which express the full human mutant ATXN3 gene.⁵³ Primary cerebellar cultures showed a positive staining for the microtubule-associated protein 2 (MAP2) and for the cerebellar neurons marker Purkinje cell protein 4 (PCP4), as well as for ATXN3, by microscopy analysis (Figure 4E). To evaluate the therapeutic potential of the engineered EVs in this model, primary cerebellar MJD YAC84.2 cells were incubated with two doses of EVs, added at days 10 and 12 of culture. A significant decrease of mutATXN3 of approximately 49% was observed upon incubation with EVs containing the mirSilencer with ExoMotif (Figure 4F). EVs carrying mirSilencer without ExoMotif showed a tendency to downregulate the levels of mutATXN3 mRNA approximately 29% relative to control condition.

Overall, these results suggest that engineered EVs carrying the mirSilencer with the ExoMotif and expressing the RVG peptide are functionally active at downregulating the levels of mutATXN3 mRNA in Neuro2A cells and in MJD murine cerebellar neurons.

Delivery of mirSilencer to the brain significantly decreases mutATXN3 mRNA in a dual-luciferase MJD mouse model

The promising results of therapeutic EVs at efficiently downregulating mutATXN3 mRNA in neurons in vitro led us to study the potential of this platform for in vivo purposes. For that aim, we developed a novel dual-luciferase MJD mouse model upon intracerebellar stereotaxic injection of LVs encoding FLuc associated with mutATXN3 and RLuc (Figure 5A). This model allows double-luminescence readout, where FLuc emits light 10 min after intraperitoneal (i.p.) injection of luciferin, which directly correlates with the mutATXN3 levels, and RLuc emits light 30 s after intravenous (i.v.) injection of ViviRen (coelenterazine analog) (Figure S6A). The suitability of this animal model to monitor the mutATXN3 mRNA levels in vivo was evaluated upon co-injection of LVs encoding the dual-luciferase MJD reporter system and either the mirSilencer with ExoMotif or mirScramble in the cerebellum. In vivo bioluminescence showed a significant and robust 59% reduction of FLuc-mutATXN3 luminescence levels in the mirSilencer condition relative to mirScramble (Figure 5B).

Figure 5.

Delivery of mirSilencer to the brain significantly decreases mutATXN3 mRNA in dual-luminescent MJD mouse model

(A) Generation of a dual-luminescent MJD mouse model upon intracerebellar injection of LV encoding dual-luciferase reporter (FL/RL) associated with mutATXN3. (B) Co-injection in the cerebellum of LV encoding the dual-luciferase MJD reporter and the mirSilencer with ExoMotif or mirScramble. In vivo bioluminescence assessment showed a significant decrease of FLuc-mutATXN3 luminescence by 59% in the mirSilencer condition relative to control (mirScramble). Results are expressed in mean ± SEM of arbitrary units (n = 3/4). Data were compared performing unpaired t test, ∗p < 0.05. (C) Schematic representation of daily intranasal administration of EVs. (D) Therapeutic EVs carrying mirSilencer with ExoMotif or mirScramble were administered intranasally daily for 1 month in a dose of 2 × 10⁹ EVs/animal/day. Evaluation of mirSilencer distribution throughout the brain upon intranasal administration showed the highest fold change of mirSilencer in the olfactory bulb, followed by the brainstem, cerebellum, and the remaining brain. (E1) Schematic representation of cerebellum processing for RNA and bioluminescence. (E2) Levels of mutATXN3 mRNA in the cerebellum homogenates were significantly reduced when mirSilencer with ExoMotif EVs (violin graph in blue) were administered compared with the scramble condition. (E3) Dual-luciferase assay in cerebellar homogenates showed a significant decrease of mutATXN3-luciferase activity in the EVs carrying mirSilencer with ExoMotif condition (violin graph in red) compared with scramble EVs. Results are expressed in mean ± SEM of arbitrary units (n = 6/8). Statistical significance: ∗p < 0.05 and ∗∗p < 0.01.

To explore a non-invasive route of administration, we evaluated whether RVG-EVs would reach the cerebellum of wild-type mice upon intranasal administration. CD63-GFP EVs expressing PDGFR-RVG were intranasally administered twice a day for 2 weeks (Figure S7A). Interestingly, we observed GFP fluorescence in the cerebellum, suggesting that EVs reach this brain region (Figure S7A).

Moreover, to further investigate whether RVG-EVs containing mirSilencer with ExoMotif administered through intranasal route would reach the cerebellum and knock down mutATXN3 in a dual-luminescent MJD mouse model, a dose of 2 × 10⁹ therapeutic EVs was administered in each animal daily for 1 month (Figure 5C). mirSilencer distribution throughout the brain showed the highest fold change in the olfactory bulb, followed by brainstem, cerebellum, and the remaining brain regions, suggesting that mirSilencer-containing EVs can reach the major regions affected in MJD, namely the brainstem and cerebellum (Figure 5D). Unexpectedly, treatment monitoring after 15 and 30 days did not demonstrate significant differences between conditions in living animals, possibly due to technical limitations (such as skull, skin, and fur interference) (Figure S8A). Animals were sacrificed after 30 days from the beginning of administrations, and cerebellum homogenates were split for RNA processing and bioluminescence analysis (Figure 5E1). Interestingly, mutATXN3 mRNA levels were significantly downregulated by 38% when mirSilencer with ExoMotif EVs were administered compared with scramble EVs (Figure 5E2). These findings were also corroborated at the protein level by a dual-luciferase assay in cerebellum homogenates that showed a significant downregulation of 24% in the luciferase activity in the condition with respect to EVs carrying mirSilencer with ExoMotif when compared with scramble EV conditions (Figure 5E3).

As all the above points have demonstrated, these results showed that long-term administration of engineered EVs via the intranasal route in an MJD mouse model (1) deliver the mirSilencer with ExoMotif to the cerebellum and (2) downregulate mutATXN3 mRNA, thus turning out to be a promising therapeutic approach to alleviate MJD in vivo.

Discussion

The present study demonstrates that a miRNA-based silencing sequence targeting mutATXN3 mRNA embedded in a miR-155 scaffold is (1) packaged into EVs, (2) significantly enriched upon association with the ExoMotif GGAG and the hnRNPA2B1 protein, and (3) more efficiently delivered to neuronal cells when the corresponding EVs are decorated with the RVG peptide on their surface. Engineering EV-packaging cells with modified miRNA-based silencing sequences, hnRNPA2B1, and the RVG peptide enabled production of EVs with the capacity to efficiently silence mutATXN3 in cell lines and primary cultures of cerebellar neurons of an MJD transgenic mouse model. Finally, a daily non-invasive intranasal administration of these therapeutic EVs into a dual-luminescent MJD mouse model enabled their efficient delivery into the most affected brain regions and significantly silenced mutATXN3 expression, suggesting that our bioengineered EVs can stand as a promising therapeutic strategy for MJD.

Many efforts have been made to develop RNA interference (RNAi) strategies to ameliorate the neuropathology and rescue the disease phenotype in MJD animal models.^{5,9,12,13,54,55} RNAi is a naturally occurring mechanism that involves sequence-specific downregulation of mRNA by simply destroying it or avoiding its translation into protein.⁵⁶ The vast majority of RNAi technologies used to downregulate mutATXN3 mRNA are based on the delivery of shRNAs or siRNAs through intracranial injection.^{8,9,10,11,12,13,15,16} However, intracranial injections are an extremely invasive procedure with safety issues and limited distribution across the different diseased brain regions. Less invasive administration routes have therefore been exploited to deliver silencing sequences to the brain with success in the context of MJD.¹⁹ Nevertheless, the use of cell-derived lipid membranes, such as EVs, as a delivery vehicle for RNAi has still been poorly explored.^20,57,58 In fact, so far only two studies have explored the therapeutic potential of EVs in MJD animal models using either native mesenchymal stem cell (MSC)-derived EVs⁵⁹ or miR-6780-5p-enriched EVs derived from butylidenephthalide pre-conditioned human olfactory ensheathing cells,⁶⁰ both with promising results at alleviating motor behavior phenotypes. To the best of our knowledge, the use of engineered EVs as a vehicle to deliver artificial miRNA-based silencers to downregulate mutATXN3 mRNA in MJD mouse models had not been previously investigated.

In this work, three different cell lines (HEK293T, KUM10, and SH-SY5Y) were analyzed regarding miRNA incorporation into secreted EVs. mir-155 was found to be enriched in all cell lines being used as a scaffold for our mirSilencer. Moreover, two miRNAs containing the ExoMotif GGAG were identified as highly enriched into EVs from all cell lines: miR-451 and miR-601. Remarkably, miR-451 is at least 1,000-fold enriched in EVs when compared with their progenitor cells, suggesting miR-451 as an efficient scaffold candidate to incorporate silencing sequences. Indeed, this strategy was shown to be efficient in loading SOD1 siRNAs into EVs, reducing the therapeutic dose of siRNAs and, consequently, the toxicity in SOD1 G93A mice.⁴³ A distinct study performed intracranial injection in Huntington and SCA3 disease models by directly infusing an AAV5-encoding miR-451 scaffold with silencing sequences targeting huntingtin (miHTT) and ataxin-3 (miATXN3) in non-human primates (NHPs).⁶¹ In this case, miR-451 packaging properties into EVs allowed long-term expression of the artificial miRNA in EVs secreted in CSF for up to 2 years.⁶¹ Nevertheless, high packaging efficiency into EVs may not necessarily correlate with high therapeutic efficiency with respect to mRNA downregulation in target cells, due to the possibility of redirection for EV secretion upon cell internalization without reaching the target mRNA.⁶²

The loading of small RNAs in EVs depends on multiple factors: (1) the presence of ExoMotif sequences, such as GGAG, CCCU, GGCU, UGGA, and CGGGAG sequences^36,37,38,63; 2) the presence of RNA-binding proteins, such as hnRNPA2B1, SYNCRIP, Y-box protein 1, HuR, Lupus La protein, Alyref, and Fus^{36,37,38,39,40,63,64}; (3) the secondary structure of miRNAs^39,40; and (4) cell type-specific mechanisms.^{36,37,38,39,40,63} Our miRNA-based silencing sequence targeting mutATXN3 mRNA was based in an miR-155 scaffold⁴⁹ that has been described in previous studies to be naturally enriched into EVs.^65,66,67 To further increase our miRNA packaging efficiency by an endogenous loading mechanism, we associated the ExoMotif GGAG and hnRNPA2B, mediating a 3-fold enrichment when compared with the control condition. However, EV-based delivery of miRNA with the ExoMotif promoted a modest improvement of the downregulation of mutATXN3 mRNA when compared with the condition without ExoMotif, suggesting that packaging of miRNA into EVs may not be the only influencing factor, and highlighted the involvement of other mechanisms to increase the efficiency of miRNA delivery through EVs.

We then hypothesized that a limiting step for therapeutic delivery and efficacy of our EV-miRNA strategy would relate to their internalization process into recipient cells. We observed limited internalization rates of EVs into Neuro2A cells, encouraging us to further decorate their surface with the PDGFR-RVG fusion protein. The RVG peptide was described to target neurons through a specific ligand-receptor-mediated transcytosis mechanism, relying on the binding to the AchR, which is expressed in neurons. Many studies used RVG-targeting properties to reach the brain either by directly associating siRNA with RVG,^52,68 by decorating liposomes to enable brain targeting,^19,57,58,69 or by expressing them on EVs for the same purpose.^{20,21,22,45,70} LAMP2B is the transmembrane domain typically used to express the RVG peptide on the surface of EVs.⁴⁴ Instead, we used the PDGFR transmembrane domain to associate with the RVG peptide, an extremely efficient system already used to deliver EV-AAV to the brain.⁵¹ Indeed, as expected, RVG improved EV internalization, which was more efficient in Neuro2A cells than in other non-neuronal cells lines, with a prominent peak of internalization at 6 h. Even so, we found that RVG-EV incubation with a non-neuronal cell line (bEnd.3) promoted more internalization compared with their respective controls (not expressing RVG on the surface), suggesting that other internalization mechanisms may also occur in the presence of RVG. Overall, RVG-EVs showed an improved internalization profile in all tested cell lines, suggesting that the use of RVG may improve delivery of therapeutic cargo, particularly to neuronal cells. Indeed, RVG-EVs in association with the ExoMotif and hnRNPA2B1 achieved around 30% and 50% downregulation of mutATXN3 in a dual-luciferase reporter cell line and in primary cerebellar cultures, respectively, compared with a 20% downregulation of mutATXN3 mRNA when using EVs without RVG. Taken together with previous findings, we hypothesize that the major limiting step for the downregulation of mutATXN3 mRNA is the limited internalization of EVs in recipient cells. Therefore, future studies should address the internalization of EVs by exploring their native properties from different cell sources or by modifying their surface with other fusogenic entities.

The RVG peptide is typically expressed on the surface of nanoparticles to target the brain upon i.v. administration.^{20,21,22,50,58,71} Despite the success of this strategy, some concerns regarding the efficiency upon i.v. injection have been raising due to the high liver retention.⁷² Additionally, the AchR is also expressed on macrophages,⁷¹ which may increase the engulfment of RVG-EVs by resident macrophages in liver.⁵¹ An alternative non-invasive route is intranasal administration, which bypasses the BBB and results in less systemic adverse effects. This route is being increasingly used for brain delivery of small molecules,^73,74,75,76 MSCs,⁴⁸ olfactory ensheathing cells,^77,78 and EVs.^47,48,79,80 A comparison between i.v. and intranasal (i.n.) administrations of EVs to reach the mouse brain was performed using neuroimaging with gold nanoparticle-labeled EVs. These in vivo results suggested that i.n. administration was more effective than i.v. injection.⁷⁹ Indeed, i.n. administration of MSC-derived EVs was previously shown to be neuroprotective and immunomodulatory in a 3xTg animal model of Alzheimer’s disease (AD). Curiously, this administration route was chosen to the detriment of i.v. or intra-CSF single administration due to the additional advantage of feasible multiple administrations through a non-invasive procedure.⁴⁷

There are two main mechanisms described for the delivery of therapeutics from nasal cavity to the brain: (1) one involving the trigeminal nerve, also known as intraneuronal pathway, which requires axonal transport of the therapeutics throughout several days, and (2) the other involving the extraneuronal pathway, involving the extracellular bulk flow along perineural and perivascular channels, and biofluids (such as CSF) directly to brain parenchyma, constituting a faster process.^73,80,81,82 The trigeminal nerve connects the olfactory bulb to the brainstem,⁸³ and it was shown to express AchR,^84,85 being a probable target for the RVG nanoparticles.⁸⁶ We successfully delivered RVG-EVs through the i.n. route to different brain regions after performing repeated daily administrations for 1 month. Indeed, we could detect the presence of EVs in the cerebellum and brainstem, regions linked to the olfactory bulb by the trigeminal nerve. We hypothesize that this effect may be enhanced by the RVG peptide, which facilitates EV internalization by AchR-expressing cells in trigeminal nerves. Interestingly, a recent study in larger mammals pointed out limitations in brain distribution of EVs through the i.n. route, suggesting its reconsideration for use in Macaca nemestrina or humans.⁸⁷

Remarkably, we saw a significant downregulation in mutATXN3 mRNA within the cerebellum, one of the primary regions affected in MJD, suggesting that a daily administration of EVs can be a useful strategy to deliver silencing sequences to the brain upon i.n. administration. These findings corroborate a previous study that shows i.n. delivery of neuropeptide Y (NPY) to be effective at mitigating MJD motor impairment phenotype and the neuropathology in an MJD transgenic mouse model with severe cerebellar atrophy.⁸⁸

Besides the conventional MJD/SCA3 mouse models used in pre-clinical studies for motor and neuropathology evaluation, here we reported the development of a novel dual-luminescent mouse model with tremendous advantages to evaluate in vivo target engagement: (1) it expresses FLuc associated with mutATXN3 gene that emits light 10 min after i.p. injection of luciferin, while RLuc emits light 30 s after i.v. injection of ViviRen (analog of coelenterazine used for in vivo studies) working as a housekeeping control; (2) it allows therapy monitorization and dose adjustment in living mice without the need of sacrifice; (3) it can be applied to a specific brain region through intracranial injection; and (4) it can be applied in other genetic-based diseases. We should mention that the lesion caused by the intracranial injection in the brain parenchyma may compromise BBB integrity and increase EV recruitment to the injected site, as described before by Zhuang et al.⁸⁰ Other MJD/SCA3, such as YACQ84 SCA3, animal models should be further tested after establishing bioreactors of 3D cells lines for scaling up EV production.

In this study, we engineered EVs to pack our miRNA-based silencing sequences with high efficiency and increased their internalization properties by decorating EV surfaces with the RVG peptide. The use of the ExoMotif associated with our artificial miRNA, together with the expression of hnRNPA2B1 and PDGFR-RVG proteins, allowed us to generate therapeutic EVs with the capacity to efficiently downregulate mutATXN3 species in different cellular models. Additionally, we demonstrated that EVs carrying silencing sequences can reach the cerebellum and ameliorate MJD through a significant downregulation of mutATXN3, constituting a promising therapeutic strategy for this and other neurodegenerative disorders.

Materials and methods

Animals

All animal experimental protocols were approved by the European Union Directive 86/609/EEC for the care and use of laboratory animals. This study is part of a research project that was approved by the Center for Neuroscience and Cell Biology ethics committee (ORBEA_66_2015_/22062015 and ORBEA_289_) and the Portuguese authority responsible for the regulation of animal experimentation, Direcção Geral da Agricultura e Veterinária (DGAV 0421/000/000/2015).

Researchers received adequate training (Federation of European Laboratory Animal Science Associations [FELASA]-certified course) and certification from Portuguese authorities (Direcção Geral de Alimentação e Veterinária) to perform the experiments. MJD YAC84.2 and C57BL/6 mice (Charles River Laboratories) were maintained with unlimited access to water and food under a 12 h light/dark cycle. Male and female mice ranging from 8 to 10 weeks in age were randomly assigned to experimental groups.

Lentiviral production and titer assessment

LVs encoding for the mirSilencer, mirScramble, hnRNPA2B1, and PDGFR-RVG plasmids were produced in the HEK293T cell line, as previously described in Carmona et al. and de Almeida et al.^35,89 Briefly, cells were seeded and 24 h later were transfected with a four-plasmid system. Six hours after transfection, cells were washed with PBS and incubated in new culture media. LV isolation was performed 48–72 h later upon UC at 70,000g followed by pellet resuspension in 1% PBS/BSA. Viral particle was evaluated by assessing HIV-1 p24 antigen levels by ELISA 2.0 (Retro Tek, 0801002), in accordance with the manufacturer’s instructions. Concentrated viral stocks were stored at −80°C until use.

Stereotaxic injection into the mouse brain

C57BL/6J mice of 4–5 weeks of age were anesthetized through i.p. injection of a mixture of ketamine (75 mg/kg, Nimatek, Dechra) and medetomidine (0.75 mg/kg, DOMTOR, Esteve).

Mice were stereotaxically injected into the striatum with the following coordinates relative to bregma: anteroposterior: 0.6 mm, lateral: +1.8 mm, ventral: 3.3 mm, and tooth bar: 0 mm, with concentrated LVs in a final volume of 2 μL/injection containing 400 ng p24 antigen. For cerebellar injections (Lobule V), bregma and lambda were aligned, and we used the following coordinates relative to lambda: anteroposterior: −2.4 mm, lateral: 0 mm, ventral: −2.9 mm, and tooth bar: 0 mm. LVs were injected in a final volume of 4 μL/injection containing 600 ng p24 antigen. The infusion was performed at an injection rate of 0.25 mL/min using a 10 mL Hamilton syringe; 5 min after the infusion was completed, the needle was retracted 0.3 mm and allowed to remain in place for an additional 3 min prior to its complete removal.³⁵ The skin was closed using a 6-0 Prolene suture (Ethicon, Johnson and Johnson, Brussels, Belgium).

In vivo bioluminescence analysis

Stable lentiviral transduction in the cerebellum was monitored by assessing FLuc bioluminescence periodically using IVIS Lumina XR equipment upon injection of D-Luciferin (PerkinElmer). For each determination, mice were i.p. injected with D-Luciferin (100 mg/kg) and anesthetized with 2.5% isoflurane in 100% oxygen. Bioluminescence images were acquired 10–20 min after D-Luciferin injection. To evaluate RLuc expression, i.v. injection of ViviRen (coelenterazine substrate modified for in vivo analysis) was administered, and the signal collected 30 s after injection. Analysis was performed using Living Image software (v.4.10, Xenogen), and quantification of the bioluminescent signal was obtained from a region of interest (ROI) drawn around the cranium. Values are expressed as average radiance relative to control.

Mouse tissue preparation for immunofluorescence

Mice were sacrificed under lethal administration of ketamine and xylazine, followed by intracardiac perfusion with PBS and fixation with 4% paraformaldehyde (PFA)/PBS (Sigma). Brains were post-fixed in 4% PFA/PBS for 48 h at 4°C, followed by incubation in 30% sucrose/PBS for 48 h at 4°C. Brains were then frozen at −80°C and sliced in cryostat (Leica CM3050S, Leica Microsystems at −20°C). Sagittal sections of 35 μm were collected in a serial mode in PBS/Azide (0.05 μM) for further free-floating immunofluorescence.

Immunofluorescence

Free-floating immunofluorescence was initiated by incubating the selected brain sections for 1 h in blocking and permeabilizing solution, 0.1% Triton X-100/10% normal goat serum (NGS) in PBS, at room temperature (RT). Sections were incubated overnight at 4°C with rabbit polyclonal anti-GFP antibody (1:1 000, Thermo Scientific). Sections were washed in PBS and incubated for 2 h at RT with the corresponding secondary antibody Alexa Fluor 564 (1:200, Invitrogen). Sections were washed with PBS and incubated with DAPI (1:5,000; Sigma), then washed and mounted with mounting medium (Dako) on gelatin-coated slides.

Mouse cerebellar primary culture

Primary cultures of MJD YAC84.2 pups (post-natal day 6 [P6]–P7) cerebellar neurons were prepared from (P6–P7) post-natal pups. Cerebella were dissected and dissociated with trypsin (0.01%, Sigma, T0303) for 15 min (inversion each 5 min) at 37°C and DNase (0.045 mg/mL, Sigma, D5025) in Mg2+-free Krebs buffer (120 mM NaCl, 5 mM KCl, 1.2 mM KH₂PO₄,13 mM glucose, 15 mM 4-(2-hydroxyethyl)pipera-zine-1-ethanesulfonic acid [HEPES], 0.3% BSA [pH 7.4]).

Cerebella were then washed with Krebs buffer (with Mg2+) containing trypsin inhibitor (0.3 mg/mL, Sigma, T9128) to stop trypsin activity. Cells were dissociated in this solution and centrifuged. Pellet was resuspended first with a pipet tip, followed by a syringe with a needle of 21G, then filtered through a strainer of 40 μm and resuspended in Basal Medium Eagle supplemented with 25 mM KCl, 30 mM glucose, 26 mM NaHCO3, 1% penicillin-streptomycin (100 U/mL, 100 mg/mL), and 10% fetal bovine. Cells were plated on 48- or 24-well plates coated with poly-D-lysine. 24–48 h after the isolation, cytosine arabinoside 10 μM final concentration was added to cultures. Cultures were maintained up to 15 days in a humid incubator (5% CO2/95% air at 37°C).

Immunocytochemistry

Cell cultures were washed and fixed with 4% PFA/PBS. After permeabilization and blocking in PBS/0.1% Triton X-100/3% BSA, cells were incubated with primary antibodies overnight at 4°C. The following primary antibodies were used diluted in blocking solution: mouse anti-β3 tubulin clone 38F4 (1:500; Life Technologies), mouse anti-MAP2 (1:250, M1406, Sigma), quail anti-ATXN3 antibody (1:1,000, HBT018-100, HenBiotech), rabbit polyclonal anti-PCP4 (C15) (1:200 Santa Cruz), and rabbit polyclonal anti-GFP antibody (1:1,000, Thermo Scientific). Cells were washed with PBS and incubated for 2 h at RT with the secondary antibodies Alexa Fluor 488, -564, and -647 (anti-rabbit, anti-mouse, 1:200 Invitrogen, and anti-chicken, 1:250 Life Technologies). Cells were washed with PBS and incubated with DAPI (1:5 000; Sigma), then washed and mounted in mounting medium (Dako) on gelatin-coated slides. Cells were visualized in a Zeiss Axio Imager Z2 and Zeiss LSM 510 Meta confocal microscope (Carl Zeiss MicroImaging), equipped with EC Plan-Neofluar 40×/1.30 Oil DIC M27 (420462-9900) and Plan-Apochromat 63×/1.40 Oil DIC M27 (420782-9900) objectives and ZEN Image software.

Cell line culture and transduction

HEK293T, bEnd.3, KUM10, and Neuro2A cells were maintained in standard DMEM (Sigma) supplemented with 10% fetal bovine serum (Life Technologies) and 1% penicillin-streptomycin (Gibco) and grown at 37°C and 5% CO2. SH-SY5Y were maintained in DMEM-F12 (Sigma) supplemented with 10% fetal bovine serum (Life Technologies) and 1% penicillin-streptomycin (Gibco) and grown at 37°C and 5% CO2. Cells were plated and transduced 24 h after plating with LVs encoding each construct (400 ng p24 per 200,000 cells). 24 h later, the medium was replaced with regular medium, and cells were cultured and expanded in their standard conditions. Fluorescence and genomic DNA analysis was used to monitor stable cell line generation.

Isolation of EVs by dUC

For EV isolation, cells were cultured in depleted fetal bovine serum (dFBS) (previously centrifuged at 100,000g for 18 h at 4°C). Medium was collected from cells at 80% confluency after 48–72 h and centrifuged at 300g for 10 min, followed by a 2,000g centrifugation for 10 min to remove cells and death cells. To remove cellular debris, medium was then centrifuged at 16,500g for 1 h in thin-wall polyallomer tubes (Beckman Coulter), SW28Ti rotor (Beckman Coulter) in Centrifuge Optima XE-100.

Supernatant was then filtered with a 0.22 μm sterile syringe filter (Merck Millipore) to remove particles larger than 220 nm from the media. Supernatant was placed into new thin-wall polyallomer tubes (Beckman Coulter) to pellet EVs at 100,000g for 2 h. The pellet was then washed abundantly with particle depleted PBS (dPBS) at 100,000g for 2 h to eliminate contaminating proteins. Pellet was resuspended to a final volume between 50 and 100 μL.

Western blotting

Total protein from cells and EVs was extracted using RNA immunoprecipitation (RIPA) buffer (50 mM Tris-base; 150 mM NaCl; 5 mM EGTA; 1% Triton X-100; 0.5% sodium deoxycholate; 0.1% SDS) containing cOmplete Mini proteinase inhibitor (Roche) and supplemented with 0.2 mM PMSF (phenylmethylsulphonyl fluoride), 1 mM DTT (dithiothreitol), 1 mM sodium orthovanadate, and 5 mM sodium fluoride. Protein concentration was determined by Bradford assay according to manufacturer instructions (Bio-Rad Laboratories). Protein samples were denatured (95°C for 10 min) with 6× sample buffer containing 0.375 M Tris (pH 6.8; Sigma-Aldrich), 12% SDS (Sigma-Aldrich), 60% glycerol (Sigma-Aldrich), 0.6 M DTT (Sigma-Aldrich), and 0.06% bromophenol blue (Sigma-Aldrich). Samples were resolved by electrophoresis on 10% or 12% SDS-PAGE and transferred onto polyvinylidene fluoride (PVDF) membranes (GE Healthcare). Total protein labeling was performed using No Stain Labeling Reagent (Invitrogen) according to the manufacturer’s protocol. Membranes were blocked by incubation in 5% non-fat milk powder in 0.1% Tween 20 in Tris-buffered saline (TBS-T) and incubated overnight at 4°C with primary antibodies ALIX (BD Biosciences, 611620, 1:1,000), calnexin (Santa Cruz, sc-11397, 1:1,000), CD63 (DSHB, AB528158, 1:500), Flotillin-1 (BD Biosciences, 610820, 1:1,000), HSC70 (GeneTex, GTX101144, 1:1,000), Lamp-2 (Santa Cruz, sc18822, 1:1,000), and TSG101 (BD Biosciences BD612696, 1:1,000). Membranes were then washed 3 times in TBS-T for 10 min each and incubated with an alkaline phosphatase-linked secondary goat anti-mouse/anti-rabbit antibody (1:10,000; Thermo Scientific Pierce) at RT for 1 h. Bands were visualized with enhanced chemifluorescence (ECF) substrate (GE Healthcare) in chemifluorescence imaging (ChemiDoc Imaging System, Bio-Rad). Analysis was carried out based on the optical density of scanned membranes in ImageLab v.5.2.1 (Bio-Rad).

RNA extraction, cDNA synthesis, and RT-PCR

Total RNA was isolated with miRCURY RNA isolation kit (Exiqon), Total RNA Purification Plus Kit (Norgen), and Total RNA isolation Kit (Macherey-Nagel) according to the manufacturer’s instructions. RNA was quantified using a NanoDrop 2000 Spectrophotometer (Thermo Scientific) and stored at −80°C.

Specific cDNAs for miRNA quantification were synthetized using a TaqMan MicroRNA Reverse Transcription Kit combined with specific TaqMan MicroRNA Assays (Applied Biosystems) for each miRNA according to manufacturer’s instructions. qPCR was performed using TaqMan Universal PCR Master Mix II with UNG (Applied Biosystems) in a StepOnePlus Real-Time PCR System (Applied Biosystems).

cDNA synthesis for mRNA quantification was performed with iScript cDNA Synthesis Kit (Bio-Rad) from 0.5 to 1 μg total RNA. Real-time quantitative PCR was performed with the Sso Advanced SYBR Green Supermix Kit (Bio-Rad) using the StepOnePlus Real-Time PCR System (Applied Biosystems). Reactions were performed in duplicated or triplicated. The amplification rate for each target was evaluated from the cycle threshold (Ct) numbers obtained with cDNA dilutions. Differences between control and experimental samples were calculated using the 2^−ΔΔCt method. TaqMan Probes for miRNAs are miR-575 (ID001617), miR-451 (ID001141), miR-198 (ID002273), miR-601 (ID001558), miR-887 (ID002374), miR-125a-3p (ID002199) and mir-181a-5p (ID000480), U6 snRNA (ID001973), and snoRNA202 (ID001232). Exiqon primers are as follows: RNU5G (hsa, mmu, rno), RNU1A1 (hsa, mmu, rno), mirScramble: ID: 715657-1, and mirSilencer: IDs: 715661-1 and 715653-1.

The following primers were used: FLucFwd: 5'-CTCACTGAGACTACATCAGC-3' and FLucRev: 5'-TCCAGATCCACAACCTTCGC-3'; RLucFwd: 5'-GGAATTATAATGCTTATCTACGTGC-3' and RLucRev: 5'-CTTGCGAAAAATGAAGACCTTTTAC-3'; hATXN3 Fwd: 5'-TCCAACAGATGCATCGACCA-3' and hATXN3 Rev: 5'-ACATTCGTTC CAGGTCTGTT-3'; mGAPDH Fwd: 5'-TGGAGAAACCTGCCAAGTATGA-3' and mGAPDH Rev: 5'-GTCCTCAGTGTAGCCCAAG-3'; and hGAPDH Fwd: 5'-TGTTCGACAGTCAGCCGCATCTTC-3' and hGAPDH Rev: 5'-CAGAGTTAAAAGCAGCCCTGGTGAC-3'.

Artificial miRNAs against mutATXN3 allele

The engineering of an allele specific artificial miRNA against mutATXN3 mRNA was recently demonstrated by our group.⁴⁹ miRNA155-based scaffold was used to embed the following oligomer sequences: 5'-TGATAGGTCCCGCTGCTGCTGC-3' to encode mirSilencer and 5'-CAACAAGATGAAGAGCACCAA-3' to encode mirScramble.

Dual-luciferase reporter assay

The dual luciferase reporter constructs with FLuc associated with mutATXN3 (FLuc-mutATXN3), and RLuc was used to evaluate target engagement of artificial miRNA. Cells were washed with PBS and frozen at −80°C or directly processed. Cell processing was performed according to the manufacturer’s instructions (Dual-Luciferase Reporter Assay System). Briefly, cells were lysed with passive lysed buffer (PLB) and 20 μL loaded in white 96-well culture plates (Lumitrac 200) and an opaque 96-well plate (Corning).

Firefly luminescence activity was measured on Synergy H1 Hybrid Multi-Mode Reader (BioTek) and FLUOstar Omega Microplate Reader (BMG LABTECH) after automatic injection of 100 μL Luciferase Assay Buffer II (LARII). Renilla luminescence activity was used as a normalization control and was measured after automatic injection of 100 μL Stop & Glo Reagent. Integration times were 10 s for FLuc signal capture and 5 s for RLuc signal capture. Each sample was loaded in duplicate, and at least 2 reads were performed.

Transmission electron microscopy

EVs isolated by dUC were fixed with 2% PFA/PBS and allowed to absorb on Formvar-carbon coated grids (TAAB Laboratories) for 5 min. The excess liquid was blotted off the film surface using a filter paper (Whatman). Then, the grids were contrasted with uranyl acetate 2%, and after 1 min, the excess stain was blotted off, and the sample air dried. Observations were carried out using a Tecnai G2 Spirit BioTwin electron microscope (FEI) at 100 kV.

Nanoparticle tracking analysis

Number of EVs diluted in PBS was assayed using the NTA v.2.2 Build 0375 instrument (NanoSight NS300 instrument, Malvern Instruments). Particles were measured for 30 s, and the number of particles (30–800 nm) was determined using NTA Software 2.2. Samples were diluted 1:1,000 in PBS prior to analysis. The following photographic conditions were used: frames processed (1,498 of 1,498 or 1,499 of 1,499); frames per second (24.97 or 24.98 f/s); calibration (190 nm/pixel); and detection threshold (6 or 7). Number of particles per frame was within the recommended range of 20–100 particles/frame for NanoSight NS300.

Data availability

All original data are available from the authors under request.