Extracellular vesicle-mediated gene editing for the treatment of non-syndromic progressive hearing loss in adult mice

Editor’s Summary:

The clinical translation of gene therapy has been challenging, due to limitations from current delivery approaches. Herein, we report an efficient non-viral genome editor delivery approach using single guide RNA (sgRNA):CRISPR-associated protein 9 (Cas9) ribonucleoprotein (RNP) complexes mediated by extracellular vesicles (EVs) for in vivo gene therapy. By leveraging a high-throughput microfluidic droplet-based electroporation system (μDES), we achieved a 10-fold enhancement in loading efficiency and more than a 1000-fold increase in processing throughput for loading RNP complexes into EVs compared to conventional high-voltage pulsed electroporation. μDES generates uniform microdroplets containing EVs and RNPs, applying direct current (DC) controlled low voltage (up to 60V) to transiently permeabilize membranes and enable efficient cargo encapsulation while maintain EV integrity at both protein and morphological level.. In the Myo7a^WT/Sh1 mouse model of autosomal dominant progressive hearing loss which potentially represent Myo7a-associated DFNA11 hearing loss in humans, we demonstrated the effective delivery of RNP mediated by EVs into cochlear hair cells by cross-sectional and whole-mount confocal imaging. The posterior semicircular canal injection of RNP-EVs in Myo7a^WT/Sh1 (4-week old) resulted in a notable reduction of mRNA expression of Myo7a^Sh1 allele and evidence of hearing recovery as measured by auditory brainstem responses (ABR), compared to untreated side of ears and EV only control groups. This study highlights the potential of μDES-produced RNP-EVs for gene editing as a treatment for progressive non-syndromic hearing loss in patients.

One Sentence Summary:

High-throughput loading of sgRNA:Cas9 RNPs targeting Myo7a^Sh1 into extracellular vesicles rescues progressive hearing loss in an adult mouse model.

Introduction

Genome editing is an emerging and powerful therapeutic tool for treating diverse diseases. Sensorineural hearing loss affects more than 450 million people worldwide and up to 50% of cases have a genetic origin (1–3). To date, gene delivery to the inner ear has been largely pursued with engineered adeno-associated viral vectors (AAVs) and cationic lipid nanoparticles (LNPs) (4–8). Although Cas9, an RNA-guided nuclease used for precise genome editing, can selectively disrupt autosomal dominant mutant alleles linked to hearing loss, their effective delivery remains a challenge because of the risk of prolonged expression of single guide RNA (sgRNA):Cas9 ribonucleoproteins (RNPs) and off target effects (9, 10). In addition, AAVs have limited packaging capacity (~4.7 kb), preventing delivery of larger endonucleases such as high-fidelity Cas9 derived from Streptococcus pyogenes in a single vector (11, 12). Larger vector systems like lentiviruses with higher packaging capacity ~10 kb, pose risks of insertional mutagenesis and oncogenesis(13). LNP and liposome-mediated delivery of Cas9-based gene editing has been extensively evaluated for transfection effectiveness to post-mitotic cells such as neurons and hair cells, but the tolerability in the inner ear is only partially defined (14–16). An alternative may be extracellular vesicle (EV)-based gene delivery, which is emerging as a safe and biocompatible approach for addressing some of the above-mentioned challenges (17–20). The first-in-human use of umbilical cord mesenchymal stromal cell-derived EVs (hUC-MSC-EVs) demonstrated potential to attenuate inflammatory side effects from cochlear implantation (17, 21). The hUC-MSC-EVs were also shown to attenuate inflammation induced by noise trauma in a mouse model, which indicates superior otoprotective properties and promising biodistribution in the cochlea (17). However, EV-mediated delivery of sgRNA:Cas9 RNP complexes for gene editing in the inner ear has not yet been fully explored.

Efficiently loading sgRNA:Cas9 RNP complexes into EVs has been a major roadblock for developing EV-based gene therapy. Conventional approaches, such as chemical transfection, sonication, or bulk electroporation, often suffer from low loading efficiency, heterogeneous cargo distribution, and damage to vesicle integrity (22). These methods also lack scalability and reproducibility, limiting their translational potential (23, 24). In addition, using engineered cells as EV producers constrains the type and amount of cargo that can be packaged into vesicles. For instance, sgRNA:Cas9 RNP complexes may indue genome editing in producer cells, thus, requiring additional molecular modules for selective EV packaging to ensure specific functional delivery (25–27). Such sophisticated design substantially increases system complexity and may hinder large-scale production. More recently, microfluidic or extrusion-based techniques have been explored to enhance cargo loading (28, 29), but maintaining EV stability and biological functionality during processing remains a critical barrier. These challenges highlight the need for a robust and scalable strategy that enables efficient incorporation of large biomolecular cargos, such as sgRNA-Cas9 RNP complexes, while preserving the native structure and function of EVs..

MYO7A encodes the unconventional myosin VIIA protein in auditory and vestibular hair cells that plays an essential role in the development of sensory hair cells and signal transduction (30–33). Mutations in this gene lead to 39 – 55% of the total cases of Usher syndrome type 1B (USH1B), as well as non-syndromic, autosomal dominant and autosomal recessive hearing loss (DFNA11 and DFNB2, respectively) (34–36) and familial presbycusis (37). Gene editing enables precise correction or selective inactivation of pathogenic Myo7a mutants, potentially overcoming the limitations from gene replacement approaches, including cargo packaging constraints and uncontrolled transgene expression (38). Therefore, EV-mediated delivery offers a safe and non-viral alternative to AAVs for gene editing in the inner ears, which reduces the long-term risks on insertional mutagenesis and associated immune response. Together, EV-mediated gene editing for correcting dominant Myo7a mutations holds the great therapeutic potential.. Our previous research showed that delivery of Myo7a^WT cDNA prevented hearing loss in Myo7a^WT/Sh1 mice with one mutant (Shaker1, or Sh1) allele (a loss-of-function mutation) and one wild-type (WT) allele (38). We hypothesize that timely removal of pathologic Myo7a alleles by EVs delivery could prevent the progression of hearing loss from the Myo7a^WT/Sh1 heterozygous mouse model.

Herein, we developed an efficient and scalable EV loading platform - Microfluidic Droplet-based EV Electroporation System (μDES) with continuous flow-based droplet generation and low-voltage electroporation, for developing EV mediated gene therapy which is demonstrated in vivo by treating a mouse model of progressive autosomal dominant hearing loss caused by Myo7a mutation. This platform could offer an opportunity to enable the loading of mutation-specific gene editing machineries into EVs, in turn, enabling customization of treatment based on patients’ heterogeneous mutational backgrounds, which holds potential for overcoming current challenges in gene therapy.

Results

High throughput and highly efficient EV electro-transfection via the μDES platform

We developed the μDES platform for enhancing EV transfection efficiency, loading capacity, and throughput. The μDES platform (Fig. 1A) streamlines droplet generation as numerous bioreactors with continuous-flow electroporation controlled by a low voltage DC power supply. Within the droplets, the cargo transportation into EVs relies on the electrical mobility of the cargo themselves, electric flux, and concentration gradient. On such a small scale, mass transport for crossing transient pores in the EV membrane can be completed in milliseconds (Fig. 1B) (39, 40). A homogeneous electric distribution can be achieved within water-in-oil droplets, which carry low conductive buffer as the aqueous phase (39, 41, 42). Multi-physics COMSOL simulation (Fig. 1C-E) illustrates the uniform electric field distribution across individual microdroplets and the continuous flow dynamics within the microchannel. The electric field lines extend from the high- to low-voltage electrodes (Fig. 1C), ensuring consistent pulse exposure for each droplet, while the velocity profile confirms stable and evenly distributed flow (Fig. 1D), with faster velocity at the channel center and slower flow in the enlarged electroporation chamber. The power distribution map (Fig. 1E) further shows that the electric field intensity is concentrated within the electroporation chamber, where EV loading occurs, while the surrounding regions experience minimal field exposure. Together, these simulations demonstrate that the μDES design enables uniform and localized electric field application under controlled flow conditions, supporting efficient and reproducible EV electroporation. The device uses a 3D printer as detailed in the Supplementary Materials and Methods and fig. S1) The continuous generation of uniform droplets at a fast speed enables large-scale processing of EVs for cargo loading, which has been demonstrated by collecting droplets in a large petri dish and 2 mL Eppendorf tubes (Fig. 1F) in uniform droplet sizes (Fig. 1G). Using fluorescent polystyrene beads of 100 nm size as a reference for nanosized EVs, we were able to demonstrate consistent and efficient encapsulation of polystyrene beads within droplets (Fig. 1H). The low power electroporation did not alter the droplet size and quantity (Fig. 1I and J).. Following electro-transfection, the cargo-loaded EVs in the aqueous phase can be harvested via low-speed centrifugation phase separation to fully remove the oil phase (Fig. 1A). For efficient removal of excessive sgRNA:Cas9 RNP cargo from loaded EVs, we employed Ni Sepharose high performance agarose beads to selectively capture the His-tagged RNPs in the suspension sample solution.

Fig. 1. High throughput and efficient EV electro-transfection via the μDES platform.

(A) Images of the μDES device with an illustration of continuous-flow droplet generation, droplet-based electroporation, and cargo-loaded EV harvesting and purification. Created with BioRender.com (B) Schematic illustration of droplet-based electroporation for EV cargo loading under uniform electric field distribution as demonstrated by multi-physics COMSOL simulation (C). (B) Created with BioRender.com (D) Multi-physics COMSOL simulation analysis of the continuous flow profile and electric field profile (E) showing uniformity for precision control. (F) Image depicting the large-scale collection of high throughput droplets containing cargo loaded EVs. (G) Microscopic image of continuous flow generated droplets. (H) Fluorescence microscopic image of 100 nm fluorescent polystyrene beads encapsulated within generated droplets, demonstrating consistent and efficient EV encapsulation within droplets. (I and J) The droplet size is consistent and uniform before (I) and after (J) electric field application for transfection (30 V). The insert scale bar is 1000 μm. (K) Evaluation of EV cargo loading rates among different transfection methods using fNTA. Data are from three or more independent experiments and shown as mean ± SD. (L) The recovery rate of μDES produced EVs compared to conventional cuvette electro-transfection, which requires 1500V for electroporation. Data are shown as mean ± SD from three independent experiments. (M) Quantitative measurements of Cas9 proteins from loaded RNP EVs normalized by EV particle number compared among different transfection methods. Values are means of two or more technical replicates from three independent experiments. Data is shown as mean ± SD. (N) Quantitative PCR analysis of sgRNA copy number from loaded RNP EVs normalized by total EV RNAs. The electro-transfection was done by using the μDES platform in eight independent experiments (Each has three to four technical replicates). The native EVs and μDES prepared EVs without RNP cargos both served as negative control groups. EGFP-Cas9 were pre-assembled with sgRNAs at a 1:2 molar ratio before loading. EVs were purified from HEI-OC1 hair-like cell cultures and mixed with sgRNA: EGFP-Cas9 RNPs in electroporation low conductivity buffer. A final concentration of ~10¹⁰ /mL of EVs was used for the Neon cuvette electroporation control group, μDES system, and chemical transfection methods. Data are represented as the mean ± SD. One-way ANOVA with Dunnett’s multiple comparison test was performed for statistical analysis. Scheme was created by BioRender.com.

The platform achieved up to 80.3% cargo loading rate for large proteins into EVs including sgRNA:Cas9 RNP complexes, showing higher EV loading efficiency than other conventional transfection methods including direct incubation, lipofection, and cuvette electro-transfection (Fig. 1K). We also compared the EV recovery rate after electroporation using μDES and cuvette-based methods (Fig. 1L). The μDES platform achieved a higher mean recovery rate 88% at 30V voltage compared to 79% from cuvette-based method. The cleavage bioactivity of sgRNA:Cas9 RNPs as illustrated in fig. S2A was not impacted by introducing the electroporation process (fig. S2B and 2C). Band intensities of the cleaved fragments were quantified by ImageJ by calculating the ratio of cleaved products to total input and comparing the control group (no electroporation) and the cuvette-based group (fig. S2B) with the μDES-processed group (fig. S2C). The quantification results (fig. S2D) showed that the gene-editing activity of μDES-processed sgRNA:Cas9 RNPs was preserved, with up to 93% cleavage activity relative to the control group. We quantified EV loading of EGFP-Cas9 proteins following μDES using fluorescent nanoparticle tracking (fNTA) normalized to the particle number per injection volume (Fig. 1M), demonstrating a more than 5-fold higher protein signal per 10⁸ EVs from the μDES group compared to conventional cuvette electro-transfection, 4-fold for Saponin and 3-fold for Lipofectamine method. To characterize the reproducibility of the μDES platform’s loading performance, we used a qPCR-based standard curve of chemically modified sgRNA-1 (fig. S3A) with the corresponding amplification curve (fig. S3B) and verified specific amplification of sgRNA after the assay (fig. S3C) with gel electrophoresis assay. This standard curve was then applied to quantify sgRNA loading in RNP-EVs in Fig. 1N. The loading capacity and reproducibility from ten independent μDES produced RNP-EVs were shown in Fig. 1N compared to native EVs and μDES-prepared EVs without RNP cargos served as negative control groups. The MicroWestern characterization (fig. S3D) also confirmed the successful loading of Cas9 proteins into EVs following μDES processing. The stability analysis of the resulting RNP EVs stored in −80 ⁰C for 28 days showed 84% retention in terms of EV particle numbers (fig. S4A and 4B) and 67% retention of sgRNA from RNP-EVs (fig. S4C). A higher protein-to-sgRNA ratio may help mitigate sgRNA degradation, as proteins are generally more stable molecular forms. These characterizations of RNP-EV formulation established the foundation for subsequent studies targeting Myo7a mutations in vitro and in vivo. In summary, the μDES platform demonstrated high EV cargo loading efficiency, throughput, consistency, and EV stability.

High biocompatibility of μDES-produced RNP-EVs

The μDES platform maintained a consistent flow rate and uniform electroporation period for each droplet, which is essential for the reproducible and efficient cargo loading into EVs. We first evaluated whether the μDES platform preserved key physicochemical properties of EVs from HEI-OC1 cells, a previously reported mouse hair-like cell line (43), after electroporation for the proof-of-concept. Size distribution (Fig. 2A, fig S5A to 5D) and zeta potential (Fig. 2B) of EVs were analyzed by nanoparticle tracking analysis (NTA). The size distribution of EVs processed by the μDES platform and cuvette electroporation was comparable to that of native EVs, whereas EVs prepared using saponin or lipofection methods showed a slight decrease in mean particle size. Size distribution from fNTA assessed EVs from different resources is comparable to NTA in scatter mode assessed EVs indicating detectable RNP-EVs exhibit a comparable size profile (fig. S5A to 5D). The zeta potential of EVs in Fig. 2B showed that surface charge of EVs treated by μDES remained comparable to that of native EV controls, indicating that μDES did not alter surface charge of EVs. Yet cuvette treated EVs showed notable decrease while chemical reagents including saponin and lipofectamine showed increase regarding zeta potential. The protein contents (Fig. 2C) were then assessed with automated MicroWestern assays with the same total protein input quantified by MicroBCA assays. . The μDES produced RNP EVs exhibited similar band intensity of essential EV-associated proteins including CD81, TSG101 and Alix, relative to native EVs. CD81 is a canonical EV surface marker, whereas TSG101 and Alix are marker proteins associated with the endosomal sorting complex required for transport (ESCRT) pathway. Notably, μDES produced RNP EV showed a distinct amount of encapsulated Cas9 proteins (Fig. 2C). We also assessed CD81, TSG101, Alix in EVs in which RNPs were loaded using 0.2% saponin, lipofectamine 3000 (Lipo3000) or commercial electroporation (Neon), respectively. These RNP-EVs displayed reduced abundance of CD81, Alix, and TSG101 compared to native EVs with no sgRNA:Cas9 RNP loaded (fig. S6A to 6C). The reductions in TSG101 and Alix were also observed for MSC-EVs which has great clinical translation potential (fig. S7A and 7B). Immuno-gold nanoparticle (AuNP) staining and TEM imaging showed the same cup-shaped morphology and surface expression of CD81 between RNP-EVs and native EVs, without obvious adsorption of sgRNA:Cas9 RNPs on the EV surface (Fig. 2D), suggesting that most RNPs were loaded inside of EVs. We next evaluated the cellular uptake efficiency of μDES-produced EGFP-RNP MSC EVs compared with EVs prepared by other loading methods, including Neon electroporation, saponin and lipo3000 and RNP-LNP as a positive control shown in Fig. 2E. HEI-OC1 cells were incubated for 1 hour with equal concentrations of EGFP-RNP EVs and cell uptake was assessed by flow cytometry. It showed that μDES group exhibited the highest percentage of EGFP-positive cells in Fig. 2E, as confirmed by quantitative analysis of EGFP RNP uptake signal in Fig. 2F. We obtained similar observation of highest cell uptake of using EGFP-RNP HEI-OC1 EVs (fig. S8A and 8B). Furthermore, a quantitative assessment of cell viability using EVs from both hUC-MSCs and HEI-OC1 cells showed that unlike the LNP groups, both RNP MSC-EVs and RNP HEI-OC1 EVs demonstrated the ability to promote growth of HEI-OC1 cells (Fig. 2G), suggesting superior cell biocompatibility. Both the μDES-produced RNP-EVs and native EVs showed a similar trend of improving HEI-OC1 cell viability compared with untreated controls in Fig. 2G. We observed an increase in cell nuclei and phalloidin after ~60-hour incubation with mCherry mRNA MSC-EVs compared to mCherry mRNA LNPs (fig. S9). Next, we incubated μDES-produced mCherry mRNA MSC-EVs with HEI-OC1 cells to assess cell growth, compared to mCherry mRNA LNPs. At the same cellular seeding density, the MSC-EV-exposed cells exhibited visibly increased growth at 60 hours, allowing sufficient time for mCherry detection and assessment of long-term effects, compared to the RNP LNP-treated cells. The cellular uptake behavior of μDES-produced RNP*^EGFP EVs (EGFP-fused sgRNA:Cas9 RNP MSC-EVs) was assessed by imaging analysis at the 1-hour time point which exhibited higher uptake for cytoplasmic release and gradual entry into the nucleus (indicated by the white arrows in Fig. 2I), compared to the RNP LNP group (Fig. 2I and fig. S10). The data suggested that μDES-produced EVs may provide improved delivery of sgRNA:Cas9 RNP, associated with improved biocompatibility.

Fig. 2. Characterization of μDES RNP EVs demonstrates high biocompatibility.

(A) Characterization of EV size and zeta potential (B) after transfection via different methods. The original native EVs serve as the control. Three or more technical replicates were measured for each experiment and means were used in the plot. Data are shown as mean ± SD. (C) MicroWestern blotting analysis of essential EV proteins (CD81, TSG101, Alix, Cas9) derived from μDES produced RNP HEI-OC1 EVs in serial dilution, with native EVs as the control group. (D) Immune gold nanoparticle (AuNP) staining TEM imaging analysis on μDES produced RNP HEI-OC1 EVs with native EVs as the control, for molecular detection in terms of CD81 surface marker expression and Cas9 surface adsorption. Scale bar = 200 nm. (E) Flow cytometry analysis on HEI-OC1 cellular uptake of RNP LNPs, MSC EVs, and EGFP-RNP MSC EVs engineered by different methods. ~40,000 HEI-OC1 cells were treated with the same concentration of EGFP- RNPs (~1 μM) in particles for 1 hour. (F) Flow cytometry mean fluorescent intensity (MFI) from the HEI-OC1 cells treated with RNP*^EGFP MSC-EVs using different loading methods. LNPs were used as the control. Values and error bars represent mean ± SD. (G) Biocompatibility analysis using PrestoBlue assay with HEI-OC1 hair-like cells (~10⁴), dosed with the LNP group (w/o RNP), MSC-EV group (w/o RNP), and HEI-OC1 EV group (w/o RNP) in ~10⁹ particles. Values represent Min to Max with all data points from three or more biological replicates. (H) Confocal images of HEI-OC1 cells (10,000 cells) after 60 hours of incubation with μDES produced mCherry mRNA MSC EVs (20 μg in 6 × 10¹⁰ EV) and mCherry mRNA LNPs as the control group (20 μg in ~10¹⁰ LNPs). Scale bar = 100 μm. (I) Confocal imaging analysis of particle uptake in HEI-OC1 cells dosing with RNP*^EGFP LNPs and RNP*^EGFP MSC-EVs in ~10⁹ particles. The white arrow indicates the cytoplasmic release and entry into the nucleus. Scale bar = 10 μm.

Gene editing design for allele-specific editing of pathologic Myo7a^Sh1

To target autosomal dominant mutation c.Myo7a1505G>C mutation in the Myo7a^WT/Sh1 mouse model of autosomal dominant progressive hearing loss, two different sgRNA sets associated with Streptococcus pyogenes Cas9 were designed for the in vitro and in vivo editing investigation. In each sgRNA set, we designed full-length and truncated forms of Myo7a^Sh1-targeted sgRNAs (Fig. 3A and table S1) to individually form the sgRNA:Cas9 RNP complexes for electro-transfecting primary fibroblast cells with the same transfection efficiencies (fig. S11A). The in vitro gene editing results of T7 endonuclease 1 assay demonstrated efficient cleavage in the primary Myo7a^WT/Sh1 and Myo7a^Sh1/Sh1 fibroblasts with no detectable cleaved bands observed in Myo7a^WT/WT fibroblasts (fig. S11B to 11E) which is further confirmed with the quantification of band intensity shown in fig. S12A. The indel percentage of sgRNA-1 was reduced to about 25% in Myo7a^WT/Sh1 fibroblast cells, from about 45% in homozygous Myo7a^Sh1/Sh1fibroblasts, suggesting selective targeting of the Myo7a^Sh1 allele for editing in vitro. Sanger sequencing analysis by TIDE (fig. S12B and 12C) (44) and next generation sequencing (NGS) analyses by CRISPResso2 (fig. S12D) (45) were performed to confirm the allelic cleavage specificity and editing efficiency. The sequence percentage showed that sgRNA-1 and sgRNA-2 led to higher overall cleavage activity than their respective truncated versions (Fig. 3B and fig. S12D). Unedited Myo7a^Sh1 sequences were reduced to about 20% and the edited sequences at the Myo7a^Sh1 allele accounted for the remaining about 80% when fibroblasts were treated with sgRNA-1:Cas9 RNP (Fig. 3B). To quantify the targeting specificity of sgRNA designs, we sorted the mutation pattern (5’-CCG-3’) and wild type pattern (5’-CGG-3’) separately in the edited DNA sequences. By designing PAM sequences of sgRNAs which have closer proximity to the Sh1 mutation (46), about 95% specificity in the edited sequences pool was achieved for in vitro editing, and the wild-type allele was mostly intact after editing from sgRNA-1:Cas9, sgRNA-2:Cas9, and their truncated sgRNAs (Fig. 3C). Indel profiling revealed that most sgRNA:Cas9 RNP-induced variants were deletions for sgRNA-1 and insertions for sgRNA-2 (Fig. 3D and 3E). The sequence table of sgRNA-1 (Fig. 3F) and sgRNA-2 (fig. S12F) further confirmed the indel profiling analysis. The sequence table of Tru-sgRNA-1 and Tru-sgRNA-2 showed a lower edited sequence percentage but similar deletion and insertion to their full-length counterparts (fig. S12G and 12H). The editing efficiency of sgRNA-2:Cas9 RNP at the Myo7a^Sh1 allele was slightly lower than that of sgRNA-1:Cas9 RNP. The allele-specific editing efficiencies of the truncated versions of sgRNA-1 and sgRNA-2 were halved compared to their full-length counterparts (Fig. 3B), which aligned with the global editing results (fig. S11B and H) (8). and F). The most common mutation type was either a single base deletion or a single base insertion causing frameshift mutations in the coding sequence of MYO7A, which could result in disruption of the translation of MYO7A^Sh1 proteins (fig. S12I) (47). The results also indicated that in frame mutations may generate truncated MYO7A mutants with less impact on the integrity of open reading frame, such as 1-residue and 6-residue deletions (Fig. 3F and fig. S12I). It is also important to highlight that sgRNA-1 and Tru-sgRNA-1 contained fewer frameshift mutations that possibly interfered with gene integrity than that from sgRNA-2 and Tru-sgRNA-2 (Fig. 3D and fig. S12E). Taken together, these data suggested that full-length sgRNA-1 and sgRNA-2 provided better allele-specific editing properties than their truncated version, thereby prompting further in vitro and in vivo validation of RNP-EVs. Because sgRNA-1 led to slightly higher editing efficiency and less impact on genome integrity at the Myo7a^Sh1 allele, sgRNA-1 was used for in vivo validation.

Fig. 3. Design and characterization of a sgRNA:Cas9 RNP system for allele-specific editing on pathologic Myo7a^Sh1.

(A) Two different sgRNAs and their truncated version were designed to target the allele harboring Myo7a^Sh1 mutation (c. Myo7a1505G>C; p.R502P). Scheme was create with BioRender.com (B) The percentages of unedited and edited sequences in Myo7a^Sh1 reads were analyzed using CRISPResso2 from gDNA isolated from fibroblasts treated with sgRNA-1, Tru-sgRNA-1, sgRNA-2, and Tru-sgRNA-2 RNPs. Data are from three biological replicates and values represent mean ± SEM. (C) Indel percentages of Myo7a^Sh1 allele and Myo7a^WT allele in the total edited sequences were analyzed from gDNA of fibroblasts treated with sgRNA-1, Tru-sgRNA-1, sgRNA-2, Tru-sgRNA-2 RNPs. Data are from three biological replicates and values represent mean ± SD. (D) Pie chart analysis of mutation types in gDNA from fibroblasts treated with sgRNA-1:Cas9 RNP. (E) Representative distribution of indel size of edited sequences from sgRNA-1 and sgRNA-2. Negative values mean deletions and positive mean insertions. (F) Representative allele frequency tables of Myo7a^WT and Myo7a^Sh1 alleles from NGS analysis of gDNA from fibroblasts treated with sgRNA-1 RNP. Reference sequence is the mutant Myo7a^Sh1 allele. (G) Quantification of indel percentage of Myo7a^WT and Myo7a^Sh1 allele in gDNA from fibroblasts treated with sgRNA-1:Cas9 RNP MSC EVs produced by μDES and Cuvette electroporation with RNP-LNP as a control and untreated cells as negative control(-Ctrl.). The dosing concentration is 10⁷ EVs per fibroblast cell, while LNPs were using the same concentration of RNP (~1 μM) instead of particle number. All graphs show the mean ± SEM and three individual biological replicates. Two-way ANOVA with Dunnett’s correction was performed for statistical analysis and multiple comparisons. ns, P > 0.05; * P < 0.05; ** P < 0.01; *** P < 0.001; **** P < 0.0001.

For further in vitro editing analysis, we employed EVs for delivering sgRNA-1:Cas9 RNPs (~1 μM) to primary Myo7a^WT/Sh1 fibroblast cells. A 5.24% editing percentage on the Sh1 allele was achieved using μDES produced RNP MSC-EVS, which was significantly (P < 0.05) higher than the 1.64% achieved by conventional cuvette electro-transfected RNP MSC-EVs and comparable to the LNP-mediated editing efficiency which is widely used as a standard non-viral platform for gene-editing delivery (Fig. 3G). Low editing efficiencies were observed in all groups on Myo7a^WT allele, as expected (Fig. 3G). Sequence table of sgRNA-1:Cas9 RNP showed most abundant mutation type is 1bp deletion (fig. 13A). Both types of transfection resulted in similar mutation patterns in Myo7a^Sh1 (fig. S13B and 13C) and a similar abundance of frame shifting and in frame mutations in the coding sequence of the MYO7A protein (fig. S13D and 13E), indicating consistent gene editing behavior. Therefore, μDES-produced RNP MSC-EVs demonstrated good editing efficiency, suggesting the potential for precise gene editing in vivo.

Arrest of progressive hearing loss by μDES-produced RNP-EVs in adult mice

WT C57BL/6J Myo7a^WT/WT and heterozygous Myo7a^WT/Sh1 mice initially have normal hearing. However, heterozygous Myo7a^WT/Sh1 animals will gradually lose hearing ability with substantial impairment evident by testing at 6 months of age(38). Therefore, we considered Myo7a^WT/Sh1 heterozygotes to be an ideal model in which we could assess whether μDES-produced RNP MSC-EVs could serve as a gene editing delivery platform to the inner ear. For clinical translation potential, we characterized both HEI-OC1 EVs and MSC-EVs in terms of cellular production rate (fig. S14A), EV size (fig. S14B), and polydispersity index (PDI) (fig. S14C). MSC-EVs exhibited higher cellular production rate while having similar size and PDI as HEI-OC1 EVs (fig. S14A to 14C). Analysis of proteome identifications (PRIDE) database(49) revealed that HEI-OC1 EVs contain 28 unique proteins, whereas MSC-EVs contain over 2,000 (fig. S14D) from either umbilical cord or bone marrow (fig. S14E to 14G). Because human umbilical cord-derived MSC EVs (hUC MSC-EVs) have been reported to have higher manufacturing potential (50), we chose to primarily use hUC-MSC-EVs for sgRNA:Cas9 RNP delivery in vivo as a clinically translatable route.

To characterize the biodistribution of EVs in the inner ear, three groups of Postnatal day 30 (P30) Myo7a^WT/Sh1 mice were injected with RNP*^EGFP LNP, μDES-produced RNP*^EGFP MSC-EVs, or RNP*^EGFP HEI-OC1 EVs, respectively via the posterior semicircular canal (PSCC), which delivers into the perilymphatic space of the inner ear (Fig. 4A). By two hours, the μDES-produced RNP MSC-EVs and RNP HEI-OC1 EVs displayed superior tissue delivery, particularly to the targeted outer and inner hair cells, compared to RNP LNPs, which were limited to auditory nerve bundles with no obvious delivery to inner or outer hair cells (Fig. 4B). We next administrated μDES-produced RNP MSC-EVs (10⁹ particles in about 1 μL) and RNP LNP as a control group with the same dosage via the PSCC into the left ears of P30 Myo7a^WT/Sh1 mice, whereas the right ears remained untreated as controls. Four weeks after injection, the organ of Corti was extracted for Sanger sequencing, NGS, and qPCR-based validation of the modification of both genomic DNA and mRNAs. qPCR revealed that both RNP LNPs and RNP MSC-EVs caused disruption to genomic compared to the untreated controls (Fig. 4D). Quantification of the Sh1/WT ratio showed that RNP MSC-EVs induced more substantial disruption of the Myo7a^Sh1 genomic DNA than RNP LNPs (Fig. 4E). Disruption of the genomic DNA was further confirmed by NGS, which also demonstrated measurable editing response from RNP MSC-EV-treated Myo7a^Sh1 (fig. S15A). Measurement of Myo7a^Sh1 mRNA expression using qPCR showed that treatment with the RNP MSC-EVs led to a significant (P < 0.0001) reduction of the Myo7a^Sh1 expression compared to untreated mice and compared to RNP LNP-treated mice, which showed no changes in gene expression (Fig. 4F). There were no significant (P > 0.05) changes in the expression of the Myo7a^WT within treated organs of Corti in any groups (Fig. 4G and fig. S15B). The μDES-produced RNP HEI-OC1 EVs showed detectable increase in modified sequences on the Myo7a^Sh1 gene (fig. S15C and 15D), indicating the feasibility of in vivo gene editing using μDES-produced RNP EVs. The NGS data showed a higher indel percentage at Myo7a^Sh1 from μDES-produced RNP EV groups than that from the RNP LNP group (Fig. 4H), whereas the MSC EV only group did not introduce any gene editing activities (fig. S15B). The gene editing differences between EVs and LNPs could potentially be due to the reduced tissue penetration ability of the LNPs used for delivery (Fig. 4B). Fig. 4I, featuring the top 5 edited sequences, illustrates the diversity of mutations induced by μDES-produced RNP-MSC EVs in vivo. The editing efficiency was higher at the mRNA level, especially given that Myo7a mRNA was highly transcribed in hair cells, leading to a higher signal (10, 51). Whole-mount images of organs of Corti further revealed the localization of RNP*^EGFP-loaded EVs to OHCs and IHCs (Fig. 4J). Confocal imaging of midmodiolar sections showed that MSC-EVs delivered EGFP cargo throughout the cochlea, with dense fluorescence detected in all turns from basal to apical (fig. S16A). Anti-MYO7A labeling of hair cells in the inner ear also revealed a strong colocalization of MYO7A and EGFP in the hair cells of both auditory and vestibular systems (fig. S16B).

Fig. 4. Characterization of μDES loaded RNP EVs for specific gene editing in vivo.

(A) Schematic illustration of administration route and experimental image of PSCC in the isolated mouse inner ear highlighting the PSCC site within the inner ear. The injection site is clearly outside of the region of the organ of Corti where hair cells are located, showing the minimally invasive injection. Created with BioRender.com (B) Confocal imaging analysis of the biodistribution in the organ of Corti via PSCC injection of RNP*^EGFP LNP (left panel), μDES produced RNP*^EGFP MSC EVs (middle panel), and RNP*^EGFP HEI-OC1 EVs (right panel) into P30 Myo7a^WT/Sh1 mice ears individually. Scale bar = 30 μm. (C) The injection schedule and timeline for the in vivo investigation of gene editing efficiency through the comparison between RNP LNPs, RNP MSC-EVs, and MSC-EV only groups. hr means 1 hours; 1w means 1 week. (D) Relative fold change analysis of gDNA for Myo7a^WT and Myo7a^Sh1 alleles using the Tfrc gene as the reference gene. Data represents four or more biological replicates and are mean ± SD. (E) Relative fold change (2^-ΔΔCt) analysis of the Sh1/WT ratio in the gDNA isolated from organ of Corti in Myo7a^WT/Sh1 mice treated with RNP MSC EV, RNP LNP and untreated contralateral sides as the control. Data are presented as mean ± SD with four or more biological replicates. ns P > 0.05 and * P < 0.05 are derived from One-Way ANOVA with Dunnett’s multiple comparison. (F) Relative fold change (2^-ΔΔCt) analysis cDNA of Myo7a^Sh1 in the mRNA isolated from organ of Corti in in Myo7a^WT/Sh1 mice treated with RNP MSC-EV, RNP LNP and untreated contralateral sides as the control. Data represents four or more biological replicates and are mean ± SD. Statistical analyses were performed using unpaired t test with Welch’s correction to assess differences between treated and untreated groups. ns, P > 0.05; * P < 0.05, ** P < 0.01; *** P < 0.001; **** P < 0.0001. (G) Relative fold change (2^-ΔΔCt) analysis of cDNA of Myo7a^WT in mRNA isolated from organ of Corti treated with RNP MSC-EV (+) and untreated contralateral sides as control (−). Data represents four biological replicates and are mean ± SD. Statistical analyses were performed using unpaired t test with Welch’s correction to assess differences between treated and untreated groups. ns, P > 0.05; * P < 0.05, ** P < 0.01; *** P < 0.001; **** P < 0.0001. (H) Indel percentages of Myo7a^Sh1 and Myo7a^WT alleles analyzed by CRISPResso2 of cDNA from mRNA isolated from the organ of Corti treated with RNP MSC EV, RNP LNP, native EVs and untreated contralateral sides serving as “Before Treatment” group. Data are presented as mean ± SEM from three or more biological replicates. (I) Mutant sequence table of Top 5 editing sequences from NGS analysis of cDNA from mRNA isolated from the organ of Corti across independent experiments. Reference sequence is the mutant sequence, Sh1. Del means deletion, Sub means substitution while the number means the changes in nucleotides in each sequence. (J) Whole-mount confocal images of the organ of Corti showing cellular localization of EGFP labeled RNP cargos. Merged fluorescence signals of the organ of Corti highlight nuclear staining with Hoechst (blue), cytoskeleton through F-actin (red), and the delivered EGFP cargo (green) which is more enriched in the IHCs. Scale bar = 50 μm.

We further monitored hearing ability 6 months after treatment via. auditory brainstem response (ABR) measured with threshold in SPL (sound pressure level) defined so that at least one of the waves could be identified in 2 or more repetitions of the recording (Fig. 5A and 5B). P30 mice within the wide range of sound frequency from 4 kHz to 32 kHz, displayed about 20 dB better hearing thresholds in the left ears that were treated with μDES-produced RNP MSC-EVs, compared to the untreated right ears, nearly comparable to P30 wild type controls (Fig. 5B to 5D). In contrast, the groups treated with RNP LNPs and MSC-EVs only did not experience prevention of hearing loss across 4 kHz to 32 kHz (Fig. 5C, fig. S17A and 17B). . Neonatal heterozygous mice treated at P4 with μDES-produced RNP HEI-OC1 EVs also displayed significant (P < 0.001) prevention of hearing loss, as compared to P30 wild-type mice as the baseline (fig. S17C and 17D). Myo7a^WT/Sh1 mice at 180 days (P180) exhibited notable hearing loss while Myo7a^WT/W mice showed only slight hearing threshold increase for the wide type. Treatment of μDES-produced RNP MSC-EV significantly (P < 0.001) prevented hearing loss (Fig. S17E).

Fig. 5. In vivo genome editing for arresting progressive hearing loss.

(A) Schematic illustration of animal auditory testing methods. Created with BioRender.com (B) Representative auditory brainstem responses (ABRs) recorded from heterozygous Myo7a^WT/Sh1 mice left ear treated with μDES produced RNP MSC EVs in month 6, and right ear without treatment. The red arrow showed the hearing threshold. (C) Heterozygous Myo7a^WT/Sh1 mice (n = 6, P30) were treated with RNP MSC EVs, native EVs, or RNP LNPs via PSCC injection on the left ear (~10⁹ particles in ~1 μL with ~0.9 μM RNP proteins), with the right ear as an untreated control. Wild-type normal-hearing mice at P30 served as baseline control of good hearing. The ABR hearing was monitored 6 months after injection from sound frequency 4 kHz to 32 kHz. The data are presented as a box-and-whisker plot showing Min to Max values. Each point represents ABR threshold for each animal. (D) The before-and-after comparison of hearing thresholds from Shaker-1 heterozygous Myo7a^WT/Sh1 mice (n = 6, 30 days old) revealed notable improvement from RNP MSC-EV-treated ears across sound frequencies from 4 kHz to 32 kHz compared to P30 C57BL/6J wild-type mice as the baseline. Statistical analysis was performed using Two-way ANOVA with Dunnett’s multiple comparison test to assess the differences between three groups. ns, P > 0.05; * P < 0.05; ** P < 0.01; *** P < 0.001; **** P < 0.0001. The before-after graph displays data from different biological replicates using distinct symbols, with lines connecting paired measurements. (E) Confocal images from immunohistochemistry oxidative stress analysis of markers 4-HNE and 3-NT stained in the inner ear hair cells in organ of Corti, at 3 and 6 months of age mice (n = 6) from both hearing impaired heterozygotes (Myo7a^WT/Sh1 ) and normal hearing wide type (Myo7a^WT/WT ), as well as the RNP MSC-EV treated group. (F) Quantitative analysis of the expression of oxidative stress markers 3-NT and 4-HNE (G) in the inner ear hair cells from organ of Corti, at 3 and 6 months of age (n = 6) from both hearing disable heterozygotes (Myo7a^WT/Sh1) and normal hearing wide type (Myo7a^WT/WT ) mice without any treatment, compared to RNP MSC-EV treated group. Data are presented as mean ± SEM. P < 0.0001 was performed by unpaired t test was used to compare 6m Treated and untreated 6m Het groups.

Our previous work has shown a correlation between cochlear oxidative stress and genetic hearing loss (52, 53). Therefore, we evaluated the Myo7a^WT/Sh1 mouse inner ear for the presence of the oxidative stress markers 4-hydroxynonenal (4-HNE) and 3-nitrotyrosine (3-NT). Myo7a^WT/Sh1 mice showed oxidative stress marker labeling throughout the organ of Corti and a higher expression of both oxidative stress markers at 3 and 6 months of age in the inner ear hair cells (Fig. 5E to 5G), in contrast to WT mice. In treated mice, expression of both 4-HNE and 3-NT was significantly (P < 0.0001) reduced at 6 months of age, associated with preserved ABRs after gene editing. Overall, the results indicated that EV-mediated sgRNA:Cas9 RNP delivery and in vivo editing could represent a promising approach with potential for clinical translation.

Assessment of in-vivo on-target specificity and safety of gene editing

To assess the on-target specificity and potential off-target effects of gene editing, we first conducted a computational off-target analysis using guide RNA design checker from Integrated DNA Technology (IDT), which predicted the top 18 potential off target sequences, ranked by off-target score (table S2). We then used PrimerQuest from IDT to design primers for the amplification of the first nine predicted off-target sequences (table S3). The amplicons were subsequently used for NGS and CRISPResso2 analyses. The indel percentage was calculated by aligning the amplicons with the corresponding sgRNA sequences in table S2 using the CRISPResso2 pipeline. The analysis showed that on-target editing efficiencies of 29.65% with off-target editing response of 0.067% at Off-target 5 and lower than 0.016% at all other predicted off-target sites (Fig. 6A). To further quantify genome-wide off-target editing, we employed the CIRCLE-seq pipeline, modified from the reported approach(54) to include additional PCR and qPCR to amplify and confirm the on-target sequences after random enzymatic digestion of gDNA isolated from the Myo7a^WT/Sh1 mice (Fig. 6B). The sequencing results were then processed through the CIRCLE-seq analytical pipeline, which identified off-target reads on the order of one to two dozen each, whereas on-target sequences were enriched to almost one million (Fig. 6C). The sequences identified by CIRCLE-seq did not overlap with those sequences in the initial computational analysis, however, which highlights the essential role of CIRCLE-seq in uncovering potential off-target editing sites (54). We utilized electroporation to transfect sgRNA-1:Cas9 RNPs into fibroblasts isolated from Myo7a^WT/WT mice and then isolated gDNA for the library preparation of Myo7a amplicons. The resulting NGS analysis using CRISPResso2 demonstrated less than a 1% off-target editing rate (Fig. 6D).

Fig. 6. Evaluation of in-vivo gene editing on-target specificity and safety.

(A) NGS sequencing analysis of editing efficiency of 9 in silico-predicted off-target sites in gDNA from fibroblasts treated with sgRNA1:Cas9 RNP genes with the on-target site Myo7a^Sh1 allele as positive control. Data are presented as mean ± SEM. Each point represents an individual sequencing experiment. (B) Schematic representation of the genome-wide off-target analysis workflow using optimized CIRCLE-seq pipeline modified from a previous report (54). Created with BioRender.com (C) Sequence table of NovaSeq X NGS results from CIRCLE-seq analysis of gDNA isolated from Myo7a^WT/Sh1 mice. A dot indicates no mismatch between the detected sequences and the sgRNA-1 sequence. Mismatched nucleotides are aligned to their respective positions within the sgRNA-1 sequence. (D) Indel sequence analysis of the most abundant sequences generated from CRISPResso2, which is derived from wild-type fibroblasts (Myo7a^WT/WT) without treatment and electro-transfected RNP group as the control. (E) Safety assessment using wild type C57BL/6J healthy mice (Myo7a^WT/WT) (n=3) via injecting high dose RNP MSC-EVs and native EVs (~10¹⁰ particles in ~1 μL) at 2 months of age, with ABR monitoring for 6 months. The untreated mice served as the control. Data are shown as a violin plot with each point representing an individual animal.

We also assessed the safety of μDES-produced RNP MSC-EVs in healthy WT mice (Fig. 6E). The high dose RNP MSC-EVs and native EVs (10¹⁰ particles in about 1 μL) were injected into the left ear of P60 WT mice for monitoring hearing via ABR for 6 months continuously, with untreated WT mice as controls. The hearing was measured every two months per mouse after injection. We did not observe any changes in ABRs suggesting tissue damage (Fig. 6E).

To understand the potential impact of in vivo gene editing with in-frame mutations on protein structures, we employed the Alphafold2 pipeline with relaxation to characterize the structural similarity to residue 401–561 in the wild-type motor domain which harbored the MYO7A^R502P mutation along with Cas-induced mutants (fig. S18 and S19) (55). Considering that the loop region was mostly tolerant to indels in general(56, 57) and on the basis of AlphaFold2’s prediction(55, 56, 58), we expected that 1-residue deletions on the loop region were unlikely to reduce substantial structural integrity compared to the WT MYO7A domain (fig. S18). Similarly, the pLDDT of 6-residue deletion of the loop region also showed high tolerability to deletions and high foldability, whereas the binary alignment to WT MYO7A exhibited a slightly differently positioned 6-residue deletion domain (fig. S19). By reducing the percentage of the Myo7a^Sh1 allele by Cas9-mutagenesis, the Myo7a^WT allele-to-Myo7a^Sh1 allele ratio was notably both in vitro and in vivo as reported previously. Therefore, functional WT MYO7A proteins should dominate over the possible variants, if there are any as previously reported(38), which would benefit prevention of hearing impairment. Although it is still challenging to predict gradual or marked shifts in the function of final in-frame mutants based on computational toolkits (56), the residual expression and the associated high structural integrity suggest potentially limited impacts of in-frame mutants on the biological function of hair cells for hearing. Collectively, these findings support the therapeutic potential of μDES-produced MSC-EVs carrying sgRNA and Cas9 RNPs for gene editing, providing a clinically translatable strategy for treating genetic mutations in the inner ear.

Discussion

Successful utilization of EVs as therapeutics for clinical translation relies on the capacity of cargo loading and production throughput (50). Here, we have developed a μDES platform with a 10-fold enhancement of loading efficiency and more than a 1000-fold increase in processing throughput for loading RNP complexes into EVs, compared to conventional transfection methods. The EV membrane consists of a lipid bilayer with similar protein composition as the donor cell, which can be transiently permeabilized by pulsed electric shock (5). However, nanosized EVs contain more compact membrane curvatures and demonstrate strong Brownian motion (59). By using microfluidic droplet-based electroporation, we fully take advantage of efficient mass transport in confined space to maximize EV loading capacity, in turn, deriving optimal gene editing efficiency in vitro and in vivo. Fast continuous flow through of droplets also prevents direct contact of EVs with electrodes and avoids any thermal damage to EVs to retain their natural integrity and stability. In contrast to the kilovolt high voltage used in conventional workflows, this technology substantially minimizes Joule heating and high voltage risk by using only about 30 volts while still maintaining effective loading efficiency. The uniform electric field can be precisely controlled in a compact droplet space, which can greatly improve consistency and transfection efficiency. In our system, pharmaceutical-grade fluorinated oil FC40 and associated 2% fluoroSurfactant (w/w) were employed due to their low conductivity, chemical inertness and stability, and easy removal (60, 61). The inclusion of pharmaceutical grade trehalose in the aqueous phase further stabilized EVs and reduced aggregation or leakage during electroporation (62–67). Adjusting the flow conditions allowed control of droplet size and throughput, providing improved processing capacity compared with conventional cuvette-based electroporation. Such a platform can offer an amenable approach for scale-up, enabling future good manufacturing practice (GMP)-grade manufacturing of cargo-loaded EVs. This approach represents a promising step in producing gene editing cargoes for treating deafness-related mutations via non-viral vector-mediated delivery.

Our previous research found that introducing Myo7a^WT cDNA prevented hearing loss in Myo7a^WT/Sh1 mice and partially restored hearing function in Myo7a^Sh1/Sh1 mice (38). It has been reported that Myo7a-deficiency leading to gradual hearing loss requires more than 70% reduction of Myo7a in postnatal mice (33), suggesting that Myo7a is not fully haploinsufficient during adulthood. This threshold supports the rationale of our removal of Myo7a^Sh1 strategy, indicating Myo7a^WT expression alone may be sufficient to maintain auditory function. Therefore, we hypothesized that timely disruption of the mutant Myo7a^Sh1 allele in the presence of an intact Myo7a^WT allele might slow the onset or progression of hearing loss observed in Myo7a^WT/Sh1 mice. Selectively disrupting the Myo7a^Sh1 but not the Myo7a^WT allele through gene editing could therefore serve as an alternative therapeutic strategy to gene replacement that addresses the late-onset phenotype modeled by Myo7a^WT/Sh1 mice, and also provide proof of concept for the gene delivery and editing specificity of our μDES-produced RNP EV delivery platform.

Genome editing using Cas9-based gene editing technology is a promising approach to correct genetic defects underlying autosomal dominant hearing loss. Encapsulating sgRNA:Cas9 RNPs within EVs provides a transient way to target the inner ear yet potentially durable approach for precise genome modification in the inner ear. Two sets of sgRNAs were designed to target Myo7a^Sh1 allele and both full-length sgRNA1:Cas9 and sgRNA2:Cas9 RNPs demonstrated efficient editing activity. sgRNA1:Cas9 RNP showed slightly more effective editing efficiency thereby being selected for in vivo testing. Our results suggest that EVs can achieve efficient tissue penetration and functional delivery in the inner ear of Myo7a^WT/Sh1 mice with improved distribution in the inner ear related to SM102 LNP. Importantly, the observed improvement in auditory thresholds and reduction of oxidative stress markers indicate that editing of Myo7a^Sh1 by μDES-produced RNP EVs can slow the progression of hearing loss. The ability to deliver cargo into a clinically accessible space in adult animals to access both inner and outer hair cells has been challenging, and our PSCC administration route proved delivery success with minimal tissue invasion. Together, these findings highlight the feasibility of EV-mediated RNP delivery as a foundation for developing gene-editing therapies which could be targeting applied to a range of different genetic defects associated with hearing loss, with Myo7a mutations as a showcase.

Allele-specific gene editing via CRISPR/Cas9 was recently shown to successfully prevent hearing loss by disrupting the mutated transmembrane channel-like gene family (Tmc1) allele in Beethoven mice using a variety of delivery approaches including AAV vectors (8), RNP lipid complexes (8, 68), and RNP liposomes (69). Neonatal mice are often used in such studies, yet results could vary in terms of delivery distribution when comparing with the adult mice, because hearing matures after postnatal day 14. Using RNP EVs, we demonstrated successful prevention of hearing loss phenotype in adult animals as well as neonatal animals. In addition, we utilized a delivery approach (injection via the PSCC to the perilymph space) that can be translated clinically. Based on human cochlear implantation data, in the mature ear, violation of the scala media could result in loss of residual function. Thus, we believe delivery to the perilymph space to be essential (70). Interestingly, we achieved very limited delivery of RNP LNPs following PSCC-based administration to the inner ear hair cells, with the majority displayed in the basal region of the spiral ganglion. As shown by others (71), EVs derived from different organisms, even from different kingdoms, can cross-traffic, which is supported by our data using EVs from both mouse HEI-OC1 cells and human UC-MSCs in vivo, which showed effective distribution to a wide range of cells within the mouse inner ear, including inner and outer hair cells. This opens the door for more cell source options to expand EV production.

There are some limitations to this study. Although we obtained a consistent and robust prevention in hearing phenotype after the treatment of RNP EVs targeting Myo7a^Sh1. However, the function of Myo7a editing and associated potential long-term safety induced by RNP-mediated gene editing needs further investigation, particularly their potential effects on hearing function in mice models carrying different Myo7a mutations. Additionally, while μDES-produced RNP EVs demonstrated measurable editing activity, further in vivo validation and larger sample sizes are needed to confirm statistical significance and long-term efficacy. Future studies will address these aspects to strengthen the translational potential of this platform. In addition, enhancing the targeting ability of RNP EVs to IHCs and OHCs may improve editing outcomes and enable the delivery of gene editing tools such as base editors and prime editors via EVs to the inner ear. For future clinical applications, further studies are needed to develop GMP compliant RNP EVs from μDES, including conducting preclinical evaluations in nonhuman primates and human inner ear organoids. The reduced retention of intact sgRNA after 28 days and RNP after 70 days highlights the need for further optimization of formulation stability and storage conditions to enhance the platform’s long-term usability and support translational development of the platform. Despite these limitations, μDES-produced RNP EVs have potential for the delivery of gene editing therapeutics.

In conclusion, We developed a microfluidic electro-transfection platform to efficiently load EVs with gene editing RNPs, called μDES, that offers high-throughput and scalability while minimizing impact on EV integrity and RNP function. The resulting sgRNA:Cas9 RNP EVs protected against hearing loss, with low off-target editing and no evidence of ototoxicity by ABR, in a mouse model of autosomal dominant progressive hearing loss. These findings highlight the potential of EV-based delivery of gene editing and potentially other cargoes for inner ear-specific applications.

MATERIALS AND METHODS

Study design

This study aimed to develop a droplet-based microfluidic electroporation system, μDES, to encapsulate sgRNA:Cas9 RNP within EVs and investigate the gene therapy potential of μDES-produced RNP EVs with a disease model of autosomal dominant progressive hearing loss caused by a missense mutation in Myo7a (Myo7a^Sh1). We evaluated the performance of continuous droplet generation and streamlined electroporation of EVs with CRISPR RNP cargos. The electroporation conditions of μDES were optimized to minimize the impact on resultant RNP EVs while maximizing loading efficiency for gene delivery. The biophysical properties, morphology, and stability of RNP EVs were characterized with nanoparticle tracking analysis, MicroWestern blotting and immunogold TEM. The in vitro cleavage activity of sgRNA:Cas9 RNP targeting and disrupting Myo7a^Sh1 mutations was assessed within fibroblast cells isolated from mice carrying Myo7a^Sh1 using T7E1 assays, Sanger sequencing, and NGS. The μDES RNP EVs were compared to conventional transfection methods in vitro using direct incubation, chemical transfection, and cuvette-based electroporation. To assess therapeutic efficacy in vivo, μDES RNP EVs were administered via PSCC into Myo7a^WT/Sh1 mice (4-week old) with progressive hearing loss.. In vivo gene editing efficiency was evaluated at both the gDNA and cDNA levels using qPCR and NGS. The auditory function of treated mice was assessed through ABR measurements at P180 following treatment at P30. In addition, immunostaining images of the organ of Corti were obtained from P180 mice. Safety assessments, including genome-wide off target analysis, evaluation of off-target editing rates of RNP EV, and hearing phenotypes in wild-type mice, were also conducted.

Mice and ethics approval

Homozygous Shaker-1 (Myo7a^Sh1/Sh1) mice were purchased from Jackson Laboratory and maintained in a breeding colony. The animals were maintained under 12-hour light/dark cycle and received a standard rodent diet. All animal care and procedures were approved by the Institutional Animal Care and Use Committee University of Kansas University Medical Center (ACUP No. 22–05-241). Evaluation of EV-mediated gene editing was tested in Shaker-1 mice aged P4 – P180 with both sexes. All treated mice were included in the analysis. Outcomes measures included hearing testing using auditory evoked brain stem responses, and histology/immunofluorescence staining.

Detailed experimental procedures, including the use of cell lines, primary fibroblast cells and animal models are provided in the corresponding sections of the Materials and Methods or Supplementary Materials and Methods.

Materials and reagents

All chemical reagents and materials were purchased from Thermo Fisher Scientific unless otherwise specified. Modified sgRNAs were synthesized and quantified by Synthego. The Cas9 proteins from Streptococcus pyogenes used in the study NLS-Cas9-NLS (Z03469), EGFP-Cas9-NLS nucleases (Z03467) were purchased from Genscript. Anti-Cas9 antibody (Clone 4A1) was also purchased from Genscript. All DNA oligos were purchased from Integrated DNA Technologies. Collagenase IV was obtained from STEMCELL Technologies. SYLGARD^™ 184 Silicone Elastomer Kit was purchased from Dow Silicones Corp. Master Mold for PDMS device was obtained from CADworks3D. Q5 high-fidelity DNA polymerase (M0491) and AMV reverse transcriptase (M0277) was purchased from New England Biolabs. Human bone marrow-derived conditioned culture medium and human umbilical cord-derived conditioned culture medium for EV isolation were obtained from EriVan Bio, RoosterBio respectively. SM-102 LNP in ethanol was a generous gift from Fan Zhang, in the College of Pharmacy at the University of Florida. 6nm Goat anti-Mouse and anti-Rabbit IgG, Immunogold reagents were purchased from AURION.

sgRNA design and synthesis

sgRNA sequences were designed using the Custom Alt-R CRISPR-Cas9 guide RNA provided by IDT (https://www.idtdna.com/site/order/designtool/index/CRISPR_SEQUENCE) and Horizon CRISPR Design Tool (https://horizondiscovery.com/en/ordering-and-calculation-tools/crispr-design-tool). Two sgRNAs that harbored the single mutation in Myo7a gene were selected, and truncated versions of the corresponding sgRNAs that were 3nt shorter were also investigated accordingly (72) shown in Table S1. The sequences were then synthesized by Synthego Inc. with the following modifications, 2’-O-Methyl at three first and last bases, 3’ phosphorothioate bonds between first three and last two bases (73). sgRNAs were then dissolved in nuclease-free TE buffer (10mM Tris, 1mM EDTA, pH 8.0) to make the final concentration of 100 μM and stored in −20°C for future use.

Cell culture and electro-transfection with RNP complexes

Mouse primary fibroblast cells were isolated from ear punches from wild-type C57BL/6J Myo7a^WT/WT, heterozygous Myo7a^WT/Sh1 and homozygous Myo7a^Sh1/Sh1 mice. After euthanasia, a small piece of external ear was cut out and sterilized with 70% ethanol for 5 min in a sterile 15 mL conical tube. Small ear punches were then washed with Hanks’ Balanced Salt Solution (HBSS, American Type Culture Collection) without Ca²⁺ or Mg²⁺ and then dissected into smaller pieces with razor. Next, mashed tissue suspension was further treated with ~270U of prewarmed collagenase IV (STEMCELL Technologies) in a cryotube vial for 90 min at 37°C followed by 0.05% trypsin-EDTA for 20 min at 37°C. Isolated fibroblast cells after filtering through a 70-μm mesh filter were then cultured in 10% fetal bovine serum containing DMEM (Gibco) and supplemented with 1% GlutaMax with 5% CO₂ at 37°C. Fibroblast cells were validated with Sanger sequencing (Genewiz) before electro-transfection. For electro-transfection, the Neon transfection platform (Thermo Fisher Scientific) was optimized and used for loading sgRNA:Cas9 RNP into the fibroblast cells according to the manufacturer’s instruction. Briefly, 60K fibroblast cells detached from the flask with 0.25% trypsin-EDTA were resuspended in R buffer in 10μL Neon electroporation kit. The sgRNA:Cas9 complexes were then added into the electroporation solution to make 2 μM Cas9 and 6 μM sgRNA in the final electroporation buffer. The electroporation was performed with 1500 V, 30 ms with 1 pulse and then cells were gently put into a 12-well plate containing prewarmed culture medium, incubated at 37°C, 5% CO₂ for 96 hours.

Preparation and isolation of EVs

HEI-OC1 cells were subcultured in 10% fetal bovine serum supplemented DMEM (Gibco) in T-175 flasks at 33°C, 10% CO₂ to reach 60% confluence. The culture medium was then replaced with 10% exosome-depleted fetal bovine serum containing DMEM for EV production after washing cells twice with cold Dulbecco’s phosphate-buffered saline (DPBS) buffer without Ca²⁺, Mg²⁺. After 48 hours, the cell culture medium was collected in 50 mL sterile conical tubes and then centrifuged at 350 rcf, 2000 rcf, 10,000 rcf for 5 min, 20 min, 30 min respectively, to remove cells, cell debris and microvesicles. The resulting medium was then stored in −80°C or further ultracentrifuge.

hUC-MSC (C43001UC, RoosterBio) were cultured with RoosterNourish^™-MSC-XF culture medium (KT-016, RoosterBio) according to the manufacturer’s instruction until the cells reached to 80% confluency. The medium was then replaced with RoosterCollect^™-EV (M2001, RoosterBio) for EV collection after washing with DPBS twice. The cell culture medium was then collected after 48hr and followed the same steps as described below for the isolation of hUC-MSC-EV. hBM-MSC-EV conditioned medium after 48 hours incubation was collected and provided by EriVan Bio, which followed the same steps as described below for the isolation of hBM-MSC-EV.

To isolate a certain size range of EVs, 35% (w/w) sucrose in cold DPBS buffer without Ca²⁺, Mg²⁺ was prepared freshly and sonicated for 5 min to fully dissolve the sucrose bed. The resulting sucrose solution was then filtered with 0.2 μm PVDF membrane filter before use. 3 mL of 40% sucrose solution was added to form the sucrose bed beneath the 15 mL cell culture medium in polycarbonate thick-walled tubes (Thermo Fisher Scientific). After 90 min ultracentrifuge at 100,000 rcf, the 3 mL sucrose bed was transferred to a new ultracentrifuge tube and then disrupted in another 15 mL cold DPBS buffer for another 90 min ultracentrifuge. The resulting EV pellets were resuspended in 1 mL cold DPBS buffer and filtered with 0.2 μm PVDF membrane filter before quantification and size distribution analysis with ZetaView^® nanoparticle tracking analysis system. All centrifugation steps above were conducted at 4°C.

Protein profiling analysis of natural EVs

To characterize the protein compositions of natural EVs from different parental cells, we used the published datasets of HEI-OC1-EVs(74), hBM-MSC-EVs, hUC-MSC-EVs(75) for analysis and comparison. Of note, HEI-OC1-EVs are derived from a murine cell line while MSC-EVs originate from human cells. We firstly converted gene identifiers from HEI-OC1 EV datasets to human gene IDs by using the g:Profiler online tool (https://biit.cs.ut.ee/gprofiler/orth). We then filtered the HEI-OC1 EV datasets and MSC-EV datasets through EV database described previously.(75) The gene IDs from different EVs were then uploaded to the Venn Diagram webtool (https://bioinformatics.psb.ugent.be/webtools/Venn/) to generate the overlapped gene ID list. The resulting gene IDs were further analyzed by functional profiling against gene ontology database and then the top 8 gene terms and the number of unique proteins were reported and visualized using Prism Graphpad 9.0.

Droplet generator on chip

The 3D structure of the microfluidic device was designed and drawn by SOLIDWORK CAD. The resin mold for PDMS casting containing the features of 1 mm electrode channels, and 150 μm in width as nozzle for droplet generation, was printed by a μMicrofluidics Printer (CADworks3D, 30 μm resolution) shown in Fig. S1. Briefly, CAD files were opened by utility.exe that is connected to the μMicrofluidics Printer. The software setting for printing was as follows: 50 μm as thickness, 0.1 for grid size, 40% Power ratio. The microstructure was then sliced and launched for printing. The resulting resin mold after the printing was then soaked in fresh 100% Ethanol or isopropanol for 20 – 30 min 2–3 times to remove the free resin before the final UV curing step. Then the resin mold was dried with compressed Nitrogen. The soaking-drying cycles must be repeated several times until there was no shiny free resin on the microstructure. Each side of resin mold was then cured in the Creative Cure Zone (CADworks3D) for another 10 min twice for the final photopolymerization and solidification of microstructures. The resin mold was then ready for PDMS casting. PDMS was prepared using the standard 10:1 (base to curing agent) ratio. The PDMS mixture was stirred continuously at least 3 min and degassed for at least 30min before being poured into the 3D-printed molds and baked at 75°C for 3 hours. After the surface activation of molded PDMS pieces using a corona discharger and the microscope plate, PDMS molds were then assembled and permanently bound onto the microscopic plate as the droplet-based electro-transfection device. Electrodes made with 90% foot platinum and 10% iridium (SUREPURE CHEMICALS LLC) were tailored to L shape to fit into electroporation chambers designed in μDES and then carefully inserted into the electroporation sites to ensure precise alignment. The 1/16 OD, 1/32 ID tubings were then inserted into the inlets and outlets in μDES. To avoid any potential leakage that can result in the unstable flow rate in μDES, the additional PDMS mix was added to the area surrounding the inlets, outlets and electroporation sites for crosslinking to solidify as one monolithic piece before use. The device was then baked in the oven at 75°C for another 30min for the following electroporation.

COMSOL Simulation

The proposed microfluidic device was fine-tuned using the COMSOL Multiphysics software package. The device’s mathematical model involves fluid flow and electromagnetism. For both models a standard linear triangular extra-fine mesh was assigned to the geometry. To observe the geometric evolution of our droplets, we used the computational fluid dynamics system (CFD) module using the laminar two-phase flow. For this simulation, the oil phase material was defined as FC-40 with a density of 1850 kg/m³ and a dynamic viscosity of 0.0018 Pa/s. Microfluidic flows are defined by the Navier Stokes equation where $\rho$ is the density of the fluid, $u$ is the velocity of the field, $t$ is time and $P$ is the pressure field:

$$\rho \left(\frac{\delta u}{\delta t}+u·\nabla u\right)=\nabla P+\mu {\nabla }^2+f$$

The physics of the electroporation system of the device were defined using the AC/DC module to simulate the electric field distribution in the microfluidics electroporation model. The material of the droplet was defined as electroporation buffer with a conductivity of 1 × 10⁻⁴ S/m. The electrical conductivity of the electrode was defined by the composition of the Platinum-iridium wire as 9.43 × 10⁶ S/m. The measurement of FC40 supplemented with 2% FluoroSurfactant was performed in the voltmeter and the conductivity of oil phase was approximated as 1.044 S/m. In steady conditions, the flow of electric currents within a conducting fluid follows Ohm’s law:

$$J=\sigma E+J_e$$

Here, $J$ represents the total current density within the material while $\sigma$ represents the electrical conductivity measured in S/m. $E$ is the electrical field strength measured in V/m and $J_e$ represents the current density.

Moreover, to mathematically describe the electric field that acts upon the droplet within the microfluidic device, we used the induced potential difference $\mathrm{\Delta }{\mathrm{\Psi }}_i$ at a point $M$ of the droplet membrane at any time $t$:

$$\mathrm{\Delta }{\mathrm{\Psi }}_i\left(M,t\right)=fg\left(\lambda \right)Er\cos \left(\theta \right)\left(1-e^{\frac{-t}{{\tau }_c}}\right)$$

(76)

The results of this simulation were then used to adjust and optimize the device’s design for the intended application.

Bioactivity assays of electroporated sgRNA:Cas9 RNPs

For conventional cuvette electroporation, the Thermo Neon system was used. The designed 6 μM sgRNAs were assembled with 2 μM Cas9 nucleases at room temperature in Neon R buffer for 10 min and then treated with an electric field with 1500 V, 30 ms and 1 pulse with a 10 μL electroporation tip in Neon system. The resulting sgRNA:Cas9 RNPs were then transferred to a 1.5 mL Eppendorf tube and recovered for 10 min at room temperature. ~200 ng of purified Myo7a amplicons were then added to 0.5 μM sgRNA:Cas9 RNPs and incubated at 37°C for 45 min. 2 μg proteinase K (Thermo Fisher Scientific) was used to degrade Cas9 proteins for 15 min at 56°C and followed by 2 μg RNase A (Thermo Fisher Scientific) to completely deactivate sgRNA:Cas9 RNPs at 37°C. The final solutions were heated at 75°C for 10 min to end the enzymatic digestions. The final cleaved products were analyzed in 0.8 – 2% E-gel electrophoresis system (Thermo Fisher Scientific) and imaged with Typhoon imager with Cy2 channel.

Neon electro-transfection system

EVs in DPBS buffer were first transferred into Neon R buffer (Thermo Fisher Scientific) by using 30K cutoff ultrafiltration column to the final concentration of 10¹⁰ EVs/mL. Basically, EVs were added into the pre-washed 30 KDa cutoff column and centrifuge at 7000 rcf for 8 min and then the concentrated EV (~80 μL) were resuspended in 400 μL Neon R buffer to centrifuge again under the same condition. The resulting EVs in Neon R buffer (~60 μL) were placed on ice immediately for later use. To electro-transfect sgRNA:Cas9 RNP into EVs, sgRNAs and EGFP-Cas9 nuclease were first pre-mixed together in 45 μL Neon R buffer and self-assembled at room temperature for 10 min and then added into EVs solution to obtain 6 μM of EGFP-Cas9 and 9 μM of sgRNA in the ready-to-electro-transfection solution (~106 μL). To stabilize the membrane, 4.4 μL of 1250 mM trehalose in DPBS buffer without Ca²⁺ and Mg²⁺ was added to the ready-to-electro-transfection solution to have 50 mM trehalose in the final solution. The addition of trehalose increased the viscosity of the electro-transfection solution therefore, to maintain the viscosity balance in the Neon electroporation system, 120 μL of 1250 mM trehalose was added to 3mL Neon electrolytic buffer. 1500 V, 20 ms, 1 pulse was used to electro-transfect sgRNA:Cas9 RNP into HEI-OC1 derived EVs with the 100 μL Neon platform. The resulting RNP-EVs were gently transferred to a 1.5 mL protein low binding Eppendorf tube for membrane recovery at room temperature for 10 min. And then 500 μL DPBS buffer at room temperature was added to the EV solutions followed by further membrane recovery at 37°C for 20 min. The resulting RNP-EVs were stored in −20°C for the downstream purification and analysis.

High-throughput droplet-based μDES

Water-in-oil droplets were generated at the flow-focusing junction inside the μDES. We first used FITC-conjugated 100 nm polystyrene beads as EV-like nanoparticles to characterize the encapsulation rate and effectiveness. A microfluidic pressure flow controller (PreciGenome LLC) was used to generate the droplets with a diameter of around 1000 μm at 3.0 − 3.8 μL/min of the aqueous solution and 1.5 − 2 μL/min of the oil phase. The resulting droplets containing fluorescent polystyrene beads were collected in the 6-cm petri dish and visualized under the inverted microscope Cytation 5 (BioTek). For electroporation of EVs, as described previously, 10¹⁰/mL EVs were first transferred to Cytoporation^® media T (Biochrom Ltd.) and then mixed with 5 μM sgRNA:Cas9 RNP (sgRNA:Cas9 molar ratio=1.5:1). The resulting mixture was delivered into the device as the dispersed aqueous phase. The oil phase contained FC-40 mixed with 2 weight % 008-FluoroSurfactant (RAN Biotech). The droplet generation condition was described above. The electroporation on device was performed ranging from 10 – 60 V by using a direct current-based power supply (GW INSTEK) and the resulting emulsion was collected within an Eppendorf tube or microplate and then analyzed under the inverted microscope Cytation 5 (BioTek). The isolation of the aqueous phase containing RNP-EVs from the oil phase was performed under the centrifuge at 2000 − 3000 rcf for 5 − 10 min at room temperature. The aqueous phase was then collected by the pipettes and transferred to a new Eppendorf tube for downstream purification. The resulting RNP-EV was stored in −20°C for the downstream purification and analysis.

Chemical transfection methods for RNP-EV

The same amount of RNP mentioned above was first premixed and then coincubated with 2% saponin as reported before.(77) For loading of RNP to EVs with lipofectamine 3000, 5 μM RNP was mixed with 1 μL lipofectamine 3000 and incubated for 20 min at room temperature according to the manufacturer’s protocol for DNA plasmid transfection. The resulting RNP-EVs were then purified as described previously.

Preparation of RNP-LNP

The same amount of RNP mentioned above was first premixed and then added to 1/10 volume of premixed SM102-LNP (Cayman Chemical) according to the manufacturer’s instruction for the mRNA loading. For loading of RNP to EVs with lipofectamine 3000, 5 μM RNP was mixed with 1 μL lipofectamine 3000 and incubated for 20 min at room temperature according to the manufacturer’s protocol for DNA plasmid transfection. The resulting RNP-EVs were then purified as described below.

Purification of delivery vehicles RNP-EVs and RNP-LNPs

The RNP-EVs containing His tagged EGFP were purified by applying ~100 μL Ni Sepharose high performance beads (GE Healthcare) to remove excessive free sgRNA:His tagged EGFP Cas9 RNP complexes in the solution. An equal volume of beads was incubated with RNP-EVs at 4°C for 0.5 – 1 hours on the rocker until reaching binding equilibrium with sgRNA: His tagged EGFP-Cas9. After Ni sepharose beads captured His tag-RNPs in solution, the beads were washed with 10× volume of cold DPBS buffer and then pre-equilibrated in DPBS buffer for 10 mins to collect EVs. The purified RNP-EVs were subsequently concentrated to a 2-fold increase using a 30 KDa ultrafiltration column and stored in −80°C for further analysis. In addition, to purify the RNP-EVs assembled by using surfactant 0.2% saponin, the resulting RNP-EV was first washed with 10× volume of DPBS without Ca²⁺ and Mg²⁺ by using 10 KDa ultrafiltration column before the incubation with Ni Sepharose beads. The purification of RNP-LNPs follows the same protocols.

Characterization of RNP-EVs

Nanoparticle tracking analysis

The size and particle number of purified EVs and RNP-EVs were analyzed by nanoparticle tracking analysis (NTA) using the ZetaView^® (PARTICLE METRIX, Germany) supplied with a multilaser system. Briefly, to measure the size and number of EVs in the scatter mode, the EV solutions was diluted with DPBS to reach the measuring concentration of 4 – 8 × 10⁷ /mL via a blue laser (488 nm). Subsequently, 1 – 2 mL diluted EV solutions were then injected into the measuring cell and each measurement of every sample was repeated 2 – 3 times. Data analysis was carried out in ZetaView software with the following software settings for capture and analysis: Sensitivity= 80, Shutter= 100, Min brightness= 20, Max brightness= 1000. To quantify the EVs that were encapsulated with EGFP or sgRNA:EGFP-Cas9, fluorescent NTA analysis was performed by increasing sensitivity to 90 and switching to a 505 nm filter while keeping the other parameters unchanged.

Zeta potential

The zeta potential of the purified RNP-EVs was measured via electrophoretic light scattering (Litesizer 500, Anton Paar). Briefly, 25 μL of the final EV solution was diluted 1:20 with 10% DPBS buffer and injected into a disposable folding capillary cuvette, and each zeta potential measurement was conducted five times. The final measurement was conducted with 2min for pre-equilibrium, under room temperature by selecting PBS as the referenced conductivity.

Loading efficiency of RNP complexes

We used 10 mg/mL EGFP-Cas9-NLS protein from Genscript to detect EGFP fluorescence, enabling the generation of a standard calibration curve. This, in turn, allowed for the accurate quantification of Cas9 proteins in the formed RNP complex loaded into EVs. Each measurement was performed in triplicate to ensure rigor and reproducibility. The standard curve of fluorescent intensity of EGFP-Cas9 in DPBS buffer was measured by Cytation 5 with a serial dilution of EGFP-Cas9 in a 96-well microplate. 35 μL of the final EV solution was diluted in 1:1 with DPBS buffer and then added to the microplate. The fluorescence intensity of the resulting solution was measured under the same condition as the standard curve to accurately quantify Cas9 protein. To exclude contamination from EV-associated proteins, we leveraged EGFP-tagged Cas9 proteins to measure fluorescence from purified RNP EVs. The fluorescence background from the native EVs without loading of EGFP-Cas9 was also measured in the same particle number for signal subtraction and elimination of noise from EV protein readouts (readout values below 200). To further quantify the loading efficiency of sgRNA:EGFP-Cas9 in EVs, 25 μL of the final EV solutions diluted 1:40 with PBS buffer was then injected into ZetaView NTA (ParticleMetrix, Germany) for the quantification of EGFP⁺ EV. Basically, the diluted EVs were measured in both scattering mode and fluorescence mode by using particle number/sensitivity measurement in fluorescence NTA with laser 488nm. By using reference polystyrene beads labeled with fluorescein (Applied Microspheres, Netherlands), the sensitivity scale was set at 96–98 with 100% fluorescent labeled beads. The final percentage of EGFP⁺ EVs was normalized against that from the reference beads under the same sensitivity in the fluorescence mode.

Automated microWestern blotting of RNP-EV

Total protein extracts from natural EVs and purified RNP-EVs were prepared in one volume of lysis solution, RIPA buffer (Thermo Fisher Scientific), followed by 5 min sonication, and vortex for 30 s. Protein concentration was then determined by the Micro BCA Protein Assay (Thermo Fisher Scientific). To quantify the total proteins from RNP-EVs assembled by surfactant saponin, the resulting RNP-EVs were first washed as described previously to remove saponin as much as possible. However, it is important to note that the saponin which contributes dark purple as blank greatly interferes with the sensitivity of the Micro BCA assay. In contrast, the Bradford assay showed reduced background signals after the multiple washing steps and the quantification of total protein of RNP-EVs from saponin used EV only as the reference group for a better accuracy. Protein extracts (~150 ng per lane) were added and separated in the cartridge compatible with Wes Instrument (Bio-Techne). Simple Western was performed and imaged according to the manufacture’s procedure. anti-CD81 (D5O2Q, Cell Signaling Technologies), anti-Alix (NBP1–49701, Novus Biologicals) and anti-TSG101 (T5826, Sigma-Aldrich) antibodies were used for the MicroWestern assays.

Transmission electron microscopy

10 μL of the EVs suspension with 10⁹ particles/mL was vortexed for 30s and then placed on parafilm. The glow discharged dark side of a 400-mesh TEM grid (Electron Microscopy Sciences) was incubated with the EV droplet on the parafilm for 15 min at room temperature. The TEM grid was then air dried at room temperature. 2% Uranyl acid was used for the negative staining of EVs on the TEM grip for 5 min at room temperature. The TEM grid was then washed with deionized water for 10 s and air dried for TEM imaging on a Tecnai G2 Sprit TWIN 120kV with UltraScan 1000 (2k × 2k) CCD camera.

Immunogold-electron microscopy

The TEM grid was pretreated with 1% poly-L-Lysine for 15 min and then air dried. 10 μL of the EVs suspension with 10⁹ particles/mL was incubated with primary antibodies (anti-Cas9 (A01935, Genscript) or anti-CD81 (D5O2Q, Cell Signaling Technologies)) diluted 1:25 in the PBST buffer containing DPBS, 0.1% BSA and 0.01% Tween 20 for 1 hour at 4°C. The resulting solution was then added to the TEM grid, dried at room temperature followed by washing in DPBS three times and 0.1% BSA. The TEM grid was then incubated in the prepared 1% paraformaldehyde for 5 min, followed by washing in PBST buffer twice for 5 min each. The grid was then incubated with secondary antibody (goat anti-mouse and goat anti-rabbit conjugated to 6 nm AuNP) diluted 1:10 in PBT buffer for 1 hour at room temperature, then washed by passing over three droplets of deionized water. The grid was post-stained with 2% uranyl acetate and imaged on a Tecnai G2 Sprit TWIN 120 kV with UltraScan 1000 (2k x 2k) CCD camera.

RT-qPCR quantification of sgRNA

About 10⁸ EVs or RNP-EVs were first treated with 2U Proteinase K (Thermo Fisher Scientific) for 15min at 37°C and then proteinase K was inhibited by adding 1X Halt protease inhibitor cocktail. 2U RNase A (Thermo) was then added to RNP-EVs, native EVs, and μDES-EVs, and incubated at 37°C for 15min. The total RNA content from the treated EVs was then isolated by following the manufacturer’s protocol of the miRNeasy purification kit (QIAgen). The total EV RNA was then quantified by a Ribogreen quantification kit (Thermo Fisher Scientific). We used synthetic sgRNA-1 as the template to obtain standard curve for the quantification of sgRNA molecules. The sgRNA-1 The reverse transcription of sgRNA was conducted at 54°C for 30 min with AMV Reverse Transcriptase by following the manufacturer’s instructions. The resulting cDNA of sgRNAs were then directly used for the qPCR quantification with PowerTrack^™ SYBR Green Master Mix for QuantStudio (TM)7 Flex System. The final cDNA was aliquoted to four individual qPCR reactions by using the primers in Table S1.

Stability investigation of RNP-EV

After electroporation through μDES and purification as described above, 15 μL RNP MSC-EV was aliquoted into Eppendorf tubes with particle concentration at 10¹⁰ /mL supplemented with 25 mM Trehalose in DPBS without Ca²⁺ or Mg²⁺. The resulting RNP EV was then stored in −80°C for the stability test over time, 0 Day, 7 Day, 14 Day, 28 Day and 70 Day. Generally, after defrosting the RNP EV, the particle number was first measured using ZetaView software for three different replicates with the same condition as described elsewhere. The total RNA from the resulting RNP EV solutions were then isolated by following the manufacturer’s protocol of the miRNeasy purification kit (QIAgen). The amount of total RNA was measured using Quant-iT^™ RNA Assay Kits (Thermo Fisher Scientific) and qPCR analysis and quantification was performed as described elsewhere.

Biocompatibility investigation of LNP and EV

About 10⁶ EVs and LNPs per HEI-OC1 cell in a 96-well plate were used for the biosafety test. EVs were treated with the μDES strategy described before but without the encapsulation of RNP. LNPs were also prepared according to the manufacturer’s protocol. The resulting EVs and LNPs were incubated with HEI-OC1 cells in 96-well plate for 72 hours. The PrestoBlue reagent (Thermo Fisher Scientific) was then added to each well to quantify the viability of HEI-OC1, a hair-like cell line, based on the manufacturer’s instructions. The fluorescence intensity was subsequently measured at 560/590 nm using a Cytation 5 microplate reader (BioTek).

Indel analyses

T7E1 assay

200 ng purified amplicons were denatured and re-annealed by following the manual of the GeneArt Cleavage detection kit. The re-annealed amplicons were then incubated with T7 endonuclease I for 30 min at 37°C. The final solution was then immediately added to the well of 0.8 – 2% SYBR safe prestained E-gel and ran for 26 min in the E-gel running system (Thermo Fisher Scientific). The gel was then imaged with a Typhoon imager using the Cy2 channel. The band intensity was quantified with ImageQuant and Indel % was obtained by using equation: (1- (1- fraction cleaved)^1/2).

Sanger sequencing and ICE analysis

Purified amplicons with more than 200nt flanking sequences were quantified with Cytation 5 and then diluted to the final concentration of 5 ng/μL. The resulting amplicons together with the reverse primer (5’-GCGTAGGAGTTGGACTTGATAG −3’) were then submitted to Sanger sequencing provided by Genewiz. The sequencing data was opened with UGENE software for imaging and visualization of amino acid sequences. The DNA sequencing data were then uploaded to ICE, the online quantitative tool of Indel analysis of sequencing traces of sgRNA:Cas9 RNP-mediated editing. The indels was calculated by comparing the frequency of sequence alterations in sgRNA:Cas9 RNP-treated samples with that in untreated controls. The same sgRNAs were entered into ICE for both full length sgRNAs and the corresponding truncated counterparts due to the same cutting site.

HTS and CRISPResso2 bioinformatic analysis

The sequences of amplicons listed in Table S1 were obtained by NGS. The Indel analysis and mutation types were identified with the online genome editing bioinformatic tool, CRISPResso2 by following the steps published before.(8) For the analysis of coding sequences in CRISPResso2, the exon sequences were: 5’-CCTTGGGGAACTTGCTCTCCTCATCGATGAGGGAGATGACGTTCATAGGCGGGTT GGCAATCATGTCCAGTGCTTCCTGGTTGTCAGTGAACTCAATGTGCAGCCAGTCGATGCTCTCCAGGTCGTACTCCC-3’. The sequences of sgRNAs and the associated truncated counterparts were used in the CRISPResso2 analysis.

For allele-specific indel analysis, the wild-type Myo7a (Myo7a^WT) and Myo7a^Sh1 sequences from CRISPResso2 were further analyzed by using the LEN and SUBSTITUTE Function with 5’-CCG-3’ for WT reads associated with EXCEL. The allele-specific indel % was calculated by using the equation: <edited Myo7a^WT sequences/(edited Myo7a^WT sequences and edited Myo7a^Sh1 sequences)>.

For mRNA analysis, CRISPResso2 analysis was performed similarly to the indel analysis mentioned above. To quantify intact, non-edited reads, we used the following sequences: 5’-AGGCCGGAA-3’ and 5’-TTCCGGCCT-3’ for WT reads, and 5’-AGGCGGGAA-3’ and 5’-TTCCCGCCT-3’ for Sh1 reads.

In silico off target analysis and CIRCLE-seq

The off-target prediction by using the off-target algorithm from Integrated DNA Technology was performed by entering the Shaker-1 DNA sequence. The potential off targets based on the sgRNA-1 sequence were reported based on the ranking scores (table S2). The off targets with lower scores have a higher chance to be edited. Therefore, the top 9 off targets were amplified by using the primers listed in Table S3. CRISPResso2 was used to quantify the indel percentage based on the corresponding sgRNAs.

For the genome-wide off target analysis, CIRCLE-seq was performed by following the protocol published previously(54) using the DNA oligo probes and primers listed in table S4, with some modifications as shown below on the published CIRCLE-seq method due to the discontinuity in some reagents and lack of thermal cycler to perform the exact annealing steps required by the CIRLCE-seq protocol. First, 60 μg genomic DNA was purified from primary fibroblasts and then enzymatically sheared for 10 min by following the manual instructions (NEB Ultra II FS kit, E7805)(78). We performed an additional 10 PCR cycles to amplify the sequences of interest and qPCR to ensure the existence of the sequence of interest after intramolecular circulation reaction. Circularized gDNA was then incubated with sgRNA1-Cas9. Then the sequencing libraries were prepared and sequenced on the NovaSeq X platform. Off target sites were identified using previously published CIRCLE-seq pipelines, with a minor modification setting.

Flow cytometry of cellular uptake

40,000 HEI-OC1 cells were seeded to a 48-well plate overnight for cell attachment. Cells were then incubated with 1 μM of EGFP-RNP encapsulated with MSC-EVs, HEI-OC1 EVs and LNPs with particle concentration 5 ×10⁹ - 10¹⁰/mL while the same particle concentration (5 × 10⁹ /mL) of MSC-EVs and HEI-OC1 EVs. The treated cells were then put on the orbital shaker at 100 rpm for 15 min at 33°C. The treated cells were then put into the incubator for another 45 min at 33°C, 10% CO₂. The treated cells were washed with 500 μL cold DPBS twice and then dissociated from the plate by incubating 200 μL of non-enzymatic dissociation buffer (ATCC) for 5 min at 37°C. Then equal volume of DPBS supplemented with 0.05% BSA was used to resuspend the cells and then washed robustly with DPBS twice before flow cytometry. Cellular uptake of different particles in HEI-OC1 cells were then measured and analyzed by using the Accuri C6 flow cytometer. The results were further analyzed and imaged using FlowJo V10.9.

Confocal imaging of cellular uptake

3,000 – 5,000 HEI-OC1 cells were seeded on a cover glass slip which was pre-treated with 1% poly-L-Lysine and cultured for 48 hours at 33°C, 10% CO₂. 1 mL culture medium was added to each well and then the cells were incubated to allow the cells to attach overnight. 100 μL of 3 – 7 × 10¹⁰ /mL RNP*^EGFP-EVs, RNP*^EGFP-LNPs with 1 μM sgRNA1:EGFP-Cas9 RNP were then added to incubate with cells for one hour at 33°C, 10% CO₂ followed by the removal of the culture medium. The treated cells were then washed twice with DPBS without Ca²⁺ and Mg²⁺ (Thermo Fisher Scientific). The cells were fixed with 4% formaldehyde in DPBS for 15 min and permeabilized with 0.5% Triton X-100 in DPBS for 5 min followed with DPBS washing twice. The cells were then incubated with 2% BSA and 22.52 mg/mL glycine in DPBS solution for 1 hour to reduce the non-specific binding of primary antibodies. The cells were then washed with DPBS twice and incubated with 1:500 anti-EEA1 antibody (ab2900, Abcam), anti-Cas9 antibody (A01935, Genscript) in 1% BSA mixture overnight at 4°C. The secondary antibodies, Alexa 555 anti-mouse antibody (A-21127, Thermo Fisher Scientific) and Alexa 647 anti-Rabbit antibody (A-21245, Thermo Fisher Scientific) were then incubated with the cells for 1 hour at 4°C after being washed twice. The cells were stained with nucleus staining dye, DAPI with the final concentration of 0.2 μg/mL for 5 min. After final DPBS washing steps, ProLong^™ gold anti-fade mountant was added to the center of the microscope slides and then the side of the cover slip with treated cells were attached onto the slide. Clear nail polish was added onto the four corners of the glass cover slip to stabilize the glass. The glass slide was then ready for confocal imaging with the Zeiss Confocal LSM800 Microscope. The different samples were taken with the exact same imaging settings including gain power, image resolution et al.

The confocal microscope imaging of cellular uptake of mCherry mRNA delivered by MSC-EVs and LNPs was conducted with the same confocal imaging conditions. However, the incubation was conducted with 20 μg mCherry mRNA encapsulated with (6 × 10¹² /mL) μDES MSC-EV and (~10¹² /mL) LNP with 10,000 HEI-OC1 cells.

Gene editing by RNP LNPs and RNP MSC-EVs in the fibroblasts

About 5,000 primary fibroblast cells isolated from ear punches from Myo7a^WT/Sh1 mice were seeded into a 48-well plate overnight under culturing conditions described previously. The old cell culture media was removed from the well before the addition of RNP LNPs and RNP MSC-EVs. 50 μL of ~1 μM RNP in RNP LNPs and μDES-produced RNP MSC-EVs, and Cuvette RNP MSC-EVs were added into 100 μL of regular fibroblasts culture media supplemented with 1% penicillin-streptomycin and 10 μM chloroquine and then incubated with fibroblasts for 30 min at 37°C in the orbital shaker at 100 rpm. Incubation was then undertaken in the regular cell incubator at 37°C, 5% CO₂. The fibroblasts were dosed twice in total with the same dosing concentration. After 96 hours of incubation, gDNA was isolated from the fibroblasts using the GeneArt^™ Genomic Cleavage Detection Kit (Thermo Fisher Scientific) and following the manufacturer’s protocol. The amplicons were prepared using Q5 high fidelity DNA polymerase (New England Biolabs) for Illumina Miseq 2 × 250 bp sequencing.

In vivo delivery of RNP-EVs to the inner ear

For the surgical procedure, P4 and P30 Myo7a^WT/Sh1 mice were anesthetized with an intraperitoneally administered mixture of ketamine (100 mg/kg), xylazine (5 mg/kg) and acepromazine (2 mg/kg). A dorsal postauricular incision was made and the bulla exposed. For EV delivery, the posterior semicircular canal was exposed as previously described(79). EV or LNP injections consisted of sgRNA-1:Cas9 ribonucleoproteins in 1 μL of volume with a particle concentration of 10¹² /mL (delivered using a Hamilton micro syringe with 0.1 μL graduations). All treatments were carried out in the left ear with the right ear serving as untreated control. P4 Myo7a^WT/Sh1 mice were anesthetized with cold as described in reference (80). The posterior semicircular canal was exposed in the postauricular space, and a sharp pick was used to open the canal. EV at a dose of 1 μL was injected using a microsyringe as described above. Genotyping was carried out at P21 as described. Histological evaluation was carried out at 3 and 6 months of age and hearing outcomes measured at 6 months of age. For safety evaluation, EVs were delivered to the posterior semicircular canal of Myo7a^WT/WT mice. Postoperative acepromazine (2 mg/kg) was administered by the veterinary staff.

ABR measurement

ABR thresholds were recorded using the Intelligent Hearing Systems Smart EP program (IHS). Animals were anesthetized as described above and kept warm on a heating pad (37°C). Needle electrodes were placed on the vertex (+), behind the left ear (−) and behind the opposite ear (ground). Tone bursts were presented at 4, 8, 16 and 32 kHz, with a duration of 500 μs using a high frequency transducer. Recording was carried out using a total gain equal to 100K and using 100 Hz and 15 kHz settings for the high and low-pass filters. A minimum of 128 sweeps were presented at 90 dB SPL. The SPL decreased in 10 dB steps. Near the threshold level, 5 dB SPL steps using up to 1024 presentations were carried out at each frequency. The threshold was defined as the SPL at which at least one of the waves could be identified in 2 or more repetitions of the recording.

Immunohistochemistry and histology

Mice were anesthetized with phenobarbital (585 mg/kg) and phenytoin sodium (75 mg/kg) (Beuthanasia-D Special, Schering-Plough Animal Health Corp.) via intraperitoneal injection (i.p.) and sacrificed via intracardiac perfusion with 4% paraformaldehyde in DPBS. The temporal bones were removed, later the stapes extracted, and the round window was opened. The temporal bones were then postfixed overnight in 4% paraformaldehyde in DPBS at 4°C. After rinsing in DPBS three times for 30 min, the temporal bones were decalcified in 10% EDTA (ethylene diamine tetraacetic acid) for 48 hours. The temporal bones were rinsed in DPBS and embedded in paraffin. Seven μm sections were cut in parallel to the modiolus, mounted on Fisherbrand Superfrost/Plus Microscope Slides (Fisher Scientific) followed by an overnight drying process. Samples were deparaffinized and rehydrated in DPBS two times for 5 min, then washed three times in 0.2% Triton X-100 in DPBS for 5 min and finally in blocking solution 0.2% Triton X-100 in DPBS with 10% fetal bovine serum for 30 min at room temperature. After blocking, specimens were stained with anti-4 hydroxynonenal (R&D Systems MSB 3249) or anti 3-nitrotyrosine (MA1–2770) or anti-Myosin VIIa (Abcam, ab150386) diluted 1:100 in blocking solution. The tissue was incubated for 48 hours at 4°C in a humid chamber. After three rinses in 0.2% Triton X-100 in DPBS, immunofluorescent detection was carried out with anti-rabbit IgG (1:50; Alexa Fluor 488 nm; Invitrogen Inc.). The secondary antibody was incubated for 6 hours at room temperature in a humid chamber. The slides were rinsed in 0.2% Triton X-100 in DPBS three times for 5 min and finally coverslipped with ProLong Gold antifade reagent (Invitrogen Molecular Probes). Mean fluorescence intensity for the evaluation of gene expression in inner hair cells, outer hair cells, and ganglion cells were analyzed by immunofluorescence using ImageJ software. Regions of interest (ROI) corresponding to the cells were selected manually within ImageJ. The intensity of fluorescence within the same ROI in multiple images was measured and background fluorescence was subtracted to obtain accurate signal values. The intensity value was measured three times and averaged, and fluorescence intensity was expressed as arbitrary units/ROI.

Whole mount preparation

Cargo-loaded EVs were delivered and animals anesthetized and sacrificed as described earlier. The temporal bones were removed, the stapes extracted, and the round window was opened. The temporal bones were postfixed overnight in 4% paraformaldehyde in DPBS at 4°C. After rinsing in DPBS three times for 30 min, the temporal bones were decalcified in 10% EDTA (ethylene diamine tetraacetic acid) for 48 hours. The temporal bones were rinsed in DPBS. The cochlea was removed from the temporal bone and decalcified bone removed with fine forceps. The organ of Corti was then carefully stripped away from the modiolus. The organ of Corti whole mounts were placed in ProLong Gold antifade reagent (Invitrogen Molecular Probes) and cover slipped for the subsequent confocal imaging.

Inner ear tissue dissection for NGS and qPCR

Both sides of the organ of Corti were collected 6 months after the PSCC administration of RNP-EVs on the left side, whereas the right side remained untreated. DNA and RNA samples were collected by following the manufacturer’s instruction of the QIAgen MagAttract HMW DNA Kit and miRNAeasy Mini Kit with DNAse treatment respectively. The resulting nucleic acid samples were then quantified with Agilent Genomic DNA Tape assay and RNA Screen Tape assay respectively. cDNA library of mRNA for NGS and qPCR were prepared by using the amplicons for mRNA of Myo7a shown in table S1.

qPCR of genomic DNA was performed following the protocol published before.(81) Briefly, 2.5 ng of genomic DNA was used for each well in 382-well Armadillo PCR plate (Thermo Fisher Scientific) and 250 nM qPCR probes shown in table S1. For qPCR of cDNA of mRNA from the organ of corti, cDNA of total RNA was first synthesized by M-MLV reverse transcriptase (New England Biolab) following the manufacturer’s instruction by using a random primer mixed with 2 pmol of mRNA_Myo7a_RP. The following qPCR reaction was using the probes 18S rRNA_Mouse_FP and 18s rRNA_Mouse_RP for 18S rRNA, cMYO7A-Sh1_qPCR_FP and cMYO7A_qPCR_RP for Myo7a^Sh1 while cMYO7A-WT_qPCR_FP and cMYO7A_qPCR_RP were used for Myo7a^WT. The detailed sequences are listed in table S1. The fold change analysis was performed as described previously(82).

In silico analysis of principal indel variants

We first sorted and grouped the new alleles into the 5 principal mutation types. The Cas9-mutagenesis-derived mRNA sequences that could be transcribed from the indel alleles were entered into Expasy translation tool (https://web.expasy.org/translate/) and the potential polypeptide sequences were collected. The domain selection was based on the features of the unconventional myosin-VIIa (P97479) UniProt database (https://www.uniprot.org/uniprotkb/P97479/entry). Based on the loci where the mutation identified in the annotated domain region, myosin motor domain, we selected residue 401–561 for further characterization and analysis. The name entry was based on their deletion. The in-frame mutants were characterized with Alphafold2 to predict the structures and obtain PDB files (https://colab.research.google.com/github/sokrypton/ColabFold/blob/main/AlphaFold2.ipynb, accessed on X, October 2023) with the following parameters: num_relax=1; template_mode=none; rank_num=3 while the other as default. The resulting PDB files were used for the binary structure alignment and color by predicted pLDDT value by using command line <spectrum b, yellow_orange_cyan_deepblue, minimum 57, maximum 98> in Pymol software.

Statistical Analysis

Statistical analyses were performed using GraphPad Prism 9.0.0. Data are presented as biological replicates and shown as means±SEM or range as indicated in the corresponding figure legends. The sample sizes, statistical tests used, and P values, when significant, are also described in detail in the corresponding figures and figure legends. Statistical analyses were performed using unpaired two-tailed Student’s t-tests for comparisons between two groups. For experiments involving more than two groups or multiple variables, one-way or two-way analysis of variance (ANOVA) tests was used as appropriate. Dunnett’s post hoc test was applied following ANOVA when comparing multiple treatment groups to a single control while accounting for multiple comparisons. For all analyses, P < 0.05 was assigned as statistically significant.

Supplementary Material

Supplementary Materials and Methods

Figs S1 to S19

Tables S1 to S4

Data File S1

MDAR Reproducibility Checklist

References (93-97)

Data and materials availability:

All data are present in the paper or the Supplementary Materials. Source data for the figures in this study are available from the corresponding author upon reasonable request. NGS sequencing data are available at NCBI BioProject ID: PRJNA1041447