Engineering ARMMs for improved intracellular delivery of CRISPR-Cas9

Abstract

CRISPR-Cas9-based gene editing holds enormous promise for therapeutic applications, but its effectiveness is often limited by inefficient delivery methods. This study explores the potential of arrestin domain-containing protein 1 (ARRDC1)-mediated microvesicles (ARMMs)—a type of extracellular vesicles formed at the plasma membrane—as a novel platform for packaging and delivering CRISPR-Cas9 complexes. We achieved efficient Cas9 packaging into ARMMs by directly fusing Cas9 with ARRDC1. Two different ARRDC1-Cas9 fusion constructs were designed, and both demonstrated gene-editing efficiency comparable to unmodified Cas9. The fusion with a shorter version of ARRDC1 (sARRDC1), which includes only the minimal motifs required for vesicle budding, proved particularly effective in enhancing Cas9 packaging. Additionally, the incorporation of vesicular stomatitis virus glycoprotein (VSV-G) further improved ARMMs budding and Cas9 encapsulation. We tested gene editing in U2OS cells with an exogenous GFP gene and in human neuronal cells targeting the endogenous amyloid precursor protein (APP) gene, which is associated with the Alzheimer’s disease. The combination of ARMMs and VSV-G resulted in high editing efficiency, with ARMMs targeting the APP gene in neuronal cells significantly reducing pathogenic amyloid peptides. These results highlight ARMMs as a versatile and effective platform for CRISPR-Cas9 delivery, with strong potential for therapeutic applications in neurodegenerative and other genetic diseases.

1. Introduction

CRISPR-Cas9-based gene editing is revolutionizing genetic medicine by enabling precise modifications to treat a wide range of human diseases.^1-3 This technology facilitates the correction of genetic mutations, insertion of therapeutic genes, and deletion of harmful sequences in patient cells. For instance, in diseases like sickle cell anemia and beta-thalassemia, CRISPR-Cas9 has been used to reactivate fetal hemoglobin production, offering a potential cure by compensating for defective adult hemoglobin.^4,5 In cancer therapy, CRISPR-edited T cells have been engineered to enhance their ability to recognize and destroy cancer cells, as shown in clinical trials targeting resistant B-cell malignancies.^6,7 Furthermore, recent studies underscore its growing potential for treating neurodegenerative diseases.^8,9 Despite the tremendous promise of CRISPR-Cas9—owing to its precision, efficiency, and versatility in correcting genetic defects—several major challenges remain. Key obstacles include effective delivery to target cells, minimizing off-target effects, and avoiding immune responses. Addressing these issues is crucial for realizing the full therapeutic potential of CRISPR-based gene editing.

Several delivery methods have been developed for CRISPR-Cas9, each with distinct advantages and limitations. Viral vectors, such as adeno-associated viruses (AAVs), are widely used for their ability to target specific tissues. However, their cargo capacity is limited, which restricts the size and number of genetic elements they can deliver.^10-12 Lipid nanoparticles offer the advantage of encapsulating larger genetic payloads and allowing for repeated dosing, but certain formulations may be toxic to cells or provoke inflammatory responses.^13,14 Electro-poration, which uses electrical pulses to introduce CRISPR components into cells, is highly effective for ex vivo applications, but it can also cause significant cell damage.^15,16 Ribonucleoprotein (RNP) complexes, which deliver the Cas9 protein and guide RNA directly, help reduce off-target effects and minimize immune responses, making them well-suited for ex vivo gene editing. However, their use in vivo remains challenging due to issues with delivery efficiency and stability.¹⁷

Extracellular vesicles (EVs), including exosomes and microvesicles, have emerged as a promising platform for gene delivery due to their natural origin, biocompatibility, and ability to cross biological barriers. For example, exosome-mediated delivery of CRISPR-based gene editors has demonstrated potential in animal models of neurodegenerative diseases.^18,19 A novel subset of EVs, ARRDC1-mediated microvesicles (ARMMs), are generated by direct budding from the cellular plasma membrane.²⁰ These vesicles have been shown to serve as versatile delivery vehicles for various macromolecules, including mRNA, proteins, and CRISPR-Cas9 complexes.²¹ ARMMs can package and deliver the CRISPR-Cas9 system to recipient cells through direct fusion of the Cas9 protein with the WW domains of ITCH proteins, employing the interaction between ARRDC1 and WW-domain-containing proteins.^20,21 However, the efficiency of this two-construct system is somewhat limited by the strength of the WW-domain interaction and the need to balance the functions of both Cas9 and ARRDC1. These factors likely contribute to suboptimal efficiency in both packaging and gene editing.

In this study, we present a novel ARMMs-based delivery system for CRISPR-Cas9, utilizing the direct fusion of ARRDC1 with Cas9 proteins. We identified a truncated version of ARRDC1 that is more efficient at recruiting Cas9 into ARMMs. The packaging capability of ARMMs for the Cas9/sgRNA complex was further optimized through VSV-G pseudotyping. Gene editing efficiency was assessed by targeting the exogenous GFP gene in engineered U2OS cells (which contain a single copy of the GFP gene) and the endogenous amyloid precursor protein (APP) gene in human neurons. Our findings establish a highly efficient strategy for the intracellular delivery of CRISPR-Cas9, offering improved gene editing capabilities.

2. Materials and methods

2.1. Cell culture

HEK293T cells (ATCC, catalog # CRL-3216) and U2OS-gfp cells (Millipore Sigma, catalog # CLL1136) were cultured in DMEM (high glucose, Gibco) supplemented with 10 % fetal bovine serum (FBS, Gibco) and 100 μg/mL PenStrep (Gibco). ReNcell^® CX human neural progenitor cell line (ReNcells) was purchased from Millipore Sigma (catalog # SCC007) and maintained in laminin-coated plates using ReNcells NSC maintenance media (Millipore Sigma, catalog# SCC005) supplied with epidermal growth factor (EGF) and fibroblast growth factor (FGF) according to supplier’s instructions. Human induced pluripotent stem cells derived glutamatergic neurons (iPSC-GlutaNeurons) were obtained from Fujifilm Cellular Dynamics (catalog #R1061) and seeded in 96-well plate precoated with poly-l-ornithine and Matrigel at a density of 8 × 10^4 cells per well according to the manufacturer’s instruction.

2.2. Plasmids

Single guide RNAs (sgRNAs) targeting genes eGFP (5′-GGCGAGGGCGATGCCACCTA-3′) or human APP (5′-ATCCATTCAT-CATGGTGTGG-3′) were cloned into PX330 vector (Addgene, catalog # 42230) at BbsI site according to previous literature.^22,23 Full length (1299 bp) DNA fragment of ARRDC1 (sequence details in Supplementary Text S1) was amplified by PCR from pEGFP-N1.ARRDC1 construct (Addgene, catalog # 38320) and cloned into PX330 constructs after the sgRNA cloning at AgeI site. Alternatively, short version of ARRDC1 (sequence details in Supplementary Text S2) were synthesized and inserted at the AgeI site. Sequences of all the cloned constructs were verified by Sanger sequencing using proper primers. Plasmid pMD2.G were purchased from Addgene (catalog # 12259). All the plasmids with high purity were prepared using Endotoxin-free plasmid maxi kit (Qiagen, catalog # 12362).

2.3. Transfections

Plasmid constructs were transfected into HEK293T cells and ReNcells using TurboFect transfection reagent and the Neon Electroporation system, respectively. In detail, HEK293T cells were seeded in 10-cm dishes at a density of 2 × 10^6 cells/plate and cultured overnight. Subsequently, 3 μg of DNA was transfected into the cells using 9 μL of TurboFect transfection reagent, following the manufacturer’s instructions. For VSV-G co-transfection, different ratios of VSV-G plasmid were co-transfected with Cas9 constructs, respectively, with the total amount of plasmid remaining the same. For ReNcells, electroporation was performed in Neon Buffer R with 1 μg of construct using the parameters 1300 V, 10 ms, and 3 pulses. The medium was replaced with fresh corresponding medium after 24 h. Seventy-two hours post-transfection, genomic DNA or whole cell lysates were collected for downstream assays.

2.4. ARMMs production

To produce ARMMs, modified Cas9 constructs with either full length or short ARRDC1 were transfected in HEK293T cells respectively using TurboFect transfection reagent. Briefly, HEK293T cells were seeded in 15-cm plates at a density of 7.5 × 10^6 cells/plate and cultured overnight. Then 15 μg of DNA was transfected into the cells using 45 μL of TurboFect transfection reagent, following the manufacturer’s instructions. Twenty-four hours later, the medium was replaced with 35 mL of fresh DMEM supplemented with 10 % of exosome-depleted FBS (Thermo Fisher Scientific, catalog # A2720803) and cells were maintained for additional 48 h.

2.5. ARMMs collection and characterization

Seventy-two hours post-transfection, the cell culture supernatant was collected and subjected to purification following the Minimal Information for Studies of Extracellular Vesicles (MISEV) guidelines.²⁴ In brief, the conditioned medium was processed by two consecutive rounds of centrifugation at 500g and 2000 g for 10 min each. The medium was then passed through a 200 nm filter (Acrodisc) and subjected to ultracentrifugation using the SW32Ti rotor in an XE90 centrifuge (Beckman) at 174,900 g for 2 h. The supernatant was aspirated, and the pellets enriched with ARMMs were resuspended in ice-cold PBS. The purity and yield of ARMMs were measured by Nanoparticle Tracking Analysis (NTA) using the NanoSight NS300 instrument (Malvern).

2.6. ARMMs incubation with different cell lines

U2OS-gfp cells and ReNcells were seeded in 48-well plates with a density at 2 × 10^4 cells/well and cultured overnight. Cell culture medium was replaced with fresh medium with ARMMs at different concentrations. Human iPSC-GlutaNeurons were seeded in 96-well plate at a density of 8 × 10^4 cells for 72 h before replacing the medium with different concentrations of ARMMs. Cells were incubated with ARMMs for 48 h followed by downstream assays.

2.7. Cytotoxicity assay and flow cytometry

Cytotoxicity was determined by measuring the release of LDH in the cell culture medium after the transfection using the CytoTox 96^® Non-Radioactive Cytotoxicity Assay kit (Promega, catalog #G1780) according to the manufacture’s instruction. The GFP signal of U2OS-gfp cells were determined by flow cytometry, where the cells were trypsined from the plate and washed with PBS. The GFP signal was determined by CytoFlex (Beckman), and the data were analyzed using FlowJo^™ (BD Biosciences, version 10.4).

2.8. Immunoblots analysis and antibodies

Cells and vesicle fractions were lysed in M-PER mammalian protein extraction reagent (Thermo Fisher Scientific) supplemented with a protease inhibitor cocktail (Roche). Cell lysates or vesicles resuspended in lithium dodecyl sulfate sampling buffer (Novex) were resolved on a 4–12 % NuPAGE gel (Novex) and transferred onto a PVDF membrane (Bio-Rad). Blots were probed with primary antibodies in Tris-buffered saline containing 0.1 % Tween 20 and 5 % nonfat milk overnight, followed by HRP-conjugated anti-rabbit antibody (Cell Signaling, 7074S, RRID: AB_2099233, 1:2000) or anti-mouse antibodies (Cell Signaling, 7076S, RRID: AB_330924, 1:2000). Primary antibodies used include anti-Flag (Millipore Sigma, F1804, RRID: AB_262044, 1: 2000), anti-APP-Y188 (Abcam, AB32136, RRID: AB_2289606, 1: 3000), anti-APP-2E9 (Millipore Sigma, MABN2295, RRID: AB_3107050, 1: 1000), anti-vinculin (Abcam, AB129002, RRID: AB11144129, 1: 2000), anti-CD9 (Cell Signaling technology, 13174, RRID: AB_2798139, 1: 1000), anti-VSV-G (Santa Cruz Biotechnology, SC-365019, RRID: AB_10846802, 1: 2000), anti-GFP (Cell Signaling Technology, 2956, RRID: AB_1196615, 1: 2000), and anti-GAPDH (Santa Cruz Biotechnology, SC-47724, RRID: AB_627678, 1: 2000).

2.9. T7 Endonuclease I assay

Genomic DNA extraction was performed using Qiagen DNeasy Blood and Tissue Kit (catalog # 69504) following the manufacture’s instruction. DNA fragments containing the gene editing sites with different sgRNAs were PCR amplified with proper primers and the PCR products were further purified with Monarch^® PCR and DNA cleanup kit (New England Biolabs, T1030S). The T7EI assay was performed following the manufacturer’s instructions (New England Biolabs). Briefly, 200 ng of genomic PCR product was combined with 2 μl of NEBuffer 2 (New England Biolabs) and diluted to a final volume of 19 μl. The samples were hybridized by denaturation at 95 °C for 5 min, followed by a ramp down to 85 °C at −2 °C/s and then to 25 °C at −0.1 °C/s. After hybridization, 1 μl of T7EI (M0302, New England Biolabs) was added, mixed thoroughly, and incubated at 37 °C for 15 min. The reaction was terminated by adding 1.5 μl of 0.25 M EDTA, and the products were visualized on a 2 % agarose gel.

2.10. Genomic DNA sequencing, and data analysis

The purified amplicon from genomic PCR containing the gene editing site was sequenced with proper primers using Sanger sequencing on a 3730xl DNA Analyzer (Thermo Fisher Scientific) at ETON Bioscience Inc. (San Diego, CA) and the sequencing results were analyzed using the TIDE platform.²⁵ Amplicons containing the gene editing site of APP in iPSC GlutaNeurons were PCR amplified with specific primers and adapters. These fragments were sequenced using Amplicon-EZ (100–500 bp) by GENEWIZ/Azenta Life Sciences, yielding approximately 75,000 to 250,000 reads per sample. The data were analyzed and visualized with NGS Genotyper (v 1.4.0). Sequences with misalignments, gaps, or insertions around the expected cut site were classified as NHEJ events. The frequency, length, and position of matches, insertions, deletions, and mismatches were all tracked in the resulting aligned sequences.

2.11. Determination of Aβ40 and Aβ42 peptides

Human Aβ40 and Aβ42 were detected using ELISA kits (KHB3481 and KHB3544) following the manufacturer’s instructions. Briefly, supernatants from ReNcells or human iPSC GlutaNeurons were collected and diluted (5x for ReNcells, 2x for iPSC GlutaNeurons). The diluted supernatants and detection antibodies were added to the provided wells and incubated for 3 h at room temperature with shaking. After washing (4x), anti-Rabbit IgG HRP solution was added and incubated for 30 min at room temperature. The stabilized chromogen was then added and incubated for another 30 min at room temperature in the dark. After adding the stop solution, absorbance at 450 nm was measured using a luminescence microplate reader, and peptide concentrations were calculated from a calibration curve with peptide standards.

2.12. Statistical analysis

Statistical analysis and plotting were performed using Prism software (GraphPad, version 10.1.0). An unpaired, two-tailed Student’s t-test was used to compare two groups, while one-way ANOVA followed by Tukey’s multiple comparison test was used for multiple groups. A P-value <0.05 was considered significant.

3. Results

3.1. Packaging of CRISPR-Cas9 complex via ARMMs

ARMMs have previously been shown to facilitate the packaging and delivery of CRISPR-Cas9 complexes relying on ARRDC1’s interaction with WW-domain proteins.²¹ To investigate whether ARMMs can independently package CRISPR-Cas9, we created two fusion constructs by linking ARRDC1 to the N-terminus of Cas9 (Fig. 1A). Due to the large size of Cas9 (approximately 160 kDa), we designed one construct with full-length ARRDC1 (A1-Cas9) and another with a truncated version of ARRDC1 (sA1-Cas9), retaining only the essential domains required for vesicle budding, namely the N-terminal arrestin domain and the PSAP and PPXY motifs (Fig. 1A). Both A1-Cas9 and sA1-Cas9 retained comparable gene-editing activity to unmodified Cas9 (Fig. S1), indicating that the fusion did not impair CRISPR-mediated gene editing. To evaluate vesicle budding, we transfected HEK293T cells with these constructs and analyzed the extracellular vesicles (EVs) produced using nanoparticle tracking analysis. As shown in Fig. 1B, EVs from A1-Cas9 and sA1-Cas9-transfected cells were significantly more abundant than those from Cas9-transfected or non-transfected cells, with sA1-Cas9 generating the highest number of EVs.

Fig. 1. Packaging of Cas9 into ARMMs.

(A) Schematic drawings of original construct with wild type Cas9 and modified constructs with direct fusion of full length- or shortened ARRDC1 at N-terminus of Cas9. (B) Nanoparticle Tracking Analysis (NTA) of EVs collected from HEK293T cells transfected with different Cas9 plasmids. Data were obtained from five replicates for each condition. *P < 0.05, ***P < 0.005, ****P < 0.0001. (C) Western blotting results showing the Cas9 packaged in ARMMs from HEK293T cells. ARMMs were isolated via ultracentrifugation. Western blotting was done on the EVs along with the whole cell lysates with indicated antibodies. MW, molecular weight. (D) Western blotting data were analyzed via image quantification using ImageJ software. Each band intensity of Flag signals in EV was normalized with the band intensity of cell lysate (left) and CD9 signals (right), respectively. Western blotting was repeated three times to obtain the data for each band. *P < 0.05, **P < 0.01, ***P < 0.005.

We then used Western blotting to assess the incorporation of Cas9 protein (Flag-tagged) into EVs. As shown in Fig. 1C, Flag signals were detected in cell lysates, confirming the presence of Cas9. Flag signals were also observed in EV fractions, but only in those from A1-Cas9 and sA1-Cas9-transfected cells. Statistical analysis revealed stronger Flag signals in the sA1-Cas9 group compared to A1-Cas9, with both absolute and relative intensities normalized to CD9 serving as loading controls (Fig. 1D). These results indicate that both full-length and truncated ARRDC1 fusions promote Cas9 packaging into ARMMs, with sA1-Cas9 demonstrating superior efficiency.

3.2. VSV-G pseudotyped ARMMs packaging CRISPR-Cas9

The fusogenic glycoprotein of vesicular stomatitis virus (VSV-G) was initially used as a substitute for the envelope glycoprotein in virus-derived vectors and has been extensively employed for pseudotyping to improve viral vector delivery.^26-28 We next explored whether VSV-G can pseudotype ARMMs and improve budding and packaging (Fig. 2A). Due to potential VSV-G cytotoxicity,^29,30 we first optimized its concentration by co-transfecting Cas9 constructs with GFP as a filler, using up to 25 % VSV-G in HEK293T cells (Fig. 2B). Significant cytotoxicity was observed with 10 % VSV-G as evidenced by representative cellular images (Fig. 2C) and lactate dehydrogenase (LDH) assay results (Fig. 2D). Western blot analysis showed stronger Flag signals with the presence of VSV-G co-transfection. However, non-specific signals (GFP, vinculin, GAPDH) in EVs were also observed in high VSV-G groups (Fig. 2E). Therefore, we choose 5 % VSV-G for subsequent experiment. We then assessed EV budding and Cas9 packaging in cells co-transfected with Cas9 constructs and either 2.5 % or 5 % VSV-G. NanoSight nanoparticle tracking analysis revealed a dose-dependent enhancement of EV budding with increasing VSV-G concentration (Fig. 2F). Importantly, transfection with unmodified Cas9 and VSV-G alone did not result in Cas9 packaging into EVs, confirming the ARRDC1 dependence for Cas9 incorporation (Fig. S2A and B). Additionally, VSV-G pseudotyping did not alter the size of EVs packaging Cas9 proteins as evidenced by the OptiPrep-based density gradient fractionation Western blotting results (Fig. S2C).

Fig. 2. Pseudotyping ARMMs with VSV-G.

(A) Schematic drawing of the pseudotyping strategy for ARMMs with VSV-G. ARMMs were generated from transfection shortened ARRDC1 fused Cas9 construct (sA1-Cas9) in HEK293T cells and further collected via ultracentrifugation. (B) Experimental settings to determine the effects of different percentage of VSV-G on ARMMs budding and Cas9 packaging. (C) Representative cellular images of HEK293T cells after 72-hr transfection with different groups of plasmids. Scale bar = 100 μm. (D) Cytotoxicity results from HEK293T cells after 72-hr transfection with different groups of plasmids. LDH data was normalized to the non-transfected negative control group. *P < 0.05, ****P < 0.0001. (E) Western blotting results showing the Cas9 packaged in ARMMs and effects of different percentage of VSV-G on the packaging efficiency of Cas9 and unspecific packaging of GFP in ARMMs. ARMMs were isolated from HEK293T cells via ultracentrifugation. Western blotting was done on the EVs along with the whole cell lysates with indicated antibodies. (F) Nanoparticle Tracking Analysis (NTA) of EVs collected from HEK293T cells transfected with sA1-Cas9 alone and with two different percentage of VSV-G. Data were obtained from five replicates for each condition. ****P < 0.0001. (G) Quantification of Cas9 molecules in ARMMs from sA1-Cas9 alone and sA1-Cas9 with 5 % VSV-G co-transfection. The calculation was based on Western blotting results that detect Flag signals from recombinant dCas9 protein standards along with EVs. Western blotting was performed three times. *P < 0.05.

To further evaluate Cas9 packaging efficiency into EVs, we compared the incorporation of Cas9 into vesicles using the sA1-Cas9 constructs in the presence of VSV-G. Flag signals from equal amounts of isolated EVs were quantified along with recombinant Cas9 protein standards (Fig. S2D). Using a standard calibration curve for Cas9, we found that VSV-G significantly enhanced Cas9 packaging efficiency, with the sA1-Cas9 and VSV-G combination achieving the highest packaging efficiency—approximately 1000 Cas9 molecules per vesicle (Fig. 2G). These results demonstrate that VSV-G enhances Cas9 packaging into EVs, with the truncated ARRDC1 version leading to greater EV budding and packaging efficiency.

3.3. Delivery of CRISPR-Cas9 via ARMMs in U2OS cells

It has been demonstrated that the Cas9/sgRNA complex packaged into ARMMs can be delivered to recipient cells,²¹ we assessed whether ARMMs packaging Cas9 through direct fusion with ARRDC1 can deliver their cargo and exhibit gene editing activity. ARMMs were produced via transfections targeting the GFP gene with Cas9/sgRNA complexes. These ARMMs were incubated with U2OS cells containing a single GFP gene copy³¹ and gene knockout efficiency was measured by changes in GFP expression (Fig. 3A). After incubation, a significant increase in GFP-negative cells was observed compared to the untreated control group (Fig. 3B). No significant differences were observed between A1-Cas9 and sA1-Cas9 without VSV-G pseudotyping. However, when 5 % VSV-G was used, a marked increase in GFP-negative cells was detected, with sA1-Cas9 demonstrating higher editing efficiency (Fig. 3B). To further assess the impact of ARMMs on gene editing, we tested varying concentrations of ARMMs from sA1-Cas9 transfections (up to 5 × 10⁵ EVs per cell) in U2OS-GFP cells. Western blotting confirmed the presence of Cas9 in recipient cells and a reduction in GFP signals, indicating successful ARMMs uptake and gene editing activity (Fig. 3C). Notably, a concentration-dependent increase in GFP-negative cells was observed, with the highest editing efficiency achieved at 5 % VSV-G (Fig. 3D). Both T7E1 assays (Fig. S3A) and direct DNA sequencing (Fig. 3E) confirmed successful gene editing of the GFP target gene following ARMMs treatment. These findings demonstrate that ARMMs, packaged with Cas9/sgRNA via direct ARRDC1 fusion, can be efficiently delivered to recipient cells while retaining gene editing capability. Furthermore, ARMMs from sA1-Cas9 exhibited enhanced editing efficiency when pseudotyped with VSV-G.

Fig. 3. Gene editing via ARMMs targeting exogenous gene in U2OS-gfp cells.

(A) Schematic drawing of experimental design for detecting the potential gene editing via ARMMs delivery. Cas9 constructs with direct fusion of full length (A1-Cas9) or shortened ARRDC1 (sA1-Cas9) and VSV-G were used for the transfection in HEK293T cells. Single guide RNA targeting GFP gene was designed. ARMMs were collected via ultracentrifugation and incubated with U2OS-gfp cells. (B) Flow cytometry data showing the decrease of GFP signals in U2OS-gfp cells incubating with ARMMs at the concentration of 2E5 EVs per cell for 48 h. ARMMs were collected from HEK293T cells via ultracentrifugation after transfections with ARRDC1 (full length or shorten) directly fused Cas9 constructs. GFP-negative cell percentage were summarized and effects of VSV-G pseudotyping were compared. ****P < 0.0001. (C) Western blotting results showing the uptake of Cas9 in recipient U2OS-gfp cells and the change of GFP signals. Cells were incubated ARMMs from sA1-Cas9 with and without VSV-G co-transfection at two different concentrations for 48 h and the whole cell lysates were collected for Western blotting. (D) Statistical results summarizing the concentration-response of GFP changes in U2OS-gfp cells incubated with ARMMs for 48 h ****P < 0.0001. (E) Statistical results showing the editing efficiency in U2OS-gfp cells incubated with different concentrations of ARMMs packaging CRISPR-Cas9 targeting gfp gene. The amplicon contains the predicted editing site was amplified from genomic DNA and subjected to Sanger sequencing. The editing efficiency was quantified from TIDE online platform.

3.4. Gene editing via ARMMs delivery in human neuronal cells

We have demonstrated that ARMMs can efficiently deliver the CRISPR-Cas9 gene editor to recipient cells, achieving effective knockout of an exogenous gene. Notably, ARMMs derived from sA1-Cas9 showed higher editing efficiency compared to A1-Cas9. To investigate the potential of ARMMs as a delivery system for CRISPR-Cas9 targeting an endogenous gene, we used human neuronal cells in vitro, given the growing interest in gene therapy for these cells. APP gene, encoding amyloid precursor protein, was selected as the target gene due to its role in neurodegenerative diseases; modifications at the C-terminus of APP protein have been reported to reduce pathogenic activity by altering cleavage processes.^32-34 We incubated human neural progenitor ReNcell CX cells (ReNcells) and human iPSC-derived neurons with ARMMs packaging Cas9/sgRNA and assessed knockout efficiencies as well as the downstream functional assay by detecting the concentration of amyloid beta peptides (Aβ40 and Aβ42) resulting from beta-/gamma-cleavage of APP protein (Fig. 4A). Initially, VSV-G pseudotyping effects on ARMMs’ knockout efficiency in ReNcells were evaluated using a consistent ARMMs concentration. Results showed higher efficiency with ARMMs pseudotyped with 2.5 % or 5 % VSV-G compared to non-VSV-G or higher percentage groups (Fig. S4). Based on previous packaging efficiency findings, 5 % VSV-G was used for subsequent experiments. ReNcells were incubated with various ARMM concentrations, cell lysates were collected for Western blotting and PCR products from genomic DNA at the knockout site were sequenced. As shown in Fig. 4B, editing efficiency was concentration-dependent, reaching up to 90 % at 5X10⁵ EVs/cell, significantly surpassing plasmid direct transfection via electroporation (Fig. S5). Furthermore, Western blotting confirmed Flag signals of Cas9 proteins in ReNcells and a decrease in both C- and N-termini of APP, indicating successful ARMM-mediated gene editing (Fig. 4C). In addition, ELISA results also showed reduced Aβ40 and Aβ42 concentrations post-ARMMs incubation (Fig. 4D).

Fig. 4. Gene editing via ARMMs targeting endogenous gene in human neural cell models.

(A) Schematic drawing of experimental design using ARMMs to deliver CRISPR-Cas9 complex targeting amyloid precursor protein (APP) in human neural progenitor cells (ReNcells) and matured neurons (iPSC derived neurons). Cells were incubated with VSV-G pseudotyped ARMMs from sA1-Cas9 and the gene editing efficiency as well as the functional effects on amyloid peptide formations were determined. (B) Statistical results summarizing the concentration-response of gene editing efficiency in ReNcells incubating with ARMMs targeting APP. The amplicon contains the predicted editing site was amplified from genomic DNA and subjected to Sanger sequencing. The editing efficiency was quantified from TIDE online platform. (C) Western blotting results showing the uptake of Cas9 in ReNcells and the change of APP protein. Cells were incubated with different concentrations of ARMMs for 48 h and whole cell lysates were collected for Western blotting. (D) ELISA results showing the concentrations of amyloid beta peptides. Cells were incubated with ARMMs for 48 h and the supernatant medium was collected for ELISA detecting amyloid beta-40 and −42 peptides. **P < 0.01, ****P < 0.0001. (E) Western blotting results showing the uptake of Cas9 in iPSC neurons and the change of APP protein. Cells were incubated with different concentrations of ARMMs for 48 h and whole cell lysates were collected for Western blotting. (F) Statistical results summarizing the concentration-response of gene editing efficiency in iPSC neurons incubating with ARMMs targeting APP. Genomic DNA was collected from cells after 48-hr incubation with ARMMs and the amplicon containing the predicted editing site was amplified and further sequenced. The sequencing results were further quantified via TIDE platform. (G) Genomic deep sequencing results of APP gene in iPSC neurons incubating with ARMMs showing the Indel percentage (Top), major mutated APP loci resulting from CRISPR editing via ARMMs (Middle), and predicted translational products (post-editing) for the major mutant alleles observed in deep sequencing (Bottom). (H) ELISA results showing the concentrations of amyloid beta peptides from iPSC neurons after incubation with ARMMs. Cells were incubated with ARMMs for 48 h and the supernatant medium was collected for ELISA detecting amyloid beta-40 and −42 peptides. ***P < 0.005, ****P < 0.0001.

Furthermore, we tested ARMM-mediated gene editing in mature human iPSC-derived neurons. Western blotting results confirmed the uptake of ARMMs in iPSC neurons and decreased APP protein signals further indicated gene editing occurred with ARMMs incubation (Fig. 4E). Gene editing efficiency from direct DNA sequencing further showed the concentration-dependent response with ARMMs incubation in recipient iPSC neurons (Fig. 4F). In addition, genomic deep sequencing revealed efficient APP editing as evidenced by the indel percentage (Top), major mutated APP loci resulting from CRISPR editing via ARMMs (Middle) and predicted translational products (post-editing) for the major mutant alleles in Fig. 4G. Similarly, ELISA results also showed significant reduced Aβ40 and Aβ42 concentrations post-ARMMs incubation in iPSC neurons (Fig. 4H).

Importantly, we further determined the potential off-target activity in top four computationally predicted off-target sites (Fig. S6A). No editing was observed using T7 endonuclease assays in ReNcells with direct transfection with Cas9/sgRNA constructs (Fig. S6B) or with ARMMs incubation (Fig. S6C). Similarly, no editing from the direct DNA sequencing was seen in two APP homologues APLP1/2 that undergo similar processing as APP (Fig. S6D-F). Together, these findings demonstrate that ARMMs packaging the CRISPR-Cas9 complex can be effectively delivered to both human neural progenitor cells and mature neurons, achieving high-efficiency gene editing targeting APP.

4. Discussion

In this study, we demonstrated that ARRDC1-mediated microvesicles (ARMMs) can effectively deliver the CRISPR-Cas9 complex, leveraging a direct fusion of ARRDC1 to the N-terminus of Cas9 protein. This fusion enables efficient packaging of the CRISPR-Cas9 complex, with a truncated ARRDC1 enhancing vesicle budding and overall packaging efficiency. Furthermore, VSV-G pseudotyping was found to significantly improve vesicle budding and Cas9 incorporation into ARMMs. These Cas9-loaded ARMMs exhibited robust gene-editing activity in both exogenous and endogenous genes across a range of human cell types, including U2OS cells and human neuronal cells (such as neural progenitor cells and mature neurons). Notably, ARMM-mediated editing of the APP gene in human neuronal cells led to a reduction in amyloid peptide levels, suggesting potential therapeutic applications for neurodegenerative diseases. Together, these findings position ARMMs as a novel and highly efficient platform for delivering CRISPR-Cas9-based gene editing, particularly for human disease therapies.

ARMMs have previously been engineered as versatile delivery systems for a variety of biological cargos, including RNA, proteins, and ribonucleoprotein complexes like Cas9/sgRNA.²¹ Prior strategies typically involved engineering the Cas9 protein with WW domain fusions from the ITCH protein, relying on interactions between the PPXY motifs in ARRDC1 and the WW domains to facilitate vesicle budding and cargo packaging.²¹ In contrast, our study enhances this approach by directly fusing ARRDC1 to the N-terminus of Cas9, eliminating the need for external interactions and addressing the limitations posed by multiple constructs or competition between other WW domain-containing proteins. This modification significantly improves the efficiency of packaging the CRISPR-Cas9 complex into ARMMs while preserving the gene-editing activity of Cas9, as shown by its comparable performance to wild-type Cas9 (Fig. S1). Importantly, the ARRDC1-Cas9 fusion construct demonstrated equivalent gene-editing efficiency in vitro, targeting the exogenous GFP gene in U2OS cells (Fig. 3), compared to ARMMs using WW domain-ARRDC1 interactions.²¹ This system represents the first single-construct approach capable of driving ARMM budding and effectively packaging the CRISPR-Cas9 complex without compromising gene-editing functionality.

Further improvement was achieved by generating ARMMs using a truncated version of ARRDC1 (sA1-Cas9), which exhibited enhanced vesicle budding and CRISPR-Cas9 packaging efficiency. This advancement represents a significant step forward compared to traditional CRISPR-Cas9 delivery systems. Viral vectors, for example, face limitations in packaging capacity (usually under 4 kb) and are associated with risks of genetic mutations and immunogenicity.^35,36 In contrast, ARMMs overcome these limitations by offering efficient cargo packaging and cellular uptake without the drawbacks of viral systems. Our findings show that sA1-Cas9 fusion constructs integrated into ARMMs not only facilitate highly efficient Cas9 delivery but also outperform conventional viral and non-viral delivery methods. Moreover, the ubiquitous expression of ARRDC1 suggests that ARMMs are likely to be well-tolerated across diverse cell types, making them a promising platform for the therapeutic delivery of gene-editing machinery and other biomolecules.

Additionally, we improved the efficiency of vesicle budding and Cas9 incorporation by incorporating VSV-G pseudotyping. Consistent with previous reports, VSV-G is crucial for enhancing genome-editing activity by promoting the escape of delivered ribonucleoproteins from endosomes in recipient cells.³⁷ Other studies have shown that EVs without VSV-G suffer from rapid cargo degradation in recipient cells.^38,39 In our study, even at lower VSV-G concentrations, optimal packaging efficiency was achieved, minimizing the cytotoxicity seen with higher concentrations used in previous studies.³⁷ These improvements enabled ARMMs to package approximately 1000 Cas9 proteins per vesicle (Fig. 2G), a substantial increase compared to non-viral delivery methods, which typically package only tens of Cas9 molecules per vesicle.⁴⁰ While viral platforms, such as lentivirus-like particles, can package up to seven hundred Cas9 molecules,⁴¹ ARMMs offer additional advantages, including lower toxicity, reduced immunogenicity, and the preservation of functional vesicles. Together, these features make ARMMs a competitive and promising platform for therapeutic delivery.

Furthermore, we demonstrated the successful application of ARMMs to target the APP gene in human neuronal cells, an important step toward advancing extracellular vesicle-based gene therapies. Gene editing in neuronal cells, especially for endogenous targets, has traditionally been challenging due to issues with efficient delivery and minimizing off-target effects. While viral vectors like AAVs and lipid nanoparticles are often employed to cross the blood-brain barrier, they still carry risks of immune activation and off-target effects.^42,43 Exosome-based delivery systems have shown promise for gene editing in neuronal models, such as targeting α-Synuclein in Parkinson’s disease, but these systems face limitations in cargo capacity and scalability.¹⁸ In contrast, ARMMs offer a scalable, non-viral, and efficient delivery platform, as demonstrated by our successful editing of the APP gene and reduction of amyloid peptide formation in human neuronal cells (Fig. 4). These results are consistent with previous in vivo findings.³⁴ Together, these results highlight the potential of ARMMs as a scalable, non-viral, and less invasive tool with enhanced loading and gene-editing efficiency, representing a significant improvement over existing delivery methods.

While our findings are promising, several limitations should be considered. First, our study focused on in vitro cell cultures, and further research is needed to assess the efficiency and safety of ARMMs in vivo, where factors such as tissue-specific delivery and gene-editing efficiency may present additional challenges. Second, the mechanism through which VSV-G enhance Cas9 packaging into ARMMs remain unknown, and the use of VSV-G, which binds to ubiquitously expressed LDL receptors,⁴⁴ results in a lack of specificity in tissue targeting. Future studies could address the latter limitation by incorporating tissue-specific targeting elements, such as single-chain variable fragment (scFv) antibodies, to enhance ARMM specificity. Finally, the efficiency of ARMM-mediated gene editing may vary depending on the guide RNAs used for different genetic targets, highlighting the need for extensive optimization and testing of various CRISPR components.

In summary, this study demonstrates that ARMMs represent a highly efficient and scalable platform for CRISPR-Cas9 delivery, with significant potential for therapeutic applications in neurodegenerative diseases and other human disorders. By leveraging a direct fusion of ARRDC1 with Cas9 and incorporating VSV-G pseudotyping, we have optimized ARMMs for efficient gene editing with minimal toxicity and immunogenicity. Our findings provide a foundation for using ARMMs to deliver not only gene editors but potentially other therapeutic cargoes for genetic medicine.

Supplementary Material

Supplementary data to this article can be found online at https://doi.org/10.1016/j.vesic.2025.100082.