In vivo human T cell engineering with enveloped delivery vehicles

Viruses and virally derived particles have the intrinsic capacity to deliver molecules to cells, but the difficulty of readily altering cell-type selectivity has hindered their use for therapeutic delivery. Here, we show that cell surface marker recognition by antibody fragments displayed on membrane-derived particles encapsulating CRISPR–Cas9 protein and guide RNA can deliver genome editing tools to specific cells. Compared to conventional vectors like adeno-associated virus that rely on evolved capsid tropisms to deliver virally encoded cargo, these Cas9-packaging enveloped delivery vehicles (Cas9-EDVs) leverage predictable antibody–antigen interactions to transiently deliver genome editing machinery selectively to cells of interest. Antibody-targeted Cas9-EDVs preferentially confer genome editing in cognate target cells over bystander cells in mixed populations, both ex vivo and in vivo. By using multiplexed targeting molecules to direct delivery to human T cells, Cas9-EDVs enable the generation of genome-edited chimeric antigen receptor T cells in humanized mice, establishing a programmable delivery modality with the potential for widespread therapeutic utility.

Viruses and virally derived particles have the intrinsic capacity to deliver molecules to cells, but the difficulty of readily altering cell-type selectivity has hindered their use for therapeutic delivery.Here, we show that cell surface marker recognition by antibody fragments displayed on membrane-derived particles encapsulating CRISPR-Cas9 protein and guide RNA can deliver genome editing tools to specific cells.Compared to conventional vectors like adeno-associated virus that rely on evolved capsid tropisms to deliver virally encoded cargo, these Cas9-packaging enveloped delivery vehicles (Cas9-EDVs) leverage predictable antibody-antigen interactions to transiently deliver genome editing machinery selectively to cells of interest.Antibody-targeted Cas9-EDVs preferentially confer genome editing in cognate target cells over bystander cells in mixed populations, both ex vivo and in vivo.By using multiplexed targeting molecules to direct delivery to human T cells, Cas9-EDVs enable the generation of genome-edited chimeric antigen receptor T cells in humanized mice, establishing a programmable delivery modality with the potential for widespread therapeutic utility.
Therapeutic interventions involving genome editing require the safe and effective delivery of molecules into target cell nuclei [1][2][3] .Although such capability would transform both clinical and research applications, current non-viral delivery is limited to cells treated ex vivo [4][5][6] , tissues targeted by local administration 7,8 or the liver because of its natural propensity for molecular uptake 8,9 .Recent lipid nanoparticle formulations have been described with tropism for non-hepatic cells or organs 10,11 , but expansion of in vivo genome editing applications will probably require multiple approaches for molecular delivery to specific cells or organs inside the body following systemic administration.
Retargeting the tropism of viruses or viral vectors is an established delivery strategy involving the surface display of a cell-selective targeting molecule alongside a viral glycoprotein required for cell entry by fusion at the plasma membrane or in the low-pH environment of the endosome [12][13][14][15] .Recent progress leverages a mutant form of the vesicular stomatitis virus glycoprotein (VSVG), VSVGmut, that maintains endosomal fusion activity but lacks native low-density lipoprotein receptor binding affinity [16][17][18] .Pairing VSVGmut with cell-specific targeting molecules can redirect lentiviral transgene delivery and has enabled high-throughput screening of T cell and B cell receptor libraries to study receptor-antigen interactions 19,20 .
Particles cloaked in cellular membrane fragments-such as retrovirus-like particles (VLPs), extracellular vesicles and biomimetic nanoparticles-are gaining in popularity for the delivery of molecular https://doi.org/10.1038/s41587-023-02085-z Cas9 RNP complexes produced by over-expressing Cas9 fused to the carboxy-terminal end of the viral Gag polyprotein during VLP production, but cell-selective VLP targeting has relied on cell infection strategies evolved by enveloped viruses 22 .To test whether VLPs could be reformulated as programmable EDVs, we first cloned a CD19 targeting antibody as an scFv fused to the stalk and transmembrane domain of CD8a, a strategy commonly used in CAR architecture 28 (Fig. 1a and Supplementary Fig. 1a,b).Given that Cas9-VLPs bud from the plasma membrane of transfected producer cells, we reasoned that co-expression of the scFv fusion and VSVGmut together with lentiviral components that are necessary for Cas9 RNP encapsulation would generate Cas9-EDVs possessing both receptor specificity and endosomal escape capability, respectively.
To analyze the receptor-mediated function of Cas9-EDVs, we generated a HEK293T cell line that co-expresses both the B cell ligand CD19 and EGFP, enabling assessment of genome editing in on-target EGFP + cells and off-target EGFP − bystander cells (Fig. 1b).We produced Cas9-EDVs containing single guide RNA (sgRNA) targeting the β-2 microglobulin (B2M) gene and outwardly displaying either VSVG, CD19 scFv+VSVGmut or a control scFv+VSVGmut that should not recognize the target cells in this experiment.In an ~3:1 mixture of HEK293T and CD19 EGFP 293T cells, the VSVG Cas9-EDVs mediated genome editing in both CD19 + and CD19 − populations, whereas the CD19-scFv Cas9-EDVs induced the knockout of B2M only in CD19 + cells (Fig. 1c).No B2M knockout was observed in either the CD19 + or CD19 − cells using the control scFv+VSVGmut Cas9-EDVs.Antibody-targeted Cas9-EDV activity was titratable, with up to 74% of target cells exhibiting B2M knockout and little to no editing detected in bystander, non-target cells (Fig. 1d,e).Antibody-targeted Cas9-EDVs produced genome edits in target cells present at 2-92% of a cell mixture, whereas bystander cell editing was unchanged (Supplementary Fig. 1c).Together, these results demonstrate the ability of EDVs to deliver functional molecular cargo in a receptor-mediated fashion.

Programmable cell-specific genome editing with Cas9-EDVs
Receptor-mediated delivery of genome editing molecules could enable targeted engineering of any cell type as a function of its surface antigens.To test this possibility, we investigated the modularity and programmability of Cas9-EDVs to direct genome editing in HEK293T cells displaying various plasma membrane proteins normally expressed by human immune cells, including CD20, CD4 and CD28 (Fig. 1b and Supplementary Fig. 2a).Cas9-EDVs displaying VSVGmut and scFv-based CD20, CD4 and CD28 targeting molecules, generated in both variable heavy (VH)-linker-variable light (VL) and VL-linker-VH orientations (Supplementary Table 1) 29 , induced up to 80% genome editing that was titratable and selective for ligand + over ligand − cells (Supplementary Fig. 2b).Not all scFvs produced the same level of on-target cell editing, but in no case was an scFv-displaying EDV able to induce more than minimal editing in off-target cells lacking the cognate surface receptor protein.Furthermore, in all engineered cell mixtures, ligand + and ligand − cells were similarly susceptible to genome editing when treated with control Cas9-EDVs that express the VSVG fusogen (Supplementary Fig. 2b).A panel of CD19, CD20 and CD4 antibody-targeted EDVs only mediated genome editing in cells expressing their matched ligand and not in mismatched ligand-expressing cells, demonstrating that delivery requires antibody-antigen interactions (Supplementary Fig. 2c).

Cas9-EDV optimization and study of nonessential components
To enhance Cas9-EDV yield and per-particle editing efficiency ahead of in vivo administration, we screened multiple N-terminal nuclear localization signal (NLS) additions to Cas9 and found that 2× p53-derived NLS together with 3× nuclear export sequences (NESs) appended to the C-terminal end of Gag 8 most improved the editing efficiency of antibody-targeted Cas9-EDVs (Supplementary Fig. 3a-d and Supplementary Table 2).Editing efficiency was further increased by expressing sgRNA from both the Gag-NES-NLS-Cas9 and Gag-pol plasmid backbones, as opposed to expression from a separate plasmid (Fig. 2a,b).We speculate that these optimizations improved Cas9 RNP loading into EDVs during assembly as well as enhanced Cas9 RNP nuclear import in target cells.Optimized Cas9-EDVs maintained receptor-mediated delivery specificity except at the highest doses tested (Supplementary Fig. 3e), and Cas9-EDV titration produced a 36-fold enrichment for genome editing on-target cells (79.7%) versus bystander cells (2.2%) (Supplementary Fig. 3f).Optimized Cas9-EDVs pseudotyped with the broadly transducing VSVG glycoprotein also demonstrated improved genome editing activity when tested on cytokine-stimulated primary human CD34 + cells and cytokine-stimulated and activated primary human T cells ex vivo (Fig. 2c,d).Surprisingly, optimized VSVG-pseudotyped Cas9-EDVs mediated genome editing in resting primary human T cells (Fig. 2e), which are difficult to edit using standard electroporation approaches.This suggests that Cas9-EDVs may be an effective strategy for genome editing T cells in the absence of cellular activation, stimulation and expansion.
We next investigated whether the internal composition of Cas9-EDVs affects editing efficiency, focusing on the lentiviral capsid that forms during proteolytic virion maturation 30 (Fig. 2f).To probe the role of the capsid in Cas9-EDV delivery, we employed GS-CA1, a small-molecule inhibitor of nuclear import and/or subsequent uncoating of HIV-1 capsid cores 31,32 (Fig. 2g).Treatment of target cells with increasing concentrations of GS-CA1 blocked the integration of a lentiviral transgene, which relies on nuclear import from the capsid (Fig. 2h), but did not negatively impact the genome editing efficiency of Cas9-EDVs (Fig. 2i) or electroporated Cas9 RNPs (Supplementary Fig. 3g).In a separate experiment, we generated a Cas9-EDV variant that relies on the tobacco etch virus (TEV) protease to release Cas9 from Gag ('TEVp-Cas9-EDVs'; Supplementary Fig. 3h), preventing the HIV-1 protease-dependent virion maturation required for capsid assembly.Despite the demonstrated loss of Gag proteolytic processing (Supplementary Fig. 3i), TEVp-Cas9-EDVs maintained genome editing activity in treated cells proportional to the amount of Cas9 generated (Supplementary Fig. 3j).These results suggest that the capsid is not required for packaging and delivering Cas9 RNP complexes into target cell nuclei.

Optimized Cas9-EDV characterization
We next performed characterization of Cas9-EDVs to better understand particle composition and genome editing activity.Cas9-EDVs are similar in diameter to lentiviral vectors (LVs) (Supplementary Fig. 4a,b), and multiple scFvs are detectable on the surface of antibody-targeted particles (Supplementary Fig. 4c).Interestingly, we could detect Cas9-independent packaging of over-expressed sgRNA into Cre recombinase-packaging Cas9-EDVs, but sgRNA packaging was enhanced ~330-fold in Cas9-containing particles (Supplementary Fig. 4d).While others have shown the unintended packaging of cellular RNAs and proteins into retroviral vectors 23,33 , which probably occurs in Cas9-EDVs, we could not detect unintended Cas9-EDV-mediated delivery of plasmids from producer cells to treated cells (Supplementary Fig. 4e,f).
Using synthetic sgRNA as a standard curve (Supplementary Fig. 4g), we estimate that Cas9-EDVs produced in one 10 cm plate contain ~2.66 × 10 11 sgRNA molecules (Supplementary Fig. 4g) distributed among approximately 5.65 × 10 10 Cas9-EDV particles (Supplementary Fig. 4i).Benchmarking Cas9-EDVs against Cas9 RNP nucleofection, we found that treatment with 0.2 μl of 30-fold concentrated Cas9-EDVs resulted in the equivalent amount of genome editing as ~23 pmol of nucleofected RNP (Supplementary Fig. 4j-l).Finally, we tested the biodistribution of wild-type VSVG versus VSVGmut pseudotyped LVs and found that relative to the amount of vector in the serum, both   22 are highlighted in teal.c-e, Genome editing activity of optimized VSVGpseudotyped Cas9-EDVs in primary human CD34 + cells (c) and activated (d) and resting primary human T cells (e).B2M or TRAC genome editing was assessed by amplicon sequencing 7 days post treatment; n = 3 technical replicates were assessed for all conditions except for the untreated resting human T cells (n = 2).f, Schematic of potential intra-particle Cas9-EDV configurations for packaged Cas9 RNPs following proteolytic maturation.g, Schematic of the compound GS-CA1 inhibiting either the nuclear import and/or uncoating of an HIV-1 capsid.h, An mNeonGreen lentiviral vector was used to transduce HEK293T cells at the indicated multiplicity of infection (MOI) in the presence of GS-CA1 or DMSO.The percent of mNeonGreen-positive cells was assessed by flow cytometry 3 days post treatment.TU, transducing units.i, B2M-targeting Cas9-EDVs, pre-titered such that the highest treatment dose would result in approximately 50% of cells B2M − , were used to transduce HEK293T cells in the presence of GS-CA1 or DMSO.B2M expression was assessed by flow cytometry 3 days post treatment.Error bars, s.e.m.Unless otherwise noted, n = 3 technical replicates were used in all experiments; four-parameter non-linear regression curves are plotted in a, h and i.

Multiplexed targeting molecules for human T cell engineering
Human T cells are important targets for in vivo genome engineering applications because of their use in treating cancer and other diseases.Using CD25 expression as a marker, we found that the co-display of CD3 and CD28 targeting molecules on Cas9-EDVs triggered T cell activation and cellular expansion similar to T cells pretreated with commercially available CD3 or CD28 coated magnetic beads 34 or engineered lentiviruses 19 (Fig. 3a,b).CD3 + CD28 scFv Cas9-EDV treatment also led to robust levels of genome editing (Fig. 3c).
Further screening of CD3 and CD45 scFvs revealed additional Cas9-EDV targeting molecules that enabled genome editing of the human Jurkat T cell line (Fig. 3d and Supplementary Fig. 5a).The minimal human T cell editing observed with CD45-targeted Cas9-EDVs may result from a lack of CD45 internalization from the plasma membrane following monoclonal antibody engagement 35 (Supplementary Fig. 5b,c).Primary human T cells were susceptible to genome editing using CD3-targeted Cas9-EDVs and, to a lesser extent, CD4-targeted Cas9-EDVs, but not Cas9-EDVs pseudotyped with off-target control scFv targeting molecules (Fig. 3e and Supplementary Fig. 5d).Immunophenotyping of T cells post Cas9-EDV treatment showed that CD3-targeted Cas9-EDVs direct genome editing in CD4 + and CD8 + subsets of T cells as expected, whereas CD4 scFv-targeted Cas9-EDVs specifically mediated genome editing in the CD4 + subset population (Fig. 3e and Supplementary Fig. 5d).Interestingly, multiplexing CD3 and CD4 scFv targeting molecules on the same Cas9-EDVs led to higher levels of editing than Cas9-EDVs displaying either CD3 or CD4 scFv targeting molecules alone (Fig. 3e and Supplementary Fig. 5d).This observation was antigen-specific, as multiplexing CD3 targeting molecules with off-target control targeting molecules did not enhance genome editing.Receptor cross-linking or aggregation can lead to endocytosis and subsequent lysosomal degradation 36,37 , possibly explaining the synergistic increase in genome editing by engaging both CD3 and CD4 receptors.

T cell targeted Cas9-EDVs enable genome engineering in vivo
Antibody-targeted EDVs have the unique flexibility of enabling cell-specific delivery of either genome editors alone, lentiviral-encoded transgenes alone or genome editors and transgenes together.We next investigated the ability of antibody-targeted EDVs to perform cell-targeted engineering of human CAR T cells in vivo, an advance that could negate the delays and costs associated with ex vivo CAR T manufacturing 38,39 .Using immunodeficient mice engrafted with human peripheral blood mononuclear cells (PBMCs) to mimic a humanized immune system, we tested T cell-targeted vectors for their ability to generate either CAR T cells (LV) or gene-edited CAR T cells (Cas9-EDV  T cell-targeting Cas9-EDVs containing the CAR transgene (n = 4) or T cell-targeting lentivirus containing the CAR transgene (n = 3) were systemically administered and in vivo cell engineering was assessed 10 days post treatment (Fig. 4b).CAR-transduced T cells were observed in all mice in which human cells successfully engrafted, as detected by mCherry expression (Fig. 4c,d and Supplementary Fig. 6a-c).In the two Cas9-EDV-treated mice that successfully engrafted with human https://doi.org/10.1038/s41587-023-02085-zT cells (n = 2 out of 4), we observed 1.67% and 1.51% modified alleles in the CAR-transduced T cells, compared to 0.04% and 0.04% in the CAR − T cells isolated from the same mice (Fig. 4e,f).As expected, no modified alleles were observed in cells isolated from mice treated with the T cell-targeted lentivirus.We repeated this experiment with mice humanized with PBMCs from a different donor and with more mice per treatment group, and we again observed CAR T cells generated in vivo in eight out of eight mice treated with T cell-targeted Cas9-EDVs and eight out of eight mice treated with T cell-targeted lentivirus (~0.5% vs ~5% CAR + T cells, respectively) (Fig. 4g and Supplementary Fig. 6d-f).Again, we observed genome editing only in mice (n = 4 out of 8) treated with Cas9-EDVs, with higher levels of genome editing in CAR-transduced T cells than in CAR − T cells (Fig. 4h, i).
Treatment with the T cell-targeted Cas9-EDV and lentivirus was well tolerated, with no weight loss observed (Supplementary Fig. 6g).
Although mCherry + F4/80 + Kupffer cells/macrophages were observed, no mCherry + β-catenin-expressing hepatocytes were detected in the liver (Supplementary Fig. 7a-d).Together, these results indicate that antibody-based targeting of Cas9-EDVs is a strategy that maintains cell-selective and tissue-specific delivery of transgenes and genome editors in vivo.The primary objective of our humanized mouse experiments was to assess Cas9-EDVs for their ability to mediate cell-targeted genome editing and transgene delivery in vivo.Given that human CD19 + B cells, in addition to T cells, engrafted in the second mouse cohort, we additionally assessed in vivo CAR T cell killing activity.Variable levels of CD19 + B cells were observed in Cas9-EDV-treated mice, and no CD19 + B cells were detected in mice treated with antibody-targeted lentivirus, demonstrating in vivo CAR T cell-mediated cytotoxicity (Fig. 5a and Supplementary Fig. 8a).This analysis suggests a model in which antibody-derived targeting molecules can direct molecular cargo to specific cells in vivo to successfully reprogram cell activity (Fig. 5b).Diverse T cell clonotypes were observed for CAR-transduced T cells isolated from mice in both groups (Fig. 5c), suggesting that multiple cells were engineered in vivo and did not arise solely through expansion of a single engineered cell.Given that clonotype diversity correlated with the number of CAR T cells analyzed (Supplementary Fig. 8b), the clearance of B cells in the lentiviral group was probably attributable to a higher number of CAR T cells generated during the initial in vivo transduction.Taken together, these findings offer an approach for generating genome-engineered cells with complex edits that could prove valuable for a wide range of clinical applications in the future.

Discussion
The EDVs created here combine the molecular packaging and cellular transduction capabilities of a lentivirus with the cell-surface recognition properties of antibodies to deliver Cas9 protein, sgRNAs and transgenes into specific human cell types, both ex vivo and in vivo.Co-display of scFv antibody fragments and the VSVGmut fusogen on the Cas9-EDV envelope provides selectivity of cell transduction.Antibody-directed Cas9-EDVs mediate genome editing in targeted human cells over bystander, non-target cells in vitro and in humanized mice without transducing hepatocytes, thus avoiding a common barrier to selective in vivo delivery owing to passive liver uptake.
Cas9-EDVs enable complex cell engineering in specific cells, as shown by in vivo generation of gene-edited human CAR T cells, with important advantages relative to other in vivo delivery methods.First, unlike VLP-mediated delivery 8 , EDVs can be administered systemically for cell-type specific receptor-mediated delivery of multiple cargo molecules, including protein, RNA and DNA.Second, in contrast to viral vector-based methods for delivering DNA-encoded molecules [40][41][42] , EDVs provide transient delivery of preassembled genome editors whose short lifetime limits off-target editing.In addition, both adeno-associated virus and lentiviral delivery can involve random transgene integration 43,44 that could be avoided in the future using Cas9 RNP-mediated genome editing for targeted transgene knock-in.Third, distinct from reported viral vectors 45 and lipid nanoparticles 40 , antibody-targeted EDVs do not induce detectable transduction of liver hepatocytes, which could help avoid toxicity by minimizing the effective concentration necessary for therapeutic benefit.Finally, in contrast to previous in vivo CAR T cell generation reports using retargeted retroviruses 46,47 , there is no need for T cell activation before vector  No treatment-related toxicity was observed following T cell-targeted vector administration in vivo; however, it will be important for future studies to assess the impact of vector dose on triggering aberrant T cell activation and proliferation.Clonotype analysis of in-vivo-generated CAR T cells indicated that multiple independent transduction events occurred in vivo.The incomplete B cell aplasia observed for the T cell-targeted Cas9-EDV dose tested here is probably explained by the number of Cas9-EDV-generated CAR T cells being below a threshold needed for B cell ablation in the first 10 days post systemic administration.At ten days post systemic administration, we observed that 5% of all T cells were CAR + in mice that were administered the T cell-targeted lentivirus compared to 0.5% in mice administered the T cell-targeted Cas9-EDV.Although this does not tell us about initial in vivo transduction rates, which could be occluded because of CAR T cell expansion, it does suggest that 5% CAR T cells at 10 days correlates with B cell aplasia in this humanized mouse model.Future experiments will assess B cell ablation either at later time points post Cas9-EDV administration to allow more time for cellular expansion or test higher doses of Cas9-EDVs to generate more CAR T cells initially.To test higher doses, future work will investigate methods for boosting the yield of antibody-targeted Cas9-EDVs, as the current process is less effective than VSVG-pseudotyped Cas9-EDV production.Additionally, it is likely that the packaging of Cas9 RNPs alongside a transgene inhibits successful gene transduction by Cas9-EDVs.Future work will also focus on optimizing the co-delivery of Cas9 RNP complexes and transgenes to improve the efficiency of Cas9-EDV gene delivery in vivo.
Two aspects of Cas9-EDV composition and cell targeting efficiency were unexpected and warrant further analysis.First, the capsid is not needed and potentially inhibits Cas9-mediated genome editing, showing that EDV-delivered Cas9 RNP complexes with nuclear localization tags are sufficient to promote nuclear access.Limiting viral structural components to those essential for EDV production may improve the per-particle delivery efficiency of Cas9-EDVs.Second, the finding that not all scFv-based targeting molecules result in equivalent levels of genome editing in target cells suggests that productive EDV delivery requires more than antigen binding.For example, CD45 does not undergo internalization upon antibody binding, which may explain the minimal editing achieved by CD45-scFv Cas9-EDVs (refs.35,48).In addition, differences in scFv delivery may result from suboptimal targeting molecule display on Cas9-EDVs.
The results reported here merge the single-treatment potential of genome engineering with cell-specific delivery of preassembled genome editors to provide an approach to selective cell editing ex vivo and in vivo.Although this report focuses on the engineering of human immune cells (T cells), future work will be extended to non-immune cells, with a particular focus on the targeted engineering of tissue-resident stem cells in vivo.These findings also shed light on fundamental aspects of fusion-based cargo delivery that may be further uncovered through the investigation of EDV-based molecular trafficking, offering the potential to use EDVs for fundamental research as well as therapeutic delivery applications.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41587-023-02085-z.
Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material.If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.To view a copy of this license, visit http://creativecommons. org/licenses/by/4.0/.© The Author(s) 2024 https://doi.org/10.1038/s41587-023-02085-z

Plasmid construction
VSVGmut (K47A R354A VSVG) sequence was human codon-optimized and synthesized as a gBlock (Integrated DNA Technologies, IDT) and then cloned into the pCAGGS expression plasmid.To generate the CD19 scFv-1 expression plasmid, the sequence encoding the CD8a signal peptide, myc epitope tag, scFv and CD8a stalk and transmembrane domain of a-CD19-4-1BBζ-P2A-mCherry 22,49,50 was subcloned into pCAGGS.This plasmid was subsequently used as an entry plasmid for cloning all other scFv antibody fragments; the CD8a signal peptide, myc tag and scFv sequences were dropped out by EcoRI/Esp3I restriction digest (New England Biolabs, NEB) and new DNA sequences encoding CD8a signal peptide and scFv were inserted.This cloning strategy resulted in removing the N-terminal myc epitope tag and adding a serine amino acid residue between the scFv and CD8a hinge domains.A flexible linker (GGGGSGGGGSGGGGSS) was used to link VH and VL domains of source monoclonal antibody sequences.If the antibody source sequence was already an scFv, then the linker from the source sequence was used.Except for CD19 scFv-1, all antibody fragment sequences were human codon-optimized and synthesized as eBlock Gene Fragments (IDT).Lastly, a CD19 scFv expression plasmid with 2× strep-tag was generated by removing the myc tag from CD19 scFv-1 and inserting the 2× strep-tag.InFusion cloning (Takara Bio) was used to generate all plasmids.Additional information on the scFv targeting molecules and sequence sources can be found in Supplementary Table 1.
Sequences for EDV production plasmids are in Supplementary Table 2.

Cas9 RNP electroporation
B2M-targeting crRNA (IDT) and tracrRNA (IDT, no.1072534) were resuspended in IDT duplex buffer to 160 μM, combined at a ratio of 1:1 and annealed at 37 °C for 30 min.Cas9 RNPs were formed by combining the annealed crRNA and tracrRNA and 40 μM Cas9-NLS (UC Berkeley QB3 MacroLab) at a molar ratio of 2:1 and incubating at 37 °C for 15 min.Electroporation was performed using a 96-well format 4D-nucleofector (Lonza) with 200,000 cells per well.HEK293T cells were electroporated with the SF buffer and the CM-130 pulse code, and primary human T cells were electroporated with the P3 buffer and the EH-115 pulse code.Cells were immediately resuspended in pre-warmed media, incubated for 20 min and transferred to culture plates.

Cas9-EDV titer quantification
The QuickTiter Lentivirus Titer Kit (Lentivirus-Associated HIV p24) (Cell Biolabs) was used to quantify Cas9-EDV particle number.Cas9-EDVs were diluted 1:1,00-100,000, and ELISA was performed according to the manufacturer's directions.Absorbance at 450 nm was measured by a plate reader (BioTek).Cas9-EDV p24 content was calculated by comparison to serial dilution of a p24 standard, and guidance from the manufacturer was used to convert p24 quantity into particle number (Cell Biolabs).

Western blot analysis
Cas9-EDVs were mixed with Laemmli buffer containing 10% 2-mercaptoethanol and heating at 90 °C for 5 min.Proteins from whole cell lysates were separated by 4%-20% SDS-PAGE gel (Bio-Rad) and transferred to an Immun-Blot low fluorescence PVDF membrane (Bio-Rad).Membranes were blocked and incubated with primary antibodies at 4 °C overnight followed by secondary antibodies at room temperature for 1 h.Primary and secondary antibodies are listed in Supplementary Table 3. Imaging was performed using the Odyssey imaging system (LI-COR).

GS-CA1 experiments
To generate lentiviruses encoding EF1a-mNeonGreen, 3.5-4 million Lenti-X cells were plated in a 10 cm tissue culture dish (Corning) and transfected with 1 μg pCMV-VSV-G (Addgene plasmid no.8454), 10 μg psPax2 (Addgene plasmid no.12260) and 2.5 μg of EF1a-mNeonGreen lentiviral transfer plasmid using polyethylenimine (Polysciences) at a 3:1 PEI:plasmid ratio.Lentiviral-containing supernatants were collected 2 days post transfection and passed through a 0.45 μm PES syringe filter (VWR).Lentiviral supernatants were concentrated 20× with Lenti-X Concentrator (Takara Bio) according to the manufacturer's instructions, resuspended in Opti-MEM (Gibco), aliquoted and frozen at −80 °C for future use.B2M-targeted VSVG Cas9-EDVs were produced as described above.The mNeonGreen lentivirus stock was pre-titered on HEK293T cells; Cas9-EDV and mNeonGreen lentivirus samples were diluted in Opti-MEM in a twofold dilution series.Aliquots of 50 μl of each dilution series were mixed with 15,000 HEK293T cells in 50 μl cDMEM in triplicate in a 96-well plate.For the lentiviral sample, the percentage of mNeonGreen + cells was assessed by flow cytometry 3 days post transduction.Wells in which the percent of mNeonGreen + cells was ≤25% were used to calculate the transducing units per ml.The genome editing activity of the Cas9-EDV stock was pre-titered similarly, except that B2M expression was assessed by flow cytometry at 3 days post treatment to calculate Cas9-EDV volume that resulted in approximately 50% of cells negative for B2M expression.

Humanized mouse experiments
All animal studies and procedures were conducted in accordance with the established National Institutes of Health guidelines for animal care and use and were approved by the UC Berkeley Animal Care and Use Committee.All experimental and control animals were housed under the same conditions as approved by the Berkeley Office of Laboratory Animal Care.Mice were housed at ambient room temperature in a humidity-controlled animal facility with free access to water and food.Mice were maintained on a 12:12 h light:dark cycle (lights on from 07:00 to 19:00 h).
Human peripheral blood mononuclear cell-engrafted NSG mice (745557, 6-8 weeks of age) were purchased from Jackson Laboratory.'Experiment 1' and 'experiment 2' mouse cohorts were engrafted with cells from unique human donors.Following anesthesia induction with 2-3% isofluorane, 100 μl of Cas9-EDVs, lentivirus or PBS was administered by retro-orbital injection.Mice were euthanized with CO 2 10 days post treatment and the spleen and liver were dissected.See 'Immunofluorescent staining and imaging' for downstream liver sample processing.Single-cell suspensions of spleen were prepared by gently bursting the organ in 4 ml DMEM (Corning) in a well of a 6-well dish using the back of a 3 ml syringe (Thermo Fisher).Splenocytes were passed through a 100 μm cell mesh (Corning), brought up to 25 ml with DPBS (Gibco) and spun at 300×g for 10 min.Erythrocytes were lysed by resuspending splenocytes in 5 ml 1× BD Pharm Lyse lysing solution (BD Biosciences) for 5 min before being brought up to 25 ml with DPBS.Cells were then pelleted at 300×g for 10 min, resuspended in 10 ml DPBS and counted using a Countess 3 automated cell counter (Thermo Fisher).Cells were cryopreserved in freeze media (Bambanker) for downstream cell sorting.For flow cytometry analysis, CD45 + cells were isolated from splenocytes using the EasySep Release Human CD45 Positive Selection Kit for Humanized Mouse Samples (StemCell Technologies) on an EasySep EasyEights Magnet (StemCell Technologies) according to the manufacturer's instructions.Immunophenotyping was performed using anti-human CD4-FITC, anti-human CD4-PE-Cyanine7, anti-human CD8-PE-Cyanine7 and anti-human CD19-FITC (Supplementary Table 3), and cells were assessed for CAR-2A-mCherry expression using an Attune NxT flow cytometer with 96-well autosampler (Thermo Fisher).For cell sorting, cryopreserved splenocytes were thawed and stained with anti-human CD4-FITC and anti-human CD8-FITC (Supplementary Table 3) in PBS containing 1% BSA, and an SH800S cell sorter (Sony Biotechnology) was used to sort mCherry + and mCherry − primary human T cells that were either expressing CD4 or CD8.

TCR sequencing
Splenocytes from three Cas9-EDV-treated mice and three lentivirustreated mice in humanized mouse experiment 2 were sorted into mCherry + and mCherry − primary human T cells using an SH800S cell sorter (Sony Biotechnology).RNA was extracted from sorted cells using the RNeasy Mini Kit (Qiagen) according to the manufacturer's instructions.TCR a/b libraries were prepared from each sample using the SMART-Seq Human TCR (with UMIs) kit (Takara Bio) according to the manufacturer's instructions.Samples were pooled and sequenced on an Illumina MiSeq using 2×300 bp paired-end chemistry and MiSeq Reagent Kit v.3 (MS-102-3003, Illumina), yielding an average sequencing depth of 1.75 million reads.FASTQ files were analyzed with Cogent NGS Immune Profiler Software (v.1.5),using a UMI cutoff of 3. Clonotypes were visualized using Cogent NGS Immune Viewer (v.1.0)and a custom Python script.

Immunofluorescent staining and imaging
Humanized mice were killed with CO 2 and livers were dissected.Livers were fixed in 4% paraformaldehyde (Electron Microscopy Sciences) for 24 h and transferred to 30% sucrose (Fisher Chemical).After 3 days, livers were embedded into cryoblocks using the Tissue Plus O.C.T. Compound (Fisher HealthCare).Liver sections (20 μm) were cut with the Leica CM 3050 S Cryostat, placed on Superfrost Plus microscope slides (Fisher Scientific) and stored at −80 °C.
Sections were blocked with a buffer including 5% normal goat serum (Sigma-Aldrich), 2% BSA (Sigma-Aldrich), 0.03% Triton X-100 (Fisher Scientific) and 0.05% sodium azide (Sigma-Aldrich) for 1 h at room temperature.Sections were stained for 2 h at room temperature with primary antibodies as follows: mouse anti-β-catenin and rat anti-F4/80; then mouse anti-β-catenin and rabbit anti-hCD3 (Supplementary Table 3).Tissue sections were washed three times with 1× PBS and stained with secondary antibodies for 1 h (Supplementary Table 3).Tissue sections were washed again three times with 1× PBS and treated with DAPI (Sigma-Aldrich) for 10 min.Then,

Fig. 1 |
Fig.1| Cell-specific genome editing with antibody-targeted Cas9-EDVs.a, Schematic scFv targeting molecules (blue) and VSVGmut (orange) on the exterior surface of a Cas9-EDV.Cas9-EDVs package pre-formed Cas9-sgRNA complexes to avoid genetically encoding genome editors within a viral genome.b, Experimental outline and schematic of the lentiviral vector used for engineering HEK293T EGFP cells that express heterologous ligands on the plasma membrane (for example, CD19).To promote cellular engineering by single lentiviral integration events, engineered cell mixtures were generated through low multiplicity of infection to achieve <25% EGFP + cells.Engineered cell mixtures were challenged with B2M-targeting Cas9-EDVs to test targeting

Fig. 2 |
Fig. 2 | Optimization of Cas9-EDVs for enhanced genome editing activity in primary human cells.a, Genome editing activity comparison of CD19 antibodytargeted Cas9-EDV variants packaging B2M-targeted Cas9 RNPs.Expression of B2M protein was assessed by flow cytometry 7 days post treatment in CD19expressing target cells.b, Diagram of the optimized Gag-Cas9 and Gag-pol Cas9-EDV production plasmids; features updated from a previous study22 are highlighted in teal.c-e, Genome editing activity of optimized VSVGpseudotyped Cas9-EDVs in primary human CD34 + cells (c) and activated (d) and resting primary human T cells (e).B2M or TRAC genome editing was assessed by amplicon sequencing 7 days post treatment; n = 3 technical replicates were assessed for all conditions except for the untreated resting human T cells (n = 2).f, Schematic of potential intra-particle Cas9-EDV configurations for packaged

Fig. 3 |
Fig. 3 | Multiplexed antibody targeting and editing of primary human T cells.a,b, Treating resting human T cells with Cas9-EDVs co-displaying CD3 and CD28 scFvs results in cellular activation (a) and proliferation (b) as measured by flow cytometry detection of CD25 3 days post treatment and fold expansion relative to the untreated T cell count, respectively.CD25 expression and cellular proliferation was observed for CD3/CD28 scFv Cas9-EDVs, regardless of whether they packaged Cas9 RNPs targeting PDCD1 or a non-targeting control.c, Genome editing 3 days post treatment, as detected by amplicon next-generation sequencing.For a-c, Cas9-EDVs were concentrated 62× and 50 μl was used to treat 30,000 resting T cells.CD3 scFv-1 and CD28 scFv-2 were tested.d, Screening

Fig. 4 |
Fig. 4 | Programmable human cell delivery generates gene-edited CAR T cells in vivo.a, Summary of T cell-targeted Cas9-EDVs and lentivirus tested in PBMChumanized mice.Both particles display multiplexed scFvs (CD3 scFv-3, CD4 scFv-1 and CD28 scFv-2).The Cas9-EDV vector co-packages a lentiviral-encoded CAR-2A-mCherry transgene and Cas9 RNP complexes to disrupt the TRAC gene; the lentivirus encodes the CAR-2A-mCherry transgene.b, Experimental schematic for testing T cell-targeted Cas9-EDVs and lentivirus in PBMC-humanized mice by intravenous (I.V.) retro-orbital injections.c, Representative flow cytometry plots demonstrating that CAR-expressing human T cells are detectable in the spleens of PBMC-humanized mice 10 days post administration of 1.5 × 10 9 Cas9-EDV (n = 2 animals) or lentivirus (n = 3 animals) but not in mice administered PBS (n = 3), quantified in d. e,f, Gene editing is observed in CAR − and CAR + human T cells isolated from mice treated with T cell-targeted Cas9-EDVs (n = 2 animals)and T cell-targeted lentivirus (n = 3 animals).One CAR + lentivirus sample was excluded in f because of failing sequencing.g, CAR-expressing human T cells are detectable in the spleens of PBMC-humanized mice 10 days post administration of 6.2 × 10 8 Cas9-EDV (n = 8 animals) or lentivirus (n = 8 animals) but not in mice administered PBS (n = 4 animals).P values calculated by means of Dunnett's multiple comparison test after Brown-Forsythe and Welsh one-way ANOVA.h,i, Genome editing is observed in CAR − and CAR + human T cells isolated from mice treated with T cell-targeted Cas9-EDVs (n = 8 animals per group).Significance calculated by two-sided unpaired t-test.Comparison in i is not significant (P > 0.05).For all plots, black lines indicate the median of the data set.LOD, limit of detection as defined by the average modified reads from lentiviral-treated samples.

Fig. 5 |
Fig. 5 | Functional dynamics of cellular engineering in vivo.a, Depletion of CD19 + B cells is observed post administration of T cell-targeted lentivirus (experiment 2).Human CD45 + cells were isolated from PBMC-humanized spleens 10 days post systemic administration of T cell-targeted Cas9-EDV (n = 8 animals), lentivirus (n = 8 animals) or PBS (n = 4 animals), and the percentage of CD19expressing cells was assessed by flow cytometry.P values calculated by means of Dunnett's multiple comparison test after ordinary one-way ANOVA.**P < 0.01.