INTRODUCTION
Ex vivo chimeric antigen receptor T (CAR-T) cell therapy has shown remarkable success in treating both malignant and autoactivated B cells
1. This therapy has led to high rates of long-lasting remission in patients with relapsed or refractory cancers or autoimmune diseases who previously had limited treatment options. Despite its success, challenges remain for the current FDA-approved autologous CAR-T therapy, which is personalized and requires the collection of a patient’s T cells, genetic modifications to express CARs, and lymphodepletion prior to CAR-T cell infusion
2. Off-the-shelf CAR-T therapy, also called allogeneic CAR-T, uses T cells from donors or derived from iPSC, providing immediate availability of the CAR-T cell product
1. However, issues of allogeneic CAR-T include life-threatening graft-versus-host diseases (GvHD) and rapid elimination by the host immune system
3.
In vivo CAR-T cell therapy involves delivering the CAR sequence directly into the patient’s T cells within their body. CAR sequence is packed by viral or non-viral delivery platforms, which is also “off-the-shelf” while bypassing the GvHD of allogeneic CAR-T cell therapy. In this context,
in vivo CAR-T is no longer a personalized cell therapy but a ready-to-use gene therapy. Thus, it faces the same challenge as gene therapy: efficiently and specifically delivering the CAR sequence to T cells. Several delivery platforms have been used to develop
in vivo CAR-T therapy. Lipid nanoparticles (LNP) packing mRNA-encoding CAR sequence have been conjugated with antibodies, including anti-CD3, CD8, and CD5, to achieve redirection to T cells rather than the liver
in vivo4-9. Although the COVID-19 vaccine has proved the safety and scalability of this platform, antibody-conjugated LNPs face challenges in large-scale manufacturing and unclear long-term clinical safety following intravenous administration. Lentivirus (LV) with an engineered envelope with single-chain variable fragments (scFv) against CD3, CD4, CD7, CD8, or TCR could also be redirected to T cells
in vivo10-16. LV integrates the CAR sequence into the host cell genome, allowing stable transgene expression through T-cell expansion while also raising safety concerns on sequence integration into the genome of bystander cells and insertional mutagenesis. Virus-like particles (VLPs) resemble viruses but lack viral genetic material, making them non-infectious
17,18. VLP allows
in vivo delivery of Ribonucleoprotein (RNP) instead of DNA. A recent study engineered VLP by pairing the display of a mutated VSVG with anti-CD3, CD4, or CD28 scFv on Cas9-EDVs (enveloped delivery vehicles) to target T cells
in vivo and deliver genome editor and CAR sequences
19. Although the concept is well-established, the less efficient generation of CAR-T cells
in vivo and the complicated manufacturing process limit their applications in clinics.
Recombinant AAV vectors (AAVs) have been widely used in gene therapy due to their safety profile and ability to induce long-term expression of transgenes in non-dividing cells. Although wild-type (WT) AAVs naturally accumulate in the liver by systemic injection, engineering the variable regions of the capsid protein of AAV has proved to be effective in redirecting AAV to extrahepatic organs or tissues (e.g., the central nervous system and muscle) or increasing transduction efficiencies to non-hepatic cells
20. For example, DART-AAV inserted a designed ankyrin repeat protein (DARPin) targeting murine CD8 into the GH2-GH3 loop of AAV2 capsid protein VP1 and resulted in over 20-fold increase in transduction activity in murine T cells compared to AAV2
10. Ark313 is an AAV6 variant targeting murine T cells, which was engineered through three rounds of directed evolution of a capsid library with 7-mer random insertion in the variable region IV
21. Infusion of Ark313-generated CAR-T cells demonstrated better efficacy in controlling tumor growth than retrovirus-generated CAR-T cells. In a follow-up study, intravenous Ark313 injection successfully transduced murine T cells in vivo
22. However, the question of whether an engineered AAV variant could enable the
in vivo targeting of human T cells, while staying away from the liver and other organs, through systemic delivery, has yet to be explored.
Here, we demonstrated that an engineered AAV6 variant, AAV6-M2, could effectively engineer human T cells both
in vitro and
in vivo.
In vitro, AAV6-M2 successfully delivered the CD19 CAR sequence into both activated and resting human T cells, leading to robust, antigen-dependent cytotoxicity. Further, evidence from whole-genome CRISPR screening, targeted gene knockout, cryo-electron microscopy (cryo-EM) structural analysis, and molecular docking collectively underscores the critical role of CD62L, a cell surface marker of less-differentiated T cells
23, in facilitating AAV6-M2 transduction into resting T cells without prior activation. Systemic administration of AAV6-M2-CAR into Humanized Immune System (HIS) mice resulted in the generation of up to 77.5% of CD8
+ CAR-T cells six weeks post AAV injection. In the HIS mice with systemic lupus erythematosus (SLE), this approach successfully ameliorated lupus pathologies, correlating with the depletion of both circulating and tissue-resident B cells. Notably, systemic injection of AAV6-M2 led to approximately two orders of magnitude lower liver accumulation compared to WT AAV, in both mice and cynomolgus macaques, highlighting a favorable safety profile. Collectively, this study positions AAV6-M2 as a clinically translatable vector for
in vivo CAR-T therapy and broadens the therapeutic scope of AAV-based gene delivery from inherited disorders to autoimmune diseases.
RESULTS
AAV6 variants show superior transduction efficiencies in human primary T cells
To design a capsid library with an improved likelihood of identifying variants with specific tropism for human T cells, we collected functional peptides reported from publicly available databases and literature. For these peptide sequences, we randomly selected fragments with 6–12 amino acids and inserted them between residues A581 and T593 of the AAV6 capsid protein VP3 (variable region VIII or VR VIII)
24, yielding a diverse library comprising ~480,000 AAV6 variants. Following capsid library screening in human primary T cells, we identified variants that exhibited enhanced transduction efficiencies relative to WT AAV6 (AAV6-WT). Among them, two top-performing candidates, designated AAV6-M1 and AAV6-M2, were selected based on their transduction efficiencies and low coefficient of variation in transduction performance across replicates (Supplementary Fig. S1a).
To validate the transduction efficiencies of the leading candidates in human primary T cells, we packaged self-complementary AAVs encoding EGFP using the AAV6-M1 and AAV6-M2 capsids and compared their performance to AAV6-WT (Fig. 1a). Transduction efficiency was quantified by counting the percentage of EGFP+ T cells. Across a range of multiplicities of infection (MOI: 1 × 102, 1 × 103, and 1 × 104), both AAV6-M variants outperformed AAV6-WT (Fig. 1b). This superiority was consistent when assessing both the percentage of EGFP+ cells and the mean fluorescent intensity (MFI) (Fig. 1c, d). Notably, at an MOI of 1 × 103, the AAV6-M1 and AAV6-M2 achieved EGFP+ rates of 91.8% and 80.2%, respectively — matching or exceeding the transduction efficiency observed with AAV6-WT at an MOI one order of magnitude higher (1 × 104). Moreover, the engineered AAV6-M2 exhibited packaging efficiency comparable to that of AAV6-WT, whereas AAV6-M1 showed a reduced packaging yield (Supplementary Fig. S1b). The manufacturability of AAV6-M2 was further verified in a scaled-up 3-liter bioreactor, demonstrating comparable packaging efficiency to AAV6-WT, a dominant full-particle peak, and uniform capsid morphology (Supplementary Fig. S1c–e).
CD62L mediates enhanced transduction of AAV6-M2 in human T cells
AAV6-M variants were assembled using engineered capsid proteins containing mutations between residues A581 and T593 of VP3, where the corresponding peptides are located in the 3-fold protrusion of AAV particles and are positioned to interact directly with host cells. We hypothesized that these variants utilized distinct cell-surface factors to enter human T cells, compared to AAV6-WT. To investigate this, we performed a whole-genome CRISPR knockout screen in Jurkat cells to identify genes contributing to the enhanced transduction efficiencies of AAV6-M variants (Fig. 2a). In brief, Jurkat cells stably expressing SpCas9 (Jurkat-Cas9) were first transduced with the lentiviral CRISPR Brunello library at a low MOI to deliver one sgRNA per cell, followed by transduction with either AAV6-WT-EGFP, AAV6-M1-EGFP, or AAV6-M2-EGFP. We then sorted and collected the top and bottom 20% Jurkat-Cas9 cells according to EGFP expression. Compared to the top 20%, sgRNAs enriched in the bottom 20% of cells represented gene knockouts that disrupted the AAV attachment, internalization, or intracellular trafficking in human T cells.
We analyzed the top 10 genes enriched in each of the three CRISPR screens. All screens identified genes that are known to be essential for AAV transduction and trafficking, including
KIAA0319L (
AAVR) and
TM9SF2, validating the effectiveness of our screening approach. A few genes, such as
NOTCH1, were specifically enriched in the screen using AAV6-M1 relative to AAV6-WT (Supplementary Fig. S2a). However, NOTCH1 is broadly expressed across multiple organs, including endocrine tissues, brain, and liver, making it less favorable for selective
in vivo T-cell targeting. Nevertheless, among the top 10 genes specifically enriched in the AAV6-M2 screen,
SELL stood out (Fig. 2b, c).
SELL encodes a cell surface protein and is almost exclusively expressed in immune cells. Also known as CD62L or L-selectin, SELL is essential for the binding and rolling of lymphocytes on endothelial cells, facilitating their migration into secondary lymphoid organs. Interestingly, a recent study engineered lentiviral envelopes with CD62L-specific scFv to enable CAR delivery into less differentiated human T cells
25.
Indeed, knocking out CD62L (SELL) from Jurkat cells significantly reduced the transduction efficiency of AAV6-M2 but not AAV6-WT (Fig. 2d). The reduction was comparable to that observed upon AAVR knockout, suggesting a critical and specific role of CD62L in mediating AAV6-M2 entry into T cells. Moreover, we used two independent sgRNAs to knock out CD62L in primary human T cells from three donors. Following CD62L knockout, the transduction efficiency of AAV6-M2 was markedly diminished, while AAV6-WT transduction remained largely unaffected (Fig.2e; Supplementary Fig. S2b). Accordingly, overexpression of CD62L in CD62L-negative cells enhanced AAV6-M2 binding (Supplementary Fig. S2c). Collectively, both CRISPR screening, targeted gene knockout, and CD62L overexpression experiments underscore the essential role of CD62L in AAV6-M2 transduction into human T cells — a mechanism that appears unique to this AAV variant and is not shared by AAV6-WT.
To further investigate how the engineered loop may enhance the interaction between AAV6-M2 and CD62L, we resolved the cryo-EM structure of AAV6-M2 at 1.8 Å resolution (PDB: 9VI4; Fig. 2f; Supplementary Fig. S2d). The VP protein monomer was accurately modeled, with both the backbone and side chains in the VR VIII loop clearly resolved (Supplementary Fig. S2e). Structural alignment showed that the backbone conformation of AAV6-M2 was nearly identical to that of AAV6-WT (PDB: 4V86) (Supplementary Fig. S2f); however, the side chain conformations within the VR VIII loop differ substantially (Supplementary Fig. S2g, h). To explore potential interactions with CD62L, we performed molecular docking between AAV6-M2 and human CD62L (PDB: 5VC1; lectin/EGF domains). The engineered loop was predicted to insert into the EGF domain of CD62L, where three negatively-charged residues (E590, E591, and E592) within the VR VIII loop of AAV6-M2 engage a positively-charged surface pocket through electrostatic interactions (Figs. 2g, h). Mutating each one of these glutamic acids individually, or all of them collectively, to alanine abolished the binding affinity between AAV6-M2 and human T cells, further highlighting the critical roles of these residues (Supplementary Fig. S2i). These structural observations suggest that the engineered VR VIII loop appears to contribute to enhancing the engagement between AAV6-M2 and CD62L. The engineered loop, however, did not increase but slightly reduced neutralization by human serum compared to AAV6-WT (Fig. 2i; Supplementary Fig. S2j).
AAV6-M2 transduces resting human T cells in vitro
Considering both the efficiencies of viral packaging and the specificity of the binding partner, we focused on AAV6-M2 in the subsequent studies. We first replaced the transgene with CD19 CAR to evaluate the expression kinetics and cytotoxicity of AAV6-M2-mediated CAR-T cells in vitro (Supplementary Fig. S3a). Following transduction of bead-activated human T cells (aT) with AAV6-M2-CAR, the proportion of CAR+ T cells peaked at day 3 and gradually declined by day 10 (Fig. 3a). We observed comparable T cell viability and proliferation relative to the non-AAV control (Fig. 3b; Supplementary Fig. S3b). When co-cultured with CD19-expressing NALM6 cells, AAV6-M2-transduced T cells exhibited robust cytotoxicity by 20 h post-transduction and achieved complete B cell killing by day 3 (Fig. 3c, d; Supplementary Fig. S3c). In contrast, AAV6-WT-transduced T cells showed minimal cytotoxicity. These results demonstrate that AAV6-M2 effectively delivers CD19 CAR to active human T cells, enabling antigen-specific cytotoxicity in vitro.
Next, we performed site-specific CD19 CAR insertion using two AAV6-M2 vectors to generate stable CAR-expressing T cells, a strategy expected to prolong CAR-T presence during T cell proliferation. The AAV vector-1 provides the CD19 CAR expression cassette, which is integrated into the 1st exon of TRAC, where the AAV vector-2 generates a double-strand break by enAsCas12f-mediated cleavage (Fig. 3e). Following dual AAV transduction, transient CAR+ T cells (TCR+CAR+) emerged rapidly but declined after day 3 (Fig. 3f), coinciding with the onset of robust T-cell expansion (Fig. 3g). In contrast, stable CAR+ T cells (TCR−CAR+) gradually increased over time, expanding in parallel with the proliferating T cell population and exhibiting sustained CAR expression. Together, these results demonstrate that AAV6-M2 enables efficient generation of stable CAR+ T cells through precise genomic integration of the CAR cassette, thereby supporting prolonged CAR-T expression during T-cell proliferation.
However, since the majority of human T cells are in a resting state in vivo, we sought to evaluate the delivery capability of AAV6-M2 in non-activated T cells. In two parallel experiments, non-activated human T cells were transduced with AAV6-M2-EGFP and subsequently either activated or left unstimulated (Supplementary Fig. S3d, e). We then monitored the EGFP expression over time. Interestingly, in the absence of activation, the percentage of EGFP+ cells gradually increased, reaching ~40% by day 15 (Fig. 3h). In contrast, post-transduction activation rapidly elevated the percentage of EGFP+ cells to ~80% within three days, followed by a decline to levels similar to those observed in non-activated cells by the endpoint of the experiment (Fig. 3i). Adding AAV6-M2-EGFP into non-activated human peripheral blood mononuclear cells (PBMC) also showed successful transduction in both CD4+ and CD8+ T cells (Supplementary Fig. S3f). These results indicate that AAV6-M2 is capable of delivering a transgene into human T cells without the need for prior or subsequent activation.
To investigate the antigen-specific cytotoxic potential of core circulating T cell subsets following AAV6-M2-CAR transduction, we co-cultured naïve (Tn), central memory (TCM), and effector memory (TEM) T cells with NALM6 cells (Fig. 3j). In the absence of bead-based activation, Tn cells effectively eliminated NALM6 cells between days 6 and 9 post-co-culture, with some variability observed among donors (Fig. 3k). The cytotoxic activities of TCM and TEM were more variable, with two donors exhibiting complete NALM6 clearance between days 6 and 14 (Supplementary Fig. S3g, h). Upon antigen stimulation, both Tn and TCM underwent phenotypic conversion into TEM (Supplementary Fig. S3i–k). Collectively, AAV6-M2 enables effective CAR delivery to multiple resting T cell subsets and supports their differentiation into functional effectors capable of eliminating target cells in an antigen-dependent manner.
AAV6-M2 mediates in vivo CAR-T cell generation and B cell depletion
The AAV-mediated CAR-T cell generation encouraged us to explore the in vivo application of AAV6-M2. We are curious whether the superior transduction efficiency of AAV6-M2 to human T cells could enable the generation of CAR-T cells in vivo through systemic administration to bypass the ex vivo manufacturing process of the conventional CAR-T cell therapy. We employed an optimized HIS mouse model (NCG-X-hIL15), in which human hematopoietic stem cells (hHSCs) were engrafted into immunodeficient mice and various human immune subsets were developed (Fig. 4a). Following engraftment, the HIS mice with comparable levels of human T and B cells were randomly grouped (Supplementary Fig. S4a, b) and systemically injected with AAV6-WT or AAV6-M2 at a dose of 3 × 1013 vg/kg and sacrificed after 28 or 42 days. The same CD19 CAR-EGFP construct used in the in vitro assay was applied in the in vivo experiment.
In peripheral blood, CAR+ T cells emerged by week 2 following AAV6-M2-CAR injection, whereas very few CAR+ T cells were detected in the AAV6-WT-CAR group (Fig. 4b). These CAR+ cells were predominantly CD8+, with minimal CD4+CAR+ T cells detected (Fig. 4c). Concurrently, human B cells, the targeted cells of these CAR+ T cells, decreased over time. The AAV6-M2-CAR group showed a notable decline from baseline to week 4, with B cell counts in most mice approaching the detection threshold (Supplementary Fig. S4c, d). While B cell levels were variable at baseline and across individual mice, the overall trend indicated a greater extent of reduction in the AAV6-M2-CAR group by the study endpoint, compared to the AAV6-WT-CAR group.
In bone marrow, spleen, liver, and lung, the human B cells were largely undetectable in AAV6-M2-treated mice, whereas significantly more B cells remained in mice treated with AAV6-WT (Fig. 4d). Correspondingly, CAR+ T cells were readily detected in these organs in the AAV6-M2-CAR group, but were largely absent in the AAV6-WT-CAR group (Fig. 4e). Mirroring the observation from the peripheral blood, CAR+ T cells were predominantly CD8+, with up to 77.5% CAR+CD8+ present in the bone marrow at week 6 (Fig. 4f; Supplementary Fig. S4e). Collectively, AAV6-M2 successfully delivered CD19 CAR into the human T cells in vivo, which could be detected in the peripheral blood two weeks post-AAV injection, and effectively depleted both circulating and tissue-resident B cells.
To assess the differentiation state within CAR+CD8+ T cells in vivo, we analyzed Tn, TCM, TEM cells, and terminally differentiated effector memory re-expressing CD45RA (TEMRA) subsets across tissues. CAR+CD8+ T cells were predominantly of TEM phenotype, while Tn and TCM subsets were minimally detected in these tissues (Fig. 4g; Supplementary Fig. S4f–i). Given our in vitro data showing that Tn and TCM cells adopt a TEM phenotype following AAV-CAR transduction and antigen encounter (Supplementary Fig. S3i–k), the tissue-resident TEM cells observed in vivo are likely derived from these early-stage subsets and reflect the active killing of human B cells. Together, these findings suggest that the effective B cell depletion in the AAV6-M2-CAR group is accompanied by a dominant presence of CAR+CD8+ TEM cells.
Next, we examined the transduction specificity of AAV6-M2 in vivo. Among various immune cell types in tissues and blood, CD8+ T cells were preferentially converted to CAR-T cells compared to CD4+ T cells. CAR+ cells were barely found in B cells, monocytes, and NK cells (Fig. 4h). As systemically injected AAV tends to accumulate in the liver, we also examined the biodistribution of the AAV vector. Compared to the AAV6-WT, the engineered AAV6-M2 shows significant de-targeting from the liver and reduced viral accumulation across all organs (Fig. 4i). Furthermore, in cynomolgus macaques, AAV6-M2 showed minimal biodistribution to peripheral tissues, with very low levels of viral genomes detected, including in the liver (Fig. 4j). Together, these findings demonstrate that AAV6-M2 selectively targets human T cells while minimizing off-target liver transduction in both humanized mice and non-human primates, supporting its favorable safety profile for in vivo T cell engineering.
AAV6-M2 ameliorated lupus pathologies in SLE HIS mice via B cell depletion
To further evaluate the efficacy of in vivo generated CAR-T cells in a disease context, we employed a SLE HIS mouse model induced by topical application of R848, a TLR7/8 agonist (Fig. 5a). SLE HIS mice were randomly assigned to SLE and SLE+AAV groups, with comparable levels of human CD45 cells, human B cells, IgG, and anti-dsDNA antibody (Supplementary Fig. S5a–d). Following systemic administration of AAV6-M2-CAR at week 0, we monitored the B-cell levels in peripheral blood from week 2 to week 8. A rapid and sustained decline in circulating B cells was observed in the SLE+AAV group beginning at week 2, with significantly lower B cell levels compared to the untreated SLE group (Fig. 5b).
To evaluate the tissue-wide impact of CAR-T activity beyond peripheral blood, we quantified human B cells across multiple organs at the study endpoint. Flow cytometry revealed a significant reduction in the percentages of human B cells in the spleen, bone marrow, liver, lung, and kidney of the SLE+AAV group compared to the untreated SLE group (Supplementary Fig. S5e). Although the percentages of B cell reduction in the bone marrow did not reach statistical significance due to an outlier, a clear downward trend was observed, with significantly lower B cell counts detected in the AAV6-M2-treated animals (Fig. 5c). Subtype analysis further demonstrated that the depletion encompassed all major B cell subsets, including transitional, naïve, memory B cells, and plasmablast cells (Supplementary Fig. S5f). Notably, transitional B cells, which represent the majority of developing human B cells in bone marrow, were nearly undetectable following AAV6-M2-CAR administration (Supplementary Fig. S5g). Naïve and memory B cells and plasmablast cells were also significantly reduced in the bone marrow (Supplementary Fig. S5h–j). In the spleen, we found a similar B cell count depletion (Fig. 5d) and a marked reduction of splenic plasmablast and plasma cells (PB&PC), as evidenced by a significant drop in CD38+CD138+ populations (Supplementary Fig. S5k, l), which are the primary source of autoantibody production in SLE and contribute directly to immune complex deposition and tissue inflammation. Together, bone marrow and spleen analysis inform the direct on-target activity of CAR-T therapy, indicating systemic suppression of B cell development and differentiation.
Lupus nephritis is a major SLE complication associated with progression to renal failure and an overall worse prognosis. In the kidney, the reduction of glomerular IgG deposition in the SLE+AAV group indicated a marked alleviation of immune complex-mediated damage and served as a critical readout for therapeutic efficacy (Fig. 5e; Supplementary Fig. S5m). Consistent with this, circulating levels of anti-dsDNA and anti-Smith (Sm) antibodies significantly declined following AAV treatment (Fig. 5f, g), and serum blood urea nitrogen (BUN) and creatinine (CREA) levels also significantly reduced, indicating improved renal function (Fig. 5h, i). In the lung, histological analysis by hematoxylin and eosin (H&E) staining showed preserved alveolar architecture and reduced inflammatory infiltration in the SLE+AAV group, in contrast to the mild to moderate perivascular and interstitial inflammatory infiltrates observed in the SLE group (Fig. 5j; Supplementary Fig. S5o). Flow cytometry confirmed the depletion of human B cells in the lung, consistent with the observed histological improvement (Supplementary Fig. S5n). In the liver, mild immune cell infiltration was present in the SLE group but completely absent in the SLE+AAV group, aligning with the efficient clearance of tissue-resident B cells (Supplementary Fig. S5p, q). Together, these data demonstrate that administering AAV6-M2-CAR not only eliminates systemic and tissue-resident B cells but also alleviates multi-organ inflammation and protects organ function in a humanized SLE model.
DISCUSSION
This study demonstrates that a clinically validated delivery platform, the recombinant AAV vector, can be engineered to efficiently and specifically deliver transgenes into human T cells both in vitro and in vivo. We showed that AAV6-M2, an engineered AAV6 variant, successfully generates both transient and stable CAR-expressing T cells in vitro. Through a combination of CRISPR screening, cryo-EM, and molecular docking, we identified CD62L, a surface marker expressed on less-differentiated T cells, as a key facilitator of AAV6-M2, but not AAV6-WT, entry into human T cells. This unique tropism enables AAV6-M2 to deliver CD19 CAR into resting human T cells and trigger antigen-dependent cytotoxicity upon encountering CD19+ NALM6 cells. In vivo, systemic administration of AAV6-M2-CAR into HIS mice resulted in up to 77.5% of CAR+CD8+ T cells six weeks post-injection. In HIS SLE mice, human B cells were effectively depleted in both the peripheral blood and organs, accompanied by improvements in lupus-associated pathologies. Notably, AAV6-M2 exhibited striking liver detargeting, with viral genome accumulation in the liver reduced by more than two orders of magnitude in both mouse and cynomolgus macaque, compared to the WT AAV6 and AAV9.
While LV, LNP, and VLP have been explored for in vivo T cell engineering, this study is, to our knowledge, the first to demonstrate that systemic injection of a single AAV vector can generate persistent and functional human CAR-T cells in vivo. Given that AAV is used in eight FDA-approved gene therapies, this finding marks an advance in the application of AAV vector for in vivo T cell engineering. Natural AAVs are known to accumulate in the liver following intravenous injection. Therefore, achieving both liver detargeting and T cell specificity is critical to bringing AAV into the realm of in vivo CAR-T therapy. By changing amino acids in the VR VIII loop, which protrudes to the surface of the AAV viral particle, AAV6-M2 exhibited strong tropism to human T cells while largely escaping from the liver. CRISPR-based receptor screening and validation confirmed that CD62L is required for AAV6-M2 transducing T cells, even without prior activation. The CD62L surface marker decorated naïve and early-memory T cells with higher plasticity and greater ability to proliferate, which are favored by CAR-T cell therapy. A higher viral copy number of AAV6-M2 in CD62L+ T cells than that in CD62L– T cells in the spleen of HIS mouse provides preliminary in vivo evidence of CD62L preference (Supplementary Fig. S6a).
The interaction between the engineered VR VIII loop of AAV6-M2 and the EGF domain of CD62L was further supported by cryo-EM structural analysis and molecular docking. However, this does not explain why AAV6-M2 exhibits reduced liver uptake, where the liver sinusoidal endothelial cells (LSECs) and Kupffer cells are known to non-specifically sequester AAV particles from circulation. Notably, the engineered AAV6-M2 introduces three consecutive glutamic acids (E) to the VR VIII loop. These residues not only facilitate docking into a positively charged pocket in the EGF domain of CD62L, but also alter the capsid's surface electrostatic potential by contributing an additional ~120 negatively-charged amino acids (Supplementary Fig. S6b).
Indeed, the net loss of positive charge for AAV does alter the electrostatic properties of capsids, likely modulating its interactions with cellular receptors, antibodies, and other serum proteins
26. Because circulating AAV particles can access the space of Disse, the increased negative surface charge of AAV6-M2 may reduce nonspecific interactions with the negatively-charged extracellular matrix, thereby potentially contributing to reduced hepatic AAV accumulation. Similar principles have been observed for LNP, where increasing negative surface charge can reduce hepatic sequestration and shift biodistribution toward the spleen, primarily by weakening nonspecific electrostatic interactions in the plasma and liver
27,28. Consistent with this notion, a preliminary comparison of AAV6-M2 with Ark312 and Ark315, two AAV6 variants also engineered for human T cell targeting
29,30, suggests that AAV6-M2 exhibits lower liver accumulation relative to these variants (Supplementary Fig. S6c–e). Together, these observations raise the possibility that an increased number of negative surface charges may contribute to reduced hepatic viral genome levels; however, further investigation will be required to elucidate the mechanisms underlying liver detargeting and to inform future capsid engineering efforts.
Similar to the LNP, LV, and VLP platforms, our AAV-based
in vivo CAR-T system demonstrates proof-of-concept but still requires further optimization through future biotechnological advancements. First, CAR sequences delivered as episomal DNA are subject to dilution as T cells proliferate. Although robust CAR-T cell signals were detected six weeks after AAV injection, site-specific CAR integration into the T cell genome using CRISPR may further extend CAR expression and provide more durable immune surveillance against target cells. Dual-AAV delivery represents a potential strategy to balance the need for stable CAR expression with the risks associated with random integration from lentiviral vectors. The differing kinetics of CAR-T cells generated via LNP, AAV, or lentiviral approaches warrant detailed investigation, as they likely dictate the most appropriate therapeutic indications for each platform. For example, a single dose of our current AAV-based
in vivo CAR-T was sufficient to clear human B cells and ameliorate lupus-like pathology in HIS mice, whereas more persistent CAR-T cells may be required to control tumor progression. Second, leveraging AAV as a delivery platform to treat non-rare diseases, particularly tumors prone to relapse, offers exciting opportunities while also revisiting the long-standing challenge of redosing. LNP has been dosed repeatedly within a short interval (e.g., one week), boosting peripheral CAR-T cell levels
9. Encouragingly, anti-CD40-mAb and CD20×CD3 bispecific antibodies have been used in non-human primates to suppress anti-AAV antibody response and enable repeat administration
31. IgG degrading enzymes (e.g., Imlifidase) have also been actively investigated in preclinical and clinical studies for prophylactic immunomodulation in
in vivo gene therapy, enabling AAV administration in the presence of pre-existing antibodies or allowing for re-administration in non-human primates
32. Whether such strategies can similarly enhance the antigen-dependent cytotoxicity of AAV-delivered CAR-T cells remains a promising direction for future investigation. Finally, the use of T-cell-specific, compact, and potent promoters and enhancers may further enhance selective and robust CAR expression in the desired cell types, thereby increasing the safety margin.
We demonstrate that an engineered AAV6 variant, exhibiting superior transduction efficiency in human T cells and markedly reduced liver targeting, can deliver CAR sequences via systematic administration and generate functional CAR-T cells in vivo. This study establishes a novel in vivo CAR-T platform built upon a clinically validated and widely used delivery vector. Moreover, it broadens the therapeutic potential of AAV-mediated gene transfer beyond inherited disorders, extending into the treatment of acquired and immune-related diseases.
MATERIALS AND METHODS
AAV packaging
An ITR-containing GOI plasmid was utilized for packaging with different AAV capsid plasmids and adenovirus helper plasmid using polyethyleneimine (24765-1, Polysciences). To pack AAV, HEK 293T cells were seeded in 150 mm plates; The GOI, capsid, and helper plasmids were added at a ratio = 1:1:1 (Supplementary Table S1). After 72 h, the transfected HEK 293T cells were collected in SAN digestion buffer and lysed by three rounds of rapid freeze/thawing, followed by a 1 h incubation at 37 °C with 100 units/mL Benzonase (Yeasen, #20156ES60). AAV was further purified following cell harvest and PEG precipitation using iodixanol (StemCell Technologies, OptiPrep, #07820) gradient ultracentrifugation. The purified AAV was treated with DNaseI (Invitrogen, #AM2238) and Proteinase K (Qiagen, #W0013). The titer of AAV was determined by qPCR with ChamQ Universal SYBR qPCR Master Mix (Vazyme, #Q711) via LightCycler®96 (Roche). Relative quantity was estimated compared to a serial dilution of a plasmid standard of known concentration. Primers for AAV titration were listed in Supplementary Table S2.
For large-scale AAV preparation, AAV particles were produced in a 3-liter bioreactor (Duoning) using WayneLVPro™ HEK293 cells (QUACELL) transfected with a three-plasmid system at a 1:1:1 ratio (GOI: capsid: helper plasmid). At 72 h post-transfection, cells were harvested, and the AAV particles were purified by iodixanol gradient ultracentrifugation, followed by concentration and buffer exchange using tangential flow filtration. Final AAV products were resuspended in DPBS and stored at –80 ºC.
The quality of purified AAV was assessed through: (1) Purity analysis by SDS-PAGE (SurePAGE, 4–20%, GenScript) with Coomassie Blue staining (GenScript); (2) Genomic titer quantification via absolute qPCR using ChamQ Universal SYBR qPCR Master Mix (Vazyme Q711) on a LightCycler® 96 System (Roche) was performed and titers were calculated from a standard curve generated by serial dilutions of linearized plasmid DNA containing the transgene expression cassette (concentration verified by Nanodrop™); (3) Ratio of full/empty capsid was determined by analytical ultracentrifugation (Beckman Optima AUC A/I, An-50 Ti rotor, 20 °C, 16,000 rpm).
Five GOIs were used in this study:
scAAV-EGFP: CAG-EGFP
ssAAV-Cas12f-sgRNA-EGFP: U6-sgRNA-CMV-Cas12f-EGFP
ssAAV-HDR-CD19CAR-EGFP: EF-1α-CD19CAR-TagBFP
ssAAV-CD19CAR-EGFP: EF-1α-CD19CAR-EGFP
33scAAV-EGFP-Barcode: CAG-EGFP-barcode
In vitro culture and transduction of primary cells and cell lines
Human PBMCs and primary T cells
Peripheral blood from healthy donors was obtained with informed consent and institutional ethical approval (Westlake University, #20240318MLJ001) from Liquan Hospital (Shanghai, China), Boren Hospital (Beijing, China), and the Affiliated People’s Hospital of Ningbo University (Ningbo, China).
PBMCs were isolated by density gradient centrifugation using Ficoll-Paque. Human primary T cells were subsequently purified from PBMCs using the Pan T Cell Isolation Kit (Miltenyi Biotec, #130-096-535) according to the manufacturer's instructions.
To conduct transduction in activated T cells, human T cells were activated using Enceed™ T cell Activation reagent (L00899, Genescript) for 24h in RPMI 1640 supplemented with 10% FBS and IL-2 at 100 U/mL (Jiangsu Kingsley Pharmaceutical Co., Ltd.). AAV vectors were then added at the indicated MOIs. Cells were maintained in IL-2-containing medium (100 U/mL) throughout the culture period.
To conduct transduction in cells without activation, human PBMCs and resting T cells were cultured in RPMI 1640 medium supplemented with 10% FBS, human IL-7 (5 ng/mL), and IL-15 (100 U/mL). Throughout the culture period, the medium was half changed every three days. To assess transduction efficiency in post-transduction activated T cells, activated human T cells were cultured in RPMI 1640 with 10% FBS and human IL-2 (100 U/mL) following activation. Cells were washed three times thoroughly at designated time points to remove residual AAV.
EGFP or CAR expression was assessed by flow cytometry at designated time points following AAV transduction. CAR expression was detected by Biotin-SP–Goat Anti-Mouse IgG, F(ab’)2 fragment-specific antibody (Jackson ImmunoResearch, #115-065-006), followed by PE-conjugated anti-biotin antibody (BioLegend, #409003).
Jurkat cells
The Jurkat cells were purchased from ATCC (TIB-152) and cultured in RPMI1640 medium with 10% FBS. To generate Jurkat-SpCas9 cells, parental Jurkat cells were transduced with the LentiCas9-Blast vector (Addgene, #52962) and selected with 2 µg/mL blasticidin. Monoclonal SpCas9-expressing lines were established by limited dilution and expanded under continuous 0.5 µg/mL blasticidin selection.
NALM6 cells
The NALM6 cell line was kindly provided by Dr. Xv Li’s laboratory at Westlake University and maintained in RPMI 1640 medium with 10% FBS.
Cytotoxicity assay of CAR-T cells
To evaluate CAR-T cytotoxicity, activated human T cells were transduced with AAV6-WT-CAR or AAV6-M2-CAR at an MOI of 1 × 104. At day 2 post-transduction, NALM6 cells were co-cultured with transduced T cells at a 1:1 ratio (T cell:B cell, T:B). After 24 h of co-culture, cells were harvested and stained with anti-human CD45 and anti-human CD19 antibodies to distinguish NALM6 cells from human T cells. Cytotoxicity was calculated using the following formula:
Cytotoxicity (%) = 100 × (CD19+% in non-AAV control − CD19+% in AAV-transduced cells) / CD19+% in non-AAV control.
To evaluate CAR-T cytotoxicity in resting human T cell subtypes, Central memory T cells (TCM; CCR7+CD45RO+), effector memory T cells (TEM; CCR7−CD45RO+), and naïve T cells (TN; CCR7+CD45RO−) were isolated from pre-cultured human T cells using fluorescence-activated cell sorter (BD FACSAria Fusion) and maintained in RPMI 1640 medium supplemented with 10% FBS, IL-7 (5 ng/mL), and IL-15 (100 U/mL). Each subset was transduced with recombinant AAV (AAV) at an MOI of 1 × 105. At 24 h post-transduction, cells were washed three times with RPMI 1640 and co-cultured with NALM6 cells at a T:B ratio of 3:1. From that point forward, the percentages of NALM6 cell (CD45–CD19+), TCM, TEM, Tn, and TEMRA (CCR7−CD45RO−) populations in each well were assessed by flow cytometry every three days.
Serum blocking assay
Serum samples were collected from 36 healthy donors at Zhejiang Cancer Hospital. All procedures, including serum collection and experimental assays, were conducted under ethical approval from both Zhejiang Cancer Hospital and Westlake University (#20250527MLJ001). PBMC from four donors were used in this experiment.
To assess the inhibition of AAV transduction by pre-existing neutralizing antibodies, serial 10-fold dilutions of human serum (prepared in RPMI 1640 supplemented with 2% FBS, starting at 1:2.5) were prepared in an 8-tube strip and mixed with equal volumes of AAV6-M2-EGFP or AAV6-WT-EGFP (diluted in RPMI 1640 + 2% FBS) in a U-bottom 96-well plate. The virus-serum mixtures were incubated at 37 °C with 5% CO2 for 1 h, resulting in final serum dilutions ranging from 1:5 to 1:5,000 across four serial dilution steps.
After incubation, 5 × 104 human PBMCs, pre-activated via Enceed™ T cell Activation reagent (Genescript, #L00899) and human IL-2 for 24 h, were added to each well. Each well contained a final volume of 200 μL RPMI 1640 with 2% FBS and 100 U/mL IL-2. AAVs were added at an MOI of 5 × 103. Positive control wells contained AAV without serum, and negative control wells contained neither AAV nor serum.
At 48 h post-transduction, cells were stained with anti-human CD45 (BioLegend, #304012) and anti-human CD3 (BioLegend, #317308), and EGFP expression in CD3+CD45+ T cells was analyzed by flow cytometry. The EGFP+ percentage was normalized using the formula:
(Sample EGFP% − Negative Control EGFP%) / (Positive Control EGFP% − Negative Control EGFP%) × 100%.
Donors were grouped based on the highest dilution factor at which the EGFP+ percentage dropped below 50%, and the number of donors in each group was plotted for AAV6-WT and AAV6-M2, respectively.
Site-specific CAR insertion
The first AAV vector packaged sequences that encode CD19 CAR linked to tagBFP, with the two expression cassettes connected by a self-cleaving T2A sequence. The CD19 CAR cassette was flanked by two 400-bp homology arms corresponding to the insertion sites generated by a CRISPR system delivered by the second AAV. The second AAV vector packaged sequences encoding an engineered AsCas12f-HKRA (enAsCas12f-HKRA), a nuclear localization signal, EGFP, T2A, and an sgRNA targeting the first exon of TRAC.
The dual AAV vectors were co-transduced into activated human T cells at an MOI 1 × 105. Transduced cells were cultured in RPMI 1640 medium supplemented with 10% FBS and 100 U/mL IL-2. The percentages of GFP+ and BFP+ cells were monitored by flow cytometry post-transduction. Cells were also stained with anti-TCR α/β (Biolegend, #306727) to assess TCR knockout, and the CAR expression was evaluated by Biotin-SP–Goat Anti-Mouse IgG, F(ab’)2 fragment-specific antibody (Jackson ImmunoResearch, #115-065-006), followed by PE-conjugated anti-biotin antibody (BioLegend, #409003). Stable CAR+ T cells were defined as TCR−CAR+ cells.
Genomic insertion of the CAR sequence was verified by PCR amplifying the flanking regions surrounding the insertion sites.
Genome-wide CRISPR-Cas9 screening
Preparing the LV Brunello library
The transfer plasmid (LentiGuide-Brunello-mKate2), the pMD2.G (Addgene, #12259) envelope plasmid, and the psPAX2 (Addgene, #12260) packaging plasmid were mixed at the mass ratio 5:2:3 and incubated with PEI for 15 min at room temperature. The mixture was dropped to HEK293T cells at 80% confluency. Lentiviral supernatant was collected at 48 and 72 h post-transfection, filtered through a 0.45-μm filter (Millipore, #SLHV033RB), and then concentrated by ultracentrifuging at 70,000× g at 4 °C for 2 h. Finally, the concentrated LV was aliquoted and stored at –80oC.
CRISPR-Cas9 screening in Jurkat-Cas9 cells
LentiGuide-Brunello-mKate2 was transduced in Jurkat with stable Cas9 expression (Jurkat-Cas9). The screening was performed in two Jurkat-Cas9 mono clones. For each replicate, 7 × 107 Cas9-expressing Jurkat cells (~300X coverage) were transduced with the lentiviral library at MOI ≤ 0.3 by infection and incubated overnight. Flow cytometry confirmed transduction efficiencies at around 30% in each replicate for mKate2+% expression at 72 h after LV transduction. Cas9-expressing Jurkat cells (Jurkat-Cas9) were selected and expanded in a culture medium with puromycin. When mKate2+% achieved above 95%, 1 × 107 cells per replicate were transduced with AAV6-WT-EGFP or AAV-M-EGFP at the indicated MOI to allow the percentage of EGFP+ cells to reach over 80%. At 72 h after AAV transduction, cells collected for the top 20% and bottom 20% were sorted via Flow Cytometer. BD FACSAria™ Fusion and SONY MA900 were used for cell sorting.
After sorting, genomic DNA was extracted by TIANamp Genomic DNA Kit (TIANGEN, DP304-03). Then, sgRNA sequences were amplified using Q5 master mix (2×) (NEB, #M0544S) and purified using SPRI beads (Beckman, #A63882) for NGS. PCR primers with P5 and P7 adapters (NGS-Lib-KO-F and NGS-Lib-KO-R) were listed in Supplementary Table S2. sgRNA sequences used in validation were listed in Supplementary Table S3.
Screening data analysis
The CRISPR screening was analyzed using MAGeCK
34. The difference in genes between the bottom 20% of cells and the top 20% of cells was evaluated by the read count of corresponding gRNAs. In addition, the negative gRNAs were input as background with the parameter (--control-sgrna). The MAGeCK score and log
2 fold change were used to rank genes from the CRISPR screening. The log
2 fold change was calculated as the log
2 transformed ratio between the normalized read counts in the bottom 20% sample and the top 20% sample. Normalization was based on the sequencing depth of each sample.
Validating candidate genes in Jurkat cells
Candidate gene-knockout cell lines were generated using Jurkat-Cas9 cells transduced with LV encoding the indicated sgRNA under the U6 promoter. From day 2 after LV transduction, Jurkat cells were selected with puromycin (2 μg/mL) for 7–10 days. Then, the genomic DNA of the Jurkat cells was extracted for PCR amplification of the target gene sequence. The indel% of the target gene was assessed via SYNTEGO analysis based on Sanger sequencing of PCR products. Primers used in Sanger sequencing for knock-out validation were listed in Supplementary Table S2. Before AAV transduction, CD62L– cells were sorted via BD FACSAria™ Fusion for high purity. For each group, a total of 5 × 104 Jurkat cells were transduced with AAV6-WT-EGFP or AAV6-M2-EGFP. 48 h after AAV transduction, cells were collected and tested for EGFP expression via flow cytometry.
Validating candidate genes in human T cells
Following 24 h of activation, human T cells were nucleofected using the P3 Primary Cell 4D-Nucleofector™ X Kit (Lonza, #V4XP-3032) and a 4D-Nucleofector™ X Unit (Lonza, #AAF-1003X). SgRNAs were synthesized by GenScript (Supplementary Table S3). RNP complexes were prepared by incubating 30 pmol of recombinant Cas9 protein (Takara, #632641) with 75 pmol of sgRNA at 25 °C for 10 min. A total of 1 × 106 activated T cells were electroporated with RNPs per well using the EO-115 program. After nucleofection, cells were resuspended in fresh culture medium supplemented with cytokines. Three days post-knockout, AAV6-WT-EGFP or AAV6-M2-EGFP were added to the culture at an MOI of 1 × 104. Another three days later (day 6 post-nucleofection), cells were collected and stained with anti-CD62L antibody. EGFP expression was analyzed by flow cytometry in both CD62L+ and CD62L− T cell populations.
Cryo-EM structure
For cryo-EM sample preparation, 3.5 μL of the AAV sample (1.3 × 1014 vg/mL) was applied to a glow-discharged Quantifoil R1.2/1.3 Cu 300 mesh grid. Grids were blotted for 4 s with a blot force of 10 following a 6 s waiting time using a Vitrobot Mark IV (Thermo Fisher Scientific) under 100% humidity at 4 °C. The grids were then plunge-frozen in liquid ethane cooled by liquid nitrogen.
Data collection was performed on a Titan Krios transmission electron microscope (Thermo Fisher Scientific) operating at 300 kV, equipped with a Gatan K3 Summit direct electron detector and a GIF Quantum energy filter (20 eV slit width). Movie stacks were recorded in super-resolution mode at a nominal magnification of 130,000× using EPU software, with a preset defocus range of −1.4 to −2.0 μm in Aberration-Free Image Shift (AFIS) mode. Each stack was exposed for 0.85 s (32 frames, 0.0266 s per frame), with a total electron dose of ~50 e−/Å2.
A total of 11,707 movie stacks were collected. Of these, 9,621 micrographs were manually selected and processed using CryoSPARC v4.6.2. Patch motion correction and Patch CTF estimation were applied. An initial round of auto-picking yielded ~157,325 particles with a box size of 640 pixels. After iterative rounds of template-based particle picking and 2D classification, 156,281 particles were selected for ab initioreconstruction (C1 symmetry). Homogeneous refinement, followed by local CTF refinement and non-uniform refinement, produced a final 3D reconstruction at an overall resolution of 1.8 Å.
A structural model of AAV6-M2 was initially generated using the AlphaFold3 prediction
35. Manual model adjustment and refinement were performed in COOT
36, followed by atomic-level refinement using Phenix
37. Structural figures were prepared using ChimeraX
38.
Receptor docking
Flexible docking between AAV6-M2 and the transmembrane receptor L-selectin was performed using the HADDOCK2.4 web server
39. The input structures included the AAV6-M2 trimer (PDB: 9VI4) and the lectin/EGF domains of L-selectin (PDB: 5VC1). All small molecules were removed from the L-selectin structure, except for the calcium ion. The engineered loop region of one AAV6-M2 monomer and the EGF domain of L-selectin were defined as active residues for docking. The top-ranked docking model was visualized using ChimeraX
38.
Mouse experiments
HIS mice generation
NOD/ShiLtJGpt-Prkdc
em26Cd52Il2rg
em26Cd22kit
em1Cin(V831M)Il15
em1Cin(hIL15)/Gpt (NCG-X-hIL15) mice (Strain NO. T037155) and NOD/ShiLtJGpt-Prkdc
em26Cd52 Il2rg
em26Cd22 Rosa26
em1Cin(hCSF2&IL3&KITLG)/Gpt (NCG-M) mice (Strain NO. T036669) were purchased from GemPharmatech Co., Nanjing, China. These mice were housed under specific pathogen-free conditions at the Model Animal Research Center, Medical School of Nanjing University, with all experiments conducted in compliance with institutional guidelines under an approved animal protocol (LY-02). HIS mice were generated as previously described
40. Briefly, human hematopoietic stem and progenitor cells (HSPCs) were purified from fetal liver using the human CD34 MicroBead Kit (Miltenyi Biotec, #130-046-702) following ethical approval (Drum Tower Hospital, protocol #2021-488-01/02). Newborn NCG-M mice received 80 cGy sublethal irradiation followed by intrahepatic injection of 5 × 10
4 CD34
+CD38
– HSCs within 4–6 days after birth, whereas non-irradiated NCG-X-hIL15 mice underwent the same HSC transplantation procedure. Successful immune reconstitution was confirmed at 10 weeks post-engraftment by flow cytometry detection of human immune cells (CD45
+, CD3
+, CD19
+, CD14
+, CD56
+) in peripheral blood, with engraftment defined as > 1 × 10
5 human CD45
+ cells/mL.
In vivo CAR-T generation in HIS mice
Humanized NCG-X-hIL15 mice demonstrating sufficient T cell engraftment (≥ 15% human CD3+ T cells in peripheral blood human CD45+ cells) were randomly assigned to two experimental groups. These groups received intravenous administration of either AAV-M2-CAR19-EGFP or AAV6-CAR19-EGFP at a dose of 3 × 1013 vg/kg. Peripheral blood CART signal and B cell dynamics were monitored at bi-weekly intervals via flow cytometry. Mice were sacrificed 4 weeks and 6 weeks post-AAV injection for comparative immune cells profiling in indicated tissues, including CAR-T cell transduction efficiency (EGFP+ in human CD3+ T cells, CD4+ T cells, CD8+ T cells, Tn cells, TCM cells, TEM cells, TEMRA cells) and B cell depletion. Flow cytometry (Agilent NovoCyte Penteon, Santa Clara, USA) was used for cell population detection. Antibodies were listed in Supplementary Table S4. Gating strategies were summarized in Supplementary Fig. S7a.
Evaluation of lupus pathologies in AAV6-M2-CD19CAR-treated HIS SLE mice
HIS-SLE mouse model was generated as previously described
41. Briefly, we topically administered the TLR7/8 agonist R848 to NCG-M HIS mice for eight consecutive weeks to induce systemic lupus-like manifestations. Following five weeks of R848 treatment, mice demonstrated comparable levels of human immune reconstitution (quantified by peripheral blood human CD45
+ cell counts) and elevated serum autoantibody titers (quantified by anti-dsDNA antibody) were randomly allocated to receive either AAV6-M2-CD19CAR (3 × 10
13 vg/kg, i.v.) or PBS control. Mice were euthanized at 8 weeks post-AAV administration for endpoint analysis. Then, serum autoantibody levels were measured by ELISA: anti-dsDNA antibody (COIBO BIO, #CB13357-Hu) and anti-Sm antibody (COIBO BIO, #CB19966-Hu) following the manufacturer's instructions. Serum biochemical profiling was quantified using an automated clinical analyzer (HITACHI 3500, Tokyo, Japan) following manufacturer-recommended protocols. Subsequently, B cell subsets (Transitional B, Naïve B, Plasmablast, Memory B) in indicated tissues were further analyzed by flow cytometry. All antibodies used were listed in Supplementary Table S4. Gating strategies were summarized in Supplementary Fig. S7b.
Quantification of AAV accumulation in the liver of HIS mice
Various organs from HIS mice were collected at the endpoint of the experiment. For each organ, a 50–80 mg minced tissue sample was homogenized, and genomic DNA was extracted using the TIANamp Genomic DNA Kit (TIANGEN, #DP304-03). Absolute quantification of AAV-delivered CAR copies was performed via qPCR using 100 ng of gDNA per reaction. Primers (QCAR-F, QCAR-R) and probes (QCAR-Probe) targeting the CAR transgene were designed as listed in Supplementary Table S2. PCR amplification was conducted using Hot Start Taq 2× Master Mix (NEB, #M0496S) with the following thermocycling conditions: 95 oC for 30 s, followed by 40 cycles of 95 oC for 15 s, 52 oC for 30 s, and 68 oC for 20 s. Ct values were converted to AAV vector genome copies (Vg) based on a standard curve.
Quantification of AAV accumulation in human T cell subpopulation in the spleen of HIS mice
Humanized NCG-X-hIL15 mice received intravenous administration of AAV6-M2-CAR19-EGFP at a dose of 3 × 1013 vg/kg. Mice were sacrificed at 12 h post AAV injection, and spleen was taken for cell sorting. Four cell populations, including human CD4+CD62L+ cells, human CD4+CD62L– cells, human CD8+CD62L+ cells, human CD8+CD62L– cells were sorted for 1~2 × 105. Then, the genomic DNA of each cell population was extracted by DNeasy Blood & Tissue Kit (Qiagen, 69504). Absolute quantifications of AAV6-M2-CAR19-EGFP copies and human beta-actin copies were performed via qPCR using 30 ng of gDNA per reaction. PCR amplification was conducted using Hot Start Taq 2× Master Mix with the following thermocycling conditions: 95 oC for 30 s, followed by 40 cycles of 95 oC for 15s, 52 oC for 30 s, and 68 oC for 20 s. Ct values were converted to AAV vector genome copies (Vg) based on a standard curve.
Quantification of AAV accumulation in the liver of B6 mice
C57BL/6J mice were purchased from the Animal Center of Westlake University and housed under standard, individually ventilated, pathogen-free conditions at the Laboratory Animal Resource Center of Westlake University. All animal experiments were conducted in accordance with the Institutional Animal Care and Use Committee (IACUC) guidelines approved by Westlake University. Six-week-old mice were administered scAAV6-WT-EGFP, scAAV6-M2-EGFP, scAAV6-Ark312-EGFP, or scAAV6-Ark315-EGFP via intravenous injection at a dose of 5 × 10
12 vg/kg. Phosphate-buffered saline (PBS) was injected intravenously as a negative control. At day 10 post-injection, mice were euthanized, and livers were harvested for EGFP expression analysis using the PHOTON IMAGER™ OPTIMA system, and EGFP signals from individual livers were quantified. The AAV viral copies in the liver were also quantified by qPCR using a similar approach as described in the HIS mice section. The capsid sequences of Ark312 and Ark315 were obtained from published patents
29,30.
H&E staining
Liver and lung tissues were fixed in 4% paraformaldehyde (Beyotime Biotechnology, P0099) at room temperature for 24 h, dehydrated through a graded ethanol series, cleared in xylene, and embedded in paraffin. Tissue sections were cut using a microtome, mounted on glass slides, and dried overnight. H&E staining, sections were deparaffinized in xylene, rehydrated through a descending ethanol series (100% to 70%), stained with Harris’ hematoxylin for 3 min, and counterstained with eosin Y for 1 min. Slides were then dehydrated through an ascending ethanol series, cleared in xylene, and coverslipped. Stained sections were examined using an Olympus microscope (VS200, Olympus, Tokyo, Japan).
Cynomolgus macaque experiments
The non-human primate (NHP) experiments were conducted at the NHP facility of Innostar Biotechnology and approved by Innostar’s Institutional Animal Care and Use Committee (IACUC). A 2-year-old cynomolgus macaque was used in this study. Four weeks after intravenous administration of AAV (1 × 1012 vg/kg), the animal was euthanized with 10 mg/kg ketamine hydrochloride and 5 mg/kg xylazine hydrochloride, followed by PBS perfusion. To quantify viral genome copies, 50–100 mg of tissue samples were homogenized in Buffer ATL (QIAGEN) using a tissue grinder. Genomic DNA and viral DNA were extracted using the DNeasy Blood & Tissue Kit (QIAGEN) according to the manufacturer’s instructions. Vector genomes were amplified from the extracted DNA using primers flanking the transgene and the Q5 High-Fidelity Master Mix (NEB, #M0544S). Amplicons containing Illumina adapter sequences were quantified using Qubit, pooled at equal mass, and sequenced on an Illumina NovaSeq X Plus platform.
Flow cytometry data collection and analysis
Flow cytometry (Agilent NovoCyte Penteon and CytoFLEX) was used to analyze cells. Flow cytometry data were analyzed via FlowJo.
Statistical analysis
Statistical analysis was performed using GraphPad Prism v10, and the specific tests were indicated in the figure legend.
The Author(s) 2026. Published by Higher Education Press. This is an Open Access article distributed under the terms of the CC BY license (https://creativecommons.org/licenses/by/4.0/).
