跳到正文
原文
Nature Biotechnology· Lingjuan Hong·· 4 天前精选AI 评分80

新型AAV载体提高T细胞和NK细胞基因编辑效率

A CD7-targeted AAV variant enables efficient T and NK cell engineering in vivo

AI 导读

研究开发了一种新型AAV载体AAV.APR31,显著提高了T细胞和NK细胞的基因编辑效率,包括CAR-T、TCR-T和CAR-NK细胞工程。该载体在多种细胞类型中表现出高效且特异的基因编辑能力,优于现有载体。

推荐理由

研究揭示了一种新型AAV载体,显著提高了T细胞和NK细胞的基因编辑效率,有望提升免疫疗法效果。

正文

Main

T cells are at the forefront of cancer immunotherapy1,2,3,4,5,6,7. The US Food and Drug Administration has approved seven chimeric antigen receptor (CAR) T cell products for treating hematological malignancies, including B cell acute lymphoblastic leukemia (B-ALL)8,9, large B cell lymphoma10,11,12,13, mantle-cell lymphoma14, follicular lymphoma15 and multiple myeloma16,17, as well as the first T cell receptor (TCR) T cell therapy product, Tecelra (afamitresgene autoleucel), for treating a solid tumor, metastatic synovial sarcoma. Numerous clinical trials to treat solid tumors underway7,18,19. Several approaches have been used to produce human CAR-T cells and other cell-based immunotherapies20,21. The most widely used gene delivery systems are lentiviruses or gammaretroviruses, which are the basis for all currently approved CAR-T products. However, these vectors integrate semirandomly, which can lead to transgene silencing22 and clustering of integration near transcriptional start sites23,24, posing a risk of insertional mutagenesis and oncogenic transformation25,26,27,28,29,30. Safety considerations motivate the development of gene delivery strategies that provide improved control over transgene expression and enhanced targeting specificity.

AAV-based vectors may provide an attractive alternative because of their favorable safety profile and compatibility with precise genome engineering. Targeted transgene integration in primary human T cells has been successfully achieved through the use of AAV and CRISPR technologies31,32. However, there is a need for AAV serotypes that enable more efficient transduction of human T cells. AAV6 has shown acceptable delivery efficiency only at very high doses, with multiplicity of infection (MOI) ranging from 1 × 105 to 1 × 106 viral genomes (vg) per cell31,33,34. Such high MOI can result in toxicity to T cells, trigger immune responses35,36 and increase the cost and potential off-target effects37.

Here, we evolved AAV capsids directly in primary human T cells and identified a family of variants carrying an APR motif in variable region IV (VR-IV), among which AAV.APR31 (also termed HONG31) emerged as a lead candidate with markedly enhanced tropism for human T and NK cells. Mechanistic studies revealed that AAV.APR31 exploits a CD7-dependent entry route while retaining the canonical AAVR pathway, defining a layered-entry mechanism that accounts for its potent yet lineage-restricted lymphoid transduction. Functionally, a single capsid AAV.APR31 supports transient transgene expression, CRISPR-mediated targeted knock-in and AAV-Sleeping Beauty (SB) transposon delivery, thereby enabling CAR-T, TCR-T and CAR-NK cell engineering ex vivo, as well as direct in vivo CAR-T cell generation in both hematologic and solid tumor models. Under matched conditions in primary human T cells, AAV.APR31 further outperformed the concurrently reported human T cell-tropic capsid AAV-hT7 (ref. 38). Together, these findings show that AAV.APR31 is an evolved lymphotropic AAV capsid that, unlike existing single-modality delivery systems, enables transient and stable engineering across T and NK cells.

Results

Directed evolution identifies a family of APR-motif-enriched AAV capsid variants with human T cell tropism

We engineered an AAV capsid library using AAV6 as a starting serotype, given its prior use in human T cell gene-editing applications31,34,39. We performed saturation mutagenesis on a pseudotyped AAV2/6 genome, comprising the AAV2 Rep region and the AAV6 Cap region flanked by AAV2 inverted terminal repeats (ITRs) (Fig. 1a). We targeted VR-IV of the VP1 capsid protein (amino acids 454–460), a known surface-exposed domain involved in host interaction and antibody neutralization40. For functional selection, we screened the library directly in primary human T cells. Peripheral blood mononuclear cells (PBMCs) from healthy donors were activated using human CD3/CD28/CD2 T cell activator and transduced with the AAV capsid library at MOI = 1,000 (MOI = 500 for the third round of selection), that is, 1,000 vg per cell. To enrich for capsid variants capable of successful cellular entry, cells were subjected to stringent washes to remove surface-bound and unbound viruses. Internalized viral DNA was extracted; then, the VR-IV region was amplified by PCR and subsequently recloned into the AAV6-based plasmid backbone (the wild-type VR-IV DNA sequence was replaced with EGFP to avoid any capsid6 contamination during the library cloning) to generate a new capsid library for the next round of selection (Fig. 1a). After three rounds of selection, high-throughput sequencing revealed over 3 million unique capsid variants in the initial library, which decreased to ~1 million variants after selection (Extended Data Fig. 1a–c).

Fig. 1: Directed evolution identifies a family of AAV capsid variants with human T cell tropism.

a, Schematic of AAV capsid evolution strategy in human T cells. A pooled library of AAV6 capsid variants was evolved in activated human T cells. b, The NGS data analysis of parental and each round of the evolved capsid library. Bubble plots show the outcome of capsid evolution and each bubble represents a unique capsid variant at the amino acid level. Bubble size indicates the fold change of variants (n = 2 donors). c, Correlation analysis of enrichment of the top 1,000 AAV variants after the third round of selection across donor A (x axis) and donor B (y axis). Each point represents a unique variant, categorized as AAV6 (pink), APR variants (green) or non-APR variants (gray). The dashed black line represents the line of best fit. The Pearson correlation coefficient (r = 0.813) and the corresponding P value (P < 2.2 × 10−16) are indicated on the plot. d, Enrichment of variants over selection rounds in human T cells. Line graph showing the percent of reads for different variant groups across three rounds of selection in donor A and donor B. e, Motif enrichment analysis of the top 200 enriched capsids from the third-round evolved library. f, The 7-mer sequence of the top ten enriched APR-containing variants.

The quantification analysis revealed substantial enrichment of a family of variants harboring the APR motif at positions 455–457 of VR-IV, which eventually dominated the library after three rounds of selection (Fig. 1b), including AAV.APR31 with an amino acid insertion 454-AAPRANE-460. This enrichment pattern was consistently observed in libraries from two independent donors (Fig. 1b). Notably, wild-type AAV6 was also among the top enriched variants early in the screen, although the underlying mechanism remains unclear. Correlation analysis of the top 1,000 AAV variants after the third round of selection across the two donors showed a strong positive correlation (Fig. 1c and Extended Data Fig. 1d), demonstrating the reproducibility and robustness of the screening strategy. We further quantified the relative abundance of AAV6, APR variants and non-APR variants in the pool. Notably, the AAV.APR variants rapidly emerged and dominated the population (Fig. 1d), while the non-APR variants progressively decreased because of their lower transduction efficiency (Fig. 1d). Motif analysis of the top 200 enriched capsids revealed strong enrichment of the APR motif at positions 2–4 within the 7-mers (Fig. 1e,f and Supplementary Table 1). Even among the top 1,000 variants from both donor screens, the APR-containing capsids demonstrated substantially higher enrichment compared to the non-APR counterparts (Extended Data Fig. 1e,f).

AAV.APR31 enables efficient transgene delivery in human and mouse T and NK cells

To determine whether APR-containing AAV variants outperform AAV6 in T cell transduction, we focused on AAV.APR31, one of the most enriched variants. First, we assessed its AAV packaging yield by producing a self-complementary AAV (scAAV) vector carrying an EGFP cassette (scAAV-CAG-EGFP) and a single-stranded AAV (ssAAV) vector carrying a CD22-CAR cassette (Fig. 2a). AAV.APR31 yielded 1.02 × 1014 vg per L and 1.63 × 1013 vg per L of medium for EGFP and CD22-CAR cassette packaging, respectively, compared to 2.29 × 1014 vg per L and 2.87 × 1013 vg per L for parental AAV6 under matched production conditions (Fig. 2b). Although the packaging yield of AAV.APR31 was slightly reduced relative to AAV6 (2.05-fold and 1.76-fold reductions for scAAV-CAG-EGFP and ssAAV-EFS-CD22-CAR production, respectively), the variant maintained robust vector production across distinct transgene cassettes, suggesting that the 454–460 motif substitution remains compatible with efficient AAV production. Next, we evaluated the cell-surface binding and uptake of the new variant. Activated human T cells were incubated with the virus at 4 °C to allow binding without endocytosis. After stringent washes, vg per cell was quantified by qPCR. AAV.APR31 bound significantly more vector than AAV6 (20.3 vg per cell versus 5.8 vg per cell at MOI = 50, 48.3 vg per cell versus 6.3 vg per cell at MOI = 100 and 252.7 vg per cell versus 49.6 vg per cell at MOI = 1,000) (Fig. 2c). To assess uptake, human T cells were incubated with virus in a 37 °C incubator for 2 h, revealing dramatically higher internalization by AAV.APR31 (37.9 vg per cell versus 2.5 vg per cell at MOI = 50, 103.4 vg per cell versus 3.6 vg per cell at MOI = 100 and 1,319.3 vg per cell versus 34.9 vg per cell at MOI = 1,000) (Fig. 2d).

Fig. 2: Characterization of AAV.APR31 variant and transgene delivery ability.

a, Schematic of scAAV-CAG-EGFP and ssAAV-EFS-CD22-CAR constructs. b, AAV packaging ability of AAV6 and AAV.APR31. AAV titer was quantified as vg by qPCR. Data are presented as the mean values ± s.e.m. (n = 9 biological replicates for scAAV (EGFP), P = 0.0005; n = 5 biological replicates for ssAAV (CD22-CAR), P = 0.0307). c, The number of AAVs bound to the human T cell surface, quantified as vg by qPCR. Data are presented as the mean values ± s.e.m. (n = 3 biological replicates, P = 0.0004). d, Number of internalized AAVs by a 2-h incubation of AAV and human T cells in a 37 °C incubator. Data are presented as the mean values ± s.e.m. (n = 3 biological replicates, P < 0.0001). e, Schematic of AAV transduction and EGFP detection in Jurkat. f, Representative flow cytometry analysis of EGFP expression following transduction of Jurkat cells with AAV6 or AAV.APR31 at the indicated MOIs. g, EGFP percentage and intensity quantification of transduced Jurkat cells. MFI, mean fluorescence intensity. Data are presented as the mean values ± s.e.m. (n = 3 biological replicates, P < 0.0001). h, Schematic of human PBMC isolation, activation, AAV transduction and flow cytometry in human T cells. i,j, Flow cytometry analysis of EGFP expression following transduction of human T cells with AAV6 or AAV.APR31 at the indicated MOIs. Left, fluorescence histograms. Right, EGFP quantification of transduced cells. Donor 300F (i) and donor E (j) were used on day 5 after activation (n = 1 replicate with six different MOIs). k, Representative flow cytometry analysis of EGFP expression following transduction of human T cells with AAV6 or AAV.APR31 at MOIs of 250, 500 and 2,500. T cells were used for transduction on day 2 after activation. l, The quantification of T cell transduction efficiency of AAV6 and AAV.APR31 in four healthy donors. T cells were used for transduction on day 2 after activation. Data are presented as the mean values ± s.e.m. (n = 4 donors, P = 0.003). m, The quantification of T cell transduction efficiency of AAV6 and AAV.APR31 in seven healthy donors. T cells were used for transduction on day 5 after activation. Data are presented as the mean values ± s.e.m. (n = 7 donors, P < 0.0001). In this figure, two-sided unpaired t-tests (b) and ordinary two-way analyses of variance (ANOVAs; c,d,g,l,m) were used to evaluate statistical significance. *P < 0.05, **P < 0.01, ***P < 0.001 and **** P < 0.0001.

Source data

We then compared functional transduction in Jurkat cells, an immortalized human T lymphocyte cell line. We packaged scAAV-CAG-EGFP either with AAV.APR31 or AAV6 and then transduced Jurkat at MOIs ranging from 50 to 1,000; EGFP expression was measured by flow cytometry 48 h later (Fig. 2e). Notably, AAV.APR31 achieved 26.47% EGFP-positive cells at an MOI of 100 versus 0.84% for AAV6, increased to 58.07% versus 5.07% at an MOI of 250 and exceeded 90% at an MOI of 1,000 versus 18.13% for AAV6 (Fig. 2f,g). To independently validate these findings at the transcript level, we quantified EGFP mRNA by reverse transcription (RT)–qPCR in the same experimental conditions. Consistent with the flow cytometry data, EGFP transcript levels were markedly higher in AAV.APR31-transduced Jurkat cells compared to AAV6 across all MOIs tested (Extended Data Fig. 2a), confirming that the superior transduction efficiency of AAV.APR31 is reflected at both mRNA and protein levels. Lastly, we evaluated primary T cells from multiple donors (Fig. 2h). Even at a relatively low MOI of 250, AAV.APR31 outperformed AAV6 transduced at a much higher MOI of 10,000 (donor 300F: 16.7% versus 13.1%, respectively; donor E: 17.2% versus 15.0%, respectively) on day 5 after donor human T cell activation (Fig. 2i,j and Extended Data Fig. 2b,c). To assess the applicability of AAV.APR31 across diverse genetic backgrounds and at different stages of T cell activation, we repeated the transduction experiments using primary human T cells from 11 additional donors. Four donors were transduced at an early time point (day 2), whereas seven donors were transduced at a later time point after T cell activation (day 5) (Fig. 2k–m). Consistently, AAV.APR31 exhibited robust transduction efficiency across all samples. We also evaluated the ability of AAV.APR31 to transduce mouse CD8+ T and NK cells using an EGFP reporter. AAV.APR31 demonstrated markedly higher transduction efficiency compared to parental AAV6 in both cell types (Extended Data Fig. 2d,e).

AAV.APR31-mediated potent site-specific transgene knock-in in human T cells

To assess the dependence on dosing, we targeted the first exon of the TRAC locus (a gene encoding the TCR alpha chain) to create universal allogeneic T cells31,41,42 (Fig. 3a), which eliminates endogenous TCR signaling interference43. Human T cells were electroporated with Cas9 ribonucleoprotein (RNP) and then transduced with either AAV6 or AAV.APR31 carrying an EGFP or CD22-CAR cassette (Fig. 3a,b). At the lowest MOI (250 vg per cell), AAV.APR31 achieved 30.95% EGFP knock-in versus 4.34% with AAV6 and approximately 60% knock-in at an MOI of 1,000 on day 5 after electroporation (Fig. 3c). We next delivered a larger (1,803 bp) CD22-CAR cassette to different donors. At an MOI of 100, AAV.APR31 yielded CD22‑CAR expression in 18.7% (Fig. 3d) and 14.85% (Fig. 3e) of T cells, compared to 1.49% and 2.21% for AAV6 (Fig. 3d,e and Extended Data Fig. 2f,g). Even at an MOI of 50, AAV.APR31 maintained 7.65% versus 1.47% in AAV6 (Fig. 3e). At a higher MOI of 1,000, the differences in knock-in efficiencies between AAV.APR31 and AAV6 were more pronounced (57.8% versus 14.2%, 45.05% versus 12.8%; Fig. 3d,e). To confirm that the observed EGFP and CD22-CAR signal reflects stable genomic integration rather than residual episomal expression, cells transduced with AAV alone (without electroporation) were analyzed in parallel; no detectable EGFP or CD22-CAR expression was observed on day 5 under these conditions (Extended Data Fig. 2h,i), demonstrating that the EGFP+ or CD22-CAR+ population in the editing experiments represents bona fide knock-in events.

Fig. 3: AAV.APR31-mediated highly efficient gene knock-in in primary human T cells.

a, Schematic of EFS-EGFP-poly(A) and EFS-CD22-CAR-poly(A) constructs in the AAV backbone and knock-in strategy in the TRAC locus. b, Schematic of the transgene knock-in pipeline in primary human T cells. c, Flow cytometry analysis of EGFP expression following RNP electroporation and AAV transduction of human T cells. Left, fluorescence histograms. Right, EGFP quantification of transduced cells (n = 2 donors). d,e, CD22-CAR integration efficiency was analyzed by flow cytometry in four different donor sources of activated T cells. The knock-in efficiency was quantified by the number of CD22-CAR-positive cells across a range of MOIs (n = 2 donors for each experiment). f, CD22-CAR integration efficiency was analyzed by flow cytometry in nonprestimulated bulk CD8+ T cells (n = 2 donors). The knock-in efficiency was quantified by the number of CD22-CAR-positive cells across a range of MOIs on day 7 after AAV transduction (n = 2 donors). g, Schematic of CD22-CAR knock-in at the TRBC1 locus. h, Quantification of CD22-CAR knock-in efficiency across the indicated MOIs. Data are presented as the mean values ± s.e.m. (n = 4 donors, P < 0.0001). i, Schematic of BFP knock-in at the GAPDH locus. j, Quantification of BFP knock-in efficiency across the indicated MOIs. Data are presented as the mean values ± s.e.m. (n = 3 donors, P < 0.0001). k, Representative flow cytometry plots showing dual AAV-mediated delivery of EGFP and BFP to the TRAC and GAPDH loci, respectively, at MOIs of 250 and 1,000. l, Quantification of EGFP +BFP+ double-positive cells across four MOIs (n = 1 donor with four different MOIs). m, Representative flow cytometry plots showing simultaneous knock-in of MSLN-CAR and CD22-CAR into the TRAC and GAPDH loci, respectively, at MOIs of 250 and 1,000. n, Quantification of MSLN-CAR+CD22-CAR+ double-positive cells across four MOIs (n = 1 donor with four different MOIs). In this figure, ordinary two-way ANOVAs (h,j) without adjustment were used to evaluate statistical significance. ****P < 0.0001.

Source data

To assess whether AAV.APR31 results in aberrant increases in vector genome integration, we quantified knock-in CAR copy number per cell by qPCR in AAV6-transduced and AAV.APR31-transduced T cells. Given that our approach targets a single genomic locus, the theoretical maximum integration is two copies per cell (one per allele). We observed that the mean CAR copy number in AAV.APR31-transduced cells was comparable to that of AAV6-transduced cells (1.34 versus 1.69 copies per cell; Extended Data Fig. 2j) and neither group exceeded the expected diploid maximum of two copies per cell. These findings indicate that the superior knock-in efficiency of AAV.APR31, reflected by a higher frequency of CAR+ cells, is driven by an increased proportion of successfully edited cells rather than by supraphysiological multicopy integration events. Furthermore, the absence of copy numbers > two suggests that AAV.APR31 does not promote off-target random integration, consistent with the site-specific nature of homology-directed repair (HDR)-mediated knock-in.

Deliberate ex vivo pre-stimulation with anti-CD3/anti-CD28 is commonly used to render T cells permissive to HDR-mediated knock-in; however, less differentiated T cells can produce more potent CAR-T cells44. Therefore, we assessed whether AAV.APR31 could support knock-in in bulk CD8+ T cells that were not deliberately prestimulated ex vivo (Fig. 3b). Because HDR is restricted to cycling cells, we characterized the activation state of the cells before editing. Compared to antibody-activated T cells (day 2: 91.9% CD69+, 95.0% CD25+; day 4: 41.3% CD69+, 94.5% CD25+), nonprestimulated cells showed only partial activation (40.3% CD69+, 5.7% CD25+) (Extended Data Fig. 2k,l) while retaining a partially detectable Ki67+ fraction (Extended Data Fig. 2m). Under these conditions, AAV.APR31 mediated efficient knock-in that markedly exceeded AAV6 (13% versus 1.99% at MOI = 100; 35.8% versus 2.95% at MOI = 1,000) (Fig. 3f). As expected, knock-in in nonprestimulated CD8+ T cells was lower than in prestimulated T cells, consistent with the higher activation and larger cycling fraction of the latter rendering them more permissive to HDR.

To test this knock-in system at other genomic loci, we performed DNA knock-in within the TRBC1 and GAPDH loci of the human T cell genome (Fig. 3g,i). We observed locus-dependent knock-in efficiencies, likely reflecting RNP cutting rates. At TRBC1 exon 1, AAV.APR31 produced knock-in efficiencies of 24.08% at MOI = 250 (versus 4.12% with AAV6) and 48.3% at MOI = 1,000 (versus 10.34% with AAV6) (Fig. 3h). At GAPDH exon 9, AAV.APR31 reached 36.4% and 56.87% of BFP-positive cells at MOIs of 250 and 1,000, respectively (Fig. 3j).

Delivering multiple genetic elements to multiple desired genomic loci can be technically challenging, especially in cells that are difficult to transfect, such as primary human T and NK cells. To evaluate multiplex editing using AAV.APR31, we simultaneously delivered two HDR templates targeting the TRAC and GAPDH loci. Dual knock-in was achieved in 26.1% of cells at MOI = 250 (versus 0.6% for AAV6) and 50.8% at MOI = 1,000 (versus 15.7% for AAV6) (Fig. 3k,l). To further test the efficiency of dual knock-in by AAV.APR31, we simultaneously targeted MSLN-CAR and CD22-CAR into the TRAC and GAPDH loci, respectively. AAV.APR31 enabled dual knock-in in 34.4% of cells at MOI 250 (versus 0.5% for AAV6) and 40.6% at MOI 500 (versus 2.65% for AAV6) (Fig. 3m,n). The total knock-in efficiency of the TRAC locus (MSLN-CAR) reached 56.8% of cells at MOI = 250 (versus 3.8% for AAV6) and the GAPDH locus (CD22-CAR) reached 55.0% of cells at MOI = 250 (versus 9.0% for AAV6) (Fig. 3m).

Genome-scale CRISPR screens identify essential genes for AAV.APR31 transduction in human T cells

To elucidate the mechanistic alterations of AAV.APR31 that improves its transduction efficiency over AAV6, we performed genome-scale CRISPR knockout (KO) screens to identify genes essential for AAV.APR31-mediated transduction in Jurkat cells, an immortalized T cell line, with comparable transduction efficiency in human primary T cells. To distinguish T cell-specific factors, we carried out a parallel CRISPR screen in NALM-6, a human B lymphocyte line, serving as a lineage control (Fig. 4a). First, we generated Cas9-expressing Jurkat and NALM-6 cell lines (Jurkat-Cas9 and NALM-6-Cas9) and transduced them with the Brunello sgRNA library45. Following puromycin selection, the resulting heterogeneous cell populations, each containing distinct gene KOs, were transduced with AAV.APR31 carrying an EGFP cassette (Fig. 4a). Cells were then sorted and sequenced on the basis of EGFP expression into populations with high (top 18%) and low (bottom 18%) fluorescence (Fig. 4b).

Fig. 4: Genome-wide CRISPR KO screening identifies essential genes for AAV.APR31 transduction.

a, Schematic of genome-wide KO screens in Jurkat (T lymphocytes) and NALM-6 (B lymphocytes) cells to identify genes specifically for AAV.APR31 transduction in T lymphocytes. b, Flow sorting strategy of Jurkat-Cas9 and NALM-6-Cas9 cells (transduced with Brunello library) after scAAV-CAG-EGFP transduction. c, Scatter plots of median sgRNA log2 fold change (EGFPlow/high sorting bins) for each gene in two representative replicates of Jurkat screen. FC, fold change. d, sgRNA log2 fold changes for genes of the top enriched bin in the Jurkat screen. Bars represent the mean log2 fold change for each sgRNA across three independent replicates. The density plots above represent the distribution of all sgRNAs. e, Comparison of gene log2 fold change (median sgRNA, mean of three replicates) in Jurkat and NALM-6 screens. f, Cas9 RNP-mediated CD7 KO in Jurkat. Representative flow plots measuring CD7 expression after Cas9 RNP electroporation. g, Representative fluorescence histograms of Jurkat cells transduced with scAAV-CAG-EGFP after CD7 KO. h, Quantification of EGFP fluorescence intensity in Jurkat cells after CD7 KO (MOI = 1,000). Data are presented as the mean ± s.e.m. (n = 3 biological replicates, AAVS1-sg versus CD7-sg1: P < 0.0001, AAVS1-sg versus CD7-sg2: P < 0.0001). i, Cas9 RNP-mediated CD7 KO in human T cells. Representative flow plots of CD7 expression after Cas9 RNP electroporation. j, Quantification of EGFP fluorescence intensity in human T cells after CD7 KO (MOI = 10,000; (n = 2 donors). k, Quantification of EGFP fluorescence intensity in Jurkat cells that were pretreated with anti-human CD7 antibody. Data are presented as the mean values ± s.e.m. (n = 3 biological replicates, q < 0.0001 (IgG), q < 0.0001 (0.4 μg ml−1), q < 0.0001 (1 μg ml−1), q < 0.0001 (2 μg ml−1), q < 0.0001 (4 μg ml−1)). l, Quantification of EGFP intensity in human T cells that were pretreated with anti-human CD7 antibody (MOI = 10,000). Data are presented as the mean values ± s.e.m. (n = 3 donors, q = 0.0009 (IgG), q = 0.4589 (0.4 μg ml−1), q = 0.6839 (1 μg ml−1), q = 0.4589 (2 μg ml−1), q = 0.5301 (4 μg ml−1)). m, Representative flow plot of NALM-6 with CD7 overexpression. n, Quantification of EGFP fluorescence intensity in NALM-6 and NALM-6-CD7 cells after scAAV-CAG-EGFP transduction at different MOIs. Data are presented as the mean values ± s.e.m. (n = 3 biological replicates, CD7−-AAV.APR31 versus CD7+-AAV.APR31: P < 0.0001, CD7+-AAV6 versus CD7+-AAV.APR31: P < 0.0001). o, Quantification of EGFP expression in Jurkat cells transduced with scAAV-CAG-EGFP preincubated with recombinant human AAVR PKD1–PKD5 domain protein at the indicated concentrations. Recombinant EGFP protein was included as a negative control (MOI = 1,000). p,q, Representative BIAcore sensorgrams showing the binding of recombinant human AAVR PKD1–PKD5 domain protein to immobilized AAV6 (p) and AAV.APR31 (q) at the indicated analyte concentrations. The associated rate constant (Ka), dissociation rate constant (Kd) and equilibrium dissociation constant (KD) are indicated in each panel. RU, resonance units. r, Primary human CD8+ T cells, CD4+ T cells, NK cells, B cells and monocytes were transduced with scAAV-CAG-EGFP at an MOI of 250. EGFP expression was assessed using flow cytometry 48 h after transduction. Data are presented as the mean values ± s.e.m. (n = 3 donors, CD8+ T cells, q < 0.0001; n = 3 donors, CD4+ T cells, q < 0.0001; n = 3 donors, NK cells, q = 0.0248; n = 2 donors, B cells; n = 2 donors, monocytes). s, The AlphaFold 3-predicted structures of VR-IV domains of AAV6 and AAV.APR31 and the extracellular domain of CD7. t, The conformation and interactions (highlighted in cyan) of the AAV.APR31 monomer and CD7, with the VR-IV domain protruding toward CD7, establishing a strong interaction with corresponding residues in the extracellular domain of CD7. u, The core of the interaction interface between the AAV.APR31 VR-IV domain and CD7. The interaction residues are highlighted in orange and cyan. In this figure, two-sided unpaired t-tests (h), multiple two-sided unpaired t-tests (q value; k,l,r), and two-way ANOVAs with Tukey’s multiple-comparison tests (adjusted P value; n) were used to evaluate statistical significance. NS, not significant; *P < 0.05, ***P < 0.001 and ****P < 0.0001.

Source data

Subsequent sgRNA quantification confirmed the enrichment of several known and novel host factors. In the Jurkat screen, genes such as CD7, TM9SF2, KIAA0319L, VPS52, VPS51, VPS16, ATP2C1, TRAPPC1, GPR108, KPNA6, NR4A1, HELQ, KMT2D, SASH3, COG4, OSBPL8 and SLC35B2 were strongly enriched (Fig. 4c,d and Supplementary Table 2). In contrast, the NALM-6 screen revealed strong enrichment of TM9SF2, KIAA0319L, ST3GAL4, ATP2C1, NAA10, VPS52, VPS51, VPS29, JTB, LUC7L3, WDR24, GPR107, GPR108 and MAX (Extended Data Fig. 3a,b and Supplementary Table 3). Notably, top-ranked hits overlapped between the two screens, including KIAA0319L (also known as AAVR) (Fig. 4e), a well-established and essential receptor for AAV infection46, as well as TM9SF2, GPR108, VPS52, VPS51 and ATP2C1 (Fig. 4e), all of which have previously been implicated in AAV biology46,47. These targets highlighted several cellular components and molecular functions that were positively associated with AAV infection, including the Golgi apparatus subcompartment, the vesicle tethering complex, the histone methyltransferase complex, the methyltransferase complex, phospholipase binding and the MLL3/4 complex, which can all contribute to AAV.APR31 transduction (Extended Data Fig. 3c). The recurrence of these canonical AAV infection-related genes strongly supports the robustness and specificity of our screens. Comparative enrichment analysis between the Jurkat and NALM-6 screens highlighted CD7 (also known as GP40) as a top candidate specifically enriched in the Jurkat cells but not in NALM-6 cells (Fig. 4e). CD7 encodes a transmembrane protein belonging to the immunoglobulin superfamily and serves as a lineage marker, with expression detected on T cells48,49 and NK cells50. These findings suggest that the interaction between CD7 and AAV.APR31 enhances its transduction efficiency specifically in CD7-positive Jurkat cells but not in CD7-negative NALM-6 cells, supporting the potential role of CD7 as a lineage-specific entry factor.

To determine whether CD7 was required for AAV.APR31 transduction, we generated Jurkat cell lines with CRISPR–Cas9 KO of CD7 or AAVS1 (negative control) (Fig. 4f). Then, 5 days after electroporation with Cas9 RNPs, cells were infected with scAAV‑CAG-EGFP at an MOI of 1,000 vg per cell. AAVS1‑KO cells remained fully transducible (~100% EGFP-positive), whereas CD7‑KO markedly reduced AAV.APR31 transduction, while AAV6 was only modestly affected (Fig. 4g,h). We validated these findings in primary human T cells from two donors and showed that disruption of CD7 substantially impaired AAV.APR31‑mediated transduction (Fig. 4i,j and Extended Data Fig. 4a). To assess whether this CD7 dependency is conserved across species, we performed CRISPR-mediated Cd7 KO in primary mouse T cells, which similarly resulted in a significant but partial reduction in AAV.APR31 transduction efficiency (Extended Data Fig. 4b). Moreover, pretreatment of Jurkat and human T cells with an anti‑human CD7 antibody significantly inhibited AAV.APR31 transduction but had little effect on AAV6 (Fig. 4k,l and Extended Data Fig. 4c). The results jointly confirmed that disruption of interactions between AAV.APR31 and CD7, either by CD7 KO or antibody blocking, significantly hindered AAV.APR31 transduction in Jurkat cells and human and mouse T cells. Notably, CD7 disruption did not completely abolish AAV.APR31 transduction, suggesting that CD7 is not the sole determinant of viral entry. To further confirm the role of CD7 in AAV.APR31-mediated transduction, we overexpressed CD7 in a CD7-negative cell line, NALM-6 (Fig. 4m), and then evaluated whether the gain-of-expression of CD7 could recover AAV.APR31 transduction. As expected, NALM-6 CD7-expressing cells showed significantly increased AAV.APR31 transduction compared to wild-type CD7-negative NALM-6 cells in the same pool (Fig. 4n). To explore the relationship between CD7 expression level and AAV.APR31 transduction efficiency, we compared Jurkat (CD7-high), K‑562 (CD7-low) and NALM‑6 (CD7-negative) cells (Extended Data Fig. 4d). We found that AAV.APR31 transduction efficiency correlated directly with CD7 abundance on the cell membrane, whereas AAV6 remained unchanged across all lines (Extended Data Fig. 4e–g).

To investigate the CD7-independent entry pathway that accounts for residual transduction, we examined the role of the canonical AAV receptor AAVR. Preincubation of AAV6 and AAV.APR31 with soluble human AAVR polycystic kidney disease (PKD) domains 1–5 (Extended Data Fig. 4h), the key extracellular structure that interacts with the AAV capsid51,52, progressively blocked transduction of both vectors in Jurkat cells in a concentration-dependent manner (Fig. 4o), demonstrating that AAV.APR31 retains dependence on the canonical AAVR entry pathway. Surface plasmon resonance (SPR) analysis further revealed that AAVR binds AAV.APR31 with an affinity of 397 nM, compared to 230 nM for parental AAV6 (Fig. 4p,q), indicating that the APR motif substitution modestly reduces but does not abolish AAVR binding. Together, these data support a layered-entry model in which AAV.APR31 uses AAVR-mediated canonical entry as a baseline pathway, while CD7 engagement provides an additional and dominant enhancement specifically in T and NK cells, explaining both the residual transduction in CD7-negative contexts and the substantially elevated efficiency in CD7-expressing lymphoid cells.

To assess whether this CD7-dependent tropism extends to primary human immune cells and confers selectivity in a physiologically relevant context with multiple cell types, we transduced purified primary human T cells, NK cells, B cells and monocytes separately under cell-type-appropriate culture conditions at a fixed MOI of 250 vg per cell. AAV.APR31 achieved markedly higher transduction efficiency than AAV6 in all CD7-expressing populations: 61.7% versus 8.2% in CD8+ T cells, 49.9% versus 6.7% in CD4+ T cells and 35.8% versus 9.8% in NK cells (Fig. 4r). In contrast, both AAV.APR31 and AAV6 showed comparably low transduction in CD7-negative B cells (1.44% versus 1.25%) and monocytes (0.66% versus 0.41%), with no significant enhancement by AAV.APR31 over AAV6 in these populations (Fig. 4r). The magnitude of AAV.APR31-mediated improvement across cell types correlated with their respective CD7 expression levels, consistent with the pattern observed across cell lines and further supporting CD7 as the primary determinant of AAV.APR31’s preferential lymphoid tropism.

We next assessed whether amino acid insertion 454-AAPRANE-460 changes the spatial conformation of the VR-IV domain between AAV6 and AAV.APR31. We performed AlphaFold 3 prediction of the structures of the capsid monomer for AAV6 and AAV.APR31, as well as the structure of CD7 (Fig. 4s). The structural modeling revealed a new conformation of the VR-IV domain in AAV.APR31 monomer, with close interaction between the VR-IV domain of AAV.APR31 and CD7 extracellular structure (Fig. 4t,u). This interaction was not observed between AAV6 and CD7. The VR-IV domain of AAV.APR31 engaged with CD7 through key residues within the interaction interface (highlighted in orange and cyan), particularly within the APR motif, which extends toward CD7 and creates a complementary contact interface (Fig. 4t,u). The VR-IV domain centered on the APR motif preferentially formed electrostatic interactions or hydrogen bonds with residues located on the surface of CD7. The overall complementary protein surface interactions between AAV.APR31 monomer and CD7 further support the enhanced transduction ability of AAV.APR31 to human T cells.

AAV.APR31-mediated therapeutic immune cell generation for cell therapy

We next applied this system to engineer various therapeutic transgenes in primary human T cells and NK cells, including solid tumor-targeting CAR-T cells (MSLN-CAR and GD2-CAR), bispecific or tandem (CD19.20) CAR-T cells, TCR-T cells and CAR-NK cells (Fig. 5a). We designed HDR templates targeting exon 1 of the TRAC locus based on an AAV backbone (Fig. 5b) and the donor templates were packaged using either AAV6 or AAV.APR31. For solid tumor-targeting CAR-T cells, we generated MSLN-CAR and GD2-CAR-T cells using varying AAV doses and compared the performance of AAV6 and AAV.APR31. AAV.APR31 achieved markedly higher knock-in efficiencies, yielding 36.8%, 57.1%, 66.3%, 71.0% and 72.4% MSLN-CAR-positive T cells at MOIs ranging from 100 to 1,000 on day 4 after transduction (Fig. 5c). In contrast, AAV6 only achieved 0.92%, 3.48%, 8.25%, 11.7% and 18.9% CAR-positive T cells under the same conditions (Fig. 5c). A similar trend was observed in the generation of GD2-CAR-T cells, where AAV.APR31-mediated DNA knock-in exhibited significantly improved efficiency compared to AAV6 (Fig. 5d).

Fig. 5: AAV.APR31-mediated therapeutic immune cell engineering in human T and NK cells.

a, Schematic of the therapeutic immune cell generation flow chart using AAV6 or AAV.APR31. b, Schematic of MSLN-CAR, GD.2-CAR, CD22-CAR, CD19.20-CAR and NY-ESO-1 TCR structures in AAV backbone. c–e, MSLN-CAR (c), GD.2-CAR (d) and CD19.20-CAR (e) integration into the TRAC locus in activated human T cells. The knock-in efficiency was quantified by the number of CAR-positive cells across a range of MOIs. Left, fluorescence histograms. Right, quantification of CAR-positive cells (n = 1 donor for each construct with five different MOIs. f, NY-ESO-1 TCR knock-in efficiency in the TRAC locus. The knock-in efficiency was quantified by the number of EGFP-positive cells across a range of MOIs. Left, fluorescence histograms. Right, quantification of TCR-positive cells (n = 1 donor with five different MOIs. g, Representative fluorescence histograms of CD22-CAR in primary human NK cells. h, The knock-in efficiency of CD22-CAR into the TRAC locus was quantified by the number of CD22-CAR-positive NK cells across a range of MOIs. Data are presented as the mean values ± s.e.m. (n = 4 donors, P < 0.0001). i, Schematic of the AAV-SB system and its delivery procedure. j, Representative flow plots of human T cells transduced with the AAV-SB system. k, The integration efficiency was quantified by the number of EGFP-positive cells across a range of MOIs on day 7 after AAV transduction (n = 1 donor with three different MOIs). l,m, Head-to-head comparison of AAV6, AAV-hT7 and AAV.APR31 in primary human T cells. l, Transient scAAV-CAG-EGFP transduction efficiency (CD3+EGFP+) in primary human T cells transduced with AAV6, AAV-hT7 or AAV.APR31 at an MOI of 250. Data are presented as the mean values ± s.e.m. (n = 3 donors, AAV-hT7 versus AAV6: P < 0.0001, AAV.APR31 versus AAV6: P < 0.0001, AAV.APR31 versus AAV-hT7: P = 0.0002). m, TRAC locus site-specific knock-in efficiency of CD22-CAR in primary human T cells codelivered with CRISPR–Cas9 RNP and the indicated AAV donor template at an MOI of 250. Data are presented as the mean values ± s.e.m. (n = 4 donors, AAV-hT7 versus AAV6: P < 0.0001, AAV.APR31 versus AAV6: P < 0.0001, AAV.APR31 versus AAV-hT7: P = 0.0028). n, Correlation analysis of log2 fold changes in gene expression between AAV6-generated and AAV.APR31-generated CAR-T cells, cells were collected at two different time points (16 h versus 7 days, P < 2.2 × 10−16). In this figure, ordinary two-way ANOVAs (h), two-sided unpaired t-tests (l,m) and two-sided Pearson correlation analyses (n) were used to evaluate statistical significance. **P < 0.01, ***P < 0.001 and ****P < 0.0001.

Source data

Targeting a single tumor antigen has shown limitations in clinical settings, as some persons relapse or do not respond to CAR-T therapy because of antigen loss or downregulation12,53, such as the CD19 antigen54,55,56. Bispecific CAR-T cells can overcome this by synergistically enhancing tumor cell killing when both target antigens are present57,58,59. To evaluate the delivery efficiency of bispecific CAR in human T cells, we designed a tandem scFv construct targeting CD19 and CD20, which was packaged using either AAV6 or AAV.APR31 (Fig. 5b). Consistently, AAV.APR31 achieved superior knock-in efficiency with the CRISPR system, generating 46.1%, 68.8%, 75.6%, 78.5% and 79.5% CD19.20-CAR-positive T cells at MOIs ranging from 100 to 1,000 on day 4, even with the large knock-in fragment (2,645 bp) (Fig. 5e).

Unlike CAR-T cells, TCR-T cells recognize intracellular tumor antigens presented by MHC molecules, expanding the spectrum of targetable antigens. To explore the broader applicability of AAV.APR31, we cloned an NY-ESO-1 TCR construct along with an EGFP marker into the AAV backbone. Flow cytometry analysis indicated efficient NY-ESO-1 TCR knock-in, achieving up to 21.1% at MOI = 100 and 41.1% at MOI = 1,000 (Fig. 5f). To evaluate transgene stability, we performed flow cytometry analysis on day 7 after electroporation and AAV transduction, which showed even higher knock-in efficiencies and stable expression levels (Extended Data Fig. 4i).

Unlike T cells, NK cells are innate lymphoid cells critical for controlling viral infections and tumors60. They do not require HLA matching and typically do not cause graft-versus-host disease, making them ideal candidates for ‘off-the-shelf’ cell therapies, such as CAR-NK cells. We generated CD22-CAR knock-in NK cells with AAV.APR31 using primary human NK cells (Fig. 5a). AAV.APR31 achieved high knock-in efficiencies, yielding 10.6%, 20.8%, 34.0%, 39.3% and 48.4% CD22-CAR-positive NK cells (four donors) at MOIs ranging from 100 to 1,000 on day 5 after transduction (Fig. 5g,h). In contrast, AAV6 only achieved 5.7%, 6.6%, 9.5%, 12.7% and 13.8% CD22-CAR-positive NK cells under the same conditions (Fig. 5g,h).

To produce therapeutic immune cells efficiently, we previously developed a gene delivery system, MAJESTIC61, which enables high-quality and high-yield production of diverse therapeutic immune cell types. However, the MAJESTIC system requires high AAV MOIs to achieve optimal delivery efficiency. To assess whether AAV.APR31 could enhance this system at lower AAV doses, we packaged a SB transposon construct into AAV (AAV-SB) using either AAV6 or AAV.APR31 and then codelivered it with SB transposase mRNA by electroporation (Fig. 5i). AAV.APR31 achieved substantially higher integration efficiencies, yielding 52.5%, 59.7% and 58.7% of EGFP-positive cells at MOIs of 250, 500 and 1,000, respectively, while AAV6 only achieved 1.86%, 3.25% and 7.02% under the same conditions (Fig. 5j,k). To confirm that the observed EGFP signal reflects genuine transposon integration rather than residual episomal expression, we transduced T cells with AAV-SB-EGFP in the absence of SB transposase mRNA. Under these conditions, EGFP expression was not detectable above background on day 7 after transduction (Extended Data Fig. 4j), demonstrating that episomal AAV does not contribute to the signal at the time points used for analysis.

We benchmarked AAV.APR31 against the recently reported AAV-hT7 variant evolved from AAV6 (ref. 38) in primary human T cells using two independent readouts. For transient transgene delivery, AAV.APR31 achieved 85.17% CD3+EGFP+ cells compared to 67.67% for AAV-hT7 and 12.67% for parental AAV6 (Fig. 5l). For site-specific knock-in, AAV.APR31 yielded 67.83% CD22-CAR+ cells compared to 51.3% for AAV-hT7 and 18.35% for AAV6 (Fig. 5m). In both assays, AAV.APR31 significantly outperformed AAV-hT7, demonstrating superior T cell delivery efficiency across both transient expression and stable genomic integration modalities.

To further assess the impact of viral transduction on T cells, we performed transcriptome profiling of CAR-T cells generated using AAV6 or AAV.APR31. CD22-CAR constructs were knocked into exon 1 of the TRAC locus through Cas9 RNP electroporation, followed by AAV transduction (MOI = 1,000) in T cells from two different donors. Cells were harvested at 16 h and 7 days after transduction for RNA extraction and sequencing to capture early and late transcription alterations. Principal component analysis (PCA) revealed that samples from each donor clustered together at the early time point, irrespective of the treatment condition, indicating donor-specific transcriptomic features (Extended Data Fig. 5a). By day 7, all samples, including RNP only, RNP + AAV6 and RNP + AAV.APR31 groups, clustered together, suggesting that donor-related variability diminished over time (Extended Data Fig. 5a). Differential gene expression analysis at this later time point of day 7 showed a highly similar distribution of differentially expressed genes (DEGs) between AAV6 and AAV.APR31 groups (Extended Data Fig. 5b and Supplementary Table 4). The gene expression at two different time points (16 h versus 7 days) under AAV6 and AAV.APR31 showed consistently upregulated or downregulated genes with no discordant direction (|LFC| > 1 and q < 0.05; Fig. 5n, Extended Data Fig. 5c,d and Supplementary Table 4). At the early time point of 16 h, hierarchical clustering of top T cell function-related genes highlighted variation in activation and inhibitory-associated genes (PDCD1, CTLA4 and HAVCR2), as well as naive and memory-associated genes (TCF7, SELL, LEF1, KLF2, KLF3 and FOXO1), between the two donors (Extended Data Fig. 5e–g). By contrast, at the later time point of day 7, inhibitory-associated genes (PDCD1, LAYN, HAVCR2, LAG3, CD244, CTLA4 and TIGIT), cytotoxic genes (GZMA, GZMB, GZMK, PRF1 and TNF) and lactate metabolism-related genes (SLC2A3, SLC2A8, GAPDH, PGK1, LDHA and IDH2) showed similar expression patterns (Extended Data Fig. 5e,h,i). At both time points, samples were clustered by donor source rather than AAV serotypes by hierarchical clustering with T cell function-related gene sets. These findings indicate that CAR-T cells generated using AAV6 or AAV.APR31 exhibit similar functional and transcriptional phenotypes within the same donor, whereas interdonor variability remains a dominant source of heterogeneity, likely because of inherent genetic differences.

CD7 expression profile across human tissues and cell types

To comprehensively define the expression landscape of CD7 and assess the suitability of AAV.APR31 for more translational applications, we analyzed public bulk and single-cell RNA-seq (scRNA-seq) datasets across various human tissues and cell types. Cross-tissue bulk RNA-seq analysis using the Human Protein Atlas (HPA)62 revealed that CD7 expression was highest in lymphoid and digestive organs, including the thymus, lymph node, appendix, spleen, small intestine, tonsil and stomach (Extended Data Fig. 6a). Moderate levels were observed in the lung and liver (Extended Data Fig. 6a), whereas most nonlymphoid tissues, such as the kidney, heart, prostate, skeletal muscle, ovary and testis, exhibited negligible expression (Extended Data Fig. 6a). This expression pattern was independently validated by pseudobulk analysis in a pan-tissue single-cell atlas across different sexes, tissues and datasets63, which reproduced the enrichment of CD7 in lymphoid and digestive tissues (Extended Data Fig. 6b and Supplementary Fig. 1a).

To resolve the cellular origin of CD7 expression within these tissues, we performed gene expression and cell proportion analyses at the single-cell level. CD7 expression was restricted primarily to T lymphocytes (CD8+ and CD4+ T cells), NK cells and innate lymphoid cells (Extended Data Fig. 6c). The high expression of CD7 in immune and digestive system organs was attributed to a high proportion of these cell types (Extended Data Fig. 6d and Supplementary Fig. 1b). Cell-type-specific expression across tissues confirmed this lineage specificity, showing uniformly high expression of CD7 in CD4+ and CD8+ T cells and innate lymphoid cells across tissues, while B cells, myeloid cells, endothelial cells, epithelial cells and stromal cells consistently lacked CD7 expression (Extended Data Fig. 6e). These findings demonstrate the preferential restriction of CD7 to lymphoid lineages.

We next focused on primary lymphoid organs. In the thymus, scRNA-seq revealed distinct developmental clusters, including common lymphoid progenitor, double-negative and double-positive progenitors, as well as mature subsets such as naive CD4+ and CD8+ T cells, Treg cells and γδ T cells (Extended Data Fig. 6f). CD7 expression was detectable throughout the T cell developmental trajectory, with the strongest signals in common lymphoid progenitor, double-negative, double-positive and mature T cell subsets, as well as NK cells and innate lymphoid cells (Extended Data Fig. 6f and Supplementary Fig. 1c). In contrast, nonlymphoid cell types such as B cells, conventional dendritic cells, monocytes and macrophages displayed consistently minimal or no expression. In the bone marrow, CD7 expression was likewise confined to lymphoid clusters (T cells, innate lymphoid cells and NK cells) with negligible levels in myeloid populations (for example, neutrophils, monocytes and macrophages), mesenchymal stromal cells or B cells (Extended Data Fig. 6g and Supplementary Fig. 1d). scRNA-seq analysis of the central nervous system revealed no appreciable CD7 expression across major resident cell types, including neurons, astrocytes, microglia, oligodendrocytes and endothelial cells, aside from rare tissue-resident T cells (Supplementary Fig. 1e,f). Together, these data establish CD7 as a robust lineage-specific marker, with expression tightly restricted to T lymphocytes, their progenitors, innate lymphoid cells and NK cells across human tissues. While AAV.APR31 entry is not exclusively dependent on CD7 (as canonical AAVR-mediated pathways contribute to baseline transduction), the lineage-restricted expression of CD7 suggests preferential tropism toward T and NK cells in vivo.

To evaluate the in vivo tissue tropism of AAV.APR31 following systemic delivery, we administered firefly luciferase (FFLuc)-packaged AAV6 and AAV.APR31 intravenously in mice and monitored bioluminescence signals longitudinally using an in vivo imaging system (IVIS). AAV6 exhibited strong liver tropism, as evidenced by robust hepatic bioluminescence signals. In contrast, AAV.APR31 showed a markedly reduced liver signal, indicating partial detargeting from the liver following capsid engineering (Extended Data Fig. 7a,b). As bioluminescence imaging lacks the sensitivity to resolve transduction differences in organs with lower signal, we further quantified FFLuc mRNA by RT–qPCR across six additional tissues to obtain a more comprehensive organ-level biodistribution profile. AAV6 showed significantly elevated FFLuc mRNA in the lung compared to PBS controls, whereas AAV.APR31 exhibited significantly lower lung transduction than AAV6, demonstrating reduced off-target delivery to the lung (Extended Data Fig. 7c). In the spleen and kidney, both vectors showed comparable transduction levels with no significant difference between groups (Extended Data Fig. 7c). In the heart and brain, FFLuc mRNA levels were comparable to PBS controls in all groups, indicating negligible transduction of these tissues by either vector (Extended Data Fig. 7c). In skeletal muscle, AAV.APR31 showed a trend toward reduced transduction compared to AAV6, although this difference did not reach statistical significance (Extended Data Fig. 7c). Together, these data demonstrate that AAV.APR31 exhibits an altered tissue tropism profile relative to parental AAV6, with reduced transduction in the liver and lung while maintaining comparable levels in other tissues examined.

Preclinical antitumor efficacy of AAV.APR31-generated CAR-T cells

To evaluate the antitumor activity of CAR-T cells generated with AAV6 or AAV.APR31 under the different MOIs, we generated antigen-specific CD22-CAR-T cells by transient expression and conducted coculture assays using various bulk T cell-to-cancer cell ratios (Extended Data Fig. 7d,e). The coculture data consistently demonstrated that AAV.APR31-derived CAR-T cells mediated significantly higher cytolysis of NALM-6-GL leukemia cells compared to AAV6-derived CAR-T cells. At 24 h, the killing efficiency of AAV.APR31-derived CAR-T cells was markedly superior across all tested MOIs and effector-to-target ratios (for example, 5.8% versus 31.2% at MOI = 250, 4.3% versus 42.3% at MOI = 500 and 11.9% versus 53.5% at MOI = 1000 at a 1:2 ratio; Extended Data Fig. 7f). Similar trends were observed at 48 h, where AAV.APR31-derived CAR-T cells maintained significantly higher cytolytic activity (for example, 20.9% versus 53.3% at MOI = 250, 40.0% versus 73.7% at MOI = 500 and 44.5% versus 83.7% at MOI = 1,000 at a 1:2 ratio; Extended Data Fig. 7f). However, there was no killing discrepancy if cytolysis was normalized by the percentage of CAR-positive cells (Extended Data Fig. 7g). These differences in cytotoxicity are attributable to the higher fraction of CAR-expressing T cells achieved by AAV.APR31, as the superior transduction efficiency directly translates into greater functional output at the population level.

To further benchmark AAV.APR31 against the current clinical standard for CAR-T manufacturing, we compared the antitumor functionality of CAR-T cells generated by AAV.APR31, AAV6 and lentiviral vectors (Fig. 6a). To ensure a fair functional comparison, CAR-T cells from all three systems (Fig. 6b) were normalized to equivalent CAR-positive cell numbers before functional assessment. In vitro coculture cytotoxicity assays demonstrated comparable tumor killing activity across all three systems, with no difference in cytotoxicity among CAR-T cells derived from AAV.APR31, AAV6 and lentiviral vector at matched effector-to-target ratios (Fig. 6c and Extended Data Fig. 7h). To validate these findings in vivo, CAR-T cells manufactured by each system were evaluated in an SK-OV-3 xenograft solid tumor model (Fig. 6a). Notably, all three systems produced CAR-T cells capable of achieving complete tumor elimination, with tumor-free outcomes observed across all treatment groups (Fig. 6d,e and Extended Data Fig. 7i). These results establish that AAV.APR31-engineered CAR-T cells are functionally similar to lentiviral vector-manufactured CAR-T cells in both in vitro and in vivo settings.

Fig. 6: Preclinical antitumor evaluation of ex vivo and in vivo generated CAR-T cells.

a, Schematic overview of CD22-CAR-T cell generation with either AAV6 or AAV.APR31 and subsequent in vitro cytotoxicity assays. b, Representative flow plots of measuring CD22-CAR+ T cells. c, Cytolysis of NALM-6-GL leukemia cells in coculture assays at 24 h. Data are presented as the mean values ± s.e.m. (n = 3 replicates per donor, from three donors in total). d, Tumor bioluminescence imaging quantification in NOG mice bearing SK-OV-3-PL tumors (5 × 106 cells per mouse) following treatment with PBS or MSLN-CAR-T cells manufactured using lentiviral vector, AAV6 or AAV.APR31 (3 × 106 CAR-T cells per mouse). Luminescence signals were measured longitudinally from day 4 to day 19 after CAR-T cell administration. Data are presented as the mean values ± s.e.m. (n = 5 mice per group, PBS versus lentivirus: P < 0.0001, PBS versus AAV6: P < 0.0001, PBS versus AAV.APR31: P < 0.0001). e, Survival curves of SK-OV-3-PL tumor-bearing NOG mice treated as described in d (PBS versus lentivirus: P = 0.0044, PBS versus AAV6: P = 0.0044, PBS versus AAV.APR31: P = 0.0044). f, Schematic of the in vivo T cell delivery efficiency experiment. Human T cells (3 × 106) and scAAV-CAG-EGFP packaged with AAV6 or AAV.APR31 capsids (5 × 1011 vg per mouse) were administered intravenously into NOG mice on day 0. Spleens were isolated on day 2 for flow cytometric analysis of EGFP expression (n = 5 mice per group). g, Representative flow cytometry plots showing EGFP expression in human CD3+ T cells isolated from spleens of mice treated with PBS, AAV6 or AAV.APR31 at 48 h after injection. Numbers indicate the percentage of EGFP+ cells within the gated population. h, Quantification of EGFP+ cells within human CD3+, CD4+ and CD8+ T cell populations isolated from spleens of mice treated with PBS, AAV6 and AAV.APR31. Data are presented as the mean values ± s.e.m. (n = 5 mice per group, all q < 0.0001). i, Schematic of in vivo CAR-T generation. NOG mice bearing established NALM-6-GL leukemia were adoptively transferred with human T cells, followed by tail-vein injection of AAV6 or AAV.APR31 encoding CD22-CAR (AAV dose: 5 × 1011 vg per mouse). j, Quantification of total bioluminescence of leukemia-bearing mice. Data are presented as the mean values ± s.e.m. (n = 4 mice per group, PBS versus T cell only: P = 0.474, PBS versus T cell + AAV6: P = 0.011, PBS versus T cell + AAV.APR31: P < 0.0001, T cell only versus T cell + AAV6: P = 0.321, T cell only versus T cell + AAV.APR31: P < 0.0001, T cell + AAV6 versus T cell + AAV.APR31: P = 0.008). k, Survival curve of NALM-6-GL-induced leukemia-bearing NOG mice after T cell and AAV adoptive transfer (PBS versus T cell only: P = 0.0091, PBS versus T cell + AAV6: P = 0.0091, T cell + AAV6 versus T cell + AAV.APR31: P = 0.0266, T cell only versus T cell + AAV.APR31: P = 0.0062, T cell only versus T cell + AAV6: P = 0.9216). l, Schematic of the experimental design. NOG mice were engrafted subcutaneously with human AsPC-1 pancreatic cancer cells (day 0) and received human T cells on day 9. Mice were subsequently treated with scAAV-MSLN-CAR packaged in AAV6 or AAV.APR31 on days 9, 11 and 13. m, Tumor growth curves of tumor-bearing mice after receiving T cell and AAV adoptive transfer. Data are presented as the mean values ± s.e.m. (n = 5 mice per group, PBS versus T cell only: P = 0.8741, PBS versus T cell + AAV6: P = 0.9427, PBS versus T cell + AAV.APR31: P < 0.0001, T cell only versus T cell + AAV6: P = 0.5562, T cell only versus T cell + AAV.APR31: P < 0.0001, T cell + AAV6 versus T cell + AAV.APR31: P < 0.0001). In this figure, two-way ANOVAs with Tukey’s multiple-comparison tests (adjusted P value; d,j,m), log-rank (Mantel–Cox) tests (e,k) or two-sided unpaired t-tests (h) were used to evaluate statistical significance. *P < 0.05, **P < 0.01 and ****P < 0.0001.

Source data

To quantitatively assess in vivo transgene delivery efficiency independent of antigen-driven T cell dynamics, we performed an experiment in which human T cells were adoptively transferred into non-tumor-bearing humanized NOG mice, followed by systemic administration of scAAV-CAG-EGFP packaged with AAV.APR31 or AAV6 capsids (Fig. 6f). EGFP was selected as a reporter to enable clean, unconfounded quantification of transduction efficiency. At 48 h after AAV administration, spleens were isolated and analyzed by flow cytometry. AAV.APR31 achieved significantly higher EGFP expression in human T cells compared to AAV6 (CD3+ T cells: 45.0% versus 18.2%, CD4+ T cells: 44.9% versus 18.2%, CD8+ T cells: 50.0% versus 20.8%; Fig. 6g,h).

To evaluate the translational potential of AAV.APR31 for direct in vivo CAR-T generation, human T cells were first adoptively transferred into immunodeficient mice bearing established NALM-6-GL leukemia, followed by systemic administration of AAV.APR31 or AAV6 encoding a CD22-specific CAR (Fig. 6i). The AAV.APR31 group exhibited potent cytotoxicity, significantly reducing leukemia burden as quantified by bioluminescence imaging and leading to markedly improved survival compared to controls (Fig. 6j,k). By contrast, the AAV6 group showed minimal antitumor effects and no survival benefit, comparable to T cell only (Fig. 6j,k). To further assess its applicability in solid tumors, we established a human pancreatic cancer (AsPC-1) model and evaluated the antitumor efficacy of AAV.APR31-generated MSLN-CAR-T cells (Fig. 6l). Consistent with its superior transduction efficiency, the T cell + AAV.APR31 group markedly suppressed tumor growth compared to the T cell + AAV6 and control groups, resulting in substantially reduced tumor burden (Fig. 6m).

Discussion

Despite major advances in adoptive cell therapy1,2,3, efficient and scalable engineering of human T cells remains a formidable challenge. Primary human T cells are inherently resistant to genetic manipulation because of their quiescent nature, limited permissiveness to viral vectors and sensitivity to cytotoxic stress during ex vivo manipulation. Viral vectors such as gammaretrovirus and lentivirus pose limitations including semirandom genomic integration, whereas nonviral methods, such as mRNA, HDR DNA templates and transposon systems, offer improved safety but are often limited by low efficiency, transient expression and cellular toxicity. Directed evolution of AAV capsids has identified variants with desired tropisms in vitro and in vivo40,64,65,66,67,68,69,70,71,72,73,74,75,76. Recently, Ark313, an AAV6-derived variant with enhanced tropism for mouse T cells was reported77. However, Ark313 completely lost infectivity in human T cells compared to AAV6. In contrast, AAV.APR31 reported here enabled efficient delivery across human and mouse T and NK cells and outperformed AAV6 across a range of MOIs and transgene constructs.

Our genome-scale CRISPR screens uncovered CD7 as a lineage-specific and dominant entry enhancer for AAV.APR31 transduction, in contrast to AAV6. Moreover, CD7 overexpression in CD7-negative B lymphocytes (NALM-6) substantially increased AAV.APR31 transduction. Structural modeling further indicated that the engineered VR-IV domain centered on the APR motif in AAV.APR31 creates a novel binding interface that aligns with a complementary surface on CD7, enabling direct capsid-receptor engagement.

A major challenge in gene editing of T cells and NK cells has been achieving high knock-in efficiency without excessive viral vector dosing or cytotoxicity. AAV.APR31 appears to address this issue by mediating efficient gene delivery at MOIs nearly three orders of magnitude lower (102–103 vg per cell) than those required for AAV6 (105–106 vg per cell)31,33,34. In most cases, AAV.APR31 achieves 50–80% knock-in efficiency in an MOI range of 102–103. For instance, AAV.APR31-mediated delivery of CD22-CAR, MSLN-CAR and GD2-CAR constructs into human T cells yielded over 70% knock-in efficiency at an MOI of 1,000, while AAV6 yielded 10–18% under the same conditions. This efficiency advantage was preserved even with large cargo sizes, as demonstrated by the successful integration of a ~2.6-kb bispecific CD19.20-CAR cassette and a ~3-kb NY-ESO-1 TCR cassette.

Beyond CAR-T cells, AAV.APR31 also facilitated efficient generation of NY-ESO-1 TCR-T cells and CAR-NK cells, demonstrating its compatibility with both adaptive and innate immune cell populations. Additionally, the integration of AAV.APR31 into our previously reported MAJESTIC immune cell engineering system further improved the efficiency of transposon-based gene delivery. Although simultaneous NK cell transduction could be considered off-target in CAR-T applications, we argue that this is more likely advantageous. NK cells provide rapid, antigen-independent cytotoxicity that complements antigen-specific CAR-T responses and their coengineering may enhance antitumor efficacy particularly in settings of antigen loss or tumor heterogeneity.

CAR-T cell generation in vivo has proved challenging. Lipid nanoparticles78,79,80 and lentiviral-based vectors81,82,83,84 have been explored for this purpose but issues such as liver toxicity and high immunogenicity still pose notable barriers. The AAV.APR31 capsid is optimized to engage CD7, a receptor selectively expressed in T cells, NK cells and their progenitors, with negligible expression in the reproductive and central nervous systems, as well as most nonhematopoietic tissues, which may minimize the off-target risks. Consistent with this expression pattern, transduction of CD7-negative primary human immune cells, including B cells and monocytes, was comparably low between AAV.APR31 and AAV6, with no enhancement in CD7-negative populations. Furthermore, in vivo biodistribution analysis in mice following systemic administration revealed reduced transduction in the liver and lung relative to AAV6, with no significant difference in the heart, brain or skeletal muscle. However, nonlymphoid tissues could be transduced through the canonical AAV receptor AAVR, identified in our CRISPR screen and by SPR as binding AAV.APR31. Thus, AAV.APR31 operates through a layered-entry mechanism; a retained AAVR pathway provides a CD7-independent baseline throughout the body, upon which CD7 confers a lineage-restricted enhancement in T and NK cells. Therefore, we interpret the in vivo profile as favorable lineage-restricted detargeting relative to AAV6 rather than absolute organ-level specificity. We demonstrated that AAV.APR31 enables both transient transgene expression and permanent site-specific CAR integration by HDR, offering versatility for different applications. Systemic administration of AAV.APR31 encoding a CD22-CAR or MSLN-CAR generated potent CAR-T cells in vivo that effectively controlled aggressive leukemia or pancreatic cancer and prolonged overall survival, results not observed with parental AAV6. These results support further research on AAV.APR31 for potential clinical applications. Beyond oncology, AAV.APR31 may contribute to CAR-T or CAR-NK therapy for autoimmunity85,86,87,88,89, inflammation and fibrosis, for example, cardiac fibrosis, by selectively eliminating activated fibroblasts, which has shown evidence in reversing pathological remodeling and improving heart function78.

Concurrent with our study, two independent groups reported engineered AAV variants for in vivo T cell targeting. Lu et al. described AAV6-M2, which achieves CAR expression in human T cells in a humanized mouse model through a VR–VIII–CD62L interaction mechanism90. However, the specific capsid mutations are undisclosed, precluding independent validation or direct comparison. Nyberg et al. independently evolved an AAV6 capsid library under selective pressure from human serum neutralizing antibodies, identifying AAV-hT7, an APR motif-containing variant, with CD7 as the key host receptor for T cell transduction38. Our selection strategy, which prioritized direct T cell binding affinity rather than antibody evasion, independently converged on the same APR motif enrichment and CD7 dependency, providing mutual scientific validation of this receptor axis as a robust principle for T cell-targeted AAV engineering. However, the two studies differ substantially in several important respects. Among the top 200 enriched variants, our screen recovered 165 APR-containing capsids compared to 70 reported by Nyberg et al., suggesting that T cell-directed selection more efficiently drives convergence toward CD7-engaging architectures. Notably, despite this quantitative difference, the specific APR-containing sequences identified in the two studies share no sequence overlap whatsoever (Supplementary Fig. 2), demonstrating that independent evolutionary trajectories under distinct selective pressures can converge on the same functional motif through entirely different sequence solutions. Furthermore, whereas Nyberg et al. focused primarily on human T cell transduction, our study extends validation to human NK cells and mouse T and NK cells, demonstrating that CD7-dependent tropism is conserved across lymphoid lineages and species. Lastly, in a direct head-to-head comparison under matched conditions, AAV.APR31 achieved higher T cell transduction efficiency than AAV-hT7 (Fig. 5l,m), consistent with a selection strategy optimized for direct cellular engagement. Together, this orthogonal convergence on the APR–CD7 axis across independent studies, combined with the distinct variant landscapes, broader biological validation and superior functional performance reported here, establishes the APR motif as a generalizable structural module for immune-cell-targeted AAV engineering.

We note a limitation regarding direct benchmarking against other in vivo delivery systems. Although AAV.APR31-engineered cells are functionally equivalent to lentiviral-manufactured CAR-T cells and AAV.APR31 outperforms AAV6 both in vitro and in vivo, we did not perform head-to-head in vivo comparisons against fundamentally different modalities such as enveloped delivery vehicle-based systems. However, the relevant delivery component can be compared in a controlled manner; in site-specific integration systems, including that of Nyberg et al., an AAV serves as the DNA donor vector and, under matched in vitro conditions in primary human T cells, AAV.APR31 is a more efficient donor vector than AAV-hT7. Because the AAV-hT7-based system has been shown to outperform lentivirus in vivo, these data indirectly suggest that an AAV.APR31-based system would perform comparably, although in vivo determinants beyond donor delivery efficiency were not assessed here. The principal advantage of AAV.APR31 does not only depend on such combinations, as it achieves direct in vivo CAR-T generation and tumor control through single-vector systemic administration. Systematic cross-system benchmarking remains an important direction for future work.

Methods

Mouse model

Animal care and experimental procedures adhered to the guidelines established by the Nanjing University and Nanjing Medical University Institutional Animal Care and Use Committee (IACUC; IACUC-2605005-1, IACUC-2604019). Mice 8–12 weeks of age were used in this study. NOD.Cg-Prkdcscid Il2rgtm1Sug/JicCrl (NOG) mice were purchased from Charles River and maintained in-house for assessing T cell-based antitumor therapeutic efficacy. Mouse health was monitored daily after tumor induction. C57BL/6J mice were kindly provided by L. Huang (Institute of Modern Biology, Nanjing University) for T and NK cell isolation. All mice were housed under standard conditions with a 12-h light–dark cycle (7:00 a.m. to 7:00 p.m. light phase).

Cell culture

The 293T (CTCC-001-0188), NALM-6 (CRL-3273) and AsPC-1 (CTCC-400-0260) cell lines were purchased from commercial sources (American Type Culture Collection and MeisenCTCC). K-562 (SCSP-5054), Jurkat (SCSP-513) and SK-OV-3 (TCHu185) cell lines were kindly provided by the Stem Cell Bank, Chinese Academy of Sciences. NALM-6-GL (GFP+ and luciferase+) cells were generated by transducing NALM-6 cells with GFP–luciferase lentivirus first and then purified by enriching for GFP+ cells by fluorescence-activated cell sorting. SK-OV-3-PL (puromycin+ and luciferase+) cells were generated by transducing SK-OV-3 cells with puromycin–luciferase lentivirus first and then purified by enriching for puromycin+ cells by drug selection. The 293T cells were cultured in DMEM (Gibco) supplemented with 10% FBS (Hyclone or Gibco) and 1% penicillin–streptomycin, hereafter referred to as D10. NALM-6, NALM-6-GL, K-562, AsPC-1 and Jurkat cells were cultured in RPMI-1640 (Gibco) medium supplemented with 10% FBS and 100 U per ml penicillin–streptomycin.

Human PBMCs were purchased from Milestone Biotechnologies or HYCELLS and cultured in X-VIVO 15 medium (Lonza) supplied with 5% FBS (Hyclone), 50 μM β-mercaptoethanol, 100 U per ml human IL-2 (Novoprotein), 5 ng ml−1 of IL-7 (Novoprotein) and 5 ng ml−1 of IL-15 (Novoprotein) at a density of 1 × 106 cells per ml. NK cells were cultured in CTS NK-Xpander medium (Gibco) supplied with 5% FBS (Hyclone) and 500 U per ml human IL-2 (Novoprotein). T cells were activated with ImmunoCult human CD3/CD28/CD2 T cell activator (StemCell) at a concentration of 25 μl ml−1 X-VIVO 15 medium. Primary human CD14+ monocytes were purchased from Milestone Biotechnologies and cultured in RPMI-1640 supplemented with 10% FBS, 1% penicillin–streptomycin, 2 mM L-glutamine and 50 ng ml−1 recombinant human GM-CSF (Novoprotein). Primary human B cells were purchased from Milestone Biotechnologies and cultured in RPMI-1640 supplemented with 10% FBS, 1% penicillin–streptomycin, 2 mM L-glutamine, 49 µM β-mercaptoethanol and 50 ng ml−1 recombinant human BAFF (Novoprotein). In this study, multiple healthy donors were involved in various experiments; the donor information is indicated in the figure legends.

Mouse T and NK cells were isolated from the spleens of C57BL/6J mice using the EasySep mouse T cell isolation kit and the EasySep mouse NK cell isolation kit (StemCell Technologies), respectively, according to the manufacturer’s protocols. Mouse T cells were cultured at 1 × 106 cells per ml, the culture plate was precoated with 2 μg ml−1 anti-CD3ε antibody (BioLegend) for 1 h at 37 °C and RPMI-1640 medium was supplemented with 10% FBS, 1% penicillin–streptomycin, 2 ng ml−1 IL-2, 1 μg ml−1 anti-CD28 (BioLegend), 2.5 ng ml−1 IL-7 and 50 ng ml−1 IL-15. Mouse NK cells were cultured at a density of 1 × 106 cells per ml in RPMI-1640 medium supplemented with 10% FBS, 2 mM L-glutamine, 1% penicillin–streptomycin, 49 mM β-mercaptoethanol, 50 ng ml−1 IL-2 and 50 ng ml−1 IL-15. Cytokines were purchased from Novoprotein or PeproTech.

Constructs

Plasmid used for transiently expressing EGFP or CAR constructs were based on a scAAV backbone where EGFP was driven by a CAG promoter, scAAV-CAG-EGFP (Addgene, 83279). To site-specifically knock-in EGFP, CD22-CAR91, MSLN-CAR92, GD2-CAR (patent no. US2019/048085A1), CD19.20 bispecific CAR93 and NY-ESO-1 TCR94 constructs into the TRAC exon 1 locus, we generated various plasmids: pLY154 (AAV-TRAC-LHA-EFS-EGFP-RHA), pLY061 (AAV-TRAC-LHA-EFS-CD22-CAR-RHA), pLY119 (AAV-TRAC-LHA-EFS-MSLN-CAR-RHA), pLY150 (AAV-TRAC-LHA-EFS-GD2-CAR-RHA), pLY161 (AAV-TRAC-LHA-EFS-CD19.20-CAR-RHA) and pLY162b (AAV-TRAC-LHA-EFS-EGFP-T2A-NY-ESO-1-RHA). All constructs were individually cloned into an AAV plasmid containing a pair of homologous arms (left HA and right HA) that targeted the TRAC exon 1 locus39.

To knock in the CD22-CAR construct into the TRBC1 locus, two different plasmids were constructed by replacing TRAC-targeting LHA and RHA sequences from the AAV-LHA-EFS-CD22-CAR-RHA plasmid with TRBC1-targeting homologous arms. Finally, we got plasmid pLY158 (AAV-TRBC1-LHA-EFS-CD22-CAR-RHA). For the GAPDH locus BFP knock-in, we constructed plasmid pHW011a (AAV-LHA-recoded exon 9-P2A-BFP-3′UTR-RHA) using a similar strategy to that previously described95.

For the AAV-SB-mediated gene delivery, we used previously developed MAJESTIC technology to deliver EGFP constructs. The same cloning strategy and detailed delivery protocols were performed as described in our previous work61.

To generate the AAV-hT7 capsid, the VR-IV of the parental AAV6 capsid was modified by replacing the native heptapeptide sequence 454GSAQNKD460 with the APR motif-containing sequence 454HAPRVEE460, as described by Nyberg et al.38. Briefly, site-directed mutagenesis was performed on the AAV6 capsid plasmid using overlapping PCR. The resulting plasmid was verified by Sanger sequencing across the entire VR-IV insertion region to confirm the correct amino acid substitution and the absence of unintended mutations. The verified AAV-hT7 capsid plasmid was then used for AAV vector production following the same protocol as described for AAV.APR31.

AAV production

The 293T cells were prepared as described above. D10 medium was replaced with 13 ml of prewarmed, FBS-free DMEM. For each dish, cells were transiently transfected with a mixture containing 5.2 μg of transfer plasmid, 8.9 μg of RepCap plasmid and 10.4 μg of pDF6 plasmid, premixed with 130 μl of polyethylenimine (1 mg ml−1) in 450 μl of Opti-MEM. Then, 6 h after transfection, 7 ml of prewarmed D10 medium was added to each dish. Then, 72 h after transfection, the transfected cells were dislodged and collected for AAV purification96. Viral titers were quantified using RT–qPCR with a pair of primers targeting the specific sequence in the AAV vector.

AAV transduction

Human T cells were activated with ImmunoCult human CD3/CD28/CD2 T cell activator (StemCell) for 48 h. After activation, the T cells were resuspended in fresh T cell medium and seeded at a density of 1 × 106 cells per ml. AAV transduction was typically performed on days 2–5 after activation of T cells. Before AAV addition, all AAV groups (raw virus) were normalized to the same titer using complete X-VIVO 15 medium to ensure consistent MOI and volume across all groups. Unless otherwise specified, the AAV-containing medium was replaced after 16 h of viral incubation.

Lentivirus production and transduction

For lentiviral packaging, low-passage HEK293T cells were used. Approximately 2 × 107 HEK293T cells were seeded per 150-mm dish 1 day before transfection. Before transfection, the D10 medium was carefully replaced with 13 ml of prewarmed Opti-MEM (Invitrogen). The cells from each 150-mm dish received a transfection mixture containing 20 μg of transgene plasmid, 15 μg of psPAX2 (Addgene), 10 μg of pMD2.G (Addgene) and 90 μL of Lipo293 transfection reagent (Beyotime) in 450 μl of Opti-MEM. This mixture was vortexed briefly and incubated at room temperature for 10–15 min before being added to the cells. Then, 6 h after transfection, 13 ml of Opti-MEM medium was replaced with 20 ml of prewarmed D10 medium. Viral supernatant was harvested 48 h after transfection and purified with 10% PEG8000 (Promega) precipitation by mixing raw viral supernatant with 40% PEG8000 solution; then, the mixture was incubated at 4 °C for 6 h. The lentivirus was centrifuged at 2,000g for 30 min, resuspended with cold PBS or medium, then aliquoted and stored at −80 °C.

AAV6 capsid library generation

To generate the AAV6 capsid library, we performed saturated mutagenesis of the VR-IV region (454-GSAQNKD-460) of the AAV6 capsid as previously reported40. To perform a saturation mutagenesis at the VR-IV region, we randomly synthesized an oligo as NNSNNSNNSNNSNNSNNSNNS and then used PCR to introduce the library. The PCR1 oligos were 5′-CCTGTATTACCTGAACAGAACTCAAAATCAGTCCNNSNNSNNSNNSNNSNNSNNSTTGCTGTTTAGCCGTGGGTC-3′ and 5′-GGAATTTTGGCCCAAATGGGAC-3′; the PCR2 oligos were 5′-GCACGGTTCAAGTCTTCTCGGAC-3′ and 5′-GGACTGATTTTGAGTTCTGTTCAGGTAATACAGG-3′. PCR1 and PCR2 were performed using a plasmid template (AAV6-VR-IV-Stop) that prereplaced the GSAQNKD motif with seven tandem stop codons to prevent amplification of wild-type sequences. Next, we cloned the pLY037a (P5-Rep_deadGFP-Cap) plasmid as a library cloning backbone to ensure no residual functional wild-type capsid after double enzymatic digestion. We digested pLY037a with Pfl23II (BsiWI) and SdaI (Sbf I) (Thermo Scientific) enzymes, then conducted gel purification and combined PCR1 and PCR2 amplicons in an equimolar ratio. The one-step assembly of PCR inserts and plasmid backbone was performed using the NEBuilder HiFi DNA Assembly Master Mix (New England Biolabs). The assembled products were then purified and precipitated with two volumes of ethanol. The purified products were electroporated into Endura electrocompetent cells (LGC Biosearch Technologies) and the electroporated competent cells were recovered in recovery medium for 1 h before being plated into agar plates (245 mm × 245 mm). The library size was evaluated by performing a series of dilutions of recovered cells. We pooled all single colonies from agar plates and did plasmid DNA preparation using a NucleoBond Xtra Midi EF kit (Macherey-Nagel). All oligonucleotides used for PCR in this study were synthesized by Tsingke.

To produce the capsid library virus, low-passage 293T cells were cotransfected with 19.25 μg of pDF.6 helper and 8.75 μg of pLY038a (P5-Rep_CapLib-IV) library plasmids per 150-mm plate. The general transfection, medium change and viral purification were similar to those described above, except that the culture medium was collected on days 3 and 4 and the cells were collected on day 4 for AAV extraction and purification. The virus from the medium and cells was pooled together at the end.

Directed AAV evolution

Activated human T cells were transduced with the pooled capsid library virus at an MOI of 1,000 vg per cell (500 vg per cell for the third round of selection) and then incubated in a 37 °C CO2 incubator for 6 h. Unbound AAVs were removed by triple washing of T cells with PBS and total cellular genomic DNA and viral DNA were extracted from the PBS-washed T cells using a QIAamp DNA blood kit (Qiagen). The extracted DNA mixture was used as the PCR template for the second round of library cloning. The enriched capsid region was PCR amplified and reassembled into the digested pLY037a backbone as described above. The protocols for the second and third rounds of capsid library cloning, Endura competent cell electroporation, AAV packaging and purification, T cell activation, viral transduction and DNA extraction were the same as described above.

AAV6 capsid library readout and sequencing

Three rounds of evolved AAV capsid pools and parental library (virus pool before selection) were used for readout and next-generation sequencing (NGS). The parental library was first digested with DNase I (37 °C for 30 min, 75 °C for 10 min) and proteinase K (50 °C for 60 min, 95 °C for 20 min); then, all samples were subjected to PCR readout, including the digested parental library and DNA mixtures extracted from the first, second and third rounds of selection. Two rounds of PCR were performed to capture and barcode the target capsid region. For the round 1 PCR, primer sets AAV6-Cap_F2 (5′-GCACGGTTCAAGTCTTCTCGGAC-3′) and AAV6-Cap_R1 (5′-GGAATTTTGGCCCAAATGGGAC-3′) were used; the PCR program was 98 °C 30 s for one cycle, followed by 18 cycles of 98 °C for 10 s, 60 °C for 5 s and 72 °C for 10 s, with a final extension at 72 °C for 2 min. For the round 2 PCR (barcoding PCR), primer sets IV-BC-F and IV-BC-R (Supplementary Table 5) were designed for NovaSeq sequencing and demultiplexing (containing Illumina adaptors and unique indices) and round 1 PCR products were used as templates. The PCR program was 98 °C for 30 s for one cycle, followed by 15 cycles of 98 °C for 10 s, 60 °C for 5 s and 72 °C 5 s, with a final extension at 72 °C for 2 min. All PCR reactions were conducted using 2× Phanta Flash Master Mix (Vazyme) and amplicons were run on a 2% agarose gel and gel-purified using a QIAquick gel extraction kit (Qiagen). Samples were quality-verified and quantified at an equimolar ratio before pooling. The pooled library was sequenced by the NovaSeq X Plus platform.

AAV6 capsid screening data analysis

For the raw reads with FASTQ format, Cutadapt (version 4.4)97 was used to trim the adaptors of the paired-end sequence. Next, filtering the reads with low quality was performed by Sickle version 1.33 (Fass. Sickle: a sliding-window, adaptive, quality-based trimming tool for FastQ files, version 1.33; https://github.com/najoshi/sickle) using the argument -t 25. Then, 5′ and 3′ anchor sequences flanking the 21-bp VR were trimmed by Cutadapt. Lastly, only the 21-bp reads concordant in both read 1 and read 2 with no N base were kept and translated into an amino acid sequence for subsequent analysis. The different variants with synonymous DNA sequences encoding for the same amino acid sequence were merged.

Statistically, the number of each variant in each round was normalized to the total size of each sample and this value was used as the first indicator to quantify the absolute number of each variant, which was then expressed as a percentage. To remove the bias of oligo synthesis and assess the expansion ability of variants, we divided the percentage of each variant in each round by the parental percentage and termed the fold change. The mean of the percentage and fold change of each variant between two donors was calculated by arithmetic mean and geometric mean, respectively. These two indicators were used to evaluate potential AAV variant candidates.

Flow cytometry

For the transient EGFP detection after scAAV-CAG-EGFP transduction, T cells were collected and PBS-washed on day 2 after viral transduction; then, T cells were stained with antibodies targeting T or NK cell lineage markers in MACS buffer (0.5% BSA and 2 mM EDTA in PBS) for 30 min on ice. For the CD22-CAR-T cell measurement, T cells were incubated with CD22-Fc protein (Novoprotein) in PBS for 30 min on ice, then stained with anti-human IgG Fc and other antibodies (such as anti-human CD3) and incubated on ice for 30 min. For CAR constructs containing a Flag tag, CAR-T cells were stained directly with an anti-DYKDDDDK tag antibody. If T cells need to be stained with other antibodies, a cocktail of antibodies could be prepared first and then added to the T cells. Cells were washed with MACS buffer after staining and all samples were analyzed on a BD LSRFortessa cell analyzer or BD FACS Aria III cytometer. The representative gating strategy for flow cytometry, as shown in Supplementary Figs. 3 and 4. Flow cytometry data were analyzed using FlowJo (version 10.8.1). All flow cytometry antibodies were purchased from BioLegend or BD Biosciences.

Viral binding and internalization detection

Human T cells were activated with ImmunoCult human CD3/CD28/CD2 T cell activator for 48 h. For the viral binding assay, the cells were precooled at 4 °C for 30 min to inhibit cellular uptake before infection. The cooled cells were then exposed to AAV6 or AAV.APR31 packaged virus (containing CD22-CAR expression cassette) at specific MOIs for 2 h at 4 °C to allow viral binding without internalization. Unbound virions were removed by washing the cells three times with ice-cold PBS. Viral and cellular DNA were subsequently extracted using the genomic DNA extraction kit (Qiagen). For the viral uptake detection, activated T cells were directly incubated with AAV at specific MOIs in an incubator for 2 h (37 °C, 5% CO2) to enable the uptake of bound virions. Following incubation, the cells were treated with 0.05% TrypLE Express (Gibco) for 5 min and then triple-washed with ice-cold PBS to remove any uninternalized virus. DNA extraction was then performed for both viral and cellular components; then, 25 ng per reaction of DNA was used for RT–qPCR analysis to quantify the viral copy number in each cell (forward: 5′-GCAGAGCGCACATCGCCCAC-3′; reverse: 5′-AAGGCGGAGCCAGTACACGA-3′) and a housekeeping gene RPPH1, known to have two copies per cell, was used as an internal control (forward: 5′-AGCTGAGTGCGTCCTGTCACT-3′; reverse: 5′-TCTGGCCCTAGTCTCAGACCTT-3′)98.

Nucleofection

All electroporation experiments were conducted using a Neon NxT electroporation system (Thermo Fisher). For a 10-μl electroporation reaction, 5 × 105 activated T cells or NK cells were resuspended in 10 μl of genome-editing buffer and combined with preprepared Cas9 RNP. The RNP was prepared by mixing 700 ng of sgRNA with 0.5 μl of Cas9 protein (Integrated DNA Technologies) (0.3 μl of 61 mM Cas9 protein + 0.2 μl buffer R) and the mixture was incubated at room temperature for 15 min. The RNP dose could be scaled up according to the cell number. The 10-μl mixture of cells and RNP was then loaded into the Neon pipette without any bubbles. The electroporation parameters were set to 1,600 V and 10 ms for three pulses. After electroporation, the cells were immediately transferred to a 48-well plate containing prewarmed medium. The HDR donors were delivered using AAV by adding them into electroporated T cells within 1 h after electroporation; specific MOIs were used depending on specific assays. All sgRNAs were synthesized by GenScript or Tsingke and their sequences are listed in Supplementary Table 5.

In vitro mRNA preparation

The SB transposase mRNA was commercially synthesized using in vitro transcription by CYNBIO. The mRNA was modified with pseudo-U, capped with Cap1, polyadenylated and then subjected to silica membrane purification. The mRNA was aliquoted and stored at −80 °C.

Bulk RNA-seq and data processing

As described above, human T cells were activated with ImmunoCult human CD3/CD28/CD2 T cell activator for 48 h. On day 4 after T cell activation, T cells were collected for sgTRAC/Cas9 RNP electroporation and pLY061 AAV transduction. T cells were collected at 16 h and 7 days after AAV transduction for RNA extraction and RNA-seq. RNA-seq was performed on the Illumina platform and purity-filtered sequencing reads were first trimmed using Cutadapt (version 4.4)97 to remove adaptor sequences. The trimmed reads were then aligned to the human reference genome (GRCh38) using STAR (version 2.5.2b)99. Gene-level read counts were obtained with featureCounts (version 2.0.8)100. Gene expression values were calculated by log2-transforming the FPKM values derived from raw counts using a custom script. PCA was performed using factoextra (version 1.0.7) and 95% confidence intervals were plotted for each group. Heat maps of selected T cell-related genes were generated using the Pheatmap package (version 1.0.12). Differential expression analysis was carried out using DESeq2 (version 1.42.0)101 with the design formula ‘donor + group’ to identify DEGs at the 16-h and 7-day time points. The T cell functional modules used in the heatmap were from Chu et al.102. The Benjamini–Hochberg method was used to correct for multiple testing. The Jaccard index was used to quantify the overlap between DEGs in different AAV serotypes and Cas9-only groups. All data processing and visualization were performed using the tidyverse toolkit (version 2.0.0).

Genome-wide CRISPR screens

To identify essential targets that assist the infection of human T cells by AAV.APR31 variant, a genome-wide CRISPR KO screen was performed in Jurkat cells (immortalized human T lymphocytes). In the meantime, a parallel CRISPR screen was carried out in NALM-6 cells (human B lymphocytes), serving as a lineage control. Jurkat and NALM-6 cells were transduced with Cas9-Blast lentivirus to generate stable Cas9-expressing cell lines, Jurkat-Cas9 and NALM-6-Cas9 cells for the screening experiment. The Brunello sgRNA library (Addgene, 73178) was amplified using Endura Electrocompetent Cells (Endura) according to the manufacturer’s protocol. Electroporation was carried out using a Gene Pulser Xcell electroporation system (Bio-Rad) in a 0.1-cm-gap cuvette (Bio-Rad). The Brunello sgRNA library includes 76,441 gRNAs targeting 19,114 genes, along with 1,000 nontargeting controls. The lentivirus was packaged as described above. The screening was conducted in triplicate, with each replicate maintaining over 750-fold coverage to ensure sufficient representation of each sgRNA. Then, 5 days after Brunello lentiviral transduction, Jurkat-Cas9 and NALM-6-Cas9 cells were transduced with scAAV-CAG-EGFP at MOIs of 1 × 104 and 5 × 104, respectively. After 16 h of viral incubation, the medium was replaced. At 48 h after AAV transduction, Jurkat-Cas9 and NALM-6-Cas9 cells were collected and resuspended in MACS buffer for cell sorting. Specific cell populations were sorted on the basis of EGFP expression. A total of 6.5 × 107 cells from each replicate were used for sorting, which was performed using a BD FACS Aria cell sorter.

Genomic DNA was extracted from sorted cells and cell pellets using the QIAamp DNA blood mini kit (Qiagen). The sgRNA library readout was conducted in two rounds of PCR. For PCR1, approximately 2 μg of genomic DNA per 50-μl PCR reaction was used, with each sample undergoing six independent PCR reactions. A region containing the entire sgRNA cassette was amplified using specific primers: readout-PCR1-F (5′-aatggactatcatatgcttaccgtaacttgaaagtatttcg-3′) and readout-PCR1-R (5′-gagccaattcccactcctttcaagac-3′). PCR1 products from each biological sample were pooled and used as templates for PCR2 (barcoding PCR). Each biological sample was barcoded separately with a unique combination of barcode primers (Supplementary Table 5). All PCRs were performed using 2× Phanta Flash master mix (Vazyme). The thermocycling parameters were as follows: 98 °C for 10 s, followed by 20 (PCR1) or 28 (PCR2) cycles of 98 °C for 5 s, 62 °C for 5 s and 72 °C for 10 s, with a final extension at 72 °C for 1 min. The PCR2 products were gel-purified using the QIAquick gel extraction kit (Qiagen), then quantified and pooled. Amplicon libraries were spiked with 5–20% PhiX and sequenced on the Illumina NovaSeq X Plus platform.

Analysis of pooled CRISPR screen data

Raw FASTQ files were quality-filtered using fastp103,104 with the parameter ‘--qualified_quality_phred 20’, followed by adaptor trimming using Cutadapt. To remove 3′ and 5′ adaptors flanking the sgRNA spacer sequences, the following settings were applied: ‘-e 0.2 20 -g GGCTTTATATATCTTGTGGAAAGGACGAAACACCG -a GTTTTAGAGCTAGAAATAG’. After trimming, only the 20-bp sgRNA spacer sequences were retained for subsequent analysis. Reads were then quantified using MAGeCK (version 0.5.9.5)105 count with parameters ‘--sgrna-len 20--norm-method none’. Raw read counts were normalized to the total read number of each sample. Normalized counts from EGFPhigh and EGFPlow bins were compared using MAGeCK test with options ‘-norm-method none’, ‘-control-sgrna’ and ‘-paired’, pairing samples from the same sorting experiment and using nontargeting sgRNAs as controls. Genes were considered significant hits when the median absolute log2 fold change was >0.5 and the false discovery rate was <0.05. Functional enrichment of DEGs was assessed using clusterProfiler (version 4.2.2)106 with default parameters to identify enriched Gene Ontology terms.

CD7 preblocking

Jurkat cells or activated human T cells (5 × 105 cells) were incubated with anti‑human CD7 antibody (Biolegend or Elabscience) at specific concentrations at 37 °C for 30 min, washed twice with PBS and then seeded into 48‑well plates. Cells were transduced with scAAV-CAG‑EGFP (Jurkat, MOI = 1,000; human T cells, MOI = 10,000) at 37 °C for 16 h. The medium was then replaced and cultures were continued for 32 h before measuring EGFP expression by flow cytometry.

AAVR PKD1–PKD5 domain expression and purification

The coding sequence of the extracellular PKD1–PKD5 domain of human AAVR was cloned into the pET-28a(+) expression vector with a C-terminal 6×His tag (plasmid pLYSB032). The construct was transformed into Rosetta2(DE3) competent cells and plated on Luria–Bertani (LB) agar containing kanamycin (50 μg ml−1) at 37 °C for 12–16 h. A single colony was inoculated into 5 ml of LB medium and cultured overnight at 37 °C with shaking. The overnight culture was diluted 1:100 into fresh LB medium and protein expression was induced with 1 mM IPTG when the optical density at 600 nm reached 0.6–0.8. Cells were cultured at 18 °C for 16 h with shaking and then harvested by centrifugation at 8,000g for 10 min at 4 °C; cell pellets were stored at −80 °C until purification.

All purification steps were carried out on ice. Cell pellets were resuspended in lysis buffer (20 mM NaH2PO4, 500 mM NaCl and 10 mM imidazole, pH 7.4) and lysed by sonication (300 W, 2 s on, 2.5 s off, 30 min) in an ice-water bath. The lysate was clarified by ultracentrifugation at 20,000g for 30 min at 4 °C, passed through a 0.22-μm filter and incubated with Ni-NTA Sepharose resin overnight at 4 °C with gentle agitation. The resin was washed sequentially with wash buffer 1 (20 mM NaH2PO4, 500 mM NaCl and 20 mM imidazole, pH 7.4) and wash buffer 2 (20 mM NaH2PO4, 500 mM NaCl and 40 mM imidazole, pH 7.4). His-tagged AAVR protein was eluted with elution buffer (20 mM NaH2PO4, 500 mM NaCl, 300 mM imidazole and 5% glycerol, pH 7.4) and dialyzed against storage buffer (20 mM HEPES and 20 mM NaCl, pH 7.4). The protein was concentrated using an Amicon Ultra centrifugal filter unit (30-kDa molecular weight cutoff), passed through a 0.22-μm filter, aliquoted and stored at −80 °C until use. Protein purity was confirmed by SDS–PAGE.

AAV neutralization assay

scAAV-CAG‑EGFP was preincubated with recombinant soluble human AAVR PKD1–PKD5 domain protein at the indicated concentrations on ice for 30 min before transduction. Recombinant EGFP protein (1 μM) was included as a negative control to confirm that inhibition was specific to AAVR-AAV interaction. Pretreated AAV was then added to Jurkat cells (5 × 105 cells per well in a 48-well plate) at an MOI of 1,000 and incubated at 37 °C. Culture medium was replaced after 16 h and EGFP expression was quantified by flow cytometry 48 h after AAV addition.

SPR analysis of binding

SPR analyses were carried out using a Biacore T200 instrument (GE Healthcare) at 25 °C with a flow rate of 30 μl min−1 in PBS running buffer. AAV6 or AAV.APR31 viral particles were diluted in sodium acetate buffer (pH 4.5) and immobilized on a CM5 sensor chip by standard amide coupling chemistry. Serial dilutions of recombinant soluble AAVR PKD1–PKD5 domain protein were injected as the analyte. Binding kinetics and equilibrium dissociation constants (KD) were determined using the Biacore T200 BIA evaluation software (GE Healthcare).

CD7 expression analysis in human tissues and cell types

HPA bulk RNA-seq data62 were used to evaluate CD7 expression at the bulk level across various normal human tissues. The log1p-normalized tags per million values were used for visualization. For single-cell analysis, data from the cross-tissue multicellular single-cell atlas63 were accessed to determine CD7 expression across different tissues and systems. Pseudobulk expression values were calculated by averaging the log1p counts per million across cells for each cell type. To investigate CD7 expression in the bone marrow, thymus, and brain at the single-cell level, scRNA-seq data were downloaded from the DISCO database107. Data analysis and visualization were performed using Scanpy (version 1.10.1)108 and Seaborn (version 0.13.2)109 in Python.

Vector copy-number quantification

CAR-T cells from three independent donors were isolated on day 14 after Cas9 RNP electroporation and AAV transduction and washed with PBS. Genomic DNA was extracted using the QIAamp DNA blood mini kit (Qiagen). CAR copy number was determined by qPCR using primers targeting the EFS promoter (forward: 5′-gcagagcgcacatcgcccac-3′; reverse: 5′- aaggcggagccagtacacga-3′). The housekeeping gene RPPH1 (two copies per diploid genome) was used as an internal reference (forward: 5′-AGCTGAGTGCGTCCTGTCACT-3′; reverse: 5′-TCTGGCCCTAGTCTCAGACCTT-3′). Each 20-µl reaction contained 10 µl of NovoStart SYBR qPCR SuperMix plus (low ROX premixed) (Novoprotein), 8 µl of diluted genomic DNA (1.25 ng μl−1) and 1 µl of each primer. Thermocycling conditions were as follows: 94 °C for 3 min, followed by 40 cycles of 94 °C for 10 s and 58 °C for 30 s. The CAR copy number per genome was calculated using the ΔΔCt method normalized to RPPH1.

Kill assays

To interrogate CAR-T cell killing efficacy, 1 × 105 NALM-6-GL cells (GFP+ and luciferase+) were seeded per well in 96-well plates, with three technical replicates per MOI condition. CD22-CAR-T cells were then added in a total volume of 200 μl of T cell medium according to various effector-to-target ratios (CAR-T cell to cancer cell or bulk T cell to cancer cell, depending on specific experiments). Cytolysis was quantified at 24 h and 48 h after cocultures using luciferase-based bioluminescence. Specifically, 150 μg ml−1 D-luciferin (Yeasen) was added to each well with a multichannel pipette and luminescence was measured using a Synergy H1 multimode reader (BioTek).

In vivo antitumor assessment of CAR-T cells generated by different delivery approaches

To evaluate the antitumor efficacy of CAR-T cells generated using different gene delivery systems (lentivirus, AAV6 and AAV.APR31), female NOG mice were subcutaneously inoculated with 5 × 106 SK-OV-3-PL cells. Then, 2 weeks after tumor implantation, tumor-bearing mice received 3 × 106 of MSLN-CAR-T cells through tail-vein injection. Tumor progression was longitudinally monitored by bioluminescence imaging using an IVIS. Briefly, mice were administered 100 μl of 150 μg ml−1 D-luciferin before imaging and total photon flux was quantified to assess tumor burden. Animal survival was assessed following an approved death-as-endpoint protocol. Mice were killed by carbon dioxide asphyxiation once they reached an endpoint according to the IACUC-approved protocols.

In vivo evaluation of liver tropism

To compare the liver tropism of AAV6 and AAV.APR31 following systemic delivery, FFLuc-expressing AAV vectors packaged with either AAV6 or AAV.APR31 capsids were administered to mice through intravenous injection of 2 × 1010 vg of AAV per mouse. Mice were subjected to longitudinal bioluminescence imaging using an IVIS. Before imaging, animals received 100 μl of 150 μg ml−1 D-luciferin substrate to enable detection of luciferase activity. Bioluminescence signals were acquired under identical imaging settings for all groups and quantified the whole-body total flux.

In vivo transgene delivery in T cells

To evaluate the in vivo T cell delivery efficiency of AAV.APR31, human T cells (3 × 106 per mouse) and scAAV-CAG-EGFP packaged with AAV6 or AAV.APR31 capsids (5 × 1011 vg per mouse) were administered into NOG mice through intravenous injection on the same day. Then, 48 h after injection, spleens were isolated and dissociated into single-cell suspensions. Cells were stained with live/dead, anti-human CD3, CD4 and CD8 antibodies and analyzed by flow cytometry. EGFP expression was assessed within gated human CD3+, CD4+ and CD8+ T cell populations. Five mice were included per group.

In vivo CAR-T cell generation and antitumor assessment

To assess the antitumor efficacy of CAR-T cells generated with AAV.APR31 in vivo, NOG mice were intravenously injected with 1 × 105 NALM-6-GL cells through the tail vein. Then, 3 days later, tumor-bearing mice received 8 × 106 of human T cells through tail-vein injection, followed by tail-vein injection of 5 × 1011 vg of AAV per mouse (2 × 1013 vg per kg). Human IL-2 (dose: 2.5 μg per mouse) was intraperitoneally injected on day 0 and day 3 after T cell infusion. Tumor progression was monitored and evaluated by bioluminescence imaging. For the pancreatic malignancy model, 3 × 106 AsPC-1 cells were subcutaneously injected into NOG mice. Then, 9 days later, tumor-bearing mice received 8 × 106 human T cells through tail-vein injection, followed by systemic delivery of scAAV-MSLN-CAR packaged in AAV6 or AAV.APR31 at three time points (5 × 1011, 2 × 1011 and 2 × 1011 vg per mouse on days 9, 11 and 13, respectively). Human IL-2 (dose: 2.5 μg per mouse) was administered intraperitoneally on each AAV injection day. Animal survival was assessed following an approved death-as-endpoint protocol. Mice were killed by carbon dioxide asphyxiation once they reached an endpoint according to the IACUC-approved protocols.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

Data generated or analyzed during this study are included in the article and its Supplementary Information. Raw and processed sequencing data from AAV evolution and CRISPR screen experiments are available from the Gene Expression Omnibus under accession number GSE302084 and RNA-seq sequencing data are available from Genome Sequence Archive under accession number HRA012600. HPA data were downloaded from the HPA database (https://www.proteinatlas.org/humanproteome/tissue/data#hpa_tissues_rna). The cross-tissue multicellular single-cell atlas was downloaded from Zenodo (https://doi.org/10.5281/zenodo.15554080)110. scRNA-seq data of bone marrow, thymus and brain were downloaded from the DISCO database (https://disco.bii.a-star.edu.sg/download). Data and materials that support the findings of this research are available to the academic community from the corresponding author upon reasonable request, through direct sharing, material transfer agreements or public repositories. Source data are provided with this paper.

Code availability

The code used for data analysis and the generation of figures related to this study is available from GitHub (https://github.com/GennongChen/AAV_evolution_APR).

References

  1. Bashor, C. J., Hilton, I. B., Bandukwala, H., Smith, D. M. & Veiseh, O. Engineering the next generation of cell-based therapeutics. Nat. Rev. Drug Discov. 21, 655–675 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  2. Irvine, D. J., Maus, M. V., Mooney, D. J. & Wong, W. W. The future of engineered immune cell therapies. Science 378, 853–858 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  3. June, C. H., O’Connor, R. S., Kawalekar, O. U., Ghassemi, S. & Milone, M. C. CAR T cell immunotherapy for human cancer. Science 359, 1361–1365 (2018).

    Article  CAS  PubMed  Google Scholar 

  4. Tscharke, D. C., Croft, N. P., Doherty, P. C. & La Gruta, N. L. Sizing up the key determinants of the CD8+ T cell response. Nat. Rev. Immunol. 15, 705–716 (2015).

    Article  CAS  PubMed  Google Scholar 

  5. Zhang, N. & Bevan, M. J. CD8+ T cells: foot soldiers of the immune system. Immunity 35, 161–168 (2011).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  6. Baker, D. J., Arany, Z., Baur, J. A., Epstein, J. A. & June, C. H. CAR T therapy beyond cancer: the evolution of a living drug. Nature 619, 707–715 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  7. Labanieh, L. & Mackall, C. L. CAR immune cells: design principles, resistance and the next generation. Nature 614, 635–648 (2023).

    Article  CAS  PubMed  Google Scholar 

  8. Maude, S. L. et al. Tisagenlecleucel in children and young adults with B-cell lymphoblastic leukemia. N. Engl. J. Med. 378, 439–448 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  9. Schultz, L. M. et al. Disease burden affects outcomes in pediatric and young adult B-cell lymphoblastic leukemia after commercial tisagenlecleucel: a pediatric real-world chimeric antigen receptor consortium report. J. Clin. Oncol. 40, 945–955 (2022).

    Article  CAS  PubMed  Google Scholar 

  10. Abramson, J. S. et al. Lisocabtagene maraleucel for patients with relapsed or refractory large B-cell lymphomas (TRANSCEND NHL 001): a multicentre seamless design study. Lancet 396, 839–852 (2020).

    Article  CAS  PubMed  Google Scholar 

  11. Neelapu, S. S. et al. Axicabtagene ciloleucel CAR T-cell therapy in refractory large B-cell lymphoma. N. Engl. J. Med. 377, 2531–2544 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  12. Schuster, S. J. et al. Tisagenlecleucel in adult relapsed or refractory diffuse large B-cell lymphoma. N. Engl. J. Med. 380, 45–56 (2019).

    Article  CAS  PubMed  Google Scholar 

  13. Nastoupil, L. J. et al. Standard-of-care axicabtagene ciloleucel for relapsed or refractory large b-cell lymphoma: results from the US lymphoma CAR T consortium. J. Clin. Oncol. 38, 3119–3128 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  14. Wang, M. et al. KTE-X19 CAR T-cell therapy in relapsed or refractory mantle-cell lymphoma. N. Engl. J. Med. 382, 1331–1342 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  15. Fowler, N. H. et al. Tisagenlecleucel in adult relapsed or refractory follicular lymphoma: the phase 2 ELARA trial. Nat. Med. 28, 325–332 (2022).

    Article  CAS  PubMed  Google Scholar 

  16. Munshi, N. C. et al. Idecabtagene vicleucel in relapsed and refractory multiple myeloma. N. Engl. J. Med. 384, 705–716 (2021).

    Article  CAS  PubMed  Google Scholar 

  17. Berdeja, J. G. et al. Ciltacabtagene autoleucel, a B-cell maturation antigen-directed chimeric antigen receptor T-cell therapy in patients with relapsed or refractory multiple myeloma (CARTITUDE-1): a phase 1b/2 open-label study. Lancet 398, 314–324 (2021).

    Article  CAS  PubMed  Google Scholar 

  18. Majzner, R. G. & Mackall, C. L. Clinical lessons learned from the first leg of the CAR T cell journey. Nat. Med. 25, 1341–1355 (2019).

    Article  CAS  PubMed  Google Scholar 

  19. Young, R. M., Engel, N. W., Uslu, U., Wellhausen, N. & June, C. H. Next-generation CAR T-cell therapies. Cancer Discov. 12, 1625–1633 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  20. Ellis, G. I., Sheppard, N. C. & Riley, J. L. Genetic engineering of T cells for immunotherapy. Nat. Rev. Genet. 22, 427–447 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  21. Atsavapranee, E. S., Billingsley, M. M. & Mitchell, M. J. Delivery technologies for T cell gene editing: applications in cancer immunotherapy. EBioMedicine 67, 103354 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  22. Ellis, J. Silencing and variegation of gammaretrovirus and lentivirus vectors. Hum. Gene Ther. 16, 1241–1246 (2005).

    Article  CAS  PubMed  Google Scholar 

  23. Mitchell, R. S. et al. Retroviral DNA integration: ASLV, HIV, and MLV show distinct target site preferences. PLoS Biol. 2, E234 (2004).

    Article  PubMed  PubMed Central  Google Scholar 

  24. Ciuffi, A. Mechanisms governing lentivirus integration site selection. Curr. Gene Ther. 8, 419–429 (2008).

    Article  CAS  PubMed  Google Scholar 

  25. Morgan, R. A. & Boyerinas, B. Genetic modification of T cells. Biomedicines 4, 9 (2016).

    Article  PubMed  PubMed Central  Google Scholar 

  26. Labbe, R. P., Vessillier, S. & Rafiq, Q. A. Lentiviral vectors for T cell engineering: clinical applications, bioprocessing and future perspectives. Viruses 13, 1528 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  27. Fraietta, J. A. et al. Disruption of TET2 promotes the therapeutic efficacy of CD19-targeted T cells. Nature 558, 307–312 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  28. Shah, N. N. et al. Clonal expansion of CAR T cells harboring lentivector integration in the CBL gene following anti-CD22 CAR T-cell therapy. Blood Adv. 3, 2317–2322 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  29. Perica, K. et al. CD4+ T-cell lymphoma harboring a chimeric antigen receptor integration in TP53. N. Engl. J. Med. 392, 577–583 (2025).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  30. Verdun, N. & Marks, P. Secondary cancers after chimeric antigen receptor T-cell therapy. N. Engl. J. Med. 390, 584–586 (2024).

    Article  CAS  PubMed  Google Scholar 

  31. Eyquem, J. et al. Targeting a CAR to the TRAC locus with CRISPR/Cas9 enhances tumour rejection. Nature 543, 113–117 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  32. Sather, B. D. et al. Efficient modification of CCR5 in primary human hematopoietic cells using a megaTAL nuclease and AAV donor template. Sci. Transl. Med. 7, 307ra156 (2015).

    Article  PubMed  PubMed Central  Google Scholar 

  33. Mansilla-Soto, J. et al. HLA-independent T cell receptors for targeting tumors with low antigen density. Nat. Med. 28, 345–352 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  34. Wang, J. et al. Highly efficient homology-driven genome editing in human T cells by combining zinc-finger nuclease mRNA and AAV6 donor delivery. Nucleic Acids Res. 44, e30 (2016).

    Article  PubMed  Google Scholar 

  35. Ertl, H. C. J. Immunogenicity and toxicity of AAV gene therapy. Front. Immunol. 13, 975803 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  36. Li, C. & Samulski, R. J. Engineering adeno-associated virus vectors for gene therapy. Nat. Rev. Genet. 21, 255–272 (2020).

    Article  CAS  PubMed  Google Scholar 

  37. Wang, D., Tai, P. W. L. & Gao, G. Adeno-associated virus vector as a platform for gene therapy delivery. Nat. Rev. Drug Discov. 18, 358–378 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  38. Nyberg, W. A. et al. In vivo site-specific engineering to reprogram T cells. Nature 652, 712–721 (2026).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  39. Ye, L. et al. A genome-scale gain-of-function CRISPR screen in CD8 T cells identifies proline metabolism as a means to enhance CAR-T therapy. Cell Metab. 34, 595–614 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  40. Tse, L. V. et al. Structure-guided evolution of antigenically distinct adeno-associated virus variants for immune evasion. Proc. Natl Acad. Sci. USA 114, E4812–E4821 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  41. Stadtmauer, E. A. et al. CRISPR-engineered T cells in patients with refractory cancer. Science 367, eaba7365 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  42. Zhang, J. et al. Non-viral, specifically targeted CAR-T cells achieve high safety and efficacy in B-NHL. Nature 609, 369–374 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  43. Yang, Y. et al. TCR engagement negatively affects CD8 but not CD4 CAR T cell expansion and leukemic clearance. Sci. Transl. Med. 9, eaag1209 (2017).

    Article  PubMed  PubMed Central  Google Scholar 

  44. Ghassemi, S. et al. Rapid manufacturing of non-activated potent CAR T cells. Nat. Biomed. Eng. 6, 118–128 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  45. Doench, J. G. et al. Optimized sgRNA design to maximize activity and minimize off-target effects of CRISPR–Cas9. Nat. Biotechnol. 34, 184–191 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  46. Pillay, S. et al. An essential receptor for adeno-associated virus infection. Nature 530, 108–112 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  47. Meisen, W. H. et al. Pooled screens identify GPR108 and TM9SF2 as host cell factors critical for AAV transduction. Mol. Ther. Methods Clin. Dev. 17, 601–611 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  48. Aruffo, A. & Seed, B. Molecular cloning of two CD7 (T-cell leukemia antigen) cDNAs by a COS cell expression system. EMBO J. 6, 3313–3316 (1987).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  49. Aandahl, E. M. et al. CD7 is a differentiation marker that identifies multiple CD8 T cell effector subsets. J. Immunol. 170, 2349–2355 (2003).

    Article  CAS  PubMed  Google Scholar 

  50. Rabinowich, H., Pricop, L., Herberman, R. B. & Whiteside, T. L. Expression and function of CD7 molecule on human natural killer cells. J. Immunol. 152, 517–526 (1994).

    Article  CAS  PubMed  Google Scholar 

  51. Zhang, R. et al. Adeno-associated virus 2 bound to its cellular receptor AAVR. Nat. Microbiol. 4, 675–682 (2019).

    Article  CAS  PubMed  Google Scholar 

  52. Meyer, N. L. et al. Structure of the gene therapy vector, adeno-associated virus with its cell receptor, AAVR. eLife 8, e44707 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  53. Locke, F. L. et al. Long-term safety and activity of axicabtagene ciloleucel in refractory large B-cell lymphoma (ZUMA-1): a single-arm, multicentre, phase 1-2 trial. Lancet Oncol. 20, 31–42 (2019).

    Article  CAS  PubMed  Google Scholar 

  54. Orlando, E. J. et al. Genetic mechanisms of target antigen loss in CAR19 therapy of acute lymphoblastic leukemia. Nat. Med. 24, 1504–1506 (2018).

    Article  CAS  PubMed  Google Scholar 

  55. Shah, N. N. & Fry, T. J. Mechanisms of resistance to CAR T cell therapy. Nat. Rev. Clin. Oncol. 16, 372–385 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  56. Hamieh, M. et al. CAR T cell trogocytosis and cooperative killing regulate tumour antigen escape. Nature 568, 112–116 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  57. Shah, N. N., Maatman, T., Hari, P. & Johnson, B. Multi targeted CAR-T cell therapies for B-cell malignancies. Front Oncol 9, 146 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  58. Majzner, R. G. & Mackall, C. L. Tumor antigen escape from CAR T-cell therapy. Cancer Discov 8, 1219–1226 (2018).

    Article  CAS  PubMed  Google Scholar 

  59. Zah, E., Lin, M. Y., Silva-Benedict, A., Jensen, M. C. & Chen, Y. Y. T cells expressing CD19/CD20 bispecific chimeric antigen receptors prevent antigen escape by malignant B cells. Cancer Immunol. Res. 4, 498–508 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  60. Wolf, N. K., Kissiov, D. U. & Raulet, D. H. Roles of natural killer cells in immunity to cancer, and applications to immunotherapy. Nat. Rev. Immunol. 23, 90–105 (2023).

    Article  CAS  PubMed  Google Scholar 

  61. Ye, L. et al. AAV-mediated delivery of a Sleeping Beauty transposon and an mRNA-encoded transposase for the engineering of therapeutic immune cells. Nat. Biomed. Eng. 8, 132–148 (2024).

    Article  CAS  PubMed  Google Scholar 

  62. Uhlen, M. et al. Tissue-based map of the human proteome. Science 347, 1260419 (2015).

    Article  PubMed  Google Scholar 

  63. Shi, Q. et al. Cross-tissue multicellular coordination and its rewiring in cancer. Nature 643, 529–538 (2025).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  64. Korbelin, J. et al. A brain microvasculature endothelial cell-specific viral vector with the potential to treat neurovascular and neurological diseases. EMBO Mol. Med. 8, 609–625 (2016).

    Article  PubMed  PubMed Central  Google Scholar 

  65. Deverman, B. E. et al. Cre-dependent selection yields AAV variants for widespread gene transfer to the adult brain. Nat. Biotechnol. 34, 204–209 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  66. Bryant, D. H. et al. Deep diversification of an AAV capsid protein by machine learning. Nat. Biotechnol. 39, 691–696 (2021).

    Article  CAS  PubMed  Google Scholar 

  67. Tabebordbar, M. et al. Directed evolution of a family of AAV capsid variants enabling potent muscle-directed gene delivery across species. Cell 184, 4919–4938 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  68. Chan, K. Y. et al. Engineered AAVs for efficient noninvasive gene delivery to the central and peripheral nervous systems. Nat. Neurosci. 20, 1172–1179 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  69. Ravindra Kumar, S. et al. Multiplexed Cre-dependent selection yields systemic AAVs for targeting distinct brain cell types. Nat. Methods 17, 541–550 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  70. Huang, Q. et al. An AAV capsid reprogrammed to bind human transferrin receptor mediates brain-wide gene delivery. Science 384, 1220–1227 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  71. Huang, Q. et al. Targeting AAV vectors to the central nervous system by engineering capsid-receptor interactions that enable crossing of the blood–brain barrier. PLoS Biol. 21, e3002112 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  72. Gonzalez, T. J. et al. Cross-species evolution of a highly potent AAV variant for therapeutic gene transfer and genome editing. Nat. Commun. 13, 5947 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  73. Yao, Y. et al. Variants of the adeno-associated virus serotype 9 with enhanced penetration of the blood–brain barrier in rodents and primates. Nat. Biomed. Eng. 6, 1257–1271 (2022).

    Article  CAS  PubMed  Google Scholar 

  74. Dalkara, D. et al. In vivo-directed evolution of a new adeno-associated virus for therapeutic outer retinal gene delivery from the vitreous. Sci. Transl. Med. 5, 189ra176 (2013).

    Article  Google Scholar 

  75. Korbelin, J. et al. Pulmonary targeting of adeno-associated viral vectors by next-generation sequencing-guided screening of random capsid displayed peptide libraries. Mol Ther 24, 1050–1061 (2016).

    Article  PubMed  PubMed Central  Google Scholar 

  76. Yang, L. et al. A myocardium tropic adeno-associated virus (AAV) evolved by DNA shuffling and in vivo selection. Proc. Natl Acad. Sci. USA 106, 3946–3951 (2009).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  77. Nyberg, W. A. et al. An evolved AAV variant enables efficient genetic engineering of murine T cells. Cell 186, 446–460 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  78. Rurik, J. G. et al. CAR T cells produced in vivo to treat cardiac injury. Science 375, 91–96 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  79. Hunter, T. L. et al. In vivo CAR T cell generation to treat cancer and autoimmune disease. Science 388, 1311–1317 (2025).

    Article  CAS  PubMed  Google Scholar 

  80. Billingsley, M. M. et al. In vivo mRNA CAR T cell engineering via targeted ionizable lipid nanoparticles with extrahepatic tropism. Small 20, e2304378 (2024).

    Article  PubMed  Google Scholar 

  81. Hamilton, J. R. et al. In vivo human T cell engineering with enveloped delivery vehicles. Nat. Biotechnol. 42, 1684–1692 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  82. Pfeiffer, A. et al. In vivo generation of human CD19-CAR T cells results in B-cell depletion and signs of cytokine release syndrome. EMBO Mol. Med. 10, e9158 (2018).

    Article  PubMed  PubMed Central  Google Scholar 

  83. Agarwal, S. et al. In vivo generation of CAR T cells selectively in human CD4+ lymphocytes. Mol. Ther. 28, 1783–1794 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  84. Coradin, T. et al. Efficient in vivo generation of CAR T cells using a retargeted fourth-generation lentiviral vector. Mol. Ther. 33, 4953–4967 (2025).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  85. Wang, X. et al. Allogeneic CD19-targeted CAR-T therapy in patients with severe myositis and systemic sclerosis. Cell 187, 4890–4904 (2024).

    Article  CAS  PubMed  Google Scholar 

  86. Muller, F. et al. CD19 CAR T-cell therapy in autoimmune disease—a case series with follow-up. N. Engl. J. Med. 390, 687–700 (2024).

    Article  PubMed  Google Scholar 

  87. Schett, G., Mackensen, A. & Mougiakakos, D. CAR T-cell therapy in autoimmune diseases. Lancet 402, 2034–2044 (2023).

    Article  CAS  PubMed  Google Scholar 

  88. Mougiakakos, D. et al. CD19-targeted CAR T cells in refractory systemic lupus erythematosus. N. Engl. J. Med. 385, 567–569 (2021).

    Article  PubMed  Google Scholar 

  89. Jorgensen, L. V., Christensen, E. B., Barnkob, M. B. & Barington, T. The clinical landscape of CAR NK cells. Exp. Hematol. Oncol. 14, 46 (2025).

    Article  PubMed  PubMed Central  Google Scholar 

  90. Lu, Z. et al. An AAV variant enables human T cell engineering in vivo. Vita 1, 125–139 (2026).

    Article  Google Scholar 

  91. Haso, W. et al. Anti-CD22-chimeric antigen receptors targeting B-cell precursor acute lymphoblastic leukemia. Blood 121, 1165–1174 (2013).

    Article  CAS  PubMed  Google Scholar 

  92. Good, C. R. et al. An NK-like CAR T cell transition in CAR T cell dysfunction. Cell 184, 6081–6100 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  93. Shah, N. N. et al. Bispecific anti-CD20, anti-CD19 CAR T cells for relapsed B cell malignancies: a phase 1 dose escalation and expansion trial. Nat. Med. 26, 1569–1575 (2020).

    Article  CAS  PubMed  Google Scholar 

  94. Roth, T. L. et al. Reprogramming human T cell function and specificity with non-viral genome targeting. Nature 559, 405–409 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  95. Allen, A. G. et al. A highly efficient transgene knock-in technology in clinically relevant cell types. Nat. Biotechnol. 42, 458–469 (2024).

    Article  CAS  PubMed  Google Scholar 

  96. Ye, L. et al. In vivo CRISPR screening in CD8 T cells with AAV–Sleeping Beauty hybrid vectors identifies membrane targets for improving immunotherapy for glioblastoma. Nat. Biotechnol. 37, 1302–1313 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  97. Martin, M. Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet J. 17, 10–12 (2011).

    Article  Google Scholar 

  98. Kolacsek, O. et al. Reliable transgene-independent method for determining Sleeping Beauty transposon copy numbers. Mob. DNA 2, 5 (2011).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  99. Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29, 15–21 (2013).

    Article  CAS  PubMed  Google Scholar 

  100. Liao, Y., Smyth, G. K. & Shi, W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923–930 (2014).

    Article  CAS  PubMed  Google Scholar 

  101. Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

  102. Chu, Y. et al. Pan-cancer T cell atlas links a cellular stress response state to immunotherapy resistance. Nat. Med. 29, 1550–1562 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  103. Chen, S. Ultrafast one-pass FASTQ data preprocessing, quality control, and deduplication using fastp. iMeta 2, e107 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  104. Chen, S., Zhou, Y., Chen, Y. & Gu, J. fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics 34, i884–i890 (2018).

    Article  PubMed  PubMed Central  Google Scholar 

  105. Li, W. et al. MAGeCK enables robust identification of essential genes from genome-scale CRISPR/Cas9 knockout screens. Genome Biol. 15, 554 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

  106. Wu, T. Z. et al. clusterProfiler 4.0: a universal enrichment tool for interpreting omics data. Innovation (Camb.) 2, 100141 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

  107. Li, M. et al. Rediscovering publicly available single-cell data with the DISCO platform. Nucleic Acids Res. 53, D932–D938 (2025).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  108. Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol. 19, 15 (2018).

    Article  PubMed  PubMed Central  Google Scholar 

  109. Waskom, M. L. seaborn: statistical data visualization. J. Open Source Softw. 6, 3021 (2021).

    Article  Google Scholar 

  110. Shi, Q. & Chen, Y. Data for ‘Cross-tissue multicellular coordination and its rewiring in cancer’. Zenodo https://doi.org/10.5281/zenodo.15554080 (2026).

Download references

Acknowledgements

We thank all members of the L.Y. laboratory for various discussions. We thank the Institute of Modern Biology, the Chemistry and Biomedicine Innovation Center, the State Key Laboratory of Pharmaceutical Biotechnology and Nanjing Medical University core facilities for technical support.

Funding

L.Y. is supported by the Nanjing University Institute of Modern Biology Startup Fund, the National Natural Science Fund for Excellent Young Scientists Fund Program (overseas), the Fundamental Research Funds for the Central Universities (2024300391), National Natural Science Foundation of China (82473328), the Jiangsu Provincial Innovation and Entrepreneurship Team (JSSCTD202347) and the Jiangsu Specially Appointed Professor Fund.

Author information

Author notes

  1. These authors contributed equally: Lingjuan Hong, Gennong Chen, Yongjun Liu, Yuanfang Cao, Xueting Li.

Authors and Affiliations

  1. Institute of Modern Biology, Department of Hematology, Nanjing Drum Tower Hospital, Nanjing University, Nanjing, China

    Lingjuan Hong, Gennong Chen, Yongjun Liu, Yuanfang Cao, Xueting Li, Weijie Li, Yumu Pan, Huahe Wang, Gaomeng Ru, Siyu Yang, Ming Cheng, Tongtong Lv, Yunfan Zhu, Shan Huang, Xiaoxiao Nian, Hongling Ou & Lupeng Ye

  2. Department of Genetics, Yale University School of Medicine, New Haven, CT, USA

    Lingjuan Hong & Monkol Lek

  3. School of Basic Medicine, Nanjing Medical University, Nanjing, China

    Chengkun Wang

  4. Department of Laboratory Medicine, Yale University School of Medicine, New Haven, CT, USA

    Sean X. Gu

  5. State Key Laboratory of Pharmaceutical Biotechnology, Nanjing University, Nanjing, China

    Lupeng Ye

Authors

  1. Lingjuan Hong
  2. Gennong Chen
  3. Yongjun Liu
  4. Yuanfang Cao
  5. Xueting Li
  6. Weijie Li
  7. Yumu Pan
  8. Huahe Wang
  9. Gaomeng Ru
  10. Siyu Yang
  11. Ming Cheng
  12. Tongtong Lv
  13. Yunfan Zhu
  14. Shan Huang
  15. Xiaoxiao Nian
  16. Hongling Ou
  17. Chengkun Wang
  18. Sean X. Gu
  19. Monkol Lek
  20. Lupeng Ye

Contributions

L.Y. and L.H. conceptualized and designed the study. L.H., Y.L., Y.C., X.L. and L.Y. led and performed experiments. G.C. led the NGS data analysis. L.Y., L.H., Y.C., Y.L., X.L. and W.L. analyzed the data except for NGS. W.L., Y.P., H.W., G.R., S.Y., M.C., T.L., S.H., X.N., Y.Z. and H.O. assisted with the experiments. C.W. provided technical advice on gene editing and assisted with the animal experiments. S.X.G. provided technical and conceptual advice on CAR-T. M.L. provided conceptual guidance, supervision and critical revision of the paper. L.Y. and L.H. prepared the paper with inputs from all authors. L.Y. secured funding and supervised the work.

Corresponding author

Correspondence to Lupeng Ye.

Ethics declarations

Competing interests

A patent has been filed related to this study, with L.Y., L.H., Y.L., G.C., Y.C., and X.L. listed as inventors. The other authors declare no competing interests.

Peer review

Peer review information

Nature Biotechnology thanks Sebastian Kobold, Ralf Schmidt and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Fig. 1 Directed evolution identifies a family of APR-enriched capsid variants.

Related to Fig. 1. (a-c) NGS analysis of parental and evolved libraries from three rounds of selection with total and unique read counts. (d) Correlation analysis of AAV variants in two donors. Scatter plot comparing the log2-transformed fold-change (FC) of reads of the top 1000 AAV variants after round 3 selection across Donor A (x-axis) and Donor B (y-axis). Each point represents a unique variant, categorized as AAV6 (pink), APR variants (green), or non-APR variants (gray). The dashed black line represents the line of best fit. The Pearson correlation coefficient (r = 0.952) and the corresponding P value (< 2.2e-16) are indicated on the plot. (e) Enrichment of the parental top 1000 variants over selection rounds in human T cells. Line graph showing the percent of reads for different variant groups across three rounds of selection in Donor A and Donor B. (f) Box plot comparing the enrichment of APR-variants and non-APR-variants in donor A and donor B, showing the top 1000 enriched variants. The pink lines indicated the position of AAV6. Individual data points are overlaid as dots. The Wilcoxon t test was used to evaluate statistical significance, the P values were indicated in the plots.

Extended Data Fig. 2 AAV.APR31 enables efficient transgene delivery in primary human T cells and mouse T and NK cells.

Related to Fig. 2; Fig. 3. (a) RT-qPCR quantification of EGFP mRNA in Jurkat cells after scAAV-CAG-EGFP transduction. Data are presented as mean values +/- SEM, n = 3 biological replicates. (b, c) The percentage of EGFP-positive T cells following transduction of human T cells with AAV6 or AAV.APR31 at the indicated MOIs (T cells were used at day 5 after activation). 2 donors were used for transduction efficiency evaluation. (d) Left: Representative flow cytometry analysis of EGFP expression following transduction of mouse CD8+ T cells with AAV6 or AAV.APR31 at the indicated MOIs. Right: EGFP quantification of transduced cells. Data are presented as mean values +/- SEM, n = 3 biological replicates. (e) The percentage of EGFP-positive mouse NK cells following transduction of AAV6 or AAV.APR31 at the indicated MOIs. Data are presented as mean values +/- SEM, n = 3 biological replicates. (f) CD22-CAR knock-in efficiency was analyzed by flow cytometry in donor B T cells at day 5 after AAV transduction. The knock-in efficiency was quantified by the number of CD22-CAR-positive cells across a range of MOIs. n = 1 donor with 4 different MOIs. (g) CD22-CAR knock-in efficiency was analyzed by flow cytometry in donor E T cells at day 7 after AAV transduction. The knock-in efficiency was quantified by the number of CD22-CAR-positive cells across a range of MOIs. n = 1 donor with 6 different MOIs. (h, i) Episomal transgene expression analysis in the absence of genome editing. Human T cells were transduced with AAV-EFS-EGFP (h) or AAV-EFS-CD22-CAR (i) without CRISPR/Cas9 RNP delivery. EGFP or CD22-CAR expression was assessed by flow cytometry at day 5 post-transduction. n = 1 donor for each experiment. (j) Vector copy number (VCN) analysis of CD22-CAR knock-in T cells generated using AAV6 or AAV.APR31. VCN was quantified by qPCR targeting the integrated transgene cassette at the TRAC locus. Data are presented as mean values +/- SEM, n = 4 donors. (k) Representative flow cytometry plots of CD69 and CD25 expression in non-pre-stimulated (maintained in IL-2-containing medium without anti-CD3/anti-CD28 stimulation) and anti-CD3/anti-CD28-stimulated bulk human CD8+ T cells. Representative of n = 3 donors. (l) Quantification of CD69 and CD25 on non-pre-stimulated and anti-CD3/anti-CD28 antibody stimulated bulk CD8+ T cells. Antibody-stimulated cells were analyzed on day 2 and day 4 after stimulation. Data are presented as mean values +/- SEM, n = 3 donors. (m) Representative flow cytometry plots of Ki-67 expression in non-pre-stimulated and anti-CD3/anti-CD28-stimulated bulk human CD8+ T cells. Representative of n = 3 donors. FMO: fluorescence-minus-one controls. In this figure, ordinary two-way ANOVA tests (a, d), multiple two-sided unpaired t tests (adjusted P value) (e), and two-sided unpaired t tests (j, l) were used to evaluate statistical significance. Significance notes: ns - not significant, ** P < 0.01, *** P < 0.001, **** P < 0.0001.

Source data

Extended Data Fig. 3 Genome-wide CRISPR knockout screening in NALM-6 cells and GO enrichment analysis.

Related to Fig. 4. (a) Scatter plots of median sgRNA log2-fold change (EGFP low/high sorting bins) for each gene in NALM-6. (b) sgRNA log2-fold changes for genes of interest in the NALM-6 screen. Bars represent the mean log2-fold change for each sgRNA across three independent replicates. The density plot above represents the distribution of all sgRNAs. (c) GO enrichment analysis of positive regulators of AAV.APR31 infection in Jurkat screen.

Extended Data Fig. 4 CD7 is a key receptor for AAV.APR31 transduction in human and mouse T cells.

Related to Fig. 4; Fig. 5. (a) Cas9 RNP-mediated CD7 knockout in human T cells (Donor 059 K). Representative flow cytometry detection of CD7 expression after Cas9 RNP electroporation. (b) Quantification of EGFP percentage and MFI in Cas9/sgNTC or Cas9/sgCd7 RNP-treated mouse T cells. Data are presented as mean values +/- SEM, n = 3 biological replicates. (c) Quantification of EGFP percentage in human T cells that were pre-treated with anti-human CD7 antibody (MOI = 10000). Data are presented as mean values +/- SEM, n = 3 donors. (d) Representative flow cytometry detection of CD7 expression in Jurkat, K-562, and NALM-6. (e-g) Representative fluorescence histograms and quantification of EGFP fluorescence intensity of Jurkat (CD7 high), K-562 (CD7 low), and NALM-6 (CD7 negative) cells after scAAV-CAG-EGFP transduction. n = 1 replicate with 4 different MOIs. (h) SDS-PAGE analysis of purified recombinant human AAVR PKD1-5 domain protein. The expected molecular weight is indicated. (i) Quantification of MSLN-CAR, GD2-CAR, CD19.20-CAR, and NY-ESO-1 TCR knock-in efficiency at day 7 after electroporation and AAV transduction. n = 1 donor with 5 different MOIs. (j) EGFP expression in human T cells transduced with AAV-SB-EGFP in the absence of SB100× mRNA. EGFP expression was assessed by flow cytometry at day 7 post-transduction. n = 2 donors. In this figure, ordinary two-way ANOVA tests (b) and multiple two-sided unpaired t tests (adjusted P value) (c) were used to evaluate statistical significance. Significance notes: ns - not significant, **** P < 0.0001.

Source data

Extended Data Fig. 5 RNA-seq analysis of CD22-CAR T cells generated with AAV6 or AAV.APR31.

Related to Fig. 5. (a) Principal component analysis (PCA) of transcriptomes from CAR T cells generated by AAV6 and AAV.APR31 across two donors and two time points (16 h and d7). (b) The distribution of transcriptomic gene log2 fold change of CAR T cells generated by AAV6 or AAV.APR31. (c, d) The correlation analysis of transcriptomic gene log2 fold change of CAR T cells produced by AAV6 or AAV.APR31 at 16 h and d7 after AAV transduction. (e-i) The hierarchical clustering analysis of the immune function-related genes in CD22-CAR T cells.

Extended Data Fig. 6 CD7 expression profile across human tissues and cell types.

(a) Human bulk RNA-seq data analysis and CD7 expression pattern across different human tissues and organs. (b) Human scRNA-seq data analysis and CD7 expression pattern across different human tissues and organs at the pseudo-bulk level. (c, d) The major cell type distribution in different human tissues and organs. (e) CD7 expression heatmap across various human tissues and major cell types. (f) Thymus scRNA-seq data analysis to map CD7 expression pattern in cell types. (g) Bone marrow-specific scRNA-seq data analysis to map CD7 expression pattern in cell types.

Extended Data Fig. 7 Off-target analysis and pre-clinical anti-tumor evaluation of AAV.APR31.

Related to Fig. 6. (a) Representative IVIS bioluminescence images of PBS-, AAV6-, and AAV.APR31-injected mice at the indicated time points following intravenous administration of luciferase-packaged vectors (2 × 1011 vg per mouse). (b) Longitudinal quantification of bioluminescence signals from mice in (a). Data are presented as mean values +/- SEM, n = 5 mice per group. (c) Quantification of firefly luciferase (FFLuc) mRNA levels by RT-qPCR in lung, spleen, kidney, heart, brain, and skeletal muscle harvested from PBS-, AAV6-, and AAV.APR31-injected mice. Expression levels are normalized to PBS controls. Data are presented as mean values +/- SEM, n = 3 mice per group. (d) Schematic of the ex vivo CD22-CAR T cell manufacturing and cytotoxicity assay workflow. (e) Representative flow cytometry plots showing CD22-CAR expression in AAV6- and AAV.APR31-transduced T cells at MOI 250, 500, and 1,000. Numbers indicate the percentage of cells in each quadrant. (f) In vitro cytotoxicity of bulk T cells against NALM-6-GL target cells at 24 h (upper panels) and 48 h (lower panels) co-culture, at total T cell-to-cancer cell ratios of 1:2 and 1:1, across MOI 250, 500, and 1,000. Data are presented as mean values +/- SEM, n = 3 replicates. (g) Cytolysis normalized to CAR-positive cell percentage for AAV6- and AAV.APR31-generated CD22-CAR T cells at 24 h co-culture. (h) In vitro cytotoxicity of lentiviral vector-, AAV6-, and AAV.APR31-manufactured CD22-CAR T cells against NALM-6-GL target cells at 48 h co-culture across three independent donors (Donor 054 K, Donor 0718B, Donor 1120 A), at effector-to-target ratios of 1:2, 1:5, and 1:10. (i) Representative IVIS bioluminescence images of SK-OV-3-PL tumor-bearing NOG mice treated with PBS, lentiviral-, AAV6-, or AAV.APR31-manufactured MSLN-CAR T cells at the indicated time points. Red crosses indicate mice that reached humane endpoints. In this figure, ordinary two-way ANOVA (f, g), two-way ANOVA with Tukey’s multiple comparisons tests (adjusted P value) (b), and two-sided unpaired t tests (c) were used to evaluate statistical significance. Significance notes: ns - not significant, * P < 0.05, ** P < 0.01, **** P < 0.0001.

Source data

Rights and permissions

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 licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence 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 licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Hong, L., Chen, G., Liu, Y. et al. A CD7-targeted AAV variant enables efficient T and NK cell engineering in vivo. Nat Biotechnol (2026). https://doi.org/10.1038/s41587-026-03344-5

Download citation

  • Received: 21 February 2026

  • Accepted: 14 September 2026

  • Published: 16 November 2026

  • Version of record: 16 November 2026

  • DOI: https://doi.org/10.1038/s41587-026-03344-5

来源:Nature Biotechnology · nature.com