Directed evolution of a family of AAV capsid variants enabling potent muscle-directed gene delivery across species

Parts used

Key findings

• DELIVER (directed evolution of AAV capsids leveraging in vivo expression of transgene RNA) selects functional AAV capsid variants based on transgene mRNA expression in the target tissue rather than on the presence of vector genome DNA, with each capsid variant carrying its own 7-mer-inserted capsid coding sequence as its own selectable transgene. Selecting on mRNA yields far fewer enriched variants than selecting on vector genome DNA, and across mouse liver and muscle there is essentially no correlation between a capsid variant's abundance in the injected virus library or its vector genome DNA level in a tissue and its transgene mRNA expression level in that same tissue — indicating that physical entry of a capsid variant into a cell does not predict whether it will functionally transduce that cell.
• Two rounds of in vivo DELIVER selection in C57BL/6J mice, screening a starting library of more than 5,000,000 unique AAV9-derived capsid variants each carrying a random 7-amino-acid peptide inserted between positions 588 and 589 of the capsid, converge on a family of muscle-tropic variants that all share an RGD (arginine-glycine-aspartic acid) motif at the first three positions of the inserted peptide: all 12 of the top variants enriched for expression in skeletal and cardiac muscle under either the CK8 or MHCK7 muscle-specific promoter contain this motif. The variant carrying the RGDLTTP peptide insertion, designated MyoAAV 1A, is the top-ranked hit under the MHCK7 promoter and one of the two highest-ranked hits overall.
• Systemic (intravenous) administration of MyoAAV 1A-CMV-EGFP at 1E+12 vg (~4E+13 vg/kg) in C57BL/6J mice produces 10- to 29-fold higher EGFP mRNA expression across skeletal muscles and 6.3-fold higher expression in heart than the same dose of AAV9, while liver EGFP mRNA is 2.8-fold lower — a liver-detargeted, muscle- and heart-selective transduction profile not seen in lung, kidney, spleen, or brain, where MyoAAV 1A transduces similarly to or less efficiently than AAV9.
• MyoAAV 1A also outperforms AAV9 outside of high-dose systemic muscle delivery: intramuscular injection into the tibialis anterior of C57BL/6J mice produces 14-fold higher EGFP mRNA than AAV9 at the same dose, and in vitro transduction of primary myotubes is 35- to 52-fold more efficient than AAV9 in human myotubes from four independent donors and 23-fold more efficient in mouse myotubes.
• MyoAAV 1A delivered systemically to mdx-Ai9 mice transduces muscle satellite (stem) cells — assessed by Cre-dependent tdTomato labeling and FACS — with 2.6-fold and 3.2-fold higher efficiency than AAV9 and AAV8, respectively.
• The enhanced, liver-detargeted muscle transduction profile of MyoAAV 1A, established in C57BL/6J mice, is conserved in BALB/cJ and DBA/2J mice: systemic injection of MyoAAV 1A-CMV-EGFP in both strains reproduces the same pattern of increased muscle transduction and reduced liver targeting.
• Systemic delivery of MyoAAV 1A carrying a SaCas9/dual-gRNA construct targeting the mdx Dmd mutation produces exon-23-deleted Dmd mRNA at 3.4%–25% of total Dmd transcript across muscles of mdx mice, compared with 1.3%–8.7% for the same construct delivered by AAV9, alongside 6.7- to 19-fold higher SaCas9 mRNA and 3.5- to 7.8-fold higher gRNA expression. This translates into more extensive dystrophin protein restoration and significantly higher tibialis anterior specific force with less eccentric-contraction-induced force loss than in AAV9-CRISPR-treated mdx mice.
• Systemic delivery of MyoAAV 1A carrying human MTM1 under the MHCK7 promoter, at a dose of 2E+12 vg/kg — 50- to 150-fold lower than the dose used in an ongoing human XLMTM clinical trial — rescues Mtm1 knockout mice from the disease's otherwise lethal phenotype: AAV9-hMTM1-injected Mtm1 knockout mice reach humane endpoint between 11 and 21 weeks after injection, whereas MyoAAV 1A-hMTM1-injected Mtm1 knockout mice survive with body weight, activity, and survival trajectories similar to wild-type littermates, along with higher hMTM1 mRNA/protein expression, higher muscle-specific force, and lower liver vector genome accumulation than AAV9-treated animals.
• MyoAAV 1A transduction depends on RGD-binding integrin heterodimers on the target cell surface: overexpressing human α8β1, αVβ1, αVβ3, αVβ6, or αVβ8 in HEK293 cells increases MyoAAV 1A transduction efficiency relative to pUC19-transfected controls, and overexpressing αVβ6 — and to a lesser extent α8β1 or αVβ1 — increases MyoAAV 1A binding to the cell surface. The pan-αV integrin antagonists CWHM-12 and GLPG-0187 each block MyoAAV 1A transduction of mouse and human primary myotubes in a dose-dependent manner, with no dose-dependent effect on AAV9. Among these integrins, αVβ6 has the highest affinity for MyoAAV 1A: pre-incubating virus with soluble αVβ6 protein, or treating human myotubes with anti-αVβ6 antibody, dose-dependently blocks MyoAAV 1A transduction, whereas αVβ1, αVβ3, and αVβ8 recombinant proteins do not.
• MyoAAV 1A transduction also requires the AAV receptor (AAVR), the endocytosed membrane protein needed for transduction by most AAV serotypes other than AAV4 and AAVrh32.33: MyoAAV 1A transduces AAVR-knockout HEK293FT cells far less efficiently than AAVR-expressing cells. Overexpressing RGD-binding integrin heterodimers in AAVR-knockout cells increases MyoAAV 1A transduction relative to mock-transfected AAVR-knockout controls but does not restore transduction to the level seen when AAVR itself is present, indicating that integrin heterodimers and AAVR act at distinct, non-redundant steps of MyoAAV 1A cell entry.
• MyoAAV 1A transduction and binding, like that of its AAV9 parent, is enhanced by removing terminal sialic acid from the target cell surface: treating HEK293 cells with neuraminidase increases MyoAAV 1A transduction 172-fold and cell-surface binding 4-fold, and subsequently masking the exposed terminal β-1,4 galactose with Erythrina cristagalli lectin inhibits both binding and transduction of MyoAAV 1A, as well as AAV9, while having no effect on AAV2.
• Varying the amino acids flanking the RGD motif (positions 586–588 and 592–595 of capsid hypervariable region VIII) while holding the RGD motif itself fixed, followed by two further rounds of in vivo DELIVER selection in C57BL/6J and mdx mice at two doses, identifies a second-generation 10-mer insertion, GPGRGDQTTL, as the most highly selected muscle-tropic sequence at both doses; this variant is designated MyoAAV 2A. Unlike MyoAAV 1A, which depends primarily on αVβ6, MyoAAV 2A binds αVβ1, αVβ3, αVβ6, and αVβ8 with higher affinity across the board, and antibody blockade of αVβ6 alone is significantly less effective at inhibiting MyoAAV 2A transduction of human myotubes than it is for first-generation variants such as MyoAAV 1A.
• Systemic delivery of MyoAAV 2A-CMV-EGFP at a low dose (2E+11 vg, ~8E+12 vg/kg) in C57BL/6J mice transduces skeletal muscle 10- to 80-fold and heart 17-fold more efficiently than AAV9, with 2.5-fold lower liver transgene expression, and transduces human primary myotubes 128-fold more efficiently than AAV9 and 4.1-fold more efficiently than MyoAAV 1A.
• Systemic delivery of MyoAAV 2A carrying a microdystrophin transgene (CK8-microdystrophin-FLAG) at a low dose of 2E+13 vg/kg to DBA/2J-mdx mice produces 7.6- to 15-fold higher microdystrophin mRNA and 12- to 46-fold higher vector genomes per diploid genome in skeletal muscle than the same dose of AAV9-microdystrophin, with liver vector genome levels 2.5-fold lower than AAV9 and roughly equal to muscle levels (versus more than 40-fold liver-over-muscle enrichment for AAV9), and produces significantly greater recovery of muscle-specific force after eccentric-contraction injury than AAV9.
• Independent application of DELIVER directly in cynomolgus macaques, starting from the same random-7-mer capsid library design used in mice, converges on RGD-motif-containing capsid variants (the MyoAAV 3 series) as the top muscle-tropic hits after two rounds of in vivo selection, showing that RGD-dependent muscle tropism is not an artifact specific to selection in mice. The optimal amino acids flanking the RGD motif, however, are species-dependent: a round of RGD-fixed selection performed directly in macaques, using the top 120,000 RGD-fixed variants previously identified in mice, yields a distinct set of top hits (the MyoAAV 4 series, including MyoAAV 4A, 4C, and 4E) whose flanking sequences differ from those of the mouse-optimized MyoAAV 2A. MyoAAV 2A, despite being the most potent second-generation variant for mouse muscle, does not transduce NHP muscle with high efficiency after systemic administration, whereas the NHP-selected variants MyoAAV 3A, 4A, 4C, and 4E outperform AAVrh74 and AAV9 for muscle transduction in both cynomolgus macaques and mice.
• The NHP-selected MyoAAV 3A, 4A, 4C, and 4E variants share the same receptor dependencies established for the mouse-selected MyoAAV 1A and 2A: transduction of human primary myotubes by all four variants is blocked in a dose-dependent manner by the pan-αV integrin antagonist GLPG-0187 and by anti-αVβ6 antibody, all four bind αVβ6 with the highest affinity among the αV-containing integrins tested (followed by αVβ8 and αVβ3), and all four require AAVR for transduction.

Abstract

Replacing or editing disease-causing mutations holds great promise for treating many human diseases. Yet, delivering therapeutic genetic modifiers to specific cells in vivo has been challenging, particularly in large, anatomically distributed tissues such as skeletal muscle. Here, we establish an in vivo strategy to evolve and stringently select capsid variants of adeno-associated viruses (AAVs) that enable potent delivery to desired tissues. Using this method, we identify a class of RGD motif-containing capsids that transduces muscle with superior efficiency and selectivity after intravenous injection in mice and non-human primates. We demonstrate substantially enhanced potency and therapeutic efficacy of these engineered vectors compared to naturally occurring AAV capsids in two mouse models of genetic muscle disease. The top capsid variants from our selection approach show conserved potency for delivery across a variety of inbred mouse strains, and in cynomolgus macaques and human primary myotubes, with transduction dependent on target cell expressed integrin heterodimers.