CHARIOT-AAV: Conjugation of diverse vectors to adeno-associated viruses for delivery of large genes

31 Systemic, tissue-specific delivery of large transgenes exceeding the packaging capacity of adeno-32 associated viruses (AA Vs) remains a key translational challenge for molecular therapeutics. 33 Vectors with larger capacities, such as lentiviral vectors ( LVVs) and lipid nanoparticles (LNPs) , 34 often lack adjustable, tissue -specific tropisms. Here we report CHARIOT -AA V (Crosslinked 35 Hybrid Architectures for Robust, Interchangeable, and Organ -specific Targeting with AA V), a 36 platform where diverse delivery vectors are conjugated to AA Vs, thereby achieving tissue-specific 37 tropism of AA Vs and expanded cargo capacity. AA V-AA V conjugates packaging split SpCas9 38 constructs in AA V .CAP-B10 capsids demonstrate a ~2-fold increase in brain gene editing 39 efficiency over unconjugated AA V cocktails after intravenous injection. In addition to AA V-AA V 40 conjugates, AA V-LVV and AA V-LNP conjugates achieve AAV-guided delivery of genetic payloads 41 to target cells . Furthermore, AA V-LNP conjugates enable systemic delivery of mRNAs to brain 42 endothelial cells. CHARIOT-AA V thus provides a modular platform for systemic, tissue-specific 43 delivery of diverse therapeutics beyond the limits of individual vectors. 44 45 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 3 Main 46 Systemic delivery of genetic cargo to target tissues via intravenous injection could transform 47 molecular therapies for genetic disorders and empower research with genetic models , yet it 48 remains a fundamental challenge1,2. Among delivery platforms, recombinant adeno-associated 49 viruses (AA Vs) are one of the most promising vehicles, offering low pathogenicity, limited 50 immunogenicity, and tunable tissue-specific tropism 3–11. However, AA V’s gene packaging 51 capacity is limited to ~4.8 kb 12, which must accommodate not only molecular tools but also 52 regulatory elements such as cell-type-specific promoters and organelle trafficking signals . 53 Transformative molecular tools such as CRISPR-Cas gene editing systems exceed this capacity (> 54 ~6 kb)1, and even smaller functional equivalents (e.g. CasMINI13, NanoCas14) that fit within a 55 single capsid leave insufficient room for the long regulatory sequences necessary for cell -type 56 specificity. Alternative vehicles such as l entiviral vectors (LVVs) and lipid nanoparticles (LNPs) 57 can accommodate larger cargo (~10 kb)15–17; however, these vectors offer limited tissue specificity 58 compared to AA Vs18,19. There remains a need for platform s that accommodate large and diverse 59 genetic cargo while achieving tissue specificity comparable to that of AA Vs. 60 To circumvent AA V’s packaging limit, previous work has pursued dual-AA V strategies that 61 split a large transgene across two vectors, relying on subsequent reconstitution of the full-length 62 gene or protein in target cells20,21. However, these approaches demand simultaneous transduction 63 of both constructs packaged in separate AA Vs, a requirement that potentially compromises overall 64 efficacy22,23 and necessitates elevated vector doses that raise safety concerns22,24. Meanwhile, non-65 AA V vectors, while promising 25–27, are yet to achieve tissue -specific tropism and versatility 66 comparable to AA Vs. 67 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 4 To enable systemic, tissue-specific delivery of payloads beyond the packaging capacity of 68 AA Vs, we developed a platform that employs surface chemistry to covalently couple AA V capsids 69 to diverse molecular delivery vehicles, resulting in oligomeric constructs that recapitulate the 70 tissue-specific tropism of AA Vs while expanding cargo capacity. We term this platform 71 CHARIOT-AA V (Crosslinked Hybrid Architectures for Robust, Interchangeable, and Organ -72 specific Targeting with AA V). Using this platform, we first address the co -delivery bottleneck 73 inherent in dual-AA V strategies by conjugating multiple AA V capsids to ensure simultaneous 74 transduction of split transgenes (Fig. 1a). The presence of Lys residues on the surfaces of AA V 75 capsids permits the chemical modification with complementary tetrazine (Tz) or trans-cyclooctene 76 (TCO) moieties via facile carbodiimide chemistry, enabling their crosslinking via inverse electron-77 demand Diels-Alder (IEDDA) click chemistry28. We applied this approach to create conjugates of 78 two AA V serotypes: AA V-DJ (AA VDJ), which efficiently transduces HEK293T cells, and AA V9, 79 which does not (Fig. 1b, Fig. S1)29. Each capsid packaged a distinct fluorescent reporter , NLS-80 GFP (nuclear localization signal -tagged green fluorescent protein ) in AA V9 and a red-emitting 81 fluorescent protein mRuby2 in AA VDJ, under a ubiquitous CAG promoter. In cultures transduced 82 with a 1:1 cocktail of AA V9-CAG::NLS-GFP and AA VDJ-CAG::mRuby2, negligible expression 83 of NLS-GFP was observed, while mRuby2 exhibited strong fluorescence (Fig. 1c). In contrast, 84 conjugation of AA V9 to AA VDJ enhanced NLS-GFP expression by 13.3 times (Fig. 1c , d; 85 p=4.10×10-3), demonstrating that AA VDJ escorted AA V9 into target cells. 86 The physical size of individual conjugates may influence transduction efficiency , as 87 excessively large clusters could interfere with AA V’s canonical endocytosis pathways or nuclear 88 entry30. We therefore fractionated AA V DJ-AA V9 conjugates by size via ultracentrifugal 89 sedimentation (Fig. 1e)31. Dynamic light scattering (DLS) confirmed separation into populations 90 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 5 with distinct hydrodynamic diameters (dh; Fig. S2). Transmission electron microscopy (TEM) of 91 selected fractions verified conjugation, and the observed cluster sizes correlated with dh (Fig. 1f, 92 g). Co-expression efficiency varied substantially with conjugate dh (Fig. 1h), and the fraction 93 exhibiting the highest co -expression rate (17.9-fold increase over the AA V cocktail; p=0.021) 94 contained the greatest proportion of AA V-AA V dimers (~40%), an architecture that ensures a 1:1 95 stoichiometry of the two serotypes. 96 97 98 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 6 99 100 We next employed AAV-AA V conjugates to deliver transgenes exceeding AA V’s packaging 101 capacity. As a first test cargo, we selected the SpCas9 gene editing cassette (~6.0 kb including the 102 AA V inverted terminal repeats ) and split it using split-intein domains , which mediate post -103 translational trans-splicing to reconstitute the full -length SpCas9 protein in situ (Fig. 2a )32–35. 104 Gene editing efficiency was assessed using StopLight HEK, a n established reporter cell line 105 (Methods), in which GFP is expressed upon indel formation at the target sequence (Fig. 2b)36. 106 Among inteins derived from diverse host proteins and organisms37–39,23, the Rma DnaE split intein 107 Fig. 1| Conjugated AA Vs for co-delivery of two transgenes . a, Scheme of the synthesis of AA V-AA V conjugates. AA Vs were functionalized with either Tz or TCO through NHS chemistry, followed by conjugation through inverse electron -demand Diels -Alder click chemistry. A representative TEM image of an AA V-AA V dimer is shown. Scale bar, 10 nm. b, Schematics of the proof-of-concept in vitro assay with two fluorescent proteins. In the cocktail condition, two AA Vs, AA V-DJ packaging pAA V-CAG-mRuby2 that transduced HEK 293T cells efficiently and AA V9 packaging CAG-NLS-GFP that did not transduce HEK 293T cells efficiently, were mixed and incubated with HEK293T cells. The same two AA Vs were conjugated and incubated with HEK293T cells in the AA V-AA V condition. c, Confocal images of HEK 293T cells incubated with the cocktail or AA V -AA V conjugates for 24 hours (multiplicity of infection (MOI) = 100). Blue – DAPI, Magenta – mRuby2 (AA V-DJ), Green - NLS-GFP (AA V9). Scale bars, 10 µm. d, mRuby2-positive and GFP-positive cell percentages in the two groups, as quantified from confocal images and with CellProfiler57. e, Schematic of sedimentation fractionation and representative TEM images of as -synthesized AA V-AA V conjugates. Scale bar, 100 nm. f, Representative TEM images of fractionated AA V -AA V conjugates with different hydrodynamic diameter (d h). Scale bars, 100 nm. g, Frequency of cluster sizes for the two selected fraction s (dh = 43, 83 nm) obtained from TEM images using a machine learning model (Detectron2; Methods). The dashed lines represent the average cluster size of each group. h, Percentage of cells that co-expressed mRuby2 and NLS-GFP after 24-hour incubation with the cocktail or one of the selected fractions having distinct hydrodynamic diameters, as measured by flow cytometry (MOI = 100). Bar plots are presented as mean ± standard error of the mean (S.E.M.). Statistical significance was assessed by Welch’s t-test (n = 3 biological replicates). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 7 exhibited the highest gene editing activity, with no significant reduction relative to wild-type 108 SpCas9 ( p=0.714; Fig. 2 c, Fig. S 3). The split SpCas9 constructs were separately packaged in 109 AA VDJ capsids, which were subsequently conjugated and fractionated as described above . In 110 StopLight HEK cells at 24 hours post -transduction, AA VDJ-AA VDJ conjugates achieved a 1.23-111 fold increase in gene editing efficiency over the cocktail (p=2.49×10-4; Fig. 2d) and required an 112 18.8% lower dose to reach 50% of the maximum gene editing activity (Fig. S4). Notably, the high 113 transduction efficiency of AA VDJ in HEK cells represents a stringent baseline, yet AA VDJ-AA VDJ 114 conjugates still yielded a 23% improvement, highlighting the benefit of vector conjugation. This 115 enhancement generalized to a CRISPRi transcriptional repressor, dSpCas9-KRAB-MeCP2 (~7 kb 116 total)40, for which AA VDJ-AA VDJ conjugates achieved a 1.21-fold greater gene repression than the 117 cocktail (p=0.0113; Fig. S5, S6). Together, these results demonstrate that AA V-AA V conjugates 118 enhance co-delivery of transgenes exceeding the packaging limit of individual AA V capsids. 119 Building on these in vitro results, we evaluated whether AA V-AA V conjugates could 120 enhance gene editing in vivo following systemic administration. Split SpCas9 cassettes were 121 packaged in AA V .CAP-B10 (AA VCAP-B10), an AA V serotype that efficiently crosses the blood-brain 122 barrier (BBB; Fig. 2 e). As a readout, t he Ai14 reporter mouse line was employed, in which 123 tdTomato expression is induced upon indel formation within the stop cassette (Fig. 2 e)41. A 124 moderate AA V dose (2.5×1010 vg/animal) that yields extensive but unsaturated gene editing was 125 selected to reduce systemic exposure and to reveal differences between the conjugate and cocktail 126 conditions7,22. Three weeks following retro-orbital delivery, tissues were harvested, and confocal 127 imaging was conducted on sagittal brain sections . Consistent with prior studies, tdTomato 128 fluorescence was observed throughout the brain of mice injected with AA VCAP-B10 cocktail (Fig. 129 2f, Fig. S7). However, the expression, as quantified by the density of tdTomato-positive cells, was 130 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 8 enhanced by 1.95 times in animals injected with AA VCAP-B10-AA VCAP-B10 conjugates as compared 131 to the cocktail-injected controls (p=0.0032; Fig. 2f-g). This result demonstrates that AA VCAP-B10-132 AA VCAP-B10 conjugates enhance systemic co-delivery of split SpCas9 to the brain across the BBB. 133 134 135 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 9 136 137 The modularity of the CHARIOT-AA V platform motivated its extension to non -AA V 138 vehicles for delivery of large unsplit genes. Lentiviral vectors (LVVs) integrate transgenes up to 139 ~10 kb into the host genome17, enabling stable expression throughout cell division , but do not 140 permit cell -type-specific transduction . Building on the observation that AA V tropism can be 141 conferred to another AA V serotype (Fig. 1b-d) and prior demonstration of AA V-mediated tissue 142 guidance of synthetic nanomaterials 28, we hypothesized that the targeting properties could be 143 similarly imparted to LVVs through AA V conjugation. Unlike AA Vs, LVVs are cloaked in lipid 144 envelopes, which are not readily functionalized via carbodiimide chemistry for IEDDA reaction 145 used in AA V-AA V conjugation. To overcome this, we employed SNAP-tag technology, in which 146 a genetically encoded protein tag (SNAP -tag) reacts with O6-benzylguanine (BG) to form a 147 Fig. 2| AA V-AA V conjugates for large gene delivery in vitro and in vivo. a, Split design of SpCas9-based gene editing cassette. The constructs were packaged separately in AA V DJ capsids. b, StopLight reporter system employed in in vitro gene editing assays. Upon CRISPR- Cas9-mediated indel formation at the target site, GFP expression was induced by frameshift mutations. c, Representative confocal images of StopLight HEK cells after lipofection with wild-type SpCas9, split SpCas9 ( Rma), or vehicle control (lipofectam ine 3000). mCherry fluorescence was also shown as a reference. Scale bars, 250 µm. d, Gene editing efficiency of AA VDJ-AA VDJ conjugates relative to the cocktail, as quantified by flow cytometry. e, Schematic of the in vivo experiment in Ai14 mice. AA V CAP-B10 capsids packaging the split SpCas9 constructs were conjugated and fractionated. The resulting AA VCAP-B10-AA VCAP-B10 conjugates were administered intravenously to Ai14 mice. Upon successful CRISPR -Cas9-mediated excision of the stop cassette, tdTomato expression was induced. f, Representative confocal images of the brain (cortex) harvested at 3 weeks post -injection from mice injected with AA VCAP-B10-AA VCAP-B10 conjugates or AA V cocktails (2.5 ×1010 vg/animal). Blue – DAPI, Green – NeuN, Magenta - tdTomato. Scale bars, 50 µm. g, Density of tdTomato-positive cells in the cortex, normalized to the cocktail group. Animals were administered the cocktail or AA VCAP-B10-AA VCAP-B10 conjugates (2.5×10 10 vg/animal), and tissues were harvested at 3 weeks post-injection. Bar plots are presented as mean ± S.E.M. Statistical significance was assessed by Welch’s t-test (n = 3 biological replicates). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 10 covalent bond (Fig. 3a)42. AA VDJ capsids were functionalized with BG via carbodiimide chemistry, 148 while LVVs displaying SNAP-tag were produced by incorporating a SNAP-tag-encoding plasmid 149 during virus packaging25. These LVVs, carrying a gene for GFP, also harbored mutant vesicular 150 stomatitis virus G -protein (VSV-G) that reduced their transduction in HEK cells 25. Following 151 crosslinking via the SNAP -tag-BG reaction (Fig. S 8), HEK293T cells were transduced with 152 AA VDJ-LVV conjugates or stoichiometric vector cocktails. At 24 hours post-transduction, ~13% 153 of cells treated with AA VDJ-LVV conjugates showed GFP expression, compared with < 1% in the 154 cocktail condition (p=4.24×10-4; Fig. 3b, c). A non-split SpCas9-based gene editing cassette (~10 155 kb) was also delivered within AA V-LVV conjugates, which yielded a significantly higher gene 156 editing rate than the cocktail group (Fig. S 9). These data demonstrate AA V-guided delivery of a 157 non-AA V vector carrying a large transgene. 158 Finally, w e extended CHARIOT -AA V to the delivery of non -viral vectors . Lipid 159 nanoparticles ( LNPs) accommodate large genetic cargo (up to ~10 kb ) as well as molecular 160 payloads (e.g. proteins), yet they commonly accumulate in the liver via apolipoprotein E (ApoE) 161 adsorption43. Although decoration of LNPs with poly(ethylene glycol) (PEG) bearing longer alkyl 162 chains, such as C18, prevents ApoE adsorption and hepatocyte transfection, thereby prolonging 163 circulation time19,44,45, extrahepatic targeting remains limited, with only a few organs accessible to 164 date26,27. We therefore sought to redirect LNPs beyond the liver using CHARIOT-AA V (Fig. 3d). 165 TCO-functionalized AA Vs and Tz-functionalized LNPs were conjugated via IEDDA click 166 chemistry (Fig. S10). DLS measurements showed a slight increase in hydrodynamic diameter from 167 101.6 nm (cocktail) to 108.8 nm (AA V -LNP) upon conjugation ( Fig. S 11), confirming that 168 excessive crosslinking did not occur. GFP mRNA was packaged in LNPs, and GFP-positive cells 169 were quantified by flow cytometry at 24 hours post -transfection. To minimize ApoE adsorption 170 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 11 and suppress AA V-independent transfection, serum-free medium was used during the procedure. 171 In the AA VDJ-LNP conjugate group, 43.3 ± 4.5% of HEK cells expressed GFP, compared with 0.6 172 ± 0.1% for the cocktail group (Fig. 3e), demonstrating that covalent AA V conjugation enables 173 targeted cellular uptake of LNPs. 174 Given their favorable long-term safety profile46, LNPs are a promising vehicle for delivery 175 of molecular therapeutics in vivo. To evaluate the ability of CHARIOT-AA V to guide LNPs toward 176 the brain, w e conjugated AA VCAP-B10 to LNPs formulated with C18 -PEG lipids, which suppress 177 unwanted hepatic uptake (Fig. 3f). In these conjugates, AA VCAP-B10 capsids packaged a fluorescent 178 protein mNeonGreen gene driven by a CAG promoter, and LNPs harbored Cre recombinase 179 mRNA. The ability of AA VCAP-B10-LNP conjugates to deliver molecular payloads to target tissues 180 was then evaluated in Ai14 mice, which carry a Cre-dependent tdTomato reporter allele driven by 181 the ubiquitous CAG promoter . AA VCAP-B10-LNP conjugates and unfunctionalized LNPs were 182 administered intravenously, and tissues were harvested five days post -injection. In the brain, 183 tdTomato expression localized to the vasculature, with significantly higher fluorescence intensity 184 in CD31⁺ cerebral vessels compared to LNP -only controls (p=0.0133; Fig. 3g -i, Fig. S 12), 185 suggesting that endothelial internalization was driven by the conjugated AA V CAP-B10 capsids. 186 Notably, the mNeonGreen carried by the AA VCAP-B10 was expressed exclusively in the brain 187 parenchyma rather than endothelial cells ( Fig. 3h). As expected, tdTomato expression was also 188 observed in the liver, consistent with residual hepatic uptake of LNPs despite C18 -PEG 189 incorporation (Fig. S 13). Importantly, mNeonGreen expression in the liver was minimal, 190 suggesting that AA VCAP-B10-LNP conjugates were preferentially directed toward the brain rather 191 than the liver. These data demonstrate that AA V capsid conjugation redirects non-AA V vehicles 192 toward defined cell populations in a serotype-dependent manner both in vitro and in vivo. 193 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 12 194 195 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 13 196 197 Here w e introduced CHARIOT-AA V , a tissue -specific systemic delivery platform for 198 genetic payloads beyond AA V’s packaging capacity . AA V-AA V conjugates achieved systemic 199 brain transduction with split SpCas9 across the BBB. Th is design is readily adaptable to other 200 dual-AA V systems, thereby providing a generalizable platform for large gene delivery. AA V-LVV 201 and AA V-LNP conjugates extend this capability to unsplit full-length genes up to ~10 kb . AA V-202 Fig. 3| Expansion of CHARIOT -AA V to lentiviral vectors and lipid nanoparticles . a, Schematic of the in vitro evaluation of AA V-LVV conjugates in HEK293T cells. During LVV packaging, mutated VSV-G and SNAP-tag membrane proteins were incorporated to minimize nonspecific transduction of HEK 293T cells and to render LVV surfaces reactive with BG, respectively. The LVVs packaged CMV-GFP. AA V-DJ capsids were functionalized with BG via NHS chemistry. In the cocktail group, intact AA V-DJ was used instead of BG-functionalized AA V-DJ. HEK293T cells were incubated with either LVV -AA V or LVV at a multiplicity of infection of 276. b, Representative confocal images of HEK 293T cells transduced with a cocktail of LVVs and AA Vs or AA V-LVV. HEK293T cells were fixed with 4% PFA at 24 hours post-transduction. Blue – DAPI, Green - GFP. Scale bars, 50 µm. c, Quantification of GFP - positive cells in the cocktail and AA V-LVV conditions. At 24 hours post -transduction, cells were harvested and analyzed by flow cytometry. d, Schematic of the in vitro experiment paradigm for LNP -AA V conjugates. LNPs packaged eGFP mRNA were functionalized with Tz. AA V-DJ was functionalized with NHS-PEG24-TCO. The functionalized LNP-Tz and AA V- TCO were conjugated, followed by quenching of the Tz and TCO groups. The resulting AA V- LNP conjugates were added to a culture of HEK293T cells in serum -free media to suppress ApoE-mediated transfection. At 24 hours post-transfection, HEK293T cells were harvested and analyzed for GFP fluorescence by flow cytometry. e, Percentage of GFP -positive cells in the cocktail and AA V-LNP groups. f, Schematic of brain -targeting AA V-LNP delivery in mice. LNPs packaging Cre mRNA were conjugated with AA V .CAP-B10. The AA VCAP-B10-LNP were administered to Ai14 mice intravenously (0.23 mg-mRNA/kg). The tissues were harvested at 5 days post-injection for confocal imaging. g, Confocal images of brain sagittal sections from mice injected with AA VCAP-B10-LNP (left) and LNP control (right). Scale bars, 1 mm. h, Confocal images of hippocampus, cortex, and thalamus from mice injected with AA VCAP-B10- LNP. Scale bars, 100 µm. i, tdTomato fluorescence intensity in CD31⁺ cerebral vasculature, normalized to the LNP group. Bar plots are presented as mean ± S.E.M. Statistical significance was assessed by Welch’s t-test (c, i: n = 3 biological replicates; e: n = 4 biological replicates). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 14 LNP further enables transient expression of mRNAs and the delivery of molecular payloads (e.g. 203 proteins), which are advantageous for gene-therapy applications to minimize off-target effects. For 204 even larger genes (e.g. full -length dystrophin cDNA for Duchenne muscular dystrophy therapy, 205 >11 kb), conjugation of three or more AA Vs holds promise. In preliminary studies, co-delivery of 206 three distinct fluorescent proteins was achieved in HEK293T cells using three-AA V conjugates 207 (Fig. S14), and the Rma DnaE and Npu DnaE inteins showed no cross-reactivity (Fig. S3), enabling 208 orthogonal splitting of a large gene across three vectors. 209 Following intravenous injection, AA VCAP-B10-LNP conjugates achieved spatially distinct 210 delivery: LNP payloads to brain endothelial cells and AA V payloads to the parenchyma. This dual 211 targeting holds therapeutic implications, as brain endothelial cells are a major cellular component 212 of the BBB, which is disrupted in several neurodegenerative 47–49 and neurodevelopmental 50 213 conditions. These data also provide mechanistic insight into the fate of LNPs during AA VCAP-B10-214 mediated transcytosis. The conjugated LNPs likely release their mRNA cargo within endothelial 215 cells via the LNPs’ intrinsic endosomal escape machinery. To extend LNP payload delivery beyond 216 the endothelium, future efforts could engineer LNPs with delayed endosomal escape kinetics or 217 employ alternative vector systems that lack active endosomal escape mechanisms. While scalable 218 manufacturing, including stoichiometry control, remains to be established for therapeutic 219 applications, CHARIOT-AA V reconceptualizes AA V capsids as modular targeting elements. By 220 decoupling tissue-specific tropism from cargo capacity, this platform enables tissue- and cell-type-221 specific delivery of diverse therapeutics and research tools beyond the limitations of individual 222 vectors. 223 224 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 15

225 Reagents 226 Sucrose (BioXtra grade, ≥99.5%; S7903) was purchased from Sigma -Aldrich. O6-benzylguanine 227 (BG)-GLA-NHS (#S9151S) and BG -PEG-NH2 (#S9150S) were purchased from New England 228 Biolabs. Methoxy poly(ethylene glycol)24 -NHS (mPEG24 -NHS; #BP -23970), trans-229 cycrooctene-PEG24-NHS (TCO -PEG24-NHS; #BP -26353), tetrazine -PEG5-NHS ester (Tz -230 PEG5-NHS; #BP22681), mPEG -methyltetrazine (Mw 2000; mPEG -Tz; #BP -26352), and 231 mPEG4-TCO (#BP27872) were purchased from BroadPharm. All lipid components used for lipid 232 nanoparticle (LNP) formulation were purchased from commercial suppliers. The ionizable lipid 233 ALC-0315 (#890900), 1,2 -distearoyl-sn-glycero-3-phosphocholine (DSPC ; # 850365), 1,2 -234 dimyristoyl-sn-glycero-3-phosphoethanolamine-N-[methoxy(polyethylene glycol)-2000] (DMG-235 mPEG2000; # 880151), and 1,2 -distearoyl-sn-glycero-3-phosphoethanolamine-N-236 [methoxy(polyethylene glycol) -2000] (DSG -mPEG2000; # 880152) were obtained from Avanti 237 Polar Lipids. Cholesterol was purchased from Millipore Sigma ( #C8667). The functionalized 238 PEG-lipid 1,2 -distearoyl-sn-glycero-3-phosphoethanolamine-N-[methoxy(polyethylene glycol) -239 2000]-methyltetrazine was purchased from BroadPharm (DSPE-PEG2000-methyltetrazine; #BP-240 43749). N1-Methylpseudouridine (N1mU)-modified m RNAs encoding Cre recombinase and 241 enhanced green fluorescent protein (eGFP) were purchased from TriLink Biotechnologies (#L-242 8111) and GenScript, respectively (#RP-A00009). 243 Cloning of constructs 244 NEB 5-alpha Competent E. coli (High Efficiency; NEB C2987H) was used in transformation for 245 all genetic constructs. Miniprep was performed using NucleoSpin Plasmid EasyPure (Takara Bio 246 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 16 #740727.50). All plasmid sequences were confirmed by whole-plasmid sequencing performed by 247 Plasmidsaurus. Parent plasmids are listed in Table 1, and new construct employed in this study can 248 be found in the Supplementary Information. 249 To construct StopLight HEK , synthetic gene fragments containing mCherry-F2A-stop-250 EGFP-stop-EGFP (IDT, gBlocks Gene Fragments)36 were inserted into pMK231 (AA VS1-CMV-251 MCS-PURO, Addgene #105924) via Gibson assembly using NEBuilder HiFi DNA Assembly 252 Master Mix (NEB #E2621L). 253 To create split SpCas9, t he Rma51, Npu38,52, and Cfa39 DnaE split intein genes were 254 synthesized (IDT, gBlocks Gene Fragments). Split SpCas9 sequences were amplified from pX330, 255 and these fragments were assembled into the pAA V backbone including an sgRNA cassette (see 256 Supplementary Methods). 257 The split dSpCas9-KRAB-MeCP2 constructs were generated from dCas9-KRAB-MeCP2 258 (Addgene #110821). Rma DnaE split intein sequences were amplified from the split SpCas9 (Rma) 259 constructs by PCR using Q5 Hot Start High -Fidelity 2X Master Mix (NEB #M0494S) . dCas9-260 KRAB-MeCP2 was also split at several sites ( L833/S834, Q844/S845, D850/S851, R859/S860, 261 K867/S868, P872/S873, and L909/S910) by PCR. The resulting gene fragments were assembled 262 via Gibson assembly (3 fragments for the C-terminal split construct; 5 fragments for the N-terminal 263 split construct; see Supplementary Methods). 264 The sgRNA sequence for the StopLight HEK experiment was obtained from the original 265 article36. The same sgRNA sequence was used for both split constructs. The sgRNA sequences for 266 the CAG promoter were designed using CRISPick53,54, and the top two sequences were used in the 267 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 17 two split constructs. The sgRNA sequence for Ai14 mouse experiments w as obtained from a 268 previous study41. 269 270 Table 1| Base plasmids 271 Name Source Comment pUCmini-iCAP-AA V .CAP-B10 Addgene plasmid #175004 Gifts from Viviana Gradinaru pAA V-CAG-mRuby2 Addgene plasmid #99123 Gifts from Viviana Gradinaru pAA V-CAG-NLS-GFP Addgene plasmid #104061 Gifts from Viviana Gradinaru pAA V-CAG-mTurquoise2 Addgene Plasmid #99122 Gifts from Viviana Gradinaru pAA V2/9n Addgene plasmid #112865 Gift from James M. Wilson pMK231 (AA VS1 CMV-MCS-PURO) Addgene plasmid #105924 Gift from Masato Kanemaki PX458-AA VS1 Addgene plasmid #113194 Gift from Adam Karpf AA VS1-Pur-CAG-EGFP Addgene plasmid #80945 Gift from Su-Chun Zhang N-Terminal Split Cas9 with GyrA intein Addgene plasmid #58693 Gift from Gang Bao C-Terminal Split Cas9 with GyrA intein Addgene plasmid #58694 Gift from Gang Bao dCas9-KRAB-MeCP2 Addgene plasmid #110821 Gift from Alejandro Chavez & George Church pX330-U6-Chimeric_BB-CBh-hSpCas9 Addgene plasmid # 42230 Gift from Feng Zhang pHCMV-VSVgSS-SNAP-TM Addgene plasmid #207319 Gift from Feng Zhang pMD2.G-K47Q/R354Q Addgene plasmid #207323 Gift from Feng Zhang lentiCRISPR v2 Addgene plasmid # 52961 Gift from Feng Zhang pMDLg/pRRE Addgene plasmid #12251 Gift from Didier Trono pRSV-Rev Addgene plasmid #12253 Gift from Didier Trono pLenti CMV GFP Puro (658-5) Addgene plasmid #17448 Gift from Eric Campeau & Paul Kaufman pAA V-DJ Cell Biolabs, Inc. #VPK-420-DJ pHelper Cell Biolabs, Inc. RV01-pRGS-Reporter with GFP PNA bio 272 273 Cell culture and cloning of GFP-expressing and StopLight HEK293T 274 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 18 HEK293T cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM, Gibco #10569044) 275 with 10% fetal bovine serum (FBS, Cytovia #SH30396.03HI). 1% Penicillin-Streptomycin (P/S) 276 was also supplemented except for lentiviral vector production. Cells were passaged at 90% 277 confluency with TrypLE Express (Gibco #12605028). 278 To obtain GFP-expressing HEK and Stoplight HEK cells, HEK293T cells were cultured in 279 6-well plates in DMEM with 10% FBS and 1% P/S and dissociated with TrypLE Express for 5 280 min. Then, a total of 2×106 cells were co-transfected with 10 µg of PX458-AA VS1 plasmid, which 281 expresses SpCas9, and 6 µg of AA VS1-CMV-Stoplight-Puro or AA VS1-Pur-CAG-EGFP in 100 282 µL of Opti -MEM (Gibco #31985088) . Electroporation was performed using a BTX ECM830 283 electroporator (Poring plus: voltage, 150 V; pulse length, 5 ms; pulse, 100 ms; number of pulses, 284 2. Transfer pulse: voltage, 20 V; pulse length, 100 ms; pulse, 100 ms; number of pulses, 5). After 285 electroporation, HEK293T cells were seeded in DMEM with 10% FBS and 1% P/S, and 24 hours 286 post-electroporation, transfected cells were treated with puromycin at 1.5 μg/ml to select genome-287 edited cells. Then, the cells were further purified with a cell sorter (FACS BD Melody) according 288 to mCherry or EGFP positive expression, followed by genotyping to validate the correct genome 289 insertion and sequence. 290 Evaluation of split constructs by lipofection 291 Prepared split constructs were evaluated in StopLight HEK or GFP -expressing HEK cells by 292 lipofection. Wild-type constructs (i.e. SpCas9, dSpCas9 -KRAB-MeCP2) were used as 293 benchmarks. Cells were prepared in a 96 -well plate and used at ~80% confluency. 6 µL of 294 Lipofectamine 3000 (Thermo Fisher Scientific # L3000001) and 150 µL of Opti -MEM (Gibco 295 #31985062) was mixed. In another tube, 150 µL of Opti -MEM, 6 µL of P3000 reagent, and 75 296 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 19 fmol of plasmid DNAs were mixed. The two solutions were mixed and incubated at room 297 temperature for 10 min, and then added to 3 wells (100 µL/well). At 48 hours post-lipofection, the 298 cells were harvested and analyzed by flow cytometry (BD FACSymphony A3). 299 AAV packaging 300 AA Vs were packaged in the lab following an established protocol55. Briefly, HEK293T cells were 301 cultured in DMEM with 10% FBS and 1% P/S. At 95% confluency, triple transfection with a 302 capsid plasmid, a helper plasmid, and a plasmid encoding the gene of interest within the pAA V 303 backbone was performed using PEI Max. The medium was exchanged with fresh DMEM with 5% 304 FBS at 10 hours post-transfection. The culture medium was harvested at 72 hours post-transfection 305 and fresh 5% FBS DMEM was added. At 120 hours post-transfection, both medium and cells were 306 harvested. AA V capsids inside the cells were released using salt active nuclease (HL -SAN, 307 ArcticZyme #70910 -202). AA V capsids in the medium were collected by PEG -mediated 308 precipitation. Collected AA V capsids were purified by ultracentrifugation (350,000 ×g, 2 hour s, 309 25 min, 18 °C) in an iodixanol density gradient (OptiPrep, STEMCELL TECHNOLOGIES 310 #07820). Purified AA Vs were stored in Dulbecco’s phosphate-buffered saline (DPBS) (Corning 311 #20-031-CV) with 0.1% Pluronic F -68 (Gibco #24040032) (DPBS-F68). AA V titers were 312 measured by quantitative PCR (qPCR) using an AA V real-time PCR titration kit (Takara Bio 313 #6233). 314 AAV modification 315 AA V capsids were functionalized with Tz, TCO, or BG, depending on the CHARIOT-AA V system, 316 via NHS chemistry following a previously reported protocol 28. Briefly, AA Vs were suspended in 317 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 20 pH 8.4 sodium bicarbonate solution in DPBS -F68 using filter centrifuge units (Amicon Ultra 318 Centrifugal Filter, 100 kDa MWCO, Millipore #UFC510024). 2 pmol of AA V capsids were reacted 319 with ligands at a ligand-to-lysine residue ratio of 0.1 (AA V-DJ and AA V9) and 0.075 (AA V .CAP-320 B10). The solution was allowed to react at 4°C overnight. The resulting functionalized AA Vs were 321 purified with DPBS-F68 using the filter centrifuge units. Once purified, functionalized AA Vs were 322 titrated by qPCR. 323 AAV-AAV conjugation and fractionation 324 Tz-functionalized AA V capsids and TCO-functionalized AA V capsids were reacted in DPBS-F68 325 and conjugated via the inverse electron -demand Diels-Alder cycloaddition click chemistry. The 326 ratio of the two capsids was fixed at 1:1. Typically, functionalized AA V solutions were adjusted to 327 10 nM; 50 µL of each AA V solution was then mixed so that the final concentration of each AA V 328 was 5 nM. The solution was stored at 4 °C overnight. To stop the reaction, 50 µL of mPEG4-TCO 329 (0.1 mg/mL) was first added and allowed to react for 4 hours at 4 °C. Subsequently, 140 µL of 330 mPEG2k-Tz (1 mg/mL) was added and allowed to react for another 4 hours at 4 °C. The resulting 331 AA V-AA V conjugates were purified by medium exchange with DPBS-F68 using a filter centrifuge 332 unit (Amicon Ultra Centrifugal Filter, 0.5 mL, 100 kDa MWCO, Millipore #UFC510024) . The 333 volume was adjusted to ~200 µL after the final centrifugation step, then loaded onto a sucrose 334 gradient (8%-4 mL, 10%-4 mL, 12.5%-6 mL, 15%-6 mL, 17.8%-4 mL, 20%-2 mL, 25%-2 mL, 335 and 40%-2 mL; in a 30 mL sterile, open -top ultracentrifuge tube (Beckman Coulter # C14307)). 336 The solution was centrifuged in an SW 32 Ti Swinging -Bucket Rotor (Beckman Coulter) at 337 150,000 ×g for 1.5 hours at 4 °C. The solution was fractionated into 1 mL from the bottom of the 338 tube using a syringe, and the titer of each fraction was determined by qPCR . Prior to DLS 339 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 21 measurement, the fractionated solutions were medium -exchanged with DPBS -F68 to remove 340 remaining sucrose. The titer was again determined after the medium exchange step. 341 Transmission electron microscopy of AA V-AA V conjugates was performed on an FEI G2 Spirit 342 TWIN TEM. Fifty images were collected for each sample. Conjugate size analysis was performed 343 using Detectron2 (model: Mask R -CNN X101-FPN, model ID: 139653917)56. The model was 344 trained with human-annotated data (TEM images of AA V particles, ~1000 particles in total). 345 AAV-AAV transduction test in vitro and in vivo 346 For in vitro experiments, fractionated AA V-AA V conjugates were added to a culture of HEK293T, 347 StopLight HEK, or GFP -expressing HEK cells at varying MOI values. Transduction durations 348 were 24 hours for the two fluorescent protein transduction tests (Fig. 1) and the split SpCas9 gene 349 editing tests (Fig. 2) and 72 hours for the CAG promoter suppression test (Fig. S 6). Cells were 350 harvested and measured by flow cytometry (BD FACSymphony A3). 351 For in vivo experiments, AA V-AA V conjugates were intravenously injected via the retro-352 orbital route into 6-week-old Ai14 mice (2 males and 1 female; 2.5×1010 vg/animal). Three weeks 353 post injection, the animals were euthanized and perfused with PBS and 4% paraformaldehyde 354 (PFA) in PBS. Harvested tissues were further fixed in 4% PFA overnight. Fixed tissues were 355 sectioned at a thickness of 50 µm in the sagittal plane using a Vibratome. The obtained brain slices 356 were permeabilized in 0.3% Triton X -100 for 30 min, followed by blocking with 3% normal 357 donkey serum (NDS) for 2 hours at room tempearture. Primary antibody labeling was performed 358 in 3% NDS at 4 °C overnight (Anti NeuN antibody (host: rabbit), 1:500, Proteintech #26975-1-359 AP). After three cycles of washing with PBS, the slices were stained with secondary antibodies in 360 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 22 3% NDS for 2 hours at room temperature (Donkey anti-Rabbit IgG (H+L) Highly Cross-Adsorbed 361 Secondary Antibody, Alexa Fluor 647, 1:1,000, Thermo Fisher Scientific #A-31573). The sections 362 were subsequently stained with DAPI for 30 min (1:20,000), and then mounted on glass slides 363 using Fluoromount-G. Confocal imaging was performed on Leica Stellaris 5 with a 10x objective 364 lens. CellProfiler57 was used for quantification. 365 For both in vitro and in vivo experiments, intact AA Vs were used for the cocktail conditions, and 366 the solutions were adjusted to the same titer before use. 367 Lentiviral vector production 368 HEK293T cells were prepared in DMEM with 10% FBS in three 15-cm plates and were transfected 369 with a mixture of plasmids using PEI Max (Polysciences, #24765). Packaging plasmids 370 (pMDLg/pRRE (Addgene 12251) and pRSV .Rev (Addgene 12253)), a transfer plasmid (pLenti 371 CMV GFP Puro (658 -5) (Addgene 17448)), and envelope plasmids (pMD2.G -K47Q/R354Q 372 (Addgene 207323) and pHCMV-VSVgSS-SNAP-TM (Plasmid 207319)) were mixed at the ratio 373 of pMDLg/pRRE:pRSV .Rev:transfer plasmid:pMD2.G-K47Q-R354Q:pHCMV-VSVgSS-SNAP-374 TM=1:1:3:0.75:0.25. At 8 hours post -transfection, the media was exchanged with fresh DMEM 375 with 10% FBS. At 48 hours post-transfection, the media containing lentiviral vectors (LVVs) was 376 harvested, followed by filtration with a 0.45 µm PES membrane filter. The solution was then 377 concentrated by ultracentrifugation (50,000 ×g, 30 min, 4 °C). The viral pellet was resuspended 378 in PBS. The titer of LVVs was determined by RT-qPCR (Lenti-X qRT-PVR Titration Kit; #631235, 379 Takara bio). 380 AAV-LVV conjugation and in vitro transduction test 381 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 23 SNAP-tag-functionalized LVVs and BG -functionalized AA Vs were conjugated in DPBS-F68 at 382 an AA V/LVV ratio of 100 (100 AA Vs per LLV). The reaction was stopped at different time points 383 by adding mPEG24 -BG, which was prepared by mixing BG -PEG-NH2 and mPEG24 -NHS at a 384 molar ratio of 1:1.5 in pH 8.4 sodium bicarbonate in DPBS -F68 at 4 °C overnight. Based on the 385 optimization results (Fig. S9), 4 hours was selected as the reaction time . Prepared AA V-LVV 386 conjugates were tested in HEK293T or StopLight HEK cells at an MOI of 150. 24 hours post -387 transduction, HEK cells were harvested and analyzed for GFP fluorescence by flow cytometry 388 (BD FACSymphony A3). 389 Lipid nanoparticle production 390 Lipid nanoparticles (LNPs) were prepared via microfluidic mixing as previously described 58. 391 Briefly, an ethanol phase containing ALC-0315, DSPC, cholesterol, PEG-lipid, and functionalized 392 PEG-lipid (DSPE-PEG2000-mTztetrazine) was prepared. The molar ratio of ionizable lipid, DSPC, 393 cholesterol, and total PEG -lipid (including functionalized PEG -lipid) was 50:10:38.5:1.5. For in 394 vitro experiments, the PEG -lipid component consisted of DMG -mPEG2000, whereas DSG -395 mPEG2000 was used for in vivo studies. The DSPE -PEG2000-mTz was included at a 396 concentration of 0.005-0.5 mol% of the total 1.5 mol% PEG-lipid content. 397 This ethanol phase was rapidly mixed with an aqueous solution of RNA dissolved in 10 mM citrate 398 buffer (pH 3.0) at a volumetric aqueous -to-ethanol ratio of 3:1. Following microfluidics mixing, 399 LNPs were dialyzed for at least 18 hours at 4 °C against phosphate-buffered saline (PBS) using 400 Slide-A-Lyzer dialysis devices with a 10 kDa molecular weight cut -off (ThermoFisher). 401 Nanoparticle size and polydispersity index (PDI) were measured using a Malvern Zetasizer. RNA 402 encapsulation efficiency was assessed via the Quant-iT RiboGreen assay (ThermoFisher). 403 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 24 AAV-LNP conjugation and transduction evaluation 404 Tz-functionalized LNPs and TCO-functionalized AA Vs were conjugated through the IEDDA click 405 chemistry in DPBS -F68. The number ratio of LNP particles and AA V capsids was 1:0.75 406 (LNP:AA V), and the concentrations of LNPs and AA Vs were 20 nM and 10 nM, respectively. The 407 reaction was quenched at different time points by adding Tz-PEG2k and mPEG4-TCO 408 subsequently. Typically, 100 µL of LNP solution and 150 µL of AA V solution was mixed and 409 allowed to react for 15 min at 4 °C. The functional groups were quenched with 150 µL of mPEG2k-410 Tz (0.01 mg/mL) for 4 hours at 4 °C and subsequently with 270 µL of mPEG4-TCO (0.1 mg/mL) 411 for 4 hours at 4 °C. The resulting AA V-LNP conjugates were dialyzed for overnight at 4 °C against 412 DPBS-F68 using Slide-A-Lyzer dialysis devices with a 10 0 kDa molecular weight cut -off 413 (ThermoFisher). 414 In in vitro experiments, HEK293T cells were treated with AA V-LNP conjugates in serum -free 415 DMEM. The LNP particle -to-cell ratio was 3160. Two hours post -transfection, the medium was 416 exchanged with 10% FBS DMEM. At 24 hours post -transfection, cells were harvested and 417 measured for GFP expression by flow cytometry (BD FACSymphony A3). 418 In in vivo experiments, AA V-LNP conjugates were injected into Ai14 mice intravenously via the 419 retro-orbital route (2 males and 1 female per group; 0.23 mg mRNA/animal). Five days post -420 injections, the mice were euthanized and perfused with PBS and 4% PFA. The harvested tissues 421 were further fixed in 4% PFA overnight. The brain tissues were sectioned in a sagittal plane at 50 422 µm using a Vibratome. The brain slices were permeabilized with 0.3% Triton X-100 for 30 min, 423 followed by blocking with 3% normal donkey serum (NDS) at 4 °C overnight. The tissues were 424 then labeled with anti-CD31 antibody (Rat Anti-Mouse CD31, BD Pharmingen 550274, 1:200) at 425 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 25 4 °C overnight. Secondary antibody staining was performed at room temperature on a shaker for 426 1 hour ( Donkey anti-Rat IgG (H+L) Highly Cross -Adsorbed Secondary Antibody, Alexa Fluor 427 Plus 647, Thermo Fisher Scientific A48272, 1:1000). After 2 washing cycles with PBS, the brain 428 slices were incubated with DAPI (1:20,000) for 3 min. Following 2 washing cycles, the sections 429 were mounted on glass slides using Fluoromount -G. Confocal imaging was performed on Leica 430 Stellaris 5 with a 10x objective lens. Quantification of fluorescence was conducted using 431 CellProfiler457. 432 433

434 We thank the Koch Institute's Robert A. Swanson (1969) Biotechnology Center for technical 435 support, C. Hallee (the Genomics Core) and the Flow Cytometry Core. This work made use of the 436 MRSEC Shared Experimental Facilities at MIT, supported by the National Science Foundation 437 under award number DMR -1419807. This work was funded in part by the Pioneer Award from 438 the National Institutes of Health and National Institute for Complementary and Integrative Health 439 (DP1-AT011991, P.A.), McGovern Institute for Brain Research at MIT, K. Lisa Yang and Hock 440 E. Tan Center for Molecular Therapeutics at MIT (P.A.) , NIH grants R01MH085802 and 441 R01NS130361 (M.S.), MURI grant W911NF2110328 (M.S.), the Picower Institute Innovation 442 Fund (M.S.), and the Simons Foundation Autism Research Initiative through the Simons Center 443 for the Social Brain (M.S.). R.J.M. acknowledges support from the Army Research Office under 444 award W911NF -23-2-0101. K.N. is a recipient of a scholarship from the Honjo International 445 Scholarship Foundation and the Y. Eva Tan Fellowship Program. J.L.B. acknowledges support by 446 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 26 the Schmidt Science Fellows program, in partnership with the Rhodes Trust . S.T., C.G.L., and 447 D.G.A. acknowledge support from the sponsored research funding from FUJIFILM Corporation. 448 449 Author Contributions 450 K.N., T.O. and P.A. designed the study. K.N. and E.V .P. produced AA Vs and LVVs. K.N. prepared 451 and characterized AA V-AA V , AA V-LVV , and AA V-LNP conjugates and performed in vivo 452 experiments. K.N. and T.O. performed in vitro experiments and cloning. T.O. prepared StopLight 453 HEK and GFP-expressing HEK cells. S.T. and C.G.L. prepared and characterized LNPs. K.N. and 454 E.C.G. trained the machine learning model for TEM image analysis. S.M. prepared Ai14 mice for 455 in vivo experiments. K.N., E.V .P., and J.L.B. prepared cells in vitro experiments. K.N., T.O., M.S., 456 D.G.A., and P.A. analyzed the data. All authors have contributed to the writing of the manuscript. 457 458 Competing interest statement 459 S.T. is an employee of FUJIFILM Pharmaceuticals U.S.A., Inc. D.G.A is a founder of CRISPR 460 Therapeutics, Sigilon Therapeutics, Combined Therapeutics, Orna Therapeutics, and Souffle 461 Therapeutics, and has grants from FUJIFILM Corporation and Sanofi. K.N. and P.A. are co -462 inventors on a pending US patent application related to this work. 463 464 465 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 27

466 1. Madigan, V ., Zhang, F. & Dahlman, J. E. Drug delivery systems for CRISPR -based genome 467 editors. Nat Rev Drug Discov 22, 875–894 (2023). 468 2. Mitchell, M. J. et al. Engineering precision nanoparticles for drug delivery. Nat Rev Drug 469 Discov 20, 101–124 (2021). 470 3. Wang, D., Tai, P. W. L. & Gao, G. Adeno-associated virus vector as a platform for gene therapy 471 delivery. Nat Rev Drug Discov 18, 358–378 (2019). 472 4. Li, C. & Samulski, R. J. Engineering adeno-associated virus vectors for gene therapy. Nat Rev 473 Genet 21, 255–272 (2020). 474 5. Sahel, J.-A. et al. Partial recovery of visual function in a blind patient after optogenetic therapy. 475 Nat Med 27, 1223–1229 (2021). 476 6. Wang, J.-H., Gessler, D. J., Zhan, W., Gallagher, T. L. & Gao, G. Adeno-associated virus as a 477 delivery vector for gene therapy of human diseases. Sig Transduct Target Ther 9, 78 (2024). 478 7. Goertsen, D. et al. AA V capsid variants with brain-wide transgene expression and decreased 479 liver targeting after intravenous delivery in mouse and marmoset. Nat Neurosci (2021) 480 doi:10.1038/s41593-021-00969-4. 481 8. Tabebordbar, M. et al. Directed evolution of a family of AA V capsid variants enabling potent 482 muscle-directed gene delivery across species. Cell 184, 4919-4938.e22 (2021). 483 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 28 9. Moyer, T. C. et al. Highly conserved brain vascular receptor ALPL mediates transport of 484 engineered AA V vectors across the blood-brain barrier. Molecular Therapy 33, 3902–3916 485 (2025). 486 10. Chuapoco, M. R. et al. Adeno-associated viral vectors for functional intravenous gene transfer 487 throughout the non-human primate brain. Nat. Nanotechnol. 18, 1241–1251 (2023). 488 11. Chen, X. et al. Engineered AA Vs for non-invasive gene delivery to rodent and non -human 489 primate nervous systems. Neuron 110, 2242-2257.e6 (2022). 490 12. Wu, Z., Yang, H. & Colosi, P. Effect of Genome Size on AA V Vector Packaging. Molecular 491 Therapy 18, 80–86 (2010). 492 13. Xu, X. et al. Engineered miniature CRISPR -Cas system for mammalian genome regulation 493 and editing. Molecular Cell 81, 4333-4345.e4 (2021). 494 14. Rauch, B. J. et al. Single-AA V CRISPR editing of skeletal muscle in non-human primates with 495 NanoCas, an ultracompact nuclease. Preprint at https://doi.org/10.1101/2025.01.29.635576 496 (2025). 497 15. Thelen, J. L. et al. Morphological Characterization of Self -Amplifying mRNA Lipid 498 Nanoparticles. ACS Nano 18, 1464–1476 (2024). 499 16. Geall, A. J. et al. Nonviral delivery of self -amplifying RNA vaccines. Proc. Natl. Acad. Sci. 500 U.S.A. 109, 14604–14609 (2012). 501 17. Kumar, M., Keller, B., Makalou, N. & Sutton, R. E. Systematic Determination of the 502 Packaging Limit of Lentiviral Vectors. Human Gene Therapy 12, 1893–1905 (2001). 503 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 29 18. Bouard, D., Alazard‐Dany, N. & Cosset, F. Viral vectors: from virology to transgene 504 expression. British J Pharmacology 157, 153–165 (2009). 505 19. Hou, X., Zaks, T., Langer, R. & Dong, Y . Lipid nanoparticles for mRNA delivery. Nat Rev 506 Mater 6, 1078–1094 (2021). 507 20. Mittas, D. M. et al. Dual AA V vectors for efficient delivery of large transgenes. Nat Protoc 508 (2025) doi:10.1038/s41596-025-01243-8. 509 21. Lin, J. et al. AA VLINK: A potent DNA-recombination method for large cargo delivery in gene 510 therapy. Cell (2026) doi:10.1016/j.cell.2025.12.039. 511 22. Davis, J. R. et al. Efficient in vivo base editing via single adeno-associated viruses with size-512 optimized genomes encoding compact adenine base editors. Nat. Biomed. Eng 6, 1272–1283 513 (2022). 514 23. Chen, Y . et al. Development of Highly Efficient Dual‐AA V Split Adenosine Base Editor for In 515 Vivo Gene Therapy. Small Methods 4, 2000309 (2020). 516 24. Kuzmin, D. A. et al. The clinical landscape for AA V gene therapies. Nat Rev Drug Discov 20, 517 173–174 (2021). 518 25. Strebinger, D. et al. Cell type-specific delivery by modular envelope design. Nat Commun 14, 519 5141 (2023). 520 26. Cheng, Q. et al. Selective organ targeting (SORT) nanoparticles for tissue -specific mRNA 521 delivery and CRISPR–Cas gene editing. Nat. Nanotechnol. 15, 313–320 (2020). 522 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 30 27. Chen, M. Z. et al. A versatile antibody capture system drives specific in vivo delivery of 523 mRNA-loaded lipid nanoparticles. Nat. Nanotechnol. 20, 1273–1284 (2025). 524 28. Nagao, K., Macfarlane, R. J. & Anikeeva, P. Adeno-associated viruses escort nanomaterials to 525 specific cells and tissues. bioRxiv (2025) doi: https://doi.org/10.1101/2025.04.04.647267. 526 29. Grimm, D. et al. In Vitro and In Vivo Gene Therapy Vector Evolution via Multispecies 527 Interbreeding and Retargeting of Adeno-Associated Viruses. J Virol 82, 5887–5911 (2008). 528 30. Burns, A. & Datta, S. Improving Aggregation Control of Recombinant Adeno‐Associated 529 Virus Serotype 2 (rAA V2) With Small Sugars and Ionic Salts. Biotechnology Journal 20, 530 e70157 (2025). 531 31. Ayuso, E. et al. High AA V vector purity results in serotype - and tissue -independent 532 enhancement of transduction efficiency. Gene Ther 17, 503–510 (2010). 533 32. Shah, N. H. & Muir, T. W. Inteins: nature’s gift to protein chemists. Chem. Sci. 5, 446–461 534 (2014). 535 33. Wood, D. W., Belfort, M. & Lennon, C. W. Inteins—mechanism of protein splicing, emerging 536 regulatory roles, and applications in protein engineering. Front. Microbiol. 14, 1305848 (2023). 537 34. Wang, H., Wang, L., Zhong, B. & Dai, Z. Protein Splicing of Inteins: A Powerful Tool in 538 Synthetic Biology. Front. Bioeng. Biotechnol. 10, 810180 (2022). 539 35. Tasfaout, H. et al. Split intein-mediated protein trans -splicing to express large dystrophins. 540 Nature 632, 192–200 (2024). 541 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 31 36. De Jong, O. G. et al. A CRISPR -Cas9-based reporter system for single -cell detection of 542 extracellular vesicle-mediated functional transfer of RNA. Nat Commun 11, 1113 (2020). 543 37. Fine, E. J. et al. Trans-spliced Cas9 allows cleavage of HBB and CCR5 genes in human cells 544 using compact expression cassettes. Sci Rep 5, 10777 (2015). 545 38. Truong, D.-J. J. et al. Development of an intein-mediated split–Cas9 system for gene therapy. 546 Nucleic Acids Res 43, 6450–6458 (2015). 547 39. Stevens, A. J. et al. Design of a Split Intein with Exceptional Protein Splicing Activity. J. Am. 548 Chem. Soc. 138, 2162–2165 (2016). 549 40. Yeo, N. C. et al. An enhanced CRISPR repressor for targeted mammalian gene regulation. Nat 550

15, 611–616 (2018). 551 41. Staahl, B. T. et al. Efficient genome editing in the mouse brain by local delivery of engineered 552 Cas9 ribonucleoprotein complexes. Nat Biotechnol 35, 431–434 (2017). 553 42. Juillerat, A. et al. Directed Evolution of O6-Alkylguanine-DNA Alkyltransferase for Efficient 554 Labeling of Fusion Proteins with Small Molecules In Vivo. Chemistry & Biology 10, 313–317 555 (2003). 556 43. Hosseini-Kharat, M., Bremmell, K. E. & Prestidge, C. A. Why do lipid nanoparticles target the 557 liver? Understanding of biodistribution and liver -specific tropism. Molecular Therapy 558

& Clinical Development 33, 101436 (2025). 559 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 32 44. Mui, B. L. et al. Influence of Polyethylene Glycol Lipid Desorption Rates on Pharmacokinetics 560 and Pharmacodynamics of siRNA Lipid Nanoparticles. Molecular Therapy - Nucleic Acids 2, 561 e139 (2013). 562 45. Shi, D., Toyonaga, S. & Anderson, D. G. In Vivo RNA Delivery to Hematopoietic Stem and 563 Progenitor Cells via Targeted Lipid Nanoparticles. Nano Lett. 23, 2938–2944 (2023). 564 46. Fang, E. et al. Advances in COVID -19 mRNA vaccine development. Sig Transduct Target 565 Ther 7, 94 (2022). 566 47. Sweeney, M. D., Sagare, A. P. & Zlokovic, B. V . Blood–brain barrier breakdown in Alzheimer 567 disease and other neurodegenerative disorders. Nat Rev Neurol 14, 133–150 (2018). 568 48. Gray, M. T. & Woulfe, J. M. Striatal Blood–Brain Barrier Permeability in Parkinson’S Disease. 569 J Cereb Blood Flow Metab 35, 747–750 (2015). 570 49. Zlokovic, B. V . Neurovascular pathways to neurodegeneration in Alzheimer’s disease and 571 other disorders. Nat Rev Neurosci 12, 723–738 (2011). 572 50. Osaki, T. et al. miR126-mediated alteration of vascular integrity in Rett syndrome. Mol 573 Psychiatry https://doi.org/10.1038/s41380-026-03492-9 (2026) doi:10.1038/s41380 -026-574 03492-9. 575 51. Chen, Y . et al. Development of Highly Efficient Dual‐AA V Split Adenosine Base Editor for In 576 Vivo Gene Therapy. Small Methods 4, 2000309 (2020). 577 52. Iwai, H., Züger, S., Jin, J. & Tam, P.-H. Highly efficient protein trans ‐splicing by a naturally 578 split DnaE intein from Nostoc punctiforme. FEBS Letters 580, 1853–1858 (2006). 579 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint 33 53. Doench, J. G. et al. Optimized sgRNA design to maximize activity and minimize off -target 580 effects of CRISPR-Cas9. Nat Biotechnol 34, 184–191 (2016). 581 54. Sanson, K. R. et al. Optimized libraries for CRISPR -Cas9 genetic screens with multiple 582 modalities. Nat Commun 9, 5416 (2018). 583 55. Challis, R. C. et al. Systemic AA V vectors for widespread and targeted gene delivery in rodents. 584 Nat Protoc 14, 379–414 (2019). 585 56. Wu, Y ., Kirillov, A., Massa, F., Lo, W.-Y . & Girshick, R. Detectron2. (2019). 586 57. Stirling, D. R. et al. CellProfiler 4: improvements in speed, utility and usability. BMC 587 Bioinformatics 22, 433 (2021). 588 58. Chen, D. et al. Rapid Discovery of Potent siRNA-Containing Lipid Nanoparticles Enabled by 589 Controlled Microfluidic Formulation. J. Am. Chem. Soc. 134, 6948–6951 (2012). 590 591 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 15, 2026. ; https://doi.org/10.64898/2026.04.13.718054doi: bioRxiv preprint