Intravenous functional gene transfer throughout the brain of non-human primates using AAV

Adeno-associated viruses (AAVs) promise robust gene delivery to the brain through non-invasive, intravenous delivery. However, unlike in rodents, few neurotropic AAVs efficiently cross the blood-brain barrier in non-human primates (NHPs). Here we describe AAV.CAP-Mac, an engineered variant identified by screening in adult marmosets and newborn macaques with improved efficiency in the brain of multiple NHP species: marmoset, rhesus macaque, and green monkey. CAP-Mac is neuron-biased in infant Old World primates, exhibits broad tropism in adult rhesus macaques, and is vasculature-biased in adult marmosets. We demonstrate applications of a single, intravenous dose of CAP-Mac to deliver (1) functional GCaMP for ex vivo calcium imaging across multiple brain areas, and (2) a cocktail of fluorescent reporters for Brainbow-like labeling throughout the macaque brain, circumventing the need for germline manipulations in Old World primates. Given its capabilities for systemic gene transfer in NHPs, CAP-Mac promises to help unlock non-invasive access to the brain.

4 Results 86 Using multiple non-human primate species to identify brain-enriched AAV variants 87 Our overarching goal was to develop an AAV variant efficacious in NHPs after systemic administration. 88 To do that, we used a multi-species screening and characterization strategy to select for variants with enhanced 89 BBB-crossing tropism in NHPs (Fig. 1b). Briefly, we constructed a library as previously described by inserting 90 7mer sequences after Q588 in the structural cap gene of AAV9 14-16 ( Supplementary Fig. 1a-c). We initially 91 screened this library in 2 rounds of selection in the adult marmoset (2 marmosets per round; 2 x 10 12 vector 92 genomes [vg] of viral library per marmoset via IV administration), where we identified 33,314 unique variants 93 present in the brain. 94 In the past, we used our CREATE methodology to increase stringency during selections by only 95 recovering variants that underwent cis-Cre-Lox mediated inversion 14,16 . However, since Cre-transgenic 96 marmosets are not yet available, we pursued other strategies to compensate for the loss of this additional 97 selective pressure. We previously demonstrated the utility of clustering capsid variants based on sequence 98 similarity to generate network graphs as an aid in choosing variants for further characterization 16 . Briefly, we 99 filtered variants based on user-defined performance criteria and clustered high-performing variants into network 00 graphs (Supplementary Fig. 1d-g), wherein each node is a capsid variant, and each edge represents shared 01 sequence identity between related variants (i.e., the pairwise reverse Hamming distance). We reasoned that this 02 clustering analysis would let us efficiently sample variants from our selections while (1) limiting the number of 03 animals used for individual characterization and (2) partially overcoming the absence of the CREATE selective 04 pressure. Based on these network graphs, we chose two variants out of the 33,314 recovered from the marmoset 05 for further characterization: AAV.CAP-Mac (CAP-Mac) and AAV.CAP-C2 (CAP-C2). 06 Following library selection in the adult marmoset, we used capsid-pool studies in newborn rhesus 07 macaques to assess the translatability of several engineered AAVs to Old World primates. We pooled 8 capsid 08 variants: AAV9, CAP-Mac, CAP-C2, and five other previously-engineered AAVs 15,17,51 . Each variant packaged a 09 single-stranded human frataxin transgene fused to a hemagglutinin (HA) epitope tag under control of the 10 ubiquitous CAG promoter (ssCAG-hFXN-HA) with a unique molecular barcode in the 3' UTR. This construct 11 design allowed us to assess protein expression of the virus pool via immunostaining of the HA epitope tag while 12 also quantifying the relative enrichment of each unique barcode in DNA and RNA recovered from tissue. We 13 administered 1 x 10 14 vg/kg of the virus pool to 2 newborn rhesus macaques via the saphenous vein and, at 4 14 weeks post-injection, observed robust expression of the HA epitope throughout the brain (Fig. 2a). In the cortex 15 and hippocampus, we observed single cells with clear projections that resemble the apical dendrites of pyramidal 16 cells. Furthermore, we saw increased HA epitope expression in the thalamus and dorsal striatum (Fig. 2a, insets). 17 When we quantified the relative enrichment of each barcode in the brain, we found that the CAP-Mac-delivered 18 barcode was 9 and 6 times more abundant than the AAV9-delivered barcode in the viral DNA and total RNA, 19 respectively (Fig. 2b). The CAP-C2-delivered barcode was approximately 4-fold enriched relative to the AAV9 20 barcode in both DNA and RNA extracts. Interestingly, the viral DNA levels of all other variants, which were 21 originally selected in mice, were on par with AAV9. In the liver, CAP-Mac and CAP-C2 were negatively enriched, 22 as were some of the previously-engineered controls known to be de-targeted from the liver in rodents 17 (Fig. 2c). 23

Characterization in newborn macaques and infant green monkeys: AAV.CAP-Mac
24 efficiently transduces neurons in the CNS 25 Because CAP-Mac outperformed AAV9 and other engineered variants in our pool study, we moved 26 forward with single characterization in two species of Old World primates. In the newborn rhesus macaque, we 27 intravenously administered a cocktail of CAP-Mac vectors (5 x 10 13 vg/kg total dose via the saphenous vein) 28 packaging 3 different fluorescent reporters under control of the CAG promoter. Fluorescent protein (XFP) 29 expression was observed in multiple coronal slices along the anterior-posterior axis (Fig. 3a) and was robust in 30 all four lobes of cortex and in subcortical areas like the dorsal striatum and hippocampus. While expression was 31 particularly strong in several nuclei of the thalamus (e.g., lateral and medial nuclei, lateral geniculate nucleus, 32 5 pulvinar nucleus), we noted that expression was not found in all brain regions (e.g. the amygdala). Even with a 33 ubiquitous promoter, we observed expression primarily in NeuN+ neurons (mean [XFP+NeuN+]/XFP+ between 34 47-60% across sampled brain regions) and rarely in S100+ astrocytes (mean [XFP+S100+]/XFP+ between 0-35 3%; Fig. 3b). We also attempted to deliver CAP-Mac via LP administration in newborn rhesus macaques, but 36 found that efficiency throughout the brain was noticeably decreased compared to IV administration 37 ( Supplementary Fig. 2). Expression was especially low in subcortical structures, as reported previously [41][42][43][44][45] . 38 AAV variants engineered for BBB-crossing in mice are known to have strain-dependent behavior 16,26,52-39 54 . Therefore, in parallel with the rhesus macaque experiments, we characterized CAP-Mac in green monkeys, 40 another Old World primate species. We administered either AAV9 or CAP-Mac packaging green fluorescent 41 protein under control of CAG (ssCAG-eGFP) to individual 8-month-old monkeys (7.5 x 10 13 vg/kg via the 42 saphenous vein). In the CAP-Mac-dosed green monkeys, we saw broad and strong expression in cortex and 43 various subcortical regions, including the putamen (Fig. 3c), consistent with the capsid-pool (Fig. 2a) and rhesus 44 macaque ( Fig. 3a and b) results. We saw particularly strong eGFP expression throughout the cerebellum in the 45 CAP-Mac-dosed green monkey. Except in the thalamus, CAP-Mac eGFP expression was again found primarily 46 in neurons (mean [GFP+NeuN+]/GFP+ between 33-51%) and not astrocytes (mean [GFP+ S100+]/GFP+ 47 between 3-21%; Fig. 3d). In the thalamus, 42% of GFP+ cells were neurons and 51% astrocytes. In AAV9-dosed 48 monkeys, AAV9 eGFP expression was primarily biased towards astrocytes in cortex (mean 49 [GFP+S100+]/GFP+ between 23-59%) with low neuronal transduction (mean [GFP+NeuN+]/GFP+ between 2-50 10%; Fig. 3e), which is consistent with other reports 41,45,55,56 . Notably, recovered CAP-Mac transgenes were more 51 abundant throughout the brain compared to AAV9, suggesting overall higher brain penetrance of CAP-Mac (Fig.  52 3f and Supplementary Fig. 3a). Interestingly, the cerebellum contained the fewest vector genomes per 53 microgram of DNA in both CAP-Mac monkeys despite strong eGFP expression, most likely due to the high 54 density of cells and processes within the cerebellum 57,58 . In most non-brain tissue, eGFP biodistribution and 55 expression was comparable between CAP-Mac-and AAV9-treated animals ( Supplementary Fig. 3). It should be 56 noted that the cell-type tropism differences between CAP-Mac and AAV9 in the brain may apply to non-brain 57 tissue as well, with each vector transducing distinct cell types. Even in highly homogenous cell populations, there 58 is significant viral infection variability 59-61 , so measuring AAV genomes in bulk may not fully reflect capsid 59 penetrance in tissue across variants and cell types. 60 Experimental utility of CAP-Mac to study the macaque brain 61 The NIH BRAIN initiative emphasizes the priority of developing novel tools for genetic modulation in 62 NHPs to inform further understanding of the human brain 62 . Accordingly, we explored if we could leverage CAP-63 Mac's neuronal tropism in newborn macaques to deliver genetically-encoded reporters to interrogate the brain. 64 First, we tested whether CAP-Mac can be used as a non-invasive method to define neuronal morphology. Having 65 administered a cocktail of 3 CAP-Mac vectors packaging different fluorescent proteins (Fig. 4a), we attempted 66 Brainbow-like labeling 15,49,50 in an Old World primate. We observed widespread expression of all 3 fluorescent 67 proteins in cerebellum, cortex, and the lateral geniculate nucleus of the thalamus (Fig. 4b-d). In the cerebellum 68 and thalamus, we observed a high density of transduced cells, and the highest proportion of co-localization of 2 69 or 3 fluorescent proteins. However, co-localization of multiple fluorescent proteins was rare, suggesting that co-70 infection was uncommon after systemic administration. With broad and robust expression of fluorescent proteins 71 throughout the brain, we were able to assemble morphological reconstructions of medium spiny neurons (Fig.  72 4e) and cortical pyramidal cells (Fig. 4f). 73 In a second set of experiments, we sought to use CAP-Mac to express functional GCaMP throughout the 74 CNS of infant macaques (Fig. 4g). Given the experimental complexity and limited accessibility of NHPs, when 75 designing our GCaMP experiments, we performed initial cargo screening in mice. We therefore first 76 characterized CAP-Mac in three mouse strains. We found that the neuronal bias of CAP-Mac extended to mice 77 when delivered to the adult brain through ICV ( Supplementary Fig. 5a) but not IV administration, where it primarily 78 transduced cells with vasculature morphology ( Supplementary Fig. 5b), with no apparent differences between 79 the three mouse strains. We also found that in P0 C57BL/6J mice, IV-administered CAP-Mac was expressed in 80 6 various cell types in the brain, including neurons, astrocytes, and vasculature ( Supplementary Fig. 5c). Given 81 the strong neuronal tropism of CAP-Mac following ICV administration, we used this method to screen two genetic 82 cargos (either one-component or two-component vectors) in mice prior to applying them to NHPs 83 (Supplementary Fig. 5d-g). Given our results from this cargo selection in mice, we moved forward with a one-84 component system using the CAG promoter. 85 We intravenously delivered ssCAG-GCaMP8s to newborn macaques (3 x 10 13 vg/kg via the saphenous  86 vein) and after 4-6 weeks of expression, we removed tissue for ex vivo 2P imaging. In the hippocampus, 87 thalamus, and cortex we successfully recorded field potential-evoked calcium gradients in GCaMP-expressing 88 cells (Fig. 4h). Cells were responsive to restimulation throughout the experiment and, importantly, the mean peak 89 ΔF/F of GCaMP signal increased with increases in number of field potential pulses ( Supplementary Fig. 4a). 90 Cellular calcium dynamics differed across the four sampled brain regions ( Supplementary Fig. 4b-e). Consistent 91 with our previous profiling, we saw GCaMP expression primarily in cell types with neuronal morphology 92 throughout the brain (Fig. 4i). 93 Human cultured neurons: AAV.CAP-Mac strongly transduces human neurons 94 compared to AAV9 95 Given the efficacy of CAP-Mac in penetrating the brain of infant Old World primates and motivated by our 96 observation that CAP-Mac primarily transduces neurons, we wanted to test whether CAP-Mac offered any 97 improvement over its parent capsid, AAV9, in transducing human neurons. We differentiated cultured human-98 derived induced pluripotent stem cells (iPSCs) into mature neurons (Fig. 5a) and incubated them with CAP-Mac 99 or AAV9 packaging ssCAG-eGFP at doses ranging from 0 vg/cell to 10 6 vg/cell. We found that eGFP expression 00 was noticeably increased in CAP-Mac-administered cultures compared to AAV9-administered cultures (Fig. 5b). 01 AAV9 transduction achieved an efficiency of EC50=10 4.68 vg/cell, while CAP-Mac achieved EC50=10 3.03 vg/cell 02 (Fig. 5c), a 45-fold increase in potency (P=0.0023 using two-tailed Welch's t-test). Average per-cell eGFP 03 expression measured across transduced cells fit a biphasic step function, with CAP-Mac reaching the first 04 plateau at a dose roughly two orders of magnitude lower than AAV9 (Fig. 5d). Overall, the increased potency of 05 CAP-Mac in transducing mature human neurons in vitro is consistent with the neuronal tropism we observed in 06 infant Old World primates, suggesting a similar mechanism of neuronal transduction across species. 07 Adult non-human primate tissue: an improved vector compared to AAV9 08 Infant NHPs offer several logistical advantages for AAV characterization. For instance, they are more 09 likely to be seronegative for neutralizing AAV antibodies, and their smaller body weight requires less vector to 10 be produced for a given dose. While the mammalian BBB is fully formed by birth-including intact tight junctions 11 which give rise to the BBB's unique functionality to limit passive molecular transport into the brain-dynamic 12 molecular and cellular processes occurring during development may make the BBB more permissive 63-66 . We 13 therefore wanted to characterize CAP-Mac's tropism in adult macaque to determine tropism differences across 14 developmental stages. To further de-risk our characterization, we first chose to test CAP-Mac in adult rhesus 15 macaque slices ex vivo (Fig. 6a). In the gray matter of cultured cortical slices, cargo delivered by CAP-Mac, but 16 not AAV9, co-localized with NeuN+ cells, consistent with our previous results (Fig. 6b). Unexpectedly, only 9% 17 as many CAP-Mac viral genomes were recovered as AAV9 genomes, but 3.6-fold more viral transcripts were 18 recovered from CAP-Mac-treated slices than from AAV9-treated slices (Fig. 6c). 19 While informative, ex vivo characterization does not assess BBB penetration, so we next tested CAP-20 Mac in adult macaques in vivo. We injected two adult rhesus macaques with the same AAV pool that we used 21 in infants and found that CAP-Mac-delivered genomes were 13-fold more abundant in the brain than AAV9 (Fig.  22 6d). Again, the variants originally selected in mice were all less efficient than AAV9, but CAP-C2 was 1.2-fold 23 more efficient than AAV9. To further assess protein expression, we injected CAP-Mac packaging CAG-eGFP (1 24 x 10 13 vg/kg total dose via the saphenous vein) into a 17-year-old adult rhesus macaque (Fig. 6e). At the protein 25 level, we observed CAP-Mac-delivered eGFP expression (visualized via eGFP antibody amplification) in parts 26 of the cortex and thalamus, while eGFP expression was absent in other regions of the brain. 27 7 Finally, since CAP-Mac was originally identified using in vivo selections in the adult common marmoset, 28 we also wanted to characterize the vector in the selection species. As in the adult macaque experiment, we 29 injected CAP-Mac and AAV9 into adult marmosets (3.8 and 5.8 years old). To our surprise, we found that the 30 tropism of CAP-Mac in adult marmoset was biased primarily towards the GLUT1+ vasculature (Supplementary 31 Fig. 6), consistent with our results in adult mice. 32 Discussion 33 Here we describe AAV.CAP-Mac, an engineered AAV9 variant with increased efficiency for brain-wide 34 transgene expression in multiple NHP species. 35 By comprehensively characterizing CAP-Mac in multiple rodent strains and NHP species, across ages 36 and administration routes, we found that CAP-Mac tropism varies depending on species, developmental state, 37 and route of administration (Supplementary primates, and vice versa. Interestingly, our pool studies in macaques showed that variants identified via Cre-58 independent selections in marmosets and chosen using network graphs (CAP-Mac and CAP-C2) generally 59 outperformed variants identified via Cre-dependent selections in mice (Fig. 2b). This suggests that while 60 enhancing selective pressure is important when evolving engineered AAVs in vivo, it is also vital to consider the 61 evolutionary relatedness between the selection and target species. Notably, several transgenic marmoset lines 62 are currently available 74,75 , and the generation of Cre-transgenic marmosets is underway 76 , offering the potential 63 to perform M-CREATE in NHPs. Given that the evolutionary distance between mice and marmosets (40-55 mya) 64 is slightly larger than that between marmosets and humans (35-40 mya), the observation that AAV.CAP-B10 65 and AAV.CAP-B22 retain their BBB-crossing tropisms in marmosets offers hope that NHP selections can identify 66 capsid variants efficacious in humans. 67 The overarching goal of this study was to define and disseminate a suite of genetic tools to study the 68 NHP brain, especially in Old World primates. This includes characterizing cargo that can be delivered by CAP-69 Mac, as both self and non-self proteins (e.g. GFP) are known to be immunogenic in certain contexts 77-80 . To that 70 end, we describe two functional cargos for studying the Old World primate brain: (1) a cocktail of three fluorescent 71 reporters for Brainbow-like 49,50 labeling, and (2) GCaMP8s for optical interrogation of ex vivo neuronal activity. 72 Encouragingly, our GCaMP recordings demonstrate that cells expressing CAP-Mac-delivered molecular sensors 73 are physiologically active and healthy in ex vivo rhesus macaque slices. To our knowledge, this is the first 74 description of using a non-invasive, systemic vector to deliver genetically-encoded sensors to the macaque brain, 75 a transformational technique previously limited to rodents. Notably, none of the rhesus macaques dosed in this 76 study experienced adverse events or abnormal liver function and assessment by an independent pathologist 77 confirmed that the vectors were administered safely (Supplementary Fig. 7 and Supplementary Table 6). Moving 78 forward, we expect CAP-Mac-mediated gene transfer to help illuminate circuit connectivity and neuronal function 79 in the macaque brain 81,82 and, more generally, assist major efforts such as the NIH BRAIN Initiative 62 to 80 understand the inner workings of the primate CNS. 81 In addition to CAP-Mac's utility as a tool to study the primate brain, it is also a compelling potential delivery 82 vehicle for genetic medicine in humans. It provides an unprecedented opportunity to deepen our understanding 83 of pharmacodynamics in Old World primate models 30,83,84 and its broad and uniform distribution throughout the 84 CNS opens access to subcortical and midbrain regions for neuroscience researchers, currently difficult in 85 NHPs 41-45 . Additionally, CAP-Mac's enhanced transduction of cultured human neurons supports its potential as 86 a gene-delivery vehicle in humans. Overall, the success of the capsid engineering approach we describe here 87 offers a roadmap for developing the next class of translational gene therapies with improved safety and efficacy 88 profiles. 89 Methods 90 AAV DNA library generation 91 We initially generated diversity at the DNA level, which we then used to produce transfection material to 92 produce the AAV capsid library. For the round 1 library, we introduced this genetic diversity using primers 93 containing degenerate nucleotides inserted between amino acids (AA) 588 and 589 14 sequence near the AgeI restriction enzyme sequence and were paired with a forward primer containing a 20 bp 00 5' overhang near the XbaI restriction enzyme sequence. We then inserted the PCR fragments containing the 01 diversified region into the rAAV-ΔCAP-in-cis-Lox plasmid via Gibson assembly to generate the resulting AAV 02 DNA library, rAAV-CAP-in-cis-Lox, using NEBuilder HiFi DNA Assembly Master Mix (New England Biolabs, 03 E2621). 04 AAV capsid library production 05 We generated AAV capsid libraries according to previously published protocols 16, 85 . Briefly, we 06 transfected HEK293T cells (ATCC, CRL-3216) in 150 mm tissue culture plates using transfection grade, linear 07 polyethylenimine (PEI; Polysciences, Inc). In each plate, we transfected 4 plasmids: (1) the assembled rAAV-08 Cap-in-cis-Lox AAV DNA library, which is flanked by inverted terminal repeats (ITR) required for AAV 09 encapsidation; (2) AAV2/9 REP-AAP-ΔCAP, which encodes the REP and AAP supplemental proteins required 10 for AAV production with the C-terminus of the cap gene excised to prevent recombination with the AAV DNA 11 library and subsequent production of replication-competent AAV; (3) pHelper, which encodes the necessary 12 adenoviral proteins required for AAV production; and (4) pUC18, which contains no mammalian expression 13 vector but is used as filler DNA to achieve the appropriate nitrogen-to-phosphate ratio for optimal PEI 14 transfection. During preparation of the PEI-DNA mixture, we added 10 ng of our AAV DNA library (rAAV-Cap-in-15 cis-Lox) for every 150 mm dish and combined AAV2/9 REP-AAP-ΔCAP, pUC18, and pHelper in a 1:1:2 ratio, 16 respectively (40 µg of total DNA per 150 mm dish). At 60 hours post-transfection, we purified AAV capsid library 17 from both the cell pellet and media using polyethylene glycol precipitation and iodixanol gradient 18 ultracentrifugation. Using quantitative PCR, we then determined the titer of the AAV capsid libraries by amplifying 19 DNaseI-resistant viral genomes relative to a linearized genome standard according to established protocols 85 . 20 Marmoset experiments 21 Capsid library selections 22 All marmoset (Callithrix jacchus) procedures were performed at the National Institutes of Mental Health 23 (NIMH) and approved by the local Institutional Animal Care and Use Committee (IACUC). Marmosets were born 24 and raised in NIMH colonies and housed in family groups under standard conditions of 27°C and 50% humidity. 25 They were fed ad libitum and received enrichment as part of the primate enrichment program for NHPs at the 26 National Institutes of Health. For all marmosets used in this study, there were no detectible neutralizing 27 antibodies at a 1:5 serum dilution prior to IV infusions (conducted by The Penn Vector Core, University of 28 Pennsylvania). They were then housed individually for several days and acclimated to a new room before 29 injections. Four adult males were used for the library screening, 2 each for first-and second-round libraries. The 30 day before infusion, the animals' food was removed. Animals were anesthetized with isoflurane in oxygen, the 31 skin over the femoral vein was shaved and sanitized with an isopropanol scrub, and 2 x 10 12 vg of the AAV 32 capsid library was infused over several minutes. Anesthesia was withdrawn and the animals were monitored 33 until they became active, upon which they were returned to their cages. Activity and behavior were closely 34 monitored over the next 3 days, with daily observations thereafter. 35 At 4 weeks post-injection, marmosets were euthanized (Euthanasia, VetOne) and perfused with 1X 36 phosphate-buffered saline (PBS). After the round 1 library, the brain was cut into 4 coronal blocks, flash frozen 37 in 2-methylbutane (Sigma Aldrich, M32631), chilled with dry ice, and stored at −80°C for long term storage. After 38 the round 2 library, the brain was cut into 6 coronal blocks and, along with sections of the spinal cord and liver, 39 was flash frozen and stored at −80°C for long term storage. 40 Care and Use Committee (IACUC). The day before infusion, the animals' food was removed. 47

Individual characterization of AAV in marmosets
Animals were anesthetized with ketamine (Ketaset, Zoetis 043-304, 20mg/kg), the skin over the 48 saphenous vein was shaved and sanitized with an isopropanol scrub, and 2 x 10 13 vg/kg of AAV was infused 49 over 5 minutes. The animals were monitored until they became active, upon which they were returned to their 50 cages. Activity and behavior were closely monitored over the next 3 days, with daily observations thereafter. 51 Blood samples were taken at days 1, 7, 14, 21 and 31 to measure viral concentration in plasma. 52 At 31 days post-injection, marmosets were anesthetized with ketamine as described earlier and then 53 euthanized (Euthasol, Virbac 200-071, 1mL/kg) and perfused with 1X phosphate-buffered saline (PBS). Brains 54 and organs were cut in half, and one half was flash-frozen in 2-methylbutane (Sigma Aldrich, M32631), chilled 55 with dry ice, and stored at −80°C. The other half was fixed in 4% PFA (Thermo Scientific, J19943-K2) overnight 56 and then stored at 4°C in PBS-Azide (Sigma Aldrich, S2002-100G, 0.025%). Samples were then shipped to 57 California Institute of Technology (Caltech) for analysis. 58 Viral library DNA extraction and NGS sample preparation 59 We previously reported that viral library DNA and endogenous host RNA can be isolated using TRIzol by 60 precipitating nucleic acid from the aqueous phase 14,16 . Therefore, to extract viral library DNA from marmoset 61 tissue, we homogenized 100 mg of spinal cord, liver, and each coronal block of brain in TRIzol (Life 62 Technologies, 15596) using a BeadBug (Benchmark Scientific, D1036) and isolated nucleic acids from the 63 aqueous phase according to the manufacturer's recommended protocol. We treated the reconstituted precipitate 64 with RNase (Invitrogen, AM2288) and digested with SmaI to improve downstream viral DNA recovery via PCR. 65 After digestion, we purified with a Zymo DNA Clean and Concentrator kit (D4033) according to the manufacturer's 66 recommended protocol and stored the purified viral DNA at −20°C. 67 To append Illumina adapters flanking the diversified region, we first PCR-amplified the region containing 68 our 7mer insertion using 50% of the total extracted viral DNA as a template (25 cycles). After Zymo DNA 69 purification, we diluted samples 1:100 and further amplified around the variable region with 10 cycles of PCR, 70 appending binding regions for the next PCR reaction. Finally, we appended Illumina flow cell adapters and unique 71 indices using NEBNext Dual Index Primers (New England Biolabs, E7600) via 10 more cycles of PCR. We then 72 gel-purified the final PCR products using a 2% low-melting point agarose gel (ThermoFisher Scientific,  73 16520050) and recovered the 210 bp band. 74 For the second-round library only, we also isolated the encapsidated AAV library ssDNA for NGS to 75 calculate library enrichment scores, a quantitative metric that we used to normalize for differences in titer of the 76 various variants in our library (see ref. 16 and the next section). To isolate the encapsidated viral genomes, we 77 treated the AAV capsid library with DNaseI and digested capsids using proteinase K. We then purified the ssDNA 78 using phenol-chloroform, amplified viral transgenes by 2 PCR amplification steps to add adapters and indices 79 for Illumina NGS, and purified using gel electrophoresis. This viral library DNA, along with the viral DNA extracted 80 from tissue, was sent for deep sequencing using an Illumina HiSeq 2500 system (Millard and Muriel Jacobs 81 Genetics and Genomics Laboratory, Caltech). 82 round library, the pipeline to process these datasets involved filtering to remove low-quality reads, utilizing a 86 quality score for each sequence, and eliminating bias from PCR-induced mutations or high GC-content. The 87

NGS read alignment, analysis, and generation of network graphs
filtered dataset was then aligned by a perfect string match algorithm and trimmed to improve the alignment 88 quality. We then displayed absolute read counts for each variant during the sequencing run within each tissue, 89 and all 33,314 variants that were found in the brain were chosen for round 2 selections. 90 After such that for a given sample (e.g. the injected virus library or a tissue sample), , is the absolute read count 97 of variant , is the total number of variants recovered, and ̂, is the normalized read count. 98 To construct the CAP-Mac sequence clustering graph, we filtered the round 2 NGS data based on the 99 following criteria: (1) ≥ 100 read count in the injected library sample (24,186/33,314 variants), (2) ≥ 0.7 library 00 enrichment score in more than 2 brain samples (415 variants), and (3) at least 2 more brain samples with ≥ 0.7 01 library enrichment than brain samples with < -0.7 library enrichment (323 variants). To construct the CAP-C2 02 sequence graph, we filtered the round 2 NGS data based on the following criteria: (1) ≥ 100 read count in the 03 injected library sample and (2) both codon replicates present in at least 2 brain samples with ≥ 0.7 library 04 enrichment (95 variants). These variants were then independently processed to determine pair-wise reverse 05 Hamming distances (https://github.com/GradinaruLab/mCREATE) and clustered using Cytoscape (ver. 3.9.0) 06 as described previously 16  Individual AAV production and purification 16 To produce variants for pool testing, we followed our previously published protocol 85 using 150 mm tissue 17 culture dishes. For individual AAV.CAP-Mac and AAV9 characterization in vivo and in vitro, we adopted our 18 published protocol to utilize ten-layer CellSTACKs (Corning, 3320) to efficiently produce viruses at high titer to 19 dose rhesus macaques and green monkeys. Specifically, we passaged 20 150-mm dishes at approximately 70% 20 confluency into a 10-layer CellSTACK 24 h before transfection. On the day of transfection, we prepared the DNA-21 PEI transfection mixture for 40 150-mm dishes and combined the transfection mixture with media and performed 22 a complete media change for the CellSTACK. We collected and changed media at 72 h post-transfection similarly 23 to production in 150 mm dishes. At 120 h post-transfection, we added ethylenediaminetetraacetic acid (EDTA, 24 Invitrogen, 15575020) to a final concentration of 10 mM and incubated at 37°C for 20 min, occasionally swirling 25 and tapping the sides of the CellSTACK to detach the cells. We then removed the media and cell mixture and 26 proceeded with the AAV purification protocol 85 . Of note, during the buffer exchange step after ultracentifugation, 27 we used centrifugal protein concentrators with polyethersulfone membranes (Thermo Scientific, 88533) instead 28 of Amicon filtration devices and used Dulbecco's PBS supplemented with 0.001% Pluronic ® F-68 (Gibco, 29 24040032). 30

31
All rodent procedures were performed at Caltech and were approved by the local IACUC. We purchased 32 C57BL/6J (000664), BALB/cJ (000651), and DBA/2J (000671) mice (all males, 6-8 weeks old) from The Jackson 33 Laboratory. For IV administration in mice, we delivered 5 x 10 11 vg of virus through the retro-orbital sinus 85,86 34 using a 31 G insulin syringe (BD, 328438). For intracerebroventricular administration in mice, we injected 5 x 35 10 10 or 1.5 x 10 11 vg into the lateral ventricle. Briefly, we anesthetized mice using isoflurane (5% for induction, 1-36 3% for maintenance) with 95% O2/5% CO2 (1 L/min) and mice were head-fixed in a stereotaxic frame. After 37 shaving the head and sterilizing the area with chlorohexidine, we administered 0.05 mL of 2.5 mg/mL bupivacaine 38 subcutaneously, and a midline incision was made and the skull was cleaned of blood and connective tissue. 39 After leveling the head, burr holes were drilled above the lateral ventricles bilaterally (0.6 mm posterior to bregma, 40 1.15 mm from the midline). Viral vectors were aspirated into 10 µL NanoFil syringes (World Precision 41 Instruments) using a 33-guage microinjection needle, and the needle was slowly lowered into the lateral ventricle 42 (1.6 mm from the pial surface). The needle was allowed to sit in place for approximately 5 min and 3-5 µL of viral 43 vector was injected using a microsyringe pump (World Precision Instruments, UMP3) and pump controller (World 44 Precision Instruments, Mircro3) at a rate of 300 nL/min. All mice received 1 mg/kg of buprenorphine SR and 5 45 mg/kg of ketoprofen subcutaneously intraoperatively and 30 mg/kg of ibuprofen and 60 mg/kg of Trimethoprim/ 46 Sulfamethoxazole (TMPS) for 5 days post-surgery. After 3 weeks of expression, all mice were perfused with 47 PBS and fixed in 4% paraformaldehyde (PFA). All organs were extracted, incubated in 4% PFA overnight, 48 transferred into PBS supplemented with 0.01% sodium azide, and stored at 4˚C for long-term storage. We sliced 49 the brain into 100 μm sections by vibratome (Leica Biosystems, VT1200S), mounted in Prolong Diamond 50 Antifade (Invitrogen, P36970), and imaged using a confocal microscope (Zeiss, LSM 880). 51 52 All rhesus macaque (Macaca mulatta) procedures were performed at the California National Primate 53

Rhesus macaque experiments
Research Center (CNPRC) at UC Davis and were approved by the local IACUC. Neonate macaques (0.45-1.4 54 kg) were weaned at birth. Within the first month, macaques were infused with AAV vectors either intravenously 12 (IV) or intrathecally (LP). For IV injections, animals were anesthetized with ketamine (0.1 mL) and the skin over 56 the saphenous vein was shaved and sanitized. AAV (between 2 x 10 13 and 1 x 10 14 vg/kg) was slowly infused 57 into the saphenous vein over ~1 min in < 0.75 mL of phosphate buffered saline. For LP injections, animals 58 were administered a sedative intramuscularly and the area of skin at the neck was shaved and aseptically 59 prepared. A needle was advanced into the cisterna magna to remove a small amount of CSF proportional to 60 the amount of fluid injected. Then, a sterile syringe containing the sterile preparation of the AAV (1.5 x 10 12 or 61 2.5 x 10 13 vg/kg) proportional to the amount of fluid collected was aseptically attached and slowly injected. All 62 animals were monitored during recovery from sedation, throughout the day, and then daily for any adverse 63 findings. All monkeys were individually housed within sight and sound of conspecifics. Tissue was collected 4-64 11 weeks after injection. Animals were deeply anesthetized and received sodium pentobarbital in accordance 65 with guidelines for humane euthanasia of animals at the CNPRC. All material injected into rhesus macaques 66 was free of endotoxins (<0.1 EU/mL), and protein purity was confirmed by sodium dodecyl sulphate- mounting. We diluted all antibodies and performed all incubations using PBS supplemented with 0.1% Triton 76 X-100 (Sigma-Aldrich, T8787) and 10% normal donkey serum (Jackson ImmunoResearch, 017-000-121) 77 overnight at room temperature with shaking. 78 To isolate viral DNA and whole RNA, 100mg slices from brain and liver were homogenized in TRIzol 79 (Life Technologies, 15596) using a BeadBug (Benchmark Scientific, D1036) and total DNA and RNA were 80 recovered according to the manufacturer's recommended protocol. Recovered DNA was treated with RNase, 81 restriction digested with SmaI, and purified with a Zymo DNA Clean and Concentrator Kit (D4033). Recovered 82 RNA was treated with DNase, and cDNA was generated from the mRNA using Superscript III (Thermo Fisher  83 Scientific, 18080093) and oligo(dT) primers according to the manufacturer's recommended protocol. We used 84 PCR to amplify the barcode region using 50 ng of viral DNA or cDNA as template. After Zymo DNA purification, 85 we diluted samples 1:100 and further amplified the barcode region using primers to append adapters for 86 Illumina next-generation sequencing. After cleanup, these products were further amplified using NEBNext Dual 87 Index Primers for Illumina sequencing (New England Biolabs, E7600) for ten cycles. We then gel-purified the 88 final PCR products using a 2% low-melting point agarose gel (ThermoFisher Scientific, 16520050). Pool testing 89 enrichment was calculated identically to library enrichment, but is represented in Fig 2b and c 91 Macaques were perfused with PBS and 4% PFA. The brain was sectioned into 4 mm coronal blocks 92 and all tissue was post-fixed in 4% PFA for 3 days before storage in PBS. All tissue was transferred to Caltech 93 for further processing. Brains and liver were sectioned into 100 μm slices using vibratome. Sections of spinal 94 cord were incubated in 30% sucrose overnight and embedded in Optimal Cutting Temperature Compound 95 (Scigen, 4586) and sectioned into 50 μm slices using a cryostat (Leica Biosystems, CM1950). All slices were 96 mounted using Prolong Diamond Antifade and imaged using a confocal microscope. For GFP staining of spinal 97 cord and brain slices from the LP-administered macaque, we incubated slices with chicken anti-GFP (1:500; 98

Individual characterization of CAP-Mac in newborn rhesus macaques
Aves Bio, GFP-1020), performed 3-5 washes with PBS, incubated with donkey anti-chicken IgY (1:200;  99 Jackson ImmunoResearch, 703-605-155), and washed 3-5 times before mounting. We diluted all antibodies 00 and performed all incubations using PBS supplemented with 0.1% Triton X-100 (Sigma-Aldrich, T8787) and 01 10% normal donkey serum (Jackson ImmunoResearch, 017-000-121) overnight at room temperature with 02 shaking. 03 13 For morphological reconstruction, we sectioned brains into 300 μm sections and incubated them in 04 refractive index matching solution (RIMS) 87 for 72 hours before mounting on a slide immersed in RIMS. We 05 imaged using a confocal microscope and 25x objective (LD LCI Plan-Apochromat 25x/0.8 Imm Corr DIC) using 06 100% glycerol as the immersion fluid. We captured tiled Z-stacks (1024x1024 each frame using suggested 07 capture settings) around cells of interest and cropped appropriate fields of view for tracing. Tracing was done 08 in Imaris (Oxford Instruments) using semi-automated and automated methods. 09 For neuron (NeuN) quantification, slices were stained using anti-NeuN antibody (1:200; Abcam, 10 ab177487) overnight in PBS supplemented with 0.1% Triton X-100 and 10% normal donkey serum. Slices 11 were washed 3-5 times with PBS and incubated overnight in anti-rabbit IgG antibody conjugated with Alexa 12 Fluor 647 (1:200; 711-605-152, Jackson ImmunoResearch) in PBS + 0.1% Triton X-100 + 10% normal donkey 13 serum. After 3-5 washes and mounting using Prolong Diamond Antifade, we obtained z-stacks using a 14 confocal microscope and a 25x objective. We segmented NeuN and XFP-positive cells using custom scripts in 15 Python and Cellpose (https://www.cellpose.org/) 88 . 16 Ex vivo two-photon imaging 17 Brain slices of sizes suitable for imaging were prepared with a thickness of 400 µm from larger slices 18 using a vibratome and stored in artificial cerebrospinal fluid bubbled with carbogen gas before two-photon 19 imaging, as previously described 89,90 . For testing GCaMP8s responses, electrical stimulation (4-5 V, 80 Hz, 0.3 20 second duration) with the indicated number of pulses was delivered using an extracellular monopolar electrode 21 placed 100-200 µm away from the neuron imaged. The frame rate of imaging was 30 Hz. Traces of segmented 22 ROIs were plotted as ΔF/F0 = (F(t) -F0)/F0, where F0 is defined as the average of all fluorescence value before 23 the electrical stimulation. The rise time was defined as the time required for the rising phase of the signal to 24 reach from 10% of the peak to 90% of the peak. The decay time constant was obtained by fitting sums of 25 exponentials to the decay phase of the signal. The signal-to-noise ratio (SNR) was obtained by dividing the 26 peak amplitude of the signal by the standard deviation of the fluorescence trace before the electrical 27 stimulation. 28 Characterization in adult rhesus macaque slice 29 One adult rhesus macaque (14 years and 1 month; 10.83 kg) from the Washington National Primate 30 Research Center was planned for routine euthanasia, and the brain was collected as part of the facility's Tissue 31 Distribution Program. A block of the superior temporal gyrus was sectioned into 300 μm slices and slices were 32 recovered 89 and cultured on an air-liquid membrane interface 91 as previously described. Approximately 30 33 minutes after plating slices, we administered 1-2 μL of AAV (5 x 10 13 vg/mL of AAV9 or AAV.CAP-Mac packaging 34 either ssCAG-FXN-HA or ssCAG-eGFP). Experiments were performed in biological triplicates for each condition 35 and culture medium was refreshed every 48 hours until tissue collection at 8 days post-transduction. On the day 36 of tissue collection, the slices were imaged to confirm transduction, slices were cut in half, and each half-slice 37 was flash-frozen in a dry ice-ethanol bath. Samples were stored at -20 ˚C until further processing. 38 Each half-slice was processed (one each for DNA and RNA recovery  52 All green monkey (Chlorocebus sabaeus) procedures were performed at Virscio, Inc. and approved by 53 their IACUC. All monkeys were screened for neutralizing antibodies and confirmed to have < 1:5 titer. At 54 approximately 7-8 months of age (1-1.3 kg), monkeys were dosed intravenously (see Supplementary Table 3  55 for details). Dose formulations were allowed to equilibrate to approximately room temperature for at least 10 56 minutes, but no more than 60 minutes prior to dosing. IV dose volumes were based on Day 0 body weights. 57 Animals were sedated with ketamine (8 mg/kg) and xylazine (1.6 mg/kg). The injection area was shaved and 58 prepped with chlorohexidine and 70% isopropyl alcohol, surgically scrubbed prior to insertion of the 59 intravenous catheter. Dosing occurred with a single intravenous infusion of AAV (7.5 x 10 13 or 7.6 x 10 13 vg/kg) 60 on Day 0 via a saphenous vein administered using a hand-held infusion device at a target rate of 1 mL/minute. 61 General wellbeing was confirmed twice daily by cage-side observation beginning one week prior to dosing. At 62 the scheduled sacrifice time, monkeys were sedated with ketamine (8-10 mg/kg IM) and euthanized with 63 sodium pentobarbital (100 mg/kg IV to effect). Upon loss of corneal reflex, a transcardiac perfusion (left 64 ventricle) was performed with chilled phosphate buffered saline (PBS) using a peristaltic pump set at a rate of 65 approximately 100 mL/min until the escaping fluid ran clear prior to tissue collection. Cubes of tissue were 66 collected from the left brain hemisphere and various other organs and frozen in the vapor phase of liquid 67 nitrogen for further processing for biodistribution. The right brain hemisphere was removed and cut into ~4 mm 68 coronal slices and post-fixed intact with approximately 20 volumes of 10% neutral-buffered formalin (NBF) for 69 approximately 24 hours at room temperature. 70 Genomic DNA was extracted from CNS and peripheral tissues using the ThermoFisher MagMax DNA 71 Ultra 2.0 extraction kit (Catalog number: A36570). DNA was assessed for yield by fluorometric quantification 72 with the Qubit dsDNA assay. Approximately 20 ng of DNA was loaded into each 20 μL reaction and plates 73 were run on the BioRad CFX Connect Real-Time PCR Detection System (Catalog number: 1855201). The viral 74 copy number assay was validated for specificity by detection of a single amplified product, sensitivity by 75 assessing the lower limit of detection to be greater than 10 copies per reaction, and linearity by ensuring the 76 standard curve r 2 was > 0.95. Reactions were assembled in FastStart Universal SYBR Green Master (Rox) 77 (catalogue number: 4913850001). The sequences of the primers were: forward 78 ACGACTTCTTCAAGTCCGCC, reverse TCTTGTAGTTGCCGTCGTCC. The PCR protocol used an initial 79 denaturation step of 95 ˚C for 180 seconds, followed by 40 cycles of 95 ˚C for 15 seconds, and 60 ˚C for 60 80 seconds, with an imaging step following each 60 ˚C cycle. A standard curve was generated with linearized 81 plasmid containing the GFP template sequence present in the virus from 1e8-1e0 copies, diluted in naïve 82 untreated macaque DNA samples prepared using an identical kit as the samples in this study to control for 83 matrix effects. Copies of viral DNA were calculated from the standard curve using the equation for the line of 84 the best fit. MOI values were calculated based on the measured total genomic weight of host cell DNA per 85 reaction. 86 Post fixation, tissues were placed into 10% > 20% > 30% sucrose for 24 hours each at 4 ˚C then 87 embedded in Optimal Cutting Temperature Compound and stored at -80 ˚C until cryosectioning. Tissue blocks 88 were brought up to -20 ˚C in a cryostat before sectioning into 30 μm slices and dry-mounted onto slides after 89 cryosectioning. After sectioning, the slides were left at room temperature overnight to dry. To assist in neuron 90 quantification, we stained sections with the following antibodies and concentrations: rabbit anti-GFP (1:100; 91 Millipore-Sigma, AB3080) and mouse anti-NeuN (1:500; Millipore-Sigma, MAB377). For secondary antibody 92 staining, the following secondary antibodies and concentrations were used: donkey anti-rabbit Alexa Fluor 488 93 (1:500; Invitrogen, A21206) and donkey anti-mouse Alexa Fluor 647 (1:500; Invitrogen, A31571). All antibodies 94 were diluted with 1X PBS supplemented with 0.25% Triton X-100 (PBST) and 5% normal donkey serum. 95 Primary antibody incubations were left overnight at room temperature. Sections were then washed with PBST. 96 Secondary antibody incubations were 2 hours at room temperature. The sections were washed 3x in PBST. 97 Sections were incubated in DAPI solution (1:10,000; Invitrogen, D1306) at room temperature for 5 minutes, 98 then washed. Sections were coverslipped using Prolong Diamond Antifade. 99 15 3 sections per animal were stained and imaged. Each section was imaged in triplicate with each ROI 00 having a total of 9 images. Tissue ROIs were imaged with a Keyence BZ-X800 with the following acquisition 01 parameters: GFP (1/500 s), Cy5 (1 s), DAPI (1/12 s), High Resolution, Z-stack @ 1.2 um pitch. The following 02 brain subregions were imaged frontal, parietal, temporal, occipital cortices, cerebellum, caudate, putamen, and 03 thalamus (medial, ventral lateral, and ventral posterior nuclei). A semi-automated cell counting method was 04 performed via ImageJ for quantification. Using thresholds and particle analysis, we were able to quantify NeuN 05 positive and DAPI positive cells. Using ImageJ's cell counter, we manually counted GFP-positive and GFP & 06 NeuN double-positive cells. 07   CAP-Mac is a novel vector that enables brain-wide, systemic gene transfer in non-human primates. Representative images are shown from a newborn rhesus macaque brain expressing 3 fluorescent reporters delivered intravenously using AAV.CAP-Mac (5 x 10 13 vg total dose, 4 weeks postinjection). b, Schematic of the CAP-Mac selection strategy. (1) CAP-Mac is an AAV9 variant that we selected from a library screened in the adult common marmoset. We generated diversity by introducing 21 NNK degenerate codons after Q588 in the AAV9 cap genome and produced the capsid library for in vivo selections in adult male marmosets. (2) In two rounds of selections, we intravenously administered 2 x 10 12 vector genomes per marmoset, narrowing our variant pool with each round of selection. After the first round of selection, we recovered 33,314 unique amino acid sequences in the brain. For the second round of selection, we generated a synthetic oligo pool containing each unique variant plus a codon modified replicate (66,628 total sequences). After the second round of selection, we constructed network graphs of highperforming variants, and selected two capsids-AAV.CAP-Mac and AAV.CAP-C2-to be included in pool selections in newborn rhesus macaques. (3) For pool selections, we produced 8 capsids packaging ssCAG-hFXN-HA, each with a unique molecular barcode in the 3' UTR. This construct design enabled us to assess protein expression of the pool by staining for the hemagglutinin (HA) epitope and quantify barcodes in viral DNA and whole RNA extracts. (4) We moved forward with individual characterization of AAV.CAP-Mac in various contexts (ex vivo, in vitro, in vivo) in multiple primate species.  a, Representative images of expression in cortex, thalamus, caudate nucleus, putamen, hippocampus and claustrum after intravenous administration of 1 x 10 14 vg/kg of an 8-capsid pool (1.25 x 10 13 vg/kg of each variant) packaging hemagglutinin (HA) tagged human frataxin with a unique barcode in each capsid. b, c, Unique barcode enrichments in viral DNA (left) and whole RNA (right) extracts from the brain (b) and the liver (c) of two newborn rhesus macaques. Each data point represents the fold-change relative to AAV9 within each sample of tissue. Mean ± s.e.m. shown. The red dashed line denotes AAV9 performance in pool. One-way ANOVA using Tamhane's T2 correction tested against AAV9 enrichment. (*P<0.05, **P<0.01, ***P<0.001, ****P<0.0001). Fig. 3: CAP-Mac is biased towards neurons throughout infant green monkey and newborn rhesus macaque brains. Fig. 3: CAP-Mac is biased towards neurons throughout infant green monkey and newborn rhesus macaque brains. a, Distribution of CAP-Mac expression in 2-day-old rhesus macaques (5 x 10 13 vg/kg via intravenous administration) across coronal slices showing fluorescent reporter expression (ssCAG-mNeonGreen, ssCAG-mRuby2, ssCAG-mTurquoise2) in cortical and subcortical brain regions (insets). Imaging channels of reporters are identically pseudo-colored. b, Colocalization of fluorescent reporters with NeuN (neurons) or S100b (astrocytes) in 2-day-old rhesus macaques treated with CAP-Mac. Values are reported as a percentage of all XFP+ cells. c, Representative images from 8-month-old green monkeys dosed with CAP-Mac (top) or AAV9 (bottom) packaging ssCAG-eGFP (7.5 x 10 13 vg/kg via intravenous administration). d, e, Colocalization of fluorescent reporters with NeuN (neurons) or S100b (astrocytes) in infant green monkeys treated with CAP-Mac (d) or AAV9 (e). Values are reported as a percentage of all GFP+ cells. f, Distribution of CAP-Mac and AAV9-delivered eGFP transgene in 11 brain regions of green monkeys. Each data point represents measured vector genomes per microgram of total DNA in a section of tissue from each region and monkey. Mean ± s.e.m. shown. Fig. 4: Experimental utility of CAP-Mac for functional interrogation of the newborn rhesus macaque brain. Fig. 4: Experimental utility of CAP-Mac for interrogation of the newborn rhesus macaque brain. a-f, CAP-Mac packaging three fluorescent reporters (a) to generate Brainbow-like labeling in rhesus macaque cerebellum (b), cortex (c), and thalamus (lateral geniculate nucleus) (d), enabling morphological reconstruction of neurons (e and f). g-i, Noninvasively delivering ssCAG-GCaMP8 using CAP-Mac (g) for ex vivo two-photon imaging (h) and brain-wide GCaMP expression (i).     6: Characterization in adult rhesus macaque. a, AAV in cortical slice ex vivo taken from a 14-year-old rhesus macaque. b, CAP-Mac is more efficient at transducing neurons in gray matter of cortex. c, Quantification demonstrates that CAP-Mac-delivered transgene is better at producing RNA but not DNA compared to AAV9-delivered transgene. Twotailed Welch's t-test (*P<0.05). d-f, AAV in adult rhesus macaques in vivo. d, Recovered DNA from adult macaque administered with 8-capsid pool. One-way ANOVA using Tamhane's T2 correction tested against AAV9 enrichment. (****P<0.0001). e, We injected 1 x 10 13 vg of CAP-Mac packaging a CAG-eGFP into one 17-year-old rhesus macaque to assess CAP-Mac protein expression. g, CAP-Mac-mediated eGFP expression visualized after amplification with GFP antibody. Mean ± s.e.m. shown.