Engineered AAVs for non-invasive gene delivery to rodent and non-human primate nervous systems

Quantiﬁcation of


INTRODUCTION
Gene delivery to the central and peripheral nervous systems (CNS and PNS) has greatly accelerated neuroscience research in the last decade and has the potential to translate this research into novel therapies for neurological disorders. However, the lack of potent vectors enabling non-invasive gene delivery across species is a significant bottleneck that can hinder fast progress (Ingusci et al., 2019;Pena et al., 2020;Piguet et al., 2021).
The success of gene delivery relies on a safe and efficient vector, and therefore, most in vivo applications use adeno-associated viral (AAV) vectors. AAVs offer several advantages, including stable, long-term transgene expression and low immunogenicity (Bedbrook et al., 2018;O'Carroll et al., 2021). The natural serotypes of AAV have demonstrated considerable success in targeting different cell populations within the nervous system through direct routes of gene delivery, such as intracranial (Golebiowski et al., 2017;GuhaSarkar et al., 2017), intracerebroventricular (i.c.v.) (Bey et al., 2020), intrathecal (Hirai et al., 2012), intraganglionic (Yu et al., 2013), intrasciatic (Homs et al., 2011), and intracolonic (Gore et al., 2019). These direct delivery routes suffer from limitations, however, including the need for invasive surgery. In addition, anatomical barriers may restrict surgical access (such as for nodose ganglia [NG] or dorsal root ganglia [DRG]). Finally, localized delivery can lead to incomplete coverage of a large complex system such as the enteric nervous system (ENS) or CNS, and multiple direct interventions may be needed to increase coverage (Gray et al., 2010).
An alternative, non-invasive intravenous (i.v.) route circumvents these limitations. Some natural serotypes, including AAV9, can target the CNS or PNS systemically. However, lack of specificity toward the target and low efficiency, necessitating high vector load, both potentially lead to toxicity (Bourdenx et al., 2014;Yang et al., 2014;Vogt et al., 2015;Gombash et al., 2017). Years of capsid engineering efforts have now yielded a toolbox of improved CNS capsids for rodents (Deverman et al., 2016;Kö rbelin et al., 2016;Ojala et al., 2018;Ravindra Kumar et al., 2020;Nonnenmacher et al., 2021). These include the potent vector AAV-PHP.eB (engineered using CREATE) for the CNS, but its application is restricted to select mice strains (Deverman et al., 2016;Chan et al., 2017;Hordeaux et al., 2018;Matsuzaki et al., 2018;Hinderer et al., 2018). Unlike PHP.B/eB, the recently engineered AAV-F (engineered using iTransduce) and AAV-PHP.Cs (engineered using Multiplexed-CREATE or M-CREATE) for the CNS work across mouse strains (Hanlon et al., 2019;Ravindra Kumar et al., 2020). However, the heterogeneity of the blood-brain barrier (BBB) depending on the genetic background has posed a significant challenge for developing capsids that have the potential to translate across species (Hordeaux et al., 2019;Huang et al., 2019;Batista et al., 2020). This issue is particularly acute for non-human primates (NHPs), which are commonly used as pre-clinical research models for gene therapy (Kimura and Harashima, 2020;Piguet et al., 2021). With several CNS-and PNS-based therapies in the pipeline, there is, therefore, a high demand for next-generation systemic AAV vectors with potent neurotropic behavior to achieve efficient and safe gene delivery for translational applications (Deverman et al., 2018;Tosolini and Smith, 2018;Hudry and Vandenberghe, 2019;Chen et al., 2021;Huang et al., 2021;Challis et al., 2022). Building on our success in selecting improved CNS-targeting capsids using M-CREATE (Ravindra Kumar et al., 2020;Goertsen et al., 2022), we decided to test the potential of this method for selecting improved PNS-targeting capsids outperforming the prior engineered variant, AAV-PHP.S, which requires a high dose to exhibit its potent PNS tropism via i.v. delivery (Chan et al., 2017). Compared with the CNS, the PNS is a more challenging AAV engineering target. Cell populations are sparser, and there is no strong source of selection pressure across targets (akin to the BBB for the CNS). M-CREATE is uniquely well suited to this problem as it capitalizes on deep recovery of capsid libraries across cell types/organs to select capsids enriched in areas of interest and a customized analysis pipeline that incorporates positive and negative selections to help identify variants with desired properties.
In this study, we used M-CREATE to identify a family of 7-mer containing AAV9 capsids that appeared to be biased toward PNS areas and detailed the properties of two such selected AAVs, AAV-MaCPNS1 and AAV-MaCPNS2, in rodents and NHPs. We provide in vivo validation of their tropism in mice and demonstrate their potential applications with proof-of-concept studies for functional readout and modulation of sensory ganglia. In addition to finding improved PNS-targeting capsids, we sought to address the fundamental question of the translatability of capsids selected in mouse models. To this end, we examined these capsids across the following four species commonly used in basic through pre-clinical applications: mice, rats, marmosets, and rhesus macaques. The variants discussed in this study show improved efficiency and specificity toward the PNS and translate their potent behavior across mammalian species. Interestingly, and potentially due to the heterogeneity of the BBB, these variants also show the efficient crossing of the BBB to infect the CNS in NHPs (Table S1).

RESULTS
AAV capsid selection in mice identifies two AAV variants with PNS specificity As a starting point for capsid engineering, we chose AAV9 due to its broad tropism when delivered systemically, including for the nervous system (both CNS and PNS) (Foust et al., 2009;Bevan et al., 2011). We diversified the AAV9 capsid by inserting a randomized 7-mer peptide between positions 588 and 589 ( Figure 1A) and i.v. injected the resulting virus library into adult mice of the following three Cre-transgenic lines: SNAP-Cre (for neurons), GFAP-Cre (for glia), and Tek-Cre (for endothelial cells forming the blood-organ barrier). Two weeks post injection, we processed peripheral tissues (such as the heart, small and large intestines, and DRG) and selectively extracted viral genomes from Cre+ cells using Cre-dependent PCR (see STAR Methods; Figure 1A2).
After round-1 (R1) selection, we recovered a total of $9,000 variants from the Cre lines in the following PNS tissues of interest: the heart, DRG, and small and large intestine. Of these, $10% overlapped with the CNS libraries ( Figure 1A3). We then synthesized these variants in an equimolar ratio to create a synthetic oligopool library for round-2 (R2) PNS selection. This synthetic pool also included a spike-in library of previously validated internal controls (see STAR Methods). We i.v. injected the R2 virus library into different Cre lines as follows: Nestin-Cre (for neurons), CHAT-Cre (for cholinergic neurons), TRPV1-Cre (for primary afferent thermosensitive neurons), and Tek-Cre (for Figure 1. Multiplexed-CREATE selection for AAV capsids targeting the nervous system across species (A) An overview of the capsid selection method, Multiplexed-CREATE, and characterization of selected capsids across species. The top left panel illustrates the evolution of the AAV9 capsid (PDB 3UX1) with a zoom-in (blue) of a 3-fold axis and the 7-mer-i library insertion site between residues 588-589 highlighted in red. The diagrams below demonstrate the arrangement of the acceptor vector in the absence (top) or presence (bottom) of Cre, with the corresponding orientations of the forward/reverse primers used for Cre+ selective recovery. The selection workflow involves four key steps: (1) generation of capsid library and intravenous (i.v.) delivery into transgenic mouse lines where Cre is restricted to cell types of interest (SNAP-Cre, GFAP-Cre, Tek-Cre, n = 2 mice per Cre line). (2) Two weeks post injection, viral DNA is recovered across cell types/tissues using Cre-dependent PCR and Illumina next-generation sequencing (NGS) and (3) fed into synthetic pool library production. (4) This library then goes through a second round of in vivo selection. Following this selection process, identified variants are validated for (5) virus production and (6) in vivo transduction across species. (B) Heatmaps of capsid variants' mean enrichment by Cre-dependent recovery across tissues of interest (red text) and Cre-independent recovery across offtargets (black text) after two rounds of selection. Cre lines are plotted separately (top panel, n = 3 mice per organ) or grouped by organs (bottom panel). The y axis represents capsids unique at the amino acid (aa) level, ranked by ''neuron mean,'' which is the mean of the enrichment of all targets of interest. (C) UMAP cluster representation of $9,000 variants that were recovered after two rounds of selection (UMAP parameters: n_neighbors = 15, min_dist = 0.1, n_components = 2, random_state = 42, metric = ''correlation,'' and verbose = 3). Three separable clusters are shown along with the positions of known capsids (AAV9, PHP.S, PHP.B, PHP.B4, and PHP.C1) and new capsids (MaCPNS1 and MaCPNS2). Heatmaps (below) show enrichment of representative capsids from clusters-1 and -2 across organs. NeuroResource endothelial cells, providing negative selection). Two weeks post injection, we selectively recovered capsids by Cre-dependent PCR from tissues of interest (the DRG, heart, small and large intestine, brain, and spinal cord [SC]) and other targets (the spleen, liver, lung, kidney, testis, and muscle) by performing a Cre-independent PCR (see STAR Methods; Figures 1A4 and 1B). After two rounds of in vivo selection, among all the variants that we included in the R2 library, 6,300 variants showed a bias toward one or more of the PNS tissues ( Figures 1B and S1A). We next sought to further classify the recovered variants based on distinct tropisms. We used the uniform manifold approximation and projection (UMAP) algorithm (McInnes et al., 2020), which takes into account differences among variants' enrichments across organs, to identify clusters of variants representing distinct tropisms ( Figure 1C, top panel). Members of cluster-1 (1,846 variants), which includes AAV9, showed relatively higher enrichment in off-target tissues such as the liver ( Figure 1C, bottom left panel). Cluster-2 members (7,148 variants), including our previously engineered variant PHP.S, exhibited relatively higher enrichment in the PNS and lower enrichment in off-target tissues ( Figure 1C, bottom right panel). Cluster-3 members (5 variants), including variants from the internal control, were highly enriched in the CNS.
Based on this analysis, we reasoned that cluster-2 might contain promising variants distinct from the parental AAV9, which we indeed observed by comparing their tropism directly to that of AAV9 ( Figure 1D). We identified two new capsids from cluster-2 that exhibited low off-target transduction compared with AAV9 and PHP.S ( Figure 1E) and that could package viral genomes with similar efficiency to AAV9 (see STAR Methods; Figure S1B). We will henceforth refer to the first capsid, with a 7-mer peptide insertion of PHEGSSR between the 588-89 residues of the AAV9 parent, as AAV-MaCPNS1 and the second, with an insertion of PNASVNS, as AAV-MaCPNS2 ( Figure 1F).
IV-delivered AAV-MaCPNS1/2 efficiently transduces sensory and enteric ganglia in mice with low liver transduction To characterize the transduction capability of AAV-MaCPNS1 and AAV-MaCPNS2 variants in vivo, we packaged these variants with a single-stranded (ss) AAV genome carrying a strong ubiquitous promoter, CAG, driving expression of nuclear-localized eGFP reporter and i.v. injected them into adult mice at 3 3 10 11 vg per animal (Figures 2A-2H and 2J-2L; see STAR Methods). By quantifying expression in NG that overlapped with the NeuN neuronal marker, we found that MaCPNS1 and MaCPNS2 had mean transduction efficiencies of $28% and $35%, respectively. By comparison, the efficiency of AAV9 was $16% and AAV-PHP.S $12% ( Figures 2B and 2C). Thus, MaCPNS1/2 capsids exhibit about 2-fold higher transduction of NG than previously available vectors.
Next, we investigated the efficiency of DRG transduction from selected spinal levels (thoracic and lumbar). The MaCPNS1 vector demonstrated a mean transduction efficiency of $18% and MaCPNS2 $16%, compared with $11% for AAV9 and $7% (D) Percentage of eGFP expression overlapping with the aNeuN marker in the DRG (left) where each data point shows the mean per mouse across select DRGs within thoracic and lumbar segments of the spinal cord. A one-way ANOVA, non-parametric Kruskal-Wallis test (exact p = 0.0005), and follow-up multiple comparisons with uncorrected Dunn's test are reported (p = 0.0018 for PHP.S versus MaCPNS1 and p = 0.0087 for PHP.S versus MaCPNS2). *p % 0.05 and **p % 0.01 are shown, p > 0.05 is not shown; n R 4 per group, same experimental parameters as (B). Each data point shows the mean ± SEM of DRGs across different areas of each mouse, comprising a mean of 1-2 DRGs per area with >200 aNeuN+ cells per DRG. (Right) Percentage of eGFP expression overlapping with the aNeuN marker in DRG across spinal cord areas in individual mice. A two-way ANOVA and Tukey's multiple comparisons tests with adjusted p values are reported (*p % 0.05, **p % 0.01, and ***p % 0.001 are shown; and p > 0.05 is not shown). Each data point shows the mean ± SEM of 1-2 DRG per mouse comprising >200 aNeuN+ cells per DRG. (E) Illustration of the adult mouse gastrointestinal (GI) tract, highlighting the enteric ganglia (zoom-in) that are spread across the different segments of the GI tract. (F) Percentage of cells expressing NLS-eGFP delivered by AAV vectors in the myenteric plexus across the GI tract: stomach, duodenum, jejunum, ileum, proximal colon, and cecum (ssAAV:CAG-2xNLS-eGFP genome, n R 5 per group, $8 weeks old C57BL/6J males, 3 3 10 11 vg i.v. dose per mouse, and 3 weeks of expression). A one-way ANOVA non-parametric Kruskal-Wallis test (approximate p = 0.2985), and follow-up multiple comparisons using uncorrected Dunn's test are reported (individual p > 0.05, n.s.). Each data point shows the mean ± SEM of >100 enteric ganglia per intestinal segment per mouse). (G) Percentage of cells expressing NLS-eGFP (green in H) in the myenteric plexus of small intestinal segments: duodenum, jejunum, and ileum delivered by AAV vectors: AAV9, PHP.S, and MaCPNS2 (ssAAV:CAG-2xNLS-eGFP genome, n = 3 per group, 1 3 10 11 vg i.v. dose per mouse, and 3 weeks of expression). Twoway ANOVA, Tukey's multiple comparisons tests with adjusted p values are reported (*p % 0.05, **p % 0.01, and ***p % 0.001 are shown; p > 0.05 is not shown). Each data point shows the mean ± SEM of R 2 images per mouse comprising >100 enteric ganglia. (H) Shows representative images of AAV-mediated eGFP expression across the individual intestinal segments analyzed in (G). Scale bar, 50 mm. The tissues were co-stained with aS100b (magenta) antibodies for glia and aPGP9.5 (blue) for neurons. The images are matched in fluorescence intensity to the respective AAV9 control. (I) MaCPNS2 vector-mediated expression of tdTomato (red) from ssAAV:hSyn-tdTomato in the proximal and distal segments of the colon at three different i.v. doses per mouse: 7 3 10 11 vg, 5 3 10 11 vg, and 3 3 10 11 vg (3 weeks of expression; n = 3 per group; scale bar, 200 mm; and the dotted white box inset represents a zoomed-in view of the indicated area in each image). Images in the distal and proximal colon are matched in fluorescence intensity. for AAV-PHP.S ( Figure 2B, bottom panel; Figure 2D, left), with some variability in transduction across different segments of the SC ( Figure 2D, right). Thus, compared with known capsids, MaCPNS1/2 can achieve about 2-fold higher transduction of DRG via i.v. delivery. Increasing the vector dose to 1310 12 vg, we observed that PHP.S had higher transduction in DRG compared with AAV9 whereas MaCPNS1 exhibited even higher transduction ( Figures S2A-S2C), further confirming that MaCPNS1/2 capsids transduce DRG in adult mice more efficiently than either AAV9 or PHP.S.
To investigate the transduction efficiency of the new vectors in the ENS, we assessed AAV-mediated eGFP expression in the enteric ganglia of the myenteric plexus across different areas of the GI tract-the stomach, duodenum, jejunum, ileum, proximal colon, and cecum ( Figures 2E and 2F). We observed a mean transduction efficiency of $8% by MaCPNS1, $17% by MaCPNS2, $13% by PHP.S, and $12% by AAV9 ( Figure 2F). By analyzing individual GI segments, we noticed variability in the transduction of enteric ganglia, with a bias of MaCPNS2 transduction toward small intestinal (SI) segments (duodenum, jejunum, and ileum) over other areas, indicating a modest improvement in SI ENS transduction by MaCPNS2 compared with other vectors. To more closely examine this SI bias, we slightly lowered the vector dose to 1310 11 vg and observed a 2-fold increase in the transduction of the SI by MaCPNS2 compared with AAV9 ( Figures 2G and 2H). Next, we investigated if the transduction efficiency across large intestinal (LI) segments could be improved by a higher dose (Figures 2I and S2D). Compared with the initial dose of 3310 11 vg, we observed that a MaCPNS2 dose of either 5310 11 vg or 7310 11 vg achieved robust and uniform transduction of the enteric ganglia of the myenteric plexus in the proximal colon as well as sparser transduction in the distal colon. Thus, i.v. administered MaCPNS2 offers improved transduction of enteric ganglia in the SI, with the option of additional targeting of LI segments at higher doses.
To investigate the liver transduction of MaCPNS1/2 capsids, we assessed eGFP expression in vivo ( Figures 2J-2L). MaCPNS1 exhibited 1.5-fold lower transduction compared with the same dose of AAV9 or PHP.S ( Figure 2K). Quantifying the mean brightness of the eGFP fluorescence signal revealed that the new vectors also exhibited a 1.3-to 2.1-fold lower signal compared with AAV9 or PHP.S ( Figure 2L), suggesting reduced transgene expression per cell. Collectively, these data suggest that MaCPNS1 exhibits lower transduction of hepatocytes and lower transgene expression per liver cell compared with other vectors used for PNS. In addition to the liver, we also investigated other tissues, including the CNS, given AAV9's ability to cross the BBB. At a modest i.v. dose of 3310 11 vg, CNS transduction was low for all vectors we tested, including AAV9 ( Figure S2E). Even when we increased the dose of MaCPNS2 to 5-7x10 11 vg, we did not observe CNS transduction in mice ( Figure S2F). In addition, MaCPNS1/2 transduction of cardiac muscle was similar to that of AAV9 ( Figure S2E).
In summary, we found that systemic delivery of MaCPNS1 and MaCPNS2 vectors in mice can efficiently transduce sensory ganglia (such as NG and DRG) compared with AAV9, with MaCPNS2 distinguished from MaCPNS1 by enhanced transduction of the ENS. Furthermore, MaCPNS1 and MaCPNS2 exhibited improved specificity for the PNS, with relatively lower transduction of AAV9's primary target, the liver.
Systemic MaCPNS1-mediated sensor and actuator expressions enable functional characterization of neurons in nodose ganglia and dorsal root ganglia Functional readout of PNS activity during physiological conditions is key to understanding the interaction between the brain and peripheral system (Chen et al., 2017;Zanos et al., 2018;Walters et al., 2019;Jiman et al., 2020). However, commonly used imaging with genetically encoded calcium indicators (GECIs) in the CNS has been challenging in the PNS due to the low efficiency of delivering GECI to PNS targets. To test the applicability of MaCPNS1/2 capsids to this problem, we delivered i.v. MaCPNS1 capsid packaged with a recently developed GECI, jGCaMP8s (Zhang et al., 2020), to adult mice. After three weeks of expression, we recorded calcium signals in vivo during procedures on the gut ( Figures 3A-3C). We infused the gut with either glucose solution or saline (to induce distension) while recording neuronal activity in the NG of the anesthetized mice and observed distinguishable neuronal responses in the NG (Figures 3A-3C). No significant expression was observed in the CNS or liver of these mice.
After verifying the new vectors' potential for enabling functional readout of the PNS, we took the further step of seeing whether we could achieve functional modulation with the vectors. We sought to construct a mouse pain-induction system with improved temporal control, a critical tool for understanding and potentially managing pain. To activate DRG TRPV1 neurons, we packaged Cre-dependent excitatory designer receptors exclusively activated by designer drugs (DREADD) (Krashes et al., 2011) into MaCPNS1 and i.p. injected the vector into P1 TRPV1-Cre pups ( Figure 3D). After 6 weeks of expression (Figure 3E), we gave mice an intraplantar injection of the DREADD agonist clozapine N-oxide (CNO) and evaluated the resulting pain-like behaviors (see STAR Methods). We observed an increase in the number of bouts and overall time spent lifting or licking the injected footpad in the experimental AAV-administered group compared with the control group, indicating nocifensive pain-like behavior ( Figures 3F and 3G).
Together, these results provide a proof of concept for the application of these new vectors to a wide range of studies involving the monitoring and modulation of sensory processes, including pain.
IV-administered AAV-MaCPNS1/2 efficiently transduces the PNS in adult rats Having validated the new variants' transduction profiles in mouse models, we next investigated their efficacy in another common research model system-rats. Systemic delivery of MaCPNS1/2 capsids packaged with ssAAV:hSyn-tdTomato in Sprague Dawley adults ( Figure 1A  (sympathetic chain ganglia [SCG] and inferior mesenteric ganglia [IMG]), mixed sympathetic-parasympathetic ganglia (the major pelvic ganglion [MPG]), and enteric ganglia across the SI and LI (Figures 4A-4F, S3A, and S3B). The majority of cells transduced in both DRG and TG were NF200+ (about 70%-75%), with only 5% either CGRP+ or TRPV1+ (Figures 4C and 4D). In terms of efficiency, both capsids transduced $22% of the NF200+ cells in DRG and 15%-17% of the NF200+ cells in TG ( Figure 4E). Interestingly, this cell type bias contrasts with what we observed in neonatal mice, where the vectors showed a bias toward CGRP+ neurons (Figures S2I and S2J). This profile could result from differences in species or age/time of injection as well as the different systemic routes of vector delivery (similar tropism shifts have been reported elsewhere [Foust et al., 2009]). The tropism diversity highlights the necessity of considering all relevant experimental variables to achieve the desired transduction profile.
The vectors' efficient labeling of sensory ganglia prompted us to look further into the ENS (Figures 4F and S3C). Analyzing segments of the SI and LI, we found that MaCPNS1 and MaCPNS2 were comparable at transducing the myenteric plexus and the submucosal plexus, including the vascular and periglandular plexuses. Next, we investigated projections to the SC and brainstem ( Figures 4G and S3D). Sensory nerve fibers entering the dorsal horn of the SC were densely labeled, and the ascending afferent tracts in the dorsal column were also strongly transduced by both virus vectors. In the brainstem, fibers were densely labeled in the spinal trigeminal nucleus oralis (Sp5O), which potentially projected from the TG. In addition to the transduction in the PNS and the nerve projections labeling in the CNS, we also observed some labeling of neuronal cell bodies across regions in the brain ( Figure S3D). Furthermore, we observed no expression in the liver ( Figure S3E).
Together, these results show that the potent PNS tropism of the MaCPNS1/2 vectors is conserved across rodent models tested.
IV-delivered AAV-MaCPNS1/2 transduces the adult marmoset CNS and PNS more efficiently than AAV9 Novel capsids selected in mice do not always translate to NHPs (Hordeaux et al., 2018;Matsuzaki et al., 2018). After validating the MaCPNS1/2 capsids in mice and rats, we therefore decided to assess their performance in NHPs. First, we chose the marmoset, a New World monkey, and an emerging animal model for translational research. Owing to the limited availability of these animals, in each adult animal we tested two viral capsids (AAV9, MaCPNS1, or MaCPNS2) packaging different fluorescent reporters (either ssAAV:CAG-eGFP or ssAAV:CAG-tdTomato) ( Figure 5A). Compared with the parent AAV9, i.v. delivered MaCPNS1/2 capsids carrying either ssAAV:CAG-eGFP or ssAAV:CAG-tdTomato genome showed robust transduction of the PNS and the CNS. In the PNS, we observed enhanced transduction of DRG, the SI, and the ascending fiber tracts in the dorsal column of the SC (Figures 5B-5D), similar to what we observed in rodents. In the CNS, robust brain-wide transduction by MaCPNS1/2 capsids, but not AAV9, was evident in regions including the cortex, thalamus, globus pallidus, cerebellum, and brainstem ( Figures 5E and 5F). Further antibody staining and quantification revealed that capsids mainly transduce neurons and astrocytes in the marmoset brain. In the marmoset cortex, MaCPNS2 displayed a $5.5-fold increase in neuronal transduction over AAV9 while MaCPNS1 displayed a $4-fold increase over AAV9 (Figures S4A-S4C). Among the neurons transduced, few overlapped with the PV marker, indicating the vectors' potential bias to the excitatory population ( Figure S4D). For astrocyte transduction, MaCPNS2 displayed a $25-fold increase over AAV9 in the cortex ( Figure S4B). Similar improvement in both neuronal and astrocytic transduction was also observed in the thalamus ( Figure S4C).
These results demonstrate that the MaCPNS1/2 capsids can efficiently cross the BBB in adult marmosets while still exhibiting the enhanced PNS tropism observed in rodents.

IV-delivered AAV-MaCPNS1/2 transduces the infant rhesus macaque CNS and PNS more efficiently than AAV9
Encouraged by the tropism of MaCPNS1/2 capsids in marmosets, we further assessed their transduction potential in another NHP, the rhesus macaque, an Old World monkey, and a common pre-clinical research model for AAV gene therapy. We employed the same strategy as in marmosets, i.v. administering AAV9, MaCPNS1, and MaCPNS2 capsids packaging either ssAAV:CAG-eGFP or ssAAV:CAG-tdTomato to an infant rhesus macaque ( Figure 6A). Similar to our results in marmosets, we observed that MaCPNS1/2 efficiently targeted both the CNS and PNS compared with the parent AAV9 after three weeks of expression. We saw enhanced transduction in the SC, DRG, and GI tract, including the esophagus, colon, and SI ( Figures 6B-6D), as observed in rodents and marmosets. In the SC, there was strong labeling of both ascending fiber tracts in the dorsal column as well as the peripheral fiber tracks coming into the dorsal horn. Within the gray matter of the lumbar SC, MaCPNS1 displayed a $25-fold increase in neuronal transduction over AAV9. For non-neuronal cells, MaCPNS2 displayed a $130-fold increase in transduction over AAV9 ( Figures S5A  and S5B). Similar improvement in transduction was also observed in the thoracic SC ( Figures S5C and S5D). In the lumbar and thoracic DRG, MaCPNS2 displayed $77-fold and $44fold increases in neuronal transduction over AAV9, respectively. MaCPNS1 also showed $37-fold and $22-fold increases in neuronal targeting in lumbar and thoracic DRG. Within lumbar DRG ( Figure S5E), MaCPNS2 was highly specific to neurons, with $94% of all transduced cells being neurons. In thoracic DRG, both vectors displayed high specificity in targeting neurons ( Figure S5F). In the CNS, MaCPNS1/2 capsids mediated enhanced brain-wide transduction, including in areas such as the cortex, hippocampus, putamen, and brainstem ( Figures 6E,  6F, and S6; Videos S1-S4), similar to our observations in marmosets. Both vectors transduced neurons and astrocytes but not oligodendrocytes or endothelial cells (Figures S7A and S7B). In the macaque cortex, neuron transduction by MaCPNS1 and MaCPNS2 was $6-fold and $11-fold higher, respectively, than AAV9 ( Figure S7B). Most neurons transduced by both vectors were PV-, indicating a potential bias to excitatory populations ( Figure S7D). MaCPNS2 also displayed a $44-fold increase in astrocyte transduction over AAV9 in the cortex ( Figure S7B). (C) Representative images of MaCPNS1 vector-mediated tdTomato (red) expression in DRG (left) and TG (right). The tissues were co-stained with either aNF200 (cyan), aCGRP (yellow), or aTRPV1 (blue) markers (scale bars, 100 mm). (D and E) (D) Quantification of the proportion of AAV-mediated tdTomato expressing cells that overlap with aNF200, aCGRP, and aTRPV1 markers in TG (above) and DRG (below), and (E) the proportion of aNF200 marker+ cells that overlap with the AAV-mediated tdTomato expressing cells in DRG and TG (n R 2 per group; each data point represents the average of at least 3 images from each rat; mean ± SEM is plotted for n > 2; mean is plotted for n = 2).

NeuroResource
A similar improvement in neuronal and astrocytic transduction was also observed in the thalamus ( Figure S7C). Consistent with our observations in marmoset, MaCPNS1's transduction was more biased toward neurons, whereas MaCPNS2 showed increased transduction of both neurons and astrocytes. No significant difference in liver transduction was observed between MaCPNS1, MaCPNS2, and AAV9 ( Figure S6).
These experiments demonstrate that the new capsids, MaCPNS1/2, can efficiently transduce the PNS and CNS in both New and Old World monkeys, making them useful vectors for translational research across the nervous system.

DISCUSSION
In this study, we describe systemic AAVs that address the pressing need for efficient gene delivery vectors to target the nervous system across species. By in vivo selection and data analysis, we identified a library of capsids with divergent tropism compared with their parent, AAV9. Two variants, MaCPNS1 and MaCPNS2, were noteworthy for their potent neurotropic behavior in the mouse model in which they were selected. In contrast to our previously engineered PNS-targeting variant, AAV-PHP.S, which requires a high dose to be potent, i.v. delivery of a modest dose of the new variants in adult mice showed about 2-fold improvement in the transduction of NG and DRG compared with AAV9. In addition to improved sensory ganglia transduction, the MaCPNS2 capsid showed improved transduction of the SI of the ENS. These MaCPNS1/2 capsids also stood out from AAV9 and PHP.S in their specificity for the PNS, with lower transduction in the liver. In addition to their performance in adult mice, the new vectors efficiently transduced DRG when delivered at the P1 neonatal stage in mice, via a technically easy i.p. injection. In neonatal mice, both vectors showed a significant bias toward transducing CGRP+ neurons. Experiments involving functional readout or modulation in the sensory system require a high copy number of functional proteins. This can be challenging to achieve with systemic delivery and is thought to require a combination of both an efficient vector and an engineered genetic indicator/probe (Bedbrook et al., 2019;Michelson et al., 2019;Grødem et al., 2021). In this study, we demonstrate that it is possible to use the new MaCPNS1 capsid systemically for both monitoring and modulating neuronal function. By systemically delivering a recently engineered GECI, jGCaMP8s, we were able to visualize vagal neuron calcium dynamics in response to gut glucose infusion and distension. By demonstrating the use of viral vector-mediated GECI delivery for imaging in wild-type mice, we highlight the possibility of performing similar imaging studies in species where transgenic models may not be available. Following our success with GECI sensor delivery, we extended the application of the MaCPNS1 vector to systemic delivery of a DREADD actuator to a TRPV1-Cre neonate, enabling pain induction by chemogenetic modulation of TRPV1+ neurons in DRG, which mediate thermosensation and pain (Cavanaugh et al., 2011;Mishra et al., 2011;Pogorzala et al., 2013). These proof-of-concept experiments demonstrate the potential of MaCPNS1/2 for modulating different sensory modalities with higher temporal resolution. With the rapid development of ultra-sensitive opsins (Bedbrook et al., 2019;Gong et al., 2020) and a wireless light source (Yang et al., 2021), these newly developed AAV variants could potentially open the door to less invasive modulation of hardto-access peripheral ganglia with precise temporal control.
MaCPNS1 and MaCPNS2 also show promise in rats, another commonly used research model for PNS applications (Ravagli et al., 2020;Draxler et al., 2021). Both capsids translated their potent PNS tropism across rodents, showing efficient transduction of sensory ganglia, sympathetic ganglia, parasympathetic ganglia, and enteric neurons. Detailed cell type characterization in the DRG and TG of adult rats showed that the vectors were biased toward transducing NF200+ neurons, in contrast to the bias toward CGRP+ neurons we observed in DRG in mice injected i.p. at the neonate P1 stage. Such tropism shifts have been previously noted with other AAV serotypes and emphasize the importance of considering the roles various experimental conditions play in determining a vector's tropism in a given animal model (Foust et al., 2009;Matsuzaki et al., 2018).
The conservation of the vectors' potent PNS tropism across rodents prompted us to test their performance in NHPs. First, we tested the new AAVs in a New World monkey, the marmoset, which has recently been gaining attention in the neuroscience community as a promising animal model for biomedical research (Marx, 2016;Miller et al., 2016;Jennings et al., 2016). I.v. delivery of MaCPNS1 and MaCPNS2 to adult marmosets showed potent PNS tropism. However, the vectors also efficiently crossed the BBB to transduce the CNS, making them potent vectors across the nervous system. We further validated the vectors' tropism in an Old World monkey, the rhesus macaque, which is more closely related to humans and is widely used as an animal model for pre-clinical research, including gene therapy (Jennings et al., 2016;Hudry and Vandenberghe, 2019;Bey et al., 2020). As in marmosets, i.v. delivered MaCPNS1 and MaCPNS2 efficiently transduced both the PNS and CNS in an infant rhesus macaque. The enhanced CNS tropism we observed in NHPs may be explained by the heterogeneity of the BBB across species.
The conservation of these AAV variants' potent PNS tropism across species validates the usefulness of selecting capsids in mouse models, a preferred model among capsid engineers due to the relatively fewer challenges implementing iterations of in vivo selection or capsid evolution given animal availability. However, the question of translatability for CNS tropism requires further investigation. Prior CNS-specific selections have yielded capsids that may or may not be translatable across species (Matsuzaki et al., 2018;Goertsen et al., 2022) or whose potential has yet to be tested.
(D) fibers in the dorsal column of the spinal cord (scale bar, 500 mm), (E) coronal brain sections of the midbrain (left) and hindbrain (right) (scale bar, 500 mm), and (F) select brain areas: the cortex, thalamus, globus pallidus, cerebellum, and brainstem (scale bar, 400 mm), showing AAV9 vector-mediated expression of eGFP (green) or tdTomato (red), MaCPNS1-mediated expression of eGFP (green), and MaCPNS2-mediated expression of tdTomato (red). The images are matched in fluorescence intensity to the respective AAV9 control. Zoomed-in views of selected areas (dotted white boxes) are shown on the right in (B) and (C). In (B), the zoomed-in view shows the overlap of MaCPNS1 and MaCPNS2-mediated expression with the neuronal marker Tuj1 (blue).
In summary, here we introduce new AAVs to address some significant challenges in the field of gene delivery vectors for the nervous system (Table S1). Non-invasive delivery of transgenes across the nervous system can be transformative for many applications, including basic science, as demonstrated with our previously engineered AAV vectors (Chakrabarti et al., 2020;Marvaldi et al., 2020;Reynaud-  2021; Duan et al., 2022;Monteys et al., 2021). With several therapeutic candidates now in the pipeline for various neurological disorders (Deverman et al., 2018;Hudry and Vandenberghe, 2019;Sevin and Deiva, 2021;Privolizzi et al., 2021), the new systemic AAV vectors described in this study, AAV-MaCPNS1 and AAV-MaCPNS2, offer hope to accelerate translational research as well.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

DECLARATION OF INTERESTS
The California Institute of Technology has filed and licensed patent applications for some of the work described in this manuscript, with X.C., S.R.K., and V.G. listed as inventors. V.G. is a member of the Neuron advisory board and a co-founder and board member of Capsida Biotherapeutics, a fully integrated AAV engineering and gene therapy company.

Lead contact
Further information and requests for resources should be directed to the Lead Contact, Viviana Gradinaru (viviana@caltech.edu).

Materials availability
All mouse strains used in this study are available from Jackson Laboratories. All plasmids and viral vectors generated for the study have been made available from Addgene or can be obtained from the lead contact upon request. Accession numbers are listed in the key resources table.
Data and code availability This paper did not report original code. All data are available upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS
All animal procedures in mice that were carried out in this study were approved by the California Institute of Technology Institutional Animal Care and Use Committee (IACUC), and Harvard Medical School Institutional Animal Care and Use Committee (IACUC). C57BL/6J (000664), Tek-Cre (008863) (Kisanuki et al., 2001), ChAT-IRES-Cre (006410) (Rossi et al., 2011), Nestin-Cre (003771) (Giusti et al., 2014), and TRPV1-Cre (017769) (Cavanaugh et al., 2011) mice were purchased from the Jackson Laboratory (JAX). TH-Cre mice were obtained from the European Mouse Mutant Archive (EM::00254) and crossed with wild-type C57BL/6N mice. Heterozygous TH-Cre mice were used. For capsid selection experiments, 6-8 week old male and female mice were used. For in vivo validation studies of AAV capsid variants, 6-8 weeks old male mice were used.
All procedures performed on rats in this study were approved by the Animal Ethics Committee of the University of Melbourne (Ethics Number 1814639) and complied with the Australian Code for the Care and Use of Animals for Scientific Purposes (National Health and Medical Research Council of Australia). Six male Sprague-Dawley rats (Biomedical Sciences Animal Facility, University of Melbourne) aged 7 weeks were used in this study. Rats were housed in groups of 3 with environmental enrichment under a 12-hour (h) light-dark cycle with ad libitum access to food and water. Six male animals were used in this study and received tail vein injections of AAVs.
All experimental procedures performed on marmosets were approved by the University of California, San Diego, Institutional Animal Care and Use Committee (IACUC) and in accordance with National Institutes of Health and the American Veterinary Medical Association guidelines. 2 female animals and 1 male animal were used in this study and received intravenous injections of AAVs.
All experimental procedures performed on rhesus macaques were approved by the Institutional Animal Care and Use Committee at the University of California, Davis and the California National Primate Research Center (CNPRC). Two infant female animals were used in this study and received intravenous injections of AAVs.
For all the experiments performed in this study, the animals were randomly assigned, and the experimenters were not blinded while performing the experiments in this study unless indicated otherwise.

Library plasmid preparation
The plasmids used for AAV library preparation were described previously (Deverman et al., 2016;Ravindra Kumar et al., 2020) (plasmids available from Caltech CLOVER Center upon request). Briefly, plasmid rAAV-DCap-in-cis-Lox2 ( Figure 1A) was used for building the heptamer insertion (7-mer-i) AAV library. Plasmid pCRII-9Cap-XE was used as a PCR template for the DNA library generation. Plasmid AAV2/9-REP-AAP-DCap was used to supplement the AAV library during virus production.

AAV capsid library generation
The round-1 (R1) and round-2 (R2) libraries were generated as described previously (Ravindra Kumar et al, 2020). Briefly, the R1 library involved a randomized 21-nucleotide (7xNNK mutagenesis) insertion between AAs 588-589 of AAV9 capsid. The R2 library was built using a synthetic pool method (Ravindra Kumar et al., 2020). The R2 library was composed of an equimolar ratio of $9000 variants that were recovered from the tissues of interest in R1 (DRG, heart, small and large intestine). The Spike-in variants as part of the synthetic pool library consisted of previously validated variants such as AAV-PHP.B, AAV-PHP.B4, AAV-PHP.C1 (CNS variants), AAV-PHP.S (PNS variant) and AAV9 parent.
In vivo selection and capsid library recovery For capsid selection in vivo, the virus library was intravenously administered to male and female mice of various Cre transgenic lines (n=2-3 per Cre line) at 2x10 11 vg per mouse in R1 selection, and at 1x10 12 vg per mouse in R2 selection. Two weeks post injection, mice were euthanized, the organs of interest were harvested and snap-frozen on dry ice. The tissues were stored at À80 C for longterm. To recover capsids from the tissue, the tissues were processed using Trizol, and the rAAV genomes were recovered by Credependent PCR or Cre-independent PCR as previously described (Ravindra Kumar et al., 2020). The AAV DNA library, virus library and the libraries recovered from tissue post in vivo selection were processed for NGS as also described previously (Ravindra Kumar et al, 2020).

AAV vector production for in vivo characterization
The AAV vectors were produced using an optimized vector production protocol (Challis et al, 2019). The average yield was $ 1x10 12 vg per plate.

AAV vector administration and tissue harvest in Mice
For the intravenous injection procedures in mice, the AAV vectors were injected intravenously via the retro-orbital route into 6-8 week old adult mice at a dose of 0.1-1x10 12 vg per mouse. The retro-orbital injections were performed as described previously (Yardeni et al., 2011;Challis et al., 2019). The expression times were $3 weeks from the time of injection. The dosage and expression time were kept consistent across different experimental groups unless noted otherwise.
For the intraperitoneal injection procedures in mice, neonatal pups at postnatal stage 1 (P1) were intraperitoneally injected with the AAV vectors at a dose of 3x10 11 or 1x10 12 vg per mouse. Six weeks after AAV administration, tissue collections were performed.
To harvest the tissues of interest, the mice were anesthetized with Euthasol (pentobarbital sodium and phenytoin sodium solution, Virbac AH) and transcardially perfused using 30 -50 mL of 0.1 M phosphate buffered saline (PBS) (pH 7.4), followed by 30-50 mL of 4% paraformaldehyde (PFA) in 0.1 M PBS. The organs were collected and post-fixed 24-48 h in 4% PFA at 4 C. Following this, the tissues were washed with 0.1 M PBS twice and stored in fresh PBS-azide (0.1 M PBS containing 0.05% sodium azide) at 4 C.

AAV vector administration and tissue harvest in Rat
Six 7-week-old male Sprague Dawley rats received lateral tail vein injections of either MaCPNS1 or MaCPNS2 (2x10 13 vg/kg -3 rats/ group). After a 3-week incubation period, animals were transcardially perfused with saline followed by 4% PFA, as per a published protocol (dx.doi.org/10.17504/protocols.io.bahzib76). Tissues were then dissected out and post-fixed in 4% PFA for 1 h before being washed in 0.1 M PBS (3 3 30 min). Tissues were stored in PBS-azide until processing. In one rat (MaCPNS2 injection), the perfusion fixation process was unsuccessful, so tissues were fixed only by immersion (18-24 h), prior to washing and storage as described above.

Quantification of AAV transduction in vivo
The quantification of AAV transduction across NG, DRG and GI tract was carried out by manually counting fluorescent expression resulting from the AAV genome. The Adobe Photoshop CC 2018 Count Tool was used for this purpose. To quantify expression in the liver, we used Keyence Analyzer automated cell count software. The efficiency was determined by the percentage of cells expressing EGFP or tdTomato relative to a specific cell marker, namely, NeuN, PGP9.5, DAPI, CGRP, NF200, TRPV1, Tuj1, GLUT1, Olig2 or S100b.

NGS data alignment, processing and analysis
The raw fastq DNA files were aligned to the AAV9 capsid template using custom alignment software as described previously (Ravindra Kumar et al., 2020) (https://github.com/GradinaruLab/mCREATE). The NGS data analysis was carried out using a custom data-processing pipeline with scripts written in Python https://github.com/GradinaruLab/mCREATE) and plotting software such as Plotly, Seaborn, and GraphPad PRISM 7.05. The AAV9 capsid structure model was produced with PyMOL.
The enrichment score for a variant was determined using the following formula: Enrichment score of variant ''x'' = log 10 [(Variant ''x'' RC in tissue library1/ Sum of variants N RC in library1) / (Variant 1 RC in virus library/ Sum of variants N RC in virus library)] Where N is the total number of variants in a library. The fold-change of a variant ''x'' to AAV9 = (The enrichment of ''x''-The enrichment of AAV9)/|The enrichment of AAV9|.
Data analysis for calcium imaging Imaging data was analyzed using CaImAn-MATLAB (Giovannucci et al., 2019;Corder et al., 2019) with modified MATLAB code adapted from https://github.com/flatironinstitute/CaImAn-MATLAB. Basically, imaging frames from the same animal were first registered to correct for motion. A non-negative matrix factorization (CNMF) algorithm was applied to recognize individual cells and to extract fluorescence activities. A 30 s window before the stimulus onset was used as baseline signal. To quantify signals, the mean (m) and standard deviation (s) of F0(t) over the baseline period were computed as F(t) = (F0(t) À m)/s. Cells were defined as responsive if the average DF/F(s) value during the stimulus period was more than 3 s.d. above the baseline mean activity.