Optimization of AAV vectors to target persistent viral reservoirs

The first 6 identified AAV serotypes were isolated as live viruses from human or non-human primate (NHP) cell cultures and included AAV2, the first to be vectorized. AAV2 vectors can transduce many tissue types, including liver, muscle, lung, retina, and brain [3], and most early pre-clinical and clinical trials used AAV2 vectors to treat congenital diseases, including Leber's congenital amaurosis, retinal dystrophy caused by RPE-65 deficiency [33‐36]. For example, in 2017, subretinal injection of the rAAV2-RPE65 became the first FDA-approved AAV-based gene therapy for a genetic disease [37]. More recently, the search for new serotypes with beneficial properties has resulted in PCR-based strategies that amplify AAV DNA from samples derived from humans, NHP, and other species [23, 24, 38]. To date, over a hundred variants have been isolated, mostly from humans and NHP, and many have been vectorized (Table 1). Notably, two of the first new serotypes isolated were the rhesus macaque-derived AAV8 and the human-derived AAV9 serotypes that have shown strong tropism for liver or multiple organs, respectively [39‐41]. These serotypes have gained widespread clinical interest for use in diseases such as hemophilia B, where high FIX expression levels sufficient to elicit a phenotypic correction have been achieved using AAV8 vectors in hemophilia patients [42‐50]. Ultimately, the isolation of new AAV serotypes with a wide range of tissue tropisms – primarily determined by different receptor and co-receptor usage [3, 28, 51, 52] – has significantly expanded the AAV vector tool kit with isolates that can transduce various tissues and partially evade neutralizing antibodies. This underscores the usefulness of screening for natural AAV across a wide swath of species. However, the isolation and characterization of novel variants is time-consuming, which has motivated the generation of novel engineered capsid mutants and library screening approaches to maximize the number of potential vector capsids that can target the tissue of interest.

Directed evolution has been widely used to identify new AAV capsids with beneficial properties via the screening of mutant capsid libraries, either in cell culture or in vivo. While the screening methods used have varied, to date three main approaches have been used to generate large AAV capsid libraries from which beneficial capsid variants are selected. The first approach to be used for AAV library generation relied on the observation that targeting peptides can be inserted into exposed surface loops of the AAV2 capsid at specific sites without altering capsid assembly [53]. Subsequently, it has been demonstrated that mutants generated by this screening approach can be retargeted to alternative cell types and evade antibody neutralization [53‐56]. Libraries with random peptide insertions of 7–12 amino acids at AAV2 cap position 587/588 are now widely used to identify tissue-tropic mutant capsids. This approach has even been expanded to insert peptides at the same location in other serotypes such as AAV9 [57‐62]. The second main approach used to generate AAV capsid libraries utilizes error-prone PCR of the AAV2 cap gene to create functional mutants and has been used to identify capsids with altered receptor binding affinity and the ability to evade neutralizing antibodies [63, 64]. Error-prone PCR has also been used to beneficially mutagenize other AAV serotypes. For example, surface-exposed residues found in the AAV9 cap gene were mutagenized via error-prone PCR, which enabled the identification of mutants with ablated tropism for liver and enhanced cardiac or musculoskeletal gene transfer [65]. The third main approach used to generate AAV capsid libraries pioneered DNA shuffling and PCR-mediated reassembly of randomly sheared AAV cap gene fragments from different AAV serotypes to generate capsid chimeras [66‐69]. This DNA shuffling approach has also successfully identified novel capsids with altered tropism, the ability to evade neutralizing antibodies, or both. For example, AAV-DJ, a widely used chimera of serotypes 2, 8, and 9, was selected for its ability to evade neutralizing antibodies relative to other serotypes [66], and AAV-LK03, a chimera of 7 different serotypes, was selected for its ability to transduce human hepatocytes in humanized FRG mice [70], and is currently being evaluated in a clinical trial for hemophilia A [71]. In summary, directed evolution of AAV offers researchers the capability to select capsids from large and highly diverse mutant capsid libraries that provide distinct advantages over existing vectors, and has led to the identification of many new AAV capsids with beneficial properties such as improved transduction efficiency, reduced immunogenicity, broadened tissue tropism, or refined tissue tropism for specific cell populations [62, 65, 66, 72‐77].

The utility of AAV as a gene therapy vector has fueled research to understand the basic biology of AAV. Sixty years after the discovery of AAV, structures for multiple AAV serotypes have now been resolved, allowing for a better understanding of capsid assembly and the basis for attachment to cell surface receptors [78‐81]. This information, along with details about capsid receptor interacting domains, intracellular trafficking, uncoating, genome replication, and neutralizing epitopes, has informed the rational design of AAV vectors with enhanced properties through several different approaches. One approach to rational design relies on direct modification of the existing capsid amino acid footprint to transfer beneficial characteristics from one capsid to another. This approach may involve the substitution or deletion of single amino acids or entire domains between different AAV capsids. Examples of this approach include direct modification of single surface-exposed tyrosine residues to phenylalanine, preventing tyrosine-mediated ubiquitination and degradation of capsids after internalization [82]; substitution of variable loop domain residues to create a capsid with reduced immunogenicity [83]; or the substitution/deletion of an entire receptor binding domains that confer novel tropism upon the donor or recipient capsid [84]. Another approach to rational capsid design involves inserting non-virus elements into the capsid to convey beneficial properties. This has been primarily done to retarget AAV to alternate cellular receptors. For example, small targeting molecules have been linked to the AAV capsid via direct genetic fusion to AAV capsid proteins or via chemical conjugation to surface-displayed biotin acceptor peptides or split inteins. These studies have enabled AAV retargeting to cellular receptors including CD4, CD30, CD34, CD133, Her2/neu, and EpCAM, and offer great hope for developing AAV vectors with specificity for individual cell types [85‐92]. A more recent approach to rational design has created entirely artificial AAV capsids using phylogenic analyses to reconstruct the gene sequences from common ancestors of extant AAVs. Reconstructed ancestral AAV capsids generated via this approach appear to be more thermostable and are highly potent for several different cell and tissue types [93, 94]. In a similar approach, germline endogenous viral elements (EVEs) from the Dependoparvovirus lineage have also been identified in a number of mammals, birds and marsupials [95‐99]. These EVEs are now being used to guide the rational design of novel AAV vectors based on their distinct structural elements that could provide unique vector properties or tropisms [99]. As we continue to learn more about AAV and how it interacts with cells and tissues, new rational design strategies will be developed that enable us to discover improved vectors for different therapeutic applications.

Hepatitis B virus (HBV), hepatitis C virus (HCV), and hepatitis D virus (HDV) are hepatotropic viruses that cause viral hepatitis and can establish persistent/chronic infections in the liver. In 2015, an estimated 257 million people were living with chronic HBV infection, with an estimated 5% of these co-infected with the satellite HDV, an incomplete virus that requires HBV to establish infection and worsens HBV infection severity. A further 71 million people are estimated to have chronic HCV infection. It is estimated that approximately 4% of all new cancers are caused by chronic HBV and HCV infections [100], and together they were responsible for 1.34 million deaths in 2015 due to complications from cirrhosis and hepatocellular carcinoma (HCC), a figure projected to increase by > 60% in 2040 [101]. While a curative treatment has been recently developed for HCV, chronic HBV infection remains incurable, although viremia can be suppressed with antiviral drugs, and infection can be prevented by vaccination [100, 102‐104].

Current antiviral treatments for HBV infection include nucleoside analogs (NAs) and immunomodulatory therapy, which can effectively suppress HBV replication, lower the rates of cirrhosis and HCC, and reduce the rate of mother-to-child transmission [105, 106]. However, these treatments require long-term adherence and rarely result in a permanent cure. The basis of HBV persistence is covalently closed circular DNA (cccDNA), the template for HBV genome replication and transcription in infected hepatocytes [107]. The cccDNA reservoir in hepatocytes is stable and can be maintained without the need for re-infection, as newly formed infectious particles can recycle from the cytoplasm back into the nucleus and replenish the reservoir [108]. Thus, elimination or inactivation of the cccDNA reservoir is the ultimate goal of new curative therapeutic strategies [12, 108‐110].

Gene therapy to eliminate, mutate, or silence the HBV genome has been explored pre-clinically by a number of groups in several cell culture systems and animal models [reviewed elsewhere [12, 14, 109, 111]]. Some strategies include the delivery of RNA interference (RNAi) activators to inhibit HBV gene expression or the delivery of gene-editing enzymes such as CRISPR/Cas9, zinc finger nucleases, and transcription activator-like effector nucleases that cleave cccDNA, resulting in its degradation or inactivation through the introduction of mutations. Various delivery methods have been employed, including non-viral delivery using lipoplexes and adenoviral or lentiviral vectors. Of relevance to this review, recent studies have reported that AAV vectors can effectively deliver anti-HBV therapeutics to hepatocytes in vitro and the liver in vivo [18, 28, 111‐116]. Therefore, AAV vectors are attractive gene transfer tools to deliver anti-HBV therapeutics due to their well-characterized liver transduction capabilities and safety profile.

In addition to the inherent potency and specificity of gene editing enzymes or RNAi activators for HBV, AAV vector optimization to maximize transduction and transgene expression in human hepatocytes is critical to gene therapy's effectiveness for viral infections. In recent years, significant advances in the development of liver-tropic AAV vectors have emerged from the gene therapy field that can also be harnessed to treat persistent hepatic infections [117]. Due to the liver's essential role in metabolism and systemic delivery of proteins into blood, many inherited and acquired diseases, such as hemophilia (A and B), α − 1 antitrypsin deficiency, and ornithine transcarbamylase deficiency, affect the liver and can potentially be corrected by liver-targeted gene therapy [2, 3, 72, 118]. For example, several clinical trials for hemophilia B have produced an array of data about AAV in humans, useful for improving hepatocyte tropism, transduction efficiency, and reducing immunogenicity [119]. Most of these studies have focused on engineering the AAV capsid, one of the major tropism determinants, through directed evolution, rational design, or a combination of both [120‐122].

Recent efforts to improve AAV vectors for liver-targeted applications have focused on three areas: (1) maximizing human hepatocyte tropism, (2) evasion of neutralizing antibodies, and (3) streamlining pre-clinical evaluations of liver transduction efficiency for novel AAV capsids. The first two have mostly focused on engineering the AAV capsid, which largely determines both tropism and antigenicity. The third is motivated by discrepancies in transduction levels observed in clinical trials compared to those obtained in pre-clinical studies in cell culture and animal models. Over the last few years, liver humanized mouse models have emerged as an important tool to study AAV transduction differences between serotypes at the pre-clinical stage. To effectively study antiviral therapies against HBV, liver humanized mice are crucial since HBV cannot replicate in most small mammals. In these xenograft models, primary human hepatocytes (PHHs) are transplanted into immunodeficient mice with genetic mutations that elicit murine liver injury. This deficit confers a growth advantage to PHH over murine hepatocytes, enabling high engraftment levels [123‐126]. Still, it is unclear how well these models recapitulate in vivo delivery in humans.

As previously discussed, AAV2 has successfully been used to treat blindness in clinical trials. In contrast, an AAV2 vector used for liver-targeted expression of FIX in early clinical trials for hemophilia B showed low efficacy, an interesting result given the robust transduction of human hepatocyte-derived cell lines by AAV2 vectors in vitro. A recent study suggests that tissue culture adaptations, possibly dating back to the early propagation of AAV2 in vitro, caused increased affinity for its primary cellular receptor heparan sulfate proteoglycan (HSPG). This resulted in enhanced transduction of primary hepatocytes and cell lines in vitro, while simultaneously reducing hepatocyte transduction in vivo [127]. In the same study, clade B isolates found in primary human liver samples that are similar to AAV2 showed reduced in vitro tropism for hepatocyte cells, but increased tropism for human hepatocytes in humanized FRG mice. Furthermore, these primary isolates could be adapted for tissue culture in Huh7 cells via iterative passaging with adenovirus type 5, but this resulted in attenuated in vivo transduction of hepatocytes. This study demonstrated the complexities of optimal capsid identification when using different in vitro or in vivo systems for selection.

The generation of AAV capsid libraries followed by a selective screen has significantly increased the repertoire of AAV capsids with modified transduction properties. In 2014, Lisowski et al. used the humanized liver FRG mouse model to identify AAV capsid mutants with improved human hepatocyte transduction in vivo [70]. This directed evolution methodology selected for variants with human hepatocyte receptor binding and entry capabilities. The group identified the capsid mutant AAV-LK03, which is closely related to AAV3B, differing only by eight amino acid changes. In an in vivo vector specificity analysis in humanized FRG mice, they showed that AAV-LK03 exhibits a stronger tropism for human hepatocytes in humanized mouse livers than AAV8, and AAV-LK03 is now being evaluated in a phase I/II clinical trial to treat hemophilia A (ClinicalTrials.gov: NCT03003533). Early results from this trial indicate that albeit safe at lower doses, high doses resulted in severe adverse events and subsequent loss of transgene expression in patients [71]. Recently, a rational design approach has been used to optimize AAV-LK03 for gene therapy applications via site-directed mutagenesis of surface-exposed residues. The site selection was informed by previous studies in AAV3B, which demonstrated that the elimination of specific surface-exposed serine and threonine residues enhances transduction efficiency while retaining viral tropism and cellular receptor interactions [128]. This study showed that applying rational design to library-derived variants is a promising tool for achieving superior results in clinical settings.

Since AAV-LK03 was first characterized, several studies have reported conflicting results regarding the superiority of AAV-LK03 over other liver-tropic serotypes in mice with human livers and NHP [114, 129‐131]. In one comparative study of liver gene transfer using natural and engineered AAV serotypes, AAV3B, AAV8, AAVrh10, and AAV-LK03 all transduced NHP livers and human hepatocytes. In contrast to Lisowski et al., AAV-LK03 was not found to be superior to either AAV3B or AAV8 as a potent liver-specific vector [129]. More recently, the experimental variables that could affect AAV transduction of human hepatocytes were analyzed in a study using liver humanized FRG mice [132]. This study demonstrated that NTBC cycling, PHH donor origin, and the AAV vector dose could substantially affect transduction efficiency in human hepatocytes. These experimental variables could partially explain the discrepancies in AAV vector transduction efficiency seen between different laboratories using the FRG mouse and aid in the much-needed standardization of chimeric liver mouse models.

Although enhanced liver transduction efficiency has been seen with AAV-LK03 in some studies, this mutant is still moderately sensitive to pre-existing levels of neutralizing antibodies in humans [133]. To directly counteract humoral immunity against novel AAV capsids, a recent directed evolution study screened evolved human hepatotropic AAV capsids –obtained after five rounds of selection in FRG mice– against pools of human immunoglobulins pooled from thousands of patients to select capsids that can evade neutralization [120]. In this study, independent results from two laboratories showed that the newly identified capsid mutants AAV-NP40 and AAV-NP59 display superior transduction of human hepatocytes over AAV-LK03, regardless of human hepatocyte repopulation levels in liver humanized FRG mice. Importantly, reduced seroreactivity was seen for AAV-NP40 and AAV-NP59 relative to AAV-LK03 using serum from a cohort of 50 healthy US adults of mixed gender.

In another directed evolution approach, intravenous immunoglobulin was passively transferred into liver humanized mice before administering the AAV capsid library to identify mutants that could evade AAV neutralizing antibodies and transduce human hepatocytes in vivo [134]. After four cycles of selection, mutant AAV-LP2-10, composed of capsids derived from AAV2, AAV6, AAV8, and AAV9, was the dominant isolate. Using immunohistochemistry and flow cytometry as metrics for transduction efficiency in humanized mice, AAV-LP2-10 transduced human hepatocytes at similar levels to AAV8. Of note, several studies have reported AAV8 as a poor functional transducer of human hepatocytes in vivo [70, 135]. Nevertheless, AAV-LP2-10 was able to robustly escape pooled human immunoglobulins relative to AAV1, AAV2, AAV3, AAV6, AAV8, and AAV9. In twenty serum samples from the healthy donors, mutant AAV-LP2-10 had low neutralizing antibody titers, similar to AAV9 –one of the serotypes with the lowest prevalence of anti-AAV antibodies in healthy humans [31].

Individually, directed evolution and rational design have successfully created novel AAV vectors with enhanced liver-transduction capabilities and lower immunogenicity profiles. However, the combination of both methods has gained popularity in recent years [121, 134]. This approach involves the rational design of mutant libraries with targeted mutations in residues that are likely to affect capsid function during the screening process, thus maximizing the identification of mutants with desired features. A recent study combined rational design with directed evolution to select for liver-targeted AAV3B-derived variants [122]. The library design only allowed for randomization of residues in surface-exposed VRs while keeping the backbone sequence intact to maintain structural integrity. Moreover, to reduce the likelihood of detrimental amino acid substitutions, the allowed amino acids at each mutated site were limited to those that occur naturally at each residue in 150 native serotypes. The combinatorial AAV3B capsid library was serially screened for five rounds in vitro using 3D human hepatocellular carcinoma spheroid cultures. From this screen, variant capsid AAV3B-DE5, which contains 24 amino acid substitutions compared to AAV3B, became predominant. Although validation experiments in FRG mice demonstrated that AAV3B-DE5 transduces human hepatocytes at similar levels to AAV-LK03, the seroreactivity of AAV3B-DE5 relative to the parental AAV3B capsid improved significantly, showing that extensive changes in the amino acid sequence of VRs can effectively reduce pre-existing antibody neutralization without including neutralizing antibodies during the selection process.

In summary, the identification of liver-tropic AAV vectors in pre-clinical settings using animal models and cell culture systems has been extensively optimized to select for variants that can both transduce hepatocytes at high levels and escape pre-existing immunity. As the liver is an important target for gene delivery in many gene therapy applications, humanized liver mouse models will continue to be used in the search for vectors that meet both criteria.

DRG houses neuronal cell bodies of the somatic nervous system that project from cervical, thoracic, lumbar, and sacral vertebrate and are involved in the relay of sensory information from peripheral sites to the central nervous system (CNS). DRG have been identified as a primary target for pain, spinal cord injury, and neurodegenerative disease studies. These clusters of neurons are also major 'reservoirs' for viral DNA during latent VZV and genital HSV infections [139‐141], as alphaherpesvirus DNA can be detected in multiple DRG along the spine during latent infections. This makes DRG essential targets for curative antiviral therapies requiring gene transfer.

Localized AAV delivery routes such as direct lumbar or sciatic nerve injections can facilitate efficient gene transfer to individual or multiple closely situated DRG. Following localized injection directly into the DRG or the sciatic nerve, AAV vectors including serotypes 1, 2, 5, 6, 7, 8, 9, rh10, and an engineered AAV2-retro vector can all efficiently transduce neurons of the DRG in mouse, rat, and pig animal models [142‐148]. In one of these studies, a head-to-head comparative analysis was performed in mice using seven different serotypes after localized delivery to the DRG. Serotypes 1, 5, and 6 performed best with up to 90% of neurons, both IB4+ and CGRP+, expressing the GFP transgene after 12 weeks [145]. Although localized delivery demonstrates the promise of AAV for DRG gene transfer, comprehensive gene delivery to multiple DRG situated along the spine will likely require a more generalized approach, so other studies have investigated widespread DRG gene delivery via different AAV delivery routes.

Indirect uptake via localized AAV delivery to afferent nerve terminals that project from peripheral sites has shown promise as an approach to transduce multiple ganglia. In one such study, AAV8-GFP vectors were injected into the footpad of mice, resulting in the transduction of > 90% of sensory neurons in multiple DRG that innervate the footpad [149]. Furthermore, efficient co-labeling of HSV+ neurons was seen from the AAV8-GFP vector, demonstrating the utility of this delivery approach in potential anti-HSV therapies. In a similar approach, an engineered AAV capsid vector (PHP.S) was delivered directly into the knee joint of mice, and uptake by nerves that innervate the knee from the DRG was seen, with up to 7% of lumbar L2-L5 neurons, including TRPV1+ neurons, transduced [150]. In the same study, delivery of therapeutic transgenes that target muscarinic receptors showed that it was possible to increase or decrease knee neuron excitability, thus demonstrating the use of AAV for studying the role of the DRG in pain responses. While the footpad and knee delivery routes show promise for DRG transduction in rodents, they may not translate well to large animal models. Other systemic delivery routes have also been studied as alternatives.

Delivery of therapeutic transgenes to multiple DRG via the cerebrospinal fluid (CSF) or blood offers a more straightforward route to widespread gene delivery in multiple DRG, and several groups have investigated this. When delivered via intrathecal (IT) injection, AAV vectors can efficiently transduce sensory neurons in multiple DRG [144, 147, 151‐156] and AAV serotypes 5, 6, 8, 9 and rh10, as well as an engineered AAV8 capsid (AAV8.2) containing the AAV2 phospholipase A2 domain, have all shown efficient DRG transduction following IT delivery in mouse, rat, and macaque animal models. While these studies offer promise for future studies in humans, a pre-clinical study investigating gene transfer following IT delivery of an AAV9 vector expressing α-1-iduronidase (IDUA) to infant rhesus macaques offers the most hope for the future development of AAV-mediated antiviral therapies against alphaherpesviruses in humans [151]. This study showed that 4–29% of neurons within the cervical, thoracic, and lumbar DRG of 4 study animals expressed the IDUA transgene as long as three years and eight months post AAV administration.

Intravenous delivery of AAV vectors should, in theory, enable gene delivery to any cell within close proximity of the vasculature. However, unlike IT administration, IV delivery results in the circulation of AAV vectors through organs that are not part of the nervous system, such as the liver and spleen, where vector is often sequestered at high levels before it has a chance to see the target organ. Subsequently, sensory DRG likely see a lower relative effective AAV dose via IV delivery than IT, and mostly, AAV vectors do not appear to transduce sensory ganglia as efficiently through this administration route. Unfortunately, non-specific sequestration is inherent to most naturally occurring AAV capsids. However, one approach to prevent this has been to screen AAV capsid libraries for engineered variants that can transduce sensory ganglia when delivered IV to bypass non-specific sequestration. In vivo library selection has been used to identify AAV capsids with enhanced tropism for neurons of the CNS and PNS with great success and has yielded at least one novel capsid (AAV.PHP.S) that can efficiently transduce up to 82% of DRG neurons in mice after IV delivery [59, 61, 157]. While the PHP.S capsid does not appear to transduce other sensory ganglia as efficiently [10], future in vivo library selections may identify new AAV capsids that transduce DRG neurons and other sensory ganglia with high efficiency via any delivery route.

In mammals, two trigeminal ganglia (TG) sit below the brain, and house cell bodies of tactile, proprioceptive, and nociceptive afferent somatic sensory neurons that innervate the face. These TG are responsible for relaying craniofacial sensory and motor nervous stimuli from the left or right side of the face into the CNS through a large sensory root and small motor root that enters the brainstem. Like the DRG, the TG is a target for studies of pain, but it is also a reservoir for viral DNA during latent VZV and craniofacial HSV infection [139‐141], and is a source of reactivating virus during cold sore or other head and neck HSV/VZV flare-ups. The TG is therefore an important target for curative anti-alphaherpesvirus gene therapies.

Direct injection of AAV delivery into the TG is extremely challenging due to its location immediately below the brain. Therefore, attempts to transduce sensory neurons of the TG using AAV vectors have mainly focused on indirect vector transductions methods involving delivery via projecting efferent neurons of the face or IV administration. Transduction of the TG has been attempted via delivery of AAV to the eye since efferent neurons from the TG project into the conjunctiva of the eye, but species-specific differences have been seen in delivery efficiency. In mice, the delivery of AAV to the eye has proved inefficient for TG delivery with minimal transduction of sensory neurons seen [158]. Conversely, in the rabbit, efficient transduction of TG neurons is seen following delivery to the eye, with AAV/HSV co-infection of sensory neurons also observed, demonstrating that this delivery approach could be used in the assessment of potential anti-HSV therapies [149]. As an alternative to the eye, intradermal AAV injection into the snout has been assessed in mice since efferent neurons project from the TG into the whisker pad. Following mouse whisker pad delivery, neuronal transduction in the TG can be seen from AAV1, AAV7, AAV8, or AAV9 vectors, with the highest levels (over 20%) seen with AAV1 [158]. Furthermore, it has been shown that gene editing meganucleases targeting HSV sequences can directly cleave HSV DNA in the TG of mice when AAV1 vectors are delivered via the whisker pad and a latent HSV infection is established via ocular challenge [159]. More recent efforts to target the TG have shown that IV administration of certain serotypes can provide more efficient transduction of the TG in mice. We have found that AAV8 and AAV.rh10 vectors can deliver transgenes to the TG with high efficiency when delivered IV [10]. Furthermore, when a triple serotype (AAV1/AAV8/AAV.rh10) combination HSV-specific meganuclease therapy is delivered IV, up to 55% of latent HSV-1 DNA found in the TG of mice during latent infections can be eliminated [10]. This data demonstrates the promise of AAV vectors for targeting the TG and also shows that the treatment/cure of persistent alphaherpesvirus infections may eventually be feasible.

The autonomic nervous system (ANS) is a branch of the PNS that unconsciously controls many bodily functions and can be divided into sympathetic 'fight-or-flight' neurons, parasympathetic 'rest-and-digest' neurons, and enteric neurons that control motor functions of the GI tract. The potential to transduce autonomic neurons would be useful for studying many aspects of neurological function across multiple body systems, and AAV vectors have been investigated to this end. Like somatic neurons, autonomic neurons of the sympathetic, parasympathetic, and enteric nervous systems can also harbor latent alphaherpesvirus DNA within neuronal cell bodies of distinct ganglia [160‐162], so the ability to transduce autonomic neuronal populations is also important for the development of curative antiviral gene therapeutics targeting HSV and VZV.

While more limited in number than studies in the somatic nervous system, some studies have shown that ANS neurons can be efficiently targeted with AAV vectors. In mice, AAV8 and AAV9 vectors can efficiently transduce up to 57% of myenteric and submucosal neurons of the GI tract when delivered systemically via IV injection [163, 164]. In guinea pigs, AAV8 can transduce myenteric and submucosal gut neurons when delivered IV [165]. In cynomolgus macaques, AAV9 can transduce myenteric neurons of the stomach and small and large intestine after IV administration [165]. More recently, we were able to show that the neurons of the superior cervical ganglia (SCG), a part of the sympathetic ANS, can also be transduced with high efficiency by AAV vectors delivered IV. We found that AAV8 and AAV.rh10 vectors can transduce HSV infected neurons of the SCG in latently HSV-1 infected mice with high efficiency and that latent viral loads in the SCG can be reduced by more than 90% following triple serotype (AAV1/AAV8/AAV.rh10) HSV-specific meganuclease delivery [10]. Overall, these studies demonstrate the potential of AAV vectors for use in studying diseases of the autonomic nervous system.

Despite the promise shown by AAV vectors for gene delivery to sensory neurons in vivo, several recent studies in large animals suggest that IV or IT delivery of high doses of AAV vectors may lead to asymptomatic injury to sensory neurons in vivo. Mononuclear cell infiltration and minimal to moderate asymptomatic degeneration of DRG neurons and associated axons has been seen in NHP, and more severe sensory neuronal lesions were seen in pig DRG [166‐169]. While the toxicity seen in the DRG of NHP receiving AAV has been described as subclinical in some studies [151], a more comprehensive meta-analysis of 256 NHP receiving AAV vectors revealed that DRG pathology is almost always seen in NHP following AAV delivery as it was detected in 83% of animals receiving AAV via CSF delivery, and 32% of animals receiving AAV via IV injection. Furthermore, abnormal DRG pathology was shown to be independent of the different capsids (5), promoters (5), transgenes (20), or purification methods used for each AAV vector. These studies raise legitimate concerns for future studies that target sensory neurons of the PNS via AAV vectors. However, a recent study suggests that AAV-associated DRG toxicity can be reduced when a microRNA target is included in the vector backbone that selectively knocks down transgene expression in the DRG [170]. Further NHP studies addressing the mechanism and extent of adverse AAV-mediated pathology in DRG and other sensory ganglia will be needed as AAV vectors are advanced for use in therapies targeting the PNS.