Developing New Vectors for Retinal Gene Therapy

Abstract

Since their discovery over 55 years ago, adeno-associated virus (AAV) vectors have become powerful tools for experimental and therapeutic in vivo gene delivery, particularly in the retina. Increasing knowledge of AAV structure and biology has propelled forward the development of engineered AAV vectors with improved abilities for gene delivery. However, major obstacles to safe and efficient therapeutic gene delivery remain, including tropism, inefficient and untargeted gene delivery, and limited carrying capacity. Additional improvements to AAV vectors will be required to achieve therapeutic benefit while avoiding safety issues. In this review, we provide an overview of recent methods for engineering-enhanced AAV capsids, as well as remaining challenges that must be overcome to achieve optimized therapeutic gene delivery in the eye.

ADENO-ASSOCIATED VIRAL VECTORS FOR RETINAL GENE THERAPY

Simultaneous advances in genetics, natural history studies, and gene delivery have resulted in the advancement of gene therapy. Gene therapy is rapidly becoming a promising option to treat inherited and complex retinal diseases. However, for recent advances to translate into clinically meaningful outcomes, noninvasive and precise gene delivery tools are required to efficiently deliver genes to affected cell types and to prevent, contain, or reverse the course of retinal diseases. Adeno-associated virus (AAV) vectors are the gold standard for gene delivery in the retina. However, accumulating evidence suggests that methods beyond the classical approaches of biomining and rational design will be required to fully understand AAV biology and to create sufficiently effective gene delivery vectors. To this end, new approaches have been developed to make AAVs more efficient, including high-throughput screening and machine learning (ML)-guided capsid development. Here, we discuss obstacles to AAV development in the context of intricate intraocular spaces, and we describe bioengineering methods that can be applied to overcome these challenges.

STRUCTURE OF AAV

AAVs are single-stranded DNA viruses, part of the Parvoviridae family, and within the Dependoparvovirus genus (Hastie and Samulski 2015). To replicate, AAVs require the presence of adenovirus, herpesvirus, or papilloma virus genes (McPherson et al. 1985; Weindler and Heilbronn 1991; Cotmore et al. 2019).

AAVs are small (25 nm) with a genome of 4700 base pairs containing three open-reading frames and two viral genes—rep (replication) and cap (capsid)—that encode nonstructural, structural, assembly-activating proteins (AAPs), and membrane-associated accessory proteins (MAAPs). The open reading frames are flanked by two inverted terminal repeats (ITRs). ITRs have an essential role in inducing transgene expression, vector production, and cell transduction, and they can form hairpin structures by self-annealing (Buller and Rose 1978; Wistuba et al. 1995; Sonntag et al. 2010). The rep gene encodes four regulatory proteins—Rep78, Rep68, Rep52, and Rep40—that play a role in AAV genome replication and virion assembly. The cap gene encodes three structural virion proteins, VP1, VP2, and VP3, which form the capsid with the help of the AAP. VPs are generated through alternative splicing and the use of an alternate translational start codon (ACG) (Srivastava et al. 1983). Therefore, VP1, VP2, and VP3 share a common carboxyl terminus.

The AAV capsid consists of 60 virion proteins, comprising a mixture of VP1, VP2, and VP3 at a 1:1:10 estimated ratio, organized in T = 1 icosahedral symmetry (Xie et al. 2002; Govindasamy et al. 2006). VP3 has a molecular weight of 59–61 kDa and its sequence is shared among all VPs. VP2 has a molecular weight of 64–67 kDa. VP1 has a molecular weight of 79–82 kDa and a unique 137 aa amino-terminal region (VP1u), which is critical for successful infection. Although VP3 is capable of forming the capsid by itself, the VP1/VP2 common region, as well as VP1u, are important for nuclear localization, genome release, and endosomal trafficking and escape.

WHY AAV?

AAV was first discovered over 55 years ago (Atchison et al. 1965), and has since become the main viral vector currently used in gene therapy clinical trials. AAV's potential for safely delivering genetic cargo was first demonstrated in the 1980s, when the virus's wild-type genes were removed and substituted with a transgene (Samulski et al. 1987). In the subsequent 35 years, AAV has been continuously optimized to perform the task of delivering therapeutic genes to target cells.

With the removal of the rep and cap genes, AAV is incapable of replicating in host cells, even in the presence of adenovirus or herpesvirus. Replication incompetency, along with comparatively low immunogenicity, make AAV an attractive gene therapy vector. AAV has successfully been used as a gene vector in animal models for the past three decades. Furthermore, it has been the viral vector of choice for many clinical trials over the last decade, especially for inherited retinal diseases. Thirteen naturally occurring serotypes of AAV (AAV1 to AAV13) each possess different cell tropisms (Bulcha et al. 2021). This naturally occurring toolbox of AAV vectors, with varying rates of transduction and immunogenicity, comprises a range of options for gene delivery vectors. AAV capsid proteins may also be tailored through engineering, which has resulted in 100s of additional AAV variants with unique tropisms. Furthermore, AAV can package a double-stranded, self-complimentary AAV (scAAV) genome instead of a single-stranded construct, speeding up the rate of protein expression by bypassing the complimentary strand synthesis (McCarty et al. 2001). Additionally, a large variety of AAV-compatible promoters and regulatory sequences are available, making AAV a highly customizable vector.

Besides AAV, additional viral vectors have been used for gene therapy in the eye. Adenovirus has a carrying capacity of 37,000 base pairs, accommodating large transgenes. However, the high immunogenic response to adenovirus leads to potentially adverse effects and unknown transduction rates. Furthermore, clearance of transgene-encoded protein results in short-lived therapeutic effects. Lentiviruses have also been used, and they have an 8000 base pair carrying capacity. Their genetic cargo undergoes random integration into the host cell genome, which has mixed outcomes, as random integration can have deleterious effects. While genomic integration is attractive for cells with a high mitotic index, the retina is a postmitotic tissue. Importantly, adenovirus and lentivirus have very limited tropism for photoreceptors compared to AAVs. For these reasons, AAV is the preferred viral vector for gene delivery in the retina.

OBSTACLES TO BE OVERCOME

Despite its promise for treating diseases affecting tissues across the body, numerous obstacles prevent the effective translation of AAV-mediated gene therapies to the clinic (Fig. 1). The most significant limitations of AAV vectors include limited tropism for affected cell types, limited diffusion of the vector across structural barriers such as the inner limiting membrane (ILM), limited carrying capacity and inability to efficiently package large cargos, and immune response to the capsid or gene product delivered, as well as inactivation through neutralizing antibodies (NABs).

Figure 1.

Obstacles preventing efficient retinal gene therapies. (A) Limited tropism. (B) Limited diffusion across barriers. (ILM) Inner limiting membrane. (C) Limited carrying capacity. Adeno-associated virus (AAV) vectors can package up to 4.7 kb, aside from the inverted terminal repeat (ITR) sequences. (D) Immune response. (Figure was created with BioRender.com.)

Limited Tropism

To date, over 13 naturally occurring serotypes and more than 100 AAV variants have been identified (Kotterman and Schaffer 2014). The tropism of natural serotypes is generally broad, and AAV vectors with the capacity to specifically target retinal cell types are currently lacking (Fig. 1A; Petrs-Silva and Linden 2013; Kotterman and Schaffer 2014). Transduction specificity and efficiency is determined and limited by affinity to receptors (Gao et al. 2005; Vandenberghe et al. 2011), as well as the route of administration. Subretinal injection behind the retina results in varying levels of photoreceptor and retinal pigment epithelium (RPE) transduction, depending on the AAV serotype used, with poor transduction in the inner retinal cells (Lipinski et al. 2013; Trapani et al. 2014). Subretinal injections are invasive, risky procedures that often require vitrectomies, and they are incapable of transducing the entire retinal surface due to localized delivery and limited lateral spread (Khabou et al. 2018b; Boye et al. 2020). Intravitreal injections, on the other hand, result in better transduction of cells within the inner retina and cover a larger surface area, but they also result in limited transduction in photoreceptors and RPE cells due to structural barriers that prevent diffusion, including the ILM (Fig. 1B; Dalkara et al. 2009; Trapani et al. 2014). Suprachoroidal injections, a recently developed approach involving injection with microneedles into a potential space between the choroid and the sclera, have been investigated as a potentially less invasive approach to delivering vectors to the outer retina (Yiu et al. 2020). However, suprachoroidal injections have limited ability to target the central retina and macula of large animals and have been linked to adverse immune reactions.

Penetrating the Inner Limiting Membrane

The adult retina is insulated from the vitreous body by the ILM, which histologically defines the border between the retina and the vitreous humor. The ILM is essential for normal eye development, but it is dispensable in adults, and ILM removal is considered beneficial for patients undergoing macular hole surgery. The ILM contains a thick layer of glycans in the primate retina, including heparan sulfate (HS), which is known to be essential for the binding of AAV2 and 3 to the ILM. Binding to proteoglycans prevents AAV dilution in the vitreous volume, but the benefit of binding to this proteoglycan is also a hindrance, as it may act as a sink, impeding AAV entry into the retina (Dalkara et al. 2009). This is evident from retinal gene delivery experiments in primate eyes that show high transduction rates in areas of the retina where the ILM is thinnest or injections under the ILM (Gamlin et al. 2019).

Several methods of enzymatic or surgical disruption of the ILM have been tested as a method to enhance AAV access to the retina from the vitreous (Dalkara et al. 2009; Cehajic-Kapetanovic et al. 2010); however, these methods have intrinsic risks that restrict their development for clinical use. Others have attempted injections under the ILM as a means to provide better access to the retina using AAVs (Gamlin et al. 2019). Of note, the ILM is disrupted as part of the natural history of diseases involving photoreceptor degeneration, potentially offering the possibility to overcome this physical barrier more easily in later stages of retinal degenerative diseases (Kolstad et al. 2010; Vacca et al. 2014). Capsid engineering has also been used to create variants with altered ILM interaction, leading to improved transduction (Dalkara et al. 2013; Byrne et al. 2020; Öztürk et al. 2021).

Small Packaging Capacity

An important challenge in AAV-mediated retinal gene therapy is the delivery of large genes that exceed the carrying capacity of AAV (Fig. 1C). Some strategies to overcome this obstacle include modifying the genetic payload, including gene truncation, and in vivo genome editing.

The use of dual vectors has also emerged as a promising approach to overcome the limited carrying capacity of AAVs. By dividing the open reading frame into multiple vectors, the transgene can be split into halves; each half is packaged independently in single vectors that can reassemble once the dual AAV vectors coinfect the same cell. A range of strategies have been used to reassemble and form the full-length expression cassette, including homologous recombination and trans-splicing. Studies in the retina have demonstrated examples of each of these approaches in vitro and in vivo; however, low efficiency and the production of truncated protein products remain important concerns for clinical use (Colella et al. 2014; Dyka et al. 2014; Trapani et al. 2014, 2015). To ensure efficient protein production, a cell should be infected with equal ratios of each half of the divided message. Random infection rates cause the formation of truncated protein products, and while the inclusion of microRNA (miR) target sites in the gene transcript has previously been shown to effectively reduce the production of truncated proteins (Karali et al. 2011), a decrease in the efficiency of protein expression remains a limitation of the approach.

Neutralizing Antibodies/Immune Response

AAV immunogenicity is a complex problem, and it is dependent on several factors including serotype, transgene expression, organ of interest, and dosage (Fig. 1D). The route of vector administration to the retina also influences immune response (Kotterman et al. 2015). AAVs administered via subretinal injection are less exposed to the immune system, and subretinal injections in one eye can be followed by additional injection in the other eye (Li et al. 2008b; Bennett et al. 2016). Subretinal injections have, however, been associated with retinal thinning when performed under the fovea, and this approach may not be appropriate in fragile, damaged retinas (Bennett et al. 2016). Intravitreal injections are less invasive, and have been used to deliver AAV pan-retinally (Ali et al. 1998). However, intravitreal injections are associated with increased potential for immune response due to access to the systemic circulation and transduction of cells in the anterior segment of the eye. Intravitreal delivery is also made more difficult through dilution that occurs in the vitreous, and the diffusion distance from the ILM to outer nuclear layer (ONL) that must be overcome to achieve an effective therapeutic effect, thus requiring higher viral doses and increasing the chance of triggering an immune response. Similarly, suprachoroidal injections are also associated with significant immune response and loss of transgene expression in nonhuman primates (NHPs), likely as a result of delivery of AAV to the circulatory system (Yiu et al. 2020).

The understanding of the factors contributing to AAV-induced inflammation is currently incomplete (Ail et al. 2022). These factors include the preexistence of serum antibodies against AAVs, and the generation of new antibodies in response to the AAV capsid or transgene after intraocular injections. Binding antibodies (BABs) and a subset of BABs called NABs block AAV transduction locally and systemically.

A dose-dependent increase in BABs and NABs has been documented in ocular gene therapies, across serotypes and modes of injection. Recent work also suggests a correlation between serum BAB levels with clinical grading of inflammation, especially at high doses. Notably, immune reactions to AAV are strongly dependent on capsid dosage; dose sparing through the use of more efficient engineered capsids and strong cell-type-specific promoters can be sufficient to avoid inflammation (Khabou et al. 2018a; Ail et al. 2022). Moreover, when well-tolerated doses are used, the route of administration does not seem to strongly influence the rise of binding and NABs to AAVs (Ail et al. 2022).

Although immunosuppression can partially mitigate cellular and humoral responses induced by vector administration in naive patients, preexisting NABs continue to present challenges for efficient gene delivery, particularly from intravitreal and suprachoroidal routes. Additional approaches to circumventing NABs include shielding AAV particles from antibodies by covalent attachment of polymers to the viral capsid, encapsulation of vectors inside biomaterials, or by engineered capsid variants that evade recognition by anti-AAV antibodies present in the human population (Bartel et al. 2011).

APPROACHES TO AAV ENGINEERING

In recent decades, a spectrum of approaches has been pursued to develop AAVs with improved properties that enable more effective therapies (Fig. 2). Biomining has been used to search for naturally occurring serotypes with better tropism for human tissue types. Rational mutagenesis of naturally occurring serotypes has sought to build on our increasing knowledge of AAV biology to create efficient vectors. High-throughput screening approaches, including screening of ancestral sequences, directed evolution, and single-cell RNA-seq AAV engineering (scAAVengr) screening have been used to identify promising capsids from pools of mutant capsids. Furthermore, in silico design, artificial intelligence (AI) and ML have also been applied to AAVs, accelerating the speed of capsid discovery for gene therapy. As a result of these efforts, newly engineered capsids, including rationally designed capsids and capsids designed through directed evolution are currently in use in ongoing clinical trials.

Figure 2.

Approaches to adeno-associated virus (AAV) engineering. (A) Biomining. (B) Rational mutagenesis. (C) Reconstruction and screening of ancestral AAVs. (D) Directed evolution. (E) Single-cell RNA-seq AAV engineering (scAAVengr). (F) In silico design, artificial intelligence (AI), and machine learning. (PCR) Polymerase chain reaction. (Figure was created with BioRender.com.) (See facing page for legend.)

Biomining

Soon after its discovery, AAV was considered as a promising viral vector for gene delivery, and methods to overcome obstacles to efficient transduction were begun. Biomining, one early approach to AAV development (Fig. 2A), involves the identification of naturally occurring serotypes with potentially better characteristics as gene delivery vectors. AAV1 to AAV6 were the first AAV serotypes identified (Gao et al. 2005). All AAV serotypes but AAV5 were first found as contaminants of adenovirus preparations, while AAV5 was found in a human condylomatous wart. Biomining in simian viral preparations resulted in 10 AAV variants with 96% homology to AAV1 or AAV6. However, only two of these variants showed different tropisms for human cancer cell lines when compared to either AAV1 or 6 (Schmidt et al. 2006).

In the absence of adenovirus or herpesvirus, AAV establishes latent infections either through integrating into chromosome 19 or by forming stable episomes. It was hypothesized that additional undescribed endogenous AAV variants could thus be recovered from human and NHP tissues (Gao et al. 2005; Schmidt et al. 2006). Polymerase chain reaction (PCR) was used to amplify the cap gene, which was then sequenced (Gao et al. 2005). Over 100 novel AAV variants, separated in different clades, have since been described (Gao et al. 2004; Mietzsch et al. 2021). Of these, 55 were from human cells and 55 were from NHPs. Twenty-five AAV variants were initially screened in vivo in mouse models. Of note, AAV8, AAV9, and AAVrh.10 were identified as interesting variants for gene therapy, with higher transduction efficiencies in murine hepatocytes or lung when compared to AAV2. AAV8 is currently employed in clinical trials for several retinal degenerative diseases, as it has improved infectivity of photoreceptors when delivered subretinally compared to AAV2 (Vandenberghe et al. 2011). AAV9 and AAVrh.10 are the serotypes of choice in several trials targeting the central nervous system.

Rational Mutagenesis

Crystal structures of AAV capsids have been determined for many AAV serotypes in the past decades (Mietzsch et al. 2019), leading to deeper understanding of the structural basis for receptor binding, tropism, and transduction efficiency. This information has in turn influenced strategies for rational design and modification of the capsid (Fig. 2B). AAV2 is the best studied serotype due to its early discovery and successful implementation in gene delivery, making it more amenable to rational design.

The amino acid residues involved in HS binding (R484, R487, K532, R585, and R588) have been extensively modified (Boye et al. 2016) as have key residues responsible for ubiquitination upon cell entry (Zhong et al. 2008; Petrs-Silva et al. 2009, 2011). Rational design of the AAV capsid has been explored for the purpose of avoiding proteasomal degradation and enhancing transduction efficiencies. Substitution of surface-exposed tyrosine, threonine, and serine residues, which are phosphorylated as part of the ubiquitin-dependent proteasomal degradation process, prevents this process. Surface-exposed tyrosine-to-phenylalanine (Y-F), threonine-to-valine (T-V), or serine-to-valine (S-V) mutations (or a combination thereof) increase transduction efficiency in several tissues and organs, including the retina (Kay et al. 2013). AAV1 was given the ability to bind HS through the addition of a single E531K mutation, resulting in a vector with a similar tropism as AAV6, with the ability to infect the retina from intravitreal injections in mice (Wu et al. 2006; Woodard et al. 2016). However, similar attempts to graft AAV2 HS-binding residues onto other vectors such as AAV5 and AAV8 did not impart the ability to transduce the retina from the vitreous. This result demonstrates the complexity of the factors influencing AAV tropism in a complex structure such as the retina, and the difficulty in attempting to engineer AAV through a rational approach.

Ancestral Screening

Improvements in phylogenetic analysis, DNA sequencing, and synthesis have permitted the reconstruction of ancestral proteins, including AAV capsids. Using statistical and ML methods, applied to an analysis of extant AAVs, groups have sought to infer and reconstruct theoretical ancestral capsid sequences. Instead of analyzing a single estimated ancestral sequence, two groups have built libraries of predicted sequences that would represent putative ancestral AAVs at specific phylogenetic nodes (Fig. 2C; Santiago-Ortiz et al. 2015; Zinn et al. 2015).

One group analyzed the node that encompassed the ancestor of AAV1, AAV6, and AAV7 (Santiago-Ortiz et al. 2015), to determine whether ancestral AAVs had broad or specific tropisms, and whether ancestral AAVs might be more stable than their ancestors, making them better suited as the starting material for directed evolution studies. AAV1 and AAV6 have therapeutic relevance, while AAV7 is known for its relative resistance to NABs. Ancestral variants from the screen were stable, and showed promiscuous rather than specific infectivity profiles. The authors therefore suggested that a hypothesized increase in mutational tolerance and evolvability of this library might be harnessed in directed evolution studies.

Meanwhile, another group (Zinn et al. 2015) focused on the node representing a hypothetical ancestor of AAVs 1–3 and 6–9, called Anc80, to determine whether ancestral AAVs might be more stable, more infectious, or whether the seroprevalence for NABs against this ancestral variant might be lower in the modern human population. In silico phylogenetic and statistical modeling was used to predict putative ancestral sequences of the AAV capsid protein. A library of predicted AAV capsid sequences was then synthesized and screened in HEK cells. From this screen, the Anc80L65 clone was chosen for further analysis, and was shown to outperform other vectors in photoreceptor cells. Further lack of toxicity and desirable immune reactivity showed that a combination of in silico analysis and library selection can be a useful method for assessing AAV vector properties and discovering therapeutically relevant variants.

High-Throughput Approaches to Engineering AAVs

A range of high-throughput strategies have been exploited to create new capsids over the past decade. Directed evolution (Fig. 2D) is a method involving the introduction of mutations in the cap gene to yield highly diverse AAV libraries. Selective pressure is then applied to enrich for novel variants with enhanced properties. Techniques for the introduction of mutations include the insertion of random peptide sequences into a defined location on the capsid, incorporation of peptide sequences into random locations in the cap gene, error-prone PCR and site-directed mutagenesis, and gene shuffling, which generates chimeric AAV capsids by fragmentation of the cap gene (Müller et al. 2003; Maheshri et al. 2006; Perabo et al. 2006a,b; Grimm et al. 2008). The creation of high-quality and highly diverse libraries, which thoroughly explore possible sequence space, are key to the identification of novel AAV variants with new abilities.

scAAVengr

Accurate, quantitative comparisons of the efficiency and tropism of newly engineered vectors are challenging. Large numbers of animals are typically required to characterize the performance of variants, and variation between animals and injections lead to inaccurate comparisons. These challenges are compounded in valuable large animal models such as primates, in which there is large variability between animals. Recently, an scAAVengr pipeline was developed for rapid, quantitative in vivo comparison of transgene expression in retinal cells from newly engineered AAV capsid variants across all different cell types in a tissue in parallel, and in the same animals (Fig. 2E; Öztürk et al. 2021; Xi et al. 2022). scAAVengr uses single-cell RNA-seq to identify single cells by marker gene expression, and simultaneously quantify the ability of AAV variants to drive gene expression in those cells at the single-cell level based on quantification of transgenes. The scAAVengr single-cell RNA-seq pipeline allows for highly quantitative, direct evaluation of the multiple lead candidates across all cell types in a tissue, which is critical for understanding the clinical potential of a vector.

In Silico Design, AI, and Machine Learning

Directed evolution is a high-throughput protein engineering method applicable to AAV capsids; however, the overall success rate in these screens can be low because of the complex relationship between capsid structure and the AAV transduction pathway and the many aspects of AAV structure and function that must be simultaneously optimized. High-throughput screening methodologies can be aided by in silico design and ML (Fig. 2F). ML refers to algorithmic approaches that enable automatic learning. ML allows for rules for vector design to be established directly from input data. Training data can be acquired through deep sequencing of viral libraries.

Recently, ML was used to guide capsid engineering and to interrogate AAV biology (Ogden et al. 2019). The effects of mutations across all 735 amino acid positions, including all synonymous codons for each amino acid, were analyzed to enable detection of noncoding elements. Single-codon insertions and deletions, as well as control wild-type AAV2 sequences and stop codon substitutions were also included. Based on the analysis of resulting variants, the authors predicted that an additive model built from this data would approximate the fitness of nearby variants with multiple mutations. It was hypothesized that this approach would thereby enable the design of functional variants with higher throughput than rational design and improved efficiency compared to random mutagenesis. 1271 variants were designed and tested by measuring liver biodistribution in mice. The set of AAVs designed through the ML model included a higher fraction of mutants targeted to the liver, compared to a control set of vectors with random mutations, demonstrating that machine-guided design is a valuable approach to create higher percentages of viable mutants. These methods may enable rapid and systematic optimization of AAV capsids that might have been undetectable or lost using other less informed screening efforts.

Next Generation AAVs in the Clinic

Next-generation AAVs with enhanced properties have the potential to improve clinical outcomes. A recent meta-analysis of AAV usage in clinical settings (Au et al. 2022) found that six different capsid types were used in clinical trials for eye disorders, with the majority of clinical trials using natural AAV2. The other two natural AAVs used in clinical trials are AAV5 and AAV8. Recent clinical trials are using engineered serotypes including AAV2TYF (Zhong et al. 2008), 7m8 (Dalkara et al. 2013), and 4D-R100 (Kotterman et al. 2021). AAV2TYF, a tyrosine-to-phenylalanine mutant, was designed to avoid proteasome degradation (Zhong et al. 2008). This vector has been used in three different clinical trials (NCT03316560, NCT02599922, NCT02416622). 7m8, which has been used in four clinical trials (NCT03748784, NCT04418427, NCT04645212, NCT03326336), was designed through directed evolution, and its capsid carries a 10–amino acid insertion that enables greater transduction via intravitreal injection. 4D-R100 was also optimized through directed evolution, and it is being used in two clinical trials (NCT04483440, NCT04517149).

Promoters and cis-Regulatory Elements

While generating efficient AAV capsid variants is key to the optimization of gene delivery, promoters and other cis-regulatory elements that flank the transgene also define the therapeutic outcome by regulating gene expression. Promoters are sequences located at the 5′ end of the transgene and drive gene expression. Often simply referred to as “promoters,” they are frequently a combination of transcriptional start site, promoter, enhancer, and regulatory sequences (Gray et al. 2011). RNA polymerase attaches itself to a promoter, while enhancers are sequences upstream or downstream of promoters that regulate gene expression. As the packaging capacity of AAV is limited, AAV-compatible promoters condense these functions in short fragments, often between ∼200 and 2000 bp long.

Promoters can be ubiquitous or cell-specific, and define which infected cells will express a transgene. Ubiquitous promoters, such as the cytomegalovirus (CMV) early enhancer/promoter or the synthetic CAG promoter, can drive strong gene expression in all cell types. On the other hand, cell-specific promoters can be used to target one or more cells (Khani et al. 2007; Planul and Dalkara 2017). The specificity of promoters is essential to tailor targeted expression. Promoter specificity is influenced by viral dose, injection route, the state of the retina, the species in which the promoter is used, and capsid tropism (Khabou et al. 2018a).

The search for efficient, cell-specific promoters has led to extensive efforts such as the Pleiades Promoter Project (Portales-Casamar et al. 2010), the retina-specific synthetic promoter search by Botond Roska's team (Jüttner et al. 2019), and a recently described method that identified Müller glia and amacrine cell-specific promoters (Lin et al. 2022). These projects have developed a wide range of synthetic promoters with varying abilities to target specific cell types. Furthermore, several of these studies have shown that promoters can have unexpected off-target transgene expression in unwanted cell types, which may be a source of toxicity. Promoters with excellent cell specificity may not drive high transgene expression levels, and so efficiency must also be taken into consideration.

An array of promoters restrict expression of transgenes to specific cell types in the retina. For example, the RPE65 promoter (Meur et al. 2007), the VMD2 promoter (Alexander and Hauswirth 2008; Conlon et al. 2013; Guziewicz et al. 2013), and the CD365 promoter (Sutanto et al. 2005) drive expression in RPE cells following subretinal injection in rodents. The hRK or GRK1 promoter has been used to drive expression in photoreceptors from subretinal injections in mice and pigs (Khani et al. 2007; Boye et al. 2010; Zou et al. 2011; Manfredi et al. 2013). pR2.1 and pR1.7 promoters have been used to drive expression in cones in mice, rats, ferrets, guinea pigs, and macaques (Li et al. 2008a; Ye et al. 2016). And Grm6, 4xGRm6, In4s-In3-200En-mGluR500P, and Ple22 promoters drive expression in bipolar cells in mice and marmosets via subretinal and intravitreal injection (Doroudchi et al. 2011; Cronin et al. 2014; de Leeuw et al. 2014; Gaub et al. 2014; Macé et al. 2015; Scalabrino et al. 2015; Lu et al. 2016).

The choice of cis-regulatory elements can also affect toxicity. A previous study in mice has shown that ubiquitous promoters such as CMV and CAG, as well as one RPE-specific promoter (Best1), were correlated to ocular toxicity (Xiong et al. 2019). These results indicate that promoters should be screened for toxic effects before being employed in therapeutic studies.

CONCLUDING REMARKS

Over the past several decades, significant progress has been made in the design of next-generation AAVs with enhanced abilities to deliver therapeutic genes to the retina. AAVs are the gold standard for gene delivery, although efficient and specific gene delivery remains a key bottleneck in the development of gene therapies. Significant shortcomings remain, including tropism, limited carrying capacity, and immunogenicity. A range of approaches have been implemented and have shown promise in engineering new variants with desirable properties, including rational design, high-throughput screening, and ML. Together, these approaches may yield a toolbox of AAV vectors capable of targeting each of the cell types affected by the range of retinal diseases that exist.