*Article* **Adeno-Associated Viral Vectors as a Tool for Large Gene Delivery to the Retina**

#### **Ivana Trapani 1,2**


Received: 1 March 2019; Accepted: 5 April 2019; Published: 9 April 2019

**Abstract:** Gene therapy using adeno-associated viral (AAV) vectors currently represents the most promising approach for the treatment of many inherited retinal diseases (IRDs), given AAV's ability to efficiently deliver therapeutic genes to both photoreceptors and retinal pigment epithelium, and their excellent safety and efficacy profiles in humans. However, one of the main obstacles to widespread AAV application is their limited packaging capacity, which precludes their use from the treatment of IRDs which are caused by mutations in genes whose coding sequence exceeds 5 kb. Therefore, in recent years, considerable effort has been made to identify strategies to increase the transfer capacity of AAV vectors. This review will discuss these new developed strategies, highlighting the advancements as well as the limitations that the field has still to overcome to finally expand the applicability of AAV vectors to IRDs due to mutations in large genes.

**Keywords:** AAV; retina; gene therapy; dual AAV

#### **1. Introduction**

The eye is an ideal target for gene therapy thanks to its small and enclosed structure, relative immune privilege and easy accessibility [1,2]. This has boosted attempts at developing gene therapy approaches for the treatment of a large number of inherited retinal diseases (IRDs) over the recent decades [3,4]. Confirmation of the advancements in the retinal gene therapy field came in the last two years with the approval of the first gene therapy product for an IRD, Luxturna [5]—an adeno-associated viral (AAV) vector-based therapy for a form of Leber Congenital Amaurosis [6]—in the US, first, and then in Europe. The recombinant AAV vector on which Luxturna is based is the most widely used vector for retinal gene delivery. AAV are small (25 nm), nonenveloped, icosahedral viruses belonging to the Parvoviridae family [7]. They package a linear single-stranded DNA genome of ~4.7 kb, flanked by two 145 bp long palindromic inverted terminal repeats (ITRs) [7]. These ITRs form hairpin-loop secondary structures at the strand termini and are the only viral sequences that are retained in cis in the recombinant AAV vector genome [7]. Recombinant vectors based on AAV have fast become popular in the gene therapy field because of their excellent safety profile and low immunogenicity which allows for long-term expression of the therapeutic gene, at least in post-mitotic tissues, so that most experimental therapy studies require only a single vector administration. Additionally, dozens of different AAV variants have been identified thus far, each of them with unique transduction characteristics. This allows the user to select the most appropriate AAV serotype to transduce the retinal cell layer of interest. Indeed, following subretinal delivery, virtually all the AAV serotypes tested efficiently transduced the retinal pigment epithelium (RPE), while the levels of transduction of photoreceptors, which are the main therapeutic target cells in most IRDs, varied significantly among different serotypes [4,8]. AAV5, AAV7, AAV8 and AAV9 serotypes have all been demonstrated to efficiently transduce photoreceptors [4,8]. Additional serotypes with increased retinal transduction

abilities have also been identified through either rational design or directed evolution [4,8]. This is one of the most attractive features of AAV vectors for retinal gene therapy, since alternative, both non-viral and viral, vectors tested thus far have shown more limited transduction abilities of adult photoreceptors [2,9]. For all the above described reasons, AAV have been used in many successful preclinical and clinical studies [3,4]. Clinical trial data collected over a decade have confirmed the overwhelming safety of AAV vectors delivered intraocularly and shown many instances of efficacy in treating previously incurable IRDs.

However, one of the main limitations to a broader application of AAV vectors for retinal gene therapy is their packaging capacity, which is restricted to approximately 5 kb of DNA [10]. This vector capacity is a critical issue, given the fact that approximately 6% of all human proteins have a coding sequence (CDS) that exceeds 4 kb [11] and that, in addition to the CDS of the therapeutic gene and the ITRs, a gene therapy vector needs to include, as a minimum, a promoter and a polyadenylation signal (polyA). Thus, the treatment of disorders caused by mutations in genes over 4 kb in size, including those causative of common IRDs, is currently not achievable using standard AAV vector-mediated approaches. The development of strategies to overcome AAV packaging limitation has therefore become a key area of research within the gene therapy field.

#### **2. Strategies for Large Gene Delivery**

Two types of strategies have been developed for large gene delivery via AAV: one is based on the "forced" packaging of oversized genomes (i.e., larger than 5 kb) in a single AAV vector (oversized AAV vectors); the other relies on the delivery of portions of large transgenes in two AAV vectors, which recombine through various mechanisms in the target cell, leading to the reconstitution of the full-length gene (dual AAV vectors) (Table 1).


**Table 1.** Adeno-associated viral (AAV) vector-based strategies for large gene delivery.

#### *2.1. Oversized AAV Vectors*

Several research groups have tried to encapsidate large genes in a single AAV vector [12–14]. These "oversized" AAV vectors have been found to successfully express full-length proteins in vitro and in the retina of IRD models to levels which led to significant and stable improvement of the phenotype [12,15]. However, the genome contained in oversized AAV vectors was found to be not a pure population of intact large-sized genomes but rather a mixture of genomes highly heterogeneous in size [14,16–19]. Thus, it was proposed that full-length protein expression from oversized AAV vectors was achieved, following infection, through the re-assembly of truncated genomes in the target cell nucleus [14,16–19]. The efficiency of the transduction of oversized AAV vectors in the retina in comparison to alternative platforms for large gene delivery (i.e., dual AAV vectors, discussed below) has been assessed in various studies and found to be variable. Whereas some studies found considerably high levels of transgene expression from oversized AAV vectors [14,15], others showed

efficient large protein reconstitution only upon dual AAV vector delivery [20,21]. Both the design and purification process of oversized AAV vectors were hypothesized to be critical for the success of the strategy, as the use of transgenes slightly above 5 kb can give rise to genomes with longer overlaps compared to the use of transgenes largely exceeding AAV cargo capacity, and this can drive more efficient re-assembly of oversized AAV vectors. Along this line, it was shown that the fractionation of oversized AAV vector preparations can be explored to promote selection of the genomes with the highest transduction properties in the final viral preparation [14]. However, despite the optimization and ability of this strategy to reconstitute large genes expression in vivo, consistently shown in various studies, the heterogenous nature of oversized AAV genomes poses major safety concerns, limiting their further application in clinical settings.

#### *2.2. Dual AAV Vectors*

An alternative strategy for AAV-mediated large gene delivery is the generation of dual AAV vectors. In this strategy, large transgenes are split into two separate AAV vectors that, upon co-infection of the same cell, reconstitute the expression of a full-length gene via intermolecular recombination between the two AAV vector genomes. This ideally doubles AAV cargo capacity, allowing delivery of transgenes up to about 9 kb. Various dual AAV vector strategies have been developed (referred to as trans-splicing [22], overlapping [23] and hybrid [24] dual AAV vector strategies), which differ in the mechanism they use to reconstitute the transgene.

#### 2.2.1. Trans-Splicing Dual AAV Vectors

The trans-splicing approach relies on the natural ability of AAV ITRs to concatemerize in order to reconstitute full-length genomes [22,25]. In this approach, the two vectors carry two separate halves of the transgene, without regions of sequence overlap; the 5'-half vector has a splice donor (SD) signal at the 3' end of the AAV genome, while the 3'-half vector carries a splice acceptor (SA) signal at the 5' end of the AAV genome (Figure 1).

**Figure 1.** Schematic representation of the trans-splicing dual AAV approach for large gene reconstitution. The first vector includes the promoter, the 5'-half of the coding sequence (CDS) and the splicing donor (SD) signal; the second vector includes the splicing acceptor (SA) signal, the 3'-half of the CDS and the polyadenylation signal (PolyA). Concatemerization of the two vectors, involving the right-hand inverted terminal repeat (ITR) of the first vector and the left-hand ITR of the second vector, reconstitutes the full-length gene. After transcription, splicing leads to the removal of the ITR structure at the junction point, with restoration of the full-length, mature RNA of the transgene.

*Genes* **2019**, *10*, 287

This allows splicing of the concatemerized ITR structure that forms in the middle of the therapeutic CDS following tail-to-head concatemerization of the two AAV genomes to obtain a single large mRNA molecule. This approach was first tested about 20 years ago, and historically represents the first developed approach for AAV-mediated large gene delivery. Since then, many studies have shown the efficacy of this strategy to reconstitute large genes. The major limitation of this platform, however, is that concatemerization can occur between any of the ITR of the two vectors. This may lead to the formation of both forms of circular monomers of each AAV, as well as two-vector linear concatemers in a number of orientations of which only one (i.e., tail-to-head concatemer) is productive to restore full-length gene expression [26]. Attempts at favoring the formation of concatemers in the correct orientation have been made (as discussed in the "Limitations of dual AAV vectors" paragraph). An additional limiting step of trans-splicing vectors is splicing across the ITR junction, the efficiency of which is dependent on both selection of the optimal exon–exon junction for splitting the large therapeutic gene [27] as well as the efficiency of splicing across the ITR structure [28]. To overcome the first issue, synthetic SD and SA signals have been developed, which mediate high rates of splicing independently of the gene that needs to be delivered [29]. Yet, since the sequence surrounding the splicing signals has an impact on splicing efficiency, careful selection of the splitting point is required.

#### 2.2.2. Overlapping Dual AAV Vectors

In the overlapping approach, the transgene is split into two halves sharing homologous overlapping sequences, such that the reconstitution of the large gene expression cassette relies on homologous recombination [23] (Figure 2).

**Figure 2.** Schematic representation of the overlapping dual AAV approach for large gene reconstitution. The first vector includes the promoter and the 5'-half of the coding sequence (CDS) and the second vector includes the 3'-half of the CDS and the polyadenylation signal (PolyA). A portion of the sequence of the large transgene is repeated in both vectors (at the 3' end of the CDS of the first vector and at the 5' end of the CDS of the second vector). Thus, the full-length transgene expression cassette is reconstituted through homologous recombination of the overlapping regions in the two vectors. ITR: inverted terminal repeat.

As it has been designed, the overlapping approach is the simplest in design and requires less foreign or artificial DNA elements when compared to the other approaches. However, as the success of this strategy is critically dependent upon the ability of the overlapping region to mediate efficient homologous recombination, much work is needed to determine the optimal CDS overlapping region to be used for each transgene. Furthermore, data obtained so far have also highlighted that the success of this strategy is dependent on the retinal cell type being targeted, since the efficiency of the repair mechanism on which overlapping dual AAV vectors rely for large gene reconstitution is tissue dependent, as discussed below.

#### 2.2.3. Hybrid Dual AAV Vectors

To overcome the main limitations of the previously described platforms (i.e., the lack of preference for directional tail-to-head concatemerization of the trans-splicing approach and the need for optimization of the CDS overlap for each transgene in the overlapping approach), a third transgene-independent dual AAV approach was developed: the hybrid dual AAV vectors. This approach is a combination of the trans-splicing and overlapping approaches, as it is based on the addition of a highly recombinogenic exogenous sequence to the trans-splicing vectors in order to increase recombination efficiency [24]. This recombinogenic sequence is placed downstream of the SD signal in the 5'-half vector and upstream of the SA signal in the 3'-half vector, so to be spliced out from the mRNA after recombination and transcription (Figure 3).

**Figure 3.** Schematic representation of the hybrid dual AAV approach for large gene reconstitution. The first vector includes the promoter, the 5'-half of the coding sequence (CDS), the splicing donor (SD) signal and the highly recombinogenic exogenous sequence (HR); the second vector includes the highly recombinogenic exogenous sequence, the splicing acceptor (SA) signal, the 3'-half of the CDS and the polyadenylation signal (PolyA). Joining of the two AAV vector genomes to reconstitute the full-length gene can occur through either: 1. concatemerization of the two vectors through the inverted terminal repeats (ITR), as for trans-splicing dual AAV vectors; or 2. homologous recombination mediated by the region of homology included in both vectors. In both cases, after transcription, splicing leads to the removal of the junction point, with restoration of the full-length, mature RNA of the transgene.

The hybrid dual AAV approach is potentially more effective than the other dual AAV vector approaches, since full-length gene reconstitution can occur through both homologous recombination mediated by the highly recombinogenic exogenous sequence as well as concatemerization through the ITRs [24]. The recombinogenic sequences used thus far to induce the recombination between hybrid dual AAV vectors have been derived from regions of either the alkaline phosphatase gene (AP) [24,30] or the F1 phage genome (AK) [21]. The inclusion of the exogenous sequence allows the promotion of high levels of homologous recombination between the two vector genomes, independently of the transgene to be delivered. However, similarly to the trans-splicing approach, the sequences surrounding the splicing signals still have an impact on splicing efficiency. Thus, careful selection of the splitting point is recommended to achieve maximal efficacy of large gene reconstitution.

#### **3. The Choice of the Best Platform for Large Gene Delivery to the Retina**

The efficacy of both oversized and dual AAV vectors in the retina has been evaluated in a number of studies using different reporter and therapeutic genes, such as *ABCA4* and *MYO7A* mutated in Stargardt disease (STGD1) [31] and Usher syndrome type 1B (USH1B) [32], respectively. However, literature describing these platforms is often conflicting. Initial studies in the retina reported a better performance of oversized AAV vectors compared to dual AAV strategies [14,15]. These results, however, might be due to both design and purification processes, which favor the generation of oversized vectors with high transduction properties [14], as well as to the less than optimal design of the dual AAV platform that was used as a comparison. One study, as an example, relied on the use of overlapping dual AAV vectors with a large region of overlap (1365 bases) that had not been optimized and, therefore, might potentially have a low efficiency of recombination [15]. Reconstitution from overlapping dual AAV vectors has also been found to occur at variable levels in different studies [15,20,21,26,33]. The most critical aspect of an overlapping dual AAV vector strategy is the event of recombination between the two halves of the transgene. This is influenced by both the sequence of the transgene and the cell type that is targeted, since different cell types could possibly deploy different DNA repair mechanisms. Some studies have found that long regions of overlap may lead to higher levels of transgene reconstitution [26]. However, it has recently been shown that optimization of the overlapping region is a prerequisite to achieve sustained levels of transgene expression in photoreceptors, since the efficiency of reconstitution is not directly proportional to the length of the regions of overlap [33]. It has been suggested that if the regions are too short, they might not be able to efficiently mediate interactions with the opposing viral genome, whereas longer regions of overlap may be less available for such interactions due to secondary structure formation. In line with this hypothesis, a screening of overlapping regions ranging from 23 to 1173 bp identified an overlap of 207–505 bp as the best performing for overlapping dual AAV-mediated reconstitution of *ABCA4* at therapeutic levels [33]. Thus, optimization of the overlapping region is essential to achieve sustained levels of transgene expression in photoreceptors. The targeted tissue also plays an important role in the success of the overlapping dual AAV approach since homologous recombination is typically associated with dividing cells, while low levels of homologous recombination are found in post-mitotic cells as neurons [34]. Along this line, studies have reported inefficient transduction of photoreceptors mediated by overlapping dual AAV vectors [15,21,26], whilst more efficient reconstitution was found in the RPE [21]. Other groups, however, have found efficient transduction of photoreceptors using overlapping dual AAV vectors [20,33], highlighting that the identification of highly recombinogenic regions of overlap in the transgene overcomes the limitations related to the inability of specific cell types to mediate efficient homologous recombination [33].

More consensus on the efficacy of trans-splicing and hybrid dual AAV vectors can be found in literature. A number of studies have indeed shown the ability of these strategies to reconstitute large transgenes in the retina [20,21,26,35,36] at levels which were higher compared to the other dual AAV strategies tested side by side [20,21,26], and which resulted in improvement of the retinal phenotype of animal models of IRDs [21,37]. This is possibly due to a more limited requirement of the optimization of these platforms compared to the others, since joining of the two halves of the transgene, with a discrete nature, occurs through the ITRs and/or a region of overlap known to be highly recombinogenic. Notably, the success obtained in the delivery of the large *MYO7A* gene to the retina [21] has led to the planning of a Phase I/II clinical trial, which will test the safety and efficacy of the hybrid dual AAV platform developed in the retina of USH1B patients (https://cordis.europa.eu/project/rcn/212674\_it.html). Importantly, the results of this trial will definitively shed light on the efficiency of dual AAV vectors-mediated large gene delivery in the human retina.

Prompted by the success shown by dual AAV strategies, researchers have attempted at further expanding AAV cargo capacity in the retina up to 14 kb by adding a third vector to the dual system, generating triple AAV vectors [38]. This was found to be achievable, but at the expense of efficiency. Indeed, the levels of transduction achieved in the retina of a mouse model of Alstrom syndrome with

triple AAV vectors have led to only a modest and transient improvement of the phenotype [38]. On the other hand, the levels of transduction mediated by triple AAV vectors in the large pig retina were found to be significantly higher than in the mouse retina, as also observed with dual AAV vectors [38]. These results bode well for further optimization of this platform.

#### **4. Limitations of Dual AAV Vectors**

Currently, all the dual AAV vector approaches have shown similar issues: variable success and expression of unwanted truncated products from single half-vectors. For all dual AAV platforms to be successful, a cell must necessarily be co-infected by at least one AAV vector including the 5'- and one including the 3'-half of the expression cassette. We and others have shown that co-transduction by two AAV vectors is quite efficient in the small subretinal space [11,21,36], which thus represents a favorable environment for developing dual AAV vector-based gene therapy approaches.

So far, however, all the studies performed have shown that none of the dual AAV approach matches the levels of expression achieved with a single AAV vector [21,26,36]. Various strategies have been explored to increase the efficiency of dual AAV vector-mediated large gene reconstitution.

One option is to increase vector dose and/or use AAV serotypes with higher tropism for the target cells in order to maximize rates of co-infection by both half vectors. A recent study has, however, suggested that an increase in vector dose does not proportionally correlate with increased levels of protein expression in the retina [26]. This suggests that, once efficient co-transduction is achieved, a further increase in vector genome amounts does not provide significant advantages [26]. Attempts at achieving higher levels of transduction by using alternative AAV serotypes have not been found consistently to result in higher transduction levels. Some studies have shown that use of capsid-engineered AAV variants with higher retinal transduction abilities, as tyrosine mutants capsids [39], led to higher levels of transgene expression from overlapping dual AAV vectors compared to naturally occurring AAV serotypes [20,33]. However, delivery of hybrid dual AAV vectors using an in-silico designed, synthetic vector (Anc80L65), which has also been shown to transduce retinal cells with a higher efficiency than AAV8 [40], led to almost identical levels of protein reconstitution compared to dual AAV8 vectors [26].

Another approach explored to increase transduction levels from dual AAV vectors has been maximizing the chances of both trans-splicing and hybrid AAV vectors to generate concatemers in the productive orientation, by forcing concatemerization of the ITRs, through the use of vectors carrying heterologous ITRs (i.e., ITR from different AAV serotypes at the opposite ends of the viral genome) [41]. Indeed, by generating trans-splicing vectors with heterologous ITR from serotypes 2 and 5 it has been shown that it is possible to reduce both the ability of each vector to form circular monomers and to increase directional tail-to-head concatemerization. This resulted in increased levels of transgene reconstitution compared to the use of vectors with homologous ITRs [41,42]. However, we have later shown that inclusion of heterologous ITRs in hybrid dual AAV vectors does not provide a significant advantage in full-length transgene reconstitution over the use of vectors with homologous ITRs [37]. This is consistent with the idea that hybrid dual AAV concatemerization is already partially driven in the correct orientation by the presence of highly recombinogenic regions. An additional strategy which has been used to direct AAV vectors concatemerization in the proper orientation is the use of a single-strand DNA oligonucleotide displaying homology to both of the distinct AAV genomes [43]. Alternatively, strategies that can improve dual AAV vector transduction efficiency by positively modulating AAV transduction steps, as the delivery of kinase inhibitors along with AAV vectors, have also been tested [44].

Another major drawback of dual AAV vectors, observed in some studies, is the production of truncated protein products from each of the single AAV vectors [20,21,33,37]. We and others have shown that, both in vitro and in the retina, truncated proteins from the 5 half vector that contains the promoter sequence and/or from the 3 half vector, due to the low promoter activity of the ITR, are produced. This issue can however be efficiently overcome by the use of the CL1 degron, a C-terminal destabilizing peptide that shares structural similarities with misfolded proteins and is thus recognized by the ubiquitination system [45,46]. Inclusion of this short (16 amino acids in length) degron mediates selective degradation of the truncated product from the 5 half vector [37], without either affecting full-length protein reconstitution or significantly reducing the packaging capacity of the platform. More recently, McClements et al. have shown how the design of dual AAV vectors can also influence production of truncated proteins by the generation of unintended cryptic translation start sites and/or polyA signals [33]. Thus, the design of these platforms requires multiple considerations and adaptation, which may include codon optimizations to remove cryptic genetic signals. Furthermore, given the expression of such unwanted protein products, confirmation of the safety of dual AAV vectors is an important open question. While our preliminary data have shown no evident alterations of retinal morphology and functionality in mouse and pig eyes injected with dual AAV vectors [21,37], formal toxicity studies are required to elucidate this aspect.

#### **5. Alternative Strategies to Allow AAV-Mediated Large Gene Delivery**

Additional strategies to deliver large transgenes via AAV vectors are being actively investigated. Attempts at identifying AAV vectors with expanded cargo capacity, based on either protein libraries and directed evolution [47] or site directed mutagenesis to add positively-charged residues at lumenally exposed sites within the capsid [48], have been described. Alternatively, it has been shown that oversized AAV2 vector genomes can be effectively packaged in the capsid of human Bocavirus 1 (HBoV1) [49,50], an autonomous parvovirus relative of AAV, with a 5.5 kb genome. Testing of these vectors in the retina might lead to the identification of novel suitable vectors for large gene delivery.

The development of different short regulatory elements has also been attempted to reduce the size of the expression cassette and allow delivery of transgenes that exceed the AAV packaging capacity [51–55]. However, this often led to reduced levels of transgene expression. The combination of short synthetic enhancers and promoters was found to be useful for providing increased levels of expression of large transgenes [56]. Other studies have however shown that, despite optimization, some transgenes were more difficult than others to reconstitute from oversized AAV vectors when using short promoters [57].

The use of cDNA encoding for truncated versions of large proteins, which retain their functionality (i.e., a minigene), has also been achieved with some success [58]. However, all these approaches still cannot be easily applied to a large number of genes that exceed the AAV cargo capacity, since extensive optimization and testing would be required for each one of them.

#### **6. Conclusions and Outlook**

The growing number of clinical trials that show good safety and efficacy of the subretinal delivery of AAV vectors are contributing to the establishment of AAV as vectors of choice for retinal gene transfer. Expanding AAV cargo capacity over 5 kb is however a prerequisite to allow this platform to be used as a tool for the efficient delivery of a larger number of therapeutic genes. Recent proof-of-concept studies that used dual and triple AAV vectors to deliver large genes to the retina have shown that it is feasible to transfer genes with a CDS larger than 5 kb. Yet, these studies have highlighted that there is no one-fits-all dual AAV vector system, since dual AAV approaches have shown different relative efficiency in different studies. Clearly, the tissue being targeted, as well as the transgene that needs to be delivered, drastically influences transduction efficiency. Thus, careful design of the platform for each therapeutic application is required to achieve maximal efficacy. The planned clinical trial for USH1B will help defining whether the levels of expression achieved with dual AAV vectors are therapeutically relevant in humans. While the need of manufacturing two or more vectors to treat each disorder might represent a challenge of dual/triple AAV platforms, yet the retina is a favorable tissue for development of these approaches due to the fact that it requires delivery of only a small amount of vector. This reduces the total amount of vectors that needs to be produced.

Retinal transduction with multiple AAV vectors has been shown to reach lower levels compared to a single AAV vector. These levels were not sufficient to result in therapeutic efficacy for some diseases [38]. Consequently, alternative strategies should be explored.

Systems that rely on mechanisms different than those exploited by dual AAV vectors for large gene reconstitution might be investigated, including trans-splicing of pre-mRNAs [59] or intein-mediated protein trans-splicing [60]. Genome editing is also a rapidly expanding field of research, and could represent an interesting option for correction of mutations in genes whose delivery through AAV vectors is precluded by the large CDS size. A number of aspects for this approach however still need to be further explored. First, in the retina, where homologous recombination occurs at low rates, genome editing tools for the precise correction of a mutation will most probably need to exploit alternative repair mechanisms such as non-homologous end joining used for homology-independent targeted integration [61]. The efficiency of such approaches in the retina is still unknown. Secondly, the delivery of genome editing tools in post-mitotic tissues, such as the retina, might not be as safe as delivery in more proliferative tissues, considering the fact that their expression will persist long term after a single subretinal injection.

In conclusion, important steps forward have been made towards the treatment of IRDs due to mutations in large genes, which now seems an achievable goal. The optimization of these and the newly emerging platforms will allow expansion of the number of IRDs that are treatable using AAV-mediated gene therapy.

**Funding:** This work was supported by "Università degli Studi di Napoli Federico II" under the STAR Program.

**Acknowledgments:** We thank Raffaele Castello (Scientific Office, TIGEM, Pozzuoli, Italy) for critical reading of the manuscript.

**Conflicts of Interest:** The author is the co-inventor on patent applications on the dual AAV vector platform.

#### **References**


© 2019 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Molecular Therapies for Choroideremia**

#### **Jasmina Cehajic Kapetanovic 1,2,\*, Alun R. Barnard 1,2 and Robert E. MacLaren 1,2**


Received: 31 July 2019; Accepted: 20 September 2019; Published: 23 September 2019

**Abstract:** Advances in molecular research have culminated in the development of novel gene-based therapies for inherited retinal diseases. We have recently witnessed several groundbreaking clinical studies that ultimately led to approval of Luxturna, the first gene therapy for an inherited retinal disease. In parallel, international research community has been engaged in conducting gene therapy trials for another more common inherited retinal disease known as choroideremia and with phase III clinical trials now underway, approval of this therapy is poised to follow suit. This chapter discusses new insights into clinical phenotyping and molecular genetic testing in choroideremia with review of molecular mechanisms implicated in its pathogenesis. We provide an update on current gene therapy trials and discuss potential inclusion of female carries in future clinical studies. Alternative molecular therapies are discussed including suitability of *CRISPR* gene editing, small molecule nonsense suppression therapy and vision restoration strategies in late stage choroideremia.

**Keywords:** choroideremia; gene therapy; REP1; inherited retinal disease; treatment

#### **1. Introduction**

Choroideremia is a rare X-linked recessive inherited retinal disease caused by sequence variations or deletions in the *CHM* gene which are usually functionally null mutations, leading to deficiency in Rab escort protein 1 (REP1) [1–3]. The estimated prevalence is 1 in 50,000 males. Although REP1 is expressed ubiquitously, in humans choroideremia appears only to affect the retinal pigment epithelium (RPE) layer of the eye, leading to a characteristic clinical phenotype of progressive centripetal retinal degeneration. In Ancient Greek, the name choroid derives from χ*ó*ρι*o*ν (khórion, "skin") and εἶδ*o*ς (eîdos, "resembling"). The suffix 'eremia' (ἐρημία) was added to describe the barren appearance (from the root word meaning wasteland or desert). Hence the literal translation of choroideremia is, in relation to the eye, 'the skin-resembling part is deserted'. Interestingly, the incorrect spelling '*choroideraemia'* has been used previously, but this may be based on misinterpretation of the suffix being derived from αἷμα (haima, blood), into which the 'ae' diphthong is still substituted in many non-US English usages. Importantly however, despite reference to the choroid in the name, the disease is now known to be driven primarily by the loss of the RPE, followed by the secondary degeneration of photoreceptors and choroidal atrophy [4]. Recent evidence has shed light on the molecular mechanisms of REP1 contribution to retinal degeneration in choroideremia, describing its essential role in post-translational modification of proteins and in intracellular trafficking of molecules [5]. The process affects primarily the RPE and pigment clumping is the first sign, long before photoreceptor loss. However, since the RPE has an essential role in retinal isomerization in the visual cycle which is more important for rod compared to cone function and hence rod function is impaired quite early in the disease process. As a result, the disease presents with early childhood nyctalopia, but the majority of patients retain excellent visual acuity until the very end stages of disease, presumably because Müller cells can still contribute to the cone visual cycle in the absence of RPE [6–8].

In this chapter we review advances in molecular therapies that have resulted in the development of adeno-associated vector (AAV) gene replacement therapy for choroideremia. The therapy is currently being explored in multiple clinical trials worldwide, having recently reached phase III in the development (Table 1). We discuss new insights into the clinical phenotyping and genotyping of choroideremia male patients and female carriers, including progress from the natural history studies, that will aid disease characterisation, monitoring of disease progression and interpretation of clinical trial endpoints. The review discusses current knowledge and progress in molecular mechanisms of choroideremia and the development of emerging potential therapies.


**Table 1.** Summary of interventional gene therapy clinical trials in choroideremia.

#### **2. Choroideremia Phenotype**

Choroideremia manifests with a pathognomonic fundus appearance characterised by progressive degeneration of retina and choroid (Figure 1). The degeneration starts in a ring around the mid-periphery of the retina and expands both centripetally towards the fovea and anteriorly to the pars plana [7,15,16]. The anatomical changes are accompanied by loss of functional scotopic vision and the reduction of the mid-peripheral visual field that begins during the first and second decade of life. The visual acuity is

generally well preserved until late in the disease process, usually until the fifth decade of life, when the degeneration starts to encroach onto the fovea [6–8,15,16].

It remains somewhat unclear whether the RPE, the retina and the choroid are all primarily affected, or whether one or more of these tissues is secondarily affected during the pathogenesis of choroideremia [4].

There is however mounting indirect evidence that the RPE is the primary site of the disease in choroideremia, with the inner (photoreceptor) and outer (choroidal) layers degenerating through secondary mechanisms [5]. The unique pattern of preserved retina and RPE, as seen on autofluorescence imaging (Figure 1), with sharply demarcated edges is very different from many other retinal diseases where preserved regions are more circular or oval. This appearance is, however, almost identical in dominantly inherited *RPE65* retinal diseases. Since *RPE65* is only expressed in the RPE, we know that this phenotype is a feature specific to the RPE (presumably, RPE cell death), giving indirect evidence that choroideremia is a disease driven by RPE loss. The confounding variable in choroideremia is that the REP1 protein is expressed throughout the body [17] and the name 'choroideremia' gives the impression that this is primarily a choroidal degeneration. This is not the case, however, because any disease or treatments such as cryotherapy that destroys the RPE layer alone, will eventually lead to secondary atrophy of the underlying choroid, in a similar manner. In other words, choroideremia is the phenotype of complete RPE cell loss. The other relevant factor is that male patients with choroideremia can develop choroidal new vessels (Figure 2) and this clearly shows that the choroidal vasculature has the capacity to regenerate in certain cases. Finally, we know from female carriers (Figure 3) that the pattern of RPE loss is very similar to that in carriers of ocular albinism. There is no evidence of X inactivation leading to patchy loss of the choroid independently in female carriers.

**Figure 1.** Retinal imaging in choroideremia. Widefield optomaps, Optos, Dumfernline, UK (**A**,**B**) and Heidelberg Spectralis imaging, Heidelberg, Germany (**C**–**F**) showing choroideremia phenotype in an affected male. Colour fundus photographs (**C**,**D**) show extensive retinal degeneration with choroidal atrophy and visualisation of underlying pale sclera. Fundus autofluorescence (**E**,**F**) shows typical patterns of sharply demarcated areas of remaining tissue (hyperfluorescent) against atrophic retina (hypofluorescent background). Mesopic microperimetry, MAIA CenterVue SpA, Padova, Italy (**G**,**H**) measures central retinal sensitivity that closely maps areas of residual retina as seen on autofluorescence. Sensitivity maps are shown with corresponding histograms of threshold frequencies. Spectral domain optical coherence tomography, Heidelberg, Germany (**I**,**J**) shows retinal structure in cross-section with distribution of ellipsoid zone (yellow line) and preserved inner retinal layers.

**Figure 2.** Retinal imaging in a choroideremia patient showing an area of scaring from an old choroidal neovascular membrane in the left eye. Fundus autofluorescence (**A**,**B**), fluorescein angiography (**C**,**D**), indocyanine green angiography (**E**,**F**) and spectral domain optical coherence tomography (**G**) with arrows marking the old scar. Imaging was performed with Heidelberg Spectralis, Heidelberg, Germany.

It is also possible that REP1 expression may be important for rod photoreceptor function [18]. Processing of post-mortem tissue from patients can make histological analyses difficult, and studies using advanced imaging techniques have provided somewhat equivocal results in terms of evidence of independent rod degeneration in humans in areas of the retina where the underlying RPE cells are unaffected by the disease [6,8,18–20]. Since patients with choroideremia maintain excellent visual acuity until the very late stages of the disease [6–8], it is likely that the REP1 deficiency is not a significant factor for the cone photoreceptors.

**Figure 3.** Retinal imaging in two female choroideremia carriers. Phenotype of an asymptomatic mild carrier with Snellen visual acuity of 6/5 in both eyes is shown from (**A**–**F**) and a carrier with a 'geographic-pattern' phenotype and reduced visual acuity of 6/7.5 in the right eye and 6/12 in the left eye is shown from (**G**–**L**). Fundus autofluorescence showing very early signs of fine 'salt and pepper' mottling (**A**,**B**) compared with coarse mottling and atrophic patches resembling geographic patterns (**G**,**H**). Mesopic microperimetry, MAIA CenterVue SpA, Padova, Italy showing sensitivity maps with corresponding histograms of threshold frequencies. Near-normal central retinal sensitivity is found in mild, asymptomatic carriers (**C**,**D**) compared to reduced retinal sensitivity in affected carriers especially in the left eye of the above case (**I**,**J**). OCT imaging is clinically insignificant in mild, asymptomatic carriers (**E**,**F**) whereas some disruption of retinal pigment epithelium (RPE) and ellipsoid zone is observed in the affected carrier, particularly in the left eye (**K**,**L**).

Elucidating the pattern of degeneration in choroideremia may help us understand the basis of the disease and how it progresses [16]. It is not known why the degeneration in choroideremia starts in the equatorial region before spreading anteriorly and posteriorly to reach the macula. The retinal pigment epithelial cell density is roughly similar at 5000 cells per mm<sup>2</sup> throughout the posterior eyecup. Gyrate atrophy of the choroid however may develop in a similar distribution, although this is in contrast to age-related macular degeneration, which is very much focused in the region around the fovea. In a recent study it was shown that the rate of degeneration in choroideremia followed an exponential decay function and was very similar across patients of different ages [21], but the key factor that determined the severity of the disease was the age of onset of degeneration. It may therefore be possible to predict the severity of the disease simply by measuring the residual area in a patient at a given age, because the progression is likely to be constant in the absence of treatment.

The centripetal degeneration in choroideremia has two phases by fundus autofluorescence-mottled RPE up to the edge and a more central zone of smooth RPE, both of which shrink progressively. In more advanced stages of the disease there is a total loss of smooth zone. The anatomical basis for these two zones is not immediately clear, but it may be that the slightly increased RPE cell density and much thicker choroid at the posterior pole provides some degree of protection against the metabolic stress caused by REP1 deficiency. Recent evidence suggests that there is less preserved autoflourescence area in nasal macula that may be more vulnerable to degeneration [16]. Further studies are necessary to determine whether the RPE zones can predict the health status of the overlying photoreceptors and how these might be affected following treatment.

#### **3. Choroideremia Genotype**

The choroideremia gene, *CHM* (OMIM #300390), encodes the REP1 protein, a 653 amino acid polypeptide essential for intracellular trafficking and post-translational prenylation of proteins within the human eye. Currently, there are 346 mutations registered on Leiden Open Variation Database, LOVD<sup>3</sup> (www.lovd.nl/CHM). Almost all of the identified sequence variations regardless of mechanism, are predicted to be null [3,22–25]. The mechanisms include insertions and deletions (minor, a few nucleotides, and major involving up to the entire gene length), splice site mutations, missense changes and point mutations that result in stop codons (premature termination codons). Novel mutations have recently been identified involving a deep-intronic region [26] and a promoter region [27] of the *CHM* gene.

Compared with other genetic diseases including inherited retinal disease, choroideremia has a surprisingly low number of disease-causing missense mutations. This would suggest that the REP1 protein, with 3 principal domains, has no catalytic domains with corresponding mutational hotspots within the gene. This is in contrast to genes that encode enzymes (such as *retinitis pigmentosa GTPase regulator* gene) that typically have such hotspot regions (e.g., ORF15 region). This supports the role of REP1 as a chaperone protein, enhancing activity of another protein, which is important in cell structure and stability.

Recent evidence shows that the majority of missense mutations are disproportionately found to be single point C to T transitions at C-phosphate-G (CpG) dinucleotides, spread across 5 of only 24 CpG dinucleotides in the entire *CHM* gene [25]. This is consistent with the evolutionary loss of CpG dinucleotides through destabilising methylation and subsequent deamination. Notably, the 5 locations were the only sites at which C to T transitions resulted in a stop codon. Future de novo mutations are likely to arise within these destabilised hotspot loci.

Molecular genetic testing offers means of confirming the clinical diagnosis in choroideremia and is mandatory for the inclusion in gene therapy clinical trials. It also offers a means of identifying carriers and establishing presymptomatic diagnoses in families that carry a pathogenic change. The rate of mutation detection via next generation sequencing has been reported as high as 94% [25]. In cases of unidentified mutations, it is important to request sequencing of the above mentioned deep-intronic and promoter regions, that are not routinely sequenced, to check for pathogenic variations. In addition, functional in-vitro assay that measure levels of REP1 in peripheral blood cells and its prenylation activity [17], can support clinical diagnosis and confirm variants of uncertain pathogenicity. In this regard, choroideremia is different to retinitis pigmentosa, because the unique choroideremia phenotype can justify the additional resources needed to sequence the entire CHM genomic region.

#### *3.1. Genotype–Phenotype Correlation in Choroideremia*

Although the clinical phenotype can vary in terms of the age of onset of retinal degeneration and rate of progression, no evidence has been found for genotype–phenotype correlation with regard to onset of symptoms, decline in visual acuity and visual fields [23–25], or in the residual retinal area of fundus autofluorescence [25]. The reasons for this are not fully understood, but the lack of correlation may be due to the near universal absence of REP1 irrespective of the causative mutation that range from single point missense changes to whole gene deletions. The phenotypic variation in choroideremia may in part be explained by the degree to which the absence of REP1 can be compensated by other prenylation proteins such as REP2, which shares 95% of its amino acid sequence with REP1 [26,27]. In addition, genetic modifiers and environmental factors may play roles in the onset and progression of degeneration in choroideremia.

#### *3.2. Molecular Mechanisms of Choroideremia*

The molecular mechanisms involved in the pathology of choroideremia have recently been reviewed in great detail [5]. However, some basic concepts are worth re-stating and outlining to aid understanding of the disease. The gene that is disrupted in choroideremia produces REP1 protein. Unlike in many other inherited retinal diseases, this protein is not directly involved in the process of phototransduction or in cellular signalling within the retina. Instead, REP1 is a key player in the addition of prenyl groups (prenylation) to the Rab family of GTPases (Rabs). Such hydrophobic prenyl groups are thought to be necessary to anchor Rabs to the membranes of intracellular organelles and vesicles [28].

In the absence of REP1, there is an observable deficit in the prenylation of several different types of Rabs, and their association with membranes appears to be impaired [29]. Because Rabs themselves act as important regulators of intracellular membrane trafficking, many fundamental cellular processes can potentially be impacted by this deficit. Information from a variety of sources points to a deficit in melanosome trafficking, a delay in phagosome degradation and an accelerated accumulation of intracellular deposits in RPE cells caused by loss of REP1 [4,18,30–33]. The cellular deficits of photoreceptors themselves have been less studied, but it has been suggested that there is mislocalisation of opsin and shortening of photoreceptor outer segments in mice that is independent of RPE degeneration [4].

Fortunately, the absence of REP1 does not appear to be catastrophic for all human cells, which is likely due to the fact that there is a built-in redundancy in this system, provided by the presence of the *CHML* gene [34,35]. The *CHML* gene is thought to be an autosomal retrogene of *CHM*, created by the reverse transcription of the mRNA of the original gene and reinsertion in a new genomic location that occurred sometime during vertebrate evolution. The protein product of *CHML*, known as REP2, appears to be able to largely compensate for the loss of REP1. Although a prenylation deficit of certain Rabs can be detected in several cell types of the body [29,36,37], a single report of a systemic, blood-related, clinical phenotype have not been substantiated [38,39] and loss of REP1 appears to cause cellular dysfunction and death that is limited to specific ocular tissues and manifest as a specific disease of the retina. Differential spatial expression does not provide an obvious answer, as both REP1 and REP2 are expressed ubiquitously.

In truth, the reason why absence of REP1 drives a specific degeneration of the RPE and photoreceptor cells remains a mystery. Perhaps more than other cell types, RPE and photoreceptor cells require acute and sensitive regulation of intracellular membrane trafficking to fulfil their cellular functions. Combined with the fact that there is not any appreciable post-natal replacement of these cells, it may simply be that these cell types are sensitive to the generalised, ongoing prenylation deficit, become 'worn-out' early than usual, and undergo a type of accelerated aging and cell death. Alternatively, it has been proposed that REP1 has a selective affinity to particular Rabs that are of special significance to the cell types affected in the disease. For example, it has been suggested there is a particular requirement for correctly prenylated Rab27a to mediate melanosome trafficking in RPE cells [29,40] and Rab6, 8 and 11 might be important in targeting rhodopsin-bearing vesicles to the photoreceptor outer segment [41,42]. Biochemical assays have suggested that REP1and REP2 have largely overlapping substrate specificities but differences in the association with other catalytic units within the prenylation process might contribute instead [43–45].

#### *3.3. Gene Therapy for Choroideremia*

Gene based therapies show great promise for the treatment of inherited retinal disease, including choroideremia [46]. Recent advances have paved a successful progression of gene therapy clinical trials on choroideremia (Table 1). The first phase I/II trial started in Oxford, UK in 2011, using a subretinal delivery of AAV2-REP1 in 14 male patients with choroideremia [9,10]. The two-year trial results were recently reported [11] with median gains in visual acuity (measured by Early Treatment Diabetic Retinopathy Study, ETDRS chart) of 4.5 letters in treated eyes versus 1.5 letter loss in untreated eyes across the cohort at 24 months post treatment. Six treated eyes gained more than 5 ETDRS letters. In two patients with the greatest gains in visual acuity, improvements were noted by 6 months post treatment, and sustained at up to 5 years of follow-up. Two patients in the cohort had complications, one related to surgery (retinal overstretch and incomplete vector dosing) and the other had postoperative inflammation. Both of these events resulted in protocol changes which included developing an automated subretinal injection system and a more prolonged post-operative immunosuppressive regimen.

These encouraging safety and efficacy signals prompted additional trials using the same vector (sponsored by Nightstar Therapeutics, UK) at other international sites including Canada (NCT02077361), USA (NCT02553135) and Germany (NCT02671539) all reporting similar results [12–14], following which a phase III trial started in 2017 at multiple international sites. Independent to the Nightstar led trials, another phase I/II trial (NCT02341807) using a similar AAV vector construct (without the woodchuck hepatitis virus posttranscriptional regulatory element) begun in 2015 in Philadelphia, USA. The results of this trial are expected in the coming years.

The above-mentioned early phase I/II gene therapy clinical trials recruited patients with advanced disease with early efficacy signals suggesting that vision can be restored following treatment. Reassuring safety data, following improvements in the surgical technique, prompted initiation of a phase II trial (NCT02407678) sponsored by University of Oxford that included patients with early central degeneration and normal visual acuity. The REGENERATE trial recently completed recruitment of 30 male patients with choroideremia with prediction that earlier intervention might slow down or halt the degeneration prior to irreversible structural disorganisation.

The solstice study is an observational, long-term follow up study of 100 participants that will evaluate the safety and efficacy of the AAV2-REP1 used in the above-mentioned interventional choroideremia trials.

The outcomes of clinical trials are measured in terms of clearly defined clinical endpoints, which predict the success and ultimately the approval of new treatments. These outcomes must be selected carefully to capture the most sensitive and reliable measures of the disease progression during the course of a clinical trial and will critically depend on the stage of retinal degeneration. In the reported choroideremia trials, the primary endpoint was the change from baseline in best-corrected visual acuity (BCVA) in the treated eye compared to the untreated eye with evidence of gains in vision after gene therapy in treated eyes. This suggests that BCVA can be used as a viable primary outcome in cases of advanced choroideremia, where disease process has already affected the visual acuity. Indeed, the phase III STAR trial is using BCVA as a primary outcome measure. However, in patients with early disease stage with near-normal vision, BCVA may not be the most sensitive outcome measure, especially since the visual loss in choroideremia typically progresses very slowly. Thus, for the REGENERATE trial, secondary endpoints including the measure of central visual field by microperimetry and anatomical measures such as fundus autofluorescence and optical coherence tomography may prove to be additional valuable outcomes. However, measurements of these secondary outcomes may not always be straightforward, and need to be interpreted with caution. For example, the remaining autofluorescence area may not be easily demarcated, even with the use of automated algorithms, which may influence area measurements especially following sub-retinal gene therapy which may differentially affect central (para-foveal) and peripheral areas of the treated island.

#### **4. Should We Treat Female Carriers in the Future?**

Heterozygous female choroideremia carriers often show generalized RPE mottling due to random X-inactivation (Figure 3A–F) and are usually asymptomatic or show early deficits in dark adaptation. In some carriers a coarser pattern of degeneration is seen, with patches of atrophy interspersed with

normal tissue (Figure 3G–L). Usually, a mild reduction in retinal function is observed with this carrier phenotype. Occasionally, however, female carriers manifest with more severe male-like pattern of retinal degeneration with associated deficit in visual function [47]. This is most likely the result of skewed X-inactivation, or the proportion of cells expressing the mutant X chromosome, which occurs during early retinal development.

Choroideremia gene therapy trials are currently including affected male subjects only. For the majority of female carriers who are mildly affected and asymptomatic or have minor deficits in night vision or visual fields, treatment may not be necessary. Such functional deficits are usually slowly progressing with the majority of cases being able to maintain driving standard vision. However, the more severe female carrier phenotypes, with associated visual field loss and reduction in visual acuity, are likely to benefit from gene therapy and could be included in future clinical trials. Careful characterisation and geneotype-phenotypes correlations will help with the inclusion criteria and give insight into the optimal timing for successful gene therapy.

#### **5. Alternative Therapies**

The potential therapy that has been discussed in this review is gene replacement/augmentation therapy. This is the therapy that has advanced the furthest clinically but there are other potential therapies worth considering.

Instead of adding a working copy of the *CHM* gene, it may instead be possible to alter the patient's own copy with gene editing. Techniques to achieve this, such as zinc finger nucleases or Tal-effector nucleases (TALENs), have existed for some time, but the clinical relevance of these techniques has been somewhat limited by the low editing efficiencies generally achieved. The development of CRISPR/Cas (clustered regularly interspaced short palindromic repeats/CRISPR-associated nuclease 9) technology has given gene editing a renaissance for two reasons. Firstly, gene-editing efficiency appears to be generally better, with the potential to be more clinically meaningful. Secondly, in the CRISPR/Cas9 system, most of the investigational medicinal product can remain the same and only a specific RNA guide sequence needs to be developed to target a site within the disease specific gene—this is more attractive in terms of a clinical development pathway. Gene editing therapy is most useful when there is a need to correct or silence a mutated gene, such as when a missense mutation leads to production of dominant negative or toxic gain-of-function protein, which normally manifests as autosomal dominant and semi-dominant disease [48]. Because the vast majority of mutations in choroideremia are effectively null and therefore result in no detectable protein [24] there is no compelling need to develop a gene editing approach, and simply adding a correct copy as an episomal transgene would be sufficient to result in a therapeutic effect. Correcting the genomic copy of the gene might provide higher confidence of a correct and sustained level of expression, given that the gene would be subject to regulation by its normal transcriptional regulation and epigenetic environment. However, there is evidence that expression from a transgene can be sustained for years when using the appropriate delivery vector and expression cassette [49]. For choroideremia, there is no cell type in which ectopic expression may be predicted to cause a problem, as the protein in normally ubiquitously expressed. In terms of the level of expression, we know that the level of restored REP1 expression is inversely proportional to the prenylation deficit, and so far there is no evidence of overexpression causing toxicity [50]. Although it may be theoretically possible to develop a gene editing approach for some mutations that cause choroideremia, using CRISPR/Cas9, the effectiveness of such strategies has not yet been well established in the retina. Therefore, as gene editing might offer only marginal benefits over gene replacement, it is not currently an attractive strategy of treating choroideremia.

Another therapy that has been suggested and developed is the use of drug-stimulated translational read-through (RT) of premature termination codons (PTC). Nonsense mutations arise when a point mutation converts an amino-acid codon into a PTC that can cause premature translational termination of the mRNA, and subsequently inhibit normal full-length protein expression. Occasionally, instead of translational termination, read-through occurs. Here, a partial mispairing of codon–anticodon is successful, an amino acid is incorporated and protein synthesis continues. Small molecule translational read-through inducing drugs (TRIDs) exist that form the basis of the proposed therapy [51]. Nonsense mutations are the cause of choroideremia in over 30% of patients [52], so, although this will not be appropriate for all patients, there is a significant proportion in which it might be used.

Translational read-through inducing drugs have been used in clinical trials for life-limiting congenital diseases, such as Duchenne muscular dystrophy (DMD) and cystic fibrosis. Early trials appeared to successfully suppress premature stop mutations in patients, but there were concerns over toxicity and the need for repeated intramuscular or intravenous dosing. A newer read-through drug, Ataluren (PTC124), showed a good safety profile when administered orally and the clinical benefit shown in DMD has led to its approval in the EU for this disease [53]. Although no adverse effects have been observed so far, even the approximately 48 weeks of administration given in the clinical studies do not approach the decades of treatment that would be necessary for choroideremia. Preclinical work in the lower-vertebrate, zebrafish model has been important in developing the proof-of-concept, as this is currently the only existing model of choroideremia with a nonsense mutation [54,55]. However, absence of the CHML (REP2) gene in zebrafish means that the CHM mutation is lethal—-translational read-through inducing drugs increase the lifespan of the zebrafish model but this is outcome is not directly clinically relevant. The ability of TRIDs to rescue the Rab prenylation defect in fibroblast of a patient with a particular choroideremia nonsense mutation is encouraging, despite the fact that levels of full-length REP1 protein remained below the level of detection [55]. Given the relatively slow disease progression and the potential risks and cost to the patient from long-term administration of TRIDs, it would be judicious to establish that the correction of the prenylation deficit by TRIDs is present in fibroblast from patients with the equivalent nonsense mutations in which treatment will be attempted in any clinical study [56].

It might be argued that systemic or ocular administration of TRIDs has the potential to treat a larger area of retina when compared to gene therapy, as the former might spread by local diffusion while the latter is limited by the extent of the subretinal bleb. However, to our knowledge, the local concentration achieved in the posterior segment of the eye has never been measured when TRIDs are taken orally or administered locally. The effect of TRIDs appears to often follow an inverted u-shaped dose-response curve, so the pharmacokinetics of therapy may be critical important [57]. Until such questions are addressed, it would appear that translational read-through inducing drugs do not represent a superior strategy compared to gene replacement therapy.

Recent work has identified that antisense oligonucleotides (AONs) may also provide another potential therapy for choroideremia [58]. In some cases of choroideremia, deep-intronic mutations can create a cryptic splice acceptor site that results in the insertion of a pseudoexon in the *CHM* transcript. This disrupts gene function, and specific AONs can be designed to bind to the pre-mRNA and redirect the splicing process, potentially returning it to a normal, working transcript [59,60]. For choroideremia, AONs therapy has shown some promising in vitro results but is further along the clinical development pathway for several inherited disorders, including other forms of inherited retinal dystrophy [59,60]. As AON therapy relies on particular types of mutations, it will not be relevant for all cases of choroideremia and such a strategy is most attractive when conventional gene replacement therapy is not possible because of the large size of the coding sequence of the genes involved, such as in *CEP290*-associated Leber congenital amaurosis [59,60].

The therapies above aim to slow down or stop the degeneration of the retina and RPE and are obviously the preferred choice. However, it is also worth considering strategies that might restore vision in the late stages of the disease, when the majority of photoreceptors have already been lost. Cell transplantation is an interesting strategy for the treatment of inherited retinal disease, but this might present a significant challenge in late-stage choroideremia, where RPE and choroid have been lost along with the degenerating photoreceptors. A more feasible approach may be to use some form of retinal prosthesis. Although most systems rely on surviving inner retinal layers, with intact ganglion cell nerve conduction, there is no dependence on survival of the RPE, photoreceptors or choroid. The Argus II

retinal prosthesis, an epiretinal device approved for commercial use in advanced retinal degeneration in the EU and USA, has been implanted in at least one patient with choroideremia [61]. This device has a very good safety profile and various improvements in visual function have been reported, although these vary widely between individuals [62]. Other devices exist or are in development (44-channel suprachoriodal Bionic Eye Device (NCT03406416) Melbourne, Australia and Intelligent Retinal Implant System, IRIS V1 (NCT01864486) and V2 (NCT02670980) Pixium Vision SA) that could theoretically restore much greater levels of visual function than the Argus II, however, stopping cell loss, even at a late-stage will likely still result in a better functional outcome. Another potential therapy to restore vision in choroideremia is to render the remaining cells of the retina sensitive to light by ectopically expressing light-sensitive ion channels or opsins. This strategy, known as optogenetics, has its own considerations and challenges, which will not be discussed extensively here. Suffice to say, a number of systems are in various stages of pre-clinical development and are beginning to be investigated in clinical trials [63–65]. Again, the level of vision that can be restored by this method is likely to be relatively crude, however, this is likely to be comparable to any retinal prosthesis and may offer specific benefits such as less invasive surgery and potential restoration of a wider visual field.

#### **6. Summary**

Molecular mechanisms in choroideremia are well established. Ultimately, the absence or reduced prenylation of REP1 activity disrupts intracellular trafficking pathways leading to accumulation of toxic products and premature degeneration of the retina and vision less. Logically then, replacement of REP1 to the retinal tissue, via gene-based therapy, could restore cellular function and slow down the degeneration. Multiple clinical trials are underway testing this hypothesis. The trials are using subretinal delivery of AAV2-REP1 to target surviving central islands of the retina with promising safety and early efficacy results.

Despite ubiquitous expression of REP1, a robust systemic association with choroideremia has not been identified, although the prenylation defect is visible in assays of the peripheral blood cells. This assay can be used to support the diagnosis of choroideremia. It is not known why the retina is the only part of the body that becomes clinically affected by the lack of REP1 activity. Moreover, the complex interactions between different retinal cell types during the pathogenesis of choroideremia mean that it is difficult to deconvolve the exact order in which RPE, photoreceptors and the choroid degenerate. It appears likely that the RPE is directly affected by the loss of REP1, and is a key driver of pathogenesis, but the importance of primary or secondary degeneration of photoreceptors is less clear. Elucidating these mechanisms may help us to understand what triggers the onset of clinically significant degeneration and how the rate of degeneration in each cell type might be affected following treatment.

Evidence to date has shown no apparent genotype–phenotype correlation within the spectrum of reported CHM mutations, with regard to the onset of symptoms and the rate of functional visual decline. Since variations in male phenotypes cannot be explained by mutations in *CHM* only, genetic modifiers or environmental factors must play a role in the onset and progression of degeneration in choroideremia. Ongoing natural history studies are adding insight into the progression of the disease and the characteristics of the clinical phenotype that will help to establish the optimal therapeutic window for choroideremia. Female carriers should be enrolled into natural history studies with aim to offer gene therapy (under the realm of clinical trials) to those affected by skewed X inactivation.

**Author Contributions:** Writing: J.C.K. and A.R.B. Revision: J.C.K. and A.R.B. Supervision: R.E.M.

**Funding:** This research was funded by Oxford NIHR Biomedical Research Centre, Oxford, UK and Medical Research Council UK; JCK is also funded by Global Ophthalmology Awards Fellowship, Bayer, Switzerland.

**Conflicts of Interest:** REM is a founder and receives grant funding from Nightstar Therapeutics (now Biogen Inc.). REM and ARB are consultants to Nightstar Therapeutics and REM is a consultant to Spark Therapeutics. These companies did not have any input into the work presented.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

#### *Review* **Molecular Strategies for RPGRGene Therapy**

#### **Jasmina Cehajic Kapetanovic 1,2,\*, Michelle E McClements <sup>1</sup> , Cristina Martinez-Fernandez de la Camara 1,2 and Robert E MacLaren 1,2**


Received: 31 July 2019; Accepted: 1 September 2019; Published: 4 September 2019

**Abstract:** Mutations affecting the *Retinitis Pigmentosa GTPase Regulator*(*RPGR*) gene are the commonest cause of X-linked and recessive retinitis pigmentosa (RP), accounting for 10%–20% of all cases of RP. The phenotype is one of the most severe amongst all causes of RP, characteristic for its early onset and rapid progression to blindness in young people. At present there is no cure for *RPGR*-related retinal disease. Recently, however, there have been important advances in *RPGR* research from bench to bedside that increased our understanding of *RPGR* function and led to the development of potential therapies, including the progress of adeno-associated viral (AAV)-mediated gene replacement therapy into clinical trials. This manuscript discusses the advances in molecular research, which have connected the RPGR protein with an important post-translational modification, known as glutamylation, that is essential for its optimal function as a key regulator of photoreceptor ciliary transport. In addition, we review key pre-clinical research that addressed challenges encountered during development of therapeutic vectors caused by high infidelity of the *RPGR* genomic sequence. Finally, we discuss the structure of three current phase I/II clinical trials based on three AAV vectors and *RPGR* sequences and link the rationale behind the use of the different vectors back to the bench research that led to their development.

**Keywords:** *Retinitis Pigmentosa GTPase Regulator*; gene therapy; adeno-associated viral; Retinitis Pigmentosa (RP)

#### **1. Introduction**

Inherited retinal diseases, most of which are retinitis pigmentosa (RP), affect 1 in 4000 people worldwide. The hallmark of this heterogeneous group of disorders is premature degeneration of rod and cone photoreceptors that leads to early vision loss. RP can be inherited as an autosomal recessive, dominant, X-linked, oligogenic, or mitochondrial trait. X-linked RP is one of the most severe forms of retinal degeneration and it accounts for 10%–20% of all RP cases [1–3]. To date, only 3 genes have been identified to be associated with X-linked pattern of inheritance. Mutations in the *Retinitis pigmentosa GTPase regulator* (*RPGR*) gene accounts for over 70% of X-linked RP cases whereas less common forms of the disease are caused by retinitis pigmentosa 2 (*RP2*) and 23 (*RP23* or *OFD1*) genes [4,5].

*RPGR*-related X-linked RP is characterised by severe disease in males with early onset and rapidly progressing sight loss that leads to legal blindness commonly by the fourth decade of life [2]. The classic rod-cone phenotype with peripheral pigmentary retinopathy, waxy optic disc pallor and vascular attenuation makes it often indistinguishable from other forms of RP. Less commonly, a cone-rod phenotype manifests with early central cone degeneration and accompanying loss of visual acuity. Female carriers of the *RPGR* disease are typically asymptomatic with a characteristic phenotype that manifests as a radial streak pattern originating from the fovea [6,7]. Rarely, however, skewed X-inactivation leads to more severe male-like phenotype with associated visual impairment [8].

At present, there is no approved treatment for retinitis pigmentosa caused by mutations in *RPGR*. Several treatment options have been under investigation and with the emergence of novel gene-based therapies for inherited retinal disease, this seems the most logical strategy to develop for the *RPGR* disease. Due to its severe phenotype, relatively high incidence and the fact that more commonly mutated genes such as *ABCA4* or *USH2A* are too large to be packaged into AAV vectors, the *RPGR* disease has drawn significant interest amongst scientific and clinical research communities over the last years. However, due to the inherent instability in the retina-specific RPGRORF15 isoform sequence [9–12] the production of the therapeutic AAV-mediated *RPGR* vector has been very challenging. In attempts to improve the sequence stability and fidelity several approaches have been explored including codon optimisation [13–15], which has allowed generation of vectors for use in human trials. In this review we discuss recent advances in the understanding of *RPGR* gene structure and its evolutionary conservation that has led to an improved understanding of protein's molecular function and mechanisms implicated in the pathogenesis of RPRG-related retinal dystrophy. The pre-clinical development of gene therapy vectors that has resulted in their progression into three phase I/II clinical trials is covered in detail, including discussion on three different *RPGR* cDNA sequences used in the trials.

#### **2. Structure and Function of** *Retinitis Pigmentosa GTPase Regulator (RPGR)*

The human *RPGR* gene is located on the short arm of the X-chromosome (Xp21.1). The gene exhibits a complex expression pattern with 10 alternatively spliced isoforms, five of which are protein coding [16]. The first transcript to be identified in association with X-linked retinitis pigmentosa, was the constitutive RPGREx1-19 isoform. In humans, the RPGREx1-19 isoform contains 19 exons and expresses a full-length messenger RNA transcript of 2448 bp, which generates an 815 amino acid sequence that forms ~90 kDa protein in a variety of tissues [17]. Since this initial characterisation, multiple alternative transcripts have been identified, including the retina-specific RPGRORF15 variant [10,16,18]. This variant contains exons 1–14 of constitutive RPGR with the exon ORF15 derived from alternatively spliced exon 15 and intron 15 (Figure 1A). The RPGRORF15 isoform is 3459 bp, encoding a 1152 amino acid sequence which forms a ~200 kDa protein. As with the widely expressed variant, amino acids 54–367 (exons 3–10) form a regulator of chromosome condensation 1 (RCC1)-like domain. The alternative ORF15 exon consists of a highly repetitive purine-rich sequence coding for multiple acidic glutamate-glycine repeats. This is followed by a C-terminal tail region rich in basic amino acid residues, called the basic domain.

The reason for this complex expression pattern of the RPGR protein remains largely unknown, but may be related to the functional role of its splice isoforms in various cell types. The RPGR protein is widely expressed in vertebrate tissue including eye, brain, lung, testis and kidney. In the eye, the two major isoforms, RPGREx1-19 and RPGRORF15 are predominantly localised to the photoreceptor connecting cilia [19] and less consistently, to the nuclei and photoreceptor outer segments of some species [20]. The connecting cilium is a critical junction between the inner and outer photoreceptor segments, controlling the bidirectional transport of opsin and other proteins involved in the phototransduction cascade and the overall health and viability of the photoreceptors. Attempts are ongoing to elucidate further the expression patterns of RPGR through evolutionary characterisation of RPGR domains across species and via molecular interactions of RPGR with other proteins in order to shed light on the exact role of the RPGR protein.

The RCC1-like domain, present in both major splice forms, adopts a seven-bladed β-propeller structure and it is strongly conserved across evolution, in vertebrates and invertebrates [9]. This domain has been implicated in a regulatory role of small GTPases. It is thought to enable RPGR to act as a Ran guanine nucleotide exchange factor and RPGR has been shown to upregulate the guanine nucleotide exchange factor RAB8A, associating with the GDP-bound form of RAB8A to stimulate GDP/GTP nucleotide exchange [21]. The RCC1-like region also interacts with: RPGR interaction protein 1 (RPGRIP1), which links it to the connecting cilium of photoreceptor cells [19]; the lipid trafficking protein phosphodiesterase 6D (PDE6D) [22]; two chromosome-associated proteins important for the

structural maintenance of chromosomes, SMC1 and SMC3 [23] and two ciliary disease-associated proteins nephrocystin-5 (NPHP5) [24] and centrosomal protein 290 (CEP290) [25].

**Figure 1.** *Retinitis Pigmentosa GTPase Regulator (RPGR)* gene structure and splicing variants. (**A**) Human *RPGR* gene exon-intron structure showing the combination of exons 1 to 19 to create the constitutive protein isoform, and alternative splicing of exon 15/intron 15 that creates the RPGRORF15 variant. (**B**) Mouse RPGR gene exon-intron structure showing the combination of exons 1 to 18 to create the constitutive protein isoform and alternative splicing of intron 14 creates the RPGROFR15 variant.

The retina-specific ORF15 domain is also highly evolutionarily conserved across varied species, indicating a functional importance (Table 1). However, in contrast to the RCC1-like domain, the ORF15 domain is unique to vertebrates, suggesting a role that is unique to the ciliary-derived photoreceptors of "simple" vertebrate eyes, compared to the rhabdomeric photoreceptors of "compound" invertebrate eyes. Hence, the ciliary-based transport of cargoes such as rhodopsin, which is at least 10 times more abundant in vertebrates than invertebrates, fits with this hypothesis. ORF15 homology and a region of high AG content of >80% is identifiable in a range of species although the length varies—the mouse

ORF15 is shorter than the human ORF15, Figure 1A). This purine-rich region of ORF15 (97.5% purines within 1kb in humans) encodes the glutamine-glycine rich domain that ends in a basic C-terminal domain, which is also highly conserved, suggesting that it constitutes another functional region. This basic domain, which is unique to RPGRORF15, interacts with at least two proteins, a chaperone protein nucleophosmin and a scaffold protein whirlin [26]. Neither protein is unique to vertebrate photoreceptors, but nucleophosmin is present in metaphase centrosomes during cell division, while whirlin helps to maintain ciliary structures within the eye and ear.

**Table 1. Evolutionary conservation of DNA and amino acid sequences of RPGRORF15 variants across selected species.** All data were extracted from NCBI database files with comparisons performed in Geneious Prime 2017.10.2. For *Homo sapiens*, details were extracted from gene files NG\_009553.1 and 6103 combined with mRNA file NM\_001034853.2. \* The conserved basic domain of the human *RPGRORF15* coding sequence was used for predictions of ORF15 locations in all other species sequences by homology alignment. For *Mus musculus* data, gene files NC\_000086.7 and 19893 were aligned with the basic domain of human *RPGRORF15* and the partial sequence file AF286473.1 to identify the predicted ORF15 variant. For *Canis lupus familiaris* data, gene files 403726 and AF148801.1 were aligned with the basic domain of human *RPGRORF15* and the partial sequence file AF385629.1. For *Pan troglodytes* data, files 4465569 and XM\_024352988 were used. For *Gorilla gorilla gorilla*, files 101149059, the basic domain of human *RPGRORF15* and the partial sequence AY855163.1 were combined. For *Macaca mulatta*, files 714316, the basic domain of human *RPGRORF15* and the partial sequence file AY855162.1 were combined. Finally, *Xenopus tropicalis* sequence predictions were achieved from files 733454 and XM\_018091818.1.


The function of the repetitive glutamine-glycine-rich domain itself has been difficult to establish due to its variable length and relatively poor conservation at the individual amino acid level, although the overall charge and repeat structure length remain conserved in vertebrates. However, recent evidence shows that this intrinsically disordered region is heavily glutamylated [27], a post-translational protein modification that adds glutamates to target proteins to affect their stabilisation and folding. This process is known to be essential for the function of tubulins in intracellular trafficking [28]. Furthermore, this glutamylation has been shown to be achieved by tubulin tyrosine ligase like-5

(TTLL5) enzyme, which interacts directly with the basic domain of the OFR15 to bring it into the proximity of glutamylation sites along the glutamine-glycine-rich repetitive region [29]. The role of the ORF15 region is of course critically important to photoreceptor function, because otherwise ORF15 mutations would not be pathogenic since the RPGREX1-19 variant is still expressed in these cells. Hence in-frame deletions in the ORF15 region lead to progressive loss of function as the deletion length increases [13].

#### **3. Molecular Mechanisms and Pathogenesis of RPGR-Related X-Linked Retinitis Pigmentosa (RP)**

Molecular mechanisms and pathogenesis of RPGR-related X-linked RP have been under investigation for several decades. The drive to better understand the disease process comes from the high incidence with mutations in the gene encoding RPGR accounting at least 70% of X-linked RP and up to 20% of all RP cases [2–4]. Moreover, the disease is associated with one of the most severe phenotypes among inherited retinal diseases with central visual loss occurring early in adult life [2]. This coupled with the developments in genetic therapies has given impetus to a large number of studies aimed to uncover the pathogenic mechanisms.

Despite ubiquitous expression of the constitutive RPGR variant in ciliated cells throughout the body, the RPGREx1-19 has yet to show a firm association with any human disease. The RPGR-related phenotype seems to be confined to the retina and several studies have established an essential role for RPGRORF15 in photoreceptor function and survival [10,11]. Genetic studies have shown that mutations in the RPGRORF15 result in abnormal protein transport across the connecting cilium, which can lead to photoreceptor cell death [12,30,31]. However, there are reports in the literature that describe RPGR-related X-linked retinitis pigmentosa syndrome comprising of retinitis pigmentosa, recurrent respiratory tract infections and hearing loss [32,33]. These findings point to the abnormalities in respiratory and auditory cilia in addition to the photoreceptors. In addition, as photoreceptors develop from ciliated progenitors, it has been postulated that the axoneme may play a role in their early development. Sperm axonemes were thus studies in patients with X-linked retinitis pigmentosa and a significant increase in abnormal sperm tails was observed [34]. Similar findings have been reported in another syndromic ciliopathy, the Usher syndrome [35].

Mutations in *RPGR* account for ~70% of cases of X-linked RP and have been identified across exons 1–15, yet up to 60% of mutations occur in the ORF15 region [10,30]. The repetitive nature of the glutamate-glycine region in ORF15 is prone to adopt unusual double helix DNA conformations or triplexes that are thought to promote polymerase arrest and block replication and transcription. These imperfections are likely to contribute to genome instability and account for the high frequency of mutations in this region, known as the mutation 'hot spot' of the *RPGR.* Surprisingly, no disease-causing mutations have been reported in exons 16–19 [36].

The most common mutations are small deletions that lead to frameshifts followed by nonsense mutations [30]. Within ORF15, the most common mutations are microdeletions 1–2, or 4–5 bp [10], that cause frameshifts leading to truncated forms of the protein and in particular, loss of the C-terminus. Small in-frame deletions or insertions (and missense changes) that can alter the length of ORF15 region by a few base-pairs (e.g., up to 36, equivalent to 9 amino acids in this population based study [37], are seemingly well tolerated [38]. Thus, despite being a coding region, this domain has a surprisingly high rate of tolerable indels within primate lineages, suggesting a rapidly evolving region [9]. However, recent evidence shows that larger deletions in the ORF15 region significantly affect the degree of RPGR glutamylation, which may subsequently influence its function and ability to associate with the cilium and other interacting factors [29]. Thus, frame shift mutations that lead to loss of the C-terminal basic domain are invariably disease causing [12]. In addition, mutations that lead to the loss of TTLL5 enzyme, the basic domain-binding partner that mediates RPGR glutamylation, abort glutamylation process and cause RPGR-like phenotype in humans [39]. This further supports the critical role of glutamylation in normal RPGRORF15 function. It remains intriguing that despite its

ubiquitous expression, the RPGREx1-19 is unable to compensate for the loss of function of RPGRORF15 in the retina to rescue the phenotype. It is possible that the alternative splicing in the retina could favour the RPGRORF15 variant, so the majority of transcripts will be the RPGRORF15 isoform, with few constitutive variants available to compensate. One study failed to identify the constitutive transcript in the retina [18], which supports the finding that the constitutive isoform is expressed early in development in a mouse before its levels decline and the RPGRORF15 becomes the predominant isoform [26]. Notably, the constitutive variant lacks the glutamate-glycine repetitive region and given the importance of this domain for the normal function of RPGRORF15 in the photoreceptors, perhaps it is not so surprising that the constitutive variant cannot offer the same functional benefit as the RPGRORF15 variant.

#### **4. Clinical and Genetic Diagnosis of RPGR-Related X-Linked RP**

The diagnosis of RPGR-related retinal dystrophy is made on the basis of presenting symptoms and retinal signs seen on clinical examination and various imaging modalities. In addition, study of family history showing X-linked inheritance (no male to male transmission) and genetic testing identifying the pathogenic mutation are important in confirming the diagnosis. In cases of uncertain diagnosis and unequivocal genetic test results we have adopted several important steps, which are discussed below, in order to minimise the risk of establishing an incorrect diagnosis, and administering the patient with an incorrect gene if recruited into a gene therapy clinical trial.

RPGR-related retinal dystrophy is associated with a very heterogeneous phenotype that ranges from pan-retinal rod-cone to predominant cone dystrophy (Figure 2). The phenotype is generally more severe with faster progression compared to other forms of RP and median age of legal blindness of approximately 45 years old, which is much younger than in other RP genotypes [40]. Most patients lose their peripheral vision first, followed by the loss of central vision. Recent evidence suggests that the rod-cone phenotype is found in 70% of patients, the cone-rod in 23% and the cone phenotype in 7% of patients with X-linked RPGR related retinal dystrophy [2]. The study shows that the onset of symptoms was in early childhood in rod-cone dystrophy (median age 5 years) and in third decade in cone-rod and cone dystrophy, although the age range was very wide (between 0 and 60 years). However, cone-rod and cone dystrophies were associated with a more severe phenotype and the probability of being blind at the age of 40, with visual acuity of less than 0.05 LogMAR (3/60 or 20/400) observed in 55% of patients with cone-rod and cone dystrophy compared to only 20% in rod-cone dystrophy.

The RPGR phenotype (Figure 2) has been associated with anatomical changes including central retinal thinning of the outer nuclear layer as seen on retinal cross-sections taken by optical coherence tomography [40,41]. The junction between the inner and outer photoreceptor segments, better known as the ellipsoid zone, can be used as an important predictor of central retinal function and for monitoring of disease progression. [42]. Thus, the disruption of the ellipsoid zone can be detected with corresponding early reduction in visual acuity and retinal sensitivity as measured by microperimetry. In addition, autofluorescence can be used to assess the health of the retinal pigment epithelium with early signs of hyper-autofluorescence indicating accumulation of lipofuscin and related metabolites as a by-product of photoreceptor outer segment degradation. Later in the disease process, areas of hypo-autofluorescence become evident indicating outer retinal atrophy with loss of retinal pigment epithelium cells. The RPGR phenotype is often associated with para-foveal hyper-autofluorescent rings, which decline exponentially with disease progression [43]. Constriction areas are correlated highly with baseline area and age, where younger subjects had greatest rate of progression. No correlation with genotype was observed in this study. In the cone-rod phenotype, however, the area of hypo-autofluorescence associated with a surrounding hyper-autofluorescent ring tends to increase in size with disease progression and is inversely related to electroretinogram amplitude [44]. Ongoing natural history studies are promising to shed more light on the natural progression of the RPGR disease phenotypes and provide better understanding and interpretation of clinical trial endpoints used in current interventional gene therapy trials (Table 2).

**Figure 2.** Clinical phenotypes associated with *RPGR* retinal degeneration—rod-cone phenotype (early stage (**A**–**C**) and a more advanced stage (**D**–**F**)) and cone-rod phenotype (**G**–**I**). The phenotypes are captured by Heidelberg fundus autofluorescence, (left column), MAIA microperimetry measuring central retinal sensitivity (central column; sensitivity is represented by a heat map: green/yellow normal/mildly reduced; red/purple—reduced; black—not measurable) and Heidelberg optical coherence tomography showing retinal structures in cross-section (right column). In rod-cone phenotype there is extensive peripheral retinal atrophy with relative preservation of central retina as seen on autofluorescence associated with para-foveal hyper-autofluorescent ring (**A**). This is confirmed by near normal central retinal sensitivity (**B**) and preservation of ellipsoid zone (**C**). In more advanced stages of the disease there is reduction in size of the para-foveal hyper-autofluorescent ring (**D**) with corresponding reduction in retinal sensitivity (**E**) and length of ellipsoid zone (**F**). In contrast, in cone-rod phenotype there is early loss of para-foveal photoreceptors with associated hypo-fluorescent ring and marked reduction of retinal sensitivity with corresponding loss of the ellipsoid zone.

Female carriers of *RPGR* mutations also show high phenotypic variability [7] (Figure 3). The carrier phenotype includes asymptomatic females with near-normal clinical appearance, macular pattern reflex with different degrees of pigmentary retinopathy and severely affected females with clinical phenotype that results from skewed X chromosome inactivation and is indistinguishable from the male pattern. Female carriers with male pattern dystrophy should be considered for *RPGR* gene therapy as discussed below.

The molecular diagnosis using next-generation sequencing (NGS) is usually a robust approach in determining pathogenic variants in RP. However, the ORF15 region of RPGR is not normally sequenced with NGS methods and is currently only performed upon specific request. Moreover, sequencing of the ORF15 region in *RPGR* is notoriously difficult and error-prone. Overlapping reading frames and polymorphic deletions/insertions add further complexity to the detection of true mutations. Additional precautions must, therefore, be taken with interpreting the sequencing data so that small deletions are not confused with artefacts that would lead to spurious results. In cases of uncertainty, testing should be repeated. In addition, the full RP panel should be performed to exclude other pathogenic variants including the sequencing of *RP2* and *OFD1* X-linked genes. This comprehensive molecular genetic analysis together with the *RPGR* phenotype and a clear family history of X-linked inheritance, including evidence of a carrier phenotype, forms the basis of inclusion criteria into gene therapy clinical trials. In addition, a recent study describes an in vitro assay for determining the pathogenicity of *RPGR* missense variations [45]. The strategy is based on the RPGR protein interaction network, which is disrupted by missense variations in RCC1-like domain in RPGR, and could help to differentiate between causative missense mutations and non-disease-causing polymorphisms.


**Table 2.** Summary of clinical trials for RPGR-related X-linked retinitis pigmentosa (RP).

**Figure 3. Clinical phenotype of** *RPGR* **female carriers.** Fundus autofluorescence (Heidelberg) showing a typical macular radial pattern or 'tapetal' reflex in a female carrier of an *RPGR* mutation (**A**,**B**). Random X-chromosome inactivation generates clones of normal or affected photoreceptors giving rise to this mosaic pattern. Blue reflectance (**C**,**D**) and multicoloured (**E**,**F**) modes using Heidelberg scanning laser ophthalmoscope can be very helpful in showing the macular reflex.

#### **5. Treatment Options for RPGR-Related X-Linked RP**

Several non-gene based treatment approaches have been investigated for the preservation of vision in X-linked RP including a nutritional supplement, docosahexaenoic acid [46] and a ciliary neurotrophic factor [47] both of which were unable to prevent photoreceptor degeneration and visual loss. For patients with advanced disease, electronic retinal devices have demonstrated proof-of-concept in their ability to restore crude vision [48,49]. However, the unpredictability of benefit for individual patients and the high price of these devices make it economically difficult to maintain their availability for the treatment of patients with RP. Another potential strategy, optogenetics, is under investigation and has shown promising results for vision restoration in advanced retinal degeneration [50,51].

Emerging gene-based therapy using the AAV vector is currently the most promising therapeutic strategy for RPGR X-linked RP. The size of the coding sequence of RPGRORF15 (3.5 kb) is within the AAV carrying capacity and the relatively high prevalence and disease severity have justified development of this therapy. However, the repetitive sequence of ORF15 not only makes it a hotspot for mutations but also creates challenges for therapeutic vector production. Attempts to generate AAV vectors for RPGR gene-supplementation strategies have been thwarted by the poor sequence stability of the ORF15 region and transgene production has struggled to control spontaneous mutations and maintain the complete sequence [13,52–55]. AAV gene therapy in two RPGR X-linked RP canine models that carry different ORF15 mutations [55] provided proof of concept for treating RPGR mutations within the ORF15 region. AAV2/5-mediated sub retinal gene delivery of a full-length human RPGR-ORF15 cDNA [10], driven by either the human interphotoreceptor retinoid-binding protein (hIRBP) promoter or the human G-protein-coupled receptor kinase 1 (hGRK1) promoter, prevented photoreceptor degeneration and preserved retinal function in both canine models. However, the AAV2/5.RPGR vector was found to have multiple mutations within the purine-rich exon 15 region that led to toxic effects in mice at higher doses [52] thus posing safety questions for human applications. In an attempt to improve the sequence stability, a step-wise cloning approach was used to generate the correct full-length RPGRORF15 coding sequence (the purine-rich region was generated first and then ligated to the rest of the DNA sequence) [53], which was packaged into the AAV8.GRK1.*RPGRORF15* vector and evaluated in the *Rpgr*-KO mouse. However, despite improved stability, some vector preparations were still ridden with micro-deletions that led to expression of alternatively spliced truncated forms of the RPGR protein that was mislocalised to photoreceptor inner segments and only a partial rescue of the phenotype in treated mice. The truncated forms of the protein were further investigated for their ability to rescue the RPGR phenotype in the *Rpgr*-KO mouse [13]. The short (314 out of 348 ORF15 codons deleted) and the long (126 out of 348 codons deleted) forms of the *RPGRORF15* were tested. The long form demonstrated significant improvement in the disease phenotype, whilst the short form failed to localise correctly in the photoreceptors and showed no functional rescue of the phenotype. Importantly, as discussed above, large deletions in the ORF15 region can affect the glutamylation of the protein and lead to impaired function. Indeed, a follow-up study by the same group tested these truncated vectors [29] for their glutamylation capacity. Unsurprisingly, the long form demonstrated significantly impaired glutamylation (only 30% of the full length protein), whereas the short form showed no detectable glutamylation of the RPGR protein.

To circumvent these issues, the research team of Fischer and colleagues (2017) generated a full-length, human, codon-optimised version of RGPRORF15 to stabilise the sequence, remove cryptic splice sites and increase expression levels from the therapeutic transgene [14]. This enabled reliable cloning and vector production. The resulting AAV8.coRPGRORF15 vector was shown to offer therapeutic rescue in two mouse models of X-linked RP (*Rpgr-*/*<sup>y</sup>* and *Rd9*). This vector is now being used in a Phase I/II/III gene therapy clinical trial in humans (NCT03116113). In addition, the codon optimised form of the RPGR vector used in the canine studies [15] and the truncated form of the RPGR with near-total OFR15 deletion [13] are also being tested in ongoing clinical trials (NCT03316560 and NCT03252847 respectively) as will be discussed further in the next section. A very recent study used a bioinformatics approach as an alternative method to develop a molecularly stable *RPGR* gene therapy vector [56]. The strategy identified regions of genomic instability within ORF15 and made synonymous substitutions to reduce the repetitive sequence and thus increase the molecular stability of *RPGR*. The codon optimized construct was validated in vitro in pull-down experiments and in a murine model, demonstrating production of functional RPGR protein.

#### **6. Gene Therapy Clinical Trials for RPGR-Related X-Linked RP**

The results of the pre-clinical studies described above support the use of AAV-based gene therapy for RPGR-related X-linked RP in humans, in the early to mid-stage of the disease. Ideally, patients with moderately reduced visual acuity and constricted visual fields, but a preserved central ellipsoid zone, should be recruited into gene therapy trials for best expected therapeutic benefits. Interestingly, development of RPGR therapy from bench to bedside has resulted in setting-up of three multi-centre dose-escalation gene-therapy clinical trials (see Table 2 for details). Each trial is using a different combination of AAV vector variant and *RPGR* coding sequence (Figure 4). Specifically, the Nightstar

Therapeutics (now Biogen Inc) sponsored trial (NCT03116113) is using the wild-type AAV8 vector with a human rhodopsin kinase promoter and a human codon optimised full-length RPGRORF15 cDNA sequence (AAV2/8.hRK.*coRPGRORF15*). The second trial sponsored by Meira GTx (NCT03252847) is using a wild-type AAV2/5 capsid with a truncated, non-codon optimised *RPGR* sequence under control of the human rhodopsin kinase promoter (AAV2/5.hRK.*RPGRORF15*). The third trial conducted by Applied Genetic Technologies Corporation (NCT03316560) is using mutated AAV2 capsids (capsids with single tyrosine to phenylalanine (YF) mutations) packaged with full-length, codon optimised human *RPGRORF15* sequence also driven by the rhodopsin kinase promoter (AAV2tYF.GRK1.*coRPGRORF15*).

**Figure 4. AAV vector constructs used in current gene therapy trials:** (**A**) the Nightstar Therapeutics (now Biogen Inc) trial, NCT03116113; (**B**) the Applied Genetic Technologies Corporation trial, NCT03316560; (**C**) the MeiraGTx trial, NCT03252847.

The pre-clinical studies that led to the development of vectors used in human trials were described in detail in the previous section. However, the rationale for using the three different vectors deserves further discussion. The coding sequence used in the Meira GTx trial is an abbreviated form of human *RPGRORF15* sequence. The rationale provided for using the truncated form, which arose through a spontaneous mutation resulting in deletion of one third of the ORF15 region, was because the deletion led it to become more stable, thereby reducing the rate of further recombination errors and potential mutations. Interestingly, the authors also showed that further shortening of this critical ORF15 region significantly affects the protein function, leading to mislocalisation of the protein in photoreceptors and no functional or morphological rescue in a mouse model, confirming the importance of the ORF15 region for photoreceptor function. Importantly, a further study demonstrated that the post-translational glutamylation is reduced by over 70% in this abbreviated form of the RPGRORF15, significantly affecting trafficking of molecules critical for photoreceptor function [29]. However, since RPGR is not expressed highly in photoreceptors, it is possible that over-expression of RPGR with gene therapy can compensate for the reduced trafficking ability. The truncated construct was shown to rescue the photoreceptor function in a murine model of X-linked RP [13]. However, the mouse *RPGRORF15* is naturally shorter than the human *RPGRORF15* with an abbreviated ORF15 region (see Figure 2 and Table 1) much like the engineered abbreviated human construct used in the human trial. Thus, it may not be so surprising that the abbreviated human construct led to the rescue in a murine model, as the two sequences are very similar and the murine model has a milder phenotype compared to humans. The efficacy of this shortened version of *RPGRORF15* has not been evaluated in canine models of X-linked RP and the results from human trials are awaited in anticipation.

The constructs used in the AGTC and the Nightstar Therapeutics (now Biogen Inc.) trials are very similar and encode the full-length human wild-type RPGRORF15 protein. Both constructs applied codon optimisation that was shown by Fischer et al. to confer greater sequence stability with higher expression levels than wild-type RPGR sequence, whilst not affecting the glutamylation pattern in the RPGR protein. The codon-optimised RPGR rescued the disease phenotype in two mouse models of X-linked RP [14] and was recently also validated in the RPGR canine model [15] showing transduction of both rods and cones and preserving the outer nuclear layer structure in the treated retina. The results of the phase I/II trials are expected in the near future.

#### **7. Summary**

X-linked RPGR-related RP is a heterogenous group of disorders with no clear genotype–phenotype correlation. Both rod-cone and cone-rod retinal dystrophies are seen with relatively early onset and rapid progression to blindness that is related to mutations that cause loss of function of this key photoreceptor protein. The complex expression pattern of the *RPGR* gene through cryptic splice sites that create multiple isoforms poses challenges in elucidating its function. However, mounting evidence suggests that retina-specific RPGRORF15 is unique to vertebrates and plays a crucial role in regulating protein trafficking between inner and outer segments as well as in microtubular organisation. Importantly, RPGRORF15 contains a characteristic repetitive purine-rich region that is highly glutamylated and only the glutamylated RPGRORF15 is fully functional. Thus, any mutations that reduce the glutamylation process adversely affect RPGR protein function. In addition, the ORF15 region created challenges for the researches interested in developing *RPGR* gene-based therapies as the repetitive region made it unstable and prone to mutations. The current approach in developing a codon-optimised version of the RGPRORF15 to stabilise the sequence, remove cryptic splice sites and increase expression levels from the therapeutic transgene is now being used in humans, following proof-of-concept studies in murine and canine models of X-linked RP. This approach has allowed the rapid progression towards the first in-human gene therapy trial (NCT03116113) for X-linked RP, which began in March 2017. In parallel, two additional independent research consortia have been developing gene therapies for the RPGR disease. With recent approval of gene replacement therapy Luxturna, for the treatment of *RPE65*-related retinal disease, the precedence for approval of future gene-based therapies has been set and results of the RPGR early phase clinical trials are awaited with great expectation.

**Funding:** This research was funded by Oxford NIHR Biomedical Research Centre, Oxford, UK and Medical Research Council UK; JCK is also funded by Global Ophthalmology Awards Fellowship, Bayer, Switzerland.

**Conflicts of Interest:** REM receives grant funding from Nightstar Therapeutics (now Biogen Inc.). REM is a consultant to Nightstar Therapeutics and Spark Therapeutics. These companies did not have any input into the work presented. No other authors have a conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
