Rhodopsin DNA sequence correction
DNA repair by Homologous Recombination (HR)
To date a lot of strategies are used to silence or modify the expression of a target gene, but those that are developed in order to repair the mutated sequence are relatively rare. DNA corrective strategies are all based on the use of HR.
DNA is permanently exposed to endogenous and exogenous factors that can create Double Strand Break (DSB), including for example: nucleases and metabolic products such as reactive oxygen species, ionizing radiation, ultraviolet light, chemical agents used in anti-cancer therapy. Even physical stress during mitosis is suspected to be involved in DSB. DSB are also induced more “willingly” by the cell, for example for genetic mixing and proper chromosome segregation during meiosis88,89. To repair these DSB, the cell machinery uses two main mechanisms: HR and Non Homologous End Joining (NHEJ). NHEJ naturally ligates two broken ends together. It has frequently been considered as the error-prone DSB-repair pathway that generates small insertions or deletions. However, the error rate is only about 10-3 per joining event between fully compatible DSB ends90. Unlike NHEJ, HR provides a modified homologue sequence that facilitates recombination at the broken site, and allow the replacement of the mutated sequence by the normal one (or vice versa)91,92 (Figure 11). HR occurs in three steps: (i) the resection of the DNA 5’end, (ii) strand invasion into a homologous DNA duplex and strand exchange, and (iii) resolution of recombination intermediates.
HR naturally only occurs at a rate of 10-6, as a result different kinds of endonucleases have been developed and engineered to stimulate this mechanism by creating DSB and to bring a more important specificity to the reaction93,94. Among them, Meganucleases, Zinc Finger nucleases (ZFN), Transcription Activator Like Effector nucleases (TALEN), and Clustered Regulatory Interspaced Short Palindromic Repeats (CRISPR) systems will be described, due to their involvement in past or ongoing studies in RHO repair strategies.
Extracted from Christian Biertümpfel illustration at Max Planck institute of biochemistry.
∀ Meganucleases, also called homing endonucleases. The Meganuclease family gathers 5 classes based on structures and sequences: LAGLIDADG, GIY-YIG, HNH, His-Cys box and PD-(D/E)XK95. The first one is the most common and the most described in the literature, generally encoded in introns or inteins. These proteins, naturally present in all kingdoms of life, cut double strand DNA with a great specificity due to a large recognition site ranging from 12 to 40 bp. The obvious drawback of natural Meganucleases is the requirement to have a cleavage site near the region of interest, which is naturally extremely rare because of the great specificity of their recognition site and the small natural repertoire of Meganucleases which is around 300 proteins, most of them being still uncharacterized96. Nevertheless, it is possible to create artificial Meganucleases by, for instance, fusing domains of several proteins (also called domain swapping) or mutating specific residues95, but efforts are still required to obtain an artificial Meganuclease for each target site, that also conserves the specificity and efficiency of the natural ones.
Concerning RHO, Cellectis has filed a patent on a batch of Meganuclease variants cleaving a DNA target sequence from RHO (number of patent: WO2011141825 A1). Moreover, Chan et al.5 have already demonstrated the efficiency of Meganuclease cleavage in PR in vivo, after transduction with a recombinant AAV encoding the Meganuclease. Unfortunately there is no natural Meganuclease cleavage site in the rhodopsin gene and they had to introduce one in their model. Nevertheless, they highlight the potential of HR repair in post-mitotic cells, even without introducing a DNA sequence to correct the DSB. They therefore suggested engineering a Zinc-Finger Nuclease, in order to cleave specifically RHO and to promote DSB repair by HR with the help of a non mutated RHO sequence also brought into the cell by gene transfer.
∀ Zinc Finger Nucleases (Figure 12). ZFN represent the first group of engineered nucleases composed of a sequence-specific DNA-binding domain fused to a nonspecific DNA cleavage module. The first one is usually composed of several zinc-finger units, each one containing about 30 amino acids, in a ββα configuration, with some amino acids in contact with 3-4 bases of the target DNA sequence. It is possible to obtain a more specific molecule by gathering several zinc finger motifs, usually at least six are used to recognize a sequence of 18 bp. The DNA cleavage module is usually that of a FokI endonuclease. ZFN are active as a dimer, with two zinc finger binding sites separated by a 5-7bp sequence, containing the cleavage site of FokI97.
In 2010, Greenwald et al. established the first use of a ZFN for a retinal disease gene, by engineering a ZFN designed to bind and cleave a sequence of the human RHO, associated to an ectopic donor DNA fragment homologous to the human RHO sequence, and modified to be insensitive to the ZFN 6. They showed in P23H HEK cell lines that the ZFN correctly cleaved RHO, and induced an increase of 11.7 fold of HR. ZFN could thus be a good strategy to repair mutations on RHO, taking into account that more than 120 mutations have been described to date. Only one RHO-specific ZFN may be sufficient, but only if the frequency of the induced HR does not dramatically decrease with the increase in distance from the DSB.
Moreover, these approaches usually still induce off-target effects, such as unspecific cleavage sites, that can be toxic for the cell. Another drawback is the fact that conventional DNA or RNA-based methods for delivering ZFN into the cell (using viral vectors) are restricted to certain cell types, and can induce side effect like insertional mutagenesis, toxicity, and low efficiency97. Direct delivery of ZFN into the cell can be a solution but requires a good control of the cell targeting.
∀ Transcription Activator Like Effector Nucleases (Figure 12). TALE proteins have been discovered more recently, and are naturally expressed by Xanthomonas bacteria. TALE proteins possess DNA-binding domains composed of a series of about 30 repeat domains of 33-35 amino-acid-sequence-motifs, each one containing 2 hypervariable residues that recognize a single base pair. Like ZF domains, TALE proteins can be associated to endonucleases, recombinases, transposases, to name a few in the same way. TALEN are active as dimers, and often coupled with a FokI endonuclease domain, to promote cleavage of the DNA targeted at the spacer sequence between the two TALE DNA binding domains97.
Despite the fact that TALEN opens the door to an alternative platform for engineering DNA-binding proteins more easily, this technology presents two other new drawbacks, firstly the necessity to start the binding site with a T base, and secondly the difficulty inherent in cloning TALE repeated sequences. Moreover, TALEN possess too many repeated sequences to be delivered by lentiviral transduction, being subject of rearrangements, and their large size preclude their delivery by AAV vectors.
∀ Clustered Regulatory Interspaced Short Palindromic Repeats system (also called RNA-Guided Endonucleases) (Figure 12). CRISPR systems arose in bacteria as a kind of immunity against foreign DNA from phages and plasmids98. Indeed, short sequences of foreign DNA called spacers, are introduced into the host genome at the CRISPR loci, before being transcribed and processed into short CRISPR RNA (crRNA). These last ones enable the silencing of foreign DNA, by its cleavage after the recruitment of CRISPR associated (Cas) proteins99. This is thus a programmable RNA-guided DNA endonuclease that requires co-delivery of plasmid encoding a Cas endonuclease (Cas9 is the simplest system often used), and the crRNA components97. While this technology presents an interesting potential method to manipulate expression100,101, numerous studies are ongoing to evaluate the utility of this system, including potential off-target effects.
One of the major drawbacks of all these strategies based on HR, is the necessity to bring into the cell the modified enzyme able to induce the DSB, also in addition to the correct homologous sequence to repair the targeted gene. The fact that two elements have to be introduced into the cell can represent a brake in term of the delivery method. Furthermore, in order to reduce possible off-target effects, it will be interesting to limit the duration of meganuclease expression in targeted cells, but while it is easily conceivable in mitotic cells, systems for transient expression in post-mitotic cells in vivo still need to be improved.
Because nucleases can induce potentially toxic DNA double-strand breaks in the cell, and because this methodology requires the proper functioning of HR in the target cell, which is variable upon cell cycle, recombinases and transposases have been fused to ZF or TALE domains, allowing site-specific DNA integration, excision and inversion. Nonetheless, these new technologies require improvement in performance and flexibility of these enzymes102,103.
Silencing of mutant gene expression coupled to expression of a normal copy
Zinc Finger – Artificial Transcription Factors (ZF-ATF)
As described previously, Zinc Finger DNA binding domains can be linked to enzymatic proteins such as nucleases, recombinases, transposases and so forth (See part II.1.b) but these binding motifs can also be used to specifically repress or activate gene transcription by coupling with a transcription factor. Indeed, the Zinc Finger DNA binding domain allows binding to a target promoter, and the transcription factor can then modify its activity, and thus repress or increase the expression of a specific gene. Several activator and repressor domains have already been described and are routinely used in ZF-ATF104.
Mussolino et al.7 engineered a Zing finger based mutation-independent approach to repress human RHO expression in a P347S+/- humanized mouse model. They simulated a two-step repression replacement method, with the silencing of RHO expression by an AAV vector on one hand and gene replacement on the other. In their study, they considered that the gene replacement function, which is to produce expression equivalent to a normal gene copy, is done by the endogenous mouse rhodopsin genes (Rho). They demonstrated the efficiency of ZF-transcriptional repressor on human RHO, with a decrease of RHO expression, and a slowdown of PR degeneration. Nevertheless, the method has now to be tested in a heterozygous context, with a normal human allele and a mutant one, in a mouse knockout background model. Moreover, they only evaluated potential toxicity by histology and ERG, but they did not verify the absence of off-target by analyzing gene expression on DNA chips for example.
Generally, a particular difficulty of this strategy is to correctly design Zing Finger binding domains, to allow their binding on a promoter, meaning that accessibility of the sequence and chromatin state have to be taking into account (by in silico studies or nucleases tests conducted prior, for example).
RNA interference (RNAi) is a well-conserved gene-defense mechanism across model organisms, already observed in plants105,106, but firstly described as RNAi by Andrew Fire and Creg Mello, in C. elegans in 1988107. It is based on the repression of gene expression by a short antisense RNA involved in transcriptional gene silencing at least in yeast108, but has been described in all organisms including mammals, to induce degradation of a targeted endogenous RNA, or to inhibit its translation by post-transcriptional gene silencing109.
Among the expansive world of non-coding regulatory RNA, two pathways are to date widely described in animals: short interfering RNA (siRNA) and micro RNA (miRNA). The first ones are derived from viral infection or transcription of transposable elements and are formed from long double-stranded RNA molecules, while the second ones are triggered by long, single-stranded RNA sequences that fold into hairpin structures110. Following the transcription of the dsRNA and its export to the cytoplasm, the siRNA pathway includes support by an enzyme called Dicer, which cut the RNA into a small molecule of around 21 to 23 nucleotides with a 3’overhangs of 2 nucleotides (this molecule corresponds to the siRNA). This step induces the formation of the RNAi-induced silencing complex (RISC) by recruiting, among others, Argonaute proteins. RISC separates the two strands of the siRNA, and uses it to scan mRNA. After identification of a perfect matching between the siRNA-mRNA duplex, the targeted mRNA is cleaved inducing its degradation. The miRNA pathway shares some steps and factors with the siRNA one, however it requires that firstly the long miRNA precursor is cut by the Drosha enzyme into miRNA before its export from the nucleus. RISC also differs a little from that of the siRNA pathway by its composition particularly in Argonaute proteins. The biggest difference between the two pathways is due to the non-perfect homology between the targeted mRNA and the miRNA. This pathway can also induce significant degradation of mRNA targets despite imperfect mRNA-miRNA base-pairing, but mostly, it leads to repression of translation by inhibition of translation initiation, elongation, premature termination or cotranslational protein degradation111.
Faced with this new natural way of gene silencing, RNAi strategy has totally exploded in the past two decades and has become an indispensable tool in gene function analysis. RNAi protocols now used in therapeutic strategies gather 3 types of molecules:
∀ miRNA under the control of a polymerase II promoter.
∀ chemically synthesized siRNA. These are usually administrated via cell transfection, meaning that they are transiently efficient.
∀ short hairpin RNA (shRNA), mimic pre-miRNA, and can be stably expressed after stable transfection, or transduction with a gene transfer vector. Nevertheless, shRNA are expressed under the control of a polymerase III promoter, and skip the first steps of the RNAi pathway. The solution will thus to deliver shRNA into an endogenous miRNA backbone112.
The most important drawback of RNAi is certainly the potential off-target effects, which can be due to (i) sequence-based homology with non-target RNA, (ii) aberrant processing of small RNA (iii) general cell perturbations owing to the copious presence of RNA species.
This strong expression of small RNA can also induce cell toxicity by saturation of the endogenous RNAi machinery113,114. Another difficulty met in the use of RNAi is the incomplete understanding of the mechanism involved in this RNA-induced silencing, often leading to unpredictable and disappointing efficiency115–117, which vary between experiments and laboratories.
In the particular case of rhodopsin, the first RNAi was developed to silence mouse Rho in a mutation-independent way: after validation of shRNA efficiency to induce more stable silencing compared to siRNA, mutated gene replacement was designed to maintain a natural rhodopsin expression level118. The same group then engineered shRNA targeting human RHO and tested it in humanized mouse models. They demonstrated an efficiency of at least 85% silencing of human RHO in cells transduced by an AAV2/5 encoding the shRNA, and a decrease of PR degeneration by histological analysis when the shRNA was coupled in the AAV vector to a shRNA-insensitive codon-modified human RHO replacement cDNA, in a P23H RHO+/- Rho+/- mouse model119. In order to evaluate the maximal beneficial effect of RNAi silencing/gene replacement strategy in rhodopsin context, they also studied the effect of mutated RHO silencing by AAV-shRNA injection in the RHO P347S+/- Rho+/+ mouse model. By this way, they suppressed mutated RHO expression, and considered that mouse Rho, which is not targeted by the shRNA, acts as replacement gene. They obtained a significant retardation of the disease, observed by histology and ERG8.
In parallel, to avoid the necessity of providing two molecules, an allele-dependent RNAi approach to specifically silence the P23H RHO allele in a rat model was also developed120. Nevertheless, this study did not bring convincing results, certainly due to an insufficient silencing rate, unable to slow PR degeneration.
Table of contents :
I. RETINITIS PIGMENTOSA INDUCED BY RHODOPSIN MUTATION
I.1. EYE AND VISION
I.1.b. The retina
I.1.c. Human photoreceptors
I.2. RETINITIS PIGMENTOSA
I.2.a. Remodeling of the retina and clinical manifestations
I.2.b. Genetic involvement
I.3.a. Rhodopsin synthesis, function and visual cycle
I.3.b. Structure of rhodopsin
I.3.c. Rhodopsin mutations and consequences
I.4. TREATMENTS OF RP
I.4.a. General current treatments of RP
I.4.b. Treatments of RP in the particular case of RHO mutation
II. CURRENT STRATEGIES TO PREVENT MUTANT RHODOPSIN EXPRESSION
II.1. RHODOPSIN DNA SEQUENCE CORRECTION
II.1.a. DNA repair by Homologous Recombination (HR)
II.1.b. HR-induced endonucleases
II.2. SILENCING OF MUTANT GENE EXPRESSION COUPLED TO EXPRESSION OF A NORMAL COPY
II.2.a. Zinc Finger – Artificial Transcription Factors (ZF-ATF)
II.2.b. RNA interference
III. TRANS-SPLICING OF RHODOPSIN PRE-MRNA
III.1. CIS-SPLICING: THE PROTOTYPICAL SPLICE REACTION
III.1.a. Splicing is a phase of pre-mRNA maturation
III.1.b. Steps of the splicing mechanism
III.1.c. Regulation of cis-splicing
III.2. TRANS-SPLICING BETWEEN TWO RNA MOLECULES
III.2.a. Description of the mechanism
III.2.b. Previous studies led with trans-splicing approaches
III.2.c. Strategies of PTM Design
III.2.d. Advantages and drawbacks
IV. VIRAL VECTORS FOR GENE THERAPY OF OCULAR DISEASES
IV.1. ADVANTAGES OF THE EYE AS TARGET OF GENE THERAPY WITH VIRAL VECTOR
IV.2. CHOICE OF THE VIRAL VECTOR FOR OCULAR DISEASES
IV.2.a. Classical viral vectors used in gene therapies
IV.2.b. Previous ocular gene therapies with rAAV
V. CELLULAR AND ANIMAL MODELS OF RP LINKED TO RHODOPSIN MUTATIONS.
V.1. CELLULAR MODEL OF RHODOPSIN EXPRESSION
V.1.a. Primary cultures of photoreceptors
V.1.b. Cell lines developed to study rhodopsin mutations
V.2. ANIMAL MODELS OF RHODOPSIN-INDUCED RETINITIS PIGMENTOSA
V.2.a. Animal models of rhodopsin mutation-induced retinitis pigmentosa and more specifically mouse models
V.2.b. Comparison of mouse and human eyes
I. RESULTS OF IN VITRO ANALYSIS OF PTM EFFICIENCY.
II. IMPROVEMENT AND CHARACTERIZATION OF AN IN VIVO IMAGING TECHNOLOGY TO
MEASURE POTENTIAL EFFECTS OF OUR THERAPEUTIC STRATEGY IN A HUMANIZED MOUSE MODEL.
III. COMPLEMENTARY ONGOING RESULTS.
III.1. DEVELOPMENT OF NEW PTM IN VITRO AND STRATEGIES TO OVERCOME PTM DRAWBACKS
III.1.a. A new PTM that repairs the fifth exon.
III.1.b. Production of a truncated protein in vitro after expression of the PTM alone
III.1.c. Strategies to prevent production of truncated protein.
III.1.d. Analysis of PTM translation alone in vivo.
III.2. STUDY OF THE THERAPEUTIC EFFECTS OF TRANS-SPLICING IN A HUMANIZED MOUSE MODEL
III.2.a. The humanized Rho+/- P347S RHO+ mouse model.
III.2.b. Choice of parameters that regulate PTM expression in vivo.
III.2.c. Effect of AAV2/8 bRho-PTM20 injection in Rho+/- P347S RHO+ mice.
DISCUSSION AND PERSPECTIVES
I. THE PTM SEQUENCE: THE KEY TO EFFICIENT SMART TECHNOLOGY.
II. HOW TO ACHIEVE MORE TRANS-SPLICING THAN CIS-SPLICING?
III. THE MAIN DRAWBACKS OF PTM.
IV. FOLLOWING-UP POTENTIAL BENEFICIAL EFFECTS IN A HUMANIZED MOUSE MODEL
V. ENVISAGING A THERAPEUTIC APPLICATION IN HUMANS.
VI. THE FUTURE OF THERAPEUTIC STRATEGIES TO PREVENT MUTANT RHODOPSIN EXPRESSION.
VII. CONSIDERING GENE THERAPIES FOR RETINAL DISEASES OVER THE NEXT FEW DECADES