Retinal Organoids from an AIPL1 CRISPR/Cas9 Knockout Cell Line Successfully Recapitulate the Molecular Features of LCA4 Disease

Aryl hydrocarbon receptor-interacting protein-like 1 (AIPL1) is expressed in photoreceptors where it facilitates the assembly of phosphodiesterase 6 (PDE6) which hydrolyses cGMP within the phototransduction cascade. Genetic variations in AIPL1 cause type 4 Leber congenital amaurosis (LCA4), which presents as rapid loss of vision in early childhood. Limited in vitro LCA4 models are available, and these rely on patient-derived cells harbouring patient-specific AIPL1 mutations. While valuable, the use and scalability of individual patient-derived LCA4 models may be limited by ethical considerations, access to patient samples and prohibitive costs. To model the functional consequences of patient-independent AIPL1 mutations, CRISPR/Cas9 was implemented to produce an isogenic induced pluripotent stem cell line harbouring a frameshift mutation in the first exon of AIPL1. Retinal organoids were generated using these cells, which retained AIPL1 gene transcription, but AIPL1 protein was undetectable. AIPL1 knockout resulted in a decrease in rod photoreceptor-specific PDE6α and β, and increased cGMP levels, suggesting downstream dysregulation of the phototransduction cascade. The retinal model described here provides a novel platform to assess functional consequences of AIPL1 silencing and measure the rescue of molecular features by potential therapeutic approaches targeting mutation-independent pathogenesis.


Introduction
Leber congenital amaurosis (LCA) is an autosomal recessive retinopathy which leads to rapid loss of vision in early childhood and typically presents as very poor vision, nystagmus, and reduced or absent full-field electroretinograms (ERG) [1]. Causative genetic mutations of LCA have thus far been identified in 25 genes that are typically expressed in the retina or retinal pigment epithelium [2].

Generation of Isogenic AIPL1 Knockout iPSC by Simultaneous Reprogramming and CRISPR/Cas9 Gene Editing
To establish a human, isogenic photoreceptor model of AIPL1-related LCA, we set out to generate a genotype-matched isogenic iPSC line from a healthy individual using CRISPR/Cas9 technology to promote loss of AIPL1 expression (Figure 1). First, we designed CRISPR/Cas9 guide sequences to target the early exon 1 of AIPL1 downstream of the starting codon ( Figure S1A). A Cas9-mediated double-strand break (DSB) in this region is expected to induce indel formation by the non-homologous end joining (NHEJ) repair pathway, resulting in the frameshift of the coding region and consequent gene silencing. Four gRNAs selected by in silico analysis-AIPL1 g1.1, g1.2, g1.3 and g1.4-were designed to direct the Cas9 endonuclease to AIPL1 exon 1 ( Figure S1A). The targeting ability of each guide was tested in the HEK293T cell line transfected with AIPL1 gRNA and Cas9-encoding constructs PX458 ( Figure S1B). We observed that all guides effectively induced targeted indels in AIPL1 exon 1 as detected by the T7 endonuclease I assay ( Figure S1C) with an approximate efficacy of 35% ( Figure S1D, p < 0.0001). We selected AIPL1 g1.4 for the generation of AIPL1 knockout (KO) iPSC by simultaneous reprogramming and CRISPR/Cas9 gene editing of healthy human BJ dermal fibroblasts [24] ( Figure 1A). The emergence of BJ-derived iPSC colonies was detected as early as 14 days post-reprogramming, which were then isolated and expanded within 28 days ( Figure 1B). To screen for AIPL1-edited iPSC, the genotype of single cell-derived iPSC clones was analysed by Sanger sequencing. Of a total of 66 clones, homozygous AIPL1 KO was observed in 11 clones (16.7%), while heterozygous KO was observed in 3 clones (4.5%) and 52 clones (78.8%) were unedited ( Figure 1C). We selected one iPSC clone with homozygous AIPL1 KO through a 1 bp insertion at the AIPL1 g1.4 target ( Figure 1D), resulting in a missense mutation and frameshift leading to a premature stop codon within exon 2 (p.H18Qfs*24). Immunocytochemistry analysis validated the pluripotency of both AIPL1 wild type (WT) and AIPL1 KO lines, with the detection of OCT4, TRA-1-60, NANOG and SSEA4 markers ( Figure S2A), and detection of trilineage potential toward germ layer differentiation ( Figure S2B). Finally, we validated the specificity of the AIPL1 KO by sequencing the 10 most likely off-targets predicted in silico with the highest homology to the AIPL1 g1.4 target sequence. No undesirable off-target mutations were detected in these lines ( Figure S2C). targeting ability of each guide was tested in the HEK293T cell line transfected with AIPL1 gRNA and Cas9-encoding constructs PX458 ( Figure S1B). We observed that all guides effectively induced targeted indels in AIPL1 exon 1 as detected by the T7 endonuclease I assay ( Figure S1C) with an approximate efficacy of 35% ( Figure S1D, p < 0.0001). We selected AIPL1 g1.4 for the generation of AIPL1 knockout (KO) iPSC by simultaneous reprogramming and CRISPR/Cas9 gene editing of healthy human BJ dermal fibroblasts [24] ( Figure 1A). The emergence of BJ-derived iPSC colonies was detected as early as 14 days post-reprogramming, which were then isolated and expanded within 28 days ( Figure 1B). To screen for AIPL1-edited iPSC, the genotype of single cell-derived iPSC clones was analysed by Sanger sequencing. Of a total of 66 clones, homozygous AIPL1 KO was observed in 11 clones (16.7%), while heterozygous KO was observed in 3 clones (4.5%) and 52 clones (78.8%) were unedited ( Figure 1C). We selected one iPSC clone with homozygous AIPL1 KO through a 1 bp insertion at the AIPL1 g1.4 target ( Figure 1D), resulting in a missense mutation and frameshift leading to a premature stop codon within exon 2 (p.H18Qfs*24). Immunocytochemistry analysis validated the pluripotency of both AIPL1 wild type (WT) and AIPL1 KO lines, with the detection of OCT4, TRA-1-60, NANOG and SSEA4 markers ( Figure S2A), and detection of trilineage potential toward germ layer differentiation ( Figure S2B). Finally, we validated the specificity of the AIPL1 KO by sequencing the 10 most likely off-targets predicted in silico with the highest homology to the AIPL1 g1.4 target sequence. No undesirable off-target mutations were detected in these lines ( Figure S2C).

Characterisation of Retinal Organoid Structure from AIPL1 KO iPSC
Next, we sought to evaluate the phenotypic effect of the CRIPSR/Cas9-introduced AIPL1 premature translation termination mutation. For this, AIPL1 WT and KO iPSC were differentiated into three-dimensional retinal organoids (RO) using a two-dimensional to a three-dimensional protocol for generating iPSC-derived neural retinal vesicles (NRV) [19]. In both cell lines, NRV was observed after 20-25 days of culture. Following mechanical isolation, NRV increased in diameter and developed clear lamination, and around day 170 began displaying brush borders comprising presumptive photoreceptor structures ( Figure 2A). Evaluation of AIPL1 protein abundance found consistent expression in the outer nuclear layer of mature WT RO, but not in the AIPL1 KO model ( Figure 2B). AIPL1 was detected in rhodopsin-positive rod photoreceptors ( Figure 2C) and cone arrestinpositive cone photoreceptors ( Figure 2D) as early as week 24 (day 170) in AIPL1 WT RO, but not in AIPL1 KO RO at any time point.
Additional RO characterisation found no observable differences between AIPL1 WT and KO RO for the expression of photoreceptor recoverin or photoreceptor cilia (ARL13B) during RO maturation ( Figure S3A), or in the expression of both L/M opsin and S opsin ( Figure S3B). Finally, ONL thickness ( Figure 2E) and cell apoptosis ( Figure 2F) were measured and quantified in the ONL of AIPL1 WT and KO RO with no significant difference found between model types, indicating loss of AIPL1 had minimal impact on RO development and maturation. This was confirmed at the transcriptional level, where comparable expression patterns were observed for markers of photoreceptor development (CRX, NRL) and function (recoverin (RCVRN), rhodopsin (RHO), cone opsins (OPN1SW, OPN1MW/LW) and phosphodiesterase subunits (PDE6A, PDE6B, PDE6G, PDE6C, PDE6H)) as well as markers of retinal cell types, specifically retinal ganglion cells (BRN3B, ISLET1), bipolar cells (PKC-A), Müller glia (CRALBP) and amacrine and horizontal cells (PROX-1) ( Figure S4). Moreover, AIPL1 transcription was retained in KO organoids, first detected in both WT and KO RO at 8 weeks with increased expression during RO maturation ( Figure S4).

AIPL1 Knockout Results in RO with an LCA4 Phenotype
Downstream effects of AIPL1 protein loss were investigated in the AIPL1 WT and KO RO. Quantitative gene expression analysis was carried out for both AIPL1 and PDE6 subunits ( Figure 3A), where expression was significantly increased over time, but no significant differences were measured between the WT and KO RO. Interestingly, a downward trend in AIPL1 expression was detected at each age examined in the AIPL1 KO compared to WT, but this did not reach significance. In addition, key proteins were measured in the whole RO for both WT and KO ( Figure 3B). AIPL1 was consistently detected in WT RO but below the limit of detection in KO RO. Similarly, PDE6α was significantly reduced in KO RO (p = 0.032) compared to WT RO. Recoverin was used as a control for photoreceptor cells, and no difference was observed in recoverin levels between WT and KO RO, confirming that alterations in AIPL1 and PDE6α protein levels were unrelated to RO cellular makeup. The spatial location of rod-specific PDE6 enzymes (PDE6α and PDE6β) was also examined, where they were found to be abundant above the outer limiting membrane in WT but not in KO RO ( Figure 3C). PDE6 enzymes function to hydrolyse cGMP within the retina, and the loss of these enzymes should lead to cGMP accumulation. When whole RO was quantified for cGMP concentrations, a significant increase (p = 0.041) was found in the AIPL1 KO model compared to WT ( Figure 3D). Together, these data suggest AIPL1 KO RO manifest early molecular features of the LCA4 disease phenotype, including downstream effects on post-transcriptional levels of PDE6 and cGMP levels, and would be a suitable model to examine disease progression and treatments. Additional RO characterisation found no observable differences between AIPL1 WT and KO RO for the expression of photoreceptor recoverin or photoreceptor cilia (ARL13B) during RO maturation ( Figure S3A), or in the expression of both L/M opsin

Discussion
LCA4 is an inherited retinal disease caused by loss-of-function mutations in AIPL1, leading to the loss of correctly assembled phosphodiesterase 6 (PDE6) enzymes, and

Discussion
LCA4 is an inherited retinal disease caused by loss-of-function mutations in AIPL1, leading to the loss of correctly assembled phosphodiesterase 6 (PDE6) enzymes, and subsequent lack of cGMP hydrolysis activity causing cGMP accumulation. Here, we generated an LCA4 model using CRISPR/Cas9 by introducing an early frameshift stop mutation leading to premature translation termination (p.H18Qfs*24).
RO generated from AIPL1 KO iPSC did not contain detectable levels of AIPL1 protein compared to RO from isogenic AIPL1 WT iPSC, and while transcript levels were generally lower in the KO model, this did not reach significance. The distance between the de novo terminal codon and the last exon-exon junction should have triggered the transcript degradation by nonsense-mediated decay (NMD), suggesting additional mechanisms rendering the AIPL1 transcript resistant to NMD [25]. Despite the lack of detectable AIPL1 protein, the AIPL1 KO RO model described here exhibited comparable morphology and development to AIPL1 WT RO and a similar abundance of rhodopsin-positive rods and opsin-positive cones within the outer nuclear layer. Previous in vitro models of LCA4 have exclusively utilised patient-derived iPSC to produce RO with LCA4-associated molecular features [20,21] and used CRISPR/Cas9 to correct the AIPL1 mutation to rescue these disease features [21]. These studies also determined that the LCA4 phenotype had no effect on photoreceptor cell development or relative abundance of other cell types within the retinal organoid during development, in line with the present study. In comparison, while no obvious morphological developmental changes were observed in the retina of Aipl1−/− mice, rapid photoreceptor degeneration coincident with outer segment elongation proceeded from postnatal day (P) 12 and was complete by 4 weeks [10]. Consistent with our AIPL1 KO model, reduced levels of the rod PDE6 subunits and increased cGMP levels were observed before the onset of degeneration [10]. While the temporal patterning of retinal progenitors and specification of the major retinal cell types are comparable in the developing human and mouse retina, the specification of all retinal cell types peaks before birth in the human developing retina whereas the specification of rod photoreceptors, bipolar cells and Müller glia peaks after birth in the mouse retina but before P8. Therefore, the reason why overt photoreceptor degeneration is not evident in the human LCA4 RO models is not immediately evident, but we hypothesise that the temporal window to observe photoreceptor degeneration in human RO is too short. Finally, in Aipl1−/− mice decreased expression of bipolar cell postsynaptic proteins at P8 prior to the onset of degeneration [26] suggest more nuanced dysregulation may be induced during early development prior to degeneration of rods and cones. Although not investigated within the scope of this study, it would certainly be an interesting avenue to explore bipolar cell specification and synaptic markers in human LCA4 RO in the future to better understand the early developmental LCA4 phenotype.
Importantly, AIPL1 KO was sufficient to significantly reduce PDE6α and PDE6β at the post-transcriptional level, accompanied by a significant increase in cGMP compared to isogenic WT RO which did not cause increased apoptosis in photoreceptors. These findings are in line with published patient-derived RO models of LCA4 [20,21] and in vivo Aipl1−/− mice [10] and imply an accurate and clinically relevant model has been generated in this study.
The model presented here recapitulates the molecular features of the LCA4 phenotype seen in other model types and therefore provides an alternative, complementary, model system to further interrogate the early molecular pathways involved in disease pathogenesis. Moreover, we anticipate that the model described here will be useful for testing potential gene therapies to treat LCA4 in a mutation-independent manner. One such option is adeno-associated virus (AAV) mediated gene therapy, which is currently a popular choice due to the versatility, good safety profile, and low immunogenicity reported [27,28]. AAV-RPE65 treatment was granted FDA approval as the first gene therapy for an inherited retinal degenerative disease for administration in patients with LCA2 [29,30]. AAV treatment of AIPL1 deficient mice shows good recovery by using CMV [31,32], RK [33], or RHO [32] as a promotor to drive AIPL1 expression and either AAV2/2 [31], AAV2/5 [33], or AAV2/8 [31][32][33] serotypes. Furthermore, the efficacy of different AAV serotypes has been measured in RO, with AAV7m8 [34], and AAV2/2 [35] shown to produce the strongest transduction in each respective study. Finally, AAV2/5-RP2 has previously been tested in retinal organoids, where it was able to restore a healthy phenotype [36], displaying the utility of RO to examine AAV-mediated rescue in a human in vitro system.
Another prospective use of this model results from the elevated cGMP observed, which is a shared downstream pathway in many IRD [37] with high levels of cGMP and cGMP-dependent protein kinase (PKG) linked to rod and cone cell death [38]. Recent work has shown that cGMP analogues can act as a cGMP inhibitor to rescue photoreceptor degeneration in vivo and in murine-derived cell cultures [39,40]. Additionally, downstream molecular pathways shared in IRD and related to elevated cGMP remain viable targets for the development of future therapeutics [41], including PKG [42], PARP activity [42,43] and Ca 2+ -dependent calpain activity [44]. While investigating the activation of these pathways was outside the scope of this study, it is interesting to note that cGMP-mediated PR degeneration was not observed at any time in this model during differentiation (230 days in culture), suggesting that PR degeneration may proceed at a later time point and highlighting the required duration of RO culture as a limitation of this model system. However, the observed increased cGMP demonstrates the value of this model as a highly beneficial intermediary for testing novel therapeutics targeting the elevated cGMP levels and downstream dysregulated pathways in IRD to gain meaningful data from a human-derived in vitro model prior to proceeding towards clinical trials.

Immunofluorescence and Imaging
ROs were fixed in 4% PFA, 5% sucrose in PBS for 30 min at 4 • C then dehydrated for 1 h each in 6.25%, 12.5% and 25% sucrose:PBS at 4 • C, ROs were embedded in Tissue-Tek OCT compound and stored at −80 • C until sectioned. The amount of 7 µm cryosections was mounted on SuperFrost PlusTM slides (Thermo Fisher Scientific) and stored at −20 • C. Slides were stained for ICC by first incubating in block buffer (10% donkey serum (Sigma-Aldrich, Gillingham, UK), 0.01% Triton-X (Sigma-Aldrich, Gillingham, UK) in PBS) for 1 hr at room temperature (RT), then incubating with a primary antibody for 1 h (RT; antibodyspecific dilutions; Table S3). Slides were washed 3 times with PBS and incubated with species-specific secondary antibodies for 45 min (1:1000; Table S3). If required, slides were then incubated with stains (Table S3) for 20 min before further incubation with 4,6diamidino-2-phenylindole (DAPI; 2 mg/mL) (Invitrogen, Loughborough, UK) in PBS for 5 min, then washed 3 times with PBS. Slides were dried at RT and mounted in fluorescence mounting media (Dako; Agilent, CA, USA). Slides used for TUNEL assay (In Situ Cell Death Detection Kit; Merck) were stained as per kit guidelines, then counterstained with DAPI and mounted as per the above protocol. All images were acquired using a Leica Stellaris 5 laser-scanning confocal microscope. Images were prepared using Adobe Photoshop, Image J and Adobe Illustrator CS6.

Image Quantification
ONL thickness was calculated using ImageJ by taking 10 unique measurements from 3 DAPI-stained RO. TUNEL positive cells were quantified using low magnification images where ONL length was calculated using ImageJ and positive cells were manually counted. Statistical significance was determined by unpaired Student's t test, where * denotes a p value < 0.05.

RNA Extraction and Quantitative PCR (qPCR)
RNA from iPSC and ROs was extracted using the PicoPure RNA Isolation Kit (Thermo Fisher Scientific). cDNA synthesis was performed using the High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific). 2× GoTaq Green master mix (Promega, Southampton, UK) was used for DNA amplification by PCR with standard cycling conditions for semi-quantitative PCRs. Real-time PCR reactions were set up with 2× LabTaq Green Hi Rox Master Mix (Labtech, East Sussex, UK) and validated primers (0.25 pmol/uL; sequences detailed in Table S4) and run on an Applied Biosystems QuantStudioTM 6 Flex real-time PCR system. Gene expression levels were normalised to CRX as a consistently expressed PR-specific reference gene to negate differences in size and cellular makeup between samples. Three biological replicates per sample were used to calculate averages and standard deviation, and statistical significance was determined by one-way ANOVA where *, **, *** denote a p value ≤ 0.05, ≤0.01 and ≤0.005, respectively.

Western Blotting
Protein was isolated from 3-6 pooled RO by first incubating in RIPA buffer for 15 min on ice, briefly sonicating, then centrifuging at 4 • C for 10 min at 13 k rpm. Protein concentration was quantified by Pierce BCA assay (Thermo Fisher Scientific) as per manufacturer's instructions. Protein concentrations were normalised by addition of RIPA buffer, then mixed 1:1 with 2× SDS PAGE buffer and heated at 95 • C for 5 min to facilitate protein denaturation. Samples were run on precast 4-20% gradient gels (Bio-Rad, Watford, UK) at 120 V for 2 h, or until the dye reached the end of the gel. The proteins were transferred to a nitrocellulose membrane by wet transfer, at 90 V for 1.5 h, using standard protocols. Membranes were blocked for 1 h at room temperature in 5% skim milk powder in PBS-T, then incubated overnight at 4 • C in primary antibody (Table S3). The following day membranes were incubated in species-specific HRP-conjugated secondary antibody for 1 h, then developed using Clarity MAX Western ECL (Bio-Rad, Watford, UK) for 5 min. Membranes were imaged using a ChemiDoc MP Imaging System (Bio-Rad, Watford, UK). Images were formatted using Image Lab (Bio-Rad, Watford, UK), which was also used to calculate band density. Three independent protein isolates were used per condition and statistical significance was calculated by unpaired Students t test, where * denotes a p value < 0.05.

Quantification of cGMP
cGMP concentrations were calculated for individual whole RO by ELISA (Cambridge Biosciences, Cambridge, UK) according to the manufacturer's instructions. In brief, RO were washed with PBS and then incubated in 100 µL of 0.1 M HCl for 20 min at RT and mechanically homogenised. The samples were centrifuged at 1000× g for 10 min then the supernatant was combined with 200 µL of ELISA buffer. An amount of 50 µL of this sample was used per well. Absorbance was measured at a wavelength of 420 nm. Values given are normalised to volume to give cGMP concentration per the whole RO, and statistical significance is calculated by unpaired Students t test, where * denotes a p value < 0.05.