NRL−/− gene edited human embryonic stem cells generate rod‐deficient retinal organoids enriched in S‐cone‐like photoreceptors

Abstract Organoid cultures represent a unique tool to investigate the developmental complexity of tissues like the human retina. NRL is a transcription factor required for the specification and homeostasis of mammalian rod photoreceptors. In Nrl‐deficient mice, photoreceptor precursor cells do not differentiate into rods, and instead follow a default photoreceptor specification pathway to generate S‐cone‐like cells. To investigate whether this genetic switch mechanism is conserved in humans, we used CRISPR/Cas9 gene editing to engineer an NRL‐deficient embryonic stem cell (ESC) line (NRL −/−), and differentiated it into retinal organoids. Retinal organoids self‐organize and resemble embryonic optic vesicles (OVs) that recapitulate the natural histogenesis of rods and cone photoreceptors. NRL −/− OVs develop comparably to controls, and exhibit a laminated, organized retinal structure with markers of photoreceptor synaptogenesis. Using immunohistochemistry and quantitative polymerase chain reaction (qPCR), we observed that NRL −/− OVs do not express NRL, or other rod photoreceptor markers directly or indirectly regulated by NRL. On the contrary, they show an abnormal number of photoreceptors positive for S‐OPSIN, which define a primordial subtype of cone, and overexpress other cone genes indicating a conserved molecular switch in mammals. This study represents the first evidence in a human in vitro ESC‐derived organoid system that NRL is required to define rod identity, and that in its absence S‐cone‐like cells develop as the default photoreceptor cell type. It shows how gene edited retinal organoids provide a useful system to investigate human photoreceptor specification, relevant for efforts to generate cells for transplantation in retinal degenerative diseases.

developmental time periods from multipotential retinal progenitor cells. 1 Rod photoreceptors are generated later and in greater numbers, than the cone photoreceptors needed for daylight vision. The NRL gene encodes the neural retina leucine zipper protein, 2 a conserved bZIP transcription factor that in mouse is initially expressed in nascent rod photoreceptor precursor cells after the terminal progenitor division, and persists in rod photoreceptors thereafter. 3 Genetic studies have shown that Nrl is required for the acquisition of rod identity, differentiation and homeostasis, acting via upregulation of downstream rod target genes, such as the Rhodopsin gene, Rho, or Nr2e3, which plays a complementary role in the repression of cone specification genes. [4][5][6][7][8] Nrl acts synergistically with the cone rod homeobox transcription factor, Crx to regulate Rho transcription 7 whereas ectopic expression of Nrl under the control of the Crx promoter leads to a rod-only retina. 9 From an evolutionary perspective, the emergence of the Nrl gene is thought to have facilitated the evolution of mammalian rod photoreceptors from short wavelength-sensitive, S-cone photoreceptors (also known as blue cones), which are considered to represent an ancient photoreceptor fate. 10 Conversely, in the retina of Nrl −/− mice, rods are absent and photoreceptors appear cone-like; Nrl −/− animals exhibit super-normal cone function mediated by S-cones, with atypically elevated patterns of blue light detection, concomitant with deficient scotopic vision caused by rod absence. 7,[11][12][13][14] Expression of several key molecular markers are altered in Nrl −/− retinae, which supports the hypothesized model that progenitors generate S-cones by a default pathway, and that Nrl acts like a master regulator required to induce the rod differentiation pathway and suppress the cone fate. 6,10 In humans, heterozygous missense mutations in the NRL gene are associated with dominant retinitis pigmentosa phenotypes [15][16][17][18][19] ; gainof-function mutations (at codons 49,50,51,56 in the NRL transactivation domain) lead to reduced NRL phosphorylation and enhanced activation of the rhodopsin promoter. 20 Such patients present with scotopic vision deficits at a young ages, progressing to loss of photopic response with age; clinical signs that point towards a pattern of early loss of rod photoreceptors, followed by subsequent cone cell death later in life. 19 By contrast, homozygous and compound heterozygous loss-of-function NRL mutations cause recessive enhanced S-cone syndrome (ESCS). ESCS is characterized by an abnormally increased perception of blue light stimuli, coincident with night blindness and abnormal pigmentation patterns. 21,22 The recessive NRL mutations reported in ESCS patients are often nonsense or frameshift mutations. [23][24][25][26] Unaffected relatives carrying such mutations indicate that heterozygous NRL loss-of-function is not pathogenic. [24][25][26] has also been frequently associated with NR2E3 mutation. Histological analysis performed on a postmortem retina from a 77-year-old individual with ESCS, caused by recessive NR2E3 mutation (previously diagnosed as retinitis pigmentosa) showed degeneration, lack of rods, and increased numbers of S-opsin positive cells and reduced L/Mopsin positive cells. 27 Similar analysis of NRL-deficient human retina has not been performed.
Because of its apparent importance in rod generation, models to study NRL function in human development are needed. Retinal organoids generated from human pluripotent stem cells represent an invaluable tool to model human retinal development and pathologies, as they are able to recapitulate typical features of retinogenesis such as an organized multilayered tissue structure, and long human developmental differentiation times. [28][29][30][31][32] In the present study, we generated a homozygous human embryonic stem cell (ESC) line deficient for NRL, taking advantage of CRISPR/Cas9 gene editing to design and introduce a biallelic nonsense mutation into the NRL gene. To address whether NRL has a conserved role in the establishment of the human rod photoreceptor fate, and if the S-cone pathway represents a default photoreceptor differentiation pathway as it is in mouse, NRL −/− ESCs were differentiated to form retinal organoids. We analyzed cell identity at different time points, using immunohistochemistry and gene expression via quantitative PCR. We find that NRL −/− optic vesicles (OVs) exhibit a drastic reduction in expression at the RNA and protein level, not only of NRL, but also of known NRL-target genes and characteristic rod markers, particularly at late differentiation time points. Additionally, cone markers are highly upregulated, particularly S-OPSIN, defining an increased number of S-cone-like cells. Our results suggest that the rod photoreceptor identity-defining function of NRL is conserved in this human in vitro model system, and that in the absence of NRL, human photoreceptor precursors are directed toward a default S-cone-like cell fate.

| Generation of NRL −/− ESCs and retinal organoid in vitro differentiation
To generate a pluripotent ESC line deficient in NRL, we took advantage of CRISPR/Cas9 technology to target exon number 2, the first coding exon of the NRL gene. We designed a small 127 bp singlestranded donor oligonucleotide (ssODN) 33 to introduce a stop codon at a.a. position 74, as well as an EcoRI restriction site to facilitate screening ( Figure 1A). This mutation, c.221_222insAATTC p. (Trp74*) is predicted to yield a truncated version of the NRL N-terminus,

Significance statement
Photoreceptor cells located in the retina are essential for vision and their degeneration leads to a large proportion of global blindness. Rod photoreceptors needed for night vision are prevalent, whereas cone photoreceptors needed for high acuity daylight vision are rare. This study has engineered a human pluripotent embryonic stem cell line that lacks NRL gene function. When differentiated in vitro into 3D retinal organoids, the engineered organoids contain cone cells, but no rods. This study showed that NRL is required to differentiate rod photoreceptors and represents a powerful tool to generate enriched populations of cone photoreceptor cells in the laboratory.
F I G U R E 1 A, The human NRL gene comprises three exons, with the translation initiation codon located in the second exon (blue bars). Schematic shows the gene editing strategy using a 127 nucleotide single strand oligonucleotide donor to introduce a nonsense mutation in exon 2 at amino acid position 74 and a new EcoRI restriction site (in blue), 221 bp downstream of the translational start site; a specific guide RNA (gray bar) directed a simultaneous, Cas9-double strand break. B, Targeted ESCs clones were analyzed by gDNA extraction and PCR-amplification of the targeted area (indicated by arrows in A), followed by EcoRI digestion to identify donor integration. C, Sanger sequencing confirmed the correct NRL gene sequence editing. D, CRISPR-edited NRL −/− ESCs maintained a healthy morphology and growth rate, and expressed pluripotency markers OCT4, SOX2, TRA-1-60 and NANOG, similar to the parental embryonic stem cell line (not shown). DAPI, blue. Scale bar = 100 μm. E, Parental isogenic (wild type [WT]) and NRL −/− ESCs were directed toward retinal differentiation in vitro. Week 1 to 4: Optic vesicles (OVs) displayed a round 3D neuroepithelial structure over a patch of immature retinal pigmented epithelium (arrowheads). Week 4 to 7 OVs were excised and transferred to 96 well plates, to allow maturation as floating organoids. Week 7 to 14: OVs grew in size over time. Week 35 , together with the ssODN; cells were selected and recovered colonies screened using PCR amplification of the targeted region followed by EcoRI digestion, to identify the de novo restriction site ( Figure 1B). We identified one clone homozygous for the vector insertion, and confirmed its genotype by subcloning and Sanger sequencing analysis ( Figure   1C). Off target effects were not detected at three predicted off-target sites for the sgRNA and the edited line retained a normal karyotype ( Figure S1). Targeted cells were expanded, and displayed a normal growth rate and morphology and maintained pluripotency marker expression as determined by immunocytochemistry analysis (Figure 1D).
The NRL −/− ESCs and the parental ESC line (referred to as wild type [WT]) were differentiated toward three-dimensional retinal organoids using a previously published protocol. 28 Briefly, ESCs were expanded until confluent, then deprived of pluripotency-supporting factors fibroblast growth factor (FGF) and transforming growth factor beta (TGFB), followed by a period of neural induction to promote forebrain identity. After 3 weeks, round structures of organized neuroepithelium started to be visible, and were termed OVs. OVs were harvested between week 4 and week 7 and cultured individually in nonadherent 96-well plates, in retinal differentiation medium (RDM) supplemented with taurine and retinoic acid (see Supporting Information Methods). Using this protocol, we previously showed NRL localized in rod precursors by OV culture week 10, which closely mimics human fetal retinal development. 28,36,37 No obvious differences were detected between the WT and the NRL −/− lines in terms of ability to generate OVs, vesicle number, or size of vesicles ( Figure S2). OVs displayed neuroepithelial morphology as early as differentiation week 3, often growing over a patch of presumptive retinal pigmented epithelium (RPE) ( Figure 1E, arrowheads).
After isolation, the RPE tissue typically formed a small bundle of pigmented cells at the proximal end of the OVs (arrow). The vesicles continued to grow over several weeks, and in some cases developed additional lobes of neuroepithelium; sometimes more than one organized layer was visible under the light microscope ( Figure 1E

| Late stage NRL −/− OVs lack rod photoreceptors and generate S-like cones instead
To characterize the long-term differentiation of retinal organoids, some of the markers used in early samples were tested in OV cultures at week 25 of differentiation. Very few, Ki67-positive dividing cells were observed, and VSX2, which labels both retinal progenitor cells and bipolar cells, was at this stage restricted to cells under the ONL, which are presumed to be the latter cell type ( Figure 3A Other genes analyzed were GNGT1, expressed in rod outer segments, GRK7, involved in the cone photoresponse, cone-specific phosphodiesterase PDE6H and MEF2C, putative regulator of rod to cone identity (NRL −/− N = 2 differentiations w7, n = 4; w14, n = 4; w25 n = 4. WT, N = 2 differentiations; n = 4 for each time point). Samples were normalized to GAPDH and d0 (undifferentiated stem cells), plot mean ± SD. **P < .01, ***P < .001, ****P < .0001. GAPDH, glyceraldehyde 3-phosphate dehydrogenase; OV, optic vesicles; qRT-PCR, quantitative real time PCR; WT, wild type nonsense mediated decay caused by the introduced mutation. The NRL target and rod-specific transcription factor gene NR2E3 followed a similar trend, with low, nondifferential expression at week 7, and a drastic reduction in expression in NRL −/− OVs at week 14, falling to a 100-fold difference in samples by week 25. Finally, this same trend in differential expression between samples was also observed with the rod-specific photopigment gene RHO, whose expression is regulated synergistically by NRL and CRX 5,46 ; expression in NRL −/− OVs was reduced relative to controls at week 14, and this difference was highly significant by week 25.  To investigate whether the lack of NRL might have a detrimental effect on OV integrity, resulting in reactive gliosis, a common feature of retinal dystrophies, 52 we stained mature retinal organoids sections for activated retinal Müller glia cell marker, glial fibrillary acidic protein (GFAP). 53 The staining pattern was similar across the two genotypes, being mostly evident beneath the ONL and throughout the core of the organoids. No evidence of enhanced gliosis was observed in the presumptive ONL of NRL-deficient organoids compared to WT ( Figure 5D,D 0 ). To further characterize the inner cell layer of the OVs, we used SOX9, a marker of Müller glia cells, in addition to GFAP. 53,54 In both WT and NRL −/− samples, the SOX9 positive cells were located internal to the presumptive ONL, consistent with the native retinal organization of Müller glial cells ( Figure 5E,E 0 ).

| CONCLUSION
Mammalian retinogenesis is a highly conserved biological process, 55 although striking differences remain between humans and the mouse models available. The recent expansion of CRISPR/Cas9-gene editing, human iPSCs and organoid technologies has opened up an extensive platform for researchers to model a myriad of developmental and pathological processes. [56][57][58][59] Here we report the first NRL-deficient human ESC line, which has been used to investigate the role of NRL in human photoreceptor development and show that human rods share a highly conserved developmental pathway, where NRL is required to form rod photoreceptors, and in its absence increased numbers of photoreceptor precursors acquire an S-cone-like phenotype instead. This work allows the future exploration of preclinical potential therapeutic strategies using these supernumerary human cones-like cells. 60

| DISCUSSION
By generating a human ESC line with an NRL homozygous null mutation, we were able to directly study its effect on retinogenesis in vitro.
We introduced a mutation creating a nonsense mutation p. or NRL, suggesting a common genetic basis. 22,24,26,62,63 The protocol used in our study recapitulates the formation of the neural retina, which lacks a juxtaposed RPE layer, making it not very suitable to address RPE phenotypes like clumped pigmentary retinal degeneration, nevertheless we did not observe any abnormal pigmentation arising within the NRL −/− OV structures.
In mouse, the rodless Nrl −/− retina transiently develops whorls or "rosettes" in the ONL and the photoreceptor outer segments were shorter than those of the WT. 7,11 These structural changes may result from a conformational alteration due to the loss of rods, leaving supernumerary cones lacking organization and RPE contact, causing degeneration, although in a time-restricted manner, before stabilization. 11,64,65 Similar structural abnormalities were described in retinal tissue from a patient affected by NR2E3 mutations, and attributed to the excessive number of cones produced instead of rods, which pushes the tissue and produces convex folds. 11,27 In the organoid model, the loss of NRL did not cause an abnormal histological organization of the ESC-derived OVs, nor did it have an effect on epithelial polarity compared to the controls, as seen in the histological analysis using apical marker ZO-1 ( Figure 3C, C 0 ). By week 25, both NRL-deficient and control organoids lacked mature outer segment formation, but show clear evidence of nascent segment structures extending beyond the ONL ( Figure 3G-I 0 ). Occasionally, OVs presented "rosette-like" structures, with round doughnut-like organization of neuroepithelium inside of the organoid; however, we observed this phenomenon in several lines (control, edited, and other PSCs, data not shown) and therefore attributed it to the variability of the OV organization, not the NRL mutation.
We report the consequences of edited NRL loss-of-function mutation in organoids on the expression profile of several retinal genes and markers of cell fate. VSX2, expressed by bipolar cells at later time points, is reduced relative to controls. This is noteworthy for several reasons: development of bipolar cells and rod photoreceptors are closely related in mammals 66 ; a direct relationship between VSX2 and NRL has not been described, although it has been previously proposed that VSX2 represses photoreceptor differentiation in mouse. 66,67 Our human data suggest a feedback loop, whereby absence of NRL at later stages leads to reduced VSX2 gene expression. The cone rod homeobox gene, CRX, that interacts with NRL, 46 shows comparable levels of expression by week 14 and 25. The dramatically reduced expression of the NRL targets NR2E3 and RHO in the NRL-deficient organoids relative to controls indicating loss of differentiated rod cells is consistent with NRL acting as a key regulator for rod specification. Consequently, cone photoreceptor genes showed an elevated expression in NRL −/− OVs compared to controls. ARR3, RXR, and THRB are expressed earlier in development, and the latter two are known to be involved in cone specification in mouse. 68 Expression of ARR3 and RXR is higher in NRLdeficient OVs than controls at the RNA level, which is consistent with an enhanced cone identity of cells in the NRL-deficient samples. Notably, in contrast to the direct regulation of the Mef2c promoter activity by Nrl in mice suggesting its role in rod development, 69  Nrl is known to interact with cone-specific genes to repress their activity, and has been shown to bind the promoter region of ThrB, responsible for M-cone specification 9  Most features measured in Nrl −/− cells in the mouse model suggest they are healthy and functional, 11 and they connect to rod bipolar cells, implying that the type of photoreceptor connections is not pre-established based on cell type. 14 We observed that NRL −/− OVs form organized layers that express synaptic markers, similar to control organoids, and therefore it appears that the main consequence of losing functional NRL is the change in photoreceptor identity, without compromising cellular homeostasis. Our system therefore represents an in vitro human model that generates supernumerary S-cone-like cells and L/M-cones, but not rods, rather than a human retinitis pigmentosa model, associated with harmful gain of function mutations and characterized by degenerative features.
Several groups have recently explored tampering with Nrl as a therapeutic approach to stimulate the endogenous generation of cones capable of rescuing degeneration and cone-specific visual function. Using various methods to repress, eliminate or generate a loss-of-function Nrl mutation in mouse models of retinal degeneration, retinal degeneration was prevented by reprogramming rods to cones, accompanied by a recovery of ONL architecture and an increase in cone marker expression. [71][72][73][74] Nrl-deficient cone-like mouse photoreceptors have also been transplanted subretinally in a model of retinal dystrophy and shown to restore some photopic vision. 75 The culture of human cells as organoids presents an opportunity to generate virtually unlimited specific cells for transplantation therapy, provided that these are physiologically functional. Photoreceptors must be capable of establishing synapses with second order retinal cells in order to be a relevant product for cell therapy. Previously, a study using Nrl −/− mice reported that expression of the synaptic calcium channel component CACNA1F is altered in the mutant adult, but not newborn retinas, suggesting a role for Nrl in its maturation. 67 By contrast, Nrl overexpression in mouse results in cone-deficient retinas that maintain synaptic organization with expression of the postsynaptic protein PSD95. 9 In the present work, we showed that OVs at 25 weeks, equivalent to a mid-gestational stage develop distinct layers of cells that organize in a comparable manner to the human retina. The photoreceptor cells in the NRL −/− OVs showed evidence of synaptic connections by organized distribution of several key human synaptic markers (PSD98, 29,32 synaptophysin, 50 and RIBEYE 51 in a presumptive plexiform layer, suggesting the generation of cone cells capable of establishing connectivity across cell layers; Figure 5A,A 0 ,B,B 0 ,C,C 0 ). We did not detect differences in synaptic marker distribution in control and NRL-deficient OVs suggesting capacity for photoreceptor connectivity was not diminished at this stage in the cone-rich human organoids, although specific markers for human cone synapses are not available.
In summary, our study indicates the value of using genetically modified cell lines to characterize the retinal developmental events in humans and to generate new disease models for in vitro study. Modified OVs also represent a potential source of abundant cone photoreceptor cells for use in generating a cell product suitable for cone cell replacement transplantation therapy. However, clinical application would be contingent upon regulatory approval of such genetically modified human ESCs.

| METHODS
Additional details are provided in Supporting Information Methods.

| ESC culture
Human ESCs were routinely maintained on Laminin-coated 6 well plates on NutriStem medium. When at 80% to 90% confluence, cells were lifted using EDTA 0.5 mM and replated at 10 000 cells/cm 2 .

| CRISPR/Cas9 gene ablation of NRL
To generate an NRL-deficient ESC line, we designed a small asymmetric ssODN 33 to insert a STOP codon at the a.a. position 74 of NRL.
We electroporated the ssODN donor together with a plasmid containing the NRL sgRNA and a Cas9 sequence in ESCs; selected colonies were screened for the insertion of the donor sequence, and one clone homozygous for the designed mutation (NRL −/− ) was identified.

| Retinal differentiation
Both parental MShef10 and NRL −/− ESCs were subjected to retinal differentiation using a previously published protocol. 28 Briefly, cells were cultured to confluence and cultured in Neural Inductive Media until beginning to form three-dimensional OVs. These are picked before week 7 of differentiation and further cultured in RDM.

| Immunohistochemistry
OVs were fixed in a 4% paraformaldehyde (PFA) solution, and cryopreserved to generate 12 μm sections. Slides were incubated with the primary antibody O/N at 4 C, followed by secondary antibody incubation 2 hours at R/T and 4',6-diamidino-2-phenylindole (DAPI) counterstaining. Antibodies used are listed in Table S1.

| Quantitative PCR
Triplicate samples of independently cultured OVs were isolated at week 7, 14, and 25; undifferentiated ESCs were used as control. RNA was extracted using the RNeasy Micro Kit and cDNA was synthesized using the RevertAid H Minus First Strand cDNA Synthesis Kit.
qRT-PCR was performed using Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) as reference gene and Ct values were analyzed as previously described. 76 Primers sequences are in Table S2.

CONFLICT OF INTEREST
The authors declared no potential conflicts of interest.

AUTHOR CONTRIBUTIONS
E.C.: conception and design, provision of study material, collection and/or assembly of data, data analysis and interpretation, manuscript writing, final approval of manuscript; J.T.: collection and/or assembly of data; D.H.: conception and design, provision of study material, collection and/or assembly of data, manuscript writing; A.A.: collection and/or assembly of data, data analysis and interpretation; J.L.: conception and design, provision of study material; J.C.S.: conception and design, data analysis and interpretation, manuscript writing, financial support, final approval of manuscript

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.