Eye development is a well-studied, highly-conserved process across humans and animal models. ¹ Consequently, many of the molecular mechanisms and genetic networks underlying eye morphogenesis have been clearly described in the literature. ^{2 , 3} In early oculogenesis, eye-field transcription factors (EFTFs) involved in the eye specification, including RAX and PAX6 are essential for optic vesicle formation at ~3-4 weeks of embryonic development and subsequent invagination to form the optic cup at ~5 weeks. Later, the visual homeobox system 2 gene ( VSX2) is required for establishing retinal progenitor cells in the presumptive optic cup. The role and regulation of these genes discovered in animal models have greatly informed our understanding of human ocular development, and how dysregulation of these factors can lead to developmental eye disorders. ²

From 2019-2021, approximately 12000 peer-reviewed journal articles investigating ocular development and disease were published, with roughly 2000 articles investigating ocular development and disease using animal models including zebrafish, Xenopus, mice, rats, dogs, cats and chicks. Approximately half of these papers investigated early ocular development using animals that could be replaced with a human stem-cell based model. The number of animals used varies greatly between studies due to a variety of experimental variables, however, the average is around 50 throughout the literature. This equates to roughly 100,000 animals used for experimental procedures, without considering those used for optimisation. To generate a CRISPR/Cas9-induced mutation, at least 750 adult zebrafish are required to test the efficiencies of the single guide RNAs (sgRNAs) and generate a mixed population of homozygous, heterozygous and wild type fish that can be genotyped. Homozygous fish can be phenotyped as an exclusive population with the introduced mutation. Additionally, any attempt to knockdown a gene in zebrafish using morpholinos requires at least 200 embryos per candidate gene. For context, in our laboratory, a total of 3040 zebrafish were used in the past year. Most experiments on typical animal models such as rats, zebrafish or chicks would be classified as non-recovery such as enucleation of the eye for histology, immunohistochemistry or RNA extraction, or as severe due to general anaesthesia and surgical techniques such as sub-retinal cell transplantation performed on the animals. There are many groups investigating ocular development and disease using animal models that would benefit from adopting a stem cell-based technique which does not compromise the scientific integrity of their disease models, in line with the principles of the 3Rs; replacement, refinement and reduction of animals in research.

Despite their invaluable contribution, animal models are suboptimal for studying human eye development and disease for several critical reasons: (i) Ocular morphogenesis differs between humans and animal models. ⁴ (ii) Divergence in developmental molecular mechanisms creates different patterns of gene expression across species. For example, MAB21L2 is required for eye morphogenesis and cell survival in the developing optic cup and lens in humans yet mab21l2 expression patterns in the chick, mouse and zebrafish are still unknown. ^{5 – 7} (iii) Mature anatomy of the eye can vary between species, such as the absence of the macula in rodents and other small mammals. ⁸ (iv) Disease phenotypes observed in animals do not always mimic those seen in humans. For instance, large phenotypic variability has been observed between Usher syndrome patients and mouse models in relation to vision loss. ⁹ Embryonic lethality described in many animal models (e.g., Sox2, Otx2, and Mab21l2 mutations in mice) is also inconsistent between species, demonstrating further variation. ¹⁰ As a result, there is a pressing need for physiologically relevant human models to study ocular development and disease. However, investigating mechanisms of early ocular malformations using human samples is impractical due to the inaccessibility of foetal tissue from 4 to 7 weeks of gestation. ¹¹ Consequently, the use of human induced pluripotent stem cells (hiPSCs) is an attractive option to overcome these difficulties.

hiPSCs are generated from somatic cells by delivery of the ‘Yamanaka’ factors, OCT4, SOX2, KLF-4 and C/L-MYC. ^{12 , 13} Overexpression of these transcription factors activates endogenous gene expression and reverts cells to a pluripotent state. ¹⁴ These pluripotent cells can then be differentiated into any cell type of interest, to study their cellular and genetic properties. While it is still unknown to what extent hiPSCs can entirely replace human embryonic stem cells (hESCs) due to the unique epigenetic signature contained in each individual line, it is important to note the distinct advantages of hiPSCs over hESCs. hiPSC use circumvents the ethical concerns associated with the creation of hESC lines from embryos as they are generated from somatic cells such as blood, urine or skin samples. ¹⁵ Additionally, cell lines can be generated directly from patients, creating a wide range of applications including in vitro disease modelling to further understanding of mutation-specific disease pathophysiology and providing targets for novel therapeutic development, while also facilitating a personalised approach as the reprogrammed cells retain all original somatic mutations. ^{16 , 17} In patients with a confirmed genetic diagnosis, CRISPR/Cas9 gene editing can be used to correct the mutation in their specific hiPSC line to demonstrate phenotypic recovery. ^{18 , 19} Gene editing can also be used to introduce a known mutation into wild type hiPSCs where patient cells are not available to create disease models with isogenic controls. ²⁰

This technology has been harnessed to generate self-organising in vitro optic cups which recapitulate ocular development long-term, initially relying on the spontaneous differentiation of EBs to neuroepithelium followed by intrinsic cellular cues driven by cell culture conditions modulating developmental signalling pathways such as Wnt and Notch. ²¹ Three-dimensional differentiation protocols are advantageous as they closely recapitulate retinal microarchitecture, generate a high percentage of retinal cells, and facilitate self-organization to mature ocular tissue with high fidelity to human eye development. ^{3 , 22} Pluripotent stem cell-derived organoids have become widely used for modelling eye development in vitro through the generation of long-term self-organising retinal organoids. ^{21 , 23} Accordingly, retinal organoids grown for upwards of 25 weeks have been used to model late-onset disorders such as Leber Congenital Amaurosis, retinitis pigmentosa and CRB1-associated retinal dystrophy affecting photoreceptor function. ^{24 – 27} These disease models have advanced therapeutics development through antisense oligonucleotides, adeno-associated virus gene delivery and CRIPSR/Cas9 correction of genetic mutations to rescue disease phenotypes in vitro. ^{25 – 27} However, there has been limited modelling of developmental eye disorders at earlier timepoints, partially due to the difficulty to recapitulate in vitro the interactions of the invaginating optic cup and surface ectoderm, with only microphthalmia and congenital hereditary endothelial dystrophy investigated using hiPSC-derived models at less than five weeks differentiation. ^{3 , 28 – 31} The combination of limited knowledge of early ocular development and disease, inaccessibility of human foetal tissue from early developmental time points and lack of replicable cellular models suggests a need to develop a robust protocol faithful to human ocular development to further investigate early eye disorders. Here we describe our highly reproducible modified protocol that can consistently generate hiPSC-derived early optic vesicles to investigate early human ocular development and disease.

In this study, we present an adapted protocol to generate optic vesicles that can recapitulate early ocular development in vitro. ³² Originally, embryoid bodies (EBs) are generated from hiPSCs to undergo retinal differentiation. ^{32 , 33} A consequent study showed mechanical dissociation of hiPSCs and stable cell culture conditions generated the optimal EBs for retinal differentiation but this method does not control for EB diameter. ²² Cowan et al. investigated the relationship between EB diameter and the efficiency of retinal differentiation to generate optic cups, suggesting a diameter of roughly 275 μm for optimal retinal differentiation. ³⁴ Here the use of the Aggrewell ^® plate allows for stringent control of EB size for optimal retinal differentiation.

Additionally, we present an alternative to animal models by describing an efficient adapted protocol for the generation of stem cell-derived optic vesicle-like models to elucidate key pathways regulating early eye morphogenesis and identify molecular disruptions underlying developmental eye disorders such as microphthalmia, anophthalmia and aniridia that arise in the first six weeks of ocular development. This protocol will help developmental biologists and geneticists working with animals consider replacement with a more faithful and physiologically relevant model of human eye development. This shift will significantly reduce the use of animals in the study of ocular development and disease.

In this study, we report the generation of in vitro optic vesicles using a protocol adapted from Mellough et al. and Chichagova et al. ^{32 , 33} We describe distinct morphological changes displayed by developing optic vesicles by day 35 and the expression of early ocular markers PAX6 , RAX, OTX2, VSX2, MITF and SOX2 measured by RT-qPCR and immunohistochemistry. The detection and expression patterns of these markers highlight extensive laminar formation by day 35, as described in previous studies, suggesting high replicability of this protocol to generate hiPSC-derived ocular vesicles faithful to human eye development. ^{22 , 32 , 38 – 40} Importantly, we detect the expression of more mature photoreceptor and retinal ganglion cell progenitor markers such as CRX, OTX2 and BRN3b by day 50, suggesting this protocol can generate retinal lineage-specific organoids. Similar expression patterns have been detected in human foetal tissue (HFT), suggesting a high fidelity between in vitro optic vesicles and HFT. ^{40 , 41}

Capowski et al. suggest the emergence of laminar morphology in differentiating vesicles, as observed in our study, is a key tool to identifying retinal organoids at approximately day 35. ⁴² Additionally, those structures ubiquitously expressed early ocular markers such as VSX2 suggesting a close link between morphological and molecular changes during early optic vesicle differentiation. ⁴² For long-term studies, the identification of mature cell types and retinal microarchitecture using immunohistochemistry would validate the efficacy of this protocol to generate rod and cone photoreceptors, retinal ganglion, horizontal and bipolar cells complemented with Müller glia organised into a laminar structure that contains three neuronal cells layers connected by two plexiform layers. ⁴³ This striking cellular structure of a human retina in vivo is recapitulated in retinal organoids and is indicative of mature retinal cell formation. Electroretinography, which provides an important functional readout for photoreceptors and retinal ganglion cells similar to the human retina, can be measured in vitro or in vivo upon transplantation into animal models. ^{44 , 45}

Previous studies reported that an EB diameter of roughly 275 μm as well as enzymatic rather than mechanical passage of hiPSCs prior to EB formation are optimal for most efficient optic vesicle differentiation. ^{22 , 34} Our study has shown differentiation of EBs roughly 160 μm in diameter, suggesting that the optimal EB diameter for retinal differentiation may differ between cell lines. The standardised generation of EBs of similar size is particularly difficult yet the low variance in diameter we observed suggests our protocol has improved the standardisation of EB size optimal for differentiation to optic vesicles. Although the use of Aggrewell ^® plates can standardize the size of EBs in the beginning of the process, it is much more difficult to control the size of actual vesicles differentiating in culture. This is not something we successfully achieved in this study, nor has this been documented in previous studies. Additionally, our data shows large variability in vesicle diameter at day 35, with a range of 383.80 μm-1119.29 μm reported in vesicles derived from healthy controls. Furthermore, in our study, the standard deviation in vesicle diameter at day 35 was 192.7 μm which did not significantly differ from standard deviations reported in vesicle diameter by Guo et al., suggesting vesicle size is difficult to regulate in vitro independently of the differentiation protocol. ⁴⁶ This is significant as vesicle size can impede a response to therapeutics. ⁴⁷ Novel improvements to protocols that will standardize vesicle size over the timecourse of differentiation will enhance therapeutics testing in these models and reduce variability between samples and groups.

Two-dimensional in vitro cellular models as well as animal models were the initial widespread platform for basic science research and pre-clinical therapy testing. However, the lack of a three-dimensional microenvironment and mechanical cues to guide differentiation, and the aforementioned disadvantages of animal models, reduce the relevance of these models to human disease. ⁴⁸ Although here we have described the generation of in vitro optic vesicles to model early ocular development and disease, the original protocol has generated complex photoreceptor-like cells differentiated for upwards of 22 weeks, thus can efficiently be expanded for more complex modelling of later stages of eye development and later-onset retinal diseases. ^{33 , 49}

A considerable disadvantage to hiPSC-derived optic vesicle generation is the large variability observed both between differentiation of different hiPSC lines and also between rounds of differentiation of the same hiPSC clone; thus many lines should be screened to ascertain their differentiation capacity. ^{3 , 34} Biomarkers predicting the differentiation efficiency are particularly useful such as elevated levels of epigenetic marker H3K4me2 or downregulated Meis1 transcript levels. ⁵⁰ However, pre-screening methods are labour intensive and extremely costly, particularly following the expense, time and expertise already required for hiPSC reprogramming and characterisation. ³⁴ Therefore, further technical optimisations are required to increase differentiation efficiency as well as complexity of these model systems.

Nonetheless, the use of hiPSC-derived optic vesicles and retinal organoids can greatly reduce animal use when investigating ocular development and disease. In 2019/20, approximately ~75 original research papers were published using iPSC models of ocular development or disease. This is much lower in comparison to the ~750 original research papers still using animal models. The uptake of this method amongst animal researchers would greatly increase the number of iPSC-based papers while simultaneously effectively reducing animal model experimentation due to the greater accessibility of a human-derived physiologically faithful model of human ocular development. ³ This effect has been observed locally as our group has not created zebrafish models for early developmental disorders that have been modelled using stem-cell derived optic vesicles. As mentioned above, each zebrafish knockout line requires approximately 750 fish per mutation created by CRISPR/Cas9 gene editing and 250 per morpholino knockout. The method described here has greatly reduced the number of zebrafish used in our research by at least 750 fish per patient cell line. This would contribute greatly to the 3Rs aim of replacing animal models.

Further advancements to the methodology described here, such as retina-on-a-chip and retinal differentiation using bioreactors have enhanced in vitro differentiation to reduce variability of organoid models, and more closely recapitulate native retina physiology and the human embryological environment. ^{51 , 52} The combination of in vitro retinal tissue generated by this protocol with either a microchip containing a flow system mimicking blood flow or growth in a bioreactor more closely resembling the environment for embryonic development in utero, will enhance the modelling of ocular development and disease at both early and late developmental stages. Research is ongoing to create more complex in vitro models recapitulating the anatomy of the eye, such as a photoreceptor/RPE/choroid complex using patient derived cells.

In this study, we report an adapted protocol for the generation of hiPSC-derived optic vesicles that faithfully recapitulates early human eye development. The improved standardisation of EB generation was a key adaptation of this protocol to enhance efficiency of optic vesicle generation. Next-generation sequencing techniques and future omics studies will provide novel insights into early eye development, further understanding ocular maldevelopment that occur in diseases like microphthalmia/anophthalmia or coloboma, which currently have no treatment. This model is an exciting development for further understanding of human ocular development and disease and can be an important pre-clinical platform for the development of novel therapeutics.