Introduction

The generation of a self-formed, ectodermal, autonomous multi-zone (SEAM) relies on intrinsic developmental processes to guide the differentiation and self-organisation of human induced pluripotent stem cells (hiPSCs) into ocular tissues1,2. A growing two-dimensional SEAM primordium can be identified according to the progressive delineation of four concentric zones, with cellular location being indicative of lineage, which recapitulates characteristics of whole-eye morphogenesis. An in-depth characterisation of maturing ocular SEAMs has revealed spatial patterning centred around an innermost zone, Zone 1, which is enriched with cells expressing neural markers indicative of presumptive neuroectoderm. Extending radially, cells in Zone 2 closely resemble those of the neural retina and retinal pigment epithelium (RPE), while Zone 3 is predominantly occupied by ocular epithelial cells. PAX6, a master regulator for eye development3,4 is expressed in cells across Zones 1 to 3. In contrast, at the periphery of the SEAM, Zone 4 contains cells which express epithelial markers in the absence of PAX6 and likely represents general surface ectoderm. Functionally, SEAM-derived corneal epithelial cells have been isolated and expanded into sheets and these can successfully recover function when transplanted onto experimental ocular wounds1,5. Ocular surface epithelial cells isolated from Zone 3 of the SEAM have also been grown under defined conditions to form 3-dimensional organoids which are strikingly similar to native lacrimal glands. These organoids contain specialised cellular subtypes such as acinar and ductal cells, and when transplanted adjacent to the eyes of recipient rats adopt many of the defining features of functional lacrimal glands6.

Since their discovery7, induced pluripotent stem cells have garnered significant attention owing to their unique ability to differentiate into virtually any cell type, making them a powerful tool for both disease modelling and regenerative medicine. Meanwhile, the emergence of unbiased single-cell transcriptome-wide analyses has revealed unprecedented levels of phenotypic and functional heterogeneity amongst populations of cells, and this new knowledge has far-reaching implications in both basic and translational research. In the field of ophthalmology, the creation of hiPSC-derived cellular structures for transplant surgery is an emerging possibility. This is particularly significant given both the need to develop minimally invasive surgical techniques and the scarcity of donor tissue worldwide. Here, we use single-cell analyses to interrogate developing SEAMs at key junctures throughout their formation as proxies of human eye development, in order to reveal the developmental trajectories of individual cell populations as they lose pluripotency and commit to specific ocular lineages.

Results

Initial gene expression and cell lineage identity in the formative SEAM; weeks 0-1

Human iPSCs were initially grown for 10 days in StemFit medium to preserve pluripotency before being transferred into differentiation medium in order to promote SEAM formation. Cultures were closely monitored to ensure the appearance of the characteristic zones shown in Fig. 1a–c, after which SEAMs were harvested as described previously1,2. Single-cell RNA sequencing was performed using the 10x Genomics Chromium platform. Following extensive QC, data were normalised using SCTransform v28 and clustered using the in-built Seurat9 function ‘FindClusters’, which returned 12 populations (Fig. 1d). Gene expression markers for all identity classes were extracted using the FindAllMarkers function. Notably, cells at week 0 (WK0) were highly proliferative (Supplementary Fig. 1) and strongly expressed key pluripotency markers including POU5F1 (OCT4)10,11, ESRG12, MIR302CHG13 and DPPA414 (Fig. 1e). SOX2, which functions by forming a complex with POU5F1 and activating or repressing the expression of various target genes which are required for maintaining the undifferentiated states of pluripotent cells15,16, was also expressed ubiquitously at WK0, while NANOG was expressed by only a subset of cells (Fig. 1e).

Fig. 1: scRNAseq analysis of WK0 and WK1 SEAMs.
figure 1

a Phase-contrast image of a typical differentiating SEAM after 40 days of culture, showing the four characteristic zones. Scale bar: 100 μm. b Schematic depicting relationship of each zone to the developing embryonic eye. c Experimental timeline and representative phase-contrast image showing WK0 SEAMs in culture. Scale bar: 5000 μm; inset, 500 μm. d UMAP plot of WK0 SEAMs. e Feature plots showing expression of pluripotency markers. f Feature plots showing expression of early markers of ocular differentiation. g Experimental timeline and representative phase-contrast image showing WK1 SEAMs in culture. Scale bar: 5000 μm; inset, 500 μm. h UMAP plot of WK1 SEAMs. i Violin plots showing restriction of pluripotent gene expression (POU5F1) and expression of surface ectoderm markers in clusters 5 and 6. j Dot plot showing co-expression of CNN / POM markers by cells in cluster 5 and 6. k Feature plots showing segregation of pluripotent, neuroectodermal and surface ectodermal cells in WK1 SEAMs. Z1 Zone 1, Z2 Zone 2, Z3 Zone 3, Z4 Zone 4, SF StemFit medium, DM differentiation medium, CDM corneal differentiation medium, CEM corneal epithelium maintenance medium, EFTFs eye-field transcription factors, NE neuroectoderm, NC neural crest, RC retinal cells, RPE retinal pigment epithelium, NR neural retina, LE lens, OSE ocular surface ectoderm, SE surface ectoderm, CE corneal epithelium, CjE conjunctival epithelium, EK epidermal keratinocyte, PSC pluripotent stem cell, CNN cranial neural crest, POM periocular mesenchyme.

Full size image

The shift from a pluripotent state to a state of lineage commitment is marked by significant fluctuations in gene expression and the progressive weakening of pluripotent stem cell networks. During vertebrate eye development, so-called eye-field transcription factors are expressed in overlapping domains, likely beginning with the onset of expression of OTX2 in the anterior neural plate, with the eye-field transcription factors PAX6, SIX3, SIX6, RAX and LHX2 then becoming detectable in the presumptive eye-field17,18. With this in mind, of particular interest for our analyses at this stage are clusters 7, 8, and 9 (Fig. 1d). Cells in these clusters expressed ectodermal and eye-field transcription factors, whilst concurrently displaying reduced expression of pluripotency markers (Fig. 1e, f). This suggests that the intrinsic developmental pathways pertaining to anterior ectodermal specification are already underway in those cells in which pluripotency is diminishing.

After culture for 1 week in differentiation medium (WK1) (Fig. 1g), there was significant and progressive downregulation of several key pluripotency markers (Supplementary Fig. 2), and cluster analysis revealed early divergence of cellular identity towards either surface ectodermal or neuroectodermal fates (Fig. 1h). While some cells retained expression of pluripotency markers (e.g. POU5F1), keratin family members and surface ectoderm markers such as CLDN4, CLDN7, CDH1, and EPCAM were strongly expressed in clusters 5, 6 and 10 (Fig. 1i). BMP4 was also strongly expressed by cells in clusters 5 and 6 (80.16 and 86.89% of cells, respectively) and is widely known to play a crucial role in promoting ectodermal-epithelial cell lineages and inhibiting neural fate19,20. DLX5, a downstream effector of BMP4 in non-neural ectoderm21, again showed robust expression in these cells (Fig. 1i). At this stage in the differentiation protocol, the cells can be described as forming an immature ‘pre-SEAM’. Ocular surface ectoderm cells in the pre-SEAM which display corneal epithelial commitment express PAX6, whereas PAX6-negative cells are likely to represent early general (head) surface ectoderm1. In our data, this is clearly illustrated by the strong expression of PAX6 in cluster 6, but not cluster 5 (Fig. 1i). Cells in cluster 10 were also PAX6-negative, likely representing their status as pluripotent cells which are newly differentiating towards surface ectodermal fates, but before the establishment of ocular vs. non-ocular commitment.

As the pre-SEAM continues to grow and mature in culture, there is evidence to suggest the emergence of satellite spheres containing SOX10 + /p75+ neural crest cells1,2. The periocular mesenchyme, which is derived from neural crest cells which migrate into the ocular region during development, gives rise to cells of the corneal stroma and endothelium22. Several key transcription factors that are known to be important for corneal development are expressed in the periocular mesenchyme. These include PITX2 and FOXC123, along with TFAP2B, which is thought to be required for both epithelial stratification and endothelial differentiation22. In our analysis, we find that a subset of cells within the epithelial clusters 5 and 6 express markers of cranial neural crest/periocular mesenchyme, including NGFR, PITX2, FOXC1/2, SOX10 and TFAP2B (Fig. 1j). Given that formation of the cornea depends on complex bi-directional interactions between the head ectoderm and periocular mesenchyme24, it is possible that these reciprocal events are important for corneal development within the growing SEAM. The remaining cells at WK1 formed a large supercluster comprised of clusters 0-4, 7-9 & 11-12 (Fig. 1h). These cells were transcriptionally similar and expressed SOX2, which marks progenitors, alongside early neuroectodermal markers including OTX2 and CDH2. The classical ocular neuroepithelial markers PAX6, RAX, SIX3 and LHX2 were also expressed by cells in these clusters, which are quite distinct from cells in the presumptive surface epithelial clusters expressing EPCAM and CDH1, and the pluripotent cells expressing POU5F1 (Fig. 1k).

Early gene expression and cell lineage identity in emerging SEAMs; weeks 2 & 4

During eye development, contact of the distal portion of the optic vesicle with the surface head ectoderm results in the formation of the lens placode. Subsequently, it is the invagination of this non-neural placode that generates a 3-dimensional structure, comprising the lens vesicle and optic cup. We began our analysis of 2-week-old SEAMs (WK2; Fig. 2a, b) by assessing the expression of genes reported to be involved in vertebrate lens development. Cells in clusters 1, 8 and 9 were PAX6-positive and expressed early lens markers including FOXE3, PITX3 and PDGFRA25,26,27 along with the developmentally regulated enzyme crystallin BHMT28 and CRYAB, which interacts with other crystallins to help preserve lens transparency29 (Fig. 2c). Alongside the eye-field transcription factors PAX6, RAX and SIX6, cells within clusters 1, 3, 8, and 9 also expressed VSX2, a well-established marker of multipotent retinal progenitor cells (RPCs)30,31 and the retinal pigment epithelial (RPE) determinant MITF32,33 (Fig. 2d), indicating a mixed pool of progenitors.

Fig. 2: scRNAseq analysis of WK2 and WK4 SEAMs.
figure 2

a Experimental timeline and representative phase-contrast image showing WK2 SEAMs in culture. Scale bar: 5000 μm; inset, 500 μm. b UMAP plot of WK2 SEAMs. c Dot plots showing expression of lens progenitor cell markers FOXE3, BHMT, PITX3, PDGFRA and CRYAB. d Dot plots illustrating overlapping expression domains of NR markers and RPE marker MITF. e Dot plots showing expression of glial and neural markers. f Feature plots showing populations of stem (POU5F1, ABCG2) and surface ectoderm-derived epithelial cells. Elevated PAX6 expression indicates ocular lineages. g Experimental timeline and representative phase-contrast image showing WK4 SEAMs in culture. Scale bar: 5000 μm; inset, 500 μm. h UMAP plots of WK4 SEAMs. i Feature plots showing widespread expression of PAX6, SIX6 and SFRP2 and reciprocal expression of VSX2, SOX2 and RAX in developing NR compared with MITF and PMEL in RPE. j Lens marker expression in Seurat cluster 12. k Violin plots showing contribution of neuronal cell subtypes to clusters 4 and 5. Z1 Zone 1, Z2 Zone 2, Z3 Zone 3, Z4 Zone 4, SF StemFit medium, DM differentiation medium, CDM corneal differentiation medium, CEM corneal epithelium maintenance medium, PSC pluripotent stem cell, PC progenitor cell, RPC retinal progenitor cell, NR neural retina, RPE retinal pigment epithelium. ‘i‘ denotes immature cells.

Full size image

Five clusters of proliferative cells were present in the WK2 data (clusters 5, 7, 8, 9 and 11). These cells expressed markers such as TOP2A, MKI67 and UB2EC, and cell cycle analysis demonstrated that a considerable percentage of these cells were in the S and/or G2/M phases (Supplementary Fig. 1). While cells in clusters 8 and 9 co-expressed lens and retinal progenitor cell markers, the proliferative cells of clusters 5 and 7 were transcriptionally more similar to those in cluster 0, with expression patterns indicative of progression towards a neural fate. We observed increased levels of the anterior neuroectoderm and developing forebrain marker FOXG1 in these cells, along with TTYH1, which enhances NOTCH signalling and is required for the maintenance of neural stem cells34, PTPRZ1, which is expressed by oligodendrocyte precursors35, and the proteolipid PLP1, which is expressed in myelin36 (Fig. 2e). These gene expression patterns are therefore indicative of a mixed neural and glial precursor cell population. Another NOTCH-responsive factor, HES5, which can regulate both gliogenesis and neurogenesis37, was expressed across these clusters, whereas the pro-neural HES6 was expressed in cells of cluster 10, together with neuronal markers such as TUBB3 and DCX (Fig. 2e).

The remaining four clusters at WK2 (clusters 2, 4, 6 and 11) include subpopulations of cells expressing POU5F1 and the stem cell marker ABCG238,39, and clusters representing developing surface ectoderm (Fig. 2f). Cells in cluster 6 expressed PAX6 along with the keratin family members KRT8 and KRT18 and epithelial markers including EPCAM and CDH1. TP63, which regulates epithelial cell stratification in the developing limbus and cornea40,41, was expressed by a small number of cells at this stage (Fig. 2f). Cells in cluster 2 were transcriptionally similar to those in cluster 6, but expressed little or no PAX6, indicating that these represent non-ocular epithelium, while those in cluster 11 expressed proliferative markers (Supplementary Fig. 1). Members of the claudin and S100A multi-gene families, including CLDN4 and S100A11, which are known to be expressed in ocular tissues42,43, were widely expressed amongst the epithelial populations at WK2, with markers of periocular mesenchyme including TFAP2B also present in subsets of cells interspersed within the main clusters (Fig. 2f).

At WK4 (Fig. 2g, h) there was widespread, multi-cluster expression of neuroepithelial and retinal progenitor cell markers including RAX, SIX6, SOX2, VSX2, and SFRP2 (Fig. 2i) and at this developmental timepoint no pluripotent cells were detected (Supplementary Fig. 2). Cells in cluster 3 expressed MITF, DCT and PMEL, markers associated with melanin biosynthesis in the RPE44,45, in regions where VSX2 expression was lower, denoting progressive delineation of the presumptive RPE and neural retina (Fig. 2i). Highly proliferative progenitors mark cluster 6 (Supplementary Fig. 1), while cells in cluster 12 represent those of the developing lens. FindAllMarkers revealed that compared to other clusters, cells in cluster 12 differentially expressed FOXE3, CRYAB, LIM2, BHMT and PITX3 (Fig. 2j), well-documented contributors to lens development in vivo25,27,28. The mixed pool of glial and neural progenitors seen in WK2 SEAMs was likewise present at WK4. Neuronal cells in clusters 4, 5 and 11 expressed common elements such as DCX, TUBB3 and NEFM/L, but also specific markers according to their developing subtype. For example, cells in cluster 4 expressed SLC17A7 (VGLUT1) and SLC17A6 (VGLUT2), which encode vesicular glutamate transporters and specifically mark glutamatergic neurons46 while those in cluster 5 expressed the GABAergic markers GAD1 and GAD247 (Fig. 2k), along with the DLX family members DLX1, 2, 5 and 6, which regulate GABAergic neuronal specification48 (Supplementary Fig. 3). Cells in clusters 1 and 13 expressed markers indicative of populations of differentiating glial-like cells, including oligodendrocyte-like cells expressing PLP1, FABP7 and SOX3, and astrocyte-like cells expressing PAX2 and VAX1 (Supplementary Fig. 3a, b). Surprisingly, only a small number of ocular epithelial cells were observed in our analysis of WK4 SEAMs, making up cluster 14 (Fig. 2h).

Gene expression and cell lineage identity in advanced SEAMs; weeks 6–12

After culture for 6 weeks (WK6), differentiating iPSC cells are organised into a characteristic concentric multi-zone SEAM (Fig. 3a)1. Cells which reside in Zone 2 of 6-week-old SEAMs are known to represent those of the neural retina and RPE. Cluster analysis (Fig. 3b) revealed that like at WK4, there was a progressive delineation between RPE and neural retinal cells. Cells in cluster 5, and to a lesser extent, cluster 4, abundantly expressed RPE markers including MITF, DCT, TYR and TYRP1 in a pattern which was increasingly non-overlapping with markers of neural retinal progenitor cells such as VSX2, RAX and SOX2, in clusters 0 and 1 (Fig. 3c). Clusters 7 and 8 represent proliferative progenitors (Supplementary Fig. 1). Retinal progenitor cells are multipotent, generating all six types of retinal neurons and Müller glia, in an ordered, sequential process, with retinal ganglion cells (RGCs) being the first to form. Cells in cluster 12 expressed ATOH7 and POU4F2, which are required for RGC competence and differentiation49, along with the neurotrophic receptor tyrosine kinase member NTRK1 (Fig. 3d). Cells in cluster 11 expressed ATOH7, which marks RGCs in a transitional cell state of differentiation50, and VSX1, which marks cone bipolar interneurons which connect photoreceptors to RGCs51. ATOH7 also labels progenitors which give rise to early-born cone photoreceptors52. In human development, the birth of cone photoreceptors precedes that of rods, and cones can be identified by the expression of markers such as PDE6H, which labels a cone cell-specific inhibitory subunit53. Cells in cluster 10 strongly and exclusively expressed PDE6H, along with the photoreceptor cell-type marker CRX (Fig. 3d), allowing us to identify spontaneously developing photoreceptors in iPSC-derived SEAMs. Relative to other clusters, cells in cluster 9 most strongly expressed PTN, SPARCL1, LAMP5, VIM and PLP1 (Supplementary Fig. 4a, b). This glial-like transcriptional profile combined with robust expression of PAX2, VAX1 (Fig. 3d) and SOX2 led us to identify cells in this cluster as astrocyte precursor cells. Cells in clusters 3 and 6 strongly expressed neural markers in the absence of the optic markers RAX, SIX6 and VSX2. While neurons in cluster 6 expressed the GABAergic markers GAD1 and GAD2 along with DLX-family transcription factors, those in cluster 3 abundantly expressed HOX genes, most notably of the HOXB family. Expression profiles for HOXB family members (Supplementary Fig. 4c) suggest that these cells are non-ocular and instead more representative of neurons of the hindbrain or anterior spinal cord.

Fig. 3: scRNAseq analysis of WK6 and WK8 SEAMs.
figure 3

a Experimental timeline and representative phase-contrast image showing WK6 SEAMs in culture. Scale bar: 5000 μm; inset, 500 μm. b UMAP plot of WK6 SEAMs. c Violin plots illustrating progressive delineation of developing RPE (MITF, DCT, TYRP1) and NR (VSX2, RAX, SOX2. d Violin plots showing expression patterns indicative of astrocytes (PAX2, VAX1), RGCs (POU4F2, NTRK1, ATOH7), bipolar interneurons (VSX1) and cone photoreceptors (PDE6H, CRX). e Feature plots showing expression of epithelial markers. f Violin plots showing expression of corneal (KRT5), basal epithelial (GJB2, GJB6) and mucosal (KRT4, KRT7, KRT13, MUC16, S100A9) markers in epithelial subsets. g Experimental timeline and representative phase-contrast image showing WK8 SEAMs in culture. Scale bar: 5000 μm; inset, 500 μm. h UMAP plot of WK8 SEAMs indicating contribution of clusters to SEAM zones. i Feature plots showing non-overlapping domains of NR and RPE cells in the SEAM. j Feature plots showing expression patterns in subsets of epithelial cells. k Violin plots showing expression of endothelial markers in cluster 14. Z1 Zone 1, Z2 Zone 2, Z3 Zone 3, Z4 Zone 4, SF StemFit medium, DM differentiation medium, CDM corneal differentiation medium, CEM corneal epithelium maintenance medium, PSC pluripotent stem cell, epi epithelium, PC progenitor cell, RPC retinal progenitor cell, NR neural retina, RPE retinal pigment epithelium, RGC retinal ganglion cells, PR photoreceptors, MG Müller glia. ‘i‘ denotes immature cells.

Full size image

Clusters 2, 13 and 14 at WK6 contain cells derived from the surface ectoderm. These expressed epithelial markers including EPCAM, CDH1 and TP63 (Fig. 3e). Cells in clusters 2 and 13 were transcriptionally similar, and expressed KRT5, a well-characterised marker of corneal epithelium, along with basal epithelial markers GJB2 and GJB654,55 (Fig. 3f). Cells in both clusters also robustly expressed PAX6 (Fig. 3e) and were therefore annotated as basal corneal epithelial cells. Cluster 13 represents mitotic cells of this subtype, characterised by MKI67, TOP2A and UBE2C expression (Supplementary Fig. 1). CD200, which is expressed during iPSC differentiation but absent from mature corneal epithelial cell lineages56, was present at lower levels in epithelial clusters, and this pattern was maintained during the remainder of the SEAM culture protocol (Supplementary Fig. 5). Cells in cluster 14 expressed lower levels of PAX6 but showed strong expression of S100A9 and the keratins KRT4, 7 and 13, suggesting a non-ocular mucosal epithelial identity. This was corroborated by expression of the mucins MUC4 and MUC16 in these cells (Fig. 3f).

Cells from all four zones of the developing SEAM were well represented in our WK8 analyses (Fig. 3g, h). In Zone 2, expression in cluster 4 was dominated by the RPE marker MITF and factors known to be important for melanin production and RPE homeostasis44, whereas VSX2 expression was prevalent in the developing neural retina cells of cluster 0. (Fig. 3i). Cells in clusters 5 and 6 expressed mitotic markers (Supplementary Fig. 1a) together with VSX2, PAX6 and RAX (Supplementary Fig. 6a) and were annotated as proliferative RPCs. Relative to other clusters, cells in cluster 8 most strongly expressed SPP1, PLP1 and PAX2 in combination with the RPC marker VSX2 (Supplementary Fig. 6a). SPP1 is expressed by immature Müller glia cells57,58, which are the only glial cell type born from retinal progenitors, while PAX2 has previously been shown to be expressed in chick, but not mammalian, Müller glia59. Meanwhile. PLP1, a commonly used label for enteric glia60, has also been reported to be expressed by Müller glia61. It is possible, therefore, that immature Müller glia-like cells emerge at this timepoint in the SEAM. In common with WK6, there were clusters of astrocyte-like cells expressing PAX2, VAX1 and SOX2 (cluster 13), RGCs expressing POU4F2, ATOH7, STMN2, NTRK1 and NEFM (cluster 12) and photoreceptors expressing CRX, RCVRN, RXRG, PDE6H and NEUROD1 (cluster 11). OTX2, which is expressed by both photoreceptor cells and RPE and is essential for their development62, was robustly expressed by cells in clusters 11 and 4 (Supplementary Fig. 6b, c). Clusters 3 and 9 contain DCX, NEFL and NEFM-expressing neuronal cells similar to those seen in younger SEAMs, completing the Zone 1 profile. Compared with the neural retinal cells of Zone 2, cells in clusters 3 and 9 expressed much lower levels of VSX2 and RAX, and are likely non-ocular in derivation. Meanwhile, FOXG1 was expressed in both the neuronal (Zone 1) and retinal (Zone 2) clusters, reflecting its role in forebrain development and known expression in retinal tissues63 (Supplementary Fig. 6b).

Clusters 2, 7 and 10 represent cells expressing the epithelial cell markers CDH1, EPCAM and TP63. Detailed analysis of these clusters revealed widespread expression of KRT5 and 19, with expression of the mucosal keratins KRT4, 7 and 13 restricted to a sub-population of cells (Fig. 3j). Further analysis of PAX6 expression patterns suggests that cluster 10 contains a mixed population of cells, consisting of a small PAX6-positive population of cells which co-express KRT13 and likely represent conjunctival epithelium64, alongside a larger proportion of Zone 4 PAX6-negative cells belonging to non-ocular mucosa. Cells in clusters 2 and 7 were transcriptionally similar and robustly expressed PAX6 and KRT5 in the absence of KRT13, indicating a Zone 3 corneal epithelial phenotype55,64, with substantial expression of the basal corneal epithelial cell markers GJB2 and GJB6 restricted to cells in cluster 2 (Fig. 3j). The cells in the smallest cluster at this stage, cluster 14, represent a population of endothelial-like cells. These cells expressed PAX6 and high levels of COL3A1, TM4SF1, SPRR2F, GATA6 and PODXL in addition to COL8A165, a corneal endothelial marker. Expression of NPNT, which has been reported to play a role in the migration of periocular neural crest cells during chick corneal development66, and MGP, which is expressed in both the trabecular meshwork and the corneal endothelium67, was also observed in these cells (Fig. 3k).

In WK10 SEAMs (Fig. 4a, b) there was further demarcation between the NR and RPE cell clusters, with expression of RPE markers restricted to cluster 5 (Fig. 4c). Photoreceptors expressing CRX, RCVRN, RXRG, THRB and PDE6H make up cluster 14, whilst RGCs expressing POU4F2, ATOH7 and NTRK1 are found in cluster 10. PAX2 + /VAX1+ astrocytic cells are located in cluster 6, and these cells co-expressed SOX2 and glial-like markers such as PLP1, LGI468, SPARCL1 and METRN69,70. Similar to in WK8 SEAMs, a small population of cells (cluster 15) expressed the RPC marker VSX2 together with glial cell-type markers including PAX2 and PLP1, and therefore likely represent early Müller glia progenitors (Fig. 4c). At this developmental timepoint, clusters 7 and 13 are comprised of proliferative retinal progenitor cells co-expressing VSX2, SOX2, RAX and SFRP2 in combination with mitotic markers (Supplementary Fig. 1). A small number of HOX gene expressing neurons co-expressing prototypical markers including TUBB3, DCX and NEFL/M make up cluster 16 (Fig. 4c).

Fig. 4: scRNAseq analysis of WK10 and WK12 SEAMs.
figure 4

a Experimental timeline and representative phase-contrast image showing WK10 SEAMs in culture. Scale bar: 5000 μm; inset, 500 μm. b UMAP plot of WK10 SEAMs. c Dot plot showing expression profiles of specialised cells in the SEAM. d Feature plots showing expression of epithelial markers and keratin subtypes across clusters. e Violin plots illustrating expression of basal epithelium (GJB2, GJB6), lacrimal gland (LCN2, DEFB1), skin & mucosal (DSG3, CXCL17, MUC1, MUC16), mucociliary (AGR3) and migratory cell (MMP10, POSTN) markers. f Experimental timeline and representative phase-contrast image showing WK12 SEAMs in culture. Scale bar: 5000 μm; inset, 500 μm. g UMAP plot of WK12 SEAMs. h Dot plot showing expression profiles of specialised cells in the SEAM. i Feature plots showing expression of epithelial markers and keratin subtypes across clusters. j Violin plots illustrating expression of basal epithelium (GJB2, GJB6), lacrimal gland (LCN2, DEFB1), skin & mucosal (DSG3, CXCL17, MUC4, MUC16), and migratory cell (MMP10, POSTN) markers. Z1 Zone 1, Z2 Zone 2, Z3 Zone 3, Z4 Zone 4, SF StemFit medium, DM differentiation medium, CDM corneal differentiation medium, CEM corneal epithelium maintenance medium, epi epithelium, mig migratory, PC progenitor cell, RPC retinal progenitor cell, NR neural retina, RPE retinal pigment epithelium, RGC retinal ganglion cells, PR photoreceptors, ACs astrocytes, RCs retinal cells, MG Müller glia. ‘m’ denotes mature cells.

Full size image

Epithelial cell populations derived from surface ectoderm were well represented at WK10, with 2045/4260 (48%) of retained cells belonging to this lineage. Six clusters of EPCAM and CDH1-expressing epithelial cells were present, with various keratin expression profiles illustrated in Fig. 4d. Cells in cluster 1 strongly expressed PAX6, TP63, KRT5 and the basal epithelial markers GJB2 and GJB6 (Fig. 4d, e) and were annotated as basal corneal epithelial cells. Cells in cluster 8 were PAX6+ /KRT5+ and were transcriptionally similar to those in cluster 1. However, these cells also expressed markers typically associated with cell migration and corneal wound healing, including MMP1071 and POSTN72 (Fig. 4e). It is conceivable that this transcriptional profile is applicable to cells undergoing tissue remodelling or migration during development. Cells in cluster 11 expressed PAX6, the mucosal chemokine CXCL1773 and the functional lacrimal gland markers LCN2 and DEFB16. The remaining clusters (4, 9, 12) contain cells typically found in Zone 4 of the SEAM. These cells expressed lower levels of PAX6 and are likely non-ocular in derivation. Cells in clusters 4 and 9 expressed DSG3, which is found in squamous epithelium in skin and oral mucosa74. Some cells in cluster 9 expressed KRT1 and increased levels of its heterodimer partner KRT10, both of which are expressed in suprabasal keratinocytes in the skin75, while expression of the mucociliary epithelial marker AGR376, was observed in cells in cluster 12 (Fig. 4e).

In the WK12 SEAM (Fig. 4f, g), the sequential emergence of retinal neurons continued. Clusters 0, 3 and 6, for example, contain retinal progenitor cells at varying stages of maturation. Cells in cluster 7 expressed various RPE markers including RLBP1, BEST1, MITF, MLANA, SFRP5, COL8A1, WNT2B and TYR. However, we were unable to reliably detect expression of the late RPE marker RPE65. (Fig. 4h, Supplementary Fig. 7). Cells in cluster 14 strongly expressed multiple photoreceptor markers and high levels of TLCD3B, a gene recently discovered to be associated with recessive retinal dystrophy77. RGCs in cluster 10 expressed the prototypical markers POU4F2, ATOH7, SLC17A6, STMN2 and NTRK1 together with the homeobox transcription factors ONECUT1 and ONECUT2, which function during development in the specification of RGCs and horizontal cells78 (Fig. 4h). Cluster 13 represents PAX2 + /VAX1 + /SOX2+ astrocytes, while cells in cluster 15 expressed glial-like markers including SPARCL1, METRN and PTN, together with posterior HOX family members such as HOXB8 and HOXB9 (Fig. 4h).

Clusters 1 and 11 represent PAX6-positive corneal epithelium robustly expressing KRT5 but not KRT 4, 7 or 13. While cells in both clusters expressed TP63, GJB2 and GJB6, indicating a basal origin, cells in cluster 11 also expressed KRT12, which marks terminally differentiated corneal epithelial cells79, indicating a more fully developed phenotype (Fig. 4i, j). Cells in cluster 9 were KRT7 + /KRT13+ and expressed CXCL17 together with the mucosal markers MUC4 and MUC16 and the lacrimal gland markers DEFB1 and LCN2 (Fig. 4j), suggesting that these are conjunctival epithelial and lacrimal gland cells6,80. Cells in cluster 12 expressed mitotic markers (Supplementary Fig. 1) along with PAX6 and KRT5, and are presumably proliferating corneal epithelial cells, whilst those in cluster 8 had a transcriptional profile similar to the migratory cell population described in WK10 SEAMs. The putatively annotated non-ocular clusters of stratified oral mucosal and skin epithelial cells were also present at WK12, located in clusters 2 and 4. Heatmaps showing the top 5 genes expressed by cells of each of the returned clusters at each timepoint are shown in Supplementary Fig. 8 and all results generated by FindAllMarkers are available in Supplementary Data 1.

Integrative analysis of developmental trajectories

In order to follow changes in gene expression as differentiation within the SEAM progressed, individual Seurat objects were merged and normalised to correct for heterogeneity in sequencing depth. For clarity, and due to the over-representation of neuronal cells at WK4, this timepoint was excluded from the combined analyses. UMAP representations of the combined dataset were prepared according to Seurat cluster (Fig. 5a) and original timepoint identity (Supplementary Fig. 9a), and differential expression analysis was performed to reveal specific cluster markers (Supplementary Data 2). We next used Monocle3 to identify modules of co-regulated genes which may drive differentiation (Fig. 5b, Supplementary Data 3) and to explore cellular dynamics by performing trajectory and pseudotime inference81. Figure 5c shows a clear trajectory progression along two principal pathways which emerge as cells transition from a pluripotent state at WK0 into either a surface ectodermal or neuroectodermal lineage, eventually giving rise to the multiple cell types depicted in Fig. 5a. These cell clusters are shown ordered according to Monocle3 pseudotime in Fig. 5d, beginning with pluripotent cells in cluster 10 and ending with photoreceptors in cluster 21. Pseudotime values were assigned to all cells in the UMAP space (Fig. 5e), and trajectory-variable genes were collated into co-expression modules (Supplementary Fig. 9b and Supplementary Data 4). These analyses provide a reconstruction of linked developmental trajectories from SEAM transcriptomic data and additionally create a resource to identify potentially novel drivers of module and cluster-specific cellular differentiation which may underpin mechanisms of eye development.

Fig. 5: Combined data and trajectory inference.
figure 5

a UMAP representation of combined data, grouped by Seurat cluster. b Clustered heatmap showing aggregated expression of all genes in Monocle3 modules across Seurat clusters. Co-expressed genes were clustered into modules using find_gene_modules and heatmaps were generated using the ‘pheatmap’ package. c Monocle3 trajectory inference predicted by learn_graph. Surface ectodermal vs. neuroectodermal lineages are indicated by arrows. d Boxplot showing distribution of Monocle3 pseudotime values within each Seurat cluster, with clusters reordered based on their median pseudotime values. e UMAP plot with cells coloured by pseudotime. The root of the trajectory is labelled (1). SE surface ectoderm, NE neuroectoderm.

Full size image

Discussion

Here, we provide a comprehensive prolife of the transcriptomic landscape of hiPSC-derived SEAMs as they differentiate and grow in vitro as representations of human eye development. We have identified and molecularly characterised cellular populations which form within each of the concentric zones of the growing SEAM and have observed the generation of highly specialised cell types such as RGCs and photoreceptors, highlighting the intrinsic potential of this system to spontaneously generate multiple ocular tissues. Our data also reveal that two lineages emerge soon after the cells transition from a pluripotent state. Indeed, even at the initial WK0 timepoint, presumptive neural and epithelial lineage cells can be distinguished, likely because the hiPSCs have begun to partially differentiate during their initial 10-day pre-cultivation in StemFit medium. By WK1 of SEAM formation, distinct partitions representing either presumptive surface ectoderm or neuroectoderm are present. The predominance of cells with a neuroectodermal signature at this stage is in accordance with previous immunohistochemical findings showing that the innermost zone of the SEAM, the neuroectodermal Zone 1, is the first to form1. The current data indicate an underlying genomic bifurcation (i.e. surface ectoderm vs. neuroectoderm) that signifies an early commitment to distinct cellular lineages.

By WK2 of SEAM formation there is significant overlap in clusters containing progenitors of both lens and retinal cells. In vivo, the optic vesicle (which gives rise to the neural retina and RPE) and the surface ectoderm (which forms the lens) are initially in close proximity and are influenced by similar sets of signals. It is likely that as development progresses and the tissue architecture begins to establish, these progenitor cells receive more specific cues leading to their spatial and functional delineation, and this is perhaps similarly occurring during SEAM development. In support of this notion is the initial overlap in cells expressing the RPE marker MITF and the retinal progenitor cell marker VSX2. It is well established that MITF is initially expressed uniformly throughout the optic vesicle and that repression by VSX2 is a critical step in delineation of RPE and neuronal retinal territories32, and this is strikingly well conserved during SEAM development and maturation. Mixed progenitor pools of glia and neurons are also evident at this stage. During embryonic neurogenesis, pools of multipotent progenitor cells give rise to both neurons and glial cells in a carefully coordinated manner. The intricate balance between proliferation and differentiation of these progenitors is crucial for the generation of the appropriate numbers of neurons and glia, suggesting that these clusters represent a transitional state of progenitor cells within the SEAM that are likewise actively engaged in the process of fate determination.

The large proportion of cells belonging to a neural or glial identify at WK4 was initially surprising, particularly as this was accompanied by an unexpectedly small ocular epithelial cell population. However, it should be noted that during the SEAM differentiation protocol, it is at this point (i.e. WK4) that the initial differentiation medium (DM) is substituted for corneal differentiation medium (CDM)2, which contains KGF and the ROCK inhibitor Y-27632. KGF is a well characterised mitogen which specifically simulates the growth and survival of epithelial cells82, while Y-27632 has been reported to promote the proliferation of limbal epithelial cells and acceleration of corneal wound healing83. It is conceivable, therefore, that the results obtained at WK4 reflect SEAM cultures before the addition of factors which are critical for their sustained epithelial differentiation. For this reason, and in the interest of clarity, we excluded the WK4 data from our trajectory analyses. By WK6, a stage at which growing SEAMs are easily identifiable by their characteristic concentric multi-zones1,2, epithelial cell populations are well represented. Furthermore, it is possible to identify distinct populations of corneal and mucosal epithelium, and it is likely that the growth and survival of epithelial cells from this point in the differentiation protocol is reliant on the change in media to CDM, and later, to corneal epithelial maintenance medium (CEM). This is supported by the subsequent expansion of these epithelial populations and appearance of cell clusters expressing markers of functional lacrimal gland at later stages.

SOX10 + /p75+ neural crest cells have previously been identified in satellite spheres in immunohistochemical analyses1 and our analysis has revealed evidence of neural crest-derived periocular mesenchyme in immature SEAMs, consistent with this finding. Corneal endothelial cells derive from this periocular mesenchyme, and we have identified endothelial-like cells in very limited numbers in WK8 SEAMs. However, these cells are not maintained, again indicating that additional modifications to the protocol might be needed in order to support their generation and continued development. The corneal stroma is also derived from periocular mesenchyme, and although we observed widespread expression of the stromal marker DCN, there was no significant expression of the classical stromal markers KERA and LUM. Given that DCN has also been reported to be expressed in the limbus and peripheral cornea and in the skin84, it is unlikely that these cells represent stromal keratocytes, but we cannot exclude the possibility that a small population of stromal cells are found interspersed within the epithelial clusters.

As the SEAM matures, discrete clusters representing specialised cell types such as RGCs or photoreceptors become readily identifiable due to their distinct transcriptomic signatures. Furthermore, temporal specification appears to be largely conserved during SEAM growth and maturation. For example, RGCs, developmentally the earliest born of the retinal neurons, first emerge as a distinct population in WK6 SEAMs. Retinal neuron subtypes are generated in chronologically overlapping waves, and cones, which are born shortly after RGCs, are also identifiable in the SEAM at this stage. Some cell populations in our dataset are more challenging to classify. For instance, we observed a population of cells co-expressing retinal progenitor markers like VSX2 and RAX along with markers indicating a glial-like identity. These cells likely represent immature Müller glia, considering their developmental origin from retinal progenitor cells. However, it is notable that these cells express PAX2, which is not typically associated with mammalian Müller glia59, while classical Müller glia markers such as GLUL, CLU and RLBP1 were either absent or showed no specific upregulation. Overlapping expression profiles between developing Müller glia and retinal progenitor cells are well-documented85, and transcriptomic studies of hESC-derived retinal organoids indicate early emergence of Müller glia-like cell populations during retinal organoid development86. Evaluating the upregulated genes within this cluster may shed light on the mechanisms driving functional differentiation of these immature cells. We also noted the absence of some other mature cell-type-specific markers in our data. For example, in WK12 SEAMs, while we were able to detect expression of multiple prototypical RPE markers, we were unable to reliably detect expression of RPE65, which typically marks ‘late’ RPE87. Our SEAM culture protocol drives cell autonomous ocular differentiation in the absence of exogenous stimuli, but mature RPE has successfully been generated from hiPSCs by using defined media components and by using modified protocols which promote retinal organoid formation and expansion of the RPE88,89,90. Similarly, while we were able to follow the emergence and continued development of photoreceptors in our cultures, we were unable to detect expression of rhodopsin (RHO), or the cone opsins (OPN1SW, OPN1MW and OPN1LW), suggesting that these cells may not have reached maturity or that the culture conditions are sub-optimal for promoting expression of these specific genes at the timepoints studied. Indeed, while opsin-expressing photoreceptors have been generated from hiPSCs using specific protocols, detectable levels of these markers typically emerged only after the period covered by our study91,92,93. For example, it has been reported that photoreceptors derived from hiPSCs under directed pro-neural differentiation methods do not express mature markers until week 14 of culture, while in hiPSC-derived optic cups, opsin expression is not detected until week 2194,95. The absence of opsin expression in our data, therefore, could reflect intrinsic developmental timing or be because our two-dimensional adherent cell culture protocol and standard media composition does not support expression of a full complement of mature photoreceptor markers without the inclusion of growth factors to promote continued differentiation.

The UMAP projection during SEAM retinal cell maturation reveals a distinct transition from clusters characterised by RPE markers to those of a neural retinal identity. Notably, cells clustered around the border of this transition frequently expressed genes such as GJA1, OTX1, AQP1 and ZIC1, suggestive of a ciliary margin identity. Ciliary margin-like niches have previously been reported to form in hESC and hiPSC / SEAM culture systems using modified protocols88,96 but our results imply that this region has the potential to spontaneously generate independently without culture modifications. The ciliary marginal zone as a source of retinal stem and progenitor cells has been well studied in lower vertebrates such as fish, frogs and birds, although the functional presence of an analogous structure in the mammalian eye is broadly debated97. Nevertheless, this remains an interesting focus for future study and once again highlights the spontaneous co-induction of different ocular regions in the absence of exogenous factors.

In previous work, Zone 4 of the SEAM has been identified as non-ocular surface ectoderm, owing to expression of epithelial markers such as CDH1 and TP63 in the absence of PAX6. It was proposed that these cells would likely differentiate into epidermal keratinocytes1,5. P63+ cells are able to terminally differentiate into stratified epithelium of the cornea, skin, or oral mucosa20,41, with an ocular phenotype being dependent on the expression of PAX6. Oral mucosal epithelium and ocular surface epithelium share many phenotypic and morphological characteristics, and oral mucosal epithelial cell sheets have been used to reconstruct damaged ocular surface in limbal stem cell deficiency models and have shown good efficacy in clinical trials98,99,100. Our analysis has also revealed transcriptional similarity between clusters of ocular epithelial and putative oral mucosa cells and suggests that these clusters may contain a mixed population of cells. This is illustrated by the relatively low overall expression of PAX6 in the annotated conjunctival epithelial cell cluster, although it should be noted that lower levels of endogenous PAX6 expression in the conjunctiva compared with cornea have also been previously reported98. Mixed populations of conjunctival and non-ocular epithelium have additionally been identified in the same fraction following FACS analysis to isolate specific ocular surface lineages from growing SEAMs, and HOXB4 expression was seen in cells which did not express PAX61. This is mirrored in our recent data which shows expression of several HOX genes, which are not expressed at the ocular surface, in isolated cells belonging to epithelial clusters. We have also found expression of HOX genes in other groups of cells, most notably in neuronal cell clusters. HOX genes are not expressed in the retina101, so these HOX-expressing neuronal cells represent a broader neuroectodermal differentiation capacity within the SEAM. The HOX-expressing neuronal cluster at WK6, for example, shows robust and widespread expression of HOXB members HOXB2 through to HOXB8. HOXB2 is expressed in developing hindbrain rhombomeres and plays a role in patterning during neurogenesis102, while HOXB8 is expressed more posteriorly in the dorsal spinal cord103, indicating that some cells may be defaulting to a more posterior neuroectodermal developmental pathway, possibly in response to intrinsic or extrinsic cues that are yet to be defined. One potential candidate for this is retinoic acid (RA). HOX genes are long established to be direct effectors of RA signalling in embryogenesis and development, and RA signalling confers posterior identity during patterning of the developing nervous system104. RA also plays a number of roles in vertebrate eye development105, and consistent with this, our data show that the RA-metabolising enzymes RDH10 and ALDH1A1 – ALDH1A3 are expressed in developing SEAMs.

Analysis of time-series transcriptomic data from differentiating stem cells is a powerful approach to further our understanding of the dynamic and temporal changes in gene expression which shape developmental processes106,107,108. Our results here illustrate progressive development of SEAMs from iPSCs over a 12-week period and provide detailed analyses at key junctures which follow the transition from initial pluripotency though to generation of specialised ocular tissues. Further, we provide a clear snapshot of the contribution of cells from SEAMs of different ages and zones to the range of structures generated. By reconstructing developmental trajectories and extracting modules of co-regulated genes we are also able to identify less well studied candidate genes, highlighting the usefulness of these tools in examining potential drivers of specific differentiation pathways and offering important avenues for further study. While this work has allowed for us to explore, in fine molecular detail, the formation of the SEAM in vitro and the establishment of its characteristic zones, it has also revealed remarkable complexity and the innate self-directed differentiation of ocular components along divergent lineages at a transcriptional level. Further, our study not only advances our understanding of ocular development but also establishes a robust methodological framework for examining and customising SEAM development in vitro.

Methods

SEAM cell culture

SEAMs were generated from human iPS cells (clone 201B77, RIKEN BioResource Center, Tsukuba, Japan) as described previously1,2,6. Briefly, hiPSCs were maintained on 0.5 μg/cm2 LN511E8 (iMatrix-511 silk, 892021, Nippi, Tokyo, Japan) coated culture dishes in serum-free StemFit medium (AK03N, Ajinomoto, Tokyo, Japan) for at least 2 cell passages for stabilisation. Cells were harvested using dissociation solution (DS) containing 50% TrypLE Select (13567-84, Nacalai Tesque, Kyoto, Japan) and 50% 0.5 mM EDTA/PBS solution (13567-84, Nacalai Tesque) and seeded at 4500 cells per well in 6 well culture plates (353046, Corning, NY, USA) coated with 0.5 μg/cm2 LN511E8. Cells were then cultivated for a further 10 days in StemFit medium, and differentiation was initiated by culture in differentiation medium (DM) following established protocols2. After culture for 4 weeks, DM was substituted for corneal DM (CDM) containing growth factors (KGF, 112-00813, Wako Pure Chemical Corporation, Osaka, Japan) and the ROCK inhibitor Y-27632 (030-24-26, FUJIFILM Wako Pure Chemical Corporation) to promote epithelial cell growth and survival. After an additional 4 weeks of CDM culture (8 weeks in total), the medium was changed to corneal epithelial maintenance medium (CEM) consisting of DMEM/F12 (1:1; 11320033, Life Technologies, Carlsbad, CA, USA) containing KGF, Y-27632 and B-27 supplement (17504-044, Life Technologies). Medium changes were performed once every 2-3 days throughout the SEAM cell culture period. The composition of the culture medium for each stage of SEAM culture is shown in Supplementary Table 1.

Single-cell library generation and sequencing

Whole SEAMs were harvested from individual wells at eight successive intervals spanning 0 – 12 weeks (every 2 weeks, with the addition of the WK1 sample to capture very early differentiation) in the same single series experiment using methods optimised for cells at each timepoint. Specifically, cells were harvested (1 well/timepoint) using dissociation solution (DS; WK0, 4 min), TrypLE Express (13567-84, Nacalai Tesque; WK1, 10 min) or Accutase (12679-54, Nacalai Tesque; WK2, 10 min; WK4 – WK12, 60 min). For every timepoint, harvested single-cell suspensions containing cells from all zones were sorted using a FACSAriaII cytometer (BD Biosciences, Franklin Lakes, NJ, USA) and living cells were selected by 7’AAD (559925, BD Biosciences) staining. Sorted single-cell suspensions were prepared according to current 10x Genomics guidelines and single-cell libraries were prepared using Chromium Next GEM Single Cell 3ʹ Reagent Kits with v3.1 chemistry (16 rxns PN-1000127, 10x Genomics, Pleasanton, CA, USA) following the manufacturer’s protocols. Single cells were partitioned into GEMs in a Chromium Single Cell Controller at a recommended concentration for targeted recovery of 5000 cells per timepoint, and libraries were constructed according to published user guidelines. Following library construction, libraries were sequenced with 10x Genomics dual indexing on an Illumina NovaSeq 6000 platform using a high-output flow cell to obtain paired-end reads.

Single-cell data processing and analysis

Sequencing data files were transformed into single-cell gene count matrices with Cell Ranger 6.1.1 using default parameters and mRNA reads were mapped to the human reference genome GRCh38-2020-A (Supplementary Fig. 10a, b). Data pre-processing and cell cluster analyses were performed in R using Seurat v4109,110. Briefly, quality control metrics were first applied to the raw data to filter out cells with unique feature count values less than 500 or with greater than 15% mitochondrial reads (Supplementary Fig. 10c). Any additional putative low-quality cells were filtered out post-processing and heterotypic doublets were excluded using DoubletFinder111 with parameters: PCs (principal components) = 1:50, pN (defined doublet proportion) = 0.25, pK (PC neighbourhood size) = pK_choose. pK was adjusted programmatically for each dataset to provide optimal values. Data were normalised and scaled using SCTransform v28 and dimensionality reduction was performed by PCA and UMAP embedding. Clustree112 was used to assess cluster stability and a clustering resolution of 0.8 with 50 PCs was used in all downstream analyses. Differential expression analysis was performed using the FindAllMarkers function with min.pct = 0.25 and logfc.threshold = 0.25. Clusters were annotated according to differential expression profiles with reference to canonical markers, and where indicated, were correlated with respective SEAM zones by comparison with known SEAM zone markers and cell phenotypes. For the combined analysis, individual Seurat objects were integrated using the merge function, and PrepSCTFindMarkers was applied before differential expression analysis in order to correct for heterogeneity in sequencing depth. Monocle381,113 was used for trajectory and pseudotime inference. Briefly, the as.cell_data_set function from SeuratWrappers was used to convert a Seurat object into a Monocle data structure and learn_graph() was used to fit a principal graph. The root of the trajectory was determined programmatically and order_cells() was applied to order the cells in pseudotime. Differential expression analysis was performed using graph_test() with neighbor_graph = “knn”. Modules of co-regulated genes were extracted using find_gene_modules with resolution=1e-4. To analyse genes which change as a function of pseudotime, graph_test() was passed to neighbor_graph = ”principal_graph” and find_gene_modules() applied with resolution=1e-4.

Statistics and reproducibility

Statistical analyses were performed using R version 4.1.2 (The R Project for Statistical Computing) and designated packages. Data were obtained from single cells with targeted recovery of 5000 cells per time point in a single-series experiment. To ensure reproducibility, SEAMs were cultured using established protocols and closely monitored for the appearance of concentric multi-zones. Single-cell expression data confirmed the consistent generation of ocular cell types over continuous cellular states across all samples.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.