A molecular and cellular analysis of human embryonic optic fissure closure related to the eye malformation coloboma

Ocular coloboma is a congenital eye malformation consisting of a ventrally located notch or gap in the iris, retina, choroid or optic nerve. Coloboma is often associated with microphthalmia (small under-developed eyes) (Shah et al., 2011) and accounts for 3.2 to 11.3% of cases of blindness in children (Onwochei et al., 2000). These conditions are genetically heterogeneous (Williamson and FitzPatrick, 2014; Reis and Semina, 2015; Patel and Sowden, 2017). Mutations in at least 39 human genes are associated with coloboma, although for most the number of cases identified is small. We recently showed that the majority (97%) of coloboma cases in a paediatric cohort were not explained by mutations in known genes (Patel et al., 2019). Better understanding of the genetic regulation of human eye morphogenesis will help the interpretation of genome-wide analysis of individuals with coloboma.

Human eye morphogenesis begins with the bilateral evagination of two optic vesicles from the neuroepithelium of the developing diencephalon, which invaginate asymmetrically to form bi-layered optic cups during the 5th week of gestation (O'Rahilly, 1983). The inner and outer layers differentiate into the neural retina (NR) and the retinal pigmented epithelium (RPE), respectively (Zagozewski et al., 2014). Initially the bi-layered optic cup has a fissure along its ventral aspect, known as the optic fissure, that eventually closes to create the complete eye globe. Coloboma occurs due to a failure in the process of optic fissure closure (OFC). Studies in vertebrate animal models, which have a similar but not identical eye morphogenesis to humans, have provided insights into underlying cellular and molecular processes. These studies have demonstrated that shaping of the optic cup and bringing the fissure margins into physical contact involves changing cell shape from apical to basal constriction (Nicolás-Pérez et al., 2016; Bogdanović et al., 2012), migration of cells from the outer to the inner layer (Heermann et al., 2015; Sidhaye and Norden, 2017), and different rates of cell proliferation in the two layers of the optic cup and along the dorso-ventral axis (Behesti et al., 2009; Kim et al., 2007). Invagination is preceded by patterning of the optic vesicle, based on the regional expression of transcription factors and gradients of signalling molecules (Patel and Sowden, 2017; Zagozewski et al., 2014; Fuhrmann, 2010). There is accumulating evidence of the way in which this patterning drives physical invagination (Behesti et al., 2006; Gago-Rodrigues et al., 2015).

Once the optic fissure margins (OFMs) are in contact, the neuroepithelium forming them fuses to form a continuous NR and RPE. This is one of many essential embryonic epithelial fusion processes, ranging from neural tube closure, palatogenesis and fusion of the semi-circular canal in vertebrates to dorsal closure in the Drosophila embryo, and involving the adhesion of two epithelial sheets and knitting together of their constituent cells. Epithelial remodelling typically involves cell neighbour exchange, apico-basal extension/contraction or junction matching. Cellular mechanisms directly associated with epithelial fusion include extension of actin-based filopodia and lamellipodia (Jacinto et al., 2000; Rolo et al., 2016), epithelial-to-mesenchymal transition (EMT), cell migration and apoptosis (Kim et al., 2015; Dudas et al., 2007; Fekete et al., 1997; Kobayashi et al., 2008). Yet the behaviour of the cells lining the OFMs is not well understood and has not been studied in human embryos. OFC may share cellular mechanisms with other epithelial fusion processes; however, there is one crucial difference: the cells approach each other from the basal aspect and therefore must first degrade, displace or break through the basement membrane to come in direct contact with each other. During neural tube closure, palatogenesis and semi-circular canal fusion, the epithelia approach each other from their apical aspects.

Several animal models of coloboma are characterised by a persistent basement membrane (Tsuji et al., 2012; Torres et al., 1996; See and Clagett-Dame, 2009) with the margins sometimes growing over each other (Tsuji et al., 2012; Barbieri et al., 2002). Other studies in mouse and zebrafish have suggested that cells originating from the periocular mesenchyme (POM) may play a role in basement membrane remodelling by releasing extracellular enzymes or phagocytosis (Hero, 1990; Hero et al., 1991; James et al., 2016). Electron microscopy in the mouse showed that cells lining the OFMs appear to extend cytoplasmic processes (Hero, 1990). Recent live imaging in the zebrafish reported that cells at the closure point transiently adopted protrusive morphologies (Gestri et al., 2018). Increased apoptosis has been observed in association with several closure processes. However, evidence for its role in optic fissure closure is contradictory and inconclusive (Van Nostrand et al., 2014; Lee et al., 2013; Noh et al., 2016; Gestri et al., 2018).

In summary, there are many open questions regarding cell behaviour at the OFMs. Whether and which animal model systems best represent cell behaviour in human embryos is unknown. Here, we present a new study of OFC in human embryos and examine patterns of gene expression at the OFMs using RNAseq to help resolve some of these questions and identify new disease candidate coloboma genes. We present evidence that cells lining the human optic fissure margins undergo EMT-like morphological changes at the time of fissure closure.

OFC in the human eye has previously been described to occur broadly between Carnegie stages (CS) 14 and 17 (O'Rahilly, 1983), which corresponds to 32 to 41 days post-ovulation (O'Rahilly and Muller, 1987). However, histology capturing the process of the human fissure closing to complete the eye globe is not available. Here, we performed a new histological analysis. Samples were obtained from the Human Developmental Biology Resource tissue bank (hdbr.org). Developmental ages of human embryos were approximated using morphological landmarks according to the Carnegie staging system (O'Rahilly and Muller, 1987) as more exact ages from conception cannot be determined for tissue donated from terminated pregnancies.

At CS15 (day 33), OFC had not started and POM cells were visible between the margins of the optic fissure (Fig. 1A-C; n=2 eyes from one embryo). The margins were aligned with each other but with a physical gap between them along their entire length. By CS16 (day 37) (n=3 eyes, from two embryos), the margins were in contact at a single point immediately posterior to the developing lens and closure had been initiated (Fig. 1D-F; Fig. S1A-F; see plate D). Closure at this stage only involved a few cells and had not extended across the thickness of the margins or progressed far anteriorly or posteriorly (Fig. 1G-I). Cells from the POM had been displaced from the fissure at the point of closure, but were still present between the margins anterior and posterior to this point. At CS17 (day 41), staining with an antibody against laminin (a marker of basement membrane; Byström et al., 2006; Chen et al., 2012) on serial sections (n=2 eyes from two embryos) showed the progression of fissure closure from the point of initiation. Discontinuous laminin staining was observed at the basal opposing surfaces where closure was in progress (Fig. 1K). Anterior and posterior to this, the fissure was still open with two intact basement membranes lining the margins (Fig. 1G,L, double arrows). At the extreme anterior of the optic cup, some POM cells were still present between the two margins (Fig. 1J). These analyses indicate that at CS17 fusion progresses bi-directionally toward the anterior and posterior poles from a midpoint of closure initiation. At CS18 (day 44), the fissure was open at the anterior of the cup (Fig. 1M) with an active point of closure immediately behind it, while OFC was complete closer to the point of initiation (Fig. 1N). The location of the point of initiation and directions of human OFC are shown in relation to the optic cup and anatomical planes in Fig. 1O,P.

The time scales of development in human and mouse embryos are very different. To compare gene expression in the two species, we identified time points equivalent in terms of the extent of OFC. Comparative analysis of mouse embryonic eyes based on series of histological sections of E11.5 (n=4) and E12.5 (n=3) eyes showed that at E11.5 closure had been initiated, as in human embryos, at a single point immediately posterior to the developing lens (Fig. S1G-L; see plate I). Anterior and posterior to this point, the fissure remained open (see high magnification view in Fig. S2A-C). Immunostaining for laminin showed that by E11.5 the basement membrane had begun to fragment at the point of closure (Fig. S2D-G), and by E12.5, fissure closure was nearly complete in all eyes (Fig. S2H-J). At the extreme posterior of the eyes there was a ventral indentation in the neural retina indicating where the fissure had just closed (Fig. S2J, arrow). Some marked morphological differences between species were also apparent at equivalent stages of OFC. In the mouse eye at E11.5, RPE maturation was more advanced than that in the human CS17 eye. It was already a thin mono-layer containing more pigment granules, whereas in the human it was thicker and closer in appearance to the pseudo-stratified inner neural layer of the optic cup (Fig. S2B, asterisk). In addition, in the mouse eye, the OFMs invert considerably into the fissure (Fig. S2A-C) more than in the human eye (Fig. 1K,L). These data showed that OFC occurred during day 41-44 of human embryonic development (CS17-CS18; weeks 6-7) and E11.5-12.5 in the mouse.

We sought to identify gene expression signatures confined to the small group of cells at the fissure margins that are involved in OFC. The OFMs and a corresponding control region of dorsal optic cup were isolated using laser capture microdissection from human embryonic eyes and mouse embryos (Fig. S3A). This study design aimed to identify the signature of gene expression in the OFMs and to identify those genes that were more highly expressed along the ventral (inferior) fissure compared with the opposing dorsal (superior region). As day 41-44 embryos (CS17-18 stages) have active points of epithelial re-organisation and OFC they were considered suitable for transcriptomic analyses of the process of human fissure closure. Despite the scarcity of these OFC stage human samples, high-quality RNA was successfully isolated from microdissected regions for human (n=2 embryos at CS17, n=1 embryo at CS18) and mouse (n=3 embryos, six eyes, at E11.5 and E12.5) samples, and cDNA sequencing libraries prepared for RNAseq analysis.

Expression of 11 genes with previously known expression sites was interrogated to test the validity of data generated from the laser captured RNA samples. Six of these were known to be expressed strongly in the ventral optic cup (SMOC1/Smoc1, PAX2/Pax2, ALDH1A3/Aldh1a3, VAX1/Vax1, VAX2/Vax2 and BMP7/Bmp7) (Okada et al., 2011; Torres et al., 1996; Dupe et al., 2003; Barbieri et al., 1999; Morcillo et al., 2006), and five in the dorsal optic cup (TBX2/Tbx2, TBX3/Tbx3, TBX5/Tbx5, ALDH1A1/Aldh1a1 and BMP4/Bmp4) (Behesti et al., 2006; Matt et al., 2005). Three other genes (VSX2/Vsx2, ATOH7/Atoh7 and POU4F2/Pou4f2) reported to be upregulated over time at the OFCs in mouse (Brown et al., 2009) were also included. Heatmaps for these genes showed the predicted differential expression pattern in the fissure margin and dorsal samples from both our human and mouse data (Fig. 2A,B), thereby validating the new datasets.

We next evaluated the OFC expression patterns using our recently compiled set of 39 human coloboma disease genes and 23 additional candidate disease genes identified from animal models of coloboma (Patel and Sowden, 2017). Expression analysis during OFC has not previously been performed for most of these genes. Sixty were detected in our RNAseq analysis. Notably, this excluded coloboma genes RBP4 and SHH, which are known to be expressed outside of the optic cup in early development (Chou et al., 2015; Echelard et al., 1993). Expression of the 60 detected genes was plotted as a heat map with the human coloboma disease genes highlighted in red in Fig. 2C. Similar heatmaps were also generated using their orthologous genes in the mouse expression datasets at E11.5 and E12.5 (Fig. S4A,B). As expected (Patel and Sowden, 2017), a group of the coloboma disease genes showed differential expression across the dorsal-ventral axis of the optic cup, with enriched expression in the fissure region indicating that they may be important for OFC. A second group of coloboma genes do not show ventral enrichment and are instead expressed at higher levels in the dorsal region. They include human coloboma-associated genes such as PAX6, VSX2, GDF6 and RAX, which have wider roles affecting the growth of the embryonic eye. The comparative analyses showed that more than half of the genes linked with coloboma fall into two distinct expression profile groups in both human and mouse, showing either fissure-enriched (FM) or dorsal-enriched (D, fissure-depleted) expression (CS17: FM=31.7%, D=26.7%, of 60 genes; E11.5: FM=27%, D=23.8%). These groups were smaller in the mouse at E12.5 (E12.5: FM=19%, D=9.5%). Other genes showed variable expression between samples, which may indicate a rapid temporal change in expression during fissure closure.

To identify additional genes that exhibit dorsal/fissure margin-enriched patterns of expression and therefore represent new candidate coloboma genes, a differential expression analysis was carried out (using adjusted P<0.05, Log2 FC≥1 as a cut off for significance, n=3 samples for each group). A comparison was also made between the mouse fissure margins at the two time points, as genes involved in the process of closure are likely to be dynamically regulated over this period.

In the human analysis at day 41-44 (CS17-18), 83 genes were significantly enriched at the fissure margins and 119 were significantly enriched in the dorsal optic cup (Fig. 2D). Table 1 shows the top 30 fissure- and dorsal-enriched genes in the human embryonic eyes (Fig. S5 shows a heatmap of the top 50 fissure- and dorsal-enriched genes). Table S6 shows all significantly differentially expressed genes with fold changes and P-values in all datasets analysed.

At E11.5 in the mouse, 117 genes were significantly enriched in the optic fissure margins and 171 were significantly enriched in the dorsal optic cup (Fig. 2E). At E12.5 the overall number of differentially expressed genes was smaller; 72 genes were significantly enriched at the fissure margins and 82 were significantly enriched in the dorsal optic cup (Fig. 2F). There was a partial overlap between these groups of genes: 23.9% of fissure-enriched genes at E11.5 remained fissure-enriched at E12.5; 17.5% of dorsal-enriched genes at E11.5 remained dorsal enriched at E12.5.

Tables S1 and S2 show the top 30 fissure- and dorsal-enriched genes in the mouse at the two time points along with a summary of the known functions of these genes based on gene ontology (GO) annotations. When the fissure margins at the two time points were compared, 121 genes were significantly upregulated over time and 107 were significantly downregulated (Fig. 2E). The genes showing the largest downregulation between E11.5 (beginning of OFC) and E12.5 (end of OFC) are Cplx3, Ntn1, Lum, Tmem132c and Rgs13. Those showing the largest upregulation are Cryba4, Gadd45g, Cdh4, Stk33 and Olfm1.

To perform a comparative analysis between species, the mouse genes were converted to their human orthologues and compared with the list of significantly differentially expressed genes from the human fissure margins. Eleven genes (TENM3, ALDH1A3, VAX1, BMP7, TFEC, VAX2, ZNF503, APOE, ABCA4, ATP1A2 and SMOC1) were significantly enriched in the fissure margins both in the human analysis and in the mouse analyses at either timepoint (Fig. 3A). Thirty-three genes showed significant dorsal enrichment in both species (UNC5B, EFNB1, GDF6, CTNNA2, ID1, TBX2, EFNB2, ALDH1A1, BAMBI, FGF8, LPHN2, TBX3, EDIL3, NOL4, HES5, INSM1, NRCAM, DIAPH2, GADD45A, DKK3, SULF2, TOX, BMP4, SEMA3A, DCC, ASCL1, CAMK2N1, MAB21L1, MSX2, DTX4, ID4, TBX5 and DCLK1) (Fig. 3A). Of these 44 genes, mutations in five are known causes of human coloboma (ALDH1A3, TENM3 and SMOC1 – ventrally enriched; GDF6 and MSX2 – dorsally enriched) and four are homologues of genes that, when mutated, are known to cause coloboma in animal models (BMP7, VAX2 and VAX1 – ventrally enriched; TBX2 – dorsally enriched) (Patel and Sowden, 2017; Williamson and FitzPatrick, 2014).

The identification of 44 genes showing conserved expression patterns between species was in line with the observation that the process of OFC appeared to be morphologically similar in mouse and human eyes. We used qRT-PCR to assess the expression of five of these genes (ALDH1A3, SMOC1, ATP1A2, TENM3 and TBX5) and confirmed differential expression patterns of the human genes that was consistent with the RNAseq analysis (Fig. 3B). The qRT-PCR validation indicated that the RNAseq analysis was informative. The fact that the overlap between human and mouse genes was relatively low (22%; 44/202 differentially expressed human genes identified) may be explained by the dynamic gene expression observed during the OFC process, the variability in sample age and number available for RNAseq, and the significance thresholds selected. Owing to the limited availability of human samples it was not possible to analyse additional samples or validate multiple disease candidate genes in this study; however, these new data provide a resource for future studies.

We found that the significantly enriched gene sets, in both species, were annotated with similar GO biological process terms. Fig. 3C shows Revigo semantic plots representing the most enriched GO terms; both species contained similar terms covering biological processes such as cell adhesion, axon guidance and Wnt signalling (also see Tables S3 and S4). Notably, the genes annotated in the two species were often different, which may be due to incomplete GO annotations, or because some differentially expressed genes did not pass the significance thresholds. For example, for the cadherin superfamily of cell-adhesion molecules (Paulson et al., 2014), in the mouse gene set cadherin 4 (Cdh4) and the protocadherin genes (Pcdh17 and Pcdh19) were significantly downregulated at the fissure margins, while in the human gene set, CDH6 was significantly downregulated. Cdh4, also known as r-cadherin (Babb et al., 2005, Liu et al., 2001), and Pcdh17 have been implicated in retinal differentiation (Chen et al., 2013b), whereas a closely related cadherin, Cdh2 (N-cad), has previously been implicated in coloboma, with disrupted retinal lamination, in zebrafish (Masai et al., 2003).

A complementary approach, gene set enrichment analysis (GSEA), was used to analyse signatures of biological processes in the entire transcriptomes of the fissure margins (Subramanian et al., 2005). GSEA showed conservation of pathways between the two species. The mouse early time point (E11.5) and human (CS17-18) shared the largest number of significantly positively (20) as well as negatively (4) correlated gene sets (Table S5). In addition, a number of gene sets were significantly positively correlated with the human CS17-18 fissure margins that did not show significant correlation in the mouse (Table S5).

Among the gene sets significantly correlated with OFC is the epithelial mesenchymal transition (EMT) gene set, which is of interest in relation to fissure closure because of the emerging evidence of transient changes in cell shape occurring at the fissure margins in the chick and zebrafish (Bernstein et al., 2018; Gestri et al., 2018) (Fig. 3D). The EMT gene set was positively correlated with the optic fissure region in the E11.5 mouse expression dataset (NES=1.47, P_adj=0.07) and in the CS17-18 human expression dataset (NES=1.54, P_adj=0.019) but was strongly negatively correlated in the mouse E12.5 dataset (NES=−1.78, P_adj=0.004). Enrichment plots for this gene set are shown in Fig. 3E. The E11.5 mouse and CS17-18 human datasets show an accumulation of EMT genes enriched in the fissure margins, while the E12.5 mouse dataset shows an accumulation of EMT genes enriched in the dorsal optic cup (and relatively depressed in the OFMs). In addition, three other gene sets showing a similar positive correlation with the fissure region at E11.5 in the mouse and CS17-18 in the human (mTORC1 signalling, PI3K-AKT-MTOR signalling and TNFA signalling via NFkB) represent signalling pathways known to regulate EMT in contexts other than the developing eye (Chua et al., 2007; Lamouille and Derynck, 2007; Medici and Nawshad, 2010) (Fig. 3C). Fig. 3F,G shows genes from the ‘leading edge’ of the EMT gene set in the human CS17-18 and mouse E11.5 datasets. These data suggest that a transient occurrence of an EMT-like process may facilitate OFC.

Notable examples include the following. COL3A1/Col3a1, a secreted component of the extracellular matrix (ECM) and a known marker of EMT (Zeisberg and Neilson, 2009) was enriched at the mouse and human (FC=2.4) OFMs. Col3a1 is strongly expressed in mesenchymal tissue during mouse development (Niederreither et al., 1995) and enables the correct deposition of Type I collagen (Liu et al., 1997). Type I and Type III collagens were reported to be deposited during EMT (Gonzalez and Medici, 2014). In addition, Col9a2 was dorsally enriched in the mouse and COL11A1, COL22A1, COL9A3, COL13A1 and COL25A1 were significantly dorsally enriched in the human. Expression of LUM/Lum (lumican), which encodes the leucine-rich proteoglycan ECM component lumican (Nikitovic et al., 2008) was enriched at the mouse optic fissure margins at E11.5 (FC=1.6) and strongly downregulated from E11.5 to E12.5 (FC=0.28, P_adj=1.73E⁻¹⁴). Previous studies have implicated lumican in endothelial-mesenchymal transition, cell proliferation and cell adhesion and migration (Chakravarti et al., 2000). Expression of the cadherin gene Cdh2/CDH2 was lower at the fissure margins than at the dorsal optic cup in both species. This is consistent with observations in the chick retina showing that Cdh2 expression is absent in a very distinct region around the fissure margins and in cells that delaminate from the fissure margins (Bernstein et al., 2018). However, it is contradictory to the generally accepted role of N-cadherin as a marker of the mesenchymal state (Gonzalez and Medici, 2014).

Immunohistochemistry using antibodies against ZO-1, PAX6 and vimentin was used to investigate whether cells at the closing optic fissure transiently lose their epithelial characteristics during OFC: ZO-1 is a component of tight junctions (and helps form adherens junctions), which are lost in epithelia during EMT (Gonzalez and Medici, 2014); PAX6 is a transcription factor expressed in optic cup neuroepithelium; and vimentin is a cytoskeletal protein present in mesenchymal cells and considered to be one of the essential intermediate filaments for changes in cell shape (Franke et al., 1982; Wang et al., 2018). Prior to OFC, a line of ZO-1 staining was seen along the apical surface of the folding pseudostratified neuroepithelium at the OFM (Fig. 4A, arrowheads), and extended along the apical surface of both the presumptive NR and RPE; the ZO-1 staining appeared as a thick single line in regions where the apical surfaces were closely apposed (Fig. 4B). As OFC progressed, a disorganised group of cells, no longer stained apically by ZO-1, straddled the closing fissure region (Fig. 4C, dashed box region). Counterstaining for PAX6 suggested that this group of cells delaminated from the neuroepithelium of the optic cup, and not the POM (Fig. 4D). Once OFC was complete, neuroepithelial apical-basal polarity was re-established and two continuous lines of ZO-1 staining were seen across the remodelled NR and RPE (Fig. 4E).

Although punctate vimentin staining is characteristic of mesenchymal cells but not epithelia of the early mouse embryo and is often used as a marker of EMT (Franke et al., 1982), it has also been shown to label neural precursor cell processes. We observed intense spots of vimentin staining in the POM (Fig. 4A). Vimentin also labelled the basal aspect and processes of neuroepithelium cells beyond the fissure (Fig. 4B, asterisk). At the seam where the basal surfaces of the fissure margins approach each other, vimentin puncta were present (Fig. 4B,B′, arrowheads) and as cells at the OFC point round up and lose their neuroepithelial cell shape and polarity, punctate vimentin staining continues in reorganising cells (Fig. 4C, dashed box region, magnified view in Fig. 4C′-C‴). Fluorescently labelled wheat germ agglutinin (WGA) was used to label cell membranes (Fig. 4F,G) and then Fiji-based software was used to map the relative roundness of cells from deconvolved z-stacks. Roundness heatmaps were created (Fig. 4G′, pink indicates round cells, whereas blue indicates elongated cells; see also Fig. S6) confirming that cells at the point of closure adopted a rounded morphology, while those further away were elongated and typical of a pseudostratified epithelium. Fig. S6 shows heatmaps from different regions during closure and Movies 1–4 show three-dimensional reconstructions of these cells.

Developing human eyes (n=2 individuals at CS17) were next studied in ultrastructural detail using Toluidine Blue-stained semi-thin sections and transmission electron microscopy (EM) to investigate if there was morphological evidence for an EMT-like change occurring during the process of OFC. Semi-thin sections near the point of closure showed that POM cells were displaced from in between the fissure margins (Fig. 5A). Prior to closure, the cells lining the OFMs had elongated nuclei similar to the remaining neural epithelium (Fig. S7A). However, cells at the point of closure differed in appearance from the rest of the cells in the NR or RPE. They were smaller with rounded nuclei and were not arranged in a layer, but formed a disorganised group (Fig. 5A, Fig. S7B). Where OFC had progressed further, a larger zone of cells with a rounded morphology spanning the thickness of the fissure was seen in between elongated epithelial cells at either side (Fig. 5B). EM at the point of closure showed rounded cells (yellow) alternating with elongated cells (blue) that possessed cell junctions, resembling immature tight junctions, at their apical ends (arrowhead). The rounded cells were located closer to the basal aspect of the epithelium (Fig. 5C,D, Fig. S7C). As the rounded cells alternate with elongated cells, it is unlikely that the former represent an epithelium in a different orientation (Fig. 5C).

In both eyes, some cytoplasmic material was extruded from the epithelial layer into the space between the fissure margins (Fig. 5C, arrow; Fig. S7D). The extruded material contained organelles that included mitochondria and was surrounded by a fragmented basement membrane, indicating that it originated from the neuroepithelium of the optic cup, rather than from trapped cells of the POM (Fig. S7D). It was not clear whether it included nuclei, i.e. whether whole cells or only parts of cells were being discarded. At higher magnification, cells could be seen extending cytoplasmic protrusions, which appeared to interact with cells from the opposite margin (Fig. 5E, arrowheads). These cells still maintained junctions at their apical ends (Fig. 5E, arrows). In addition, the basement membrane lining the margins appeared to break into fragments, which were seen among the rearranging cells at the point of closure. In regions where closure was complete, the cells had regained their elongated morphology and were arranged as a pseudo-stratified epithelium (Fig. 5F). Overall, the ultrastructural study suggested that the cells undergo a change in state from their original neuro-epithelial state once the POM cells are excluded and the margins come in contact with each other.

Immunostaining of the closing OFMs with phalloidin was used to determine whether the cell protrusions observed at the point of closure by EM analysis contained filamentous actin. Actin-rich foci have been recently described in the chick fissure margins (Bernstein et al., 2018). At the point of closure in an early CS17 human eye, a few cells showed strong puncta of actin staining (Fig. 6A,B, arrowheads). In the same eye in the open region of the fissure, cells in the OFMs approaching each other also showed strong puncta of actin staining (Fig. 6C,D arrowheads). At the points where the margins came in contact with each other, actin puncta were located at the basal aspects of the neuroepithelial closure points (Fig. 6E,F, arrowheads). Movie 4 shows phalloidin staining at the closing margin (green), counterstained with WGA (magenta). Cells of the neuroepithelium away from the fissure margins showed a more diffused pattern of actin staining. Antibody staining against PAX6 or PAX2 showed that the cells with punctate actin staining, displaying a mesenchymal morphology, originated from the neuroepithelium of the optic cup and not the POM (Fig. 6A,C). PAX6 was more prevalent in the anterior fissure and PAX2 more prevalent in the posterior fissure, consistent with data in the mouse (Chiang et al., 1996).

Taken together, these observations suggest that EMT-like changes in cell state occur at the human OFMs and that this is partial and transient. Eventually, as closure completes, the cells regain their epithelial organisation. There is increasing evidence in other systems for epithelial plasticity and that epithelial and mesenchymal states are not binary states (Nieto, 2013).

Evidence regarding the involvement of apoptosis in OFC from studies in animal models is contradictory. Studies have described both too much and too little apoptosis resulting in coloboma in mouse and zebrafish models (Van Nostrand et al., 2014; Lee et al., 2013; Noh et al., 2016), or that apoptosis is dispensable for fissure closure in zebrafish (Gestri et al., 2018). Here, we did not find evidence associating apoptosis with human optic fissure closure from the RNAseq or histological analysis. Using an antibody specific for apoptotic marker cleaved caspase 3 (CC3) (Nicholson et al., 1995) apoptotic cells were observed in mouse eyes at E11.5 (n=2) and E12.5 (n=4) clustering around the optic fissure, which is consistent with reports in existing literature (Ozeki et al., 2000; Noh et al., 2016). However, no apoptotic cells were observed in human eyes (n=3 eyes) at CS17. A small number of apoptotic cells were identified in the developing lens. A possible explanation for this is that apoptosis is not essential for epithelial remodelling at the human fissure but occurs in some species because of cells transiently losing their epithelial arrangement (McLaughlin et al., 2007).

In this study, we investigated for the first time the cellular organisation and gene expression during human OFC alongside a comparative analysis of the mouse eye. The morphological analysis identified post-ovulation days 41-44 (CS17-18) as the stages during which fissure closure takes place, identifying the susceptibility period during which teratogenic or genetic insults may impair OFC and cause coloboma. Fig. 7 shows a schematic representation of cell behaviour at the human optic fissure margins based on our analysis.

POM cells were cleared from the fissure before basement membrane breakdown and the initiation of OFC in human embryos. These observations are consistent with the proposal that POM cells around the optic cup secrete factors that maintain the basement membrane (Hirai et al., 1992; Ekblom et al., 1994) and their absence at the fissure contributes to basement membrane breakdown. Conversely, it has been proposed in mouse and zebrafish that cells originating from the POM contribute to basement membrane breakdown by phagocytosis or secreting enzymes (Hero, 1990; Hero et al., 1991; James et al., 2016).

Novel genes identified at the fissure margins include potential regulators of cell morphology and the actin cytoskeleton. NTN1 was one of these. In a separate study, we showed that NTN1 expression in the fusing fissure margins is conserved across multiple species and is required for successful fissure closure (Hardy et al., 2019). NTN1 is a diffusible cell signalling molecule that can induce the extension of filopodia and lamellipodia in target cells (Deiner et al., 1997). It has been shown to induce EMT along with activation of NF-κB (Yan et al., 2014; Lee et al., 2014). Both EMT and NF-κB gene sets were positively correlated with the OFC in our data.

Enrichment at the fissure of the EMT gene set (and related gene sets MTORC signalling and NF-κB signalling) was seen in the human and in the mouse at E11.5. By contrast, at E12.5 this gene set was strongly negatively correlated with the fissure margins. Histological and ultrastructural analysis showed a group of cells at the OFMs that did not have an epithelial morphology or arrangement at the closure point, whereas parallel sections (closer to the point of initiation) containing already closed fissure did not show this group of reorganising cells. Instead, the epithelium of the fused margins appeared indistinguishable from the adjacent neural retina. Although cells at the OFMs showed phalloidin staining characteristic of cell protrusions, lost ZO-1 and expressed vimentin, they did not lose expression of PAX6, a marker of neuroepithelium in the optic cup. Thus, EMT followed by mesenchymal-epithelial transition (MET) appears to underlie human OFC. This is consistent with the current understanding that cells can be plastic and move between epithelial and mesenchymal states during embryogenesis in multiple organs (Nieto, 2013; Thiery et al., 2009). Consistent with our observations, a recent study comparing the mouse fissure margin with nasal and temporal optic cup transcriptomes at E11.5 showed that the GO BP term ‘positive regulation of cell migration’ and GO MF terms related to ECM binding were strongly enriched in the fissure margin gene set (Cao et al., 2018b).

Extension of cellular protrusions were described in the mouse optic fissure (Hero, 1990) by TEM and a transient loss of epithelial morphology has been observed in the zebrafish eye (Gestri et al., 2018). A recent morphological study in the chick eye showed a group of cells delaminating from the optic fissure margins and occupying the space in between (Bernstein et al., 2018). However, there is limited evidence of the molecular regulators of this change.

Change in ECM properties, including increases in matrix stiffness, are known to facilitate changes in cell state (Greenburg and Hay, 1988), and can result from changes in matrix components secreted by cells and the secretion of extracellular enzymes. In a recent zebrafish study morpholino knockdown of Adamts16 resulted in coloboma with a failure in basement membrane breakdown (Cao et al., 2018a). We show that expression of the human orthologue ADAMTS16 (a matrix metalloproteinase; Apte, 2009) was enriched in human OFMs at CS17 (FC=10.348, P_adj=0.003). EMT is known to involve a change in the collagens secreted into the matrix (Gonzalez and Medici, 2014). We found dynamic regulation of a number of collagen genes in human and mouse at the fissure margins. Lumican is another secreted protein component of the ECM that was enriched at the fissure margins and downregulated over time in the mouse. LUM is involved in collagen fibril assembly and EMT following injury in corneal and lens epithelial cells (Nikitovic et al., 2008; Saika et al., 2003; Chakravarti et al., 2000).

We used laser capture micro-dissection to collect the entire length of the OFMs, including open, closing and closed regions. RNAseq analyses of microdissected regions from three embryos may have detected only a fraction of the genes that are differentially expressed during the dynamic OFC period. Sample variability may have led to genes that are differentially expressed transiently, or in a small number of cells, not crossing the significance threshold. If different mechanisms of closure predominate in the distal and proximal regions, such as has been suggested in the chick (Bernstein et al., 2018), this would likely be missed. As we compared gene expression across different regions of the optic cup, it is also likely that a subset of the genes we identified are players in retino-tectal mapping (Torborg and Feller, 2005). Nevertheless we found 22% of genes passing the significance thresholds in the human OF analysis showed conserved expression in the mouse OF analysis. The two species may use different mechanisms of epithelial fusion at the fissure, or use different gene regulatory pathways to achieve the same cellular changes; however, this seems unlikely as we found that key disease genes showed conserved patterns of expression. Our histological analysis confirmed active points of closure at these stages of human development. Owing to the rarity and variable quality of human embryonic tissue the number of biological replicates available for OFC analysis was limited. In future work, scRNAseq analysis may help to further elucidate different molecular cell profiles.

Human embryonic eyes were obtained from the Joint Medical Research Council UK (grant G0700089)/Wellcome Trust (grant GR082557) Human Developmental Biology Resource (http://www.hdbr.org/) with ethics approval. Samples at Carnegie Stage (CS) 15, 16, 17 and 18 were analysed. The Carnegie staging system was used to determine the age of embryo as days post ovulation based on morphological landmarks (O'Rahilly and Muller, 1987): CS15, day 33 (7-9 mm crown rump length); CS16, day 37 (11-14 mm); CS17, day 41 (11-14 mm); CS19, day 44 (13-17 mm). C57Bl6 mice were used for comparative studies of the normal mouse optic fissure. Timed matings were set up overnight and midday on the day of identifying a positive vaginal plug was considered as embryonic day 0.5. At embryonic days (E) 11.5 and 12.5, pregnant female mice were sacrificed by CO₂ inhalation followed by cervical dislocation. The uterus was removed and transferred to ice-cold PBS (phosphate-buffered saline, Oxoid BR0014G) and kept on ice before dissection.

For histology, dissected tissue was fixed in 4% w/v PFA overnight, washed in PBS (3×10 min) and dehydrated in a graded series of ethanol (Hayman) at 4°C, cleared with chloroform at room temperature, followed by incubation in paraffin at 60°C and embedded in paraffin blocks. Paraffin sections were prepared in the parasagittal plane at a thickness of 8 µm; serial sections of the eye in the plane perpendicular to the optic fissure were used to examine the process of OFC. Sections were cleared with Histo-Clear II reagent (National Diagnostics) and rehydrated with graded ethanol series (100%, 95% and 70%). Sections were then stained with Harris Haematoxylin solution (Sigma-Aldrich) for 1 min and with 1% Eosin (Sigma-Aldrich) solution for 2 min with repeated washes in distilled water. The sections were dehydrated in graded ethanol series cleared in Histo-Clear reagent II (National Diagnostics) followed by mounting with DPX mountant (Sigma-Aldrich).

For immunohistochemistry, dissected tissue was fixed in 4% w/v PFA for 1 h, washed in PBS (3×10 min), equilibrated in 30% w/v sucrose embedded in OCT medium (VWR) and snap frozen in dry ice. The frozen blocks were sectioned at 10-12 µm in the para-sagittal plane. Frozen sections were thawed and the embedding matrix was cleared with PBS at 37°C for 10 min. They were then treated with blocking solution 10% foetal bovine serum or goat serum, 1% bovine serum albumin (Sigma-Aldrich) and 0.1% Triton-X100 in PBS, for 1 h at room temperature. Primary antibody, diluted in the blocking solution, was applied overnight at 4°C (anti-laminin, 1:200, Sigma-Aldrich L9393; anti-Vim, 1:800, Dako MO725; anti-ZO-1, 1:300, Life Technologies, 402200; anti-PAX6, 1:1000, Covance-Biolegend 901301; anti-Pax2, 1:300, Biolegend 901001) Slides were washed in PBS (3×5 min). The secondary antibody (Alexa Fluor 594 goat anti-rabbit IgG or Alexa Fluor 488 donkey anti-sheep IgG) diluted 1:800 in the blocking solution was applied for 1 h at room temperature, followed by Alexa Fluor 488-conjugated phalloidin, Alexa Fluor 647-conjugated wheat germ agglutinin (WGA) and/or DAPI (4′,6-diamidino-2-phenylindole, Sigma-Aldrich) in PBS for 10 min and washing in PBS (3×10 min). The slides were mounted using Hydromount (National Diagnostics HS-106) reagent and stored at −20°C.

Fluorescence imaging was performed using a Zeiss Observer microscope with four colour fluorescence or spinning disk confocal microscope (Yokogawa, CSU22). Image analysis was carried out using Fiji/ImageJ. Z-stacks were deconvolved using five iterations of the Richardson-Lucy algorithm in DeconvolutionLab2 (Sage et al., 2017). Cell outlines were manually annotated and roundness was measured using in-built plug-ins in Fiji. Roundness heatmaps were generated by multiplying the roundness value of each cell (range 0 to 1) by 255 to derive an 8-bit colour attribution that was then applied to its pre-saved ROI. Heatmaps are displayed using Fire LUT.

Samples were fixed in 2.5% glutaraldehyde buffered in 0.1 M sodium cacodylate (pH 7.2), post-fixed in 1% osmium tetroxide and then dehydrated in ascending grades of ethanol. Tissue blocks were passed through a transitional fluid, propylene oxide, and then infiltrated overnight in Epon resin. The blocks were then transferred to BEEM capsules with fresh resin and polymerised at 60°C for 48 h. Ultrathin sections were cut with a Diatome diamond knife at 90 nm on a Leica Ultracut UCT ultramicrotome, placed on copper grids and stained with uranyl acetate and lead citrate. Examination was carried out using a JEOL 1400 transmission electron microscope and images recorded using an AMT XR80 digital camera.

Human embryonic eyes at CS17 and CS18 were dissected, and the eye along with surrounding periocular mesenchyme were embedded in OCT medium (VWR) and snap frozen on dry ice. C57Bl6 mouse embryos were collected at E11.5 and E12.5. The embryos were dissected in PBS and heads were dissected, followed by embedding in OCT medium (VWR), snap freezing on dry ice and storage at −80°C prior to sectioning. The blocks were stored at −80°C.

The embedded eyes (in the case of human embryos) or heads (in the case of mouse embryos) were sectioned in the parasagittal plane, i.e. perpendicular to the optic fissure at a thickness of 20 µm at −20°C and the sections were collected on PEN (polyethylene naphthalate) membrane-coated slides (Thermo Fisher LCM0522). Slides with sections were stored at −80°C. Slides were thawed, air dried and viewed under bright-field illumination on a laser capture microdissection microscope (PALM microbeam). The eye was orientated and optic fissure identified. The region of interest containing the fissure margins and excluding the periocular mesenchyme and lens was identified and outlined using the graphic tool of the PALM Robo software, cut out using a UV laser and catapulted with the laser into the cap of a collection tube (Zeiss, 415190-9211-000). Fissure margin regions were collected along the length of the fissure, beginning with the first section in which both margins could be clearly visualised and ending with the last section before the optic nerve head became visible. In case a section was damaged and the fissure could not be clearly identified, no tissue was collected from it. The fissure margins of the two eyes of each embryo were pooled in the same collection tube. Similarly, a control region was cut out of the dorsal optic cup and pooled using a fresh collection tube. Lysis buffer (Qiagen RNeasy Micro kit, 74004) was added to the collection tube, the caps were closed and the tube was inverted vigorously several times to rapidly lyse the tissue and prevent RNA degradation. The tube was immediately frozen on dry ice and stored at −80°C. Total RNA was extracted using standard protocols, and treated with DNAseI (Qiagen). RNA quality and concentration were determined using the Tapestation (Agilent Technologies). Only RNA samples with intact 18S and 28S bands were used for sequencing. Approximately 1.2 and 10 ng RNA was isolated per human and mouse embryo, respectively, for analysis of the fissure margin and dorsal region by RNAseq. RNA samples were amplified and cDNA libraries prepared using the SMART-Seq v4 Ultra Low Input RNA (Clontech Laboratories) and Nextera XT DNA Library Preparation (Illumina) kits, and cDNA quality was assessed using Qubit dsDNA HS Assay kit and Qubit 2.0 Fluorometer (ThermoFisher). cDNA libraries were sequenced in four lanes on the Illumina NextSeq500 platform using 43 bp paired end reads. Fissure margin and dorsal retina samples of replicate embryos within each development time point were sequenced within the same run. Sequencing yielded an average of 19 million reads (range: 13-24 million) per mouse sample and 27 million (range: 23-33 million) reads per human sample.

Following quality control, sequencing reads (FastQ) of cDNA libraries were aligned, using the Illumina RNA Seq STAR alignment tool (version 1.1.0) to the relevant reference genome (human, GRCh37/hg19; mouse, GRCm38/m10), followed by post-alignment QC. The majority of reads aligned to coding or untranslated (UTR) regions as expected (Fig. S3B,C). Further analysis was carried out in the R programming environment (version 3.3.2). Aligned reads were assigned to genes and quantified using featureCounts (Rsubread package v. 1.24.1) (Liao et al., 2014). Genes were pre-filtered based on raw counts, retaining genes that had a total of at least five reads across all samples in one analysis. Normalisation and principal component analysis was carried out using the DESeq2 (v. 1.14.1) (Love et al., 2015). A global principal component analysis (PCA) using the normalised counts showed that the human fissure margin and dorsal samples separated from each other along the first principal component of the PCA plot. Two fissure margin samples at CS17 clustered closely together, and one CS17 dorsal sample [CS17(2) D] was outlying (which may be explained by QC indicating limited 3′-bias due to degradation; Fig. S3B). The fissure margins and dorsal optic cup samples at E11.5 and E12.5, from the mouse eye also separated from each other along the first principal component of the PCA plot. (Fig. S3E). Heatmaps of normalised read counts were plotted using the pheatmap package (version 1.0.8).

Differential gene expression analysis used the DESeq2 package (v. 1.14.1). In the case of the human samples, a modification of the DESeq2 algorithm for paired analysis was used to account for the greater variability of the samples and the smaller number of replicates. A significance threshold of adjusted P-value threshold <0.05 and a log₂ fold change threshold of >1 (referred to as fissure-enriched genes) or <−1 (referred to as dorsal-enriched genes) (i.e. fold change >2 or <0.5) was applied to identify significantly differentially expressed genes. Differential expression analysis was carried out for fissure margin samples versus dorsal control samples within each time point and fissure margin samples between the two time points. Volcano plots were plotted using log₂ fold changes and adjusted P-values. Gene ontology enrichment analysis was used to annotate the significantly differentially expressed genes using the gene ontology database (http://amigo.geneontology.org/amigo/landing). A list consisting of the significantly fissure-enriched and dorsal-enriched genes from each dataset was analysed using the DAVID enrichment tool (https://david.ncifcrf.gov/summary.jsp) and the species-relevant background. The RNAseq datasets were also analysed for differential expression at the level of gene sets. Genes were ranked by log₂ fold change calculated using DESeq2, as described above. Mouse genes were converted into their human orthologues using web-based dbOrtho tool at BiodbNet (https://biodbnet-abcc.ncifcrf.gov/ and https://www.ncbi.nlm.nih.gov/gene). The list was imported into gene set enrichment software (GSEA 3.0) (Subramanian et al., 2005). GSEA evaluates small changes in a large number of genes that together may significantly influence a pathway or cell mechanism, and provides biologically meaningful results that complement simple differential expression analysis. A pre-ranked analysis was run using the 50 hallmark curated gene sets from the molecular signature database (Liberzon et al., 2015) and a false discovery rate threshold of FDR<0.25 (Subramanian et al., 2005). RNAseq data have been deposited in ArrayExpress under accession numbers E-MTAB-9426 and E-MTAB-9427.

Expression of selected genes in microdissected tissue was verified using qRT PCR. 1 ng of total RNA, treated with DNAase1 (Qiagen), was used as a template to synthesise cDNA (RevertAid H Minus First Strand cDNA Synthesis Kit, K1631). cDNA samples were diluted to the same concentration (0.5 ng/μl) with nuclease-free water. The qRT PCR procedure was carried out in 96-well plates with a total reaction volume of 20 µl per well [10 μl Power SYBR Green Master Mix, 1 μl gene specific primers (10 μm), 1 μl cDNA, 8 μl nuclease-free water]. Primers used were as follows: ALDH1A3, 5′TGAATGGCACGAATCCAAGAG3′ and 5′CACGTCGGGCTTATCTCCT3′; SMOC1, 5′AGGTCCTACGAGTCCATGTGT3′ and 5′CACTGCACCTGGGTAAAGG3′; ATP1A2, 5′GGCCGCAAATACCAAGTGGA3′ and 5′CCACACCCAGATATAGATTGTCG3′; TENM3, 5′AAGCAGACGAGTTCACTAGACA3′ and 5′TCCTCTTCGAGTTGCTGGTTC3′; TBX5, 5′CTGTGGCTAAAATTCCACGAAGT3′ and 5′GTGATCGTCGGCAGGTACAAT3′. The following PCR cycle was used for all samples: 95°C for 10 min, 40 cycles of 95°C for 15 s and 60°C for 1 min. Ct values were normalised against results for Actin (primers 5′ATAGCAACGTACATGGCTGG3′ and 5′CACCTCTACAATGAGCTGC3′).

The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health. The human embryonic and foetal material was provided by the Joint MRC/Wellcome Trust Human Developmental Biology Resource (www.hdbr.org). We thank Dale Moulding at the Imaging Facility UCL GOS ICH and Tony Brooks at UCL Genomics for technical support, the UCL Institute of Ophthalmology for use of the Laser Capture microscope, Joe Rainger and Daniel Holder for comments on the manuscript, and all staff in the Human Developmental Biology Resource.