Compensatory mechanisms render Tcf7l1a dispensable for eye formation despite its cell-autonomous requirement in eye field specification

The vertebrate eye originates from the eyefield, a domain of cells specified by a small number of transcription factors. In this study, we show that Tcf7la is one such transcription factor that acts cell-autonomously to specify the eye field in zebrafish. Despite the much reduced eyefield in tcf7l1a mutants, these fish develop normal eyes revealing a striking ability of the eye to recover from a severe early phenotype. This robustness is not mediated through compensation by paralogous genes; instead, the smaller optic vesicle of tcf7l1a mutants shows delayed neurogenesis and continues to grow until it achieves approximately normal size. Although the developing eye is robust to the lack of Tcf7l1a function, it is sensitised to the effects of additional mutations. In support of this, a forward genetic screen identified mutations in hesx1, cct5 and gdf6a, which give synthetically enhanced eye specification or growth phenotypes when in combination with the tcf7l1a mutation.


Introduction
The paired optic vesicles originate from the eye field, a single, coherent group of cells located in the anterior neural plate (Cavodeassi 2018). During early neural development, the specification and relative sizes of prospective forebrain territories, including the eye field, depend on the activity of the Wnt/β-Catenin and other signalling pathways (Beccari et al., 2013;Cavodeassi 2014;Wilson and Houart, 2004). Enhanced Wnt/β-Catenin activity leads to embryos with small or no eyes (Cavodeassi et al., 2005;Kim et al., 2000;Heisenberg at al., 2001;Houart et al., 2002). In contrast, decreasing activity of Wnt/β-Catenin signalling generates embryos with bigger forebrain and eyes (Cavodeassi et al., 2005;Glinka et al., 1998;Lekven et al., 2001;Houart et al., 2002). Although much research has focused on the molecular mechanisms involved in the specification of the eye field, little is known about what happens to the eyes if eye field size is disrupted.
A number of genes have been identified as encoding a transcription factor network that specifies the eye field (Beccari et al 2013;Zuber et al. 2003). These genes have been defined based on conserved cross species expression patterns in the anterior neuroectoderm and on phenotypes observed when overexpressed or when function is compromised (Beccari et al., 2013). Perhaps surprisingly, to date there are relatively few mutations that lead to complete loss of eyes suggesting that early stages of eye development are robust to compromised function of genes involved in eye development.
Indeed, in humans, eye phenotypes are often highly variable in terms of penetrance and expressivity even between left and right eyes (Reis and Semina, 2015;Williamson and FitzPatrick, 2014). This again raises the possibility that the developing eye is robust and can sometimes cope with mutations in genes involved in eye formation.
Genetic robustness is the capacity of organisms to withstand mutations, such that they show little or no phenotype, or compromised viability (Felix and Barkoulas, 2015;Wagner, 2005). This inherent property of biological systems is wired in the genetic and proteomic interactomes and enhances the chance of viability of individuals in the face of mutations.
High throughput reverse mutagenesis projects and the emergence of CRISPR/Cas9 gene editing techniques have highlighted the fact that homozygous loss of function mutations in many genes generate viable mutants with no overt phenotype (Varshney et al., 2015;Dickinson et al., 2016;Meehan et al., 2017). Across phyla, mutations in single genes are more likely to give rise to viable organisms than to show overt or lethal phenotypes. For instance, it is estimated that zygotic homozygous null mutations in just ~7% of zebrafish genes compromise viability before 5 days post fertilisation (Kettleborough et al., 2013) and 8-10% between day 5 and 3 months (Shawn Burgess, personal communication); and compromised viability is predicted following loss of function for about 35% of mice genes (Dickinson et al., 2016;Meehan et al., 2017). Furthermore, apparently healthy viable homozygous or compound heterozygous 'gene knockouts' have been found for 1171 genes in the Icelandic human population (Sulem et al., 2015) and for 1317 genes in the Pakistani population (Saleheen et al., 2017).
In some cases, the lack of overt phenotype may be due to redundancy in gene function based on functional compensation by paralogous or related genes (Barshir et al., 2018, Hurles, 2004, Wagner, 1996. We can assume that genes that do not express a phenotype when mutated are not lost to genetic drift because in some way they enhance the fitness of the species. For instance, even though two paralogous Lefty genes encoding Nodal signalling feedback effectors have been shown to be dispensable for survival, they do make embryonic development robust to signalling noise and perturbation (Rogers et al., 2017).
Genetic compensation for deleterious mutations is a cross-species feature (El-Brolosy and Stanier, 2017), and mRNAs that undergo nonsense-mediated decay due to mutations that lead to premature termination codons can upregulate the expression of paralogous and other related genes (El-Brolosy et al., 2018). However, only a fraction of genes have paralogues and other compensatory mechanisms must contribute to the ability of the embryo to cope with potentially deleterious mutations. One such mechanism is distributed robustness, which can emerge in gene regulatory networks (Wagner, 2005). This kind of robustness relies on the ability of the network to regulate the expression of genes and/or the activity of proteins within the network, such that homeostasis is preserved when one of its components is compromised (Davidson, 2010;Peter and Davidson, 2016).
Maternal-zygotic tcf7l1a mutant zebrafish have been previously described as lacking eyes (Kim et al., 2000). In this study, we show that expression of this phenotype is dependent on the genetic background. We find that tcf7l1a mutants can develop functional eyes and are viable, and that this is not due to compensatory upregulation of other lef/tcf genes.
Despite the presence of functional eyes, the eye field in tcf7l1a mutants is only half the size of the eye field of wildtype embryos, indicating an early requirement for tcf7l1a during eye field specification. We further show that this requirement is cell autonomous, revealing a striking dissociation between the early role and requirement for Tcf71a in eye field specification and the later absence of an overt eye phenotype. Subsequent to compromised eye field specification, tcf7l1a mutant eyes recover their size by delaying neurogenesis and prolonging growth in comparison to wildtype eyes. This compensatory ability of the developing eye was also observed when cells are removed from the wild type optic vesicles. All together, our study suggests that the loss of Tcf7l1a does not trigger any genetic compensation or signalling pathway changes that restore eye field specification; instead, the developing optic vesicle shows a remarkable ability to subsequently modulate its development to compensate for the early, severe loss of eye field progenitors.
The penetrance and expressivity of eye phenotypes appears to be dependent on complex genetic and environmental interactions (Gestri et al. 2009;Kaukonen et al., 2018;Prokudin et al., 2014). Thus, we speculated that Tcf7l1a mutant eyes may be sensitised to the effects of additional mutations. Here we show this is indeed the case and describe the isolation of three mutations from a recessive synthetic modifier screen in tcf7l1a homozygous mutant zebrafish that lead to enhanced/novel eye phenotypes when in combination with loss of tcf7la function.
In summary, our work shows that zebrafish eye development is robust to the effects of a mutation in tcf7l1a due to growth compensatory mechanisms that may link eye size and neurogenesis. Our study adds to a growing body of research revealing a variety of mechanisms by which the developing embryo copes with the effects of deleterious genetic mutations.

Results
The tcf7l1a m881/m881 mutation is fully penetrant but maternal-zygotic mutants show no overt eye phenotype and are viable The headless (hdl) m88 mutation in tcf7l1a (tcf7l1a -/from here onwards) was identified because embryos lacking maternal and zygotic (MZ) gene function lacked eyes (Kim et al., 2000). However, no overt defects were observed in zygotic (Z) tcf7l1a -/mutants, due to functional redundancy with the paralogous tcf7l1b gene (Dorsky et al., 2003). In our facility, MZtcf7l1a -/embryos showed a variable eye phenotype, ranging from eyeless, to small and overtly normal eyes, with proportions that varied in clutches from different pairs of fish (not shown). We hypothesised that genetic background effects could be responsible for either enhancing or suppressing the eyeless phenotype. To test this idea, we outcrossed tcf7l1a -/fish to ekkwill (EKW) wildtype fish and identified tcf7l1 +/carriers by PCR genotyping. After three generations of outcrossing to EKW fish, we incrossed tcf7l1 +/carriers to grow Ztcf7l1a -/adults. All MZtcf7l1a -/embryos coming from six pairings of Ztcf7l1a -/mutant fish developed eyes only slightly reduced in size compared to eyes of wildtype embryos of the same EKW strain (100%, n>100; Fig.1A,B).
The tcf7l1a m881 mutation creates a splice acceptor site in intron 7, which leads to a 7 nucleotide insertion in tcf7l1a mRNA that gives rise to a truncated protein due to a premature termination codon (Kim et al., 2000). Given that the wildtype splice site in intron 7 is still present in tcf7l1a mutants, we assessed whether the lack of phenotype in MZtcf7l1a -/mutants could be due to incomplete molecular penetrance as a result of expression of mRNA from both wildtype and mutant splice sites. The chromatogram sequence of the RT-PCR product amplifying exons 7 and 8 in wildtype, mutant and heterozygous embryos shows that only wildtype tcf7l1a mRNA is detected in wildtype embryos and only mutant mRNA containing the 7 nucleotide insertion is observed in mutants, while heterozygous embryos produce both wildtype and mutant mRNAs ( Fig.S1; Kim et al., 2000). This suggests that the mutant splice site is the only one used in tcf7l1a -/embryos. In addition, while overexpression of wildtype tcf7l1a mRNA rescues eye formation in embryos in which both tcf7l1a and tcf7l1b are knocked down, tcf7l1a m881 mutant mRNA does not, confirming that protein arising from the tcf7l1a m881 allele is not functional (not shown;Kim et at., 2000). These observations suggest that the m881 allele is indeed a null mutation and that tcf7l1a is not essential for eye formation.
Supporting a requirement for tcf7l1a to form eyes, antisense morpholino knockdown of tcf7l1a (mo1 tcf7l1a ) leads to eyeless embryos (Dorsky et al., 2003) comparable to the originally described headless MZtcf7l1a -/mutant phenotype (Kim et al. 2000). However, the target site for the morpholino used in that study shows considerable sequence homology to the translation start ATG region of other tcf gene family members (56-76%; Fig.S2A). This suggests that the mo1 tcf7l1a phenotype may be due to the morpholino knocking down expression of other tcf genes, as has been described for other morpholinos targeting paralogous genes (Kamachi et al., 2008). Indeed, injection of a different tcf7l1a morpholino (mo2 tcf7l1a ) with low homology to other tcf genes (36-45%, Fig.S2B) does not lead to an eyeless phenotype (0.4pMol/embryo, 100%, n>100; Fig.1C ,D). tcf7l1b morpholino injection on its own shows no overt phenotype (Dorsky et al., 2003) but co-injection of mo2 tcf7l1a and mo tcf7l1b gives rise to eyeless embryos (each at 0.2pMol/embryo, 78.26%, n=92; Fig.1E and see Dorsky et al., 2003). This suggests that even though mo2 tcf7l1a injection alone results in no phenotype, the morpholino does knockdown tcf7l1a.
Together, these results suggest that even though tcf7l1a -/is a fully penetrant null mutation, lack of maternal and zygotic tcf7l1a function alone does not lead to loss of eyes in all genetic backgrounds. tcf7l1a loss of function is not compensated by upregulation of other tcf genes Ztcf7l1a -/and MZtcf7l1a -/embryos develop eyes whereas embryos lacking both Ztcf7l1a and Ztcf7l1b do not (Dorsky et al. 2003). Thus, we hypothesised that enhanced expression of the paralogous tcf7l1b, or other lef/tcf genes may compensate for the absence of tcf7l1a function, as shown for other mutations (El-Brolosy et al., 2018;Rossi et al., 2015). To test this idea, we assessed the expression of all lef/tcf genes by RT-qPCR in sibling wildtype and Ztcf7l1a -/mutant embryos at the stage when the eye field has been specified (10 hours post fertilisation; hpf).
Expression levels of lef/tcf genes did not increase in Ztcf7l1a -/mutant embryos, which suggests that there is no compensatory upregulation ( Fig.2A, TableS1). As previously shown, tcf7l1a undergoes nonsense-mediated decay in mutants resulting in reduced expression levels (Kim et al., 2000;Fig2A;TableS1). lef1 and tcf7 levels did not change significantly in mutants and tcf7l1b (tcf3b) and tcf7l2 (tcf4) expression was actually reduced to 63±6% and 62±8% respectively of wildtype levels ( Fig.2A; TableS1). The otx1b and otx2 genes, which are expressed in the anterior neural plate, also showed slightly reduced expression (otx1b, reduced to 81±11% and otx2, 79±10%) suggesting the anterior neural plate may be slightly reduced in size in mutants. Indeed the domain of the neural plate encompassed by expression of emx3 around the anterior margin of the neural plate up to the mesencephalic marker pax2a (Fig.2 D , E) was reduced to 76% of wildtype size in mutants (n=11, p=0.0041, Fig.2B; TableS2). This indicates that a reduction in the size of the prospective forebrain of Ztcf7l1a -/embryos may contribute to the reduced levels of expression of tcf7l1b, tcf7l2 and otx genes. Overall these results suggest that tcf genes do not show compensatory upregulation in response to loss of tcf7l1a function.
Optic vesicles evaginate and form eyes in MZtcf7l1a -/mutants despite a muchreduced eye field.
More remarkable than the modest changes in tcf and otx gene expression was the finding that qRT-PCR showed very reduced expression of eye field genes in Ztcf7l1a -/mutant embryos ( Fig.2A; rx3 reduced to 26±1%, p=0.0002 and six3b reduced to 44±5%, p=0.0091 of wildtype levels). Consequently, the presence of overtly normal looking eyes in both Ztcf7l1a -/and MZtcf7l1a -/embryos is surprising given that rx3 -/mutant embryos lack eyes due to impaired specification/evagination of the optic vesicles (Loosli et at., 2003;Stigloher et al., 2006). We confirmed that expression of six3b and rx3 is reduced in the anterior neural plate by in situ hybridisation in Ztcf7l1a -/and tcf7l1a morphant embryos (100%,n>40; Further ISH analysis suggests that it is the caudal region of the eye field that is most affected in Ztcf7l1a -/mutants. emx3 expression directly rostral to the eye field is slightly broader in Ztcf7l1a -/mutants than wildtypes but expression does not encroach into the reduced eye field (Fig.2D,E; Fig.S4A,B, n=5 each condition). Conversely, expression of the prospective diencephalic marker barhl2 caudal to the reduced eye field was expanded rostrally at 10hpf (Fig.S4C,D n=5 each condition) and even more evidently at 9hpf ( Fig.S4E,F, 13/13 Ztcf7l1a -/-). These observations suggest a caudalisation of the anterior neural plate in Ztcf7l1a -/mutants leading to reduced eye field specification consistent with phenotypes observed in conditions in which Wnt pathway repression is reduced (Heisenberg et al., 2001;Van de Water et al., 2001).

Tcf7l1a functions cell-autonomously to promote eye field specification
Although Tcfs regulate the balance between activation and repression of the Wnt/βCatenin pathway during anterior neural plate regionalisation (Kim at el., 2000, Dorski et al., 2003, it is unclear if Tcf function in the eye field is required for cells to adopt retinal fate. To address this, we determined whether Tcf7l1a function is required cell-autonomously during eye formation by transplanting wildtype and MZtcf7l1a -/-GFP labelled (GFP+) cells into wildtype and mutant hosts and analysing the expression of rx3 when eye specification has occurred (100% epiboly; Fig.3).
Transplants of wildtype cells to MZtcf7l1a -/mutant embryos led to the recovery of rx3 expression exclusively restricted to the wildtype GFP+ cell clones (13/13 transplants, donor embryos to wildtype hosts showed no effect on rx3 expression (not shown).
Consistent with a cell autonomous role for Tcf7l1a in eye formation, overexpression of the Wnt inhibitor Dkk1 (Hashimoto et al. 2000) expanded the anterior neural plate in both wildtype and tcf7l1a morphants, but rx3 expression and eye field size remained much smaller in the enlarged anterior plate (Fig.3H-K)).
All together, these results support a cell-autonomous role for Tcf7l1a in promoting eye field specification.
Eye size in Ztcf7l1a -/embryos recovers with growth kinetics similar to wildtype embryos.
Despite a much-reduced eye field, eyes in Ztcf7l1a -/fry and adults seem indistinguishable from those in wildtype siblings. Indeed, optokinetic responses of Ztcf7l1a -/and wildtype 5dpf larvae showed no significant differences at any of the four tested spatial frequencies Eye growth in both wildtypes and Ztcf7l1a -/mutants show similar growth kinetics (Fig.4A).
This suggests that even though Ztcf7l1a -/eyes are smaller, they follow a comparable developmental time-course as wildtype eyes in the early growth phase between 24 and 36hpf but with about 8 hours delay (for instance, a 32hpf Ztcf7l1a -/eye is about the same size as a wild-type 24hpf eye).
The temporal shift in eye growth in Ztcf7l1a -/mutants is not explained by an overall developmental delay as the position of the posterior lateral line primordium (pLLP) was similar to wildtype at all stages tested (Fig.S7).

Eye size recovers after physical ablation of much of the optic vesicle
To assess if the size recovery is a general feature of eye development, we physically ablated optic vesicle cells in wildtype embryos and assessed the effect on eye growth. Cells were aspirated from one of the two nascent optic vesicles at 12hpf (6 somite stage), leaving approximately the medial half of the vesicle intact ( Fig.4M). At 36hpf there was still a clear size difference between the experimental and control eyes (Fig.4N). However, by 4dpf we observed no obvious size difference between control and experimental eyes (n=13/13, Fig.4O). Consequently, the forming eye can effectively recover from either genetic or physical reduction in the size of the eye field/evaginating optic vesicle.

Neurogenesis is delayed in tcf71a mutant eyes.
The observation that wildtype and Ztcf7l1a -/mutant eyes display similar, but temporally offset, growth kinetics led us to speculate that that retinal neurogenesis might be delayed in Ztcf7l1a -/eyes to extend the period of proliferative growth prior to retinal precursors undergoing neurogenic divisions. Indeed, between 36 and 40hpf, Ztcf7l1a -/retinas express atoh7 exclusively in the nasal half of the retina (Fig.5H, Q), a phenotype we did not see at any stage in sibling embryo eyes.
These data indicate that progression of atoh7 expression and neurogenesis is delayed by about 8-12 hours in Ztcf7l1a -/retinas compared to siblings, a timeframe comparable to the delays seen in optic vesicle growth. In line with our results in Ztcf7l1a -/embryos, eye vesicle ablated wildtype retinas also showed delayed neurogenesis compared to control nonablated contralateral eyes at 36hpf ( Our results suggest that retinal precursors in Ztcf7l1a -/eyes remain proliferative at stages when precursors in wildtype eyes are already producing neurons.

Larger eyes undergo premature neurogenesis.
Our results are consistent with the idea that neurogenesis may be triggered when the optic vesicle reaches a critical size. To explore this possibility, we generated embryos with larger optic vesicles by overexpressing the Wnt antagonist Dkk1 (Hashimoto et al., 2000).
After 36hpf, wildtype eyes gradually caught up in size as growth slowed in eyes in dkk1overexpressed embryos ( Fig.5R; TableS6). Neurogenesis was prematurely triggered by 28hpf in the eyes of dkk1 overexpressing embryos, with many more cells expressing atoh7 compared to eyes in heat-shocked control embryos (Fig.5M, N, n=7 out of 9 embryos). This result is unlikely to be due to a direct effect of dkk1 overexpression on neurogenesis as premature neurogenesis is not triggered in tg(hsp70:dkk1-GFP) w32 retinas heat-shocked at 24hpf (Fig.5O, P, n=10, 100%).
These results further support a link between eye size and the onset of neurogenesis and the size self-regulating ability of the forming eye.

ENU modifier mutagenesis screen in tcf7l1a mutant background reveals two groups of genetic modifiers.
Although eye formation can recover in tcf7l1a -/mutants despite a much smaller eye field, we speculated that eye development in these embryos might be sensitised to showing the effects of additional mutations. To test this, we performed an ENU mutagenesis screen on fish carrying the tcf7l1a mutation (Fig.6A).
Homozygous Ztcf7l1a mutant adult male fish (F0 founders) were exposed to four rounds of ENU (van Eeden et al., 1999) and then crossed with Ztcf7l1a -/adult females to generate F1 families (Fig.6A). However, possibly because of cellular stress or the synergistic cumulative effect of many mutations induced by ENU, we observed many eyeless F1 embryos. To circumvent this problem, we injected 10pg/embryo of zebrafish tcf7l1a mRNA to rescue any Tcf-dependent eyeless phenotypes in the F1 embryos (Fig.6A). Adult F1 fish were outcrossed to EKW wildtype strain. All F2 fish were tcf7l1a +/and half carried unknown mutations (m) in heterozygosity (Fig.6A). To screen, we randomly crossed F2 pairs from each family aiming for at least 6 clutches of over 100 embryos. The probability of finding double Ztcf7l1a -/-/m -/embryos for independently segregating mutations is 1/16, hence we would expect to find ~6 double mutants in 100 embryos. Here, we describe examples of synthetic lethal mutations that lead to microphthalmia/anophthalmia (U768; Fig.6C) or eyes that fail to grow (U762, U901; Fig.7, 8).
Within this interval is hesx1, which morpholino knock-down experiments had previously suggested to genetically interact with tcf7l1a (Andoniadou et al., 2011). Primers for hesx1 cDNA failed to amplify in U768/Ztcf7l1a -/eyeless embryo cDNA samples. Using a primer set that spans the hesx1 locus, we found that all U768/Ztcf7l1a -/eyeless embryos have a ~2700bp deletion that covers hesx1 exons 1 and 2 (hesx1 Δex1/2 ; Fig.S8); this was unexpected as deletions are not normally induced by ENU (see below). Sequencing of the hesx1 locus reveals that there is a polyA stretch of approximately 80 nucleotides followed by a 33 AT microsatellite repeat on the 3' end of intron 2 that may have generated a chromosomal instability that led to the deletion of exons 1 and 2 (Fig.S8). As a consequence of the deletion, hesx1 mRNA was not detected by RT-PCR or in situ hybridisation in U768 homozygous embryos (Fig.6E, H, I). We further confirmed that only U768-F2 embryos that are homozygous for both the tcf7l1a mutation and hesx1 Δex1/2 are eyeless (Fig.6B,C).
As ENU usually generates point mutations, we speculated that the deletion in hesx1 Δex1/2 was not caused by our mutagenesis but was already present in one or more fish used to generate the mutant lines. Indeed, we found the same deletion in wildtype fish not used in the mutagenesis project. To confirm that the eyeless phenotype in U768/Ztcf7l1a -/double mutants is not caused by another mutation induced by ENU, we crossed Ztcf7l1a -/fish to one such wildtype TL fish carrying hesx1 Δex1/2 . Incrossing of hesx1 Δex1/2/Δex1/2 /tcf7l1a +/adult fish led to embryos with a very small rudiment of eye pigment with no detectable lens ( Fig.6J, K). Genotyping of eyeless and sibling embryos confirmed that only double homozygosity for hesx1 Δex1/2 /Ztcf7l1a -/led to the eyeless embryo phenotype (TableS8).
The interaction between hesx1 and tcf71a mutations strikingly illustrates how the developing eye can fully cope with loss of function of either gene alone but fails to form in absence of both gene activities. Additional eyeless families that do not carry the hesx1 deletion were identified but they remain to be validated and mutations cloned.
The tcf7l1a mutation can enhance the phenotypic severity of mutants with small eyes U762 mutants, wildtype for tcf7l1a or heterozygous for the tcf7l1a mutation, show reduced eye size by 72hpf and this phenotype is considerably more severe in embryos homozygous for the tcf7l1a mutation ( Fig.7A-F). The U762 mutation was mapped by SSLP segregation analysis to a 1.69Mb interval between 15.50Mb and 17.19Mb on chromosome 24 (Fig.S9A).
Through sequencing candidate genes in the interval ( Fig.S9A; TableS9), we identified a mutation in the splice donor of cct5 (chaperonin containing TCP-1 epsilon) intron 4 (GT>GC, Fig.7G). The mutation leads to the usage of an alternative splice donor in the 3' most end of cct5 exon 4, which induces a two nucleotide deletion in the mRNA (Fig.S9B). This deletion changes the reading frame of the protein C-terminal to amino acid 176, encoding a 29aa nonsense stretch followed by a stop codon ( Fig.7H; Fig.S9B). The mutation also induces nonsense-mediated decay of the mRNA (not shown). U762 and cct5 hi2972bTg mutations failed to complement (not shown) supporting the conclusion that the mutation in U762 responsible for the tcf7l1a modifier phenotype is in cct5. Cct5 is one of the eight subunits of the chaperonin TRiC/TCP-1 protein chaperone complex, which assists the folding of actin, tubulin and many proteins involved in cell cycle regulation (Sternlicht et al., 1993;Dekker et al., 2008;Yam et al., 2008).
Homozygous U901 mutants show a slightly smaller and misshapen eye; this mutation was mapped to gdf6a (Valdivia et al., 2016). Unlike Ztcf7l1a -/mutants in which eye size recovers, eyes in gdf6a U901/U901 /Ztcf7l1a -/embryos remain smaller than in single mutants or wildtypes . This suggests that the ability to compensate eye size is compromised in absence of both gdf6a and tcf7l1a function.
Altogether, analysis of the interacting mutations reveals that although abrogation of Tcf7l1a function alone has little effect on formation of eyes, it can lead to complete loss of eye formation or more severe eye phenotypes in combination with additional mutations.
Consequently, although eye development is sufficiently robust to cope with loss of Tcf71a, mutant embryos are sensitised to the effects of additional mutations.

Discussion
In this study, we show that although Tcf7l1a is required for cells to adopt eye field identity and express rx3, tcf7l1a mutants form normal, functional eyes. This finding reveals a remarkable ability of the developing eye to form normally from an eye field that is half the size of that in the wildtype condition. Tcf function in tcf7l1a mutants is not genetically compensated by upregulation of other tcf genes nor by other genetic mechanisms that restore neural plate regionalisation and eye field formation. Instead, we find that tcf7l1a mutant optic vesicles delay neurogenesis to enable size recovery. We observe a similar effect when optic vesicle cells are physically ablated. In contrast, neurogenesis is prematurely induced in larger optic vesicles, likely depleting progenitors and slowing growth. Our results suggest that size-dependent regulation of the balance between proliferation and differentiation may buffer the developing eye against initial differences in cell number. Although the developing eye can cope with loss of Tcf7l1a function, we speculated that embryos lacking Tcf7l1a would not be robust to the consequences of additional mutations affecting eye formation. In support of this, we identify mutations in three other genes that give synthetically enhanced eye phenotypes when combined with the tcf7l1a mutation. This approach facilitates identification of genes that participate in genetic networks that make developing eyes robust to mutations that compromise eye field specification and optic vesicle growth.
The tcf7l1a mutation is fully penetrant with no apparent genetic compensation during neural plate patterning Tcf7l1 is a core Wnt pathway transcription factor that can activate or repress genes dependent upon the status of the Wnt signalling cascade (Cadigan and Waterman, 2012).
Homozygous tcf7l1 mutant mice present severe mesodermal and ectodermal patterning defects (Merrill et al., 2004), but the duplication of tcf7l1 into tcf7l1a and tcf7l1b in zebrafish has led to functional redundancy (Dorsky et al., 2003).
Although MZtcf7l1a embryos have a severe eye field specification phenotype they still develop normal eyes. We confirmed that the tcf7l1a m881 mutant allele is null, generates no wildtype transcript and that morpholino knock-down specifically of Tcf7l1a does not give an eyeless phenotype. Hence, the originally described MZtcf7l1a eyeless phenotype (Kim et al., 2000) may have been due to genetic background effects modifying the outcome of the tcf7l1a m881 allele. The fact that we were able to recover an eyeless modifier of the tcf7l1a phenotype in our own mutagenesis pilot screen lends support to this idea.
At the stage of eye specification, we did not find genetic compensation in tcf7l1a mutants by other tcf genes. Even though tcf7l1a mutants develop eyes, they do so from an eye field that is ~50% smaller than wild-type. Although we did not find evidence for genetic compensation, and despite tcf7l1 being duplicated in fish, the fact that neither gene has been lost due to genetic drift suggests that having both genes may confer enhanced fitness and robustness to zebrafish. As an example, paralogous Lefty proteins make Nodal signalling more stable to noise and perturbations during early embryogenesis (Rogers et al., 2017).
We show that Tcf7l1a has a cell-autonomous role in specification of the eye field. Tcf7l1a is required for the expression of rx3 and consequently is a bona fide eye field gene regulatory network transcription factor that functions upstream to rx3. tcf7l1a is expressed very early in the anterior neural plate and so may work alongside otx, sox, six and pax genes to regionalise the eye-forming region of the neural plate (Beccari et al 2013;Zuber et al. 2003). Considering that it is the repressor activity of Tcf7l1a that promotes eye formation (Kim et al., 2000), the most likely role for Tcf7l1a is to repress transcription of a gene that suppresses eye field formation.

Compensatory tissue growth confers robustness to eye development
We show that despite the small eye field in tcf7l1a mutants, the optic vesicles evaginate and undergo overtly normal morphogenesis. Although tcf7l1a mutant eye vesicles are still much smaller than wild-type at 24hpf, we found that their eye growth kinetics are similar.
This suggests that the mechanisms that regulate overall growth of the retina in both conditions are comparable albeit delayed in the tcf7l1a mutant retina.
Although atoh7 expression is initiated in the ventronasal retina in tcf7l1a mutants at the same stage as in wild-type eyes, the wave of atoh7 expression that spreads across the retina is delayed by approximately 8-12hrs in mutants. atoh7 is required for the first wave of neurogenesis in the retina (Brown et al., 2001;Kay et al., 2001;Wang et al., 2001) and thus, the delay we see in tcf7l1a mutants suggests that RPCs continue proliferating in mutants at stages when they are already generating neurons in wild-type eyes. We presume that the extended period of proliferative growth due to delayed neurogenesis enables the forming eye to continue growing and recover its size. We observed a similar phenomenon of delayed neurogenesis and prolonged growth when cells were ablated from the optic vesicles. Conversely, atoh7 spreads precociously in experimentally enlarged optic vesicles.
The premature neurogenesis of RPCs in these conditions may contribute to eyes achieving a final size similar to wild-type. All together, our data suggest that the timing of the spread of neurogenesis across the retina is coupled to size of the eye, thereby providing a mechanism to buffer eye size. It is intriguing that the compensatory changes in growth seen in tcf7l1a mutant and optic-vesicle ablated eyes seem to occur prior to the establishment of the ciliary marginal zone, which accounts for the vast majority of eye growth (Fischer et al., 2013).
Our results support classical embryology experiments from Ross Harrison, Victor Twitty and others (Harrison, 1929, Twitty and Schwind, 1931, Twitty and Elliott, 1934. These investigators showed that when eye primordia from small-eyed salamander species (A. punctatum) were transplanted to larger-eyed salamanders (A. Tigrinum) or vice-versa, the eye derived from the grafted tissue formed an eye of a size corresponding to the donor salamander species. Species-specific size differences are also observed in self-organising in vitro cultured eye organoids derived from mouse or human embryonic stem cells (Nakano et al., 2012). Our work, together with the experiments in salamanders and organoids, suggests that the developing eye has intrinsic size-determining mechanisms.
Size regulatory mechanisms have been previously described in other species and perhaps most extensively studied in the fly wing imaginal disc (Potter and Xu, 2001). Indeed, many models have been put forward to explain imaginal disk size control (Eder et al., 2017;Irvine and Shariman, 2017;Vollmer et al., 2017). It is evident that the final size of paired structures within individuals is remarkably similar supporting the idea that the mechanisms that control organ/tissue size are mostly intrinsic to the tissue/organ and highly robust.

Addressing eye robustness through a forward mutagenesis screen on tcf7l1a mutant background
Our results indicate that tcf7l1a mutant eyes are sensitised to the effects of additional mutations. Indeed, a homozygous deletion of the two first exons of hesx1 leads to eyeless embryos when in combination with tcf7l1a. This result also confirms our previous observations suggesting a genetic interaction between hesx1 and tcf7l1a based upon morpholino knock-down experiments (Andoniadou et al., 2007). Furthermore, both hesx1 and tcf7l1a are expressed in the anterior neural plate including the eye field, and as observed in tcf71a zebrafish mutants, hesx1 mutant mice also show a posteriorised forebrain (Andoniadou et al., 2007;Martinez-Barbera et al., 2000). These and our results suggest that Tcf7l1a and Hesx1 have similar, overlapping functions in the anterior neural plate such that the eyeless phenotype is expressed in zebrafish only when both genes are abrogated. Mutations in hesx1 lead to anophthalmia, microphthalmia, septo-optic dysplasia (SOD) and pituitary defects in humans and mice (Dattani et al., 1998;Gaston-Massuet et al., 2008;Martinez-Barbera et al., 2000;Thomas et al., 2001). Interaction of hesx1 mutations with other genetic lesions may also occur in patients carrying Hesx1 mutations, as the phenotypes in these individuals show variable expressivity (McCabe et al., 2011). In these patients, tcf7l1a should be considered as a candidate modifier for hesx1-related genetic conditions.
Gdf6a is a TGFβ pathway member (David and Massagué, 2018) that when mutated in zebrafish results in small mis-patterned eyes, neurogenesis defects and retino-tectal axonal projection errors (Gosse and Baier, 2009;French et al., 2009). In humans, mutations in GDF6 have been identified in anophthalmic, microphthalmic and colobomatous patients Mutations in cct5 in combination with tcf7l1a also led to phenotypes in which reduced eye size failed to recover. cct5 codes for the epsilon subunit of the TCP-1 Ring Complex (TRiC) chaperonin that is composed of eight different subunits that form a ring, the final complex organised as a stacked ring in a barrel conformation (Yebenes et al., 2011). In vitro studies indicate TRiC chaperonin mediates actin and tubulin folding (Sternlicht et al., 1993); however, it also assists in the folding of cell cycle-related and other proteins (Dekker et al., 2008;Yam et al., 2008). A mutation in cct2 has been found in a family with Leber congenital ameurosis retinal phenotype (Minegishi et al., 2016;Minegishi et al., 2018) and mutations in cct4 and cct5 have been related to sensory neuropathy (Pereira et al., 2017;Lee et al., 2003;Hsu et al., 2004;Bouhouche et al., 2006). Similar to our cct5 mutant, cct1, cct2, cct3, cct4 and cct8 mutant zebrafish show retinal degeneration (Berger et al., 2018;Matsuda and Mishina, 2004;Minegishi et al., 2018), suggesting that the cct5 mutant phenotype is due to abrogation of TRiC chaperonin function, and not due to loss of a cct5 specific role.
Double cct5/tcf7l1a homozygous mutant eyes degenerate prematurely and to a greater extent than cct5 single mutants, and neurogenesis is also severely compromised. This shows that the consequence of cct5 loss of function is exacerbated by the lack of tcf7l1a function, although it is currently unclear how such an interaction might occur. However, this genetic interaction does highlight that in some conditions a gene of pleiotropic function, like cct5, can lead to a specific phenotype in the eye.
Anophthalmia and microphthalmia are generally associated with eye field specification defects (Reis and Semina, 2015), but given that normal eyes can still develop from a much reduced eye field, further analysis of the genetic and developmental mechanisms that lead to small or absent eyes is warranted. Our isolation and identification of modifiers of tcf7l1a highlights the utility of genetic modifier screens to identify candidate genes underlying congenital abnormalities of eye formation. Indeed, given that Tcf7l1a itself can now be classified as a bona fide gene in the eye transcription factor regulatory network, it should be considered when screening patients with inherited eye morphological defects.

Author Contributions
RY and SW conceived the project and analysed the data; RY, FC, TH, GG, EA, AK, JR and IB performed the experiments; RY, FC, TH, HS, QS, LL and CW performed the genetic screen. RY and SW wrote the paper with input from all co-authors but primarily from FC, HS, TH, QS and GG. SW secured the funding of this project.
to RY, Wellcome Trust grants to SW and RY, an MRC Programme grant to SW and GG, Royal Society International Joint Project funding to SW and MA and FONDAP (15090007) to MA.

Animal use, mutant and transgene alleles, genotyping and heat shock
Adult zebrafish were kept under standard husbandry conditions and embryos were obtained by natural spawning. Wildtype and mutant embryos were raised at 28.5ºC and staged according to Kimmel et al. (1995). To minimise variations in staging, embryos were collected every 30 minutes and kept separate clutches according to their time of fertilisation.
For heatshock (HS) gene induction, embryos from a heterozygous Tg(hsp70:dkk1-GFP) w32 to wild type cross were moved from embryo media at 28.5ºc to 37ºC at 6hpf or 20hpf for 45minutes, and then back to 28.5ºC embryo media. Three hours post HS, embryos were separated in controls (GFP-) and HS experimental (GFP+) groups, and fixed at the stages described in results.

ENU mutagenesis and mutant mapping
Homozygous male tcf7l1a m881 fish were exposed to four rounds of ENU according to Van Eeden et al. (1999). Details of the mutagenesis pipeline are in the results section. Embryos from incrosses of carriers of the cct5 U762 or gdf6a U768 mutations, which show a phenotype as homozygous embryos independently of mutations in tcf7l1a, were identified for the described eye phenotype at 3dpf to avoid ambiguity and false positives. For rough mapping, batches of 30 mutants and 30 siblings were fixed in methanol and genomic DNA was extracted by proteinase K protocol. This gDNA was then used for bulk segregant analysis PCR to test a library of 245 polymorphic SSLP variants spanning the whole zebrafish genome (Stickney et al., 2002). SSLP markers heterozygous in the sibling samples and homozygous in the mutant sample were confirmed on gDNA samples of 12 mutant and 12 sibling individuals. Markers that showed linkage to a locus were tested on additional mutant samples, and more SSLP markers were tested for the mapped region until a genomic interval was defined.
Homozygous tcf7l1a/hesx1 U901 mutant carriers were incrossed, and eyeless embryos and siblings were fixed in methanol. Rough mapping was carried out as above but in this case sibling embryos used for bulk segregant analysis were genotyped for tcf7l1a m881 and only homozygous mutants with eyes were included in the sibling pool. tcf7l1a m881/m881 embryos injected with mo tcf7l1b phenocopy the loss of eye phenotype seen in tcf7l1a m881/m881 /tcf7l1b +/zf157tg double mutants (Young and Wilson, unpublished).

RNA extraction, reverse transcription and qPCR
Total RNA and genomic DNA were isolated from individual embryos at 10hpf following Life Technologies Trizol protocol. cDNA was synthesised by reverse transcription using The sizes of eye profiles were quantified from lateral view images of PFA-fixed embryos by delineating the eye using Adobe Photoshop CS5 magic wand tool and measuring the area of pixels included in the delineated region. The surface area was then transformed from px 2 to µm 2 . The eye profile and eye volume were calculated from confocal imaging of vsx2 in situ hybridisation stained embryos at 24hpf. The eye volume/eye profile ratio average from 10 embryos was 53.24. This ratio was used to estimate eye volume from eye profile area and assumes that the profile area to eye volume ratio is constant after 24hpf.

Confocal microscopy and image analysis
Confocal imaging was performed on a Leica TCS SP8 confocal microscope. For time lapse analyses, the stage was set in an air chamber heated to 28.5°C. Live embryos were immobilized in 1% low melting point agarose (Sigma) and 0.016% Tricaine (Sigma) to anesthetize. Image volume analysis measurement was performed on Imaris 7.7.0 and Fuji.

Cell transplantation
WIldtype or MZtcf7l1a -/embryos used as donors were injected with 50pg of GFP mRNA at 1 cell stage. At 3-4hpf, blastula stage, dechorionated donor and host embryos were mounted in 3% methylcellulose in fish water supplemented with 1% v/v penicillin/streptomycin (5,000 units penicillin and 5mg streptomycin per ml) and viewed with a fixed-stage compound microscope (Nikon Optiphot). Approximately 30-40 cells were taken from the animal pole of donors and transplanted to approximately the same position in hosts by suction using an oil-filled manual injector (Sutter Instrument Company).
Embryos were moved to 1% penicillin/streptomycin supplemented fish media and fixed at 10hpf.

Eye vesicle cell removal
Embryos were mounted in 1% low melting point agarose in Ringer's solution supplemented with 1% v/v penicillin/streptomycin. A slice of set agarose was removed to expose one of the eyes and a drop of mineral oil (sigma) was placed over the target eye to dissolve the epidermis (Picker et al., 2009). After two minutes the oil drop was removed and optic vesicle cells were sucked out with a capillary needle filled with mineral oil. Embryos were left to recover for half an hour before being released from the agarose.

Optokinetic response
Optokinetic responses were examined using a custom-built rig to track horizontal eye movements (optokinetic nystagmus) in response to whole-field motion stimuli. Larvae at 4 dpf were mounted in 1% low melting point agarose in fish water and analysed at 5 dpf. The agarose surrounding the eyes was removed to allow normal eye movements. Sinusoidal gratings with spatial frequencies of 0.05, 0.1, 0.13 and 0.16 cycles/degree were presented on a cylindrical diffusive screen 25 mm from the centre of the fish's head with a MicroVision SHOWWX+ projector. Gratings had a constant velocity of 10 degrees/s and changed direction and/or spatial frequency every 20 s. Eye movements were tracked under infrared illumination (720 nm) at 60 Hz using a Flea3 USB machine vision camera and customwritten software. Custom-designed Matlab code was used to extract the eye velocity Zuber, M.E., Gestri, G., Viczian, A.S., Barsacchi, G., and Harris, W.A. (2003). Specification of the vertebrate eye by a network of eye field transcription factors. Development 130, 5155-5167.   Table S2), and eye field volume (C) by rx3 fluorescent in situ hybridisation confocal volume reconstruction (data in Table   S3). (D-I) Expression of emx3 (arrowhead)/pax2a (D, E), six3b (arrowhead)/pax2a (F, G) and rx3 (arrowhead)/pax2a (H, I) in wildtype (D, F, H) and Ztcf7l1a -/- (E, G, I)           Genomic DNA sequence deleted in U768 fish in bold. hesx1 exons 1 and 2 highlighted in yellow. Open reading frame first codon in exon 1 is highlighted in red. Standard deviation (SD). P value calculated by an unpaired t-test with Welch's correction.

Ztcf7l1a mutants.
Prospective forebrain size data generated by measuring the volume enclosed by emx3 expression to the rostral limit of pax2a (mesencephalic marker) expression after whole mount in situ hybridisation in wildtype (+/+), tcf7l1a +/-(+/-) and Ztcf7l1a -/-(-/-) embryos at TableS3. Measurement of the volume of rx3 expression in the eye field by fluorescent in situ hybridisation in wildtype and tcf7l1a mutants.