The Pax6 master control gene initiates spontaneous retinal development via a self-organising Turing network

ABSTRACT Understanding how complex organ systems are assembled from simple embryonic tissues is a major challenge. Across the animal kingdom a great diversity of visual organs are initiated by a ‘master control gene’ called Pax6, which is both necessary and sufficient for eye development. Yet precisely how Pax6 achieves this deeply homologous function is poorly understood. Using the chick as a model organism, we show that vertebrate Pax6 interacts with a pair of morphogen-coding genes, Tgfb2 and Fst, to form a putative Turing network, which we have computationally modelled. Computer simulations suggest that this gene network is sufficient to spontaneously polarise the developing retina, establishing the first organisational axis of the eye and prefiguring its further development. Our findings reveal how retinal self-organisation may be initiated independently of the highly ordered tissue interactions that help to assemble the eye in vivo. These results help to explain how stem cell aggregates spontaneously self-organise into functional eye-cups in vitro. We anticipate these findings will help to underpin retinal organoid technology, which holds much promise as a platform for disease modelling, drug development and regenerative therapies.

I think there are a number of perturbation experiments that could more directly support the hypothesis that the role of the Pax6/Fst/Tgfb2 network is to spontaneously polarize the optic vesicle. However, one would want to see a change in the patterning (e.g. via pax6 expression), that is more subtle than either that pax6 is downregulated (e.g. Fig 5e) or pax6 is unaffected (e.g. Fig  5i).
One possibility I could imagine would be to perform some mosaic perturbation (similar to Fig 5), and show that this can change pax6 patterning throughout the optic vesicle. It would be important to see non-cell-autonomous changes. For example, could the authors use mosaic FstMO or mosaic Fst overexpression (data they may already have) to ask whether they can shift the location of the pax6 domain? The results of these experiments could then be compared to simulations in which the perturbation is applied in silico.
The experimental design in Fig 5g,h could also provide complementary evidence, if I understand it correctly. Here, endogeneous Fst is blocked by the morpholino, but constitutive (i.e. pax6independent) Fst is provided. This is a nice experiment because it doesn't completely remove Fst, but only removes its transcriptional regulation (including via pax6), which (my intuition suggests) will be important for the self-organizing abilities of the network. From the data presented (Fig 5g), it appears that the vesicle polarises fine -is this to be expected from the simulations?

BETTER DESCRIPTION OF TISSUE GEOMETRY
Throughout the manuscript, I was not always clear on the geometry/orientation of the data presented -both experimental and simulations. I would recommend: a) early on, provide a brief description and schematic of the 3D structure of the optic vesicle, the proximal-distal patterning events and their relation to the vesicle-to-cup transition. Explain how this fits in with the orientation of the in situs presented. (Also: it appears that the morphology varies between different sections, why is this?) b) When describing the simulations, can you include a longer discussion in the main text on tissue geometry and boundary conditions. In the supplementary movies, it looks like there are 1D domains with periodic boundary conditions. Can the authors justify this choice (or at least discuss it in more detail, particularly from a biological standpoint)?
Can you give more details on the Fst transgene (as used in Fig 5g,h)? 2.
Lines 15-16, page 11: it is unclear a priori that log-transforming guarantees normality for fold changes -you have to make further assumptions (or show/state that log transforming makes your data closer to normally distributed). In the supplementary movies (and in Fig 4e), brief oscillations in Pax6 are seen -is this seen in vivo/in vitro? Is this a by-product of the initial conditions chosen? 5.
When discussing the data from Fig 4g-m, it should be emphasized that these predictions (i.e. larger domain gives multiple poles, smaller domain can result in no patterning) is a rather general feature of Turing systems and not a specific test of the Pax6/Fst/Tgfb2 network being responsible.

Reviewer 2
Advance summary and potential significance to field The manuscript by Timothy Grocott and co-workers tackles a key question of developmental biology, the pattern formation in early vertebrate embryos. They use patterning of early optic chick vesicle as an experimental model. Their central hypothesis is that a lineage-specific DNAbinding transcription factor Pax6 is a cornerstone of a cross-talk of multiple signaling pathways and these are wired together as a putative Turing network. The experiments test this model through three components of TGF/BMP signaling, Bmp4 follistatin (Fst), and Tfgb2 at their mRNA levels. The main findings in the paper are summarized in a model that accounts for differences of Pax6 regulation in distal and proximal regions of the optic vesicle (Fig. 6). Overall, these findings are novel and within the scope of journal Development.

Comments for the author
There is no direct relationship established between the Pax6 and Vsx2 expression in the retina. Additional genes, such as Fgf9 and Erk1/2, should be examined. Auto-regulation of Pax6 in the optic vesicle was probed through dnPax6 mis-expression. This experiment requires additional data to determine levels of ectopic expression in relationship to their endogenous levels. Explain why Smad6 and Smad7 are tested separately in Figs. 2 and 5. The central question and model should be further probed using single cell RNA-seq approach and/or future experiments to expand and independently validate the present model should be outlined in the Discussion. Minor points: 1) Abstract: Delete references. Add, if possible model organism studied.

2)
Introduction: Add a brief description of Turing networks.

3)
Introduction: Early chick lens development is also reviewed by Gunhaga, 2011. 4) Introduction: Additional information on BMP signaling, role of Smad6/7, Fst, and etc. is needed. Explain results of an earlier paper by Grocott et al. in terms of the molecular mechanisms and the present study [18]. Edit Results when this information is presented for the first time.

5)
Results: What is the level of Fst expression in the surface ectoderm ( Fig. 1f)? 6) Results (p4): In the eye, migratory neural crest is also called periocular mesenchyme.

7)
Discussion: Mutual repression of Pax6 and Pax2 is also relevant to the early retinal development (PMID: 11003833).

8)
References are not in Development format.

First revision
Author response to reviewers' comments ##################### The following is a point-by-point response indicating how we have addressed the concerns raised (either experimentally or by changes to the text/figures). ##################### Reviewer 1 Advance Summary and Potential Significance to Field: Grocott et al propose that a Turing-like network, comprising Pax6/Fst/Tgfb2, is involved in polarising the optic vesicle along its proximal-distal axis. They suggest that in vivo this network cooperates with external signals like BMP to polarise the tissue, but that this network also allows polarity to self-organize as had been observed in retinal organoids. If convincingly shown, I would regard this hypothesis as highly significant and of broad interest to the developmental biology community.
Many aspects of this proposal are well supported by the data provided. In particular, the authors show that interactions between Pax6, Fst and Tgfb2 are consistent with the logic required for a Turing system (summarized in Fig. 4d), and very nicely use simulations to argue that this putative Turing network is sufficient to explain optic vesicle polarization in silico.
Whilst the evidence for each of the interactions in Fig. 4d is good, I think that further evidence is required to show that this network is responsible (necessary? sufficient?) for optic vesicle polarization in vivo or in vitro. In other words, I am convinced that this network exists, but given the data presented, remain unconvinced that it is the primary driver of self-organization.

FURTHER EVIDENCE FOR THE ROLE OF PAX6/FST/TGFB2 NETWORK IN OPTIC VESICLE POLARIZATION
I think there are a number of perturbation experiments that could more directly support the hypothesis that the role of the Pax6/Fst/Tgfb2 network is to spontaneously polarize the optic vesicle. However, one would want to see a change in the patterning (e.g. via pax6 expression), that is more subtle than either that pax6 is downregulated (e.g. Fig 5e) or pax6 is unaffected (e.g. Fig 5i).
One possibility I could imagine would be to perform some mosaic perturbation (similar to Fig 5), and show that this can change pax6 patterning throughout the optic vesicle. It would be important to see non-cell-autonomous changes. For example, could the authors use mosaic FstMO or mosaic Fst overexpression (data they may already have) to ask whether they can shift the location of the pax6 domain? The results of these experiments could then be compared to simulations in which the perturbation is applied in silico.
The experimental design in Fig 5g,h could also provide complementary evidence, if I understand it correctly. Here, endogeneous Fst is blocked by the morpholino, but constitutive (i.e. pax6independent) Fst is provided. This is a nice experiment because it doesn't completely remove Fst, but only removes its transcriptional regulation (including via pax6), which (my intuition suggests) will be important for the self-organizing abilities of the network. From the data presented (Fig 5g), it appears that the vesicle polarises fine -is this to be expected from the simulations? ##################### A major theme of our revised manuscript is that, in vivo, there is a wealth of confounding positional information (e.g. extrinsic BMPs, Wnts, Tgfbs, and intrinsic Shh) that heavily constrains the Pax6/Fst/Tgfb2 network. Thus, in vivo demonstration of self-organisation is experimentally intractable due to this multitude of uncontrollable variables. In other words, in vivo data of the kind presented in our previous Fig. 5 (now called Fig. 6) cannot therefore demonstrate selforganisation as the network is not free to express this behaviour.
We therefore demonstrate self-organisation in vitro using cultured optic vesicle explants. These experiments are far more tractable, since extrinsic positional information is discarded during dissection and only an intrinsic Shh gradient remains, which we eliminate by pharmacological inhibition. We thank the reviewer for requesting further evidence to support this point and trust that these new data make a stronger case for self-organisation. #####################

BETTER DESCRIPTION OF TISSUE GEOMETRY
Throughout the manuscript, I was not always clear on the geometry/orientation of the data presented -both experimental and simulations. I would recommend: a) early on, provide a brief description and schematic of the 3D structure of the optic vesicle, the proximal-distal patterning events and their relation to the vesicle-to-cup transition. Explain how this fits in with the orientation of the in situs presented. (Also: it appears that the morphology varies between different sections, why is this?)
A new panel A in Fig. 1 shows 3-D reconstructions of optic vesicles/cups at different stages of development, and the horizontal plane of sectioning is indicated.
A new panel D in Fig. 1 shows a labelled 2-D schematic of a horizontal section through the optic vesicle, corresponding to the plane of sectioning shown in Fig. 1A. A full description is provided in the updated figure legend. Please see lines 693-696 on page 22 of the revised manuscript.
Both new panels are cited from the main text.
We have also added new text explaining that "Differences in morphology of sections are due to i) slight differences in staging of embryos between HH10-and HH10+, and ii) slight obliqueness and variation in the dorsal-ventral level of the horizontal sections." Please see lines 492-494 on page 17. ##################### b) When describing the simulations, can you include a longer discussion in the main text on tissue geometry and boundary conditions. In the supplementary movies, it looks like there are 1D domains with periodic boundary conditions. Can the authors justify this choice (or at least discuss it in more detail, particularly from a biological standpoint)?
For 1-D simulations, we now explain that the single spatial dimension was intended to "represent the optic vesicle's anterior-posterior axis (comprising anterior-proximal, distal and posteriorproximal domains). Simulations were performed with both zero-flux (Fig. 5) and periodic (Supplementary Movies 1 & 2) boundary conditions to represent dissected optic vesicle explants and spherical organoids, respectively." Please see lines 154-157 on page 6 of the revised manuscript.
We have improved Fig. 4 to better explain correspondence between 1-D simulations and tissue geometry.
In Fig. 4 B The following text was added to the legend for Fig. 4: "The vertical y-axis represents the hemispherical optic vesicle's anterior-posterior axis, which is divided into anterior-proximal, distal and posterior-proximal domains". Please see lines 754-756 on page 24 of the revised manuscript.
For 2-D simulations, we now explain that "we explored both zero-flux and fixed boundary conditions, disregarding the latter as the former agreed more closely with experimental observations. It may be interpreted that adsorption of morphogens to extracellular matrix and cell surface proteins within explants prevents a significant outward flux, while the absence of morphogens from the defined bathing medium prevents an inward flux." Please see lines 190-194 on page 7 of the revised manuscript. ##################### MINOR POINTS 1. Can you give more details on the Fst transgene (as used in Fig 5g,h)?
We also explain that FstMO knocks down both Fst 300 and Fst 315 isoforms. Please see lines 263-264 on page 9 of the revised manuscript.
The following text was added to the legend for 2.Lines 15-16, page 11: it is unclear a priori that log-transforming guarantees normality for fold changes -you have to make further assumptions (or show/state that log transforming makes your data closer to normally distributed).
We now report that fold-change data was "log-transformed to bring data closer to a normal distribution (verified by Shapiro-Wilk test) prior to plotting and null hypothesis significance testing." Please see lines 456-457 on page 16 and lines 508-510 on page 18 of the revised manuscript. ##################### We apologise for this discrepancy and are very grateful to the reviewer for raising this point as it led us to a tractable in vitro assay for self-organisation as described above.
The 1-D simulation and RT-QPCR data from the previous Fig. 4k
As described above, the previous Fig. 4 g-m have been replaced by new 2-D simulated and experimental explant data.
However, we have modified the text introducing these new experiments to emphasize the generality of this feature: "Similarly, reducing tissue size limits the number rather than the size of pattern elements generated by a Turing network so that for example, a single 'spot', half a 'spot' (i.e. a gradient) or no 'spot' is generated. " Please see lines 183-185 on page 7 of the revised manuscript. ##################### ***** Reviewer 2 Advance Summary and Potential Significance to Field: The manuscript by Timothy Grocott and co-workers tackles a key question of developmental biology, the pattern formation in early vertebrate embryos. They use patterning of early optic chick vesicle as an experimental model. Their central hypothesis is that a lineage-specific DNAbinding transcription factor Pax6 is a cornerstone of a cross-talk of multiple signaling pathways and these are wired together as a putative Turing network. The experiments test this model through three components of TGF/BMP signaling, Bmp4,follistatin (Fst), and Tfgb2 at their mRNA levels.
The main findings in the paper are summarized in a model that accounts for differences of Pax6 regulation in distal and proximal regions of the optic vesicle (Fig. 6). Overall, these findings are novel and within the scope of journal Development.
Reviewer 2 Comments for the Author: There is no direct relationship established between the Pax6 and Vsx2 expression in the retina. Additional genes, such as Fgf9 and Erk1/2, should be examined.

#####################
We thank the reviewer for this observation. Although we do not state that Vsx2 is a direct target of Pax6, we accept that we could be much clearer on this point.
The interaction between Pax6 and Vsx2, and those between Tgfb2 and Mitf/Wnt2b are now indicated with broken lines in Fig. 7A-B (previously Fig. 6a-b). The legend for Fig. 7 now includes the text "Interactions indicated by broken lines may be indirect". Please see line 834 on page 27 of the revised manuscript. ##################### Auto-regulation of Pax6 in the optic vesicle was probed through dnPax6 mis-expression. This experiment requires additional data to determine levels of ectopic expression in relationship to their endogenous levels.
The main text now includes the following: "To confirm that dnPax6 was overexpressed relative to endogenous Pax6, an N-terminal riboprobe was used to collectively detect both endogenous Pax6 and exogenous dnPax6 expression (Fig. 2G)". Please see lines 117-119 on page 4 of the revised manuscript.
We apologise if this was unclear and now explain this point with a paragraph outlining Tgfbsuperfamily signalling in the introduction. Please see lines 48-57 on page 2 of the revised manuscript. ##################### The central question and model should be further probed using single cell RNA-seq approach… ##################### While we thank the reviewer for this suggestion, we agree with the editor that scRNA-Seq is beyond the scope of the current study.
Since scRNA-Seq obliterates the spatial relationships between cells, it is not obvious that this approach would address our central question regarding self-organisation of spatial patterns. In any case, we would be unable to perform these experiments within the normal timeframe for revision. ##################### …and/or future experiments to expand and independently validate the present model should be outlined in the Discussion.
The discussion now includes the following paragraph highlighting mutual repression of Pax6 and Pax2: "In Model D we accounted for intrinsic positional information by incorporating direct suppression of Pax6 expression by a ventral-high to dorsal-low gradient of Shh activity ( Fig. 5; Supplementary Information) (Ekker et al., 1995;Macdonald et al., 1995). This is a convenient abstraction however; at later stages, the ventral extent of Pax6 expression in vivo is refined via reciprocal inhibition between distal Pax6 (prospective neural retina) and ventral Pax2 (prospective optic stalk) (Schwarz et al., 2000), whose own expression is activated by ventral Shh (Ekker et al., 1995;Macdonald et al., 1995)." Please see lines 364-370 on page 12 of the revised manuscript. ##################### 8) References are not in Development format.

Other changes to the manuscript:
We have taken the opportunity to add sub-headings to the Results and Discussion sections.
The model equations have been improved to prevent divide-by-zero errors arising during simulations. These are fully described in Supplementary Information. This was a necessary change to permit 2-D simulations on arbitrary shaped domains (i.e. simulation of explant experiments, where pixels outside of the explant domain have zero concentration values). The simulations in Fig.  4 were repeated using the revised equations and were found to give qualitatively identical results.
Simulation data and Pax6 immunofluorescence data in Fig. 4  The overall evaluation is positive and we would like to publish a revised manuscript in Development, provided that the referees' comments can be satisfactorily addressed. Referee 1 requests that you include the data showing that changing the direction of the Shh gradient reverses the polarity of Pax6 expression. I agree that adding these data will help illustrate the model and strengthen the argument. Please attend to all of the reviewers' comments in your revised manuscript and detail them in your point-by-point response. If you do not agree with any of their criticisms or suggestions explain clearly why this is so.
We are aware that you may currently be unable to access the lab to undertake experimental revisions. If it would be helpful, we encourage you to contact us to discuss your revision in greater detail. Please send us a point-by-point response indicating where you are able to address concerns raised (either experimentally or by changes to the text) and where you will not be able to do so within the normal timeframe of a revision. We will then provide further guidance. Please also note that we are happy to extend revision timeframes as necessary.

Reviewer 1
Advance summary and potential significance to field Grocott et al propose that a Turing-like network, comprising Pax6/Fst/Tgfb2, is involved in polarising the optic vesicle along its proximal-distal axis. Their revisions have satisfied the concerns raised in my original review, and I recommend this article for publication.

OVERALL COMMENTS
The authors added substantial additional data and these significantly improve the manuscript.
Showing the re-orientation of the Pax6 pole in explants (and in silico) is a great demonstration of a self-regulating system. The TGFB and HH perturbations further strengthen their TGFB-centric model; Fig 5E,G,I are a highlight. I enjoyed reading the revision! MINOR COMMENTS Minor comments made previously have been addressed (the discussions/figs on geometry work well).
Small point: on line 229 it says that data "not shown"; could you include it?

Reviewer 2
Advance summary and potential significance to field The authors have addressed all major points of criticism and very much improved their manuscript. They examine very important developmental process and compare their findings with selforganizing formation of optic cups from mammalian ES cells. They probe their system through a small set of proteins and clearly demonstrate how their mutual regulation can elicit the patterning events.

Comments for the author
None.