A timer gene network is spatially regulated by the terminal system in the Drosophila embryo

Essential revisions:

1) Please clarify the presentation of Figure 8. Two reviewers found it hard to follow (see detailed comments below).

We were asked to clarify the presentation of the modelling results from Figure 8. We have moved the genotype-by-genotype description of the model output from the Results section to a new Appendix, and replaced it with a new, narrative summary of the key findings.

2) Please move the "by gene" quantitative comparisons into the main figure, perhaps at the expense of the images of embryos. This change could be limited to the figures that compare genotypes, and all images can still be included as supplements.

We were asked to move the "by gene" quantitative comparisons from the supplementary figures into the main figures that compare gene expression across genotypes (i.e., Figures 4, 6 and 7). We have incorporated the relevant plots as suggested and agree that it is useful to have this information presented in the main figures.

3) The observation about the size of Caudal germline clones should be removed or reinterpreted. The observation is quite tangential to the point of the paper and likely unrelated to anything having to do with Caudal.

We were asked to remove or reinterpret the observation about the size of the caudal germline clones. We have rewritten the relevant paragraph to make clear that we do not know whether the morphological changes are related to the loss of Cad expression or the use of the FLP-DFS system. While we agree the finding is of little relevance to the rest of our study, we have retained the observation in the manuscript as we think it is useful information for anyone exploring our embryo dataset or indeed using the FLP-DFS system themselves.

4) The manuscript should better highlight the broader implications. The data are really nice but the way it is currently written (particularly the Results section) somehow downplays the advance to the extent that it sounds more incremental and niche than it actually is.

We were asked to revise the text (especially the Results section) to better highlight the broader implications of our work. We have completely rewritten the Abstract, partially rewritten the Introduction, and completely rewritten the modelling results subsection of the Results (as mentioned in point 1). We have also streamlined the experimental results subsections by editing the text, adding sub-subheadings to signpost the content of each subsection, and more strongly highlighting the “In summary” paragraphs at the end of each subsection. We should stress that the paper is intentionally written so that a casual reader can get by with only the summary paragraphs. However, the rest of the material is necessary if readers want to be walked through the expression data in the main figures, so we prefer to retain it in the main text.

Reviewer #2 (Recommendations for the authors):

2) The ectopic activation of Dichaete in the neuroectoderm in cad mutants, and therefore inhibition of neuroectodermal activation of D by Cad is not discussed in relation to the model, which simply shows Cad activating D. Can the authors justify this? What changes in neuroectodermal expression are predicted by the model if Cad has a role in suppressing D activation?

We have now clarified in the Results text that we do not attempt to model the D neuroectodermal domain. The neuroectodermal domain is activated at the end of our period of interest and its regulation presumably depends on factors not included in our network, such as dorsoventral patterning genes; it is therefore beyond the scope of our simple model. However, in principle, we see no difficulty with Cad both activating D’s early blastoderm expression and repressing D’s later neuroectodermal expression, for example through different enhancers and/or indirect effects.

3) If I understand the model correctly, inputs such as Tll can be strong, weak, or off. As the authors mention in the Discussion that there is a phenotypic series of tll alleles, could some of these be used to test whether the model can capture the changes in the network expression patterns when the Tll input is weak instead of strong?

It would be very interesting in future to work to compare gene expression across these different alleles, and also tll dosage mutants, to further interrogate the quantitative effects of Tll. (Indeed, while analysing tll- homozygotes for this study we noticed that tail patterning was shifted posteriorly in tll- heterozygotes compared to wild-type embryos, consistent with Tll having concentration dependent effects.) However, it would be preferable to analyse quantitative effects such as these in the context of a fully quantitative model, rather than the minimalist qualitative model we have used for this study. For the avoidance of doubt, we should also stress that by modelling Tll as a qualitative variable that can be strong, weak, or off, we are not claiming that Tll literally switches between discrete “strong” and “weak” regulatory effects at a specific concentration threshold, but rather that we need at least 3 distinct levels of Tll in our model to reproduce the qualitative patterning phenomena we wish to explain.

4) In Fig 8 it would help the reader to include, for some key mutants and time points, the same visual representation of the expression patterns based on the observed staining patterns. This would help the reader make a quick and simple comparison - currently, the authors direct the reader to the Source Data 1 but the data are provided in table form.

We had considered this option when originally deciding how to present Figure 8, but we decided against it for two reasons. Firstly, rigorously converting real embryo data to the minimal “4 region” representation we used in our model would have required so many (potentially arbitrary) decisions about AP axis positions and intensity thresholds that it felt more straightforward and transparent to simply provide the cross-references to the real data in table form. In this table (now Appendix 4—table 1), we describe noteworthy qualitative aspects of the simulations, and provide figure panel references to the corresponding expression data within the paper. Secondly and more importantly, the qualitative recapitulation of real patterning dynamics by the simulations was actually so accurate that there is only really one discrepancy between the real and simulated data (a spurious delay in the activation of Fkh in the simulated hkb- mutant); as a consequence, a visual comparison would have all but duplicated the existing figure. We hope that the rewritten section of the Results now describes the performance of the model much more clearly, obviating the need for a visual comparison.