Atrophin controls developmental signaling pathways via interactions with Trithorax-like

[…] Reviewer #1 (Minor Comments):

1) Figure 1. How are the 1275 S1-3-5 peaks derived from the 1377 common peaks mentioned in the text?

The 1275 peaks in Figure 1D are only the peaks in the major chromosome arms from the 1377 common peaks. The peaks in the heterochromatin regions (e.g. chr3Rhet) are excluded, hence there are less peaks in Figure 1D than the common 1377 peaks. This has been clarified in the main text and in the figure legends.

2) Subsection “Atro modulates developmental gene expression during larval development”, first paragraph: "two strong Atro peaks" upstream of the en TSS are described as being "in the en promoter". What are the limits of the en promoter and has it been defined? Similarly, subsection “Atro directly regulates Dpp signaling via thickveins”: "Atro bind's tkv's regulatory region"; what is this region and has it been defined? Also similarly, subsection “Atro is required for Notch signaling and directly regulates Notch pathway gene expression”, second and last paragraphs: "Atro binds to the regulatory region of fng" and "regulatory regions of multiple N signaling components"; what is the evidence these are regulatory regions?

Corresponding sources for promoter and regulatory gene regions have now been cited in the main text. For genes whose regulatory regions were not formally defined, these were changed to near putative regulatory regions of potential target genes instead of regulatory regions.

3) The clonal analysis data is not always compelling; particularly the enhancement in Figure 2 B' is difficult to see in the image provided.

In the wing, endogenous En is expressed very strongly in the wing pouch. Hence, it is difficult to see a clear upregulation caused by Atro³⁵ clones. To counter this problem, we generated new Atro³⁵ wing clones and visualized En using weaker fluorescent secondary antibody to show a clearer upregulation of En. Accordingly, this new figure replaced the old Figure 2B.

4) As the suggestion of a direct complex between Atro and Trl seems to me to be a major, or the major, point of interest in the paper, why is the critical co-IP experiment in a supplementary figure?

We agree with this point, which was brought up by both reviewers 1 and 2. In response to their comment, we have moved the coIP experiment to the main figure (Figure 5D).

5) Figure 5. Although most Atro peaks overlap Trl peaks, a sizeable minority do not. Is this just a genomic data thresholding issue or is there good evidence for Atro binding in the absence of Trl? The peak the authors use in Figure 6 seems to be a clear example but it would be good if the authors could comment in general on the Atro sites that lack Trl and how they fit in their model.

We believe that Atro can interact with other transcription factors other than Trl to associate to DNA. For example, Atro was shown to physically interact with Tailless and Cubitus Interruptus (Wang et al., 2006; Zhang et al., 2013). Thus this may explain why some Atro peaks do not intersect with Trl.

In addition to the above, we also believe that some Atro peaks that do not intersect with Trl in our analysis are weaker than the peaks that do. In order to test this we used the data generated by MACS2 peak calling:

MACS2 provides a –log(pvalue) value for each corresponding peak. This value ranks the significance of each of the peaks, where MACS2 assigns a higher –log(pvalue) for highly significant peaks. Using this data, we plotted box plots for all Atro peaks that intersect Trl and all Atro peaks that do not intersect with Trl, and then compared the two. The plots show that Atro peaks that do not intersect with Trl have slightly less –log(p-value) than that of Atro peaks that intersect Trl (Wilcoxon test p-value < 0.001). MACS2 also generated a fold-enrichment value for each of the called peaks and we plotted box plots using these values as well. We saw that Atro peaks that do not intersect with Trl have less fold enrichment than Atro peaks that do (Wilcoxon test p-value < 0.001). These data suggest that the Atro peaks that do not overlap with Trl are less significant than those that intersect.

Therefore, we believe that the Atro peaks that do not intersect Trl can be explained by: 1) Atro interacting with factors other than Trl and 2) Atro peaks that do not intersect with Trl tend to be less significant than the peaks that do intersect. The above points have been added to the main text and the plots have been added as Figure 5—figure supplement 2.

Reviewer #2 (Minor Comments):

1) The choice about which data to include in the supplemental figures sometimes seems odd. For example, the co-IP data seem central to their conclusions and so I don't understand why they are relegated to a supplemental figure.

We agree with the reviewers regarding this and in response, we have moved the co-IP figure to the main figure (Figure 5D).

2) Perhaps it's there somewhere, but I can't readily find the source of the genomic data that was used to generate the graph shown in Figure 5B.

The sources of the genomic data are from modencode.org, and Philip et al., 2015 and Oh et al., 2012 for CPB and Yki, respectively. Only Atro peaks from major chromosome arms (e.g. heterochromatin excluded) were used for this figure. This information has been added to the main text as well as in the figure legends.

3) Does bioinformatic analysis give any clue as to the difference between the Trl peaks that bind Atro and those that do not?

To address this, we repeated our bioinformatics analyses on Atro for Trl peaks. We first looked at the distribution of Trl peaks that do not intersect with Atro and Atro peaks that intersect with Trl to verified genomic regions (e.g. promoters, introns). However, we could not see any obvious differences between the two sets of data.

We also performed principal component analysis (PCA) of Atro peaks and the PCA indicated that Atro peaks can be grouped into 3 different classes. The PCA is now included as a figure supplement (Figure 1—figure supplement 1). We note that more experiments are needed to reveal the biological significance of these 3 groups.

4) Given that the co-IP analysis was performed on MNase treated crude lysates, the interaction between Atro and Trl could be very indirect. Do the authors think it depends on the presence of chromatin fragments in the lysates?

We agree with the reviewer that the interaction could be indirect and we have toned down our discussion accordingly. Currently, we do not know if the Atro-Trl interaction requires the presence of chromatin but the requirement of MNase to see a coIP of Atro and Trl suggests that Atro and Trl are interacting near chromatin.

5) Some of the ChIP experiments used double cross-linking. Is this because they found that the double cross-linking enhanced their signal?

Double cross-linking was crucial for the ChIP with the SG2524 anti-Atro antibody. Without double cross-linking, the Atro ChIP could not enrich for positive controls during the ChIP optimization. This point has been clarified in the Materials and methods section.