I'm noticing in multiple figures the annotations of domains and the labelling of silk protein types are pixelated and unreadable inside the figure. This is something to note when exporting the figures in the final paper. I would consider using simpler symbols to signify the different classes of spider silk proteins containing only the name of the protein family.

Similar to my comment about not referring to extant proteins as ancestral, I would not refer to species that are currently alive as "ancient". Early-diverging is more accurate.

I noticed multiple references to these newly discovered extant proteins in an early diverging spider lineage as "ancestral proteins". I would steer clear from that language. Ancestral proteins implies that these proteins are no longer extant, and that they precede the existence of modern proteins. These are usually lost to time, but the term is reserved for proteins created with ancestral sequence reconstruction or archeologically recovered DNA. I would instead refer to these proteins as early-diverging, or belonging to spiders from an early-diverging lineage of Araneae.

I noticed that red arrows are mentioned in the figure, but I am not seeing those arrows in the figure itself. It might work to simply highlight the name of the taxon in red instead.

It is exciting to see a non-linear relationship (epistasis) between these two environmental stressors. To help with interpreting this figure, I suggest that the figure 4 legend include the analysis being done a bit further: explaining what it means for a number to be more or less negative, naming the type of analysis in the first sentence, as done for figure 5, and explaining the meaning of the axis values.

I noticed that for some scatter plots there is a clear correlation shown with a solid line (a), and for others there is no clear correlation and no linear regression shown in the figure (all herbivory resistance scatters, as well as (b). I think it would make sense to follow suit with (c) and (d) and not show dashed 'potential' selection when the p-values do not support a correlation, and the scatter plots don't have a visible correlation.

Do shriveled and yellow leaves get counted when randomly sampling leaves for herbivory damage? I'm assuming these leaves are not counted, and this is what is meant by in the methods 'excluded ambiguous damage'.

I am struggling with the Y-axis '% herbivory per leaf'. Firstly, I am confused about the range. It seems to go from around -2% to 3% damage. However, in the methods, damage values are explained as being the percentage of the leaf surface damaged by herbivory, averaged across four leaves). How can there be negative herbivory damage? Also, perhaps the Y axis title should be something like '% leaf area damaged'

I'm curious what leaf damage due to glyphosate versus leaf damage do to herbivores looks like. It would be nice to see what each of these looks like in a figure, maybe as a supplement. For both, leaf damage is used as an important measure of the effects of stressors on the health/fitness of the plant. This section would benefit from a citation to point to the previous work referenced here that shows this software can exclude one damage from the other.

This paper was exciting to begin with, promising to trace the evolutionary history behind the rare acquisition of a purple endosymbiont. It ended up being even more exciting, expanding our knowledge of Pseudoblepharisma species, piecing together deep evolutionary information cross four different bacterial and eukaryotic families to succinctly paint not just the steps that facilitated the acquisition of a purple endosymbiont, but also the subfunctionalization of two neighboring species adapted to different oxygen environments.

Some of the most important branches on this tree seem to have low support, and the two trees are rather different. Perhaps creating a concatenated alignment and using that to create a tree will help to improve branch support and clarify species' relationships.

The heme can play a role in getting capturing oxygen that might get in the way of anoxic reactions, much like in legume root nodules. It might also be doing the opposite, helping to provide oxygen for other functions matching the more aerobic environment of this species.

I think you mean 3M, instead of 3N. There are also different colorless elements in these images like the starch granules (which are labelled) but the bacteria are not labelled. Please highlight them.

I think it would be interesting to see some of these values (# of algal cells, length, etc) displayed quantitatively as bar graphs with error bars. It would facilitate comparisons to have a similar figure for P. tenue showing cyst stages, free-swimming cells. etc. It could be provided as a supplement.

The scale bar for TBCC008 is missing.

All the important branches are strongly supported, as shown by the bold branches. I appreciate the use of SH-aLRT statistics. The phylogeny clearly points to an evolutionary history: purely heterotrophic > containing green algal endosymbionts > containing purple endosymbionts. Sampling more species would be helpful in confirming whether the taxa are truly cleanly separated as 'obligate heterotrophs' and 'photosynthesic endosymbionts', but expanding the tree this much is already an extraordinary finding.

Overall, the work carried out here is exciting for the reasons mentioned. The ancestral protein reconstruction was carried out rigorously, the predicted proteins were of very high confidence across sites, and the localization of the proteins clearly shows promiscuous functions that allow us to picture the evolution of essential endomembrane system functions. I think the questions being asked are very exciting, and this is a perfect use case for ASR. I look forward to seeing the functional characterization of these ancestral proteins taken further to include the other nodes of interest, validation in different eukaryotic systems, and the characterization of more essential protein functions.

In coming papers, it would be great to see the functional characterization of these ancestral gene expanded much further. Localization is a great place to start, since much of the function of these genes is related to where they are being expressed. However, most of the functions of these extant proteins listed here were tested in the ancestral proteins. These test could be as simple as looking for morphological differences in organelles, or imaging trafficking inside of mutant cells. These are essential genes, so this could be done with inducible knockouts and gene rescues.

really appreciate seeing not just the localization, but also the mechanisms of localization being tested in distantly-related eukaryotes. It really helps to confirm the functions of specific residues/domains and further supports that these traits were foundational to the early sub-functionalization of the endomembrane proteins.

This is the first section of the text where '3KQ' and '4KQ' are explained, both in what they mean and why the mutations were carried out. However, these first come up in Figure 3. For that reason, I would move up the mention of these definitions in the text, or add it into the figure description.

Looking at figure 2a, it appears that some sites that are of low confidence using one method are identical across both methods. For these low confidence sites, it is custom to test a secondary sequence that incudes the second most likely residues. I couldn't find supplement table 2, so it is possible you've already done that but i just couldn't look at the sequences to verify, and it wasn't clear from the text either.

It is common to find multiple gene duplication events inside of taxa, especially those representing multiple highly divergent species. Make sure you looked for signs of this in the unconstrained tree to see if there might be some unexpected taxa within your four main taxa. I do agree that it is very important to do species-relationship-guided constraints in creating these trees.