small typo: "groups"

small typo: "evidence of"

How did the present study account for the type of food (formula-, breast-, and combo-fed) in infants or the timing of introduction of solid food?

it is a little unclear whether some individuals were sampled at multiple ages. The description in the methods describes them as independent cohorts, but repeated measures are mentioned here.

What type of physiological processes might increase microbial diversity during infant wakefulness? Do you think that longer wakefulness is correlated with higher sampling of the environment (ingesting microbes) and that increases microbial diversity?

It was interesting that two samples from CMB clustered with A. exaltata in the PCA shown in Fig. S6. What do you make of that, given that those samples came from a site isolated from the others?

A small note is that the figure caption for Fig. S4 appears incorrect and is the same for Fig. S6

I found it fascinating that the two populations of milkweed have such different reproductive microbiomes and that the bacteria isolated from A. syriaca may be negatively affecting pollination in the host. It would be interesting to follow up on the fungal and bacterial isolates to determine if 1) the yeast in each host differ dramatically and 2) if the yeast from A. exaltata are outcompeting bacteria or directly inhibiting their survival in the host.

How a host affects symbiont fitness is an interesting area. Do you think that the dead hosts provide a food source specifically to the symbionts as they transition to the soil environment or to many soil microbes? Is there any reason to believe that the bacterial symbiont can utilize the dead bugs to outcompete other soil microbes?

MicroSplit is an exciting technique that could drastically improve the feasibility of multiplexed imaging. I'm wondering if the amount of training data differs depending on the training mode. For example, does training mode III represent more of a ground truth because it's based on images of each channel separately and thus requires less data? Is that the preferred training mode when possible, or does the model perform equally well with all training modes?

The interpretability of this figure could be enhanced with some minor tweaks. 1) Labels for each column to distinguish plots from experiments 1, 2, or combined would be helpful if it is necessary to show all three plots for each variable. Alternatively, you could show only the combined experiments and put the plots for each experiment in the supplemental.

2) The plots show larval settlement in response to a given variable in two states: "sterilized" and "untreated." I do not see where in the methods or results these two conditions are described, and I found this confusing.

3) In addition, it's difficult to quickly get a sense of which factors significantly impacted larval settlement. Labeling the points with different letters invokes that they indicate significant differences determined by the Tukey's post hoc test. However, the table shows that many of these factors did not impact settlement as determined by the model. Making these distinctions very clear to the reader in a visual way would greatly enhance the impact of the results.

Is there any way to determine that the concentration of the predator cue was high enough to impact the oyster larvae? Is there another readout that could be used as a positive control to know that the larvae could sense the cue but did not respond?

This beautiful study highlights the power of SRS microscopy for discovering and following up on important observations. If addressed, I have a couple of questions that would provide clarity to the reader.

1. The inset graph showing the normalized pure spectra of proteins and lipids is critical for understanding the protein and lipid channels used throughout. I see where the normalization is explained in the methods, but I didn't find the details about how those curves were obtained. What was used for the pure protein and lipid samples? How many wavenumbers were sampled (just the 5 with the tick marks on the x-axis)?

2. In Figure 1, it would be helpful to add markings to point out the additional structures described in the figure caption for non-experts in C. elegans.

3. In Figure 2, the inset really demonstrates the protein buildup in the pseudocoelomic space. Did you notice that that region on the worm was particularly striking across treatments? Would comparing the quantification in those FOV show a more pronounced effect than the entire worm?

4. The schematic showing the experimental design in Fig 1G. is helpful, but I didn't see the reasoning in the text to explain why the daf-2 mutants weren't exposed to the calorie restriction treatment like for ceh-60 worms.

5. In Figure S1, were the protein or lipid quantifications ever normalized to the area of the worm?

6. What do you make of the decreased protein buildup in ceh-60 CR worms as compared to ceh-60 fed (Fig. 6F). I see that there was no additive effect of CR on the mutants in terms of lifespan, but I found it interesting that it did look significantly lower in this plot.

This is a cool application of virtual staining that would eliminate the need for a live-dead stain. Did you confirm the brightfield morphologies with a viability stain?

Examining the cell's eccentricity may be interesting if it captures something not reflected in the circularity calculation.

Was 561 nm the optimal excitation for chloroplasts in this sample? I would expect the optimal excitation to be either between 400-460 nm or 600-750 nm.

A more standard way to report the objective would be "Olympus UPlanApo 20x"

Were all images acquired on the same day, or were these compiled across multiple imaging days? Given the day-to-day variability that imaging experiments can have (sample prep and microscope performance), it would be worth noting the number of images acquired per day that ended up in the dataset. How was the splitting of images into training and test datasets determined?

Is it possible to disturb the ANG microbial community by using antibiotics? If so, it would be interesting to look at how th populations rebound during the recovery phase.

Do you suspect that populations within more tubules are mixed if you were able to resolve lower levels of phylogenetic classifications?

This observation is intriguing. Do you think that this is a rare occurrence, given how only 50-75% of the ANGs were occupied by symbionts? Any ideas what might trigger the peristaltic motion and whether it could be triggered by environmental light, like the light organ venting process?

This model sheds new light on the biophysical processes involved in ANG microbiogeography. I noticed that protrusions are lacking from the epithelial cells lining the tubules deep in the ANG (Fig 7B). Do you have any insight into whether these epithelia are microvillous like the light organ crypts or ciliated like the preceeding portions of the ANG and intergland space?

Were these slides sequential slices from the same ANG or were they ANGs from the five different animals?

It would be interesting to gain further insight into the physiological differences when grown in the two types of media.

Why do you think the multicellular cluster size decreased over time for 3 out of the 4 PS strains? Was this surprising or a sign of instability?

These cluster boundaries look hard to delineate by eye. I wonder if you considered using a stain (like Calcafluor white) to make the cell walls more obvious to the observer.

It would help orient the reader to have a visual representation of this part of the results. A schematic could show that multicellular phenotypes emerged at different times, but a subset of those that evolved under the predator-selective regime were stable. This could have the strain names with the stable multicellular ones highlighted so the reader can easily reference this figure later when looking at the results that follow up on those strains (like Figs. 3 & 4).

It's more accurate to call these micrographs rather than microscopic images.

The first few paragraphs of the results section are more suitable for the introduction. I would consider revising the introduction and integrating the information presented as results to better prepare the reader for this study's real results.

It is hard to extract much information from Fig 6A with how the spectra are currently shown. It may be more useful to show each spectrum separately and indicate the corresponding organelle. Then, indicate the three sections of the spectrum that are shown in 6B-D.

Is the y-axis showing intensity normalized?

How did you identify these peaks? Was it based on visual observation of differences in intensity across wavenumbers, or was there a more analytical approach taken?

What light source was used and what was the duration of the videos for the motility assay? Given their negative phototaxis, I'm wondering if red light was used for imaging.

How long were the cells incubated in the nitrogen-free HSM medium before the phototaxis and motility assays? We've observed variation in gametic motility depending on the time spent in water prior to the assay.

How many PCs can explain 95% of the variance? And how many features went into the PCA?

It would be useful to show a schematic of the differences in aposymbiotic and symbiotic stony corals, kind of like a graphical abstract. The abstract was hard to follow without reading the details of the study, and it would be nice to see the overall findings in a visual way. This could get across the idea that neither the bacterial communities nor the proportion of immune cells differ between symbiotic states, but the gastrodermis I cells downregulate immune genes when algae is in close contact. That begs the question, will you follow up to determine what chemical cues from the algae are sensed by the gastrodermis cells to alter their gene expression?

I don't think it's necessary to capitalize immune cell while naming the type of cluster.

How do these corals acquire their symbionts? Is it surprising that with 2 months of recovery following the bleaching even that the symbiont load remained so low? Or is there a sensitive window that has closed for corals at this stage of development?

This seems great for surveying plant physiology for ecological field work. I didn't see anywhere an estimation of the throughput though, can any numbers be assigned to this to help a researcher see how the technique would increase teh efficiency of their work?

Can you spell out these acronyms here so the reader can refer to this section when interpreting the y-axes on Fig. 5 and 6? I see that they're listed in the abbreviations, but it would handy to have them written out in this section of the results.

It would be worth reporting the sample sizes for each treatment, as that would help the reader understand why you chose different tests (like Welch's t-test). Also for D, I am assuming you actually performed an ANOVA and then a Tukey's post-hoc test, so you should report the ANOVA statistics (F value, df, p value).

I see now that the label for Darkened should really be "Dark-adapted", is that true?

Same comment as for Fig. 1 but also it would help to make the label "Dark" more parallel. Is that a "No light" treatment as control? Consider new labels to make it clearer as to whether Dark means a condition or an observation as in Fig 1, "Darkened" seems like an observation.

Interesting, what is observed by eye? Is it a darkening of a green hue?

small note - but consider labeling the panels "low" and "high" blue light and keep the info on the values for each treatment in the caption. This would improve readability.

Was the F statistic from the ANOVA significant and were the comparisons actually between drug treatment and the control? If so, I would think a Tukey's post-hoc test would be more appropriate. If the ANOVA was not significant, then why are none of the p values from the Sheffe post-hoc test not significant? I was suprised because it looks like in panel d that the control and treatments look different.

Wow, this is a really important finding that reshapes how we think about cells can efficiently and rapidly alter their architecture in response to environmental changes.

This seems like an extensive study, but the figures have so many panels and the captions are so long, that it make it hard for the reader to quickly understand the key points. It would greatly benefit the reader if you streamline the figures and reduce redundancy within the captions (across letters in the panels and also with methods sections). Because of the different time courses, perhaps a schematic showing the overall experimental design in the first panel woud be useful for some figures.

Is there a more direct way to state the findings?

Just a note for readers outside of the Arabidopsis field that EDS1 hasn't been defined yet in the abstract.

This is an interesting approach to ensure the classifier picks up on cytoplasmic features to better differentiate cell stages. You may be interested in this pub, which uses a ResNet model on brightfield timelapses of nematodes. It would be interesting to see how the brightfield data (available in the GitHub) perform with the four models explored in your study given potential differences with intracellular hallmarks that could be more apparent by DIC.

This is an interesting application of machine learning to expand the phenotyping capacity of cell types that are hard to distinguish without more intensive or destructive methods.

Do you suspect there are any intrinsic differences in the Raman or NIR spectral for M1 or M2 macrophages without the addition of the DNA-SWCNTs? It would be interesting to know to what extent this works as a label-free approach for applications without the addition of single-walled carbon nanotubes.

I found panels A-C confusing and have questions that may help you clarify to the reader. Is the grayscale image to the right of the four panels a merge of the 4 markers shown in C? If so, I might write merge on it. I was trying to connect the panels in C to the red line in A because I thought it was an example of how you might classify CD4+ T cells based on the high signal for CD3, CD 45, CD4, and CD8. But then I saw the image to the right of them. Regardless, I would switch the top two panels in C so that they read in the order of the markers in A and B (CD3 and then CD45).

This tool seems quite robust for classifying many cell types based on multiple markers. I'm wondering if you see a useful application of the tool as identifying unusual combinations of markers that deviate from the training data?

I really like the illustration of what this study found in the graphical abstract. I don't see where the schematic incorporates the hypoxia component. If that is relevant then it may be worth making that clearer.

It would be helpful to the reader to hear more about how the resolution phase was determined.

It would be interesting to determine which of the 6 different parameters explained most of the variation in lung architecture in case it only takes a couple measurements to assess response to inflammation.

I'm interested in whether you saw a difference in staining between granules that were extracellular to the MCs or the ones associated with changes in pericyte morphology.

Very much appreciate the level of detail for the choices for each step of the methods. This makes understanding the adjustable parameters and functions of RamanSPy much more approachable.

It would be helpful to hear more about how the level of robustness was determined.

I'm curious about the authors' opinion about whether a pattern needs to be apparent to the human eye in order to be detected by a computer. These statements suggest that the supervised models perform worse than autoencoders like the one described here.

I'm curious whether the authors considered using the model to identify specific organelle contacts of interest to further investigate using deconvolved images and confirming the contact.

typo, mitochondria

typo, should be imaged

This is useful to know. Do you have a sense that 15,000 images was near the minimum number needed or whether it was overkill for the training?

Texture is something that can be harder to quantify than cell shape so it is great that it can be captured by models like this one since it improves the morphologic profiles.

It is interesting to see the stratification of chloroplast signal between the mNG fluorescence and chloroplast autofluorescence. Is the bulk of the nucleus within the empty space of the chloroplast autofluorescence? What accounts for the DAPI staining throughout the cell body?

Would it be possible to synthesize this reporter with a lower MW to explore the heterogenous distribution further?

Given that you identified two evolved phages within a mouse, I'm interested in your speculation as to how frequently you think this occurs in nature. Do you think that the two phages occupy the same intestinal niche or is there evidence that they are stratified along the GI tract?

This seems like a comprehensive and sensitive readout of microbial activity. It might be worth defining SRS as Stimulated Raman Scattering as the reader could mistake it for Spontaneous Raman Scattering.

I'm curious if you can share insight into how to design an SRS imaging experiment to allow for this type of unexpected observation of the time-dependent photothermal signal.

These opposite effects on the same group of strains by each drug is quite interesting. I'm curious about your interpretation as to why the ENT-Hi community looks so similar to the inoculum at 6 h except for the increase in E. coli abundance. One might think there's no effect of ENT-Hi on the community, but something is changing relative to the inoculum to allow E. coli to grow.

Is the major shift for the LOX-Hi and LOX-Low treatments the reduction in Bacteroides that is present at 6h but stronger at 24h? If so, it might be useful to call that out specifically for the reader because the fact that Bacteroides is affected might catch their attention.

Perhaps consider commenting on the differences between supervised, semi-supervised, and unsupervised learning and which you suggest as optimal to reduce photoxicity.

Consider revising into shorter and more declarative sentences.

Revise this to ensure a proper subject.

It sounded like the topic was using D-L models to iterate during acquisition but this refers to applying the model after the imaging experiment (and data are acquired), so please clarify.

define CLEM in this section

see point about Fig. 2B

Eliminate and start with "An"

"rarer"

Consider clarifying the type of confocal referred to here and including both spinning disk and point-scanning confocal methods to this graph.

This is a great point, consider condensing this section to get to this point across sooner.

Consider flipping this sentence to first state that these two phenomena can occur independently but are related.

Consider revising this topic sentence to include photobleaching and eliminate redundancy with the previous paragraph.

typo, eliminate the '

Consider saying, "Many biomolecules undergo degradation when exposed to specific oxygen radicals (Table 1)."

This point is clear through the citations so consider sticking to the point of UV light damaging DNA.

To ensure this point is clear to readers, consider saying, "The use of light irradiation is necessary to excite fluorescence but also produces ROS. In addition to inducing oxidative stress, ROS can also trigger multiscale alterations..."

I might reword it to, "if oxidative processes are unchecked, they can lead to oxidative stress and cellular damage..."

Is this the term that you want to introduce here or is it deep-learning augmented microscopy? It is a little hard for the reader to know which is the new term.

This is an excellent point and a crucial argument for deep learning augmented microscopy.

This is an interesting potential explanation for this decrease in interest in deep learning. Can you expand upon why you suspect this is the more likely explanation? Anything from the survey that supports this?

I found this surprising at first and very helpful to know. It informs the level of detail that we should strive for when describing analysis methods.

This figure is a bit pixelated making it hard to see the text and numbers. If the resolution can be improved, it will help the reader want to study the figure.

This is useful to expand on the details of the statistical methods.

The embedded images in the Multivariate details section did not load on my screen.

What kind of time cost is associated with the training given the scale of this dataset?

This is a huge win for this technique to be able to get both Raman spectra and mass spec data from the same biological sample.

Do the peaks identified in the Raman-mapping inform any parameters for the subsequent MALDI imaging or is it that those peaks provide a basis for the segmentation.

If the biological sample has autofluorescence with 532 nm excitation and one must use a red-shifted laser for the Raman measurements, is there any reason why that would interfere with the MALDI analysis step?

This seems like a very useful tool! I think it would be valuable for the reader if the authors comment on how the tool described in this study differs from or improves upon the 10X Visium software tools that accompany this type of ST dataset.

This figure shows up as a bit pixelated when enlarged. If it can be swapped with one at higher res, that would be useful.

This is useful for breaking down the functions

This is an elegant study that experimentally untangles the role of a plant hormone in fungal colonization and development in host plants. The model proposed based on this work is very thorough as well.

Beautiful work! It is likely that this will be resolved with the journal submission, but just to watch out for the resolution on Fig 2-6, which is too low for viewing in this format.

No italics needed

extra ) here

A very interesting computer-learning technique to improve signal to noise and enhance resolution of sub-cellular structures. The impact of this work will be more directly transmitted with clearer writing. Some specific suggestions are annotated throughout the paper.

One tool that might facilitate this is: https://hemingwayapp.com/

should this be "bounds"?

perhaps "effective" is more direct

without the need for baseline data or loss of spatiotemporal resolution

consider making this a simpler, declarative statement(s)

may want to specify the type of confocal here in addition to the methods.

can you clarify how an objective with increased resolution (higher NA) leads to more scattering?

generalizability might be better here

awkward phrasing

should be "from being"

increased is a better word choice than enlarged. Enlarged means to make bigger in size not magnitude. Increased is more accurate when talking about a larger number.

more precise to say "data with high SNR"?

simplify and make more direct

can eliminate to reduce redundancy

what does fluctuated mean here?

Eliminate this.

specify the type of contamination

odd phrasing here. What is more direct verb that can be used here?

Excellent figure to demonstrate the improvement in resolution with DeepSeMi over the raw images.

can no-compromise results be defined clearly?

what kind?

tenfold

Use of DeepSeMi

suggest saying "fluorophores in different channels. They were: tagBFP-SKL (Ex/Em 402/457 nm),..." to give the excitation and emission spectra for each fluorophore.

using nonlinear microscopy or with a nonlinear microscope (and then specify which type)

which dye concentration was better?

"increasing laser power" ?

what does brittle mean here?

awkward phrasing. Perhaps something like "magnificently well-orchestrated systems of organs, tissues, and cells" or

"drastically improved" would be better

Great point, may be stronger if framed as, "DeepSeMi improves upon interpolation-based denoising methods by (fill in the blank to describe what it is based on), which allows for observation of organelle dynamics without motion artifacts"

revise use of hypens throughout. three hyphens should not be used consecutively.

awkward phrasing. Not sure what "enriched fluorescence microscope" really means

100x

not a standard way to report this.

for which laser?

awkward phrasing. Perhaps it would be better to clearly make the point that organelle interactions are dynamic and studying them with high spatiotemporal resolution is important to understand their function in the cell.

perhaps "but this practice can increase phototoxicity to the live specimens and disrupt the normal morphology" would be more clear

awkard phrasing

computationally enables a tenfold increase in the photon budget

awkward phrasing

"multifold"

Not sure that "ergo" adds much meaning here. If it is wanted here then the punctuation should be, "differentiated; ergo, between"

Is it really videography? Is there a more precise term that can be used here?

any quantitative estimate on the improvement from previous methods?

the concentration of fluorescent dyes 50-fold

assuming these are average rates, is there a +/- standard deviation that can be reported?

Very impressive with such gentle excitation

Can this be clarified?

Great use of specific language here.

is this implying that because of their transparency, there is decreased background/noise? Perhaps this should be clarified.

What was the autofluorescence excited by 488 nm indicating in the samples?

This is an excellent point and justification for this study.

What indicates cell damage from 488 nm autofluorescence?

Interesting finding. Can any additional speculation be presented here?

"An inVia Raman confocal microscope..."

Were any measurements made across the cell using parameters for the maximal spatial resolution (0.25 µm)? It would be good to note (in a supplemental figure perhaps) that there was no difference across the 3 µm cell and justify using the center point of each cell for all of the measurements.

Images would be useful prior to showing the measurements. Also, a title for the figure would be helpful.

What statistical test was used? It was not stated in the methods.

remove a

remove a

What statistical test was used? It was not stated in the methods.

Images would be useful prior to showing the measurements. Also, a title for the figure would be helpful.

Were any measurements made across the cell using parameters for the maximal spatial resolution (0.25 µm)? It would be good to note (in a supplemental figure perhaps) that there was no difference across the 3 µm cell and justify using the center point of each cell for all of the measurements.

"An inVia Raman confocal microscope..."

Interesting finding. Can any additional speculation be presented here?

What indicates cell damage from 488 nm autofluorescence?

This is an excellent point and justification for this study.

What was the autofluorescence excited by 488 nm indicating in the samples?

the concentration of fluorescent dyes 50-fold

is this implying that because of their transparency, there is decreased background/noise? Perhaps this should be clarified.

Can this be clarified?

Great use of specific language here.

awkward phrasing