Can we give some links here to data and code? Even a setting to 'turn on the code' or some such? I like to do this for transparency and other reasons we've discussed.

OK we do link the github repo -- maybe add a note about that, and about how to find the code that produces this blog for people who want to see/check?

Based on the Washington Post article, it seems as if AI software programmers don't care about copyright and licenses.

This seems to be a problem beyond copilot. I have seen this outcry regarding every popular AI software.

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

In this paper, the researchers aimed to address whether bees causally understand string-pulling through a series of experiments. I first briefly summarize what they did:

- In experiment 1, the researchers trained bees without string and then presented them with flowers in the test phase that either had connected or disconnected strings, to determine what their preference was without any training. Bees did not show any preference.

- In experiment 2, bees were trained to have experience with string and then tested on their choice between connected vs. disconnected string.

- experiment 3 was similar except that instead of having one option which was an attached string broken in the middle, the string was completely disconnected from the flower.

- In experiment 4, bees were trained on green strings and tested on white strings to determine if they generalize across color.

- In experiment 5, bees were trained on blue strings and tested on white strings.

- In experiment 6, bees were trained where black tape covered the area between the string and the flower (i.e. so they would not be able to see/ learn whether it was connected or disconnected).

- In experiments 2-6, bees chose the connected string in the test phase.

- In experiment 7, bees were trained as in experiment 3 and then tested where the string was either disconnected or coiled i.e. still being 'functional' but appearing different.

- In experiment 8, bees were trained as before and then tested on a string that was in a different coiled orientation, either connected or disconnected.

- In experiments 7 and 8 the bees showed no preference.

Strengths:

I appreciate the amount of work that has gone into this study and think it contains a nice, thorough set of experiments. I enjoyed reading the paper and felt that overall it was well-written and clear. I think experiment 1 shows that bees do not have an untrained understanding of the function of the string in this context. The rest of the experiments indicate that with training, bees have a preference for unbroken over broken string and likely use visual cues learned during training to make this choice. They also show that as in other contexts, bees readily generalize across different colors.

Weaknesses:

(1) I think there are 2 key pieces of information that can be taken from the test phase - the bees' first choice and then their behavior across the whole test. I think the first choice is critical in terms of what the bee has learned from the training phase - then their behavior from this point is informed by the feedback they obtain during the test phase. I think both pieces of information are worth considering, but their behavior across the entire test phase is giving different information than their first choice, and this distinction could be made more explicit. In addition, while the bees' first choice is reported, no statistics are presented for their preferences.

We agree with the reviewer that the first choice is critical in terms of what the bumblebees have learned from the training phase. We analyzed the bees’ first choice in Table 1, and we added the tested videos. The entire connected and disconnected strings were glued to the floor, the bees were unable to move either the connected or disconnected strings, and avoid learning behavior during the tests. We added the data of bee's each choice in the Supplementary table.

(2) It seemed to me that the bees might not only be using visual feedback but also motor feedback. This would not explain their behavior in the first test choice, but could explain some of their subsequent behavior. For example, bees might learn during training that there is some friction/weight associated with pulling the string, but in cases where the string is separated from the flower, this would presumably feel different to the bee in terms of the physical feedback it is receiving. I'd be interested to see some of these test videos (perhaps these could be shared as supplementary material, in addition to the training videos already uploaded), to see what the bees' behavior looks like after they attempt to pull a disconnected string.

We added supplementary videos of testing phase. As noted in General Methods, both connected and disconnected strings were glued to the floor to prevent the air flow generated by flying bumblebees’ wings from changing the position of the string during the testing phase. The bees were unable to move either the connected or disconnected strings during the tests, and only attempted to pull them. Therefore, the difference in the friction/weight of pulling the both strings cannot be a factor in the test.

(3) I think the statistics section needs to be made clearer (more in private comments).

We changed the statistical analysis section as suggested by the reviewer.

(4) I think the paper would be made stronger by considering the natural context in which the bee performs this behavior. Bees manipulate flowers in all kinds of contexts and scrabble with their legs to achieve nectar rewards. Rather than thinking that it is pulling a string, my guess would be that the bee learns that a particular motor pattern within their usual foraging repertoire (scrabbling with legs), leads to a reward. I don't think this makes the behavior any less interesting - in fact, I think considering the behavior through an ecological lens can help make better sense of it.

Here we respectfully disagree. The solving of Rubik’s cube by humans could be said to be version of finger-movements naturally required to open nuts or remove ticks from fur, but this is somewhat beside the point: it’s not the motor sequences that are of interest, but the cognition involved. A general approach in work on animal intelligence and cognition is to deliberately choose paradigms that are outside the animals’ daily routines-this is what we have done here, in asking whether there is means-end comprehension in bee problem solving. Like comparable studies on this question in other animals, the experiments are designed to probe this question, not one of ecological validity.

Reviewer #2 (Public Review):

Summary:

The authors wanted to see if bumblebees could succeed in the string-pulling paradigm with broken strings. They found that bumblebees can learn to pull strings and that they have a preference to pull on intact strings vs broken ones. The authors conclude that bumblebees use image matching to complete the string-pulling task.

Strengths:

The study has an excellent experimental design and contributes to our understanding of what information bumblebees use to solve a string-pulling task.

Weaknesses:

Overall, I think the manuscript is good, but it is missing some context. Why do bumblebees rely on image matching rather than causal reasoning? Could it have something to do with their ecology? And how is the task relevant for bumblebees in the wild? Does the test translate to any real-life situations? Is pulling a natural behaviour that bees do? Does image matching have adaptive significance?

We appreciate the valuable comment from the reviewer. Our explanation, which we have now added to the manuscript, is as follows:

“Different flower species offer varying profitability in terms of nectar and pollen to bumblebees; they need to make careful choices and learn to use floral cues to predict rewards (Chittka, 2017). Bumblebees can easily learn visual patterns and shapes of flower (Meyer-Rochow, 2019); they can detect stimuli and discriminate between differently coloured stimuli when presented as briefly as 25 ms (Nityananda et al., 2014). In contrast, causal reasoning involves understanding and responding to causal relationships. Bumblebees might favor, or be limited to, a visual approach, likely due to the efficiency and simplicity of processing visual cues to solve the string-pulling task. ”

As above, it worth noting that our work is not designed as an ecological study, but one about the question of whether causal reasoning can explain how bees solve a string-pulling puzzle. We have a cognitive focus, in line with comparable studies on other animals. We deliberately chose a paradigm that is to some extent outside of the daily challenges of the animal.

Reviewer #3 (Public Review):

Summary:

This paper presents bees with varying levels of experience with a choice task where bees have to choose to pull either a connected or unconnected string, each attached to a yellow flower containing sugar water. Bees without experience of string pulling did not choose the connected string above chance (experiment 1), but with experience of horizontal string pulling (as in the right-hand panel of Figure 4) bees did choose the connected string above chance (experiments 2-3), even when the string colour changed between training and test (experiments 4-5). Bees that were not provided with perceptual-motor feedback (i.e they could not observe that each pull of the string moved the flower) during training still learned to string pull and then chose the connected string option above chance (experiment 6). Bees with normal experience of string pulling then failed to discriminate between connected and unconnected strings when the strings were coiled or looped, rather than presented straight (experiments 7-8).

Weaknesses:

The authors have only provided video of some of the conditions where the bees succeeded. In general, I think a video explaining each condition and then showing a clip of a typical performance would make it much easier to follow the study designs for scholars. Videos of the conditions bees failed at would be highly useful in order to compare different hypotheses for how the bees are solving this problem. I also think it is highly important to code the videos for switching behaviours. When solving the connected vs unconnected string tasks, when bees were observed pulling the unconnected string, did they quickly switch to the other string? Or did they continue to pull the wrong string? This would help discriminate the use of perceptual-motor feedback from other hypotheses.

We added the test videos as suggested by the reviewer, and we added the data for each bee's choice. However, both connected and disconnected strings were glued to the floor, and therefore perceptual-motor feedback was equal and irrelevant between the choices during the test.

The experiments are also not described well, for my below comments I have assumed that different groups of bees were tested for experiments 1-8, and that experiment 6 was run as described in line 331, where bees were given string-pulling training without perceptual feedback rather than how it is described in Figure 4B, which describes bees as receiving string pulling training with feedback.

We now added figures of Experiment 6 and 7 in the Figure 1B, and we mentioned that different groups of bees were tested for Experiments 1-9.

The authors suggest the bees' performance is best explained by what they term 'image matching'. However, experiment 6 does not seem to support this without assuming retroactive image matching after the problem is solved. The logic of experiment 6 is described as "This was to ensure that the bees could not see the familiar "lollipop shape" while pulling strings....If the bees prefer to pull the connected strings, this would indicate that bees memorize the arrangement of strings-connected flowers in this task." I disagree with this second sentence, removing perceptual feedback during training would prevent bees memorising the lollipop shape, because, while solving the task, they don't actually see a string connected to a yellow flower, due to the black barrier. At the end of the task, the string is now behind the bee, so unless the bee is turning around and encoding this object retrospectively as the image to match, it seems hard to imagine how the bee learns the lollipop shape.

We agree with the reviewer that while solving the task in the last step during training, the bees don't actually see a string connected to a yellow flower, due to the black barrier. Since the full shape is only visible after the pulling is completed and this requires the bee to “check back” on the entire display after feeding, to basically conclude “ this is the shape that I need to be looking for later”.

Another possibility is that bumblebees might remember the image of the “lollipop shape” while training the bees in the first step, in which the “lollipop shape” was directly presented to the bumblebee in the early step of the training.

We added the experiment suggested by the reviewer, and the result showed that when a green table was placed behind the string to obscure the “lollipop shape” at any point during the training phase, the bees were unable to identify the connected string. The result further supports that bumblebees learn to choose the connected string through image matching.

Despite this, the authors go on to describe image matching as one of their main findings. For this claim, I would suggest the authors run another experiment, identical to experiment 6 but with a black panel behind the bee, such that the string the bee pulls behind itself disappears from view. There is now no image to match at any point from the bee's perspective so it should now fail the connectivity task.

Strengths:

Despite these issues, this is a fascinating dataset. Experiments 1 and 2 show that the bees are not learning to discriminate between connected and unconnected stimuli rapidly in the first trials of the test. Instead, it is clear that experience in string pulling is needed to discriminate between connected and unconnected strings. What aspect of this experience is important? Experiment 6 suggests it is not image matching (when no image is provided during problem-solving, but only afterward, bees still attend to string connectivity) and casts doubt on perceptual-motor feedback (unless from the bee's perspective, they do actually get feedback that pulling the string moves the flower, video is needed here). Experiments 7 and 8 rule out means-end understanding because if the bees are capable of imagining the effect of their actions on the string and then planning out their actions (as hypotheses such as insight, means-end understanding and string connectivity suggest), they should solve these tasks. If the authors can compare the bees' performance in a more detailed way to other species, and run the experiment suggested, this will be a highly exciting paper

We appreciate the valuable comment from the reviewer. We compared the bees' performance to other species, and conducted the experiment as suggested by the reviewer.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Smaller comments:

Line 64: is the word 'simple' needed here? It could also be explained by more complex forms of associative learning, no?

We deleted “simple”.

Methods:

Line 230: was it checked that this was high-contrast for the bees?

We added the relevant reference in the revised manuscript.

Line 240: how much sucrose solution was present in the flowers?

We added 25 microliters sucrose solution in the flowers. We added the information in the revised manuscript.

Line 266: check grammar.

We checked the grammar as follows: “During tests, both strings were glued to the floor of the arena to prevent the air flow generated by flying bumblebees’ wings from changing the position of the string.”

Statistical analysis:

- What does it mean that "Bees identity and colony were analyzed with likelihood ratio tests"?

Bees identity and colony was set as a random variable. We changed the analysis methods in the revised manuscript, and results of the all the experiments did not changed.

- Line 359: do you mean proportion rather than percentage?

We mean the percentage.

- "the number of total choices as weights" - this should be explained further. This is the number of choices that each bee made? What was the variation and mean of this number? If bees varied a lot in this metric, it might make more sense to analyze their first choice (as I see you've done) and their first 10 choices or something like that - for consistency.

This refers to the total number of choices made by each bumblebee. We added the mean and standard error of each bee’s number of choices in Table 1. Some bees pulled the string fewer than 10 times; we chose to include all choices made by each bee.

- More generally I think the first test is more informative than the subsequent choices, since every choice after their first could be affected by feedback they are getting in that test phase. Or rather, they are telling you different things.

All the bees were tested only once, however, you might be referring to the first choice. We used Chi-square test to analyze the bumblebees’ first choices in the test. It is worth noting that both connected and disconnected strings were glued to the floor. The bees were unable to move either the connected or disconnected strings during the tests, and only attempted to pull them. Therefore,the feedback from pulling either the connected or disconnected strings is the same.

- Line 362: I think I know what you mean, but this should be re-phrased because the "number of" sounds more appropriate for a Poisson distribution. I think what you are testing is whether each individual bee chose the connected or the disconnected string - i.e. a 0 or 1 response for each bee?

We agree with the reviewer that each bee chose the connected or the disconnected string - i.e. a 0 or 1 response for each bee, but not the number. We clarify this as: “The total number of the choices made by each bee was set as weights.” 

- Line 364-365: here and elsewhere, every time you mention a model, make it clear what the dependent and independent variables are. i.e. for the mixed model, the 'bee' is the random factor? Or also the colony that the bee came from? Were these nested etc?

We clarify this in the revised manuscript. The bee identity and colony is the random factor in the mixed model.

- Line 368: "Latency to the first choice of each bee was recorded" - why? What were the hypotheses/ predictions here?

The latency to the first choice was intended to see if the bumblebees were familiarizing with the testing pattern. A shorter delay time might indicate that the bumblebees were more familiar with the pattern.

- Line 371: "Multiple comparisons among experiments were.." - do you mean 'within' experiments? It seems that treatments should not be compared between different experiments.

We mean multiple comparisons among different experiments; we clarify this in the revised manuscript.

Results

Experiment 1: From the methods, it sounded like you both analyzed the bees' first choice and their total no. of choices, but in the results section (and Figure 1) I only see the data for all choices combined here.

In table 1 and in the text you report the number of bees that chose each option on their first choice, but there are no statistical results associated with these results. At the very least, a chi square or binomial test could be run.

Line 138: "Interestingly, ten out of fifteen bees pulled the connected string in their first choice" - this is presented like it is a significant majority of bees, but a chi-square test of 10 vs 5 has a p-value = 0.1967

We used the Chi square test to analyzed of the bees’ first choice. We also added the analyzed data in the Table 1.

Line 143: "It makes sense because the bees could see the "lollipop shape" once they pulled it out from the table." - this feels more like interpretation (i.e. Discussion) rather than results.

We moved the sentence to the discussion.

Line 162: again this feels more like interpretation/ conjecture than results.

We removed the sentence in the results.

Line 184: check grammar.

We checked the grammar. We changed “task” to “tasks”.

Figures

I really appreciated the overview in Figure 5 - though I think this should be Figure 1? Even if the methods come later in eLife, I think it would be nice to have that cited earlier on (e.g. at the start of the results) to draw the reader's attention to it quickly, since it's so helpful. It also then makes the images at the bottom of what is currently Figure 1 make more sense. I also think that the authors could make it clearer in Figure 5 which strings are connected vs disconnected in the figure (even if it means exaggerating the distance more than it was in real life). I had to zoom in quite a bit to see which were connected vs. not. Alternatively, you could have an arrow to the string with the words "connected" "disconnected" the first time you draw it - and similar labels for the other string conditions.

We appreciate the valuable comment from the reviewer. We changed Figure 5 to Figure 2, and Figure 4 to Figure 1. We cited the Figures at the start of the results. We also changed the gap distance between the disconnected strings. Additionally, we added arrows to indicate “connected” and “disconnected” strings in the Figure.

Figure 1 - I think you could make it clearer that the bars refer to experiments (e.g. have an x-axis with this as a label). Also, check the grammar of the y-axis.

We added the experiments number in the Figures. Additionally, we checked the grammar of the y-axis. We changed “percentages” to “parentage”. 

I also think it's really helpful to see the supplementary videos but I think it would be nice to see some examples of the test phase, and not just the training examples.

We added Supplementary videos of the testing phase.

Reviewer #2 (Recommendations For The Authors):

Below are also some minor comments:

L40: "approaches".

We changed “approach” to “approaches”.

L42: but likely mainly due to sampling bias of mammals and birds.

We changed the sentence as follows: String pulling is one of the most extensively used approaches in comparative psychology to evaluate the understanding of causal relationships (Jacobs & Osvath, 2015), with most research focused on mammals and birds, where a food item is visible to the animal but accessible only by pulling on a string attached to the reward (Taylor, 2010; Range et al., 2012; Jacobs & Osvath, 2015; Wakonig et al., 2021).

L64: remove "in this study"

We removed “in this study”.

L64: simple associative learning of what? Isn't your image matching associative too?

We removed “ simple”.

L97: remove "a" before "connected".

We removed “a” before “connected”.

L136-138: but maybe they could still feel the weight of the flower when pulling?

Because both strings were glued to the floor in the test phase, the feedback was the same and therefore irrelevant. This information is noted in the General Methods.

L161: what are these numbers?

We removed the latency in the revised manuscript.

L167/ Table 1: I realise that the authors never tried slanted strings to check if bumblebees used proximity as a cue. Why?

This was simply because we wanted to focus on whether bumblebees could recognize the connectivity of the string.

Discussion: Why did you only control for colour of the string? What if you had used strings with different textures or smells? Unclear if the authors controlled for "bumblebee smell" on the strings, i.e., after a bee had used the string, was the string replaced by a new one or was the same one used multiple times?

We used different colors to investigate featural generalization of the visual display of the string connected to the flower in this task. We controlled for color because it is a feature that bumblebees can easily distinguish.

Both the flowers and the strings were used only once, to prevent the use of chemosensory cues. We clarify this in the revised manuscript.

L182: since what?

We deleted “since” in the revised manuscript.

L182-188: might be worth mentioning that some crows and parrots known for complex cognition perform poorly on broken strings (e.g., https://doi.org/10.1098/rspb.2012.1998 ; https://doi.org/10.1163/1568539X-00003511 ; https://doi.org/10.1038/s41598-021-94879-x ) and Australian magpies use trial and error (https://doi.org/10.1007/s00265-023-03326-6).

We added the following sentences as suggested by the reviewer: “It is worth noting that some crows and parrots known for complex cognition perform poorly on the broken string task without perceptual feedback or learning. For example, New Caledonian crows use perceptual feedback strategies to solve the broken string-pulling task, and no individual showed a significant preference for the connected string when perceptual feedback was restricted (Taylor et al., 2012). Some Australian magpies and African grey parrots can solve the broken string task, but they required a high number of trials, indicating that learning plays a crucial role in solving this task (Molina et al., 2019; Johnsson et al., 2023).”

L193: maybe expand on this to put the task into a natural context?

We added the following sentences as suggested by the reviewer:

“Different flower species offer varying profitability in terms of nectar and pollen to bumblebees; they need to make careful choices and learn to use floral cues to predict rewards (Chittka, 2017). Bumblebees can easily learn visual patterns and shapes of flower (Meyer-Rochow, 2019); they can detect stimuli and discriminate between differently coloured stimuli when presented as briefly as 25 ms (Nityananda et al., 2014). In contrast, causal reasoning involves understanding and responding to causal relationships. Bumblebees might favor, or be limited to, a visual approach, likely due to the efficiency and simplicity of processing visual cues to solve the string-pulling task. ”

L204: is causal understanding the same as means-end understanding?

Means-end understanding is expressed as goal-directed behavior, which involves the deliberate and planned execution of a sequence of steps to achieve a goal. Includes some understanding of the causal relationship (Jacobs & Osvath, 2015; Ortiz et al., 2019). .

L235: this is a very big span of time. Why not control for motivation? Cognitive performance can vary significantly across the day (at least in humans).

Bumblebee motivation is understood to be rather consistent, as those that were trained and tested came to the flight arena of their own volition and were foragers looking to fill their crop load each time to return it to the colony.

L232: what is "(w/w)" ? This occurs throughout the manuscript.

“w/w” represents the weight-to-weight percentage of sugar.

L250: this sentence sounds odd. "containing in the central well.." ?? Perhaps rephrase? Unclear what central well refers to? Did the flowers have multiple wells?

We rephrased the sentence as follows: For each experiment, bumblebees were trained to retrieve a flower with an inverted Eppendorf cap at the center, containing 25 microliters of 50% sucrose solution, from underneath a transparent acrylic table

L268: why euthanise?

The reason for euthanizing the bees is that new foragers will typically only become active after the current ones were removed from the hive.

L270: chemosensory cues answer my concern above. Maybe make it clear earlier.

We moved this sentence earlier in the result.

L273: did different individuals use different pulling strategies? Do you have the data to analyse this? This has been done on birds and would offer a nice comparison.

We analyzed the string-pulling strategies among different individuals, and provided Supplementary Table 1 to display the performances of each individual in different string-pulling experiments.

L365: unclear why both models. Would be nice to see a GLM output table.

The duration of pulling different kinds of strings were first tested with the Shapiro-Wilk test to assess data normality. The duration data that conforms to a normal distribution was compared using linear mixed-effects models (LMM), while the data that deviates from normality were examined with a generalized linear-mixed model (GLMM). We added a GLM and GLMM output table in the revised manuscript.

L377: should be a space between the "." and "This".

We added a space between the “.” and “This”.

L383-390: some commas and semicolons are in the wrong places.

We carefully checked the commas and semicolons in this sentence.

Reviewer #3 (Recommendations For The Authors):

Minor comments

Line 32: seems to be missing a word, suggest "the bumblebees' ability to distinguish".

we added “the” in the revised manuscript.

Line 47: it would be good to reference other scholars here, this is the central focus of all work in comparative psychology.

We added the reference in the revised manuscript.

Line 50-61: I think the string-pulling literature could be described in more detail here, with mention of perceptual-motor feedback loops as a competing hypothesis to means-end understanding (see Taylor et al 2010, 2012). It seems a stretch to suggest that "String-pulling studies have directly tested means-end comprehension in various species", when perceptual-motor feedback is a competing hypothesis that we have positive evidence for in several species.

We mentioned the perceptual-motor feedback in the introduction as follow:

“Multiple mechanisms can be involved in the string-pulling task, including the proximity principle, perceptual feedback and means-end understanding (Taylor et al., 2012; Wasserman et al., 2013; Jacobs & Osvath, 2015; Wang et al., 2020). The principle of proximity refers to animals preferring to pull the reward that is closest to them (Jacobs & Osvath, 2015). Taylor et al. (2012) proposed that the success of New Caledonian crows in string-pulling tasks is based on a perceptual-motor feedback loop, where the reward gradually moves closer to the animal as they pull the strings. If the visual signal of the reward approaching is restricted, crows with no prior string-pulling experience are unable to solve the broken string task (Taylor et al., 2012).

However, when a green table was placed behind the string to obscure the “lollipop” structure during the training, the bees could not see the “lollipop” during the initial training stage or after pulling the string from under the table. In this situation, the bees were unable to identify the connected string, further proving that bumblebees chose the connected string based on image matching.

Line 68: suggest remove 'meticulously'.

We removed “meticulously”.

Line 99: This is an exciting finding, can the authors please provide a video of a bee solving this task on its first trial?

We added videos in the supplementary materials.

Line 133: perceptual-motor feedback loops should be introduced in the introduction.

We introduced perceptual-motor feedback loops in the revised manuscript.

Line 136: please clarify the prior experience of these bees, it is not clear from the text.

We clarified the prior experience of these bees as follow: Bumblebees were initially attracted to feed on yellow artificial flowers, and then trained with transparent tables covered by black tape (S7 video) through a four-step process.

Line 138: from the video it is not possible to see the bee's perspective of this occlusion. Do the authors have a video or image showing the feedback the bees received? I think this is highly important if they wish to argue that this condition prevents the use of both image matching and a perceptual-motor feedback loop.

We prevented the use of image matching: the bees were unable to see the flower moving towards them above the table during the training phase in this condition. But the bees may receive visual image both after pulling the string out from the table and in the initial stages of training in this condition.

Line 147: please clarify what experience these bees had before this test.

We added the prior experience of bumblebees before training as follow: We therefore designed further experiments based on Taylor et al. (2012) to test this hypothesis. Bumblebees were first trained to feed on yellow artificial, and then trained with the same procedure as Experiment 2, but the connected strings were coiled in the test.

Line 155: This is a highly similar test to that used in Taylor et al 2012, have the authors seen this study?

We mentioned the reference in the revised manuscript as follows: We therefore designed further experiments based on Taylor et al. (2012) to test this hypothesis.

Line 183: This sentence needs rewriting "Since the vast majority of animals, including dogs 183 (Osthaus et al., 2005), cats (Whitt et al., 2009), western scrub-jays (Hofmann et al.,2016) and azure-winged magpies (Wang et al., 2019) are failing in such tasks spontaneously".

We changed the sentence as suggested by the reviewer as follow:  Some animals, including dogs (Osthaus et al., 2005), cats (Whitt et al., 2009), western scrub-jays (Hofmann et al., 2016) and azure-winged magpies (Wang et al., 2019) fail in such task spontaneously.

Line 186: "complete comprehension of the functionality of strings is rare" I am not sure the evidence in the current literature supports any animal showing full understanding, can the authors explain how they reach this conclusion?

We wished to say that few animal species could distinguish between connected and disconnected strings without trial and error learning. We revised the sentence as follows:

It is worth noting that some crows and parrots known for complex cognition perform poorly on broken string task without perceptual feedback or learning. For example, New Caledonian crows use perceptual feedback strategies to solve broken string-pulling task, and no individual showed a significant preference for the connected string when perceptual feedback is restricted (Taylor et al., 2012). Some Australian magpies and African grey parrots can solve the broken string task, but it required a high number of trials, indicating that learning plays a crucial role in solving this task (Molina et al., 2019; Johnsson et al., 2023).

Line 190: the authors need to clarify which part of their study provides positive evidence for this conclusion.

We added the evidence for this conclusion as follows: Our findings suggest that bumblebees with experience of string pulling prefer the connected strings, but they failed to identify the interrupted strings when the string was coiled in the test.

Line 265: was the far end of the string glued only?

The entire string was glued to the floor, not just the far ends of the string.

you heard weird al.dont pirate or download his song!!!

This a great example of the restrictions and the guidelines coders/programmers put on their works. I wonder if people allow or figure a way to code ai to feel emotions would they eventually go berserk on there own.

This could be unrelated but I've recently seen videos of humans hyperbolic linguistic tendencies. Will we really be able to detangle "fascism" from general words for evil?

df = pd.read_csv("../datasets/penguins_classification.csv")

The curse of ANTLR

combined with the curse of LISP

= MetaLISP

.4 - insipirations

I think that is super common among parents to act like mental illness doesn't exist and that it's still problematic to talk about

Pourquoi il y a "gap: 10px;" dans le premier bout de code, et "grid-gap: 10px;" dans le deuxième ? Les deux propriétés sont elles équivalentes ?

needs to be converted to object code for computer to execute

Welcome back and in this lesson I'm going to cover object versioning and MFA delete, two essential features of S3.

These are two things I can almost guarantee will feature on the exam and almost every major project I can involved in has needed solid knowledge of both.

So let's jump in and get started.

Object versioning is something which is controlled at a bucket level.

It starts off in a disabled state.

You can optionally enable versioning on a disabled bucket, but once enabled you cannot disable it again.

Just to be super clear, you can never switch bucket versioning back to disabled once it's been enabled.

What you can do though is suspend it and if desired a suspended bucket can be re-enabled.

It's really important for the exam to remember these stage changes.

So make a point of noting them down and when revising try to repeat until it sticks.

So a bucket starts off as disabled, it can be re-enabled again, an enabled bucket can be moved to suspended and then moved back to enabled.

But the important one is that enabled bucket can never be switched back to disabled.

That is critical to understand for the exam.

So you can see many trick questions which will test your knowledge on that point.

Without versioning enabled on a bucket, each object is identified solely by the object key, its name, which is unique inside the bucket.

If you modify an object, the original version of that object is replaced.

Versioning lets you store multiple versions of an object within a bucket.

Any operations which would modify an object, generate a new version of that object and leave the original one in place.

For example, let's say I have a bucket and inside the bucket is a picture of one of my cats, Winky.

So the object is called Winky.JPEG.

It's identified in the bucket by the key, essentially its name, and the key is unique.

If I modify the Winky.JPEG object or delete it, those changes impact this object.

Now there's an attribute of an object which I haven't introduced yet and that's the ID of the object.

When versioning on a bucket is disabled, the ID of the object in that bucket are set to null.

That's what versioning being off on a bucket means.

All of the objects have an ID of null.

Now if you upload or put a new object into a bucket with versioning enabled, then S3 allocates an ID to that object.

In this case, 111, 111.

If any modifications are made to this object, so let's say somebody accidentally overrides the Winky.JPEG object with the dog picture, but still calls it Winky.JPEG.

S3 doesn't remove the original object.

It allocates a new ID to the newer version and it retains the old version.

The newest version of any object in a version-enabled bucket is known as the current version of that object.

So in this case, the object called Winky.JPEG with an ID of 2222222, which is actually a dog picture, that is the current version of this object.

Now if an object is accessed without explicitly indicating to S3 which version is required, then it's always the current version which will be returned.

But you've always got the ability of requesting an object from S3 and providing the ID of a specific version to get that particular version back rather than the current version.

So versions can be individually accessed by specifying the ID, and if you don't specify the ID, then it's assumed that you want to interact with the current version, the most recent version.

Now versioning also impacts deletions.

Let's say we've got these two different versions of Winky.JPEG stored in a version-enabled bucket.

If we indicate to S3 that we want to delete the object and we don't give any specific version ID, then what S3 will do is try a new special version of that object known as a delete marker.

Now the delete marker essentially is just a new version of that object, so S3 doesn't actually delete anything, but the delete marker makes it look deleted.

In reality though, it's just hidden.

The delete marker is a special version of an object which hides all previous versions of that object.

But you can delete the delete marker which essentially undeletes the object, returning the current version to being active again, and all the previous versions of the object still exist, accessible using their unique version ID.

Now even with versioning enabled, you can actually fully delete a version of an object, and that actually really deletes it.

To do that, you just need to delete an object and specify the particular version ID that you want to remove.

And if you are deleting a particular version of an object and the version that you're deleting is the most recent version, so the current version, then the next most recent version of that object then becomes the current version.

Now some really important points that you need to be aware about object versioning.

I've mentioned this at the start of the lesson, it cannot be switched off, it can only be suspended.

Now why that matters is that when versioning is enabled on a bucket, all the versions of that object stay in that bucket, and so you're consuming space for all of the different versions of an object.

If you have one single object that's 5 gig in size, and you have five versions of that object, then that's 5 times 5 gig of space that you're consuming for that one single object, and it's multiple versions.

And logically, you'll build for all of those versions of all of those objects inside an S3 bucket, and the only way that you can zero those costs out is to delete the bucket and then re-upload all those objects to a bucket without versioning enabled.

That's why it's important that you can't disable versioning.

You can only suspend it, and when you suspend it, it doesn't actually remove any of those old versions, so you're still built for them.

Now there's one other relevant feature of S3 which does make it to the exam all the time, and that's known as MFA delete.

Now MFA delete is something that's enabled within the versioning configuration on a bucket.

And when you enable MFA delete, it means that MFA is required to change bucket versioning state.

So if you move from enable to suspend it or vice versa, you need this MFA to be able to do that, and also MFA is required to delete any versions of an object.

So to fully delete any versions, you need this MFA token.

Now the way that this works is that when you're performing API calls in order to change a bucket to versioning state or delete a particular version of an object, you need to provide the serial number of your MFA token as well as the code that it generates.

You concatenate both of those together, and you pass that along with any API calls to interact how you delete versions or change the versioning state of a bucket.

Okay, so that's all of the theory for object versioning inside S3.

And at this point, that's everything I wanted to cover in this license.

I'll go ahead and complete the video, and when you're ready, I'll look forward to you joining me in the next.

Welcome back in this demo lesson you're going to gain some practical experience of working with the versioning feature of S3.

So to get started just make sure that you're logged in to the management account of the organization, so the general account, and then make sure that you've got the Northern Virginia region selected, so US-EAS-1.

Now there is a link attached to this lesson which you need to click on and then extract.

This is going to contain all of the files that you'll be using throughout this demo.

So go ahead and click on that link, extract it, and it should create a folder called S3_Versioning.

Once you've confirmed that you're logged in and have the right region selected, then go ahead and move to the S3 console.

So you can get there either using the recently visited services, or you can type S3 into the Find Services box and click to move to the S3 console.

Now to demonstrate versioning we're going to go ahead and create an S3 bucket, we're going to set it up for static website hosting, enable versioning, and then experiment with some objects and just observe how versioning changes the default behavior inside an S3 bucket.

So go ahead and click on Create bucket.

As long as the bucket name is unique, its specific name isn't important because we won't be using it with Route 53.

So just give the bucket a name and make sure that it's something unique.

So I'm going to use AC_Bucket_13337.

You should pick something different than me and different from something that any other student would use.

Once you've selected a unique bucket name, just scroll down and uncheck Block All Public Access.

We're going to be using this as a static website hosting bucket, so this is fine.

And we'll need to acknowledge that we understand the changes that we're making, so check this box, scroll down a little bit more, and then under bucket versioning we're going to click to enable versioning.

Keep scrolling down and at the bottom click on Create bucket.

Next, go inside the bucket, click on Properties, scroll all the way down to the bottom, and we need to enable static website hosting.

So click on Edit, check the box to enable static website hosting.

For hosting type, we'll set it to host a static website, and then for the index document, just type index.html, and then for the error document, type error.html.

Once you've set both of those, you can scroll down to the bottom and click on Save Changes.

Now as you learned in the previous demo lesson, just enabling static website hosting isn't enough to allow access, we need to apply a bucket policy.

So click on the permissions tab, scroll down, and under bucket policy click on Edit.

Now inside the link attached to this lesson, which you should have downloaded and extracted, there should be a file called bucket_policy.json, which is an example bucket policy.

So go ahead and open that file and copy the contents into your clipboard, move back to the console and paste it into the policy box, and we need to replace this example bucket placeholder with the ARN for this bucket.

So copy the bucket ARN into your clipboard by clicking this icon.

Because this ARN references objects in this bucket, and we know this because it's got forward slash star at the end, we need to replace only the first part of this placeholder ARN with the actual bucket ARN from the top.

So select from the A all the way up to the T, so not including the forward slash and the star, and then paste in the bucket ARN that you copied onto your clipboard.

Once you've done that, you can scroll down and then click on Save Changes.

Next, click on the objects tab, and we're going to upload some of the files that you downloaded from the link attached to this lesson.

So click on Upload, and first we're going to add the files.

So click on Add Files, then you'll need to go to the location where you downloaded and extracted the file that's attached to this lesson.

And once you're there, go into the folder called S3_Versioning, and you'll see a folder called Website.

Open that folder, select index.html and click on Open, and then click on Add Folder, and select the IMG folder that's also in that same location.

So select that folder and then click on Upload.

So this is going to upload an index.html object, and it's going to upload a folder called IMG which contains winky.jpeg.

Once you've done that, scroll down to the bottom and just click on Upload.

Now once the upload's completed, you can go ahead and click on Close, and what you'll see in the Objects dialog inside the bucket is index.html and then a folder called IMG.

And as we know by now, S3 doesn't actually have folders it uses prefixes, but if we go inside there, you'll see a single object called winky.jpeg.

Now go back to the bucket, and what we're going to do is click on Properties, scroll down to the bottom, and then click on this icon to open our bucket in a new browser tab.

All being well, you should see AnimalsForLife.org, Animal of the Week, and a picture of my one-eyed cat called winky.

So this is using the same architecture as the previous demo lesson where you experienced static website hosting.

What we're going to do now though is experiment with versions.

So go back to the main S3 console, scroll to the top, and click on Objects.

So because we've got versioning enabled on this bucket, as I talked about in the previous theory lesson, it means that every time you upload an object to this S3 bucket, it's assigned a version ID.

And if you upload an object with the same name, then instead of overwriting that object, it just creates a new version of that object.

Now with versioning enabled and using the default settings, we don't see all the individual versions, but we can elect to see them by toggling this Show Versions toggle.

So go ahead and do that.

Now you'll see that every object inside the S3 bucket, you'll see a particular version ID, and this is a unique code which represents this particular version of this particular object.

So if we go inside the IMG folder, you'll see that we have the same for winkeep.jpeg.

Toggle Show Versions to Disable, and you'll see that that version ID disappears.

What I want you to do now is to click on the Upload button inside this IMG folder.

So click on Upload, and then click on Add Files.

Now inside this Lessons folder, so S3 versioning, at the top level you've got a number of folders.

You have Website, which is what you uploaded to this S3 bucket, and this Image folder contains winkeep.jpeg.

So this is a particular file, winkeep.jpeg, that contains the picture of winkeep my one-eyed cat.

Now if you expand version 1 and version 2, you might be able to tell that version 1 is the same one-eyed cat, and we can expand that and say that it is actually winkeep.

Inside version 2 we have an object with the same name, but if we expand this, this is not winkeep, this is a picture of truffles.

So let's say that an administrator of this bucket makes a mistake and uploads this second version of winkeep.jpeg, which is not actually winkeep, it's actually truffles the cat.

But let's say that we do this, so we select winkeep.jpeg from the version 2 folder, and we click on Open.

Once we've selected that for upload, we scroll all the way down to the bottom and click on Upload.

That might take a few seconds to complete the upload because these are relatively large image files, but once it's uploaded you can click on Close.

So now we're still inside this image folder, and if we refresh, all we can see is one object, winkeep.jpeg.

So it looks with this default configuration of the user interface, like we've overwritten a previous object with this new object.

And if we go back to the tab which has got the static website open and hit refresh, you'll see that this image has indeed been replaced by the truffles image.

So even though it's called winkeep.jpeg, this is clearly truffles.

Now if we go back to the S3 console, and now if we enable the versions toggle, now we can see that we've got two different versions of this same object.

We've got the original version at the bottom and a new version at the top.

And note how both of these have different version IDs.

Now what S3 does is it always picks the latest version whenever you use any operations which simply request that one object.

So if we just request the object like we're doing with the static website hosting, then it will always pick the current or the latest version of this object.

But we do still have access to the older versions because we have versioning enabled on this bucket.

Nothing is ever truly deleted as long as we're operating with objects.

So let's experiment with exactly what functionality this gives us.

Go ahead and toggle show versions.

Once you've done that, select the winkeep.jpeg object and then click delete.

You'll need to type or copy and paste delete into this delete objects box and then click on delete.

Before we do that, note what it says at the top.

Deleting the specified objects adds delete markers to them.

If you need to undo the delete action, you can delete the delete markers.

So let's explore what this means.

Go ahead and click on delete objects.

And once it's completed, click on close.

Now how this looks at the moment, we're still in the image folder.

And because we've got show version set to off, it looks like we deleted the object.

But this is not what's occurred because we've got versioning enabled.

What's actually occurred is this is added a new version of this object.

But instead of an actual new version of the object, it's simply added a delete marker as that new version.

So if we toggle show versions back to on, now what we see are the previous versions of winkeep.jpeg.

So the original version at the bottom and the one that we replaced in the middle.

And then at the top we have this delete marker.

Now the delete marker is the thing which makes it look to be deleted in the console UI when we have show version set to off.

So this is how S3 handles deletions when versioning is enabled.

If you're interacting with an object and you delete that object, it doesn't actually delete the object.

It simply adds a delete marker as the most recent version of that object.

Now if we just select that delete marker and then click on delete, that has the effect of undeleting the object.

Now it's important to highlight that because we're dealing with object versions, anything that we do is permanent.

If you're operating with an object and you have versioning enabled on a bucket, if you overwrite it or delete it, all it's going to do is either add a new version or add a delete marker.

When you're operating with versions, everything is permanent.

So in this case we're going to be permanently deleting the delete marker.

So you need to confirm that by the typing or copying and pasting permanently delete into this box and click on delete objects.

What this is going to do is delete the delete marker.

So if we click on close, now we're left with these two versions of winkeep.jpeg so we've deleted the delete marker.

If we toggle show versions to off, we can see that we now have our object back in the bucket.

If we go back to static website hosting and refresh, we can see though that it's still truffle.

So this is a mistake.

It's not actually winky in this particular image.

So what we can do is go back to the S3 console, we can enable show versions.

We know that the most recent version is actually truffles rather than winky.

So what we can do is select this incorrect version, so the most recent version and select delete.

Now again, we're working with an object version.

So this is permanent.

You need to make sure that this is what you intend.

In our case it is.

So you need to either type or copy and paste permanently delete into the box and click on delete objects.

Now this is going to delete the most recent version of this object.

What happens when you do that is it makes the next most recent version of that object the current or latest version.

So now this is the original version of winky.jpeg, the one that we first uploaded to this bucket.

So this is now the only version of this object.

If we go back to the static website hosting tab and hit refresh, this time it loads the correct version of this image.

So this is actually winky my one-eyed cat.

So this is how you can interact with versioning in an S3 bucket.

Whenever it's enabled, it means that whenever you upload an object to the same name instead of overwriting, it simply creates a new version.

Whenever you delete an object, it simply adds a delete marker.

When you're operating with objects, it's always creating new versions or adding delete markers.

But when you're working with particular versions rather than objects, any operations are permanent.

So you can actually delete specific versions of an object permanently and you can delete delete markers to undelete that object.

Now it's not possible to turn off versioning on a bucket.

Once it's enabled on that bucket, you don't have the ability to disable it.

You only have the ability to suspend it.

Now when you suspend it, it stops new versions being created, but it does nothing about the existing versions.

The only way to remove the additional costs for a version-enabled bucket is either to delete the bucket and then reload the objects to a new bucket, or go through the existing bucket and then manually purge any specific versions of objects which aren't required.

So you need to be careful when you're enabling versioning on a bucket because it can cause additional costs.

If you have a bucket where you're uploading objects over and over again, specifically of their large objects, then if you have versioning enabled, you can incur significantly higher costs than if you have a bucket which doesn't have a versioning enabled.

So that's something you need to keep in mind.

If you enable versioning, you need to manage those versions of those objects inside the bucket.

With that being said, let's tidy up.

So let's go back to the main S3 console, select the bucket, click on Empty, copy and paste or type "Permanently Delete" and click on Empty.

When it's finished, click on Exit, and with the bucket still selected, click on Delete.

Copy and paste or type the name of the bucket and confirm it with the delete bucket.

I want you to build out the accounties back in the same state as it was at the start of this demo lesson.

Now at this point, that's everything that I want you to do in this demo lesson.

You've gained some practical exposure with how to deal with object versions inside an S3 bucket.

At this point, go ahead and complete this video, and when you're ready, I'll afford you joining me in the next lesson.

doesn't clearly state the mechanism of how the study is done, confusing how methodology of study.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

Referee #3

Evidence, reproducibility and clarity

The authors of the manuscript entitled "A conserved fungal Knr4/Smi1 protein is vital for maintaining cell wall integrity and host plant pathogenesis" used a weighted gene co-expression network to identify Fusarium graminearum genes highly expressed during early symptomless infection of wheat. Based on its sequence and previous studies, authors selected FgKnr4 from the early symptomless Fusarium modules. The characterization of knockout strains revealed a role in morphogenesis, growth, cell wall stress tolerance, and virulence in F. graminearum and the phylogenetically distant fungus Zymoseptoria tritici.

The methods are properly described and statistical analysis are reasonable so reproducibility is possible. The RNA-seq dataset is already published and the authors provided a repository with the code used to create the co-expression network. However, I have the following questions:

• Why only use of high confidence transcripts maize to map the reads and not the full genome like Fusarium graminearum? I have never analyzed plant transcriptome.
• The regular output of DESeq are TPMs, how did the authors obtain the FPKM used in the analysis?
• Do the authors have a southern blot to prove the location of the insertion and number of insertions in Zymoseptoria tritici mutant and complemented strains?
• Boxplots and bar graphs should have the same format. In Figures 5 B and F and supplementary figure 6.3 the authors showed the distribution of samples but it is lacking in figure 3 B and all bar graphs.
• Line 247 FGRAMPH1_0T23707 should be FGRAMPH1_01T23707

Referees cross-commenting

I agree with reviewer 1, the order in which the figures are called in the text is confusing. Regardless of figures 5C-D I am no expert in the field therefore I can only say they look like they have not been edited.

I agree with reviewer 1, data of DON mycotoxin production in infected issues is need it to support statement in line 272-273.

I agree with Reviewer 2, the criteria to exclude genes from the final selection list should be explained.

Significance

The study showed, once again, that a weighted gene co-expression network is a great method to identify new genes of interest regardless of the organism or condition even if not very popular in the fungal pathogen field yet. The study proved that functions identified in a WGCN module from a pathogen have their opposite in the host module. The authors go beyond the theory and demonstrate the effect of the highest expressed gene during the early symptomless stage of infection in maize and wheat fungal pathogens.

Fungal pathogen, RNA-seq, metabolic models, metabolism, comparative genomics

Reviewer #1 (Public Review):

Summary:

In this study, the authors re-analyzed Experiment 1 of a public dataset (Rademaker et al, 2019, Nature Neuroscience) which includes fMRI and behavioral data recorded while participants held an oriented grating in visual working memory (WM) and performed a delayed recall task at the end of an extended delay period. In that experiment, participants were pre-cued on each trial as to whether there would be a distracting visual stimulus presented during the delay period (filtered noise or randomly oriented grating). In this manuscript, the authors focused on identifying whether the neural code in the retinotopic cortex for remembered orientation was 'stable' over the delay period, such that the format of the code remained the same, or whether the code was dynamic, such that information was present, but encoded in an alternative format. They identify some time points - especially towards the beginning/end of the delay - where the multivariate activation pattern fails to generalize to other time points and interpret this as evidence for a dynamic code. Additionally, the authors compare the representational format of remembered orientation in the presence vs absence of a distracting stimulus, averaged over the delay period. This analysis suggested a 'rotation' of the representational subspace between distracting orientations and remembered orientations, which may help preserve simultaneous representations of both remembered and viewed stimuli.

Strengths:

(1) Direct comparisons of coding subspaces/manifolds between time points and task conditions is an innovative and useful approach for understanding how neural representations are transformed to support cognition.

(2) Re-use of existing datasets substantially goes beyond the authors' previous findings by comparing the geometry of representational spaces between conditions and time points, and by looking explicitly for dynamic neural representations

Weaknesses:

(1) Only Experiment 1 of Rademaker et al (2019) is reanalyzed. The previous study included another experiment (Expt 2) using different types of distractors which did result in distractor-related costs to neural and behavioral measures of working memory. The Rademaker et al (2019) study uses these two results to conclude that neural WM representations are protected from distraction when distraction does not impact behavior, but conditions that do impact behavior also impact neural WM representations. Considering this previous result is critical for relating the present manuscript's results to the previous findings, it seems necessary to address Experimentt 2's data in the present work

(2) Primary evidence for 'dynamic coding', especially in the early visual cortex, appears to be related to the transition between encoding/maintenance and maintenance/recall, but the delay period representations seem overall stable, consistent with previous findings

(3) Dynamicism index used in Figure 1f quantifies the proportion of off-diagonal cells with significant differences in decoding performance from the diagonal cell. It's unclear why the proportion of time points is the best metric, rather than something like a change in decoding accuracy. This is addressed in the subsequent analysis considering coding subspaces, but the utility of the Figure 1f analysis remains weakly justified.

(4) There is no report of how much total variance is explained by the two PCs defining the subspaces of interest in each condition, and timepoint. It could be the case that the first two principal components in one condition (e.g., sensory distractor) explain less variance than the first two principal components of another condition.

(5) Converting a continuous decoding metric (angular error) to "% decoding accuracy" serves to obfuscate the units of the actual results. Decoding precision (e.g., sd of decoding error histogram) would be more interpretable and better related to both the previous study and behavioral measures of WM performance.

(6) This report does not make use of behavioral performance data in the Rademaker et al (2019) dataset.

(7) Given there were observed differences between individual retinotopic ROIs in the temporal cross-decoding analyses shown in Figure 1, the lack of data presented for the subspace analyses for the corresponding individual ROIs is a weakness

Reviewer #2 (Public Review):

Summary:

In this work, Degutis and colleagues addressed an interesting issue related to the concurrent coding of sensory percepts and visual working memory contents in visual cortices. They used generalization analyses to test whether working memory representations change over time, diverge from sensory percepts, and vary across distraction conditions. Temporal generalization analysis demonstrated that off-diagonal decoding accuracies were lower than on-diagonal decoding accuracies, regardless of the presence of intervening distractions, implying that working memory representations can change over time. They further showed that the coding space for working memory contents showed subtle but statistically significant changes over time, potentially explaining the impaired off-diagonal decoding performance. The neural coding of sensory distractions instead remained largely stable. Generalization analyses between target and distractor codes showed overlaps but were not identical. Cross-condition decodings had lower accuracies compared to within-condition decodings. Finally, within-condition decoding revealed more reliable working memory representations in the condition with intervening random noises compared to cross-condition decoding using a trained classifier on data from the no-distraction condition, indicating a change in the VWM format between the noise distractor and no-distractor trials.

Strengths:

This paper demonstrates a clever use of generalization analysis to show changes in the neural codes of working memory contents across time and distraction conditions. It provides some insights into the differences between representations of working memory and sensory percepts, and how they can potentially coexist in overlapping brain regions.

Weaknesses:

(1) An alternative interpretation of the temporal dynamic pattern is that working memory representations become less reliable over time. As shown by the authors in Figure 1c and Figure 4a, the on-diagonal decoding accuracy generally decreased over time. This implies that the signal-to-noise ratio was decreasing over time. Classifiers trained with data of relatively higher SNR and lower SNR may rely on different features, leading to poor generalization performance. This issue should be addressed in the paper.

(2) The paper tests against a strong version of stable coding, where neural spaces representing WM contents must remain identical over time. In this version, any changes in the neural space will be evidence of dynamic coding. As the paper acknowledges, there is already ample evidence arguing against this possibility. However, the evidence provided here (dynamic coding cluster, angle between coding spaces) is not as strong as what prior studies have shown for meaningful transformations in neural coding. For instance, the principal angle between coding spaces over time was smaller than 8 degrees, and around 7 degrees between sensory distractors and WM contents. This suggests that the coding space for WM was largely overlapping across time and with that for sensory distractors. Therefore, the major conclusion that working memory contents are dynamically coded is not well-supported by the presented results.

(3) Relatedly, the main conclusions, such as "VWM code in several visual regions did not generalize well between different time points" and "VWM and feature-matching sensory distractors are encoded in separable coding spaces" are somewhat subjective given that cross-condition generalization analyses consistently showed above chance-level performance. These results could be interpreted as evidence of stable coding. The authors should use more objective descriptions, such as 'temporal generalization decoding showed reduced decoding accuracy in off-diagonals compared to on-diagonals.

Comparing this version of the Babylonian text, there is a greater focus on when Gilgamesh was with Enkidu on how he was a great companion for him. Dreams are seen as divine in Mesopotamian culture so it is interesting that Gilgamesh was able to foreshadow the presence of Enkidu ahead of time. Because of this dream, it shows that it was a part of destiny for Gilgamesh to find his equal and was a journey for his own identity to become what it is now. Not to mention, Enkidu becoming tame as time went along and reforming into societal norms shows that outsiders can be assimilated and that is what is needed in many nations in order for them to be successful and functional. One instance of us vs them situation would be for Enkidu. He was a wild man at first which was very different from "them" which were the Uruk people as they were calm and controlled. The transition for Enkidu to becoming like the others were crucial if he wanted to be a companion of Gilgamesh and also become a figure that would be respected by others. It also points to the fact that people need to be like others to some extent in order to be liked and respected. Because Enkidu and Gilgamesh are the only prominent characters that are also male, it suggests that during this time period that females were inferior to some extent and did not hold the same respect or regard because they were unable to showcase their own skills or talents. This may have affected the way that the text is written because the perspective of males can contrast those of women because they tend to be more caring and honest. Not to mention, the other translations of the text mention more about Gilgamesh's longing for immortality so the absence or lack of information on that aspect creates a more biased view of Enkidu and alters the way that Gilgamesh is viewed as well. The low point of the text has to be more in the beginning when the people are complaining about the rule of Gilgamesh because he does not contain the same qualities of a good leader that he obtains later on. With that being the case, the text reaches a high point when Gilgamesh sees Enkidu as an equal to himself and embraces him as a companion which allows him to be a much better leader and also allows the people in his land to feel better as a result. CC BY Ajey Sasimugunthan (contact)

Compared to the other versions focusing on the epic of Gilgamesh, this version looks more into Gilgamesh's cure for immortality after Enkidu's death. The "us" in this instance would be Gilgamesh and his search for a cure while the "them" would be the enemies which are trying stop him which include the forces he come along. The text is able to create this distinction by describing Gilgamesh as the main character as the one who is need of a cure because struggles to come to terms that he will die one day. Not to mention, Enkidu as a being was able to turn Gilgamesh into a noble figure who used his power for good turning him into a more likeable figure which is why the reader also roots for him to find a cure. Gilgamesh as a figure shows that in his time period, males were the ones who were seen as leaders who have strength because the other females in all versions of the text do not carry dynamic roles that showcase their personality or even their endearing qualities. There are more political and nationalistic themes compared to the Sumerian versions which illustrate how linguistics and language can play a role in how a culture might be perceived. By using the strong characteristics of Gilgamesh, the text is ultimately able to show the civilization of Uruk and create a sense of identity as a result. CC BY Ajey Sasimugunthan (contact)

Everything as in TrailMarks

https://hypothes.is/a/SkejHGDBEe-ynDfk6N2yPQ

building code

There are some things in the Tyler system that do not seem to have equivalents here. * Error Codes * Code Version * Country Codes * State Codes * Filing Security * Data Field Config * Status Codes

Filing Code Configuration decorates (Node, FilingCode) and is effective dated.

Reviewer #2 (Public Review):

Summary:

The authors build a colossal anatomical model of juvenile rat non-barrel primary somatosensory cortex, including inputs from the thalamus. This enhances past models by incorporating information on the shape of the cortex and estimated densities of various types of excitatory and inhibitory neurons across layers. This is intended to enable an analysis of the micro- and mesoscopic organisation of cortical connectivity and to be a base anatomical model for large-scale simulations of physiology.

Strengths:

• The authors incorporate many diverse data sources on morphology and connectivity.

• This paper takes on the challenging task of linking micro- and mesoscale connectivity.

• By building in the shape of the cortex, the authors were able to link cortical geometry to connectivity. In particular, they make an unexpected prediction that cortical conicality affects the modularity of local connectivity, which should be testable.

• The author's analysis of the model led to the interesting prediction that layer 5 neurons connect local modules, which may be testable in the future, and provide a basis to link from detailed anatomy to functional computations.

• The visualisation of the anatomy in various forms is excellent.

• A subnetwork of the model is openly shared (but see question below).

Weaknesses:

• Why was non-barrel S1 of the juvenile rat cortex selected as the target for this huge modelling effort? This is not explained.

• There is no effort to determine how specific or generalisable the findings here are to other parts of the cortex.

• Although there is a link to physiological modelling in another paper, there is no clear pathway to go from this type of model to understand how the specific function of the modelled areas may emerge here (and not in other cortical areas).

• In a few places the manuscript could be improved by being more specific in the language, for example:<br /> - "our anatomy-based approach has been shown to be powerful", I would prefer instead to read about specific contributions of past papers to the field, and how this builds on them.<br /> - similarly: "ensuring that the total number of synapses in a region-to-region pathway matches biology." Biology here is a loose term and implies too much confidence in the matching to some ground truth. Please instead describe the source of the data, including the type of experiment.

• Some of the decisions seem a little ad-hoc, and the means to assess those decisions are not always available to the reader e.g.<br /> - pg. 10. "Based on these results, we decided that the local connectome sufficed to model connectivity within a region.". What is the basis for this decision? Can it be formalised?<br /> - "In the remaining layers the results of the objective classification were used to validate the class assignments of individual pyramidal cells. We found the objective classification to match the expert classification closely (i.e., for 80-90% of the morphologies). Consequently, we considered the expert classification to be sufficiently accurate to build the model." The description of the validation is a little informal. How many experts were there? What are their initials? Was inter-rater or intra-rater reliability assessed? What are these numbers? The match with Kanari's classification accuracy should be reported exactly. There are clearly experts among the author list, but we are all fallible without good controls in place, and they should be more explicit about those controls here, in my opinion.<br /> - "Morphology selection was then performed as previously (Markram et al., 2015), that is, a morphology was selected randomly from the top 10% scorers for a given position." A lot of the decisions seem a little ad-hoc, without justification other than this group had previously done the same thing. For example, why 10% here? Shouldn't this be based on selecting from all of the reasonable morphologies?

• I would like to know if one of the key results relating to modularity and cortical geometry can be further explored. In particular, there seem to be sharp changes in the data at the end of the modelled cortical regions, which need to be explored or explained further.

• The shape of the juvenile cortex - a key novelty of this work - was based on merely a scalar reduction of the adult cortex. This is very surprising, and surely an oversimplification. Huge efforts have gone into modelling the complex nonlinear development of the cortex, by teams including the developing Human Connectome Project. For such a fundamental aspect of this work, why isn't it possible to reconstruct the shape of this relatively small part of the juvenile rat cortex?

• The same relative laminar depths are used for all subregions. This will have a large impact on the model. However, relative laminar depths can change drastically across the cortex (see e.g. many papers by Palomero-Gallagher, Zilles, and colleagues). The authors should incorporate the real laminar depths, or, failing that, show evidence to show that the laminar depth differences across the subregions included in the model are negligible.

• The authors perform an affine mapping between mouse and rat cortex. This is again surprising. In human imaging, affine mappings are insufficient to map between two individual brains of the same species and nonlinear transformations are instead used. That an affine transformation should be considered sufficient to map between two different species is then very surprising. For some models, this may be fine, but there is a supposed emphasis here on biological precision in terms of anatomical location.

• One of the most interesting conclusions, that the connectivity pattern observed is in part due to cooperative synapse formation, is based on analyses that are unfortunately not shown.

• Open code:<br /> - Why is only a subvolume available to the community?<br /> - Live nature of the model. This is such a colossal model, and effort, that I worry that it may be quite difficult to update in light of new data. For example, how much person and computer time would it take to update the model to account for different layer sizes across subregions? Or to more precisely account for the shape of the juvenile rat cortex?

Reviewer #3 (Public Review):

This manuscript reports a detailed model of the rat non-barrel somatosensory cortex, consisting of 4.2 million morphologically and biophysically detailed neuron models, arranged in space and connected according to highly sophisticated rules informed by diverse experimental data. Due to its breadth and sophistication, the model will undoubtedly be of interest to the community, and the reporting of anatomical details of modeling in this paper is important for understanding all the assumptions and procedures involved in constructing the model. While a useful contribution to this field, the model and the manuscript could be improved by employing data more directly and comparing simple features of the model's connectivity - in particular, connection probabilities - with relevant experimental data.

The manuscript is well-written overall but contains a substantial number of confusing or unclear statements, and some important information is not provided.

Below, major concerns are listed, followed by more specific but still important issues.

MAJOR ISSUES

(1) Cortical connectivity.

Section 2.3, "Local, mid-range and extrinsic connectivity modeled separately", and Figure 4: I am confused about what is done here and why. The authors have target data for connectivity (Figure 4B1). But then they use an apposition-based algorithm that results in connectivity that is quite different from the data (Figure 4B2, C). They then use a correction based on the data (Figure 4E) to arrive at a more realistic connectivity. Why not set the connectivity based on the data right away then? That would seem like a more straightforward approach.

The same comment applies to Section 2.4., "Specificity of axonal targeting": the distributions of synapses on different types of target cell compartments were not well captured by the original model based on axon-dendrite overlap and pruning, so the authors introduced further pruning to match data specificity. While details of this process and what worked and what didn't may be interesting to some, overall it is not surprising, as it has been well known that cell types exhibit connectivity that is much more specific than "Peters rule" or its simple variations. The question is, since one has the data, why not use the data in the first place to set up the connectivity, instead of using the convoluted process of employing axon-dendrite overlap followed by multiple corrections?

Most importantly, what is missing from the whole paper is the characterization of connection probabilities, at least for the local circuit within one area. Such connection probabilities can be obtained from the data that the authors already use here, such as the MICRONS dataset. Another good source of such data is Campagnola et al., Science, 2022. Both datasets are for mouse V1, but they provide a comprehensive characterization across all cortical layers, thus offering a good benchmark for comparison of the model with the data. It would be important for the authors to show how connection probabilities realized in their model for different cell types compared to these data.

(2) Section 2.5, "Structure of thalamic inputs" and Figure 6.

The text in section 2.5 should provide more details on what was done - namely, that the thalamic axons were generated based on the axon density profiles and then synapses were established based on their overall with cortical dendrites. Figure S10 where the target axon densities from data and the model axon densities are compared is not even mentioned here. Now, Figure S10 only shows that the axon densities were generated in a way that matches the data reasonably well. However, how can we know that it results in connectivity that agrees with data? Are there data sources that can be used for that purpose? For example, the authors show that in their model "the peaks of the mean number of thalamic inputs per neuron occur at lower depths than the peaks of the synaptic density". Is this prediction of the model consistent with any available data?

Most importantly, the authors should show how the different cell types in their model are targeted by the thalamic inputs in each layer. Experimental studies have been done suggesting specificity in targeting of interneuron types by thalamic axons, such as PV cells being targeted strongly whereas SST and VIP cells being targeted less.

(3) "We have therefore made not only the model but also most of our tool chain openly available to the public (Figure 1; step 7)."<br /> In fact it is not the whole model that is made publicly available, but only about 5% of it (211,000 out of 4,200,000 neurons). Also, why is "most" of the tool chain made openly available, and not the whole tool chain?

OTHER ISSUES

"At each soma location, a reconstruction of the corresponding m-type was chosen based on the size and shape of its dendritic and axonal trees (Figure S6). Additionally, it was rotated to according to the orientation towards the cortical surface at that point."

After this procedure, were cells additionally rotated around the white matter-pia axis? If yes, then how much and randomly or not? If not, then why not? Such rotations would seem important because otherwise additional order potentially not present in the real cortex is introduced in the model affecting connectivity and possibly also in vivo physiology (such as the dynamics of the extracellular electric field).

The term "new in vivo reconstructions" for the 58 neurons used in this paper in addition to "in vitro reconstructions" is a misnomer. It is not straightforward to see where the procedure is described, but then one finds that the part of Methods that describes experimental manipulations is mostly about that (so, a clearer pointer to that part of Methods could be useful). However, the description in Methods makes it clear that it is only labeling that is done in vivo; the microscopy and reconstruction are done subsequently in vitro. I would recommend changing the terminology here, as it is confusing. Also, can the authors show reconstructions of these neurons in the supplementary figures? Is the reconstruction shown in Figure 4A representative?

In the Discussion, "This was taken into account during the modeling of the anatomical composition, e.g. by using three-dimensional, layer-specific neuron density profiles that match biological measurements, and by ensuring the biologically correct orientation of model neurons with respect to the orientation towards the cortical surface. As local connectivity was derived from axo-dendritic appositions in the anatomical model, it was strongly affected by these aspects.<br /> However, this approach alone was insufficient at the large spatial scale of the model, as it was limited to connections at distances below 1000μm."

As mentioned above, it is not clear that this approach was sufficient for local connectivity either. It would be great if the authors showed a systematic comparison of local connection probabilities between different cell types in their model with experimental data and commented here in the Discussion about how well the model agrees with the data.

In the Discussion: "The combined connectome therefore captures important correlations at that level, such as slender-tufted layer 5 PCs sending strong non-local cortico-cortical connections, but thick-tufted layer 5 PCs not." (Also the corresponding findings in Results.)

If I understand this statement correctly, it may not agree with biological data. See analysis from MICRONS dataset in Bodor et al., https://www.biorxiv.org/content/10.1101/2023.10.18.562531v1.

Table 2 is confusing. What do pluses and minuses mean? What does it mean that some entries have two pluses? This table is not mentioned anywhere else in the text. If pluses mean some meaningful predictions of the model, then their distribution in the table seems quite liberal and arbitrary. It is not clear to me that the model makes that many predictions, especially for type-specificity and plasticity. Also, why is the hippocampus mentioned in this table? I don't see anything about the hippocampus anywhere else in the paper.

In the Discussion, "Thus, we made the tools to improve our model also openly available (see Data and Code availability section)."<br /> As mentioned before, the authors themselves write that they made "most of our tool chain openly available to the public", but not all of it.

Table S2 has multiple question marks. It is not clear whether the "predictions" listed in that table are truly well-thought-out and/or whether experimental confirmations are real.

Introduction: It would be quite appropriate to cite here Einevoll et al., Neuron, 2019 ("The Scientific Case for Brain Simulations").

Its quite amusing to think that even when it comes to his story, Joseph's life occupies more space than other characters in the bible.

Precisamente esa idea grandilocuente de qué son un programador y un problema valiosos es lo que deja desatendidas las soluciones que no suenan ambiciosas.

Preferimos terraformar marte, que el depredado Amazonas.

En contraste el software situado nos ha permitido resolver problemas para comunidades pequeñas en HackBo, nuestro hackerspace local o ayudando en la preservación lingüística en el Amazonas.

En los ejemplos, todos los problemas a resolver parecen grandilocuentos: miles de millones de personas, la información del mundo, los astronáutas. Pareciera ser que el vecino, la familia, la comunidad local, están por fuera de esos imaginarios. Al menos pensar que los problemas importantes tan bien son cotidianos y pequeños es algo que vale la pena comunicar más asertiva y reiteradamente.

Una de las cosas interesantes es que Breck cuenta en otra entrada como el software debería ahorrar tiempo a las personas, y allí revela una sensibilidad por los problemas pequeños, que le importaban a su familia y a él como niño/adolescente: tener 20 minutos más para poder jugar o ahorrarle esos 20 minutos a su familia a conectarse a internet.

Lo que creo que necesitamos es una manera de expresar software para el cuidado: de la gente, del planeta, del tiempo. Algo como un software convivial, en las líneas de las tecnologías conviviales de Ivan Illich.

What does full professional proficiency look like for someone that has been uprooted from the Spanish language and culture and has become part of the expatriate Spanish speaking community.

Author response:

The following is the authors’ response to the previous reviews.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

My main concern is still in place. It is unclear whether the proposed method can find actual goal states, and as a result it is unclear what states it finds. Table S1 mentions the model BIOMD0000000454, which is a small metabolic pathway with known equations given in "Example One" in "Metabolic Control Analysis: Rereading Reder". In this model the goal states can be calculated analytically.

Regarding your statements below: I am not concerned that your method will be less efficient than random search (or any other search..) on small models, but I think it is important for the readers to have evidence that your method is able to discover true goal states at least in small networks, used in your study. You do show that your method scales to complex models. So, in my opinion, the missing part is to show that it is able to find true goal states.

"...For simple models whose true steady-state distribution can be derived numerically and/or analytically, it is very likely that their exploration will be much simpler and this is not where a lot of improvement over random search may be found, which explains our focus on more complex models..."

We thank you for your response and for your concerns on the lack of evidence that our method is able to re-discover the true goal states of simple models when these are known a priori. We acknowledge that adding these simple cases is useful for completeness. We did not include these simple models in our main study because in most cases a basic random search over the initial conditions will lead to the re-discovery of these goal states. For instance for the mentioned model BIOMD0000000454 described in the "Example One" from the "Metabolic Control Analysis: Rereading Reder" paper, several simplifying assumptions are made such that the system only has one steady state (x1=0.056, x2=0.769, x3=4.231) which can be found analytically as shown in the paper. In that simple case, this goal state is also straightforward to find with numerical simulation as any valid initial condition will converge to it.

To address the concerns of the reviewer, we propose to add an additional "sanity check" figure in the supplementary of the revised paper (Figure S4), as well as a “sanity check” subsection in the “Methods”, to present additional experiments made on  simple models such as this one. The novel figure and subsection can be visualized on the paper’s interactive version available online https://developmentalsystems.org/curious-exploration-of-grn-competencies, and we plan to include them as such in the further revision.  We have also included the full code to reproduce this sanity check as a ‘sanity_check.ipynb’  jupyter notebook in the github repository (https://github.com/flowersteam/curious-exploration-of-grn-competencies/blob/main/notebooks/sanity_check.ipynb).

In the novel figure S4-b, we show the results of our exploration pipeline on the suggested model BIOMD0000000454 as described in the "Example One" of the paper. These results provide evidence that the curiosity search is able to find back the correct unique goal state (x1=0.056, x2=0.769, x3=4.231), as expected.

We also include a second sanity check on BIOMD0000000341 which models the dynamics of beta-cell mass, insulin and glucose dynamics. This model has two stable fixed points representing physiological (B=300, I=10, G=100) and pathological (B=0, I=0, G=600) steady states, which are the known ground truth steady states as described in Figure 3 of the "A Model of b-Cell Mass, Insulin, and Glucose Kinetics: Pathways to Diabetes" paper. Again, as expected, curiosity search is able to find back those two steady states (Figure S4-a).

As stated in our previous answer, our main study focuses on more complex models that are not limited to one or few attractors that can easily be discovered with random initial conditions. Regarding the mentioned BIOMD0000000454, maybe something that has been confusing for the reviewer is that we indeed included it in our main study but, as specified in the caption of table S4, at the difference of what is done in the "example one" of the original paper, we let the metabolite concentrations y1,...,y5 evolve in time (instead of enforcing them as constants). When doing so, the resulting dynamics of the system are more complex and exhibit a spectrum of possible steady states (unknown a priori), which differ from the previous case with a single steady state. In that case, the new attractors are not analytically easy to find and the proposed curiosity search becomes interesting as it is able to uncover the distribution of possible steady states much more efficiently than a random search baseline, as shown in the new figures S4-c and S4-d.

We hope that these new results will address the reviewer’s concerns and provide evidence to the readers on the validity of the approach on simple networks.

eLife assessment

This important study develops a machine learning method to reveal hidden unknown functions and behavior in gene regulatory networks by searching parameter space in an efficient way. The evidence for some parts of the paper is still incomplete and needs systematic comparison to other methods and to the ground truth, but the work will be of broad interest to anyone working in biology of all stripes since the ideas reach beyond gene regulatory networks to revealing hidden functions in any complex system with many interacting parts.

We thank the editors and reviewers for their positive assessment and constructive suggestions. In our response, we acknowledge the importance of systematic comparison to other methods and to the ground truth, when available. However we also emphasize the challenges associated with evaluating such methods in the context of uncovering hidden behaviors in complex biological networks as the ground truth is often unknown. We hope that our explanations will clarify the potential of our approach in advancing the exploration of these systems.

Public Reviews:

Reviewer #1 (Public Review):

Summary: This paper suggests to apply intrinsically-motivated exploration for the discovery of robust goal states in gene regulatory networks.

Strengths:

The paper is well written. The biological motivation and the need for such methods are formulated extraordinarily well. The battery of experimental models is impressive.

We thank the reviewer for sharing interest in the research problem and for recognizing the strengths of our work.

Weaknesses:

(1) The proposed method is compared to the random search. That says little about the performance with regard to the true steady-state goal sets. The latter could be calculated at least for a few simple ODE (e.g., BIOMD0000000454, `Metabolic Control Analysis: Rereading Reder'). The experiment with 'oscillator circuits' may not be directly interpolated to the other models.

The lack of comparison to the ground truth goal set (attractors of ODE) from arbitrary initial conditions makes it hard to evaluate the true performance/contribution of the method. A part of the used models can be analyzed numerically using JAX, while there are models that can be analyzed analytically.

"...The true versatility of the GRN is unknown and can only be inferred through empirical exploration and proxy metrics....": one could perform a sensitivity analysis of the ODEs, identifying stable equilibria. That could provide a proxy for the ground truth 'versatility'.

We agree with the reviewer that one primary concern is to properly evaluate the effectiveness of the proposed method. However, as we move toward complex pathways, knowledge of the “true” steady-state goal sets is often unknown which is where the use of machine learning methods as the one we propose are particularly interesting (but challenging to evaluate).

For simple models whose true steady-state distribution can be derived numerically and/or analytically, it is very likely that their exploration will be much simpler and this is not where a lot of improvement over random search may be found, which explains our focus on more complex models. While we agree that it is still interesting to evaluate exploration methods on these simple models for checking their behavior, it is not clear how to scale this analysis to the targeted more complex systems.

For systems whose true steady state distribution cannot be derived analytically or numerically, we believe that random search is a pertinent baseline as it is commonly used in the literature to discover the attractors/trajectories of a biological network. For instance, Venkatachalapathy et al. [1] initialize stochastic simulations at multiple randomly sampled starting conditions (which is called a kinetic Monte Carlo-based method) to capture the steady states of a biological system. Similarly, Donzé et al. [29] use a Monte Carlo approach to compute the reachable set of a biological network «when the number of parameters  is large and their uncertain range  is not negligible». For the considered models, the true steady-state goal set is unknown, which is why we chose comparison with random search. We added a “Statistics” subsection in the Methods section providing additional details about the statistical analyses we perform between our method and the random search baseline.

(2) The proposed method is based on `Intrinsically Motivated Goal Exploration Processes with Automatic Curriculum Learning', which assumes state action trajectories [s_{t_0:t}, a_{t_0:t}], (2.1 Notations and Assumptions' in the IMGEP paper). However, the models used in the current work do not include external control actions, but rather only the initial conditions can be set. It is not clear from the methods whether IMGEP was adapted to this setting, and how the exploration policy was designed w/o actual time-dependent actions. What does "...generates candidate intervention parameters to achieve the current goal....", mean considering that interventions 'Sets the initial state...' as explained in Table 2?

We thank the reviewer for asking for clarification, as indeed the IMGEP methodology originates from developmental robotics scenarios which generally focus on the problem of robotic sequential decision-making, therefore assuming state action trajectories as presented in Forestier et al. [65]. However, in both cases, note that the IMGEP is responsible for sampling parameters which then govern the exploration of the dynamical system. In Forestier et al. [65], the IMGEP also only sets one vector at the start (denoted ) which was specifying parameters of a movement (like the initial state of the GRN), which was then actually produced with dynamic motion primitives which are dynamical system equations similar to GRN ODEs, so the two systems are mathematically equivalent. More generally, while in our case the “intervention” of the IMGEP (denoted ) only controls the initial state of the GRN, future work could consider more advanced sequential interventions simply by setting parameters of an action policy  at the start which could be called during the GRN’s trajectory to sample control actions  where  would be the state of the GRN. In practice this would also require setting only one vector at the start, so it would remain the same exploration algorithm and only the space of parameters would change, which illustrates the generality of the approach.

(3) Fig 2 shows the phase space for (ERK, RKIPP_RP) without mentioning the typical full scale of ERK, RKIPP_RP. It is unclear whether the path from (0, 0) to (~0.575, ~3.75) at t=1000 is significant on the typical scale of this phase space. is it significant on the typical scale of this phase space?

The purpose of Figure 2 is to illustrate an example of GRN trajectory in transcriptional space, and to illustrate what “interventions” and “perturbations” can be in that context. To that end we have used the fixed initial conditions provided in the BIOMD0000000647, replicating Figure 5 of Cho et al. [56].

While we are not sure of what the reviewer means with “typical” scale of this phase space, we would like to point reviewer toward Figure 8 which shows examples of certain paths that indeed reach further point in the same phase space (up to ~10 in RKIPP_RP levels and ~300 in ERK levels). However, while the paths displayed in Figure 8 are possible (and were discovered with the IMGEP), note that they may be “rarer” to occur naturally  in the sense that a large portion of the tested initial conditions with random search tend to converge toward smaller (ERK, RKIPP_RP) steady-state values similar to the ones displayed in Figure 2.

(4) Table 2:

a. Where is 'effective intervention' used in the method?

b. in my opinion 'controllability', 'trainability', and 'versatility' are different terms. If their correspondence is important I would suggest to extend/enhance the column "Proposed Isomorphism". otherwise, it may be confusing.

a) We thank the reviewer for pointing out that “effective intervention” is not explicitly used in the method. The idea here is that as we are exploring a complex dynamical system (here the GRN), some of the sampled interventions will be particularly effective at revealing novel unseen outcomes whereas others will fail to produce a qualitative change to the distribution of discovered outcomes. What we show in this paper, for instance in Figure 3a and Figure 4, is that the IMGEP method is particularly sample-efficient in finding those “effective interventions”, at least more than a random exploration. However we agree that the term “effective intervention” is ambiguous (does not say effective in what) and we have replaced it with “salient intervention” in the revised version.

b) We thank the reviewer for highlighting some confusing terms in our chosen vocabulary, and we have clarified those terms in the revised version. We agree that controllability/trainability and versatility are not exactly equivalent concepts, as controllability/trainability typically refers to the amount to which a system is externally controllable/trainable whereas versatility typically refers to the inherent adaptability or diversity of behaviors that a system can exhibit in response to inputs or conditions. However, they are both measuring the extent of states that can be reached by the system under a distribution of stimuli/conditions, whether natural conditions or engineered ones, which is why we believe that their correspondence is relevant.

I don't see how this table generalizes "concepts from dynamical complex systems and behavioral sciences under a common navigation task perspective".

We have replaced the verb “generalize” with “investigate” in the revised version.

Reviewer #2 (Public Review):

Summary:

Etcheverry et al. present two computational frameworks for exploring the functional capabilities of gene regulatory networks (GRNs). The first is a framework based on intrinsically-motivated exploration, here used to reveal the set of steady states achievable by a given gene regulatory network as a function of initial conditions. The second is a behaviorist framework, here used to assess the robustness of steady states to dynamical perturbations experienced along typical trajectories to those steady states. In Figs. 1-5, the authors convincingly show how these frameworks can explore and quantify the diversity of behaviors that can be displayed by GRNs. In Figs. 6-9, the authors present applications of their framework to the analysis and control of GRNs, but the support presented for their case studies is often incomplete.

Strengths:

Overall, the paper presents an important development for exploring and understanding GRNs/dynamical systems broadly, with solid evidence supporting the first half of their paper in a narratively clear way.

The behaviorist point of view for robustness is potentially of interest to a broad community, and to my knowledge introduces novel considerations for defining robustness in the GRN context.

We thank the reviewer for recognizing the strengths and novelty of the proposed experimental framework for exploring and understanding GRNs, and complex dynamical systems more generally. We agree that the results presented in the section “Possible Reuses of the Behavioral Catalog and Framework” (Fig 6-9) can be seen as incomplete along certain aspects, which we tried to make as explicit as possible throughout the paper, and why we explicitly state that these are “preliminary experiments”. Despite the discussed limitations, we believe that these experiments are still very useful to illustrate the variety of potential use-cases in which the community could benefit from such computational methods and experimental framework, and build on for future work.

Some specific weaknesses, mostly concerning incomplete analyses in the second half of the paper:

(1) The analysis presented in Fig. 6 is exciting but preliminary. Are there other appropriate methods for constructing energy landscapes from dynamical trajectories in gene regulatory networks? How do the results in this particular case study compare to other GRNs studied in the paper?

We are not aware of other methods than the one proposed by Venkatachalapathy et al. [1] for constructing an energy landscape given an input set of recorded dynamical trajectories, although it might indeed be the case. We want to emphasize that any of such methods would anyway depend on the input set of trajectories, and should therefore benefit from a set that is more representative of the diversity of behaviors that can be achieved by the GRN, which is why we believe the results presented in Figure 6 are interesting. As the IMGEP was able to find a higher diversity of reachable goal states (and corresponding trajectories) for many of the studied GRNs, we believe that similar effects should be observable when constructing the energy landscapes for these GRN models, with the discovery of additional or wider “valleys” of reachable steady states.

Additionally, it is unclear whether the analysis presented in Fig. 6C is appropriate. In particular, if the pseudopotential landscapes are constructed from statistics of visited states along trajectories to the steady state, then the trajectories derived from dynamical perturbations do not only reflect the underlying pseudo-landscape of the GRN. Instead, they also include contributions from the perturbations themselves.

We agree that the landscape displayed Fig. 6C integrates contributions from the perturbations on the GRN’s behavior, and that it can shape the landscape in various ways, for instance affecting the paths that are accessible, the shape/depth of certain valleys, etc. But we believe that qualitatively or quantitatively analyzing the effect of these perturbations  on the landscape is precisely what is interesting here: it might help 1) understand how a system respond to a range of perturbations and to visualize which behaviors are robust to those perturbations, 2) design better strategies for manipulating those systems to produce certain behaviors

(2) In Fig. 7, I'm not sure how much is possible to take away from the results as given here, as they depend sensitively on the cohort of 432 (GRN, Z) pairs used. The comparison against random networks is well-motivated. However, as the authors note, comparison between organismal categories is more difficult due to low sample size; for instance, the "plant" and "slime mold" categories each only have 1 associated GRN. Additionally, the "n/a" category is difficult to interpret.

We acknowledge that this part is speculative as stated in the paper: “the surveyed database is relatively small with respect to the wealth of available models and biological pathways, so we can hardly claim that these results represent the true distribution of competencies across these organism categories”. However, when further data is available, the same methodology can be reused and we believe that the resulting statistical analyses could be very informative to compare organismal (or other) categories.

(3) In Fig. 8, it is unclear whether the behavioral catalog generated is important to the intervention design problem of moving a system from one attractor basin to another. The authors note that evolutionary searches or SGD could also be used to solve the problem. Is the analysis somehow enabled by the behavioral catalog in a way that is complementary to those methods? If not, comparison against those methods (or others e.g. optimal control) would strengthen the paper.

We thank the reviewer for asking to clarify this point, which might not be clearly explained in the paper. Here the behavioral catalog is indeed used in a complementary way to the optimization method, by identifying a representative set of reachable attractors which are then used to define the optimization problem. For instance here, thanks to the catalog, we 1) were able to identify a “disease” region and several possible reachable states in that region and 2) use several of these states as starting points of our optimization problem, where we want to find a single intervention that can successfully and robustly reset all those points, as illustrated in Figure 8. Please note that given this problem formulation, a simple random search was used as an optimization strategy. When we mention more advanced techniques such as EA or SGD, it is to say that they might be more efficient optimizers than random search. However, we agree that in many cases optimizing directly will not work if starting from random or bad initial guess, and this even with EA or SGD. In that case the discovered behavioral catalog can be useful to better initialize  this local search and make it more efficient/useful, akin to what is done in Figure 9.

(4) The analysis presented in Fig. 9 also is preliminary. The authors note that there exist many algorithms for choosing/identifying the parameter values of a dynamical system that give rise to a desired time-series. It would be a stronger result to compare their approach to more sophisticated methods, as opposed to random search and SGD. Other options from the recent literature include Bayesian techniques, sparse nonlinear regression techniques (e.g. SINDy), and evolutionary searches. The authors note that some methods require fine-tuning in order to be successful, but even so, it would be good to know the degree of fine-tuning which is necessary compared to their method.

We agree that the analysis presented in Figure 9 is preliminary, and thank the reviewer for the suggestion. We would first like to refer to other papers from the ML literature that have more thoroughly analyzed this issue, such as Colas et al. [74] and Pugh et al. [34], and shown the interest of diversity-driven strategies as promising alternatives.  Additionally, as suggested by the reviewer, we added an additional comparison to the CMA-ES algorithm in the revised version in order to complete our analysis. CMA-ES is an evolutionary algorithm which is self-adaptive in the optimization steps and that is known to be better suited than SGD to escape local minimas when the number of parameters is not too high (here we only have 15 parameters). However, our results showed that while CMA-ES explores more the solution space at the beginning of optimization than SGD does, it also ultimately converges into a local minima similarly to SGD. The best solution converges toward a constant signal (of the target b) but fails to maintain the target oscillations, similar to the solutions discovered by gradient descent. We tried this for a few hyperparameters (init mean and std) but always found similar results.  We have updated the figure 9 image and caption, as well as descriptive text, to include these novel results in the revised version. We also added a reference to the CMA-ES paper in the citations.

Reviewer #1 (Recommendations For The Authors):

I would suggest to conduct a more rigor analysis of the performance by estimating/approximating the ground truth robust goal sets in important GRNs.

Also, the use of terminology from different disciplines can be improved. Please see my comments above. Specifically, the connection between controllability in dynamical control systems and versatility used in this paper is unclear.

We hope to have addressed the reviewer's concerns in our previous answers.

Reviewer #2 (Recommendations For The Authors):

Fig 4b: I'm not sure if DBSCAN is the appropriate method to use here, as the visual focus on the core elements of the clusters downplays the full convex hull of the points that random sampling achieves in Z space. An analysis based on convex hulls or the ball-coverage from Fig. 3b would presumably generate plots that were more similar between random sampling and curiosity search. If the goal is to highlight redundancy/non-linearity in the mapping between Z and I, another approach might be to simply bin Z-space in a grid, or to use a clustering algorithm that is less stringent about core/noise distinctions.

We thank the reviewer for the suggestion. This plot is intended to convey the reader an understanding of why a method that uniformly samples goals in Z (what the  IMGEP is doing), is more efficient than a method that uniformly samples parameters in I (what the random search is doing), in systems for which there is high redundancy/non-linearity in the mapping between I and Z. We agree that binning the Z-space in a grid and counting the number of achieved bins is a way to quantitatively measure this, which is by the way very close to what we do in Figure 3 for measuring the achieved diversity. We believe however that the clustering and coloring provides additional intuitions on why this is the case: it illustrates that large regions of the intervention space map to small regions in the outcome space and vice versa.

Additional changes in the revised version:

We added a sentence in the Methods section as well as in the caption of Table S1 providing additional details about the way we simulate the biological models from the BioModels website

We fixed a wrong reference to Figure 4 in the Methods “Sensitivity measure” subsection with reference to Figure 5.

Reviewer #2 (Public Review):

In this manuscript, Yang et al. present a modeling framework to understand the pattern of response biases and variance observed in delayed-response orientation estimation tasks. They combine a series of modeling approaches to show that coupled sensory-memory networks are in a better position than single-area models to support experimentally observed delay-dependent response bias and variance in cardinal compared to oblique orientations. These errors can emerge from a population-code approach that implements efficient coding and Bayesian inference principles and is coupled to a memory module that introduces random maintenance errors. A biological implementation of such operation is found when coupling two neural network modules, a sensory module with connectivity inhomogeneities that reflect environment priors, and a memory module with strong homogeneous connectivity that sustains continuous ring attractor function. Comparison with single-network solutions that combine both connectivity inhomogeneities and memory attractors shows that two-area models can more easily reproduce the patterns of errors observed experimentally.

Strengths:

The model provides an integration of two modeling approaches to the computational bases of behavioral biases: one based on Bayesian and efficient coding principles, and one based on attractor dynamics. These two perspectives are not usually integrated consistently in existing studies, which this manuscript beautifully achieves. This is a conceptual advancement, especially because it brings together the perceptual and memory components of common laboratory tasks.

The proposed two-area model provides a biologically plausible implementation of efficient coding and Bayesian inference principles, which interact seamlessly with a memory buffer to produce a complex pattern of delay-dependent response errors. No previous model had achieved this.

Weaknesses:

The correspondence between the various computational models is not clearly shown. It is not easy to see clearly this correspondence because network function is illustrated with different representations for different models. In particular, the Bayesian model of Figure 2 is illustrated with population responses for different stimuli and delays, while the attractor models of Figure 3 and 4 are illustrated with neuronal tuning curves but not population activity.

The proposed model has stronger feedback than feedforward connections between the sensory and memory modules (J_f = 0.1 and J_b = 0.25). This is not the common assumption when thinking about hierarchical processing in the brain. The manuscript argues that error patterns remain similar as long as the product of J_f and J_b is constant, so it is unclear why the authors preferred this network example as opposed to one with J_b = 0.1 and J_f = 0.25.

To use the code, send an HTTP POST request to the /v1/DeviceCode endpoint and authenticate with this user code using tenant's Global Admin.

After release from G1, subtelomeric regions were more under-replicated than regions further away from telomeric regions.

Reviewer #3 (Public Review):

Summary:

Wang and van Ede investigate whether and how attention re-orients within visual working memory following expected and unexpected centrally presented memory tests. Using a combination of spatial modulations in neural activity (EEG-alpha lateralization) and gaze bias quantified as time courses of microsaccade rate, the authors examined how retro cues with varying levels of reliability influence attentional deployment and subsequent memory performance. The conclusion is that attentional re-orienting occurs within visual working memory, even when tested centrally, with distinct patterns following expected and unexpected tests. The findings provide new value for the field and are likely of broad interest and impact, by highlighting working memory as an action-bound process (in)dependent on (an ambiguous) past.

Strengths:

The study uniquely integrates behavioral data (accuracy and reaction time), EEG-alpha activity, and gaze tracking to provide a comprehensive analysis of attentional re-orienting within visual working memory. As typical for this research group, the validity of the findings follows from the task design that effectively manipulates the reliability of retro cues and isolates attentional processes related to memory tests. The use of well-established markers for spatial attention (i.e. alpha lateralization) and more recently entangled dependent variable (gaze bias) is commendable. Utilizing these dependent metrics, the concise report presents a thorough analysis of the scaling effects of cue reliability on attentional deployment, both at the behavioral and neural levels. The clear demonstration of prolonged attentional deployment following unexpected memory tests is particularly noteworthy, although there are no significant time clusters per definition as time isn't a factor in a statistical sense, the jackknife approach is convincing. Overall, the evidence is compelling allowing the conclusion of a second stage of internal attentional deployment following both expected and unexpected memory tests, highlighting the importance of memory verification and re-orienting processes.

Weaknesses:

I want to stress upfront that these weaknesses are not specific to the presented work and do not affect my recommendation of the paper in its present form.

The sample size is consistent with previous studies, a larger sample could enhance the generalizability and robustness of the findings. The authors acknowledge high noise levels in EEG-alpha activity, which may affect the reliability of this marker. This is a general issue in non-invasive electrophysiology that cannot be handled by the authors but an interested reader should be aware of it. Effectively, the sensitivity of the gaze analysis appears "better" in part due to the better SNR. The latter also sets the boundaries for single-tiral analyses as the authors correctly mention. In terms of generalizability, I am convinced that the main outcome will likely generalize to different samples and stimulus types. Yet, as typical for the field future research could explore different contexts and task demands to validate and extend the findings. The authors provide here how and why (including sharing of data and code).

稍后

关于在 VS Code 中配置代码片段快捷键。

I have found the GT/Lepiter GUI pretty compelling for learners in my local hackerspace and in the information science department, both spaces where I'm a facilitator/teacher. It provides a pretty focused experience and it is stripped down of the overwhelming initial experience of the Pharo/Squeak GUI. It is not well suited for "classical Smalltalkers" though. as I have been talking with some of them and they find the DX too much specific and even cumbersome for some task they usually do (it has been not our case so far).

In our last use case at the university, the students are creating a personal code repository in Fossil, with data narratives and they do a critic/annotated reading, using Hypothesis (this very technology), which is kind of a personal public wiki-like portfolio for data narratives. They put also the reading notes in their own repositories for the data stories I published previously where I introduce Smalltalk or and introduction to data representation and processing in Pharo.

This could be another approach for wikis in the classroom, that is alterative to our use of interpersonal wikis with TiddlyWiki. At some point and in a pretty organic way, the idea would be to have all them integrated and powered by "context aware" and thematic chatbots (made in Pharo).

On a similar approach, I created and actively developed from 2014 to 2019 Grafoscopio, which, while being inside Pharo and a companion of all other tools, was providing a computable outliner to do something: write computable and reproducible documents and bridge the gap between the IDE and the app for a more mature audience (a similar approach for children was previously tested in Squeak, with Etoys).

This allowed me to write the Grafoscopio Manual (2016) inside Grafoscopio or to do with the community some hacktivist republishing, like we did with the Data Journalism Handbook (2018)

Of course, being those initiatives from the so called "Global South" and being Grafoscopio my first "real program" ever, they lacked the visibility of Global North initiatives, like the ones you collected in Smart Academic Notebook, but they were acknowledge and appreciated in small/specialized communities, like the Pharo community.

With the new GUI/DX provided by Lepiter (2021), I have been migrating the Grafoscopio Lessons from the previous half decade to this technology, with the MiniDocs package and I imagine Grafoscopio becoming more a software distribution on top of Pharo/GT, providing documentation and collaboration workflows and improved outlining with packages like TiddlyWikiPharo or the Brea decoupled CMS / static site generator.

BTW, as I don't know how to add comments or suggest updates I wonder why this note is not updated with Lepiter as it provides pretty much the experience you were advocating for since 2018 and it is already in your wiki/bliki. Maybe it is just a matter of some wiki refactoring a links update.

The input to L1 on the ESTL comes from the A10 I/O PCB Terminal X22 pin number 1 (marked with an arrow on the pcb). This is the 240V output from A10 to the safety chain.

The output from the ESTL (Terminal L*) terminates on connection A1 of the K1 contactor coil, thus potentially switching K1 on or off. Dependent of course upon the condition of the safety chain.

At the same time the output from the ESTL (Terminal L*) also returns to pin 3 of A10 terminal X22 to signal to A10 the condition of the safety chain, enabling the processor to provide appropriate indications of safety chain status. (Service 72 or Error 72).

Where two ESTLs are employed (20-2/1E models) the output terminal L* on the first ESTL is connected directly to the input terminal L1 on the second ESTL thus connecting their outputs in series, meaning both ESTL ouputs must be closed for the K1 contactor to energise.

See section 24 of the training manual for electrical wiring diagrams - commencing page 181.

It is no longer possible to enter the iCombi Classic Service menu unless you have downloaded the Rational PIN Creator App and have a valid log in and password.

Attempts to enter the Service Menu will generate a QR code on the screen.

Scanning the QR code with the Rational PIN Creator app will generate a PIN which if used on the screen will enable entry to the Service Menu for technicians.

See Video Library "Access to Service menu for Technicians"

If you have undergone technical training by Rational UK or its approved training partners you can obtain a log in and password from Rational UK.

Error code description:

Faulty data communication to automatic ignition controller.

Condition for error detection:

BUS signal from automatic ignition controller is missing or is not transmitted for at least 5 seconds at a time.

Error area:

Data transfer cable, automatic ignition controller.

Relevant causes/components:

• Electrical connection to components
• Automatic ignition controller

Error code description:

Faulty data communication to the I/O board.

Condition for error detection:

BUS signal from I/O board is missing or is not transmitted for at least 5 seconds at a time.

Error area:

Data transfer cable, I/O board

Relevant causes/components:

• Electrical connection to components
• I/O board

Error code description:

Faulty data communication to pump PCB.

Condition for error detection:

BUS signal from pump PCB is missing or is not transmitted for at least 5 seconds at a time.

Error area:

Data transfer cable, pump PCB

Relevant causes/components:

• Electrical connection to components
• Pump PCB

Error code description:

Faulty data communication to bottom fan motor.

Condition for error detection:

BUS signal from bottom fan motor is missing or is not transmitted for at least 5 seconds at a time.

Error area:

Data transfer cable, control electronics for bottom fan motor

Relevant causes/components:

• eSTL has initialised
• Electrical connection to components
• Control electronics for bottom fan motor

Error code description:

Faulty data communication to the middle fan motor on floor units and the bottom fan motor on 10 tray table top units.

Condition for error detection:

BUS signal from middle fan motor on floor units and the bottom fan motor on 10 tray table top units is missing or is not transmitted for at least 5 seconds at a time.

Error area:

Data transfer cable, control electronics for middle fan motor on floor units and the bottom fan motor on 10 tray table top units

Relevant causes/components:

• eSTL has initialised
• Electrical connection to components
• Control electronics for middle fan motor

Error code description:

ICP XS & CMP XS ONLY

Malfunction when opening/closing the drain valve

Condition for error detection:

The drain valve receives a request to open/close, but the drain valve still does not detect an end position after 4 minutes.

Error area: Drain valve

Relevant causes/components:

• Drain valve
• Electrical connection to components

Error code description: The steam generator water level detection is faulty.

Condition for error detection: The steam generator is filled via valve Y1, but no “full” water level is measured by the level electrode, even after a long time.

If the measured water flow from the CDS sensor is significantly greater than the reference volume of the steam generator, this service error 14 is triggered.

Error area:

Water level detection in the steam generator, reference volume steam generator, water leakage when filling the steam generator.

Relevant causes/components:

Level electrode Electrical connection to components Conductance water/threshold value for level detection Steam generator reference volume

Error code description:

Incorrect water quantity/flow measurement when filling the steam generator.

Condition for error detection:

A flow rate in a corresponding tolerance range is expected on the CDS sensor. The water also actively enters the steam generator, the level electrode indicates a full water level, but the CDS sensor passes on an error/implausible flow or no signal at all.

Error area:

Water supply, water flow detection, fill level detection

Relevant causes/components:

• Water supply
• Solenoid valve block with CDS sensor
• Electrical connection (CDS sensor – I/O module)
• Level electrode
• Steam generator reference volume
• Incorrect CDS pulse saved

Error code description:

Faulty data communication to top fan motor.

Condition for error detection:

BUS signal from top fan motor is missing or is not forwarded for at least 5 seconds at a time.

Error area:

Data transfer cable, control electronics for top fan motor

Relevant causes/components:

ESTL has initialised Electrical connection to components Control electronics for top fan motor Other information:

Before disconnecting the unit or replacing a component, download all data from the unit to a USB stick.This data is required if further support from RATIONAL is required.

NOTE: This error is reset automatically as soon as a BUS signal is present for at least 5 seconds at a time.

NOTE: If the ESTL has triggered (Service 72), one or more Service 34 errors are often displayed because the eSTL de-energises the affected BUS participants.

After a successful reset of the Service 72 error, the Service 34 errors are also reset automatically.

Service 25

Error code description: No or insufficient water circulation is detected during an iCareSystem cleaning program, self-test or module test.

Condition for error detection: Water hitting the fan wheel during circulation increases the output of the fan motor.

If this output increase is too low, the system assumes that no or too little water is entering the cooking cabinet.

Relevant causes/components * Drain sieve * Connection between control box and cleaning box * Cleaning carried out incorrectly * Only for XS units (6-2/3): circulation pump or drain valve faulty * Cleaning pipe * Air baffle, baffle plate * Cleaning box * Fan motor * Foam brake * Faulty water outlet/water loss

It is no longer possible to enter the iCombi Pro Service menu unless you have downloaded the Rational PIN Creator App and have a valid log in and password.

Attempts to enter the Service Menu will generate a QR code on the screen.

Scanning the QR code with the Rational PIN Creator app will generate a PIN which if used on the screen will enable entry to the Service Menu for technicians.

See Video Library "Access to Service menu for Technicians"

If you have undergone technical training by Rational UK or its approved training partners you can obtain a log in and password from Rational UK.

Thermocouple B4 measures the temperature behind the top circulation fan during the initial Self test calibration process and during any subsequent manual calibration. It also monitors the cabinet temperature for humidity control above the local boiling point.

The combi-steamer checks the temperature of thermocouple B4 every second (in all modes). As soon as an implausible temperature value is measured service error code Service 20.4 is indicated..

Check the value of temperature sensor B4 in the service menu:

For iCombi Pro: Select the steam or humidity launchers - go to Diagnosis.

For iCombi Classic: Select Diagnostics and B4

If this value is approx. 615°C [1140°F], either the sensor is defective (Open Circuit) or the screw connection on the inner cabinet is loose or not adequately insulated or the electrical connection between the temperature sensor and the I/O PCB is faulty.

B2 is situated in the Control box (formerly known as the quench box) and its function is to reduce the volume of steam emmissions from the vent stack and also to reduce the temperature of waste water through the drain to a level below 65 degrees during normal oven cooking processes.

The combi-steamer checks the temperature of thermocouple B2 every second (in all modes).

As soon as an implausible temperature value is measured, service error code Service 20.2 is indicated.

Check the value of temperature sensor B2 in the service menu:

For iCombi Pro: Select the steam or humidity launchers - go to Diagnosis.

For iCombi Classic: Select Diagnostics and B2

If this value is approx. 615°C [1140°F], either the sensor is defective (Open Circuit) or the electrical connection between the temperature sensor and the I/O PCB is faulty.

In the narratives, Hippolytus is depicted as a paragon of chastity and self-discipline, qualities that define his heroism within the cultural context of ancient Greece. His rejection of Phaedra's advances is rooted in his dedication to the goddess Artemis and his adherence to a code of moral purity. This portrayal aligns with the ideal of the male hero as one who resists temptation and remains steadfast in his principles, even at the cost of his own life. Phaedra, on the other hand, embodies the complexities of female desire within a patriarchal society. In Euripides' Hippolytus, her passion for Hippolytus is portrayed as an uncontrollable force that ultimately leads to her destruction and the downfall of Hippolytus. Her role as a woman who transgresses the boundaries of acceptable female behavior highlights the dangers of unchecked female desire, reinforcing the cultural belief that women’s emotions must be controlled and contained. The tragedy of Phaedra is not just her unfulfilled love but also the societal constraints that define her actions as inherently destructive. Pausanias' reference to the myth of Phaedra and Hippolytus, as discussed in the source, offers a more subdued version of the narrative, focusing less on the psychological torment of Phaedra and more on the broader mythological context. This difference in emphasis reflects varying cultural attitudes towards gender and heroism. While Euripides explores the inner turmoil of his characters, highlighting the destructive power of female desire, Pausanias presents a more neutral account, possibly influenced by the historical and cultural lens through which he viewed the myth. When comparing the versions of the Phaedra and Hippolytus story in Euripides and Pausanias, it is evident that Euripides' version is more focused on the emotional and psychological aspects of the characters, particularly Phaedra. Euripides' portrayal of Phaedra’s inner conflict and her ultimate decision to falsely accuse Hippolytus after he rejects her advances emphasizes the tragic consequences of her unbridled passion. In contrast, Pausanias’ version is less concerned with the emotional depth of the characters and more with the events themselves, reflecting a different approach to the narrative that is more aligned with the recording of history and myth rather than the exploration of character psychology. This difference in focus can be attributed to the cultural and political contexts in which these works were created. Euripides, writing in a period of Athenian democracy, was likely influenced by the social and philosophical debates of his time, including those related to gender and the role of women in society. Pausanias, writing in a later period, may have been more influenced by the desire to preserve and record myths as part of the cultural heritage, leading to a more straightforward recounting of the story. Comparison Between Individual Works: When comparing the story of Phaedra and Hippolytus with other similar narratives, such as the story of Joseph, we see a recurring theme of male chastity and female desire. In both stories, the male hero is depicted as morally superior, resisting the advances of a woman who is driven by passion. This resistance enhances the hero’s status as a figure of virtue and integrity, while the woman’s desire is portrayed as dangerous and destructive. However, there are also significant differences: In the story of Joseph, his refusal leads to his imprisonment, but he is ultimately vindicated and rises to a position of power. In contrast, Hippolytus’ rejection of Phaedra leads to his death, underscoring the tragic nature of Greek heroism, where even the most virtuous are not immune to the whims of fate. Lastly, Euripides' language is rich in emotional intensity, capturing the turmoil and despair that drive the characters to their tragic ends. Phaedra’s monologues, in particular, offer insight into her conflicting emotions, torn between her illicit love for Hippolytus and her sense of duty and shame. I find Pausanias’ account is more straightforward and less emotionally charged. His language is more descriptive and factual, focusing on the sequence of events rather than the inner lives of the characters. This difference in linguistic style reflects the different purposes of the texts: Euripides’ play is a work of drama intended to evoke strong emotions and provoke thought, while Pausanias’ account is more concerned with documenting the myth for posterity. To critique, however, Euripides’ use of emotionally charged language and complex character interactions can be seen as a reflection of the intellectual and cultural climate of classical Athens, where issues of gender, morality, and human nature were hotly debated. Pausanias’ more restrained language, on the other hand, reflects his role as a chronicler of myths, where the emphasis is on preservation rather than interpretation. CC BY Aarushi Attray (contact)

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

In this work, the authors provide a comprehensive description of transcriptional regulation in Pseudomonas syringae by investigating the binding characteristics of various transcription factors. They uncover the hierarchical network structure of the transcriptome by identifying top-, middle-, and bottom-level transcription factors that govern the flow of information in the network. Additionally, they assess the functional variability and conservation of transcription factors across different strains of P. syringae by studying DNA-binding characteristics. These findings notably expand our current knowledge of the P. syringae transcriptome.

The findings associated with crosstalk between transcription factors and pathways, and the diversity of transcription factor functions across strains provide valuable insights into the transcriptional regulatory network of P. syringae. However, these results are at times underwhelming as their significance is unclear. This study would benefit from a discussion of the implications of transcription factor crosstalk on the functioning of the organism as a whole. Additionally, the implications of variability in transcription factor functions on the phenotype of the strains studied would further this analysis.<br /> Overall, this manuscript serves as a key resource for researchers studying the transcriptional regulatory network of P. syringae.

Thank you for your positive comments.

Reviewer #2 (Public Review):

Summary:

The phytopathogenic bacterium Pseudomonas syringae is comprised of many pathovars with different host plant species and has been used as a model organism to study bacterial pathogenesis in plants. Transcriptional regulation is key to plant infection and adaptation to host environments by this bacterium. However, researchers have focused on a limited number of transcription factors (TFs) that regulate virulence-related pathways. Thus, a comprehensive, systems-level understanding of regulatory interactions between transcription factors in P. syringae has not been achieved.

This study by Sun et al performed ChIP-seq analysis of 170 out of 301 TFs in P. syringae pv. syringae 1448A and used this unique dataset to infer transcriptional regulatory networks in this bacterium. The network analyses revealed hierarchical interactions between TFs, various network motifs, and co-regulation of target genes by TF pairs, which collectively mediate information flow. As discussed, the structure and properties of the P. syringae transcriptional regulatory networks are somewhat different from those identified in humans, yeast, and E. coli, highlighting the significance of this study. Further, the authors made use of the P. syringae transcriptional regulatory networks to find TFs of unknown functions to be involved in virulence-related pathways. For some of these TFs, their target specificity and biological functions, such as motility and biofilm formation, were experimentally validated. Of particular interest is the finding that despite conservation of TFs between P. syringae pv. syringae 1448A, P. syringae pv. tomato DC3000, P. syringae pv. syringae B728a, and P. syringae pv. actinidiae C48, some of the conserved TFs show different repertoires of target genes in these four P. syringae strains.

Thank you for your positive comments.

Strengths:

This study presents a systems-level analysis of transcriptional regulatory networks in relation to P. syringae virulence and metabolism, and highlights differences in transcriptional regulatory landscapes of conserved TFs between different P. syringae strains, and develops a user-friendly database for mining the ChIP-seq data generated in this study. These findings and resources will be valuable to researchers in the fields of systems biology, bacteriology, and plant-microbe interactions.

Thank you for your positive comments.

Weaknesses:

No major weaknesses were found, but some of the results may need to be interpreted with caution. ChIP-seq was performed with bacterial strains overexpressing TFs. This may cause artificial binding of TFs to promoters which may not occur when TFs are expressed at physiological levels. Another caution is applied to the interpretation of the biological functions of TFs. The biological roles of the tested TFs are based on in vitro experiments. Thus, functional relevance of the tested TFs during plant infection and/or survival under natural environmental conditions remains to be demonstrated.

Thank you for your comments, and we agree with the reviewer. To eliminate the artificial binding of TFs, we performed EMSA to verify the analyzed targets. Our EMSA results confirmed the analyzed binding peaks.

For the verification experiments of the biological functions of TFs, we also performed in vivo motility assay and biofilm production assay (Figures 3b-d). To further detect the biological functions of TFs, we performed plant infection assay of TF PSPPH2193 under natural environmental condition (bean leaves). As shown in Figures S6c and g, both the motility and the virulence of P. syringae in ∆PSPPH2193 strain was significantly reduced compared with WT strain. These results showed that TF PSPPH2193 positively regulated the pathogenicity of P. syringae via modulating the bacterial motility.

Reviewer #3 (Public Review):

Summary:

This study aims to understand gene regulation of the plant bacterial pathogen Pseudomonas syringae. Although the function of some TFs has been characterized in this strain, a global picture of the gene regulatory network remains elusive. The authors conducted a large-scale ChIP-seq analysis, covering 170 out of 301 TFs of this strain, and revealed gene regulatory hierarchy with functional validation of some previously uncharacterized TFs.

Thank you for your positive comments.

Strengths:

- This study provides one of the largest ChIP-seq datasets for a single bacterial strain, covering more than half of its TFs. This impressive resource enabled comprehensive systems-level analysis of the TF hierarchy.

- This study identified novel gene regulation and function with validations through biochemical and genetic experiments.

- The authors attempted on broad analyses including comparisons between different bacterial strains, providing further insights into the diversity and conservation of gene regulatory mechanisms.

Thank you for your positive comments.

Weaknesses:

(1) Some conclusions are not backed by quantitative or statistical analyses, and they are sometimes overinterpreted.

Thank you for your comments. We used hypergeometric test in this analysis. Although only one gene was enriched in some pathways, the adjusted p-value was less than 0.05. We added the details in the revised manuscript.

(2) Some figures and analyses are not well explained, and I was not able to understand them.

Thank you for your comments, and we are sorry for the confusion. We defined ‘indirect interaction’ as ‘co-association’ and ‘cooperativity’ as ‘if the common target of two TFs is from a TF’. We added the definition of "indirect interaction" and "cooperativity" in the revised legend.

For Figure S3a, the low co-association scores and large peak numbers of these top-level TFs indicated that top-level TFs preferred to solely regulate target genes, but not to co-regulate with other top-level TFs. PSPPH4700 was an example to show that top-level TFs with low co-association scores and large peak numbers tend to solely regulate target genes, but not to co-regulate with other top-level TFs. We revised the sentence to ‘For example, the top-level TF PSPPH4700 yielded over 1,700 peaks but cooperated with only 24 top-level TFs with low co-association scores about 0.05 (Supplementary Table 2b).’.

We analyzed high co-association scores of 125 TFs in three levels and further determined the co-association patterns. To identify the tendency of co-association of all these 125 TFs, the co-association patterns were classified into 4 clusters. Bottom-level TFs tend to co-regulate target genes with other TFs. We revised the sentence in the revised manuscript.

For Figure 2b, in C1, C2 and C4, many bottom-level TFs performed co-association pattern with other TFs, especially bottom TFs (showed in C4). To explore the regulatory pattern in C3, the peak locations in target genes of MexT were analyzed with those of TFs in C3. Seven top-level TFs (PSPPH1435, PSPPH1758, PSPPH2193, PSPPH2454, PSPPH4638, PSPPH4998 and PSPPH3411), three middle-level TFs (PSPPH1100, PSPPH5132 and PSPPH5144) and four bottom-level TFs (PSPPH0700, PSPPH2300, PSPPH2444 and PSPPH2580) were compared with MexT. MexT showed higher co-association scores (more than 60 scores) with more top-level-TFs. Therefore, we demonstrated that MexT performed closer co-association relationships with top-level TFs. We added the statement in the revised manuscript.

For Figure 1a, the hierarchical network showed different number of TFs in three levels (54 top-level TFs, 62 middle-level TFs and 147 bottom-level TFs), which indicated that more than half of TFs (bottom-level TFs) tend to be regulated by other TFs and then directly bound to target genes. This finding showed a downward regulatory direction of transcription regulation in P. syringae. We revised the statement in the revised manuscript.

(3) The Method section lacks depth, especially in data analyses. It is strongly recommended that the authors share their analysis codes so that others can reproduce the analyses.

Thank you for your comments, and we defined the intergenic region before each TF sequence as the promoter region. As pHM1 plasmid carries its own constitutive promoter (lacZ promoter), we amplified the TF-coding sequence and cloned into site following the promoter. The TF protein expression was activated by the promoter of plasmid. Psph 1448A was used for our main ChIP-seq. We added the details in the revised manuscript.

For Figure S3, we performed GO analysis on genes that were co-bound by TF pairs. We added the details in the revised manuscript.

We shared our analysis codes on the website (https://github.com/dengxinb2315/PS-PATRnet-code) in the Data Availability.

Recommendations for the authors

Reviewer #1 (Recommendations For The Authors):

(1) The specific strain of Pseudomonas syringae used in the study outside of the evolutionary analysis should be specified in the abstract and main text.

Thank you for your suggestion. We revised the statements in abstract and main text to specific strains.

(2) The language used throughout the manuscript should be revised for clarity, conciseness, and readability.

Thank you for your suggestion. We have revised the language used throughput the manuscript by a scientific editor who is a native speaker of English.

(2) Line 688: Replace "80C" with "-80C".

Thank you for your correction. We revised ‘80℃’ to ‘-80℃’. Please see Line 713.

(3) Line 172 - 173: The abbreviations TT, MM, BB, TM, TB, and MB need to be expanded in the main text before their use.

Thank you for your suggestion. We added the abbreviations TT, MM, BB, TM, TB, and MB in the manuscript. Please see Lines 172-174.

Reviewer #2 (Recommendations For The Authors):

Major points

(1) The name of the P. syringae strains used in each experiment/analysis should be explicitly stated (most experiments were carried out with P. syringae strain 1448A). This should also be applied to the introduction where many papers on P. syringae are cited without clear indication of strain names. I think this amendment is essential because target genes and thus biological functions of TFs could be different between P. syringae strains, as shown in the present study.

Thank you for your suggestion. We revised the P. syringae strains in the citations throughout the manuscript.

(2) How many TFs were analyzed throughout the study? Most sentences including line 22 in the abstract say 170, but I also found some say 270 (for example, line 106 and line 149). The legend of Figure 1 says 262. More detailed information is required regarding the datasets used for each analysis.

Thank you for your suggestion. The number of TFs analyzed by ChIP-seq in this research is 170, the number of TFs analyzed by HT-SELEX in our previous research is 100. Hierarchical analysis integrated data from ChIP-seq and HT-SELEX which included 270 TFs. As 8 TFs did not show hierarchical characteristic, the legend of Figure 1 said 262 TFs. We added the data source in the revised manuscript. Please see Lines 104, 147, 160 and 1082.

(3) Figure 1b: Please define "indirect interaction" and "cooperativity" in the legend as well as in the text. I only found the definition of "direct interaction".

Sorry for the missing information. We defined ‘indirect interaction’ and ‘cooperativity’ as ‘co-association’ and ‘if the common target of two TFs is from a TF’, respectively. We added the definition of "indirect interaction" and "cooperativity" in the revised legend. Please see Lines 174-176, 1084-1086.

(4) I found it very interesting that conserved TFs show different repertoires of target genes in different P. syringae strains. This suggests the rewiring of transcriptional regulatory networks in P. syringae strains, but the underlying mechanism is not explored in the current manuscript. It can be easily tested whether these conserved TFs bind to similar or different motifs by motif enrichment analysis. If they bind to similar motifs, it is possible that the promoter sequences of their target genes have diversified. Addressing or at least discussing these points would provide molecular insights into the diversification of the transcriptional regulatory networks in P. syringae. Similarly, functional enrichment analysis of target genes can be used to test whether the conserved TFs regulate different biological processes.

Thank you for your suggestion. We added the motif analysis and functional enrichment analysis of target genes of TFs (PSPPH3122 and PSPPH4127) in different P. syringae strains. We found two different motifs (AGACN4GATCAA and CGGACGN3GATCA) in 1448A and DC3000 strains, respectively. We also performed the GO analysis and found the specific functions of PSPPH3122 in Psph 1448A compared with Pst DC3000 and Pss B728a strains, including recombinase activity and DNA recombination. For PSPPH4127, we found four different motifs in four P. syringae strains. GO analysis showed its relationship with recombinase activity in Psph 1448A strain, and RNA binding, structural constituent of ribosome, translation and ribosome in Pss B728a strain. These results indicated the highly functional diversity of TFs in P. syringae. We added these points in the Results part, and Figure S9-S10 in the revised manuscript. Please see Lines 497-509.

(5) Related to point 4, it would be quite useful if a list of orthologous genes of 1448A TFs in the other tested P. syringae strains were provided. Such information may also enhance the utility of the database developed in this study.

Thank you for your suggestion. We added the list of orthologous genes of 301 Psph 1448A TFs in the other tested P. syringae strains in the Supplementary Table 5. Please see Lines 467 and Supplementary Table 5.

(6) Lines 243-246: It is unclear how these functional enrichment analyses were performed. Did you use target genes regulated by individual TFs or those coregulated by pairs of TFs? Please add more information for the sake of readers.

Thank you for your suggestion. We performed the functional enrichment analyses by hypergeometric test (BH-adjusted p < 0.05) via using target genes regulated by individual TFs. We added the details in the Results part. Please see Lines 248-252, 270, 1194-1195, 1199-1200 and 1205-1206.

Minor points

(1) Lines 167-168: I may not understand correctly, but you might want to say "downward-pointing edges" instead of "upward-pointing edges".

Thank you for correction. We revised the ‘upward-pointing edges’ to ‘downward-pointing edges’. Please see Line 166.

(2) Line 174: "physical interactions" should be amended to "direct interactions".

Thank you for correction. We revised the ‘physical interactions’ to ‘direct interactions’. Please see Line 177.

(3) Line 224: Could you please explain why bacterial growth in plant tissues is considered an example of "multi-stability"?

Thank you for your suggestion. We are sorry for the incorrect statement. We showed ‘plant intercellular spaces’ as ‘multi-stability’. We revised the sentence to ‘These auto-regulators are important and always act as repressors in scenarios of multi-stability, such as plant intercellular spaces’. Please see Lines 224-226.

(4) Line 254-257: Here, the definition of "tether binding" is introduced, but it is not very clear to me. In my understanding, tethered binding is an indirect binding of a TF to a target gene through protein-protein interaction with other TF that directly binds to the promoter of the target gene.

Thank you for your suggestion, and we agree with you. We referred to the paper published in 2012 (Wang et al., 2012) and revised the statement of ‘tether binding’ to ‘This finding suggested that these TFs indirectly regulated target genes through protein-protein interaction with other TFs that directly binds to the promoters of target genes, a phenomenon defined as tethered binding’. Please see Lines 259-262.

(5) Lines 341-343: Figure 3b shows qRT-PCR of hopAE1, not hrpR.

Thank you for your correction. We revised ‘hrpR’ to ‘hopAE1’. Please see Line 349.

(6) Lines 500 and Figure 6b: It is hard to see edges from module 12 to others. So, it would be better to provide numeric information (number of TFs and target genes) in the text.

Thank you for your suggestion. Module 12 includes 22 TFs and 318 target genes. We added the statement of numeric information about Module 12 in the revised manuscript. Please see Lines 536-537.

(7) Line 519: Figure S4b is not the EMSA data for PSPPH3798. Should it be Figure S4e?

Thank you for your correction. We revised to ‘Figure S4e’. Please see Line 545.

(8) Line 522: Figure S6b is not relevant to the statement here.

Thank you for your correction. We deleted the ‘Figure S6b’ here. Please see Line 547.

(9) Line 593: prokaryotic transcriptional regulatory networks -> eukaryotic transcriptional regulatory networks?

Thank you for your correction. We revised ‘prokaryotic transcriptional regulatory networks’ to ‘eukaryotic transcriptional regulatory networks’. Please see Line 618.

(10) Figure S3 requires images of higher resolution. Especially, values for the color codes are not readable or very hard to see.

Thank you for your suggestion. To make the images clearer, we enlarged the images, change the color codes, and divided it into three figures. Please see the revised Figures S3-S5 and corresponding Figure legends at Lines 1191-1206.

Reviewer #3 (Recommendations For The Authors):<br /> (1) Some conclusions are not backed by quantitative or statistical analyses, and they are sometimes overinterpreted.

L221: "Taken together, the simplest and most effective submodule M1 and the coregulatory submodule M13 played crucial roles in the transcriptional regulation of TFs in P. syringae."

The authors did not provide any evidence supporting the functional importance of any of these submodules. M13 is most enriched within the locked loop, but its size is much smaller than simple loops. What evidence supports the importance of this particular submodule?

Thank you for your suggestion. In eukaryote (Saccharomyces cerevisiae) and prokaryote (Escherichia coli) which have the best characterized transcriptional regulation networks, the feed-forward loop (called M13 in this article) appear numerous times in the networks and perform different biological functions. M1 appeared most frequently by an order of magnitude than other modules. We revised the sentence to ‘Taken together, the most numerous but simplest submodule M1 played a crucial role in the transcriptional regulation of TFs in P. syringae.’ Please see Lines 222-224.

L223: "...we found 92 auto-regulators...These auto-regulators are important and always act as repressors in scenarios of multi-stability, such as in plant intercellular spaces where bacteria grow (Figure 1d)(Alon, 2007). These regulators are regarded as bistable switches that further influence the expression of downstream genes."<br /> Are these claims supported by any evidence?

Thank you for your suggestion. We referred to the following articles:

(1) Alon. Nature Reviews Genetics. 2007(Alon, 2007).

That transcription factors repress the transcription of their target genes was considered as negative regulation. These negative autoregulators account for half of the repressors in E. coli and occur in many eukaryotes. The repressors controlled the concentration of the target production through suppressing its expression, which accelerated back to the steady state of cells.

(2) Becskei. et al. Nature. 2000; Rosenfeld et al. Journal of Molecular Biology. 2002 (Becskei & Serrano, 2000; Rosenfeld, Elowitz, & Alon, 2002).

Fluorescent assay confirmed that the negative autoregulatory module (negative autoregulator TetR) spent less time to the log phase than unregulated group, which reduced cell-to-cell fluctuations in the steady-state level of the transcription factor. Some negative autoregulators were showed here, such as LexA, CysB and SrlA-D.

In our research, we also identified many autoregulators including CysB and LexA2 (annotated as LexA repressor). We revised the sentence to ‘In addition, we found 92 auto-regulators in our hierarchy network. These auto-regulators are important and always act as repressors in scenarios of multi-stability, such as plant intercellular spaces (Figure 1d) (Alon, 2007). For example, LexA and CysB as negative autoregulators were indicated to reduce cell-to-cell fluctuations in the steady-state level of the transcription factor (Becskei & Serrano, 2000; Rosenfeld et al. 2002).’. Please see Lines 224-229.

L265: "This finding indicated that the bottom-level TFs, which were more easily regulated, tended to cooperate with downstream genes and other intra-level TFs."<br /> Could the authors provide more explanation to reach this conclusion from the data? Analyzing the number of highly co-accessing TFs does not sufficiently support this conclusion. The clustering of TFs (C1-C4) is incomplete, and each TF level (Top/Middle/Bottom) contains different numbers of TFs. Since the authors calculated all-by-all co-association scores for these 125 TFs, they can group these scores into 6 possible combinations (TT, TM, TB, MM, MB, BB) and show the distribution of co-association scores.

Thank you for your suggestion. We indicated that the bottom-level TFs preferred to regulate the target genes through the cooperation with other TFs. To further support the claim, we analyzed the proportion of the bottom TF interaction in all the TF pairs interactions and direct interaction based on results in Figure 1B. The interactions of bottom TFs were 43% and 49%, respectively. However, the interactions of top TFs and middle TFs were only 20% and 28%, respectively. We revised the statement ‘Based on the analysis in Figure 1B, we found that the proportions of bottom-level TF interaction in all the TF pair interactions and direct interaction were 43% and 49%. These results indicated that the bottom-level TFs tended to regulate downstream genes through cooperating with other level TFs.’ in the revised manuscript. Please see Lines 269-272.

As not every TF performed co-association with other TFs, we only collected 125 TFs with co-association scores. For the numbers of TF in each level, we divided TFs into three levels according to hierarchy height. Hierarchy height from -1 to -0.3 represented bottom level; hierarchy height from -0.3 to 0.3 represented middle level ; hierarchy height from 0.3 to 1 represents top level. Each level was equally divided by height scores. We suggested that different numbers of TFs in three levels indicated the characteristic of transcriptional regulation in P. syringae.

Thank you for your suggestion. As the co-association patterns were determined by co-association scores of the same TFs, we first grouped the co-association scores into 3 possible TF pairs (TT, MM, and BB, in Figures S3a, S4a and S5a). Our results indicated that higher co-association scores preferred to occur in bottom-level TFs. We revised the statement in the revised manuscript. Please see Lines 244-252.

(2) Some figures and analyses are not well explained, and I was not able to understand them.

Figure 1b: The terms "direct," "indirect," and "cooperativity" require further clarification as their definitions in the text (L169-183) are unclear to me. This ambiguity hampers the evaluation of the authors' discussion regarding TF-TF interactions (L561-584), an important theme of this study. The figure includes concepts discussed in later sections (e.g., cooperativity), making it difficult to understand. A diagram explaining these concepts would be highly helpful for readers to understand.

Sorry for the missing information. We defined ‘indirect interaction’ as ‘co-association’, ‘cooperativity’ as ‘if the common target of two TFs is from a TF’. We added the definition of "indirect interaction" and "cooperativity" in the revised manuscript and legend. Please see Lines 174-176 and 1085-1087.

L253: "Notably, we found that TFs at the top level, without cooperating TFs, exhibited a large number of binding peaks (Figure S3a)."

I could not understand this sentence. Did the authors mean that top-level TFs with a large number of peaks showed a low level of co-association? If so, does this data suggest that these TFs do not tend to cooperate with other TFs? I was confused by the discussion in L253-L261.

Thank you for your comment, and we agree with you. The low co-association scores and large peak numbers of these top-level TFs indicated that top-level TFs preferred to solely regulate target genes, but not to co-regulate with other top-level TFs.

Thank you for your comment. From L253-256, PSPPH4700 was an example to show that top-level TFs with low co-association scores and large peak numbers tend to solely regulate target genes, but not to co-regulate with other top-level TFs. We revised the sentence to ‘For example, the top-level TF PSPPH4700 yielded over 1,700 peaks, but cooperated with only 24 top-level TFs with low co-association scores about 0.05 (Supplementary Table 2b).’.

From L257-261, we analyzed high co-association scores of 125 TFs in three levels and further determined the co-association patterns. To identify the tendency of co-association of all these 125 TFs, the co-association patterns were classified into 4 clusters. Bottom-level TFs tend to co-regulate target genes with other TFs. We revised the sentence. Please see Lines 262-264, 265-266 and 269-272.

L287: "The analysis of the peak locations of MexT demonstrated that MexT showed closer co-association relationships with top-level TFs (Figure 2b)."

I could reach this conclusion by seeing Figure 2b. Additional explanation and/or data visualization would be appreciated.

Thank you for your suggestion. In C1, C2 and C4, many bottom-level TFs performed co-association pattern with other TFs, especially bottom TFs (showed in C4). To explore the regulatory pattern in C3, the peak locations in target genes of MexT were analyzed with those of TFs in C3. Seven top-level TFs (PSPPH1435, PSPPH1758, PSPPH2193, PSPPH2454, PSPPH4638, PSPPH4998 and PSPPH3411), three middle-level TFs (PSPPH1100, PSPPH5132 and PSPPH5144) and four bottom-level TFs (PSPPH0700, PSPPH2300, PSPPH2444 and PSPPH2580) were compared with MexT. MexT showed higher co-association scores (more than 60 scores) with more top-level-TFs. Therefore, we demonstrated that MexT performed closer co-association relationships with top-level TFs. We added the statement in the revised manuscript. Please see Lines 291-296.

Figure 6cd: What kind of enrichment analysis did the authors perform? Was any statistical test used? The figure only shows the number of genes, and sometimes the number is only 1 for a functional category. Can it be considered as significant enrichment?

Thank you for your comment. We used hypergeometric test in this analysis. Although only one gene was enriched in some pathways, the adjusted p-value was less than 0.05. We added the details in the revised manuscript. Please see Lines 533-534.

L169: "The hierarchical network revealed a downward information flow, suggesting the prioritization of collaboration between different hierarchy levels."<br /> Can the authors please explain the logic behind this statement more in detail?

Thank you for your comment. The hierarchical network showed different number of TFs in three levels (54 top-level TFs, 62 middle-level TFs and 147 bottom-level TFs), which indicated that more than half of TFs (bottom-level TFs) tend to be regulated by other TFs and then directly bound to target genes. This finding showed a downward regulatory direction of transcription regulation in P. syringae. We revised the statement in the revised manuscript. Please see Lines 167-170.

(3) The Method section lacks depth, especially on data analyses.

How did the authors define promoter regions of each gene? How were operons treated in their analyses? Was P. syringae 1448A used for their main ChIP-seq?

Thank you for your comment. We defined the intergenic region before each TF sequence as the promoter region.

As pHM1 plasmid carries its own constitutive promoter (lacZ promoter), we amplified the TF-coding sequence and cloned into the site following the promoter. The TF protein expression was activated by the promoter of plasmid.

P. syringae 1448A was used for our main ChIP-seq. We added the details in the revised manuscript. Please see Lines 705 and 727-730.

Figure S3: I am not sure how the GO analyses were done. For example, in the case of the top-level TF PSPPH4700, did the authors perform GO analysis on genes that are co-bound by PSPPH4700 and any other top-level TFs?

Thank you for your comment and we agree with you. We performed GO analysis on genes that were co-bound by TF pairs in the same level. We added the details in the revised manuscript. Please see Lines 248-252.

The analysis presented in Figure 6a needs more explanation of the methodology employed by the authors.

Thank you for your comment. We added more details for the analysis in Figure 6a. Please see Lines 514-522.

It is strongly recommended that the authors share their analysis codes so that others can reproduce the analyses.

Thank you for your comment. We shared our analysis codes on the website (https://github.com/dengxinb2315/PS-PATRnet-code) in the Data Availability. Please see Lines 800-801.

(4) Other:

Figure 3: I suggest putting additional panel labels to facilitate the interpretation of the figure.

Thank you for your suggestion. We added detailed labels in the revised Figures 3 and 4. Please see in the revised Figures 3 and 4.

I spotted several potential errors:

L106: 170 TFs?

Thank you for your comment, and we are sorry for the missing details. For the hierarchical network, we integrated the DNA-binding data of 170 TFs in this study and 100 TFs in our previous SELEX research. We added the details in the revised manuscript. Please see Lines 104, 147 and 159-160.

L592: P. syringae not E. coli?

Thank you for your comment. Here we discussed the hierarchical characteristics in E. coli. We revised the statement in the revised manuscript. Please see Line 618.

L593: eukaryotic not prokaryotic?

Thank you for your correction. Here we discussed the feedforward loops in our study. We revised the statement in the revised manuscript. Please see Line 618.

References

Alon, U. (2007). Network motifs: theory and experimental approaches. Nature Reviews Genetics, 8(6), 450-461.

Becskei, A., & Serrano, L. (2000). Engineering stability in gene networks by autoregulation. Nature, 405(6786), 590-593.

Rosenfeld, N., Elowitz, M. B., & Alon, U. (2002). Negative autoregulation speeds the response times of transcription networks. Journal of molecular biology, 323(5), 785-793.

Wang, J., Zhuang, J., Iyer, S., Lin, X., Whitfield, T. W., Greven, M. C., . . . Cheng, Y. (2012). Sequence features and chromatin structure around the genomic regions bound by 119 human transcription factors. Genome research, 22(9), 1798-1812.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

Referee #2

Evidence, reproducibility and clarity

Corridori et al introduce IGNITE, a computational framework to infer gene regulatory networks (GRNs) from scRNA-seq data leveraging the kinetic Ising model, which can be used to simulate synthetic gene expression and perform in-silico knockout experiments. Other similar frameworks exist, but none combine these three aspects together. The authors have generated a scRNA-seq of murine ESCs differentiation which they use to compare their method with others. Specifically they show that they can infer known regulatory interactions, that they can generate similar data than the original and that it can potentially predict gene expression changes in transcription factor knock-out perturbations.

Major comments:

• Many of the authors' claims are backed by qualitative results and not properly quantified. In Fig2, authors qualitatively compare intra gene correlations between genes for the original data and their prediction. Instead of just visualizing they should compute and report the Spearman correlation between the original expression and the predicted one. The Fraction of Agreement is not a good metric to compare knockout predictions since it is completely dependent on the class imbalance of signs, for example if the selected genes are 75% positive and 25% negative, a naive predictor that only outputs positive predictions will still have a high score. Instead, the authors should quantify this with Spearman correlation or RMSE and compare across methods. In FigS4a-b the authors qualitatively claim that other methods could not predict the expected cell composition, which they should quantify and report the values across methods. When comparing against the ground truth network, the fraction of correctly inferred interactions is technically the same as precision but is ignoring recall. I suggest the authors compute precision, recall and a combined F1 score to compare the evaluated methods. Authors claim that the method is scalable to a larger number of genes but no data is provided, they should show how their method compares to others when using a different number of cells and number of genes at memory usage and running time.
• The authors need to better describe which tests were performed when talking about significance, which thresholds and which corrections, if any, were employed.
• To reduce the number of dimensions of scRNA-seq data the authors use t-SNE and then from the obtained result UMAP to project the data into a lower dimensional space. This is fundamentally wrong since distances are not well preserved in t-SNE. Instead the authors should first employ PCA and then UMAP. Additionally, the authors use UMAP distances in the Slingshot pseudotime calculation. Similar to t-SNE, UMAP distances have no real meaning and should only be used for visualization purposes. Instead, the authors should provide Slingshot the obtained PCA embeddings.
• Dictys (PMID: 37537351) is a known GRN inference method that also can simulate gene expression but is missing in the benchmark, the authors should add it to the method comparison.
• The current manuscript is not reproducible since it is missing the method's code, the code to reproduce the figures and the generated scRNA-seq data.
• Authors claim that the method is scalable to a larger number of genes but no data is provided to back this claim. They should show how their method compares to others when using a different number of cells and number of genes.

Minor points:

• In the introduction, authors mention multimodal GRN inference methods but do not provide any references.
• In Table 1, CellOracle is annotated as not being able to do multiple KO which is wrong. Additionally, the authors mention that IGNITE uses no prior knowledge which is not really true since it requires pseudotime ordering. The authors should add a column to Table 1 whether methods require pseudotime.
• It is unclear what the dashed arrow of Fig1b means. Moreover, plotting gene expression values on top of UMAPs can be misleading, instead authors should plot the gene expression distributions binned by pseudotime.
• The authors report a p-value of 1.04x10-171 which is below detection limit (see PMID: 30921532). Authors should change it to an interval such as p < 2.2×10-16.
• To make CellOracle results easier to interpret and more comparable, authors should run it at the atlas level instead of at the cell type level, this way generating only one GRN. This can be achieved by assigning the same cluster label to all cells.
• Experimental values in FigS3b seem to have been repeated and do not match the previous ones for IGNITE and SCODE.
• It is unclear what the different circles mean in Fig5b.

Significance

This manuscript is an incremental and methodological work for specialized audiences. Its strengths are that the authors employ kinetic Ising model for GRN inference and that they provide a single framework capable of inferring, simulating and perturbing gene expression. The main limitations are that the claims should be better quantified and that the code and data need to be made accessible.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

Referee #1

Evidence, reproducibility and clarity

Summary

Corridori and colleagues propose IGNITE, a novel method to recover Gene Regulatory Networks (GRN) from single cell RNA-sequencing (scRNA-seq) data. Their method solves the inverse Ising problem generating a cohort of candidate GRN optimising it to minimise the difference to the input expression matrix. Authors report the IGNITE is able to predict wild type data and simulate both single and multiple gene knockouts. Authors benchmark this method on a in-house data set of differentiating pluripotent stem cells (PSC). They focus on a small set of genes known to be involved in PSC differentiation into formative cells. Authors benchmark IGNITE against state of the art tools (SCODE, MaxEnt and CELLORACLE). They evaluate IGNITE ability to predict wild type gene expression by comparing their data with experimental data and with SCODE. They conclude the tool has generative capacity comparable with SCODE. They also evaluate IGNITE ability to recover known interactions with respect to other tools without finding it to significantly outperform them.

Major comments

• Are the key conclusions convincing?

Conclusions appear convincing although model generalizability could be shown in a more thorough manner. For instance, analysing some other publicly available dataset could help demonstrate hyperparameters effects on GRN predictions and their robustness across different experiments. - Should the authors qualify some of their claims as preliminary or speculative, or remove them altogether?

Claims are well supported by data. - Would additional experiments be essential to support the claims of the paper? Request additional experiments only where necessary for the paper as it is, and do not ask authors to open new lines of experimentation.

I think the work would benefit from an additional benchmark on a different cellular system. This experiment would show how hyperparameters generalise across datasets and would provide potential users insights how to tweak them.

Also, how does the model scale with the number of genes? A benchmark on computation time and resources required to infer GRN of growing size would be valuable in the adoption of this tool.

In addition, I think the GRN comparison benchmark presented in section (3.4) would benefit from a quantitative discussion. Authors show inferred GRNs in Figure 4 and S5. For instance, measuring matrix similarity (when appropriate) would help understanding how predicted GRN compare. I understand authors attempt to do so by focusing on validated interactions and computing the fraction of correctly inferred interactions (FCI) but I think a measurement of the overall similarity (eg. Pearson correlation) would add on this.

Another comment regards the dependency between Correlation Matrices Distance (CMD) and FCI, shown in Figure 5. I understand that IGNITE GRN that maximise FCI are not the same that minimise CMD. However, it looks like GRN that maximise FCI have higher value in terms of biological information. I wonder whether optimization for one or the other metric could be left to the end user as a tunable parameter.

Authors should discuss why the expression of some genes does not follow the expected trends (Fig 1C vs Fig S1A). Out of the 24 genes they select for their analysis, at least four do not follow the expected trends: Sox2, according to literature, is a Naive gene, however, in Figure 1C its gene expression pattern is more similar to Formative late genes. Other genes with similar "unexpected" patterns are Zic3, Etv4 and Sall4.

Are the suggested experiments realistic in terms of time and resources? It would help if you could add an estimated cost and time investment for substantial experiments.

I think suggested experiments are doable as long as authors get publicly available data, i.e. the in-house dataset they generated for this study is enough to show applicability. For example datasets analysed in SCODE paper (https://doi.org/10.1093/bioinformatics/btx194) could be used as second benchmark. The point of applying the tool to another dataset is to show how it generalises across different biological systems, experiments and, potentially, sequencing technologies. - Are the data and the methods presented in such a way that they can be reproduced?

The methods section is really clear. To enable reproducibility both raw scRNA-seq data, the IGNITE source code and code written to benchmark it should be released in the public domain in appropriate repositories (eg. ENA, GitHub, Binder etc). - Are the experiments adequately replicated and statistical analysis adequate?

Yes.

Minor comments

• Specific experimental issues that are easily addressable.

Related to the Sox2 expression pattern is the binarization shown in Figure 2D. How is it possible that Sox2 is always marked as active? Could the authors clarify how these outlier behaviours emerge and propose mitigation strategies, if any?

In section 5.11.2 it is unclear if xi are in log scale or not. Since the model starts from binarized, log transformed expression values, should not generated ones be in the same scale as the input? - Are prior studies referenced appropriately?

Yes, referencing is clear. - Are the text and figures clear and accurate?

Yes, figures appear to be clear, readable and well documented both in captions and main text. - Do you have suggestions that would help the authors improve the presentation of their data and conclusions?

Section 3.3 could be improved by better describing experimental datasets. Only in the methods section it is clearly stated that experimental data for single KO experiments were retrieved from the literature.

Check typesetting:

• parenthesis missing in Eq. 1
• Leftover $ in section 3.1
• Parenthesis missing in Section 3.3
• Misplaced comma in section 5.2.1

Significance

• Describe the nature and significance of the advance (e.g. conceptual, technical, clinical) for the field.

The paper presents a method to infer GRN from scRNA-seq data alone. Applications include GRN prediction and their perturbations. This paper represents a technical advance in the field as it is the first application of the inverse Ising problem GRN inference. - Place the work in the context of the existing literature (provide references, where appropriate).

The paper itself presents the landscape of GRN inference tools using scRNA-seq data: SCODE, MaxEnt and CELLORACLE. More tools exist, for instance SCENIC (https://doi.org/10.1038/nmeth.4463) mainly relies on co-expression matrices. Other tools exist but require additional data types e.g. GRaNIE and GRaNPA (https://doi.org/10.15252/msb.202311627) leverage on physical interaction data (ATAC-seq, ChIP-seq). Similarly DeepFlyBrain uses deep neural networks to infer eGRN in Drosophila (https://doi.org/10.1038/s41586-021-04262-z). The value of tools like IGNITE and its competitors is that they do not require additional data types, which, in turn, helps in controlling experimental costs. - State what audience might be interested in and influenced by the reported findings.

The paper might be of interest to biologists interested in regulation of gene expression. The tool might turn out to be useful in planning experimental work by guiding the choice of perturbations to introduce in experimental systems. - Define your field of expertise with a few keywords to help the authors contextualize your point of view. Indicate if there are any parts of the paper that you do not have sufficient expertise to evaluate.

I am a computational biologist.

I have no sufficient expertise to evaluate the mathematical details of the method.

Le code crée des architectures linguistiques, des unités de structures qui vont constituer des unités encore plus grandes (e.g. les réseaux sociaux)

Author response:

Reviewer #1 - Public Review

This report describes work aiming to delineate multi-modal MRI correlates of psychopathology from a large cohort of children of 9-11 years from the ABCD cohort. While uni-modal characterisations have been made, the authors rightly argue that multi-modal approaches in imaging are vital to comprehensively and robustly capture modes of large-scale brain variation that may be associated with pathology. The primary analysis integrates structural and resting-state functional data, while post-hoc analyses on subsamples incorporate task and diffusion data. Five latent components (LCs) are identified, with the first three, corresponding to p-factor, internal/externalising, and neurodevelopmental Michelini Factors, described in detail. In addition, associations of these components with primary and secondary RSFC functional gradients were identified, and LCs were validated in a replication sample via assessment of correlations of loadings.

1.1) This work is clearly novel and a comprehensive study of associations within this dataset. Multi-modal analyses are challenging to perform, but this work is methodologically rigorous, with careful implementation of discovery and replication assessments, and primary and exploratory analyses. The ABCD dataset is large, and behavioural and MRI protocols seem appropriate and extensive enough for this study. The study lays out comprehensive associations between MRI brain measures and behaviour that appear to recapitulate the established hierarchical structure of psychopathology.

We thank Reviewer 1 for appreciating our methods and findings, and we address their suggestions below:

1.2) The work does have weaknesses, some of them acknowledged. There is limited focus on the strength of observed associations. While the latent component loadings seem reliably reproducible in the behavourial domain, this is considerably less the case in the imaging modalities. A considerable proportion of statistical results focuses on spatial associations in loadings between modalities - it seems likely that these reflect intrinsic correlations between modalities, rather than associations specific to any latent component.

We appreciate the Reviewer’s comment, and minimized the reporting of correlations between the loadings from the different modalities in the revised Results (specifically subsections on LC1, LC2, and LC3). We now refer to Table S4 in each subsection for this information: “Spatial correlations between modality-specific loadings are reported in Supplementary file 1c.”

For completeness, we report the intrinsic correlations between the different modalities in Supplementary file 1c (P.19):

“Lastly, although the current work aimed to reduce intrinsic correlations between variables within a given modality through running a PCA before the PLS approach, intrinsic correlations between measures and modalities may potentially be a remaining factor influencing the PLS solution. We, thus, provided an additional overview of the intrinsic correlations between the different neuroimaging data modalities in the supporting results (Supplementary file 1c).”

1.3) Assessment of associations with functional gradients is similarly a little hard to interpret. Thus, it is hard to judge the implications for our understanding of the neurophysiological basis of psychopathology and the ability of MRI to provide clinical tools for, say, stratification.

We now provide additional context, including a rising body of theoretical and empirical work, that outlines the value of functional gradients and cortical hierarchies in the understanding of brain development and psychopathology. Please see P.26.

“Initially demonstrated at the level of intrinsic functional connectivity (Margulies et al., 2016), follow up work confirmed a similar cortical patterning using microarchitectural in-vivo MRI indices related to cortical myelination (Burt et al., 2018; Huntenburg et al., 2017; Paquola et al., 2019), post-mortem cytoarchitecture (Goulas et al., 2018; Paquola et al., 2020, 2019), or post-mortem microarray gene expression (Burt et al., 2018). Spatiotemporal patterns in the formation and maturation of large-scale networks have been found to follow a similar sensory-to-association axis; moreover, there is the emerging view that this framework may offer key insights into brain plasticity and susceptibility to psychopathology (Sydnor et al., 2021). In particular, the increased vulnerability of transmodal association cortices in late childhood and early adolescence has been suggested to relate to prolonged maturation and potential for plastic reconfigurations of these systems (Paquola et al., 2019; Park et al., 2022b). Between mid-childhood and early adolescence, heteromodal association systems such as the default network become progressively more integrated among distant regions, while being more differentiated from spatially adjacent systems, paralleling the development of cognitive control, as well as increasingly abstract and logical thinking. [...] This suggests that neurodevelopmental difficulties might be related to alterations in various processes underpinned by sensory and association regions, as well as the macroscale balance and hierarchy of these systems, in line with previous findings in several neurodevelopmental conditions, including autism, schizophrenia, as well as epilepsy, showing a decreased differentiation between the two anchors of this gradient (Hong et al., 2019). In future work, it will be important to evaluate these tools for diagnostics and population stratification. In particular, the compact and low dimensional perspective of gradients may provide beneficial in terms of biomarker reliability as well as phenotypic prediction, as previously demonstrated using typically developing cohorts (Hong et al. 2020) On the other hand, it will be of interest to explore in how far alterations in connectivity along sensory-to-transmodal hierarchies provide sufficient graduality to differentiate between specific psychopathologies, or whether they, as the current work suggests, mainly reflect risk for general psychopathology and atypical development.”

1.4) The observation of a recapitulation of psychopathology hierarchy may be somewhat undermined by the relatively modest strength of the components in the imaging domain.

We thank the Reviewer for this comment, and now expressed this limitation in the revised Discussion, P.23.

“The p factor, internalizing, externalizing, and neurodevelopmental dimensions were each associated with distinct morphological and intrinsic functional connectivity signatures, although these relationships varied in strength.”

1.5) The task fMRI was assessed with a fairly basic functional connectivity approach, not using task timings to more specifically extract network responses.

In the revised Discussion on P.24, we acknowledge that more in-depth analyses of task-based fMRI may have offered additional insights into state-dependent changes in functional architecture.

“While the current work derived main imaging signatures from resting-state fMRI as well as grey matter morphometry, we could nevertheless demonstrate associations to white matter architecture (derived from diffusion MRI tractography) and recover similar dimensions when using task-based fMRI connectivity. Despite subtle variations in the strength of observed associations, the latter finding provided additional support that the different behavioral dimensions of psychopathology more generally relate to alterations in functional connectivity. Given that task-based fMRI data offers numerous avenues for analytical exploration, our findings may motivate follow-up work assessing associations to network- and gradient-based response strength and timing with respect to external stimuli across different functional states.”

1.6) Overall, the authors achieve their aim to provide a detailed multimodal characterisation of MRI correlations of psychopathology. Code and data are available and well organised and should provide a valuable resource for researchers wanting to understand MRI-based neural correlates of psycho-pathology-related behavioural traits in this important age group. It is largely a descriptive study, with comparisons to previous uni-modal work, but without particularly strong testing of neuroscience hypotheses.

We thank the Reviewer for recognizing the detail and rigor of data-driven study and extensive code and data documentation.

Reviewer #2 - Public Review

In "Multi-modal Neural Correlates of Childhood Psychopathology" Krebets et al. integrate multi-modal neuroimaging data using machine learning to delineate dissociable links to diverse dimensions of psychopathology in the ABCD sample. This paper had numerous strengths including a superb use of a large resource dataset, appropriate analyses, beautiful visualizations, clear writing, and highly interpretable results from a data-driven analysis. Overall, I think it would certainly be of interest to a general readership. That being said, I do have several comments for the authors to consider.

We thank Dr Satterthwaite for the positive evaluation and helpful comments.

2.1) Out-of-sample testing: while the permutation testing procedure for the PLS is entirely appropriate, without out-of-sample testing the reported effect sizes are likely inflated.

As discussed in the editorial summary of essential revisions, we agree that out-of-sample prediction indeed provides stronger estimates of generalizability. We assess this by applying the PCA coefficients derived from the discovery cohort imaging data to the replication cohort imaging data. The resulting PCA scores and behavioral data were then z-scored using the mean and standard deviation of the replication cohort. The SVD weights derived from the discovery cohort were applied to the normalized replication cohort data to derive imaging and behavioral composite scores, which were used to recover the contribution of each imaging and behavioral variable to the LCs (i.e., loadings). Out-of-sample replicability of imaging (mean r=0.681, S.D.=0.131) and behavioral (mean r=0.948, S.D.=0.022) loadings was generally high across LCs 1-5. This analysis is reported in the revised manuscript (P.18).

“Generalizability of reported findings was also assessed by directly applying PCA coefficients and latent components weights from the PLS analysis performed in the discovery cohort to the replication sample data. Out-of-sample prediction was overall high across LCs1-5 for both imaging (mean r=0.681, S.D.=0.131) and behavioral (mean r=0.948, S.D.=0.022) loadings.”

2.2) Site/family structure: it was unclear how site/family structure were handled as covariates.

Only unrelated participants were included in discovery and replication samples (see P.6). The site variable was regressed out of the imaging and behavioral data prior to the PLS analysis using the residuals from a multiple linear model which also included age, age2, sex, and ethnicity. This is now clarified on P.29:

“Prior to the PLS analysis, effects of age, age2, sex, site, and ethnicity were regressed out from the behavioral and imaging data using a multiple linear regression to ensure that the LCs would not be driven by possible confounders (Kebets et al., 2021, 2019; Xia et al., 2018). The imaging and behavioral residuals of this procedure were input to the PLS analysis.”

2.3) Anatomical features: I was a bit surprised to see volume, surface area, and thickness all evaluated - and that there were several comments on the correspondence between the SA and volume in the results section. Given that cortical volume is simply a product of SA and CT (and mainly driven by SA), this result may be pre-required.

As suggested, we reduced the reporting of correlations between the loadings from the different modalities in the revised Results (specifically subsections on LC1, LC2, and LC3). Instead, we now refer to Table S4 in each subsection for this information: “Spatial correlations between modality-specific loadings are reported in Supplementary file 1c.”

We also reran the PLS analysis while only including thickness and surface area as our structural metrics, to account for potential redundancy of these measures with volume. This analysis and associated findings are reported on P.36 and P.19:

“As cortical volume is a result of both thickness and surface area, we repeated our main PLS analysis while excluding cortical volume from our imaging metrics and report the consistency of these findings with our main model.”

“Third, to account for redundancy within structural imaging metrics included in our main PLS model (i.e., cortical volume is a result of both thickness and surface area), we also repeated our main analysis while excluding cortical volume from our imaging metrics. Findings were very similar to those in our main analysis, with an average absolute correlation of 0.898±0.114 across imaging composite scores of LCs 1-5.”

2.4) Ethnicity: the rationale for regressing ethnicity from the data was unclear and may conflict with current best practices.

We thank the Reviewer for this comment. In light of recent discussions on including this covariate in large datasets such as ABCD (e.g., Saragosa-Harris et al., 2022), we elaborate on our rationale for including this variable in our model in the revised manuscript on P.30:

“Of note, the inclusion of ethnicity as a covariate in imaging studies has been recently called into question. In the present study, we included this variable in our main model as a proxy for social inequalities relating to race and ethnicity alongside biological factors (age, sex) with documented effects on brain organization and neurodevelopmental symptomatology queried in the CBCL.”

We also assess the replicability of our analyses when removing race and ethnicity covariates prior to computing the PLS analysis and correlating imaging and behavioral composite scores across both models. We report resulting correlations in the revised manuscript (P.37, 19, and 27):

“We also assessed the replicability of our findings when removing race and ethnicity covariates prior to computing the PLS analysis and correlating imaging and behavioral composite scores across both models.”

“Moreover, repeating the PLS analysis while excluding this variable as a model covariate yielded overall similar imaging and behavioral composites scores across LCs to our original analysis. Across LCs 1-5, the average absolute correlations reached r=0.636±0.248 for imaging composite scores, and r=0.715±0.269 for behavioral composite scores. Removing these covariates seemed to exert stronger effects on LC3 and LC4 for both imaging and behavior, as lower correlations across models were specifically observed for these components.”

“Although we could consider some socio-demographic variables and proxies of social inequalities relating to race and ethnicity as covariates in our main model, the relationship of these social factors to structural and functional brain phenotypes remains to be established with more targeted analyses.”

2.5) Data quality: the authors did an admirable job in controlling for data quality in the analyses of functional connectivity data. However, it is unclear if a comparable measure of data quality was used for the T1/dMRI analyses. This likely will result in inflated effect sizes in some cases; it has the potential to reduce sensitivity to real effects.

We agree that data quality was not accounted for in our analysis of T1w- and diffusion-derived metrics. We now accounted for T1w image quality by adding manual quality control ratings to the regressors applied to all structural imaging metrics prior to performing the PLS analysis, and reported the consistency of this new model with original findings. See P.36, P.19:

“We also considered manual quality control ratings as a measure of T1w scan quality. This metric was included as a covariate in a multiple linear regression model accounting for potential confounds in the structural imaging data, in addition to age, age2, sex, site, ethnicity, ICV, and total surface area. Downstream PLS results were then benchmarked against those obtained from our main model.”

“Considering scan quality in T1w-derived metrics (from manual quality control ratings) yielded similar results to our main analysis, with an average correlation of 0.986±0.014 across imaging composite scores.”

As for diffusion imaging, we also regressed out effects of head motion in addition to age, age2, sex, site, and ethnicity from FA and MD measures and reported the consistency with our original results (P.36, P.19):

“We tested another model which additionally included head motion parameters as regressors in our analyses of FA and MD measures, and assessed the consistency of findings from both models.”

“Additionally considering head motion parameters from diffusion imaging metrics in our model yielded consistent results to those in our main analyses (mean r=0.891, S.D.=0.103; r=0.733-0.998).”

Reviewer #3 - Public Review

In this study, the authors utilized the Adolescent Brain Cognitive Development dataset to investigate the relationship between structural and functional brain network patterns and dimensions of psychopathology. They identified multiple components, including a general psychopathology (p) factor that exhibited a strong association with multimodal imaging features. The connectivity signatures associated with the p factor and neurodevelopmental dimensions aligned with the sensory-to-transmodal axis of cortical organization, which is linked to complex cognition and psychopathology risk. The findings were consistent across two separate subsamples and remained robust when accounting for variations in analytical parameters, thus contributing to a better understanding of the biological mechanisms underlying psychopathology dimensions and offering potential brain-based vulnerability markers.

3.1) An intriguing aspect of this study is the integration of multiple neuroimaging modalities, combining structural and functional measures, to comprehensively assess the covariance with various symptom combinations. This approach provides a multidimensional understanding of the risk patterns associated with mental illness development.

We thank the Reviewer for acknowledging the multimodal approach, and for the constructive suggestions.

3.2) The paper delves deeper into established behavioral latent variables such as the p factor, internalizing, externalizing, and neurodevelopmental dimensions, revealing their distinct associations with morphological and intrinsic functional connectivity signatures. This sheds light on the neurobiological underpinnings of these dimensions.

We are happy to hear the Reviewer appreciates the gain in understanding neural underpinnings of dimensions of psychopathology resulting from the current work.

3.3) The robustness of the findings is a notable strength, as they were validated in a separate replication sample and remained consistent even when accounting for different parameter variations in the analysis methodology. This reinforces the generalizability and reliability of the results.

We appreciate that the Reviewer found our robustness and generalizability assessment convincing.

3.4) Based on their findings, the authors suggest that the observed variations in resting-state functional connectivity may indicate shared neurobiological substrates specific to certain symptoms. However, it should be noted that differences in resting-state connectivity between groups can stem from various factors, as highlighted in the existing literature. For instance, discrepancies in the interpretation of instructions during the resting state scan can influence the results. Hence, while their findings may indicate biological distinctions, they could also reflect differences in behavior.

For the ABCD dataset, resting-state fMRI scans were based on eyes open and passive viewing of a crosshair, and are thus homogenized. We acknowledge, however, that there may still be state-to-state fluctuations contributing to the findings, and this is now discussed in the revised Discussion, on P.28. Note, however, that prior literature has generally also suggested rather modest impacts of cognitive and daily variation on resting-state functional networks, compared to much more dominating inter-individual and inter-group factors.

“Finally, while prior research has shown that resting-state fMRI networks may be affected by differences in instructions and study paradigm (e.g., with respect to eyes open vs closed) (Agcaoglu et al., 2019), the resting-state fMRI paradigm is homogenized in the ABCD study to be passive viewing of a centrally presented fixation cross. It is nevertheless possible that there were slight variations in compliance and instructions that contributed to differences in associated functional architecture. Notably, however, there is a mounting literature based on high-definition fMRI acquisitions suggesting that functional networks are mainly dominated by common organizational principles and stable individual features, with substantially more modest contributions from task-state variability (Gratton et al. 2018). These findings, thus, suggest that resting-state fMRI markers can serve as powerful phenotypes of psychiatric conditions, and potential biomarkers (Abraham et al., 2017; Gratton et al., 2020; Parkes et al., 2020).”

3.5) The authors conducted several analyses to investigate the relationship between imaging loadings associated with latent components and the principal functional gradient. They found several associations between principal gradient scores and both within- and between-network resting-state functional connectivity (RSFC) loadings. Assessing the analysis presented here proves challenging due to the nature of relating loadings, which are partly based on the RSFC, to gradients derived from RSFC. Consequently, a certain level of correlation between these two variables would be expected, making it difficult to determine the significance of the authors' findings. It would be more intriguing if a direct correlation between the composite scores reflecting behavior and the gradients were to yield statistically significant results.

We thank the Reviewer for the comment, and agree that investigating gradient-behavior relationships could offer additional insights into the neural basis of psychiatric symptomatology. However, the current analysis pipeline precludes this direct comparison which is performed on a region-by-region basis across the span of the cortical gradient. Indeed, the behavioral loadings are provided for each CBCL item, and not cortical regions.

The Reviewer also evokes concerns of potential circularity in our analysis, as we compared imaging loadings, which are partially based on RSFC, and gradient values generated from the same RSFC data. In response to this comment, we cross-validated our findings using an RSFC gradient derived from an independent dataset (HCP), showing highly consistent findings to those presented in the manuscript. This correlation is now reported in the Results section P.15.

“A similar pattern of findings was observed when cross-validating between- and within-network RSFC loadings to a RSFC gradient derived from an independent dataset (HCP), with strongest correlations seen for between-network RSFC loadings for LC1 and LC3 (LC1: r=0.50, pspin<0.001; LC3: r=0.37, pspin<0.001).”

We furthermore note similar correlations between imaging loadings and T1w/T2w ratio in the same participants, a proxy of intracortical microstructure and hierarchy (Glasser et al., 2011). These findings are now detailed in the revised Results, P.15-16:

“Of note, we obtain similar correlations when using T1w/T2w ratio in the same participants, a proxy of intracortical microstructure and hierarchy (Glasser et al., 2011). Specifically, we observed the strongest association between this microstructural marker of the cortical hierarchy and between-network RSFC loadings related to LC1 (r=-0.43, pspin<0.001).”

3.6) Lastly, regarding the interpretation of the first identified latent component, I have some reservations. Upon examining the loadings, it appears that LC1 primarily reflects impulse control issues rather than representing a comprehensive p-factor. Furthermore, it is worth noting that within the field, there is an ongoing debate concerning the interpretation and utilization of the p-factor. An insightful publication on this topic is "The p factor is the sum of its parts, for now" (Fried et al, 2021), which explains that the p-factor emerges as a result of a positive manifold, but it does not necessarily provide insights into the underlying mechanisms that generated the data.

We thank the Reviewer for this comment, and added greater nuance into the discussion of the association to the p factor. We furthermore discuss some of the ongoing debate about the use of the p factor, and cite the recommended publication on P.27.

“Other factors have also been suggested to impact the development of psychopathology, such as executive functioning deficits, earlier pubertal timing, negative life events (Brieant et al., 2021), maternal depression, or psychological factors (e.g., low effortful control, high neuroticism, negative affectivity). Inclusion of such data could also help to add further mechanistic insights into the rather synoptic proxy measure of the p factor itself (Fried et al., 2021), and to potentially assess shared and unique effects of the p factor vis-à-vis highly correlated measures of impulse control.”

Reviewer #1 (Public Review):

This report describes work aiming to delineate multi-modal MRI correlates of psychopathology from a large cohort of children of 9-11 years from the ABCD cohort. While uni-modal characterisations have been made, the authors rightly argue that multi-modal approaches in imaging are vital to comprehensively and robustly capture modes of large-scale brain variation that may be associated with pathology. The primary analysis integrates structural and resting-state functional data, while post-hoc analyses on subsamples incorporate task and diffusion data. Five latent components (LCs) are identified, with the first three, corresponding to p-factor, internal/externalising, and neurodevelopmental Michelini Factors, described in detail. In addition, associations of these components with primary and secondary RSFC functional gradients were identified, and LCs were validated in a replication sample via assessment of correlations of loadings.

This work is clearly novel and a comprehensive study of associations within this dataset. Multi-modal analyses are challenging to perform, but this work is methodologically rigorous, with careful implementation of discovery and replication assessments, and primary and exploratory analyses. The ABCD dataset is large, and behavioural and MRI protocols seem appropriate and extensive enough for this study. The study lays out comprehensive associations between MRI brain measures and behaviour that appear to recapitulate the established hierarchical structure of psychopathology.

The work does have weaknesses, some of them acknowledged. There is limited focus on the strength of observed associations. While the latent component loadings seem reliably reproducible in the behavourial domain, this is considerably less the case in the imaging modalities. A considerable proportion of statistical results focuses on spatial associations in loadings between modalities - it seems likely that these reflect intrinsic correlations between modalities, rather than associations specific to any latent component. Assessment of associations with functional gradients is similarly a little hard to interpret. Thus, it is hard to judge the implications for our understanding of the neurophysiological basis of psychopathology and the ability of MRI to provide clinical tools for, say, stratification. The observation of a recapitulation of psychopathology hierarchy may be somewhat undermined by the relatively modest strength of the components in the imaging domain. The task fMRI was assessed with a fairly basic functional connectivity approach, not using task timings to more specifically extract network responses.

Overall, the authors achieve their aim to provide a detailed multimodal characterisation of MRI correlations of psychopathology. Code and data are available and well organised and should provide a valuable resource for researchers wanting to understand MRI-based neural correlates of psycho-pathology-related behavioural traits in this important age group. It is largely a descriptive study, with comparisons to previous uni-modal work, but without particularly strong testing of neuroscience hypotheses.

Welcome back and welcome to this CloudTrail demo where we're going to set up an organizational trail and configure it to log data for all accounts in our organization to S3 and CloudWatch logs.

The first step is that you'll need to be logged into the IAM admin user of the management account of the organization. As a reminder, this is the general account. To set up an organizational trail, you always need to be logged into the management account. To set up individual trails, you can do that locally inside each of your accounts, but it's always more efficient to use an organizational trail.

Now, before we start the demonstration, I want to talk briefly about CloudTrail pricing. I'll make sure this link is in the lesson description, but essentially there is a fairly simple pricing structure to CloudTrail that you need to be aware of.

The 90-day history that's enabled by default in every AWS account is free. You don't get charged for that; it comes free by default with every AWS account. Next, you have the ability to get one copy of management events free in every region in each AWS account. This means creating one trail that's configured for management events in each region in each AWS account, and that comes for free. If you create any additional trails, so you get any additional copies of management events, they are charged at two dollars per 100,000 events. That won't apply to us in this demonstration, but you need to be aware of that if you're using this in production.

Logging data events comes at a charge regardless of the number, so we're not going to enable data events for this demo lesson. But if you do enable it, then that comes at a charge of 10 cents per 100,000 events, irrespective of how many trails you have. This charge applies from the first time you're logging any data events.

What we'll be doing in this demo lesson is setting up an organizational trail which will create a trail in every region in every account inside the organization. But because we get one for free in every region in every account, we won't incur any charges for the CloudTrail side of things. We will be charged for any S3 storage that we use. However, S3 also comes with a free tier allocation for storage, which I don't expect us to breach.

With that being said, let's get started and implement this solution. To do that, we need to be logged in to the console UI again in the management account of the organization. Then we need to move to the CloudTrail console. If you've been here recently, it will be in the Recently Visited Services. If not, just type CloudTrail in the Find Services box and then open the CloudTrail console.

Once you're at the console, you might see a screen like this. If you do, then you can just click on the hamburger menu on the left and then go ahead and click on trails. Now, depending on when you're doing this demo, if you see any warnings about a new or old console version, make sure that you select the new version so your console looks like what's on screen now.

Once you're here, we need to create a trail, so go ahead and click on create trail. To create a trail, you're going to be asked for a few important pieces of information, the first of which is the trail name. For trail name, we're going to use "animals4life.org," so just go ahead and enter that. By default, with this new UI version, when you create a trail, it's going to create it in all AWS regions in your account. If you're logged into the management account of the organization, as we are, you also have the ability to enable it for all regions in all accounts of your organization. We're going to do that because this allows us to have one single logging location for all CloudTrail logs in all regions in all of our accounts, so go ahead and check this box.

By default, CloudTrail stores all of its logs in an S3 bucket. When you're creating a trail, you have the ability to either create a new S3 bucket to use or you can use an existing bucket. We're going to go ahead and create a brand new bucket for this trail. Bucket names within S3 need to be globally unique, so it needs to be a unique name across all regions and across all AWS accounts. We're going to call this bucket starting with "CloudTrail," then a hyphen, then "animals-for-life," another hyphen, and then you'll need to put a random number. You’ll need to pick something different from me and different from every other student doing this demo. If you get an error about the bucket name being in use, you just need to change this random number.

You're also able to specify if you want the log files stored in the S3 bucket to be encrypted. This is done using SSE-KMS encryption. This is something that we'll be covering elsewhere in the course, and for production usage, you would definitely want to use it. For this demonstration, to keep things simple, we're not going to encrypt the log files, so go ahead and untick this box.

Under additional options, you're able to select log file validation, which adds an extra layer of security. This means that if any of the log files are tampered with, you have the ability to determine that. This is a really useful feature if you're performing any account-level audits. In most production situations, I do enable this, but you can also elect to have an SNS notification delivery. So, every time log files are delivered into this S3 bucket, you can have a notification. This is useful for production usage or if you need to integrate this with any non-AWS systems, but for this demonstration, we'll leave this one unchecked.

You also have the ability, as well as storing these log files into S3, to store them in CloudWatch logs. This gives you extra functionality because it allows you to perform searches, look at the logs from a historical context inside the CloudWatch logs user interface, as well as define event-driven processes. You can configure CloudWatch logs to scan these CloudTrail logs and, in the event that any particular piece of text occurs in the logs (e.g., any API call, any actions by a user), you can generate an event that can invoke, for example, a Lambda function or spawn some other event-driven processing. Don't worry if you don't understand exactly what this means at this point; I'll be talking about all of this functionality in detail elsewhere in the course. For this demonstration, we are going to enable CloudTrail to put these logs into CloudWatch logs as well, so check this box. You can choose a log group name within CloudWatch logs for these CloudTrail logs. If you want to customize this, you can, but we're going to leave it as the default.

As with everything inside AWS, if a service is acting on our behalf, we need to give it the permissions to interact with other AWS services, and CloudTrail is no exception. We need to give CloudTrail the ability to interact with CloudWatch logs, and we do that using an IAM role. Don’t worry, we’ll be talking about IAM roles in detail elsewhere in the course. For this demonstration, just go ahead and select "new" because we're going to create a new IAM role that will give CloudTrail the ability to enter data into CloudWatch logs.

Now we need to provide a role name, so go ahead and enter "CloudTrail_role_for_CloudWatch_logs" and then an underscore and then "animals_for_life." The name doesn’t really matter, but in production settings, you'll want to make sure that you're able to determine what these roles are for, so we’ll use a standard naming format. If you expand the policy document, you'll be able to see the exact policy document or IAM policy document that will be used to give this role the permissions to interact with CloudWatch logs. Don’t worry if you don’t fully understand policy documents at this point; we’ll be using them throughout the course, and over time you'll become much more comfortable with exactly how they're used. At a high level, this policy document will be attached to this role, and this is what will give CloudTrail the ability to interact with CloudWatch logs.

At this point, just scroll down; that's everything that we need to do, so go ahead and click on "next." Now, you'll need to select what type of events you want this trail to log. You’ve got three different choices. The default is to log only management events, so this logs any events against the account or AWS resources (e.g., starting or stopping an EC2 instance, creating or deleting an EBS volume). You've also got data events, which give you the ability to log any actions against things inside resources. Currently, CloudTrail supports a wide range of services for data event logging. For this demonstration, we won't be setting this up with data events initially because I’ll be covering this elsewhere in the course. So, go back to the top and uncheck data events.

You also have the ability to log insight events, which can identify any unusual activity, errors, or user behavior on your account. This is especially useful from a security perspective. For this demonstration, we won’t be logging any insight events; we’re just going to log management events. For management events, you can further filter down to read or write or both and optionally exclude KMS or RDS data API events. For this demo lesson, we’re just going to leave it as default, so make sure that read and write are checked. Once you've done that, go ahead and click on "next." On this screen, just review everything. If it all looks good, click on "create trail."

Now, if you get an error saying the S3 bucket already exists, you'll just need to choose a new bucket name. Click on "edit" at the top, change the bucket name to something that's globally unique, and then follow that process through again and create the trail.

Certainly! Here is the continuation and completion of the transcript:

After a few moments, the trail will be created. It should say "US East Northern Virginia" as the home region. Even though you didn't get the option to select it because it's selected by default, it is a multi-region trail. Finally, it is an organizational trail, which means that this trail is now logging any CloudTrail events from all regions in all accounts in this AWS organization.

Now, this isn't real-time, and when you first enable it, it can take some time for anything to start to appear in either S3 or CloudWatch logs. At this stage, I recommend that you pause the video and wait for 10 to 15 minutes before continuing, because the initial delivery of that first set of log files through to S3 can take some time. So pause the video, wait 10 to 15 minutes, and then you can resume.

Next, right-click the link under the S3 bucket and open that in a new tab. Go to that tab, and you should start to see a folder structure being created inside the S3 bucket. Let's move down through this folder structure, starting with CloudTrail. Go to US East 1 and continue down through this folder structure.

In my case, I have quite a few of these log files that have been delivered already. I'm going to pick one of them, the most recent, and just click on Open. Depending on the browser that you're using, you might have to download and then uncompress this file. Because I'm using Firefox, it can natively open the GZ compressed file and then automatically open the JSON log file inside it.

So this is an example of a CloudTrail event. We're able to see the user identity that actually generates this event. In this case, it's me, I am admin. We can see the account ID that this event is for. We can see the event source, the event name, the region, the source IP address, the user agent (in this case, the console), and all of the relevant information for this particular interaction with the AWS APIs are logged inside this CloudTrail event.

Don’t worry if this doesn’t make a lot of sense at this point. You’ll get plenty of opportunities to interact with this type of logging event as you go through the various theory and practical lessons within the course. For now, I just want to highlight exactly what to expect with CloudTrail logs.

Since we’ve enabled all of this logging information to also go into CloudWatch logs, we can take a look at that as well. So back at the CloudTrail console, if we click on Services and then type CloudWatch, wait for it to pop up, locate Logs underneath CloudWatch, and then open that in a new tab.

Inside CloudWatch, on the left-hand menu, look for Logs, and then Log Groups, and open that. You might need to give this a short while to populate, but once it does, you should see a log group for the CloudTrail that you’ve just created. Go ahead and open that log group.

Inside it, you’ll see a number of log streams. These log streams will start with your unique organizational code, which will be different for you. Then there will be the account number of the account that it represents. Again, these will be different for you. And then there’ll be the region name. Because I’m only interacting with the Northern Virginia region, currently, the only ones that I see are for US East 1.

In this particular account that I’m in, the general account of the organization, if I look at the ARN (Amazon Resource Name) at the top or after US East 1 here, this number is my account number. This is the account number of my general account. So if I look at the log streams, you’ll be able to see that this account (the general account) matches this particular log stream. You’ll be able to do the same thing in your account. If you look for this account ID and then match it with one of the log streams, you'll be able to pull the logs for the general AWS account.

If I go inside this particular log stream, as CloudTrail logs any activity in this account, all of that information will be populated into CloudWatch logs. And that’s what I can see here. If I expand one of these log entries, we’ll see the same formatted CloudTrail event that I just showed you in my text editor. So the only difference when using CloudWatch logs is that the CloudTrail events also get entered into a log stream in a log group within CloudWatch logs. The format looks very similar.

Returning to the CloudTrail console, one last thing I want to highlight: if you expand the menu on the left, whether you enable a particular trail or not, you’ve always got access to the event history. The event history stores a log of all CloudTrail events for the last 90 days for this particular account, even if you don’t have a specific trail enabled. This is standard functionality. What a trail allows you to do is customize exactly what happens to that data. This area of the console, the event history, is always useful if you want to search for a particular event, maybe check who’s logged onto the account recently, or look at exactly what the IAM admin user has been doing within this particular AWS account.

The reason why we created a trail is to persistently store that data in S3 as well as put it into CloudWatch logs, which gives us that extra functionality. With that being said, that’s everything I wanted to cover in this demo lesson.

One thing you need to be aware of is that S3, as a service, provides a certain amount of resource under the free tier available in every new AWS account, so you can store a certain amount of data in S3 free of charge. The problem with CloudTrail, and especially organizational trails, is that they generate quite a large number of requests. There is also, in addition to space, a number of requests per month that are part of the free tier.

If you leave this CloudTrail enabled for the duration of your studies, for the entire month, it is possible that this will go slightly over the free tier allocation for requests within the S3 service. You might see warnings that you’re approaching a billable threshold, and you might even get a couple of cents of bill per month if you leave this enabled all the time. To avoid that, if you just go to Trails, open up the trail that you’ve created, and then click on Stop Logging. You’ll need to confirm that by clicking on Stop Logging, and at that point, no logging will occur into the S3 bucket or into CloudWatch logs, and you won’t experience those charges.

For any production usage, the low cost of this service means that you would normally leave it enabled in all situations. But to keep costs within the free tier for this course, you can, if required, just go ahead and stop the logging. If you don’t mind a few cents per month of S3 charges for CloudTrail, then by all means, go ahead and leave it enabled.

With that being said, that’s everything I wanted to cover in this demo lesson. So go ahead, complete the lesson, and when you're ready, I look forward to you joining me in the next.

Welcome to this lesson, where I'm going to introduce the theory and architecture of CloudWatch Logs.

I've already covered the metrics side of CloudWatch earlier in the course, and I'm covering the logs part now because you'll be using it when we cover CloudTrail. In the CloudTrail demo, we'll be setting up CloudTrail and using CloudWatch Logs as a destination for those logs. So, you'll need to understand it, and we'll be covering the architecture in this lesson. Let's jump in and get started.

CloudWatch Logs is a public service. The endpoint to which applications connect is hosted in the AWS public zone. This means you can use the product within AWS VPCs, from on-premises environments, and even other cloud platforms, assuming that you have network connectivity as well as AWS permissions.

The CloudWatch Logs product allows you to store, monitor, and access logging data. Logging data, at a very basic level, consists of a piece of information, data, and a timestamp. The timestamp generally includes the year, month, day, hour, minute, second, and timezone. There can be more fields, but at a minimum, it's generally a timestamp and some data.

CloudWatch Logs has built-in integrations with many AWS services, including EC2, VPC Flow Logs, Lambda, CloudTrail, Route 53, and many more. Any services that integrate with CloudWatch Logs can store data directly inside the product. Security for this is generally provided by using IAM roles or service roles.

For anything outside AWS, such as logging custom application or OS logs on EC2, you can use the unified CloudWatch agent. I’ve mentioned this before and will be demoing it later in the EC2 section of the course. This is how anything outside of AWS products and services can log data into CloudWatch Logs. So, it’s either AWS service integrations or the unified CloudWatch agent. There is a third way, using development kits for AWS to implement logging into CloudWatch Logs directly into your application, but that tends to be covered in developer and DevOps AWS courses. For now, just remember either AWS service integrations or the unified CloudWatch agent.

CloudWatch Logs are also capable of taking logging data and generating a metric from it, known as a metric filter. Imagine a situation where you have a Linux instance, and one of the operating system log files logs any failed connection attempts via SSH. If this logging information was injected into CloudWatch Logs, a metric filter can scan those logs constantly. Anytime it sees a mention of the failed SSH connection, it can increment a metric within CloudWatch. You can then have alarms based on that metric, and I’ll be demoing that very thing later in the course.

Let’s look at the architecture visually because I'll be showing you how this works in practice in the CloudTrail demo, which will be coming up later in the section. Architecturally, CloudWatch Logs looks like this: It’s a regional service. So, for this example, let’s assume we’re talking about us-east-1.

The starting point is our logging sources, which can include AWS products and services, mobile or server-based applications, external compute services (virtual or physical servers), databases, or even external APIs. These sources inject data into CloudWatch Logs as log events.

Log events consist of a timestamp and a message block. CloudWatch Logs treats this message as a raw block of data. It can be anything you want, but there are ways the data can be interpreted, with fields and columns defined. Log events are stored inside log streams, which are essentially a sequence of log events from the same source.

For example, if you had a log file stored on multiple EC2 instances that you wanted to inject into CloudWatch Logs, each log stream would represent the log file for one instance. So, you’d have one log stream for instance one and one log stream for instance two. Each log stream is an ordered set of log events for a specific source.

We also have log groups, which are containers for multiple log streams of the same type of logging. Continuing the example, we would have one log group containing everything for that log file. Inside this log group would be different log streams, each representing one source. Each log stream is a collection of log events. Every time an item was added to the log file on a single EC2 instance, there would be one log event inside one log stream for that instance.

A log group also stores configuration settings, such as retention settings and permissions. When we define these settings on a log group, they apply to all log streams within that log group. It’s also where metric filters are defined. These filters constantly review any log events for any log streams in that log group, looking for certain patterns, such as an application error code or a failed SSH login. When detected, these metric filters increment a metric, and metrics can have associated alarms. These alarms can notify administrators or integrate with AWS or external systems to take action.

CloudWatch Logs is a powerful product. This is the high-level architecture, but don’t worry—you’ll get plenty of exposure to it throughout the course because many AWS products integrate with CloudWatch Logs and use it to store their logging data. We’ll be coming back to this product time and again as we progress through the course. CloudTrail uses CloudWatch Logs, Lambda uses CloudWatch Logs, and VPC Flow Logs use CloudWatch Logs. There are many examples of AWS products where we’ll be integrating them with CloudWatch Logs.

I just wanted to introduce it at this early stage of the course. That’s everything I wanted to cover in this theory lesson. Thanks for watching. Go ahead, complete this video, and when you’re ready, join me in the next.

Welcome back, and in this demo lesson, I want to give you some experience working with Service Control Policies (SCPs).

At this point, you've created the AWS account structure which you'll be using for the remainder of the course. You've set up an AWS organization, with the general account that created it becoming the management account. Additionally, you've invited the production AWS account into the organization and created the development account within it.

In this demo lesson, I want to show you how you can use SCPs to restrict what identities within an AWS account can do. This is a feature of AWS Organizations.

Before we dive in, let's tidy up the AWS organization. Make sure you're logged into the general account, the management account of the organization, and then navigate to the organization's console. You can either type that into the 'Find Services' box or select it from 'Recently Used Services.'

As discussed in previous lessons, AWS Organizations allows you to organize accounts with a hierarchical structure. Currently, there's only the root container of the organization. To create a hierarchical structure, we need to add some organizational units. We will create a development organizational unit and a production organizational unit.

Select the root container at the top of the organizational structure. Click on "Actions" and then "Create New." For the production organizational unit, name it 'prod.' Scroll down and click on "Create Organizational Unit." Next, do the same for the development unit: select 'Route,' click on "Actions," and then "Create New." Under 'Name,' type 'dev,' scroll down, and click on "Create Organizational Unit."

Now, we need to move our AWS accounts into these relevant organizational units. Currently, the Development, Production, and General accounts are all contained in the root container, which is the topmost point of our hierarchical structure.

To move the accounts, select the Production AWS account, click on "Actions," and then "Move." In the dialogue that appears, select the Production Organizational Unit and click "Move." Repeat this process for the Development AWS account: select the Development AWS account, click "Actions," then "Move," and select the 'dev' OU before clicking "Move."

Now, we've successfully moved the two AWS accounts into their respective organizational units. If you select each organizational unit in turn, you can see that 'prod' contains the production AWS account, and 'dev' contains the development AWS account. This simple hierarchical structure is now in place.

To prepare for the demo part of this lesson where we look at SCPs, move back to the AWS console. Click on AWS, then the account dropdown, and switch roles into the production AWS account by selecting 'Prod' from 'Role History.'

Once you're in the production account, create an S3 bucket. Type S3 into the 'Find Services' box or find it in 'Recently Used Services' and navigate to the S3 console. Click on "Create Bucket." For the bucket name, call it 'CatPics' followed by a random number—S3 bucket names must be globally unique. I’ll use 1, lots of 3s, and then 7. Ensure you select the US East 1 region for the bucket. Scroll down and click "Create Bucket."

After creating the bucket, go inside it and upload some files. Click on "Add Files," then download the cat picture linked to this lesson to your local machine. Upload this cat picture to the S3 bucket by selecting it and clicking "Open," then "Upload" to complete the process.

Once the upload finishes, you can view the picture of Samson. Click on it to see Samson looking pretty sleepy. This demonstrates that you can currently access the Samson.jpg object while operating within the production AWS account.

The key point here is that you’ve assumed an IAM role. By switching roles into the production account, you’ve assumed the role called "organization account access role," which has the administrator access managed policy attached.

Now, we’ll demonstrate how this can be restricted using SCPs. Move back to the main AWS console. Click on the account dropdown and switch back to the general AWS account. Navigate to AWS Organizations, then Policies. Currently, most options are disabled, including Service Control Policies, Tag Policies, AI Services, Opt-out Policies, and Backup Policies.

Click on Service Control Policies and then "Enable" to activate this functionality. This action adds the "Full AWS Access" policy to the entire organization, which imposes no restrictions, so all AWS accounts maintain full access to all AWS services.

To create our own service control policy, download the file named DenyS3.json linked to this lesson and open it in a code editor. This SCP contains two statements. The first statement is an allow statement with an effect of allow, action as star (wildcard), and resource as star (wildcard). This replicates the full AWS access SCP applied by default. The second statement is a deny statement that denies any S3 actions on any AWS resource. This explicit deny overrides the explicit allow for S3 actions, resulting in access to all AWS services except S3.

Copy the content of the DenyS3.json file into your clipboard. Move back to the AWS console, go to the policy section, and select Service Control Policies. Click "Create Policy," delete the existing JSON in the policy box, and paste the copied content. Name this policy "Allow all except S3" and create it.

Now, go to AWS Accounts on the left menu, select the prod OU, and click on the Policies tab. Attach the new policy "Allow all except S3" by clicking "Attach" in the applied policies box. We will also detach the full AWS access policy directly attached. Check the box next to full AWS access, click "Detach," and confirm by clicking "Detach Policy."

Now, the only service control policy directly attached to production is "Allow all except S3," which allows access to all AWS products and services except S3.

To verify, go back to the main AWS console and switch roles into the production AWS account. Go to the S3 console and you should receive a permissions error, indicating that you don't have access to list buckets. This is because the SCP attached to the production account explicitly denies S3 access. Access to other services remains unaffected, so you can still interact with EC2.

If we switch back to the general account, reattach the full AWS access policy, and detach "Allow all except S3," the production account will regain access to S3. By following the same process, you’ll be able to access the S3 bucket and view the object once again.

This illustrates how SCPs can be used to restrict access for identities within an AWS account, in this case, the production AWS account.

To clean up, delete the bucket. Select the catpics bucket, click "Empty," type "permanently delete," and select "Empty." Once that's done, you can delete the bucket by selecting it, clicking "Delete," confirming the bucket name, and then clicking "Delete Bucket."

You’ve now demonstrated full control over S3, evidenced by successfully deleting the bucket. This concludes the demo lesson. You’ve created and applied an SCP that restricts S3 access, observed its effects, and cleaned up. We’ll discuss more about boundaries and restrictions in future lessons. For now, complete this video, and I'll look forward to seeing you in the next lesson.

Welcome back.

In this lesson, I want to continue immediately from the last one by discussing when and where you might use IAM roles. By talking through some good scenarios for using roles, I want to make sure that you're comfortable with selecting these types of situations where you would choose to use an IAM role and where you wouldn't, because that's essential for real-world AWS usage and for answering exam questions correctly.

So let's get started.

One of the most common uses of roles within the same AWS account is for AWS services themselves. AWS services operate on your behalf and need access rights to perform certain actions. An example of this is AWS Lambda. Now, I know I haven't covered Lambda yet, but it's a function as a service product. What this means is that you give Lambda some code and create a Lambda function. This function, when it runs, might do things like start and stop EC2 instances, perform backups, or run real-time data processing. What it does exactly isn't all that relevant for this lesson. The key thing, though, is that a Lambda function, as with most AWS things, has no permissions by default. A Lambda function is not an AWS identity. It's a component of a service, and so it needs some way of getting permissions to do things when it runs. Running a Lambda function is known as a function invocation or a function execution using Lambda terminology.

So anything that's not an AWS identity, this might be an application or a script running on a piece of compute hardware somewhere, needs to be given permissions on AWS using access keys. Rather than hard-coding some access keys into your Lambda function, there's actually a better way. To provide these permissions, we can create an IAM role known as a Lambda execution role. This execution role has a trust policy which trusts the Lambda service. This means that Lambda is allowed to assume that role whenever a function is executed. This role has a permissions policy which grants access to AWS products and services.

When the function runs, it uses the sts:AssumeRole operation, and then the Secure Token Service generates temporary security credentials. These temporary credentials are used by the runtime environment in which the Lambda function runs to access AWS resources based on the permissions the role’s permissions policy has. The code is running in a runtime environment, and it's the runtime environment that assumes the role. The runtime environment gets these temporary security credentials, and then the whole environment, which the code is running inside, can use these credentials to access AWS resources.

So why would you use a role for this? What makes this scenario perfect for using a role? Well, if we didn't use a role, you would need to hard-code permissions into the Lambda function by explicitly providing access keys for that function to use. Where possible, you should avoid doing that because, A, it's a security risk, and B, it causes problems if you ever need to change or rotate those access keys. It's always better for AWS products and services, where possible, to use a role, because when a role is assumed, it provides a temporary set of credentials with enough time to complete a task, and then these are discarded.

For a given Lambda function, you might have one copy running at once, zero copies, 50 copies, a hundred copies, or even more. Because you can't determine this number, because it's unknown, if you remember my rule from the previous lesson, if you don't know the number of principals, if it's multiple or if it's an uncertain number, then it suggests a role might be the most ideal identity to use. In this case, it is the ideal way of providing Lambda with these credentials to use a role and allow it to get these temporary credentials. It's always the preferred option when using AWS services to do something on your behalf; use a role because you don't need to provide any static credentials.

Okay, so let's move on to the next scenario.

Another situation where roles are useful is emergency or out-of-the-usual situations. Here’s a familiar scenario that you might find in a workplace. This is Wayne, and Wayne works in a business's service desk team. This team is given read-only access to a customer's AWS account so that they can keep an eye on performance. The idea is that anything more risky than this read-only level of access is handled by a more senior technical team. We don't want to give Wayne's team long-term permissions to do anything more destructive than this read-only access, but there are always going to be situations which occur when we least want them, normally 3:00 a.m. on a Sunday morning, when a customer might call with an urgent issue where they need Wayne's help to maybe stop or start an instance, or maybe even terminate an EC2 instance and recreate it.

So 99% of the time, Wayne and his team are happy with this read-only access, but there are situations when he needs more. This is a break-glass style situation, which is named after this. The idea of break glass in the physical world is that there is a key for something behind glass. It might be a key for a room that a certain team doesn't normally have access to, maybe it’s a safe or a filing cabinet. Whatever it is, the glass provides a barrier, meaning that when people break it, they really mean to break it. It’s a confirmation step. So if you break a piece of glass to get a key to do something, there needs to be an intention behind it. Anyone can break the glass and retrieve the key, but having the glass results in the action only happening when it's really needed. At other times, whatever the key is for remains locked. And you can also tell when it’s been used and when it hasn’t.

A role can perform the same thing inside an AWS account. Wayne can assume an emergency role when absolutely required. When he does, he'll gain additional permissions based on the role's permissions policy. For a short time, Wayne will, in effect, become the role. This access will be logged and Wayne will know to only use the role under exceptional circumstances. Wayne’s normal permissions can remain at read-only, which protects him and the customer, but he can obtain more if required when it’s really needed. So that’s another situation where a role might be a great solution.

Another scenario when roles come in handy is when you're adding AWS into an existing corporate environment. You might have an existing physical network and an existing provider of identities, known as an identity provider, that your staff use to log into various systems. For the sake of this example, let’s just say that it's Microsoft Active Directory. In this scenario, you might want to offer your staff single sign-on, known as SSO, allowing them to use their existing logins to access AWS. Or you might have upwards of 5,000 accounts. Remember, there’s the 5,000 IAM user limit. So for a corporation with more than 5,000 staff, you can’t offer each of them an IAM user. That is beyond the capabilities of IAM.

Roles are often used when you want to reuse your existing identities for use within AWS. Why? Because external accounts can’t be used directly. You can’t access an S3 bucket directly using an Active Directory account. Remember this fact. External accounts or external identities cannot be used directly to access AWS resources. You can’t directly use Facebook, Twitter, or Google identities to interact with AWS. There is a separate process which allows you to use these external identities, which I’ll be talking about later in the course.

Architecturally, what happens is you allow an IAM role inside your AWS account to be assumed by one of the external identities, which is in Active Directory in this case. When the role is assumed, temporary credentials are generated and these are used to access the resources. There are ways that this is hidden behind the console UI so that it appears seamless, but that's what happens behind the scenes. I'll be covering this in much more detail later in the course when I talk about identity federation, but I wanted to introduce it here because it is one of the major use cases for IAM roles.

Now, why roles are so important when an existing ID provider such as Active Directory is involved is that, remember, there is this 5,000 IAM user limit in an account. So if your business has more than 5,000 accounts, then you can’t simply create an IAM user for each of those accounts, even if you wanted to. 5,000 is a hard limit. It can't be changed. Even if you could create more than 5,000 IAM users, would you actually want to manage 5,000 extra accounts? Using a role in this way, so giving permissions to an external identity provider and allowing external identities to assume this role, is called ID Federation. It means you have a small number of roles to manage and external identities can use these roles to access your AWS resources.

Another common situation where you might use roles is if you're designing the architecture for a popular mobile application. Maybe it's a ride-sharing application which has millions of users. The application needs to store and retrieve data from a database product in AWS, such as DynamoDB. Now, I've already explained two very important but related concepts on the previous screen. Firstly, that when you interact with AWS resources, you need to use an AWS identity. And then secondly, that there’s this 5,000 IAM user limit per account. So designing an application with this many users which needs access to AWS resources, if you could only use IAM users or identities in AWS, it would be a problem because of this 5,000 user limit. It’s a hard limit and it can’t be raised.

Now, this is a problem which can be fixed with a process called Web Identity Federation, which uses IAM roles. Most mobile applications that you’ve used, you might have noticed they allow you to sign in using a web identity. This might be Twitter, Facebook, Google, and potentially many others. If we utilize this architecture for our web application, we can trust these identities and allow these identities to assume an IAM role. This is based on that role’s trust policy. So they can assume that role, gain access to temporary security credentials, and use those credentials to access AWS resources, such as DynamoDB. This is a form of Web Identity Federation, and I'll be covering it in much more detail later in the course.

The use of roles in this situation has many advantages. First, there are no AWS credentials stored in the application, which makes it a much more preferred option from a security point of view. If an application is exploited for whatever reason, there’s no chance of credentials being leaked, and it uses an IAM role which you can directly control from your AWS account. Secondly, it makes use of existing accounts that your customers probably already have, so they don't need yet another account to access your service. And lastly, it can scale to hundreds of millions of users and beyond. It means you don’t need to worry about the 5,000 user IAM limit. This is really important for the exam. There are very often questions on how you can architect solutions which will work for mobile applications. Using ID Federation, so using IAM roles, is how you can accomplish that. And again, I'll be providing much more information on ID Federation later in the course.

Now, one scenario I want to cover before we finish up this lesson is cross-account access. In an upcoming lesson, I’ll be introducing AWS Organizations and you will get to see this type of usage in practice. It’s actually how we work in a multi-account environment. Picture the scenario that's on screen now: two AWS accounts, yours and a partner account. Let’s say your partner organization offers an application which processes scientific data and they want you to store any data inside an S3 bucket that’s in their account. Your account has thousands of identities, and the partner IT team doesn’t want to create IAM users in their account for all of your staff. In this situation, the best approach is to use a role in the partner account. Your users can assume that role, get temporary security credentials, and use those to upload objects. Because the IAM role in the partner account is an identity in that account, using that role means that any objects that you upload to that bucket are owned by the partner account. So it’s a very simple way of handling permissions when operating between accounts.

Roles can be used cross-account to give access to individual resources like S3 in the onscreen example, or you can use roles to give access to a whole account. You’ll see this in the upcoming AWS Organization demo lesson. In that lesson, we’re going to configure it so a role in all of the different AWS accounts that we’ll be using for this course can be assumed from the general account. It means you won’t need to log in to all of these different AWS accounts. It makes multi-account management really simple.

I hope by this point you start to get a feel for when roles are used. Even if you’re a little vague, you will learn more as you go through the course. For now, just a basic understanding is enough. Roles are difficult to understand at first, so you’re doing well if you’re anything but confused at this point. I promise you, as we go through the course and you get more experience, it will become second nature.

So at this point, that’s everything I wanted to cover. Thanks for watching. Go ahead and complete this video, and when you're ready, join me in the next lesson.

This is only if the user doesn't follow the hyperlinked button. I think it should be moved to the bottom of sign in or be in a troubleshooting flow only

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

Ever-improving techniques allow the detailed capture of brain morphology and function to the point where individual brain anatomy becomes an important factor. This study investigated detailed sulcal morphology in the parieto-occipital junction. Using cutting-edge methods, it provides important insights into local anatomy, individual variability, and local brain function. The presented work advances the field and will stimulate future research into this important area.

Strengths:

Detailed, very thorough methodology. Multiple raters mapped detailed sulci in a large cohort. The identified sulcal features and their functional and behavioural relevance are then studied using various complementary methods. The results provide compelling evidence for the importance of the described sulcal features and their proposed relationship to cortical brain function.

We thank the Reviewer for highlighting the strengths of our methods and findings.

Weaknesses:

A detailed description/depiction of the various sulcal patterns is missing.

We agree that adding these details for the newly described sulci is necessary and have now done so. These details are included in the Results (Page 6):

“Beyond characterizing the incidence of sulci, it is also common in the neuroanatomical literature to qualitatively characterize sulci on the basis of fractionation and intersection with surrounding sulci (termed “sulcal types”; for examples in other cortical expanses, see Chiavaras & Petrides, 2000; Drudik et al., 2023; Miller et al., 2021; Paus et al., 1996; Weiner et al., 2014; Willbrand, Parker, et al., 2022). All four sulci most commonly did not intersect with other sulci (see Supplementary Tables 1-4 for a summary of the sulcal types of the slocs and pAngs dorsal and ventral components). The sulcal types were also highly comparable between hemispheres (rs > .99 , ps < .001).”

And in four new Supplementary Tables.

A possible relationship between sulcal morphology and individual demographics might provide more insight into anatomical variability.

We have conducted additional analyses to relate sulcal incidence to demographic features (age and gender). These results are included on Pages 5-6:

“Given that sulcal incidence and patterning is also sometimes related to demographic features (Cachia et al., 2021; Leonard et al., 2009; Wei et al., 2017), subsequent GLMs relating the incidence and patterning of the three more variable sulci (slocs-d, pAngs-v, and pAngs-d) to demographic features (age and gender) revealed no associations for any sulcus (ps > .05).”

The unique dataset offers an opportunity to provide insights into laterality effects that should be explored.

We included hemisphere as a factor in all models for this exact reason. Throughout the paper, we have edited the text to ensure that these laterality effects are more apparent to readers.

Further, we have a Supplementary Results section on hemispheric effects regarding the slocs-v, cSTS3, and lTOS:

“Hemispheric asymmetries in morphological, architectural, and functional features with regards to the slocs-v, cSTS3, and lTOS comparison

We observed a sulcus x metric x hemisphere interaction on the morphological and architectural features of the slocs-v (F(4.20, 289.81) = 4.16, η2 = 0.01, p = .002; the cSTS3 is discussed in the next section). Post hoc tests showed that this interaction was driven by  the slocs-v being cortically thinner in the left than the right hemisphere (p < .001; Fig. 2a).

There was also a sulcus x network x hemisphere interaction on the functional connectivity profiles (using functional connectivity parcellations from (Kong et al., 2019)) of the slocs-v and lTOS (F(32, 2144) = 3.99, η2 = 0.06, p < .001; the cSTS3 is discussed in the next section). Post hoc tests showed that this interaction was driven by three effects: (i) the slocs-v overlapped more with the Default C subnetwork in the left than the right hemisphere (p = .013), (ii) the lTOS overlapped more with Visual A subnetwork in the right than the left hemisphere (p = .002), and (iii) the lTOS overlapped more with the Visual B subnetwork in the left than the right hemisphere (p = .002; Fig. 2b).”

As well as the other STS rami on morphology:

“It is also worth noting that there was a sulcus x metric x hemisphere interaction (F(4, 284.12) = 6.60, η2 = 0.08, p < .001). Post hoc tests showed that: (i) the cSTS3 was smaller (p < .001) and thinner (p = .025) in the left than the right hemisphere (Supplementary Fig. 8a), (ii) the cSTS2 was shallower (p = .004) and thicker (p < .001) in the right than left hemisphere (Supplementary Fig. 8a), and (iii) the cSTS1 was shallower (p < .001), smaller (p = .002), thinner (p = .001), and less myelinated (p < .001) in the left than the right hemisphere (Supplementary Fig. 8a).”

And functional connectivity of the STS rami:

“There was also a sulcus x network x hemisphere interaction (F(32, 2208) = 12.26, η2 = 0.15, p < .001). Post hoc tests showed differences for each cSTS component. Here, the cSTS1 overlapped more with the Auditory network (p < .001), less with the Control B subnetwork (p < .001), more with the Control C subnetwork (p < .001), less with the Default B subnetwork (p < .001), more with the Default C subnetwork (p < .001), more with the Ventral Attention B subnetwork (p < .001), and more with the Visual A subnetwork (p = .024) in the right than in the left hemisphere (Supplementary Fig. 8b). In addition, the cSTS2 overlapped more with the Control B subnetwork (p < .001), more with the Control C subnetwork (p < .001), less with the Default B subnetwork (p < .001), and less with the Temporal-Parietal network (p = .011) in the right than in the left hemisphere (Supplementary Fig. 8b). Finally, the cSTS3 overlapped more with the Control B subnetwork (p = .002), less with the Default B subnetwork (p = .014), more with the Default C subnetwork (p = .022), less with the Ventral Attention B subnetwork (p = .029) in the right than in the left hemisphere (Supplementary Fig. 8b).”

Reviewer #2 (Public Review):

Summary: After manually labeling 144 human adult hemispheres in the lateral parieto-occipital junction (LPOJ), the authors 1) propose a nomenclature for 4 previously unnamed highly variable sulci located between the temporal and parietal or occipital lobes, 2) focus on one of these newly named sulci, namely the ventral supralateral occipital sulcus (slocs-v) and compare it to neighboring sulci to demonstrate its specificity (in terms of depth, surface area, gray matter thickness, myelination, and connectivity), 3) relate the morphology of a subgroup of sulci from the region including the slocs-v to the performance in a spatial orientation task, demonstrating behavioral and morphological specificity. In addition to these results, the authors propose an extended reflection on the relationship between these newly named landmarks and previous anatomical studies, a reflection about the slocs-v related to functional and cytoarchitectonic parcellations as well as anatomic connectivity and an insight about potential anatomical mechanisms relating sulcation and behavior.

Strengths:

- To my knowledge, this is the first study addressing the variable tertiary sulci located between the superior temporal sulcus (STS) and intraparietal sulcus (IPS).

- This is a very comprehensive study addressing altogether anatomical, architectural, functional and cognitive aspects.

- The definition of highly variable yet highly reproducible sulci such as the slocs-v feeds the community with new anatomo-functional landmarks (which is emphasized by the provision of a probability map in supp. mat., which in my opinion should be proposed in the main body).

- The comparison of different features between the slocs-v and similar sulci is useful to demonstrate their difference.

- The detailed comparison of the present study with state of the art contextualizes and strengthens the novel findings.

- The functional study complements the anatomical description and points towards cognitive specificity related to a subset of sulci from the LPOJ

- The discussion offers a proposition of theoretical interpretation of the findings

- The data and code are mostly available online (raw data made available upon request).

We thank the Reviewer for highlighting the strengths of our methods, analyses, and applications of our findings.

Weaknesses:

- While three independent raters labeled all hemispheres, one single expert finalized the decision. Because no information is reported on the inter-rater variability, this somehow equates to a single expert labeling the whole cohort, which could result in biased labellings and therefore affect the reproducibility of the new labels.

Our group does not use an approach amenable to calculating inter-rater agreements to expedite the process of defining thousands of sulci at the individual level in multiple regions. Our method consists of a two-tiered procedure. Here, authors YT and TG defined sulci which were then checked by a trained expert (EHW). These were then checked again by senior author  (KSW) . We emphasize that this process has produced reproducible anatomical results in other regions such as posteromedial cortex (Willbrand et al., 2023 Science Advances; Willbrand et al., 2023 Communications Biology; Maboudian et al., 2024 The Journal of Neuroscience), ventral temporal cortex (Weiner et al., 2014 NeuroImage; Miller et al., 2020 Scientific Reports; Parker et al., 2023 Brain Structure and Function), and lateral prefrontal cortex (Miller et al., 2021 The Journal of Neuroscience; Voorhies et al., 2021 Nature Communications; Yao et al., 2022 Cerebral Cortex; Willbrand et al., 2022 Brain Structure and Function; Willbrand et al., 2023 The Journal of Neuroscience) across age groups, species, and clinical populations. Further, in the Supplemental Materials we provide post mortem images showing that these sulci exist outside of cortical reconstructions, supporting this updated sulcal schematic of the lateral parieto-occipital junction. For the present study, by the time the final tier of our method was reached, we emphasize that a very small percentage (~2%) of sulcal definitions were actually modified. We will include an exact percentage in future publications in LPC/LOPJ.

- 3 out of the 4 newly labeled sulci are only described in the very first part and never reused. This should be emphasized as it is far from obvious at first glance of the article.

We have edited the Abstract (shown below, on Page 1) and paper throughout to emphasize the emphasis on the slocs-v over the other three sulci.

“After defining thousands of sulci in a young adult cohort, we revised the previous LPC/LPOJ sulcal landscape to include four previously overlooked, small, shallow, and variable sulci. One of these sulci (ventral supralateral occipital sulcus, slocs-v) is present in nearly every hemisphere and is morphologically, architecturally, and functionally dissociable from neighboring sulci. A data-driven, model-based approach, relating sulcal depth to behavior further revealed that the morphology of only a subset of LPC/LPOJ sulci, including the slocs-v, is related to performance on a spatial orientation task.”

It is worth noting that we have added additional analyses that include the other three newly-characterized sulci in response to Reviewer 1. We first described the relationship between these sulci and demographic features, alongside analyses on the patterning of these sulci, which are included in the Results (Page 6):

“Beyond characterizing the incidence of sulci, it is also common in the neuroanatomical literature to qualitatively characterize sulci on the basis of fractionation and intersection with surrounding sulci (termed “sulcal types”; for examples in other cortical expanses, see Chiavaras & Petrides, 2000; Drudik et al., 2023; Miller et al., 2021; Paus et al., 1996; Weiner et al., 2014; Willbrand, Parker, et al., 2022). All four sulci most commonly did not intersect with other sulci (see Supplementary Tables 1-4 for a summary of the sulcal types of the slocs and pAngs dorsal and ventral components). The sulcal types were also highly comparable between hemispheres (rs > .99 , ps < .001). Though we characterize these sulci in this paper for the first time, the location of these four sulci is consistent with the presence of variable “accessory sulci” in this cortical expanse mentioned in prior modern and classic studies (Supplementary Methods). We could also identify these sulci in post-mortem hemispheres (Supplementary Figs. 2, 3), ensuring that these sulci were not an artifact of the cortical reconstruction process.

Given that sulcal incidence and patterning is also sometimes related to demographic features (Cachia et al., 2021; Leonard et al., 2009; Wei et al., 2017), subsequent GLMs relating the incidence and patterning of the three more variable sulci (slocs-d, pAngs-v, and pAngs-d) to demographic features (age and gender) revealed no associations for any sulcus (ps > .05).  Finally, to help guide future research on these newly- and previously-classified LPC/LPOJ sulci, we generated probabilistic maps of each of these 17 sulci and share them with the field with the publication of this paper (Supplementary Fig. 6; Data availability).”

- The tone of the article suggests a discovery of these 4 sulci when some of them have already been reported (as rightfully highlighted in the article), though not named nor studied specifically. This is slightly misleading as I interpret the first part of the article as a proposition of nomenclature rather than a discovery of sulci.

We have toned down our language throughout the paper, emphasizing that this paper is updating the sulcal landscape of LPC/LOPJ taking into account these sulci that have not been comprehensively described previously. For example, in the Abstract (Page 1), we now write:

“After defining thousands of sulci in a young adult cohort, we revised the previous LPC/LPOJ sulcal landscape to include four previously overlooked, small, shallow, and variable sulci. One of these sulci (ventral supralateral occipital sulcus, slocs-v) is present in nearly every hemisphere and is morphologically, architecturally, and functionally dissociable from neighboring sulci. A data-driven, model-based approach, relating sulcal depth to behavior further revealed that the morphology of only a subset of LPC/LPOJ sulci, including the slocs-v, is related to performance on a spatial orientation task. “

- The article never mentions the concept of merging of sulcal elements and the potential effect it could have on the labeling of the newly named variable sulci.

We emphasize that we use multiple surfaces (pial, inflated, smoothwm) to help distinguish intersecting sulci from one another. We include extra text in the Methods (Page 21):

“We defined LPC/LPOJ sulci for each participant based on the most recent schematics of sulcal patterning by Petrides (2019) as well as pial, inflated, and smoothed white matter (smoothwm) FreeSurfer cortical surface reconstructions of each individual. In some cases, the precise start or end point of a sulcus can be difficult to determine on a surface (Borne et al., 2020); however, examining consensus across multiple surfaces allowed us to clearly determine each sulcal boundary in each individual. “

Further, upon quantifying the patterning of these variable sulci, a majority of the time they are independent (described in the Results on Page 6):

“Beyond characterizing the incidence of sulci, it is also common in the neuroanatomical literature to qualitatively characterize sulci on the basis of fractionation and intersection with surrounding sulci (termed “sulcal types”; for examples in other cortical expanses, see (Chiavaras & Petrides, 2000; Drudik et al., 2023; Miller et al., 2021; Paus et al., 1996; Weiner et al., 2014; Willbrand, Parker, et al., 2022). All four sulci most commonly did not intersect with other sulci (see Supplementary Tables 1-4 for a summary of the sulcal types of the slocs and pAngs dorsal and ventral components). The sulcal types were also highly comparable between hemispheres (rs > .99 , ps < .001).”

Thus, merging sulcal elements likely had a minimal impact on the present definitions.

- The definition of the new sulci is solely based on their localization relative to other sulci which are themselves variable (e.g. the 3rd branch of the STS can show different locations and different orientation, potentially affecting the definition of the slocs-v). This is not addressed in the discussion.

As displayed in our probabilistic maps of these sulci (Supplementary Fig. 6), the cSTS components (2-4) are actually relatively consistent between individuals, and thus, future investigators can utilize these maps to help define these sulci in new hemispheres.

Nevertheless, there is, of course, individual variability in the location of these sulci, and we do agree that this point brought up by the Reviewer is important. We have now added text to the Limitations section of the Discussion (Pages 15-16):

“The main limitation of our study is that presently, the most accurate methodology to define sulci —especially the small, shallow, and variable PTS—requires researchers to manually trace each structure on the cortical surface reconstructions. This method is limited due to the individual variability of cortical sulcal patterning (Fig. 1, Supplementary Fig. 5), which makes it challenging to identify sulci, let alone PTS, without extensive experience and practice. However, we anticipate that our probabilistic maps  will provide a starting point and hopefully, expedite the identification of these sulci in new participants. This method is also arduous and time-consuming—which, on the one hand, limits the sample size in terms of number of participants, while on the other, results in thousands of precisely defined sulci. This push-pull  relationship reflects a broader conversation in the human brain mapping and cognitive neuroscience fields between a balance of large N studies and “precision imaging” studies in individual participants (Allen et al., 2022; Gratton et al., 2022; Naselaris et al., 2021; Rosenberg and Finn, 2022).”

- The new sulci are only defined in terms of localization relative to other sulci, and no other property is described (general length, depth, orientation, shape...), making it hard for a new observer to take labeling decisions in case of conflict.

To help guide future investigators, we now show these metrics for all sulci in Supplemental Figure 7 to help future groups identify these sulci with the assistance of their general morphology.

- The very assertive tone of the article conveys the idea that these sulci are identifiable certainly in most cases, when by definition these highly variable tertiary sulci are sometimes very difficult to take decisions on.

The highly variable nature of ¾ of the putative tertiary sulci (slocs-v, slocs-d, pAngs-v, pAngs-d) described here is why we focused on the slocs-v (as it is identifiable in nearly all f hemispheres). However, we have edited our language throughout the text to also emphasize the variability of these sulci. For example, in the Results (Page 5), we now write:

“In previous research in small sample sizes, neuroanatomists noticed shallow sulci in this cortical expanse (Supplementary Methods and Supplementary Figs. 1-4 for historical details). In the present study, we fully update this sulcal landscape considering these overlooked indentations. In addition to defining the 13 sulci previously described within the LPC/LPOJ, as well as the posterior superior temporal cortex (Methods) (Petrides, 2019) in individual participants, we could also identify as many as four small and shallow PTS situated within the LPC/LPOJ that were highly variable across individuals and uncharted until now (Supplementary Methods and Supplementary Figs. 1-4). Macroanatomically, we could identify two sulci between the cSTS3 and the IPS-PO/lTOS ventrally and two sulci between the cSTS2 and the pips/IPS dorsally. We focus our analyses on the slocs-v since it was identifiable in nearly every hemisphere.”

- I am not absolutely convinced with the labeling proposed of a previously reported sulcus, namely the posterior intermediate parietal sulcus.

In defining previously-identified LPC sulci, we followed the previous labeling procedure by Petrides (2019) alongside historical definitions (detailed in Supplementary Figures 1-4). Nevertheless, future deep learning algorithms using these and others data can be used to rectify discrepancies in labeling (e.g., Borne et al., 2020 Medical Image Analysis; Lyu et al., 2021 NeuroImage). We discuss these points in the Limitations section of the Discussion (Pages 16-17):

“The main limitation of our study is that presently, the most accurate methodology to define sulci —especially the small, shallow, and variable PTS—requires researchers to manually trace each structure on the cortical surface reconstructions. This method is limited due to the individual variability of cortical sulcal patterning (Fig. 1, Supplementary Fig. 5), which makes it challenging to identify sulci without extensive experience and practice. However, we anticipate that our probabilistic maps  will provide a starting point and hopefully, expedite the identification of these sulci in new participants. This should accelerate the process of subsequent studies confirming the accuracy of our updated schematic of LPC/LOPJ. This manual method is also arduous and time-consuming, which, on the one hand, limits the sample size in terms of number of participants, while on the other, results in thousands of precisely defined sulci. This push-pull relationship reflects a broader conversation in the human brain mapping and cognitive neuroscience fields between a balance of large N studies and “precision imaging” studies in individual participants (Allen et al., 2022; Gratton et al., 2022; Naselaris et al., 2021; Rosenberg & Finn, 2022). Though our sample size is comparable to other studies that produced reliable results relating sulcal morphology to brain function and cognition (e.g., (Cachia et al., 2021; Garrison et al., 2015; Lopez-Persem et al., 2019; Miller et al., 2021; Roell et al., 2021; Voorhies et al., 2021; Weiner, 2019; Willbrand, Parker, et al., 2022; Willbrand, Voorhies, et al., 2022; Yao et al., 2022), ongoing work that uses deep learning algorithms to automatically define sulci should result in much larger sample sizes in future studies (Borne et al., 2020; Lyu et al., 2021). Finally, the time-consuming manual definitions of primary, secondary, and PTS also limit the cortical expanse explored in each study, thus, restricting the present study to LPC/LPOJ. “

Assuming that the labelling of all sulci reported in the article is reproducible, the different results are convincing and in general, this study achieves its aims in defining more precisely the sulcation of the LPOJ and looking into its functional/cognitive value. This work clearly offers a finer understanding of sulcal pattern in this region, and lacks only little for the new markers to be convincingly demonstrated. An overall coherence of the labelling can still be inferred from the supplementary material which support the results and therefore the conclusions, yet, addressing some of the weaknesses listed above would greatly enhance the impact of this work. This work is important to the understanding of sulcal variability and its implications on functional and cognitive aspects.

We thank the Reviewer for their positive remarks on the implications of this work.

Reviewer #3 (Public Review):

Summary: 72 subjects, and 144 hemispheres, from the Human Connectome Project had their parietal sulci manually traced. This identified the presence of previously undescribed shallow sulci. One of these sulci, the ventral supralateral occipital sulcus (slocs-v), was then demonstrated to have functional specificity in spatial orientation. The discussion furthermore provides an eloquent overview of our understanding of the anatomy of the parietal cortex, situating their new work into the broader field. Finally, this paper stimulates further debate about the relative value of detailed manual anatomy, inherently limited in participant numbers and areas of the brain covered, against fully automated processing that can cover thousands of participants but easily misses the kinds of anatomical details described here.

Strengths:

- This is the first paper describing the tertiary sulci of the parietal cortex with this level of detail, identifying novel shallow sulci and mapping them to behaviour and function.

- It is a very elegantly written paper, situating the current work into the broader field.

- The combination of detailed anatomy and function and behaviour is superb.

We thank the Reviewer for their positive remarks on paper and our findings.

Weaknesses:

- The numbers of subjects are inherently limited both in number as well as in typically developing young adults.

We emphasize that the sample size is limited due to the arduous nature of manually defining sulci; however, we provide probabilistic maps with the publication of this work to help expedite this process for future investigators. Further, with improved deep learning algorithms, the sample sizes in future neuroanatomical studies should be enhanced. We discuss these points in the Limitations section of the Discussion (Pages 16-17):

“The main limitation of our study is that presently, the most accurate methodology to define sulci —especially the small, shallow, and variable PTS—requires researchers to manually trace each structure on the cortical surface reconstructions. This method is limited due to the individual variability of cortical sulcal patterning (Fig. 1, Supplementary Fig. 5), which makes it challenging to identify sulci without extensive experience and practice. However, we anticipate that our probabilistic maps  will provide a starting point and hopefully, expedite the identification of these sulci in new participants. This should accelerate the process of subsequent studies confirming the accuracy of our updated schematic of LPC/LOPJ. This manual method is also arduous and time-consuming, which, on the one hand, limits the sample size in terms of number of participants, while on the other, results in thousands of precisely defined sulci. This push-pull relationship reflects a broader conversation in the human brain mapping and cognitive neuroscience fields between a balance of large N studies and “precision imaging” studies in individual participants (Allen et al., 2022; Gratton et al., 2022; Naselaris et al., 2021; Rosenberg & Finn, 2022). Though our sample size is comparable to other studies that produced reliable results relating sulcal morphology to brain function and cognition (e.g., (Cachia et al., 2021; Garrison et al., 2015; Lopez-Persem et al., 2019; Miller et al., 2021; Roell et al., 2021; Voorhies et al., 2021; Weiner, 2019; Willbrand, Parker, et al., 2022; Willbrand, Voorhies, et al., 2022; Yao et al., 2022), ongoing work that uses deep learning algorithms to automatically define sulci should result in much larger sample sizes in future studies (Borne et al., 2020; Lyu et al., 2021). The time-consuming manual definitions of primary, secondary, and PTS also limit the cortical expanse explored in each study, thus restricting the present study to LPC/LPOJ.”

- While the paper begins by describing four new sulci, only one is explored further in greater detail.

Due to the increased variability of three of the four newly-classified sulci, we chose to only focus on the slocs-v given that it was present in nearly all hemispheres. In response to other reviewers, we have conducted additional analyses that also describe these new sulci and potential factors related to their incidence (Page 6):

“Given that sulcal incidence and patterning is also sometimes related to demographic features (Cachia et al., 2021; Leonard et al., 2009; Wei et al., 2017), subsequent GLMs relating the incidence and patterning of the three more variable sulci (slocs-d, pAngs-v, and pAngs-d) to demographic features (age and gender) revealed no associations for any sulcus (ps > .05).”

In addition, given that sulcal variability is cognitively (e.g., Amiez et al., 2018 Scientific Reports; Cachia et al., 2021 Frontiers in Neuroanatomy; Garrison et al., 2015 Nature Communications; Willbrand et al., 2022, 2023 Brain Structure & Function), anatomically (e.g., Amiez et al., 2021 Communications Biology; Vogt et al., 1995 Journal of Comparative Neurology), functionally (e.g., Lopez Persem et al., 2019 The Journal of Neuroscience), and translationally (e.g., Yucel et al., 2002 Biological Psychiatry) relevant, future research can investigate these relationships regarding the slocs-d and pAngs components. We have added text to the Limitations section of the Discussion (Pages 17-18) to discuss this:

“Finally, although we did not focus on the relationship between the other three PTS (slocs-d, pAngs-v, and pAngs-d) to anatomical and functional features of LPC and cognition, given that variability in sulcal incidence is cognitively (Amiez et al., 2018; Cachia et al., 2021; Garrison et al., 2015; Willbrand, Jackson, et al., 2023; Willbrand, Voorhies, et al., 2022), anatomically (Amiez et al., 2021; Vogt et al., 1995), functionally (Lopez-Persem et al., 2019), and translationally (Clark et al., 2010; Le Provost et al., 2003; Meredith et al., 2012; Nakamura et al., 2020; Yücel et al., 2002, 2003) relevant, future work can also examine the relationship between the more variable slocs-d, pAngs-v, and pAngs-d and these features.”

- There is some tension between calling the discovered sulci new vs acknowledging they have already been reported, but not named.

We have edited the manuscript throughout to emphasize our primary focus on revising the LPC/LOPJ sulcal landscape to include these often overlooked small, shallow, and variable putative tertiary sulci, rather than using the terms “discovered sulci” and “new.”

- The anatomy of the sulci, as opposed to their relation to other sulci, could be described in greater detail.

Beyond the radar plots in the main text which compare specific groupings of sulci, we now show the morphological metrics for all sulci investigated in the present work in Supplemental Figure 7.

Overall, to summarize, I greatly enjoyed this paper and believe it to be a highly valued contribution to the field.

We are glad the Reviewer enjoyed reading our paper and thank them for their positive thoughts on the potential impact of this work on the field.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

(1) The slocs-v is found in 71 subjects left and right. Is that the same subject?

No, these are different subjects.

(2) How were the 72 subjects chosen?

The subjects were randomly selected from the HCP database as describe in the methods (Page 18):

“Here, we used 72 randomly-selected participants, balanced for gender (following the terminology of the HCP data dictionary), from the HCP database (50% female, 22-36 years old, and 90% right-handed; there was no effect of handedness on our behavioral tasks; Supplementary materials) that were also analyzed in several prior studies (Hathaway et al., 2023; Miller et al., 2021, 2020; Willbrand et al., 2023b, 2023c, 2022a).”

(3) Are there effects of laterality on sulcal pattern? Table?

We now include sulcal pattern results in the Results section and Supplementary Materials; although there were no laterality effects regarding the sulcal pattern .

(4) Depiction/description of common sulcal patterns

We now include sulcal pattern results in the Results section and Supplementary Materials.

(5) Is there a relationship between sulcal patterns and demographic features?

We now include analyses on this in the Results section. There is no relationship between sulcal patterns and demographic features.

(6) Just for clarity, the sulcal features are studied and extracted in native space?

Yes, sulcal features are studied and extracted in native space, as described in the Methods section (Page 19):

“Anatomical T1-weighted (T1-w) MRI scans (0.8 mm voxel resolution) were obtained in native space from the HCP database. Reconstructions of the cortical surfaces of each participant were generated using FreeSurfer (v6.0.0), a software package used for processing and analyzing human brain MRI images (surfer.nmr.mgh.harvard.edu) (Dale et al., 1999; Fischl et al., 1999). All subsequent sulcal labeling and extraction of anatomical metrics were calculated from these native space reconstructions generated through the HCP’s version of the FreeSurfer pipeline (Glasser et al., 2013).”

(7) The authors use "Gender". Are they referring to biological sex (female/male) or socially defined characteristics (man/woman etc.)?

The term gender is referred to socially defined characteristics, as used by the HCP data dictionary (Methods page 18):

“Here, we used 72 randomly-selected participants, balanced for gender (following the terminology of the HCP data dictionary), from the HCP database (50% female, 22-36 years old, and 90% right-handed; there was no effect of handedness on our behavioral tasks; Supplementary materials) that were also analyzed in several prior studies (Hathaway et al., 2023; Miller et al., 2021, 2020; Willbrand et al., 2023b, 2023c, 2022a).”

(8) Fig 2. Grey is poorly visible compared to green and blue.

The shade of gray has been edited to be more distinguishable.

(9) The relationship between behavior and sulcal features is significant but weak.

We acknowledge that the morphological-behavioral relationship identified in the present study explains a modest amount of variance; however, the more important aspect of the finding is that multiple sulci identified in the model are recently-characterized sulci in LPC/LOPJ identified by our group and others (Petrides, 2019), and thus, the relationship would have been overlooked or lost if these sulci were not identified. We have added text to the Limitations section of the Discussion (Pages 17-18) to emphasize this point:

“It is also worth noting that the morphological-behavioral relationship identified in the present study explains a modest  amount of variance; however, the more important aspect of our findings is that multiple sulci identified in our model-based approach are recently-characterized sulci in LPC/LOPJ identified by our group and others (Petrides, 2019), and thus, the relationship would have been overlooked or lost if these sulci were not identified. “

(10) The Limitation section could be expanded.

We have added additional text to flesh out the Limitations section of the Discussion (Pages 17-18):

“It is also worth noting that the morphological-behavioral relationship identified in the present study explains a modest  amount of variance; however, the more important aspect of our findings is that multiple sulci identified in our model-based approach are recently-characterized sulci in LPC/LOPJ identified by our group and others (Petrides, 2019), and thus, the relationship would have been overlooked or lost if these sulci were not identified. Finally, although we did not focus on the relationship between the other three PTS (slocs-d, pAngs-v, and pAngs-d) to anatomical and functional features of LPC and cognition, given that variability in sulcal incidence is cognitively (Amiez et al., 2018; Cachia et al., 2021; Garrison et al., 2015; Willbrand, Jackson, et al., 2023; Willbrand, Voorhies, et al., 2022), anatomically (Amiez et al., 2021; Vogt et al., 1995), functionally (Lopez-Persem et al., 2019), and translationally (Clark et al., 2010; Le Provost et al., 2003; Meredith et al., 2012; Nakamura et al., 2020; Yücel et al., 2002, 2003) relevant, future work can also examine the relationship between the more variable slocs-d, pAngs-v, and pAngs-d and these features. “

Reviewer #2 (Recommendations For The Authors):

First, I would like to thank the authors for their important contribution to the field of sulcal studies and anatomo-functional correlates. My main comments about the work are treated in the public review, and I will only address details in this section. I have detected a number of typos which are harder to report from a document in which lines are not numbered. Could you please submit a numbered document for the next iteration?

- p2. "hominoid-specific, shallow indentations, or sulci" - can lead to misunderstanding that sulci are hominoid-specific and shallow

Sentence has been rewritten:

“Of all the neuroanatomical features to target, recent work shows that morphological features of the shallower, later developing, hominoid-specific indentations of the cerebral cortex (also known as putative tertiary sulci, PTS) are not only functionally and cognitively meaningful, but also are particularly impacted by multiple brain-related disorders and aging (Amiez et al., 2019, 2018; Ammons et al., 2021; Cachia et al., 2021; Fornito et al., 2004; Garrison et al., 2015; Harper et al., 2022; Hathaway et al., 2023; Lopez-Persem et al., 2019; Miller et al., 2021, 2020; Nakamura et al., 2020; Parker et al., 2023; Voorhies et al., 2021; Weiner, 2019; Willbrand et al., 2023b, 2023c, 2022a, 2022b; Yao et al., 2022).”

- p2. next sentence (starting with "The combination [...]": not clear that you are addressing tertiary sulci here, maybe introduce the concept beforehand?

The previous sentence (just above) has been edited to introduce putative tertiary sulci beforehand.

- p5. error in numbering of sulci relative to Fig1. (5,6,7,8 -> 6,7,8,9)

Sulcal numbering has been fixed.

-p5. reference to supp mat -> I would have expected the nomenclature used in Borne et al. 2020 to be discussed alongside with the state of the art. How would you relate F.I.P.r.int.1 and F.I.P.r.int.2 to the sulci you describe?

We thank the Reviewer for bringing up this relevant literature. The F.I.P.r.int. 1 and 2 are described as rami of the IPS, whereas the slocs and pAngs are independent, small indentations near the IPS, but not part of the complex. Nevertheless, future work should integrate these two schematics together to establish the most comprehensive sulcal map of LPC/LOPJ. We have added text to the Supplementary Methods detailing the differences between the F.I.P.r.int.1 and F.I.P.r.int.2 and slocs-/pAngs:

“slocs/pAng vs. F.I.P.r.int.1 and F.I.P.r.int.2

Recent work (Borne et al., 2020; Perrot et al., 2011) identified two intermediate rami of the IPS (F.I.P.r.int.1 and F.I.P.r.int.2) that were not defined in the present investigation. Crucially, the newly classified sulci here (slocs and pAngs) are distinguishable from the two F.I.P.r.int. in that the F.I.P.r.int. are branches coming off the main body of the IPS (Borne et al., 2020; Perrot et al., 2011), whereas the slocs/pAngs are predominantly non-intersecting (“free”) structures that never intersected with the IPS (Supplementary Tables 1-4).”

- p6. Fig 1.a. labelling discrepancy between line 1 and 2, column 4: the labels 10 and 11 from the inflated hemisphere do not match the labels 10 and 11 in the pial surface. Fig 1.b. swapped label 2 and 3 in the 4th hemisphere

These aspects of Figure 1 have been edited accordingly.

- p7. "(iii) the slocs-v was thicker than both the cSTS3 and lTOS" -> the slocs-v showed thicker gray matter?

The sentence has been adjusted (Page 7):

“(iii) the slocs-v showed thicker gray matter than both the cSTS3 and lTOS (ps < .001), “

- p9. Six left hemisphere LPC/LPOJ sulci were related to spatial orientation task performance -> missing

Fixed (Page 9):

“Six left hemisphere LPC/LPOJ sulci were related to spatial orientation task performance (Fig. 3a, b). “

- p14. "Steel and colleagues" -> missing space

Fixed (Page 14):

“Furthermore, the slocs-v appears to lie at the junction of scene-perception and place-memory activity (a transition that also consistently co-localizes with the HCP-MMP area PGp) as identified by Steel and colleagues (2021).”

- p20. Probability maps "we share these maps with the field" -> specify link to data availability

The link to data availability has been added (Page 21):

“To aid future studies interested in investigating LPC/LPOJ sulci, we share these maps with the field (Data availability). “

Reviewer #3 (Recommendations For The Authors):

No detailed recommendations not already present in the rest of the review.

Reviewer #2 (Public Review):

Summary:<br /> After manually labelling 144 human adult hemispheres in the lateral parieto-occipital junction (LPOJ), the authors 1) propose a nomenclature for 4 previously unnamed highly variable sulci located between the temporal and parietal or occipital lobes, 2) focus on one of these newly named sulci, namely the ventral supralateral occipital sulcus (slocs-v) and compare it to neighbouring sulci to demonstrate its specificity (in terms of depth, surface area, gray matter thickness, myelination, and connectivity), 3) relate the morphology of a subgroup of sulci from the region including the slocs-v to the performance in a spatial orientation task, demonstrating behavioural and morphological specificity. In addition to these results, the authors propose an extended reflection on the relationship between these newly named landmarks and previous anatomical studies, a reflection about the slocs-v related to functional and cytoarchitectonic parcellations as well as anatomic connectivity and an insight about potential anatomical mechanisms relating sulcation and behaviour.

Strengths:<br /> - To my knowledge, this is the first study addressing the variable tertiary sulci located between the superior temporal sulcus (STS) and intra-parietal sulcus (IPS).<br /> - This is a very comprehensive study addressing altogether anatomical, architectural, functional and cognitive aspects.<br /> - The definition of highly variable yet highly reproductible sulci such as the slocs-v feeds the community with new anatomo-functional landmarks (which is emphasized by the provision of a probability map in supp. mat., which in my opinion should be proposed in the main body).<br /> - The comparison of different features between the slocs-v and similar sulci is useful to demonstrate their difference.<br /> - The detailed comparison of the present study with state of the art contextualises and strengthens the novel findings.<br /> - The functional study complements the anatomical description and points towards cognitive specificity related to a subset of sulci from the LPOJ<br /> - The discussion offers a proposition of theoretical interpretation of the findings<br /> - The data and code are mostly available online (raw data made available upon request).

Weaknesses:<br /> - While the identification of the sulci has been done thoroughly with expert validation, the sulci have not been labelled in a way that enables the demonstration of the reproducibility of the labelling.

The proposed methodology is convincing in identifying and studying the relationship between highly variable sulci and cognition. This improves our refined understanding of the general anatomical variability in the LPOJ and its potential functional/cognitive correlates. This work is important to the understanding of sulcal variability and its implications on functional and cognitive aspects.

Comments on revised version:

Thank you for the elegant and informative work.

DOI: 10.1016/j.isci.2024.110585

Resource: (IMSR Cat# CRL_022,RRID:IMSR_CRL:022)

Curator: @vtello

SciCrunch record: RRID:IMSR_CRL:022

What is this?

DOI: 10.1016/j.isci.2024.110585

Resource: RRID:IMSR_CRL:027

Curator: @vtello

SciCrunch record: RRID:IMSR_CRL:027

What is this?

DOI: 10.1016/j.taap.2024.117047

Resource: (IMSR Cat# CRL_207,RRID:IMSR_CRL:207)

Curator: @vtello

SciCrunch record: RRID:IMSR_CRL:207

What is this?

J'aimerais comprendre pourquoi les balises universelles <div> à l'intérieur des balises header et footer son disparus alors que ci-dessus mentionne que l'on peut les avoir. Merci d'avance

Reviewer #2 (Public Review):

Summary:

This paper addresses an important computational problem in learning and memory. Why do related memory representations sometimes become more similar to each other (integration) and sometimes more distinct (differentiation)? Classic supervised learning models predict that shared associations should cause memories to integrate, but these models have recently been challenged by empirical data showing that shared associations can sometimes cause differentiation. The authors have previously proposed that unsupervised learning may account for these unintuitive data. Here, they follow up on this idea by actually implementing an unsupervised neural network model that updates the connections between memories based on the amount of coactivity between them. The authors use their modeling framework to simulate three recent empirical studies, showing that their model captures aspects of these findings that are hard to account for with supervised learning.

Overall, this is a strong and clearly described work that is likely to have a positive impact on computational and empirical work in learning and memory. While the authors have written about some of the ideas discussed in this paper previously, a fully implemented and openly available model is a clear advance that will benefit the field. It is not easy to translate a high-level description of a learning rule into a model that actually runs and behaves as expected. The fact that the authors have made all their code available makes it likely that other researchers will extend the model in numerous interesting ways, many of which the authors have discussed and highlighted in their paper.

Strengths:

The authors succeed in demonstrating that unsupervised learning with a simple u-shaped rule can produce results that are qualitatively in line with the empirical reports. In each of the three models, the authors manipulate stimulus similarity (following Chanales et al.), shared vs distinct associations (following Favila et al.), or learning strength (a stand-in for blocked versus interleaved learning schedule; following Schlichting et al.). In all cases, with hand-tuning of additional parameters, the authors are able to produce model representations that fit the empirical results, but that can't easily be accounted for by supervised learning. Demonstrating these effects isn't trivial and a formal modeling framework for doing so is a valuable contribution. Overall, the work is very thorough. The authors investigate many different aspects of the learning dynamics (learning rate, oscillation strength, hidden layer overlap etc) in these models and produce several key insights. Of particular value are their demonstrations that when differentiation occurs, it occurs very quickly and asymmetrically and results in anti-correlated representations, as well as the distinction between symmetric and asymmetric integration in their model. The authors thoroughly acknowledge the relative difficulty of producing differentiation in their models relative to integration, and are now more clear about why they don't necessarily view this as mismatch with the empirical data. The authors are also more clear about the complicated activation dynamics in their model and why critical ranges for some parameters can't be given -- the number of interacting parameters mean that there are many combinations that could produce the critical activation dynamics and thus the same result. Despite this complexity, the paper is very clearly written; the authors do a good job of both formally describing their model as well as giving readers a high level sense of how many of their critical model components work.

Weaknesses:

Though the u-shaped learning rule is essential to this framework, the paper doesn't do any formal investigation of this learning rule or comparison with other learning rules. The authors do have a strong theoretical interest in this rule as well as experimental precedent for testing this rule, which they now thoroughly discuss in the paper. Still, a stronger argument in support of the non monotonic plasticity hypothesis could have been made by comparing this learning rule to alternatives. Additionally, the authors' choice of strongly prewiring associations makes it difficult to think about how their model maps onto experimental contexts where associations are only weakly learned. However, the authors thoroughly acknowledge why this was necessary and discuss this limitation in the paper.

Reviewer #2 (Public Review):

Guan and colleagues address the question of how a single neuroblast produces a defined number of progeny, and what influences its decommissioning. The focus of the experiments are two well-studied RNA-binding proteins: Imp and Syp. The Authors find that these factors play an important role in determining the number of neurons in their preferred model system of VNC motor neurons coming from a single lineage (LinA/15) by separate functions taking place at specific stages of development of this lineage: influencing the life-span of the LinA neuroblast to control its timely decommissioning and functioning in the Late-born post-mitotic neurons to influence cell death after the appropriate number of progeny is generated. The post-mitotic role of Imp/Syp in regulating programmed-cell death (PCD) is also correlated with a specific code of key transcription factors that are suspected to influence neuronal identity, linking the fate of neuronal survival with its specification. This paper addresses a wide scope of phenotypes related to the same factors, thus providing an intriguing demonstration of how the nervous system is constructed by context-specific changes in key developmental regulators. The bulk of conclusions drawn by the authors are supported by careful experimental evidence, and the findings are a useful addition to an important topic in developmental neuroscience.

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public Review):

This study addresses the temporal patterning of a specific Drosophila CNS neuroblast lineage, focusing on its larval development. They find that a temporal cascade, involving the Imp and Syb genes changes the fate of one daughter cell/branch, from glioblast (GB) to programmed cell death (PCD), as well as gates the decommissioning of the NB at the end of neurogenesis.

I believe there are some inaccuracies in this summary. We address temporal patterning during larval and pupal stages until the adult stage. The Imp and Syp genes change the fate of one daughter cell/branch from survival to programmed cell death (PCD). The change from glioblast (GB) to PCD, which occurs at an early time point, is not addressed here. The main point of the paper is missing:

• Last-born MNs undergo apoptosis due to their failure to express a functional TF code, and this code is post-transcriptionally regulated by the opposite expression of Imp and Syp in immature MNs.

Reviewer #2 (Public Review):

Summary:

Guan and colleagues address the question of how a single neuroblast produces a defined number of progeny, and what influences its decommissioning. The focus of the experiments are two well-studied RNA-binding proteins: Imp and Syp. The Authors find that these factors play an important role in determining the number of neurons in their preferred model system of VNC motor neurons coming from a single lineage (LinA/15) by separate functions taking place at specific stages of development of this lineage: influencing the life-span of the LinA neuroblast to control its timely decommissioning and functioning in the Late-born post-mitotic neurons to influence cell death after the appropriate number of progeny is generated. The post-mitotic role of Imp/Syp in regulating programmed-cell death (PCD) is also correlated with a specific code of key transcription factors that are suspected to influence neuronal identity, linking the fate of neuronal survival with its specification. This paper addresses a wide scope of phenotypes related to the same factors, thus providing an intriguing demonstration of how the nervous system is constructed by context-specific changes in key developmental regulators.

The bulk of conclusions drawn by the authors are supported by careful experimental evidence, and the findings are a useful addition to an important topic in developmental neuroscience.

I cannot summarize better the paper.

Strengths:

A major strength is the use of a genetic labeling tool that allows the authors to specifically analyze and manipulate one neuronal lineage. This allows for simultaneous study of both the progenitors and post-mitotic progeny. As a result the paper conveys a lot of useful information for this particular neuronal lineage. Furthermore addressing the association of cell fate specification, taking advantage of this lab's extensive prior work in the system, with developmentally-regulated programmed celldeath is an important contribution to the field.

Beyond Imp/Syp, additional characterization of this model system is provided in characterizing a previously unrecognized death of a hemilineage in early-born neurons.

Thanks!

Weaknesses:

The main observations that distinguish this study from others that have investigated Imp/Syp in the fly nervous system is the role played in late-born post-mitotic neurons to regulate programmed cell death. This is an important and plausible (based on the presented findings) newly discovered role for these proteins. However the precision of experiments is not particularly strong, which limits the authors claims. The genetic strategy used to manipulate Imp/Syp or the TF code appears to be done throughout the entire lineage, or all neuronal progeny, and not restricted to only the late born cells. Can the authors rule out survival of the early born hemi-lineage normally fated to die? Therefore statements such as this: 

To further investigate this possibility, we used the MARCM technique to change the TF code of lastborn MNs without affecting the expression of Imp and Syp should be qualified to specify that the result is obtained by misexpressing these factors throughout the entire lineage.

We agree that our genetic manipulations affect the entire lineage or all neuronal progeny. We do not have genetic tools to gain such precision. We have changed our descriptions to specify the entire lineage or all neuronal progeny. As the reviewer raised, we were also concerned about the possibility that the overexpression of Imp or knockdown of Syp could induce the survival of the early-born hemilineage. We have two experiments that rule out this possibility:

(1) In late LL3 larvae, Imp OE or syp MARCM clones do not change the number of cells in LL3 larvae (see Guan et al., 2022), indicating that the hemilineage that died by PCD is not affected. If Imp or Syp played a role in the survival of the hemilineage, we would see at least a 50% increase in the number of MNs at this stage.

(2) The MARCM experiment using the VGlut driver to overexpress P35 or Imp allows us to manipulate only elav+ VGlut+ neurons. The hemilineage removed by PCD is elav- VGlut- and is not affected by this experiment. Consequently, the increase in MNs in adults with genetic manipulation can only be the result of the survival of the other hemilineage (elav+, VGlut+). Moreover, this experiment shows an increase in the number of neurons in the adult but not in LL3, demonstrating that the hemilineage (elav- VGlut-) is still removed by PCD with this genetic manipulation.

The authors make an observation that differs from other systems in which Imp/Syp have been studied: that the expression of the two proteins appears to be independent and not influenced by cross-regulation. However there is a lack of investigation as to what effect this may have on how Imp/Syp regulate temporal identity. A key implication of the previously observed cross-regulation in the fly mushroom body is that the ratio of Imp/Syp could change over the life of the NB which would permit different neuronal identities. Without cross-regulation, do the authors still observe a gradient in the expression pattern of time? Because the data is presented with Imp and Syp stained in different brain samples, and without quantification across different stages, this is unclear. The authors use the term 'gradient' but changes in levels of these factors are not evident from the presented data.

We have now quantified the transcriptional activity of Imp and Syp in the NB over time using smFISH. We have also quantified the relative expression of Imp and Syp protein in the NB over time by co-immunostaining. Additionally, we quantified the relative expression of Imp and Syp protein in postmitotic neurons as a function of their birth order in late LL3 larvae. All these data show an opposite temporal gradient of Imp and Syp in the NB and an opposite spatial gradient in immature neurons according to their birth order (Figure. 4). How these gradients are established in our system remains to be elucidated. 

Reviewer #3 (Public Review):

This study by Guan and co-workers focuses on a model neuronal lineage in the developing Drosophila nervous system, revealing interesting aspects about: a) the generation of supernumerary cells, later destined for apoptosis; and, b) new insights into the mechanisms that regulate this process. The two RNA-binding proteins, Imp and Syp, are shown to be expressed in temporally largely complementary patterns, their expression defining early vs later born neurons in this lineage, and thus also regulating the apoptotic elimination. Moreover, neuronal 'fate' transcription factors that are downstream of Imp and signatures of early-born neurons, can also be sufficient to convert later born cells to an earlier 'fate', including survival.

The authors provide solid evidence for most of their statements, including the temporal windows during which the early and the later-born motoneurons are generated by this model lineage, how this relates to patterns of cell death by apoptosis and that mis-expression of early-born transcription factors in later-born cells can be sufficient to block apoptosis (part of, and perhaps indicative of the late-born identity).

Other studies have previously outlined analogous, mutually antagonistic roles for Imp and Syp during nervous system development in Drosophila, in different parts and at different stages, with which the working model of this study aligns.

Overall, this study adds to and extends current working models and evidence on the developmental mechanisms that underlie temporal cell fate decisions.

I cannot summarize better the paper.

Reviewer #1 (Recommendations For The Authors):

While this is an interesting topic, I raised two issues in my original review.

(1) Against the backdrop of numerous previous studies linking many developmental regulators, including tTFs, to programmed cell death in the developing CNS, which in several cases have involved identifying key PCD genes and decoding the molecular regulatory interplay between regulators and PCD genes, this study does not provide any new insight into the regulation of developmental PCD in the CNS.

The authors have not added any new data to address this shortcoming.

I agree with the reviewer that we did not attempt to link Imp/Syp with the temporal transcription factor (tTF) cascade or spatial selectors such as Hox genes. However, this decision was intentional as our primary focus was on studying immature MNs. It is worth noting that the decommissioning of NBs by autophagic cell death or terminal differentiation, which is mediated by Imp/Syp in other lineages, has not been correlated with tTFs or spatial selectors. Although we have not directly examined the involvement of the hb + sv > kr > pdm > cas > cas-svp > Grh cascade in the decommissioning of the Lin A neuroblast, our preliminary data indicate that Hb, Sv, Pdm, and Cas are not expressed in the Lin A NB, while Grh is consistently expressed in the NB (Wenyue et al., 2022). Thus, it is less likely that this particular tTF cascade is not implicated in Lin A neuroblast decommissioning. In contrast, spatial selectors, such as the Hox gene Antp, play an opposing role compared to HOX transcription factors in abdominal NBs. In the Lin A lineage, Antp promotes survival (Baek, Enriquez, & Mann, 2013). Here, to avoid repeating what has already been described in the literature, we focused on the role of Imp/Syp in postmitotic neurons and revealed that the precise elimination of MNs is linked to the control of TFs expressed in the MNs.

(2) I raised the issue that it is unclear if Imp/Syp acts in the NB, and/or in IMC/GMC, and/or in the daughter cells generated from these.

I agree with the reviewer's concern regarding the unclear function of Imp/Syp, i.e., whether it acts in the NB, IMC/GMC, or daughter cells. To address this, one possible approach would be to attempt rescuing Imp and Syp mutants by transgenic expression in specific cell types, such as NBs, IMC/GMC, or GB/daughter cells. However, we have not conducted such experiments as we were skeptical about the outcome. Previous published work has used drivers expressed in NBs, IMC/GMC, or postmitotic neurons to decipher the function of a gene in a specific cell type. But the results of these experiments must be taken with caution. Using NB/GMC drivers to study gene function can lead to effects not only in the NB but also in its progeny, including GMC or postmitotic neurons, due to the perdurance and stability of the Gal4 and UAS-gene expression system. For instance, dpn-Gal4 UASGFP not only labels the NB but also many of its progeny, even if Dpn is only expressed in NBs. And elav-Gal4 is expressed in the NB and GMCs.

However, our overexpression of Imp in immature neurons using Vglut demonstrates that Imp promotes cell survival through an autonomous function in these neurons. This driver is only expressed in postmitotic neurons (elav+) and not in the NB, IMC/GMC, or in the hemilineage eliminated by cell death (elav-vglut-).

Reviewer #2 (Recommendations For The Authors):

Oddly knockdown of Imp in the neuroblast (Fig. 5D) only led to death at 8h APF, when Imp is no longer expressed. Do the authors have an explanation as to how the stem cell can survive until this point? A discussion would be helpful.

The simple explanation is the efficiency of RNAi. The imp-/- MARCM clones (Guan et al., 2022) lead to a stronger reduction of MNs in LL3.

A simple experiment I would recommend is to repeat the antibody stainings of staged larvae/pupae (Fig. 4) having the anti-Imp/Syp antibodies in the same brain sample, and perhaps a quantification of the ratio in the NB. Given the species in which the ABs were raised seem compatible, this should be feasible. As it stands now, there is no indication of whether the ratio of Imp vs Syp change over time.

We have now quantified the transcriptional activity of Imp and Syp in the NB over time. We have also quantified the relative expression of Imp and Syp proteins in the NB over time and quantified the relative expression of Imp and Syp proteins in postmitotic neurons as a function of their birth in late LL3 larvae. How these gradients are established in our system still remains to be 

Minor errors/suggestions:

Fig 4. Time legend at the top goes A, B, C, E, F (no D). So it doesn't match the panels below

Yes, we have made the corrections.

Sentence repeated in Intro:

The process of terminating NB neurogenesis through autophagic cell death or terminal differentiation is commonly referred to as decommissioning.

Yes, corrections have been made.

IN FIGURE 1 THEY SAY 'TYPE IB' AND IN FIGURE 2 THEY SAY 'TYPE 1B'

We have changed it to type 1b.

In Fig2A-It's hard to see lack of Elav and Fig2G-It's hard to see presence of Dcp1. Panels could be adjusted to emphasize these results

We have increased the size of the panels and made two separate panels where only the elav and Dcp1 signals are present.

Observations that the result is equivalent in all thoracic segments is expected, since all legs need the same number of neurons. This is nice to have but can be in the supplement.

Overall the figure number seems excessive, especially considering much of the results included(particularly the NB results) are findings consistent with previous papers and some is characterization of the system that does not fit well with the main focus regarding Imp/Syp (i.e death of one hemi-lineage:

Figure 5 and 6 can be joined as one.

We have combined Figures 5 and 6, showing only the T1 segments.

There is some discrepancy between graphs Fig7F and K: At LL3 the number of neurons is different for the control in 7F and the count in K

Yes, because the genetic backgrounds are not the same and we are not counting the same type of cells. In 7F, we are counting the elav+ and VGlut+ cells, whereas in Figure 7K, we are counting all the elav+ in Lin A, including those elav+ VGlut-. VGlut expression arrives a bit later after elav+, which is why we have fewer elav+ cells in 7F. In other words, VGlut MARCM clones do not label all Lin A elav+ cells. I have clarified this in the figure.

Reviewer #3 (Recommendations For The Authors):

Main comment: on the notion of Imp and Syp gradients:

p. 5, related to figure 4 - there are clearly distinct windows for predominantly (if not exclusively) Imp, and later, Syp expression in lineage 15, with a phase of co-expression.

However, based on the data shown, it is unclear whether these windows represent gradients, as repeatedly stated. If the notion of gradients is derived from other studies, on other lineages, then this would be good to clarify. Alternatively, the idea of temporally opposing gradients of Imp and Syp would need to be demonstrated for this lineage.

For example, a more accurate way to describe this study's data is given on p.7 "In conclusion, our findings demonstrate that the opposite expression pattern of Imp and Syp in postmitotic neurons precisely shapes the size of Lin A/15 lineage by controlling the pattern of PCD in immature MNs (Fig. 8)."

We have now quantified the transcriptional activity of Imp and Syp in the NB over time. We have also quantified the relative expression of Imp and Syp proteins in the NB over time. We have also quantified the relative expression of Imp and Syp proteins in postmitotic neurons as a function of their birth in late LL3 larvae. How these gradients are established in our system still remains to be identified.

Minor points:

p.6, related to figure 7: Are numbers of EDU- early born and EDU+, late born, MNs expressed as means in the main text? As written, it suggests absence of any variability, which one would expect and which is shown in Fig.7 data.

Yes, we have added averages in the text.

Methods: the author name 'Lacin' has been mis-spelled

Sorry about that, it's been corrected.

Note: This response was posted by the corresponding author to Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

Reply to the reviewers

We would first like to thank the reviewers for their careful reading and thoughtful feedback.

We have substantially revised the manuscript and included additional experimental evidence on O-GlcNAc and OGT/OGA protein levels in the placenta of embryos bearing the OGT-Y851A hypomorphic mutation.

Overall, we believe our improved manuscript provides compelling evidence that the glycosyltransferase activity of OGT, and thus the O-GlcNAc modification itself, plays a sexually dimorphic function in placental development and the developmental repression of retrotransposons in the developing embryo.

We have addressed each of the reviewers' comments below. The original comments (C) are in italic, our responses (R) in Roman font.

Reviewer #1

Evidence, reproducibility and clarity

C1: Formichetti at el. developed mice with OGT catalytic dead mutations and then studied their function during early embryogenesis. Not surprisingly, dramatic reduction in OGT activity failed to produce embryos; however, mild reduction in OGT did produce animals. The authors then use the T931 animals that have a mild reduction in activity to further characterize the function in the early embryo. Not surprisingly, male mice showed changes in gene expression, implantation sub-lethality, and an uptick in loss of retrotransposon silencing. The authors also show that an even milder reduction in OGT activity (Y851A) effects male placenta function and chromatin remodeling. Finally, the authors make a less stable OGT transgene within the mouse and again found embryogenesis issues in the males and alterations in numerous gene families including mTOR signaling and p53 function. All in all, this is an interesting study that track functions of OGT in early embryonic development. The studies are well-controlled and rigorous.

R1: We thank the reviewer for their clear understanding and their appreciation of the rigor and impact of this work.

Significance

C2: This is a good study and novel. Not only is it of interest to reproductive biologist, but it echos themes found in O-GlcNAc biology.

R1: We are pleased that the reviewer underlined the novelty of the study and its impact across fields.

Reviewer #2

Evidence, reproducibility and clarity

Comments to authors

C3: To investigate the function of OGT at specific developmental stages, the authors perturbed OGT's function in vivo by creating a murine allelic series featuring four single amino acid substitutions that variably reduced OGT's catalytic activity. The goal was to identify the direct effect of O-GlcNAcylation, using a sophisticated collection of genetic mutants to evaluate in vivo the role of this modification at early stages of development. Overall, the severity of embryonic lethality correlated with the extent of catalytic impairment of OGT, demonstrating that the O-GlcNAc modification is essential for early development.The study represents a substantial advance in our understanding of OGT and O-GlcNAcylation in mammalian development. The creation of novel murine models and inducible systems is an important contribution, providing powerful tools for future research in this field. The insights into the role of OGT's catalytic activity and its involvement in epigenetic regulation during embryonic development are noteworthy, opening new avenues for research.

R3: We thank the reviewer for their insightful comments. We are grateful for the supporting statements. Please find below detailed response to all your comments.

However, there are a few considerations and concerns:

Major:

C4: 1. An assumption of the study is that different mutations cause different levels of O-GlcNAcylation rather than alterations in substrate specificity. It might be important to test, at least in cultured cells, that the different mutations do not change the preference of OGT to modify certain proteins rather than others, which can provide alternative explanations for their findings.

R4: Thanks for asking this question, it helped us to better explain the rationale behind the choice of the Ogt amino-acid substitutions.

This is a critical point that we carefully considered in the design of the single amino-acid substitutions. Two lines of evidence support that the precise mutations created impact the catalytic rate without modifying the substrate specificity:

First, as explained in the text, the choice of the single amino-acid substitutions was driven by previous structural and enzymology knowledge. The impact of the four point mutations selected on OGT protein stability and on the Michaelis-Menten kinetic values had previously been determined experimentally (Fig. 1A legend and Martinez-Fleites, C. et al. Nature Structure Molecular Biology 2008; https://doi.org/10.1038/nsmb.1443).

There is a second important rationale that we added in the revised manuscript: the four point mutations selected are all located in the catalytic domain (specifically, H568A in the N-Cat domain and Y851A, T931A and Q849A in the C-Cat domain), while the substrate recognition is operated via two other domains namely the intervening domain (Int-D) https://doi.org/10.1038/s41589-023-01422-2) and the tetratricopeptide Repeat (TPR) superhelix (10.1021/jacs.7b13546; https://doi.org/10.1073/pnas.2303690120). Therefore, for both these reasons, it is extremely unlikely that these mutations could influence the substrate specificity.

C5.1: 2. In Fig 1D and 1H, the thresholds to define a gene or TE as differentially expressed are not strong. According to the figure legends, "any" change in terms of log2Fc was considered as DE and colored. I think the figures should illustrate better that the changes are subtle, by for example adding a dotted line (at least) in the value 0.5 of the y-axis. These subtle transcriptional changes should be reflected better in certain paragraphs where the expression of TEs are presented/and discussed as a hallmark of the absence of O-GlcNAcylation in the OGT-mutants. The same happens with Suppl Fig 3C (changes are very minor). {. Applying a stronger threshold, among the upregulated genes, only Xist will be significantly overexpressed. If a gentle threshold needs to be applied to this data, authors should at least justify the reasons behind doing so. Same for Fig2D.

R5.1: The reviewer means Figure 2D for MA plot of gene expression and Figure 2H for retrotransposons expression. These figures now include a dash line to indicate Log2FC = 0.5 (as all MA plots).

The text is explicit on the subtle changes in transcription, it reads "with 2/3 of the genes downregulated and 90% of the significant changes below 1 log__2__FC"; "most of the Ogt__T931del/Y embryos showed a low magnitude upregulation of retrotransposons".

The revised text states "Notably, most of the OgtT931__del/Y embryos showed a low magnitude (log2FC < 1) upregulation of retrotransposons".

We expand on this topic in the next response (R5.2) noting that changes in gene expression upon O-GlcNAc perturbation in different systems were previously characterized as subtle and widespread. We suggest that this phenotype may arise from the scarcely understood pleiotropic function of O-GlcNAc in fine-tuning gene expression; this phenotype could have a biological significance.

C5.2: If a gentle threshold needs to be applied to this data, authors should at least justify the reasons behind doing so. Same for Fig2D.

R5.2: Previous studies in different systems reported that O-GlcNAc perturbation causes a widespread change in gene expression of low magnitude (https://doi.org/10.1101/2024.01.22.576677, https://www.pnas.org/doi/10.1073/pnas.2218332120). We use the same thresholds as a recent functional Ogt study in ES cells to call differentially expressed genes, specifically: p<0.05 (Wald test), any FC (Li et al. PNAS 2023, https://www.pnas.org/doi/10.1073/pnas.2218332120). The p value threshold is standard; the absence of FC threshold is dictated by the insufficient knowledge of the significance of the low magnitude changes observed across many transcripts.

C6: 3. In Figure 2B, the T931del allele was recovered in the blastocyst population with a very high frequency, even higher than the male WT group (T931del: 10; WT: 3). This observation suggests that the T931del allele did not significantly affect blastocyst survival. Further clarification or additional experiments might be necessary to understand the implications of this finding on early developmental stages.

R6: This is only a hint as the numbers of blastocysts recovered were too small to perform statistics on Mendelian distribution. Thus, more experiments are needed to perform these statistical tests. These experiments are onerous because the low frequency of germline transmission is incompatible with maintaining this mutation by breeding heterozygous animals. Because of this, a new mouse line needs to be created by CRISPR-HDR targeting in the zygote in order to compute statistics on Mandelian ratios. Importantly, this question - does T931del affect blastocyst survival? - is peripheral, and the results of these experiments would not affect our conclusions in any way.

C7: 4. Similarly, in Figure 2G, there is an apparent higher expression of TE expression in the T931A/Y embryos group than in the T931del/Y group, which combined with the higher frequency of blastocyst generated in this latest group it may indicate a deeper molecular consequence after the deletion of the T931. A comparison of the transcriptome between these two cell lines help to address this possibility. Also, the authors should compare the O-GlcNAc levels of WT, T931A, and T931del mutant blastocysts by immunostaining, similar to what was done in Figure S5F.

R7: We agree that a direct comparison between the two mutations of the T931 residue would be interesting; however, this comment is very difficult to address experimentally for the reasons outlined below:

Firstly, it is not possible to perform a statistical comparison of the transcriptome T931A/Y VS. T931del/Y with the data generated because the number of hemizygous T931A/Y (n=2) is too small. Hence, it cannot be ruled out that the seemingly milder retrotransposon reactivation in one of the T931A/Y embryos could have occurred by chance.

Secondly, considering the low magnitude effect on gene expression changes upon O-GlcNAc genetic perturbation, to statistically assess the penetrance of the molecular phenotype and perform the differential expression analysis, numerous (>>3) hemizygous blastocysts of each genotype would be needed. Because females heterozygous for the T931 mutations transmit the mutant allele at very low frequency, these experiments require numerous de novo CRISPR injection sessions.

Thirdly, for the immunostaining of O-GlcNAc to be semi-quantitative, a large number of hemizygous blastocysts for each genotype would be required (note that in Figure S5F, 29 morulae per condition were imaged), thus requiring numerous CRISPR injection experiments as discussed above. Moreover, O-GlcNAc changes could be subtler than what expected based on the strong reduction of OGT activity, since as a compensatory mechanism Ogt expression is upregulated in the Ogt__T931A/del blastocysts (Fig. S2D), making a quantification even more challenging despite a high number of stained embryos.

In sum, these in vivo experiments are difficult and require sacrificing many animals (about 20 females per CRISPR injection experiment). Because the results would bring refinement to the study but would not change our conclusions, we suggest that the cost/benefit is too high.

C8: 5. In Boulard et al. 2019 O-GlcNAcylation was shown to be sufficient to modulate expression of DNA methylation-dependent TEs. It would be interesting to know (or at least discuss) if the changes in TE expression observed in OGT-mutant embryos in this study involve changes in DNA methylation. Ideally, some DNA methylation measurement optimized for low input numbers of cells would be useful.

R8: Thank you for making the link with our previous study. In the PNAS paper, we report that targeted removal of O-GlcNAc at proteins bound to specific TEs (e.g. IAPez) causes their full-blown reactivation without detectable changes in DNA methylation, thus suggesting a role of the O-GlcNAc modification for the silencing of methylated TEs downstream or independent of DNA methylation. We agree that it would be informative to quantify DNA methylation in the T931-mutant blastocysts to test if the in vitro result is the same in vivo, but this would require performing onerous microinjection sessions as explained above.

C9: 6. The data related with the OGT-degron system in MEs seem disconnected with the rest of the manuscript. While the developmental models (blastocyst, etc) elegantly assess the contribution of O-GlcNAcylation to the control of cell survival and gene expression through the use of different OGT mutants, the degron system is a system of graded depletion that unfortunately was only possible to be used in MEFs (instead of embryos). Thus, the results obtained with the degron system in MEFs are difficult to intersect with the data from the use of OGT-mutants in embryos. Even though there are obvious interesting questions that one may want to know about this OGT degron MEF system, none of them would demonstrate a direct role for O-GlcNAcylation in cellular function, the major point addressed in the developmental system. Using the degron system in embryonic stem cells might have provided a more parallel comparison. The authors should discuss this point in more detail and either use ESC instead of MEFs or provide a stronger justification for the use of MEFs over ESC.

R9: We thank the reviewer for their clear understanding of the system. The choice of primary MEF as an in vitro model was imposed by technical limitations we encountered during the study. We fully agree that ES cells is the model of choice for preimplantation embryos; thus we initially derived ES cells and obtained only one male clone bearing the AID degron system. Upon auxin addition to the culture media, OGT's level remained unchanged in ES cells. Thus, the ES cells model was not usable. To test the AID degron in a different cell type, we then derived MEFs and showed its effectiveness (Figures 4C and S4C-E), which also allowed to collect functional data on OGT's cellular function (Figures 4D-F). We took the comment on board and clarified the rationale of studying MEFs in the revised manuscript. We agree that it remains to be verified that the OGT-dependent pathways uncovered in MEFs are relevant in the preimplantation embryo. Despite this caveat, we feel the mouse model for endogenous OGT-degron, as well as the negative results in vivo and conclusions in MEFs should be shared with the community, which could take advantage of our results to refine the system.

Minor:C10: 7. In Fig 2C the color and shape codes are confusing to understand - there are some colors/shapes that are not represented in the PCA plot. The same in Fig 3H, where in the PCA plot there are pink triangles that do not match with the code legends.

R10: We apologize for the confusion with the legends of Figures 2C and 3H, that we have made unambiguous in the revised version (as well as Figures S2B,C and S3C).

C11: 8. In the figure legends of Figures 2D, 2E, 2F, and 2H, the notation should be corrected from "OgtT931A/Y" to "OgtT931del/Y".

R11: This has been corrected; many thanks for bringing it to our attention.

Significance

C12: To investigate the function of OGT at specific developmental stages, the authors perturbed OGT's function in vivo by creating a murine allelic series featuring four single amino acid substitutions that variably reduced OGT's catalytic activity. The goal was to identify the direct effect of O-GlcNAcylation, using a sophisticated collection of genetic mutants to evaluate in vivo the role of this modification at early stages of development. Overall, the severity of embryonic lethality correlated with the extent of catalytic impairment of OGT, demonstrating that the O-GlcNAc modification is essential for early development.

R12: We thank the reviewer for their clear understanding of our work and their appreciation of the biological importance of the findings.

Reviewer #3

Evidence, reproducibility and clarity

C13: This is a conceptually interesting paper that attempts to leverage the knowledge of OGT catalysis to begin to dissect OGT function. The evidence is presented I a straightforward fashion and is in general well documented. The breeding strategies are well informed and the paper draws heavily on previous work carried out in the mouse.

R13: We greatly appreciate the overall supporting review. However, we fail to understand what they mean with "the paper draws heavily on previous work carried out in the mouse". This comment may stem from a misunderstanding because this work is not based on any previously published study. Specifically, neither the seven murine alleles presented and analyzed nor the single embryo-transcriptomic data sets on which our conclusions are based have been published elsewhere.

To put this work into context, before our study there were two seminal studies published two decades ago that reported the essential role of Ogt for mouse development, but no molecular profiling was performed (10.1073/pnas.100471497, 10.1128/mcb.24.4.1680-1690.2004). The two Ogt loss-of-function alleles studied in these papers were deemed as not suitable for interrogating molecular phenotypes because they caused cell death that confounds molecular profiling and embryonic lethality at implantation, thus preventing study of the sexually-dimorphic role of Ogt placenta. To overcome this long-standing problem, we created new seven murine alleles, which allowed us to tease apart molecular phenotypes at key stages of mouse embryonic development, focusing on the blastocyst and the placenta.

Significance

C14: The paper describes tools which will help dissect the many potential roles of O-GlcNAc addition in early development. As it stands, this is a descriptive manuscript that will lead to hypothesis generation and testing and this should not be undervalued. The biological reagents produced and characterized will be of general interest to the field. Most of the findings presented represented a verification of existing ideas in the field but this is not meant as a criticism since part of the motivation for the approach was to generate a reproducible system for analyzing the biological phenomena.

R14: We thank the reviewer for their appreciation of the importance of experimentally testing ideas shared in the field without direct evidence.

However, we must respectfully disagree with the qualification of "descriptive manuscript". This qualification may stem from the particularly difficult challenge to accessing the molecular details on how the O-GlcNAc modification exerts the biological functions we report. We are fully cognizant of the limitations of the study that we discussed in the discussion section and in R20.2. However, we feel that the adjective "descriptive" is not a fair qualification because we provide numerous novel functional evidence. Specifically, we introduce two novel orthogonal in vivo perturbations for endogenous Ogt that allowed us to interrogate for the first time its function in the developing mouse embryo. These perturbations allow us to draw causative conclusions (not descriptive) on the essential role of the O-GlcNAc modification itself for preimplantation development, its sexually-dimorphic role in the placenta and its requirement in vivo for the stable repression of retrotransposons.

C15: There are perhaps some bioinformatic shortcuts taken that may need to be corrected upon thorough review. These do not lessen the overall impact of the contribution.

R15: All the code written for the bioinformatic analyses performed in this study is publicly available: https://github.com/boulardlab/Ogt_mouse_models_Formichetti2024. The reviewer needs to specify which bioinformatic analysis they suggest could be improved.

Reviewer #4 (Evidence, reproducibility and clarity (Required)):

Summary

C16: O-GlcNAcylation is the fundamental post-translational modification of numerous nuclear and cytosolic proteins. OGT is the sole enzyme catalyzing O-GlcNAc addition onto the proteins. The essentiality of OGT for early development and cellular viability has been established by using OGT-KO mice and cell lines. However, it remains to be elucidated whether the catalytic activity of OGT is required for the early development, and if the catalytic activity of OGT is required what are the functions of OGT or O-GlcNAcylation in early development due to a lack of appropriate mouse models. To overcome the technical difficulty of manipulating the levels of O-GlcNAcylation in early embryos, Formichetti et al. created the series of four mouse models (OgtY851A, OgtT931A, OgtQ849N, and OgtH568A) with different OGT activity by introducing single amino acid substitution in the catalytic domain. By analyzing the inheritance of the hypomorphic OGT alleles and the lethality of mouse embryos, they discovered OGT activity is a critical factor for early development. Subsequently, RNA-seq analyses with two mouse models showing the maternal inheritance of the hypomorphic OGT alleles indicated that sever hypo-OGT activity altered transcription and silencing of retrotransposon in preimplantation development while mild reduction of OGT's activity affected placental development in a sexually dimorphic manner rather than preimplantation development. Furthermore, to study the function of OGT at specific developmental stages, they developed a mouse model bearing endogenously AID-tagged OGT for acute degradation of OGT. Although the degron system wasn't efficient in preimplantation embryos, they discovered quick transcriptional changes upon OGT deletion in MEFs. The quality of the manuscript is good because the question to be solved was appropriately set, the approach was well designed, and their findings were interesting, although their writing was sometimes hard to understand as I raised in my following comments. Nevertheless, there are several points to be fixed before being published.

R16: We thank the reviewer for their clear understanding of our work and their appreciation of the biological importance of the findings. Your comprehensive review of the manuscript and the questions you raised were extremely helpful in improving the manuscript and fully addressing its limitations. Below, we respond to comments in full, have revised the manuscript to improve clarity and have included novel results.

Major Comments

C17: 1. Although the authors showed in vitro activity of each mutant of OGT used in this manuscript by referencing the previous literature, they never showed the levels of global O-GlcNAcylation (and OGT itself) in their established mouse embryos. Although it could be impossible to determine O-GlcNAc levels in OgtQ849N and OgtH568A embryos because of the lack of germline transmission and founder line, respectively, they could do that in OgtY851A and OgtT931A embryos. Given that Y851A and T931A mutants had similar VMAX/KM with different VMAX, it is possible that their activity is comparable or Y851A has even lower activity in vivo depending on the concentration of UDP-GlcNAc in embryos. Therefore, it is critical to assess whether in vivo OGT activity is correlated with that in vitro as expected to conclude that severity of sub-Mendelian inheritance is proportional to the reduction of activity of OGT in vivo. Moreover, since the authors developed the elegant system to deplete OGT, the activity of Q849N and H568A mutant OGT can be examined at least in cells by expressing them in MEFs with OGT-degron system. Thus, I propose determination of global O-GlcNAc levels compensated by OGT levels by western blotting in OgtY851A, and OgtT931A embryos or MEFs with the OGT degron system re-expressing the individual four mutant OGTs. If the protein amount is insufficient for western blotting in the embryos because of the sizes of the earlier stages of embryos, I believe the author could address this by utilizing immunofluorescence as shown in Figure S5.

R17: We fully agree that this is an important point that requires revision. The only mutation for which the level of O-GlcNAc and OGT can be assessed by western blot in vivo is Y851A, the other mutations resulting in embryonic lethality before the blastocyst stage.

We have included in the revised manuscript western blot analyses of protein expression for OGT, OGA and O-GlcNAc levels in the placenta of the OgtY851A mutants (new Figures 3C,D). The new data show that OGT is upregulated at the protein level in homozygous females, in good agreement with our transcriptomic analysis. Furthermore, O-GlcNAc levels were slightly reduced in homozygous and hemizygous placentae thus showing the impact of the point mutation on global O-GlcNAc levels in the placentae. Moreover, the analysis of OGA protein level unexpectedly revealed the enrichment of a previously uncharacterized OGA fast migrating isoform in hemizygous and homozygous placentae.

We agree that it would be informative to compare O-GlcNAc levels in OgtT931A versus OgtY851A embryos. A comparison implies performing the experiment at the same developmental stage, which has to be the blastocyst stage or prior because T931A/Y embryos die around implantation. The blastocyst being made of approximately 140 cells, it would require to pool many single blastocysts to obtain the necessary protein input for western blot. We are not aware of another study performing western blot with pooled blastocysts. An additional great challenge for this experiment is the necessity to genotype and sex the blastocysts before pooling. Thus, the feasibility of this experiment is uncertain.

As an alternative, the reviewer suggests measuring O-GlcNAc levels in the degron MEFs after introduction of OGT transgenes bearing the mutation studied. This experiment would not be conclusive because of residual O-GlcNAc after OGT degradation (Figure S4E). Furthermore, the O-GlcNAc proteome is dynamic during development (as shown in the developing brain by Liu et al. https://doi.org/10.1371/journal.pone.0043724), therefore the MEFs results would have limited value to explain our results in the early embryo.

In sum, available technologies to quantify O-GlcNAc (e.g. western bot, mass spectrometry) are inadequate for low input samples as the early embryo. However, our series of hypomorphic alleles backed up with in vitro enzymology measurements brings indirect evidence to this question. Specifically, the qualitative correlation between the measured OGT activity in vitro and the developmental phenotype indicates that the resulting relative levels of O-GlcNAc are consistent with in vitro measurements.

C18.1 : 2. I didn't understand why the authors couldn't find any founder lines of the OgtH568A mutant. Was that because mosaic mice with OgtH568A mutation are lethal?

R18.1: To answer to this question, it is important to recall two key features of the biological system:

1) The mutation H568A was reported to disrupt the glycosyltransferase activity completely (10.1038/nsmb.1443). Hence, OGT-H588A is catalytic dead.

2) We performed the CRISPR-HDR targeting in the 1-cell embryo.

Based on these premises, the absence of F0 with the OgtH568A mutation (0/31) suggests that introducing this mutation causes embryonic lethality in both males and females. This hypothesis is consistent with the previously reported lethality around implementation of Ogt-null alleles (10.1128/mcb.24.4.1680-1690.2004). It is possible that the sgRNA is very efficient and results in homozygous mutations in all female zygotes injected (as we have not obtained heterozygous females bearing these mutations). High efficiency of the targeted mutagenesis in the zygote results in mutants where all or the majority of cells bear the mutation (no or low mosaicism). The high number of microinjections performed (416 embryos over the 3 injection sessions) allows us to make these claims.

C18.2 : Also, I believe there was no explanation why the OgtQ849N allele showed no maternal inheritance. Was that because Q849N possesses enough activity for sustaining mosaic embryos, but not oocytes? The authors should better explain these points in the manuscript text.

R18.2: Thanks for this comment, we agree that this maternal effect phenotype demands further explanation.

The phenotype observed suggests two possibilities: either that the oocyte cannot maturate or that the cleavage-stage embryo cannot develop with the resulting lower levels of O-GlcNAc. The cleavage-stage embryo does not transcribe a catalytically active OGT before the 8-cell stage and thus relies on the OGT protein inherited from the oocyte until this stage (https://doi.org/10.1101/2024.01.22.576677).

Thank you for this comment, we added this interpretation of the result in the text:<br /> "The lack of maternal transmission of the Q849N allele from seemingly mosaic founder females is likely explained by the reliance of the cleavage stage embryo onto the oocyte payload of OGT and O-GlcNAc modified proteins. Specifically, Ogt's exons encoding for the catalytic domains are not detectable before the 8-cell stage, while OGT full-length protein is present and thus maternally inherited (Formichetti et al, 2024)."

C19: 3. The authors serendipitously found a T931del-allele in the "WT" allele of the OgtT931A line, and suggested that T931del had milder activity loss, although the lethality of embryos was greatly mitigated. Nevertheless, transcriptome analyses in male blastocysts revealed that 120 genes' expression was changed in T931del/Y males. This raised the question about which mutant OGT has higher activity, Y851A or T931del. I think comparing the activity of Y851A and T931del mutants in MEFs with OGT-degron system is important to confirm the proportional relationship between activity and phenotypic severity.

R19: We agree that it is a limitation that the effect of the T931del mutation on OGT activity has not been biochemically characterized. However, the important point here is that our assessment of phenotypic severity based on maternal inheritance of the mutant allele and embryonic lethality is based on the point mutations for which the catalytic activity has been determined, namely Y851A, T931A, Q849N and H568A, but not T931del.

We studied the serendipitously discovered T931del mutation to obtain transcriptional insights in the blastocyst. Because the deleted residue T931 is key for the binding to the donor substrate, we can reasonably assume that this mutation affects the catalytic activity, albeit to an undetermined level.

Hence, our conclusions regarding the requirement of O-GlcNAcylation for development are unaffected by the lack of biochemical knowledge on T931del.

C20.1: 4. Regarding transcriptomes of T931del/Y, the authors found the upregulation of proteasomal activity and stress granules along with the downregulation of amino acid metabolism, mitochondrial respiration, and so on. To validate the results, the authors should perform qPCR on several up- or down-regulated genes.

R20.1 : We agree that, in principle, qPCR validation is suitable. However, this validation experiment is particularly expensive in this case because of the requirement of numerous CRISPR zygote pronuclear injection sessions.

The conclusions of the RNA-seq analysis are strongly supported by a high number of biological replicates (n=10). This high number of biological replicates was essential to obtain sufficient statistical power to quantify with a high level of confidence transcriptional changes of low magnitudes (below 2-fold change, see R5.1 and R5.2).

Therefore, the qPCR validation experiment would require to repeat the CRISPR zygote pronuclear injection sessions with the same high number of animals. This represents a major investment in experimental work and the sacrificing of about 40 animals. Importantly, the RNA-seq results presented are authoritative because of a high number of biological replicates and high number of sequencing reads per sample. Thus, we argue that qPCR validation is not essential and thus the high cost of this experiment is difficult to justify.

C20.2: In addition, according to Figure S2E, the authors pointed out that at least for genes upregulated in OgtT931A embryos, the changes were not explained by a developmentally delayed transcriptome, suggesting that upregulation of these genes was the cause of developmental delay. Therefore, I strongly encourage them to discuss in the manuscript text how up-regulated genes could contribute to developmental delay.

R20.2: Throughout the manuscript, we have been cautious to avoid establishing causal relationships between the differentially expressed genes uncovered and the developmental phenotypes (e.g. delayed development). There are two main obstacles which we believe prevent us from establishing causality with the data available. Firstly, it is not possible to disentangle differentially expressed genes and developmental delay (in other words, we have no way to tell which is the cause and which is the consequence). Secondly, O-GlcNAc modifies over 5000 proteins and the developing embryo is a particularly dynamic system; thus we cannot know whether the differentially expressed promoters are direct targets of O-GlcNAc modified proteins (or alternatively secondary effect of another molecular alteration, for example of the proteome). We discuss this limitation of the study in the discussion section.

C21: 5. Regarding the transcriptome in OgtY851A mice, Y851A/Y male mice had huge transcriptomic differences, while Y851A/Y851A female mice barely had any. Although it seems to agree with the number of Ogt alleles, I wonder whether other X-linked genes expressed higher in female placenta as shown in Figure 3C could attenuate the effects of decreased OGT activity. I don't think this possibility can be excluded, unless the authors further decrease OGT activity in Y851A/Y851A female placenta and obtain the similar results as for male placenta. Or if they compared the levels of global O-GlcNAcylation between Y851A/Y and Y851A/Y851A mouse placentas and discovered they had similar levels of O-GlcNAcylation, then the authors could conclude that the number of Ogt alleles was not the reason of sexual-dimorphism. The authors should determine the levels of O-GlcNAcylation in Y851A/Y and Y851A/Y851A mouse placentas and/or at least discuss the above possibilities in the manuscript text.

R21: Thank you for the thoughtful feedback. We agree that the most likely explanation for the higher sensitivity of males placenta as compared to females to OGT reduced activity is the difference in Ogt copy number, especially because Ogt escapes X-chromosome inactivation in the placenta (new Figure S3A).

Western blot quantification of global O-GlcNAc levels was now performed (new Figures 3C,D). We measured similar level of O-GlcNAc in Y851A/Y and Y851A/Y851A placentas (lowered than WT males in both cases), but we cannot exclude that the WB does not have the dynamic range required to detect a subtle difference. In fact, female homozygous were expected to have an intermediate level between WT males and hemizygous males, and the difference between the two male genotypes (also considering sample-to-sample variability) is already small when quantified from the blot (new Figure 3D). It is possible that a X-linked modifier attenuates the impact of hypo-O_GlcNAcylation in female mutant placenta in the case of identical O-GlcNAc levels in homozygous females and hemizygous males. Thank you for the idea that we included in the revised manuscript:

"Of note, the lower sensitivity of the homozygous females' transcriptome to Ogt disruption (Fig. 3F,I and S3B) seems difficult to reconcile with their lower O-GlcNAc level comparable (lower) O-GlcNAc level to the hemizygous males (Fig. 3C). It is possible that the western blot technique is not sensitive enough to detect subtle differences in O-GlcNAcylation. An alternative hypothesis, if O-GlcNAc levels were truly identical between Y851A/Y and Y851A/Y851A, could be the existence of a modifier in female that could be a XCI-escapee."

C22: 6. In terms of the transcriptome in OgtY851A mice, similar to comment 4, the authors should confirm their transcriptomics data shown as Figure 3D by qPCR. In addition, the authors should describe the potential mechanisms by which the differentiation of precursor cells of LaTPs and JZPs were disrupted. Were master regulators of the differentiation known to be O-GlcNAcylated and loss of O-GlcNAcylation perturbed the function?

R22: As for the whole embryo discussed in R20.2, we also interpret cautiously the gene expression phenotype observed in the placenta. Specifically, we state in the manuscript that it could either be caused by an impact of lower O-GlcNAcylation on placental differentiation or by a general delay in placentation or in the development of the embryo as a whole. The hypothesis of a general delay (of the whole embryo and/or of placental formation specifically) is supported by the downregulation of essentially all markers of more differentiated cell types and the upregulation of the precursor marker. We favor this hypothesis because it is consistent with what observed with the T931 mutants and also with the enzymatic removal of O-GlcNAc in the zygote (Formichetti et al., 2024 BioRxiv). Because of the thousands of O-GlcNAcylated proteins present in the cell, it is impossible to know which is the responsible molecular mechanism, which could even start at much earlier stages.

Minor Comments

C23: 1. Regarding DFP461-463 mutant, I couldn't understand the point of this figure because the results had no difference, and the meaning of the mutation was quite different from the others. Thus, the figure was awkward and a little confusing to me. If the authors still want to include the figures, I would suggest that they should reorganize the position of the figure (maybe after figure 3 is better to show you had tried to investigate the effects of nuclear localization of OGT on the changes of transcriptomes) and add some results. Since WT OGT seems to be localized mainly in the cytosol at steady state (Figure S1B and S1C), the effect of mutation on its nuclear localization should not be obvious. Therefore, it is difficult to conclude the mutation had no effect on the nuclear localization unless the ratio of nuclear and cytosol localization is quantified. Also, I wonder whether the O-GlcNAc levels of nuclear and cytosolic proteins in the mutant cells were comparable to those in WT cells. If this is the case, the results would also support the authors' conclusion.

R23: We took the comments on board and made it clearer that the rationale for the DFP461-463 mutant was an attempt to separate OGT's nuclear and cytosolic functions. We fully agree that these results are peripheral, and thus we presented these results in Supplementary Figure 1 (not in the main figure).

The biochemical evidence presented in Fig S1C shows that the genetic substitution of DFP to AAA on endogenous OGT has no detectable impact on its nuclear localization in primary MEFs. This result is far more authoritative than the evidence provided by Seo et al. 2016 (doi: 10.1038/srep34614), which is based on the overexpression of OGT transgenes in HeLa cells. Importantly, Seo et al. 2016 did not assess the impact of their mutations on endogenous OGT.

We believe that the negative results we obtained with the DFP461-463 mouse model shall be extremely valuable for the field. Firstly, science can move forward only if both negative and positive results are shared. In this specific case, we found that mutation of endogenous OGT in MEFs yielded to a different result than previously reported overexpression of the same mutant construct in HeLa cells. Secondly, we want to make the Ogt-NLS- mouse model available for further investigations.

C24: 2. Since OGT or O-GlcNAcylation regulates chromatin status, the authors analyzed the gene expression profiles of retrotransposons in T931del/Y or T931A/Y mice. Is it possible to investigate if the release of gene silencing is also seen in non-retrotransposon genes? I assumed retrotransposons might be a well-established system to analyze gene silencing status, however, if the authors could find similar effects on genes other than retrotransposons, that would be highly valuable.

R24: This is an interesting idea. This notion refers to the activation of promoters that are normally epigenetically repressed (e.g. silent despite the presence of all trans-active factors required for their expression). Epigenetically repressed promoters include retrotransposons, imprinted genes and germline specific genes that are normally expressed in germ cells and maintained in a repressed state in somatic cells (10.1038/s41580-019-0159-6). Testing of mono-allelic expression of imprinted genes required F1-hybrid. Thus, we assessed whether well-studied germline specific genes could be realized from silencing in T931del/Y or T931A/Y blastocyst and found no evidence for it (see dot plot below). The unbiased transcriptomic analysis presented in the manuscript shows that the product of upregulated genes are enriched in mRNA processing (Figure 2E), but these genes are not normally epigenetically repressed. Thus, contrary to retrotransposons, the role of O-GlcNAc at cellular gene promoters appears not to be linked to epigenetic silencing. This could be explained by the many different protein substrates for O-GlcNAc.

C25: 3. OgtY851A mice with milder OGT activity loss didn't exhibit impaired preimplantation development, but did display postimplantation development such as placental development, suggesting that O-GlcNAcylation of proteins required for preimplantation and postimplantation development relies on different degrees of OGT activity. I wonder whether global O-GlcNAc levels in embryos in preimplantation and postimplantation developmental stages are different or not. This might include both the pattern of blotting and intensities. The results would give the authors an explanation why the dependency on OGT activity was different in two developmental stages. Can the authors provide data? If not, then the authors should at least describe hypotheses in the manuscript to address these questions.

R25: We recently reported that the subcellular patterns of O-GlcNAc are highly dynamic during preimplantation development (Formichetti et al. 2024, BioRxiv). The most striking O-GlcNAc remodeling we observed is the enrichment of nuclear O-GlcNAc as compared to cytoplasmic O-GlcNAc that is concomitant to embryonic genome activation (Formichetti et al. 2024, BioRxiv). We quantified the ratio of the nuclear/cytoplasmic signal by immunofluorescence, but absolute quantification is not possible with this method. Due to the limited number of cells of the preimplantation embryo, this analysis cannot be performed by western blot. Hence, there is no appropriate method to quantitatively compare O-GlcNAc levels between preimplantation and postimplantation embryos.

C26: 4. The authors' AID-degron system elegantly worked in MEFs but was inefficient in preimplantation embryos. I wonder if this was because of the high expression of the shorter isoform of OGT detected as OGTp78 in the author's western blot. Is it possible to examine this possibility in the embryos? Either way, the authors should describe a potential explanation for why the efficiency in the embryos was low. In addition, the authors should describe why they inserted the AID tag only into the longest OGT isoform.

R26: This is a good point. The smallest isoform OGTp78 bears the catalytic domain and thus can partially compensate for the degradation of OGTp110. Note that the level of OGTp78 is low and does not increase upon OGTp110 degradation; thus a compensation can only be partial (Figures S4A and S4D). Alternative hypotheses for the ineffectiveness of the degron system in ex vivo grown embryos include: i) the expression level of OsTIR that may be too low in the early embryo (Rosa26 promoter not being activated at EGA), ii) a possible steric hindrance of the N-ter AID tag in these cells, iii) the lower concentration of Auxin imposed by toxicity on the embryo is likely suboptimal. Testing these possibilities is very difficult in preimplantation embryos.

It is unclear how the OGTp78 isoform is produced; it was hypothesized to originate from an alternative transcription start site (https://doi.org/10.1007/s00335-001-2108-9). We initially attempted to target both isoforms by inserting the AID tag at the C-terminus, but we were unsuccessful in producing this mouse model. It is possible that the C-terminus that is near the catalytic site cannot tolerate the AID knock-in.

C27: 5. In Figure S1C, is the band detected right below OGTp78 in nuclei fractions non-specific or do both bands correspond to OGTp78 ?

R27: To answer this question, a knockout control would be needed. OGTp78 being not targeted by our AID-degron, we cannot test the specificity of these bands using our perturbation tool kit.

C28: 6. Figure 1D top row third column: hemizgous -> hemizygous

R28: Many thanks; the embarrassing typo has been corrected.

C29: 7. Figure 1D second row third column: hemyzygous -> hemizygous

R29: Thanks for bringing this other typo to our attention, it is now corrected.

Reviewer #4 (Significance (Required)):

General assessment: strengths and limitations

C30: Strength: This manuscript elegantly revealed the requirement of OGT in mammalian development by taking advantage knock-in mouse models with different OGT activity. In addition, the manuscript provided the interesting and important transcriptomics data in both pre- and post-implantation embryos of OGT mutant mice. These data sets could explain detailed mechanisms how OGT or O-GlcNAcylation regulates mammalian development in the future. Furthermore, development of AID-tagged OGT system would be a useful tool for other researchers studying OGT function.

Limitation: Although they found interesting changes in terms transcriptomes in developing mice with different OGT activity, they lack the data showing how these changes caused the observed phenotypes. In other words, there are less mechanistic insights behind the developmental problems seen in mice with different OGT activity.

In addition, although I agree the question about whether OGT activity itself is crucial for the early development of mammals has not been completely solved for a long time, I assume people thought OGT activity is actually important for the mammalian development thorough the observation of OGT-linked congenital disorders of glycosylation.

Therefore, I would say the novelty of the manuscript is a little less impactful. Furthermore, although AID-tagged OGT system revealed fundamental questions regarding the transcriptional changes upon acute depletion of OGT in cellular levels, the system was inefficient in mouse embryos. So, they showed nothing about developmental-stage specific requirements of OGT.

Advance: The manuscript can fill a current gap regarding requirement of OGT in mammalian development. Also, the manuscript developed a series of mutant mice with different OGT activity and an AID-tagged OGT mouse line. These mice provide technical advances.

Audience: The manuscript will be interested in researchers in specific fields such as glycobiology, developmental biology, and clinical fields.

Describe your expertise: Biochemistry, Glycobiology, Cell biology

R30: We are thankful for the constructive and supportive review.

We fully agree with the limitations of the study and discussed them in the manuscript. Our in vivo approach revealed the most phenotypically relevant transcriptional phenotypes resulting from OGT catalytic impairment during embryonic development. We make the mouse models created for this study available to the community to facilitate follow-up studies aiming at exploring the underlying molecular details.

As pointed out in the comments, the requirement of OGT glycosyltransferase activity for mammalian development was widely assumed by the field, but this belief was without direct experimental evidence. This study provides the first in vivo evidence for this important conclusion.

Conclusion: The reviewers' comments were tremendously useful to improving the clarity of the manuscript and adding important new in vivo evidence. We note that none of the reviewers provided any reason to doubt our important conclusions:

• The demonstration that the enzymatic activity of Ogt, thus the O-GlcNAc modification itself, is essential for preimplantation development.
• The finding that a mild reduction of OGT's activity is sufficient to perturb the silencing of multiple families of retrotransposons in the growing embryo.

• The indication, from transcriptomes of hypo-O-GlcNAcylated embryos, of a developmental retardation upon a mild O-GlcNAc perturbation.

• The discovery that OGT's rapid depletion in vitro downregulates basal cellular function, including translation. This result provides mechanistic support to the embryonic growth delay resulting from decreasing O-GlcNAc in vivo.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

Referee #2

Evidence, reproducibility and clarity

Comments to authors

To investigate the function of OGT at specific developmental stages, the authors perturbed OGT's function in vivo by creating a murine allelic series featuring four single amino acid substitutions that variably reduced OGT's catalytic activity. The goal was to identify the direct effect of O-GlcNAcylation, using a sophisticated collection of genetic mutants to evaluate in vivo the role of this modification at early stages of development. Overall, the severity of embryonic lethality correlated with the extent of catalytic impairment of OGT, demonstrating that the O-GlcNAc modification is essential for early development.<br /> The study represents a substantial advance in our understanding of OGT and O-GlcNAcylation in mammalian development. The creation of novel murine models and inducible systems is an important contribution, providing powerful tools for future research in this field. The insights into the role of OGT's catalytic activity and its involvement in epigenetic regulation during embryonic development are noteworthy, opening new avenues for research. However, there are a few considerations and concerns:

Major:

1. An assumption of the study is that different mutations cause different levels of O-GlcNAcylation rather than alterations in substrate specificity. It might be important to test, at least in cultured cells, that the different mutations do not change the preference of OGT to modify certain proteins rather than others, which can provide alternative explanations for their findings.
2. In Fig 1D and 1H, the thresholds to define a gene or TE as differentially expressed are not strong. According to the figure legends, "any" change in terms of log2Fc was considered as DE and colored. I think the figures should illustrate better that the changes are subtle, by for example adding a dotted line (at least) in the value 0.5 of the y-axis. These subtle transcriptional changes should be reflected better in certain paragraphs where the expression of TEs are presented/and discussed as a hallmark of the absence of O-GlcNAcylation in the OGT-mutants. The same happens with Suppl Fig 3C (changes are very minor). Similarly, in Fig2C, the changes in gene expression are lower than log2FC 1 (which represent the double in absolute expression). Applying a stronger threshold, among the upregulated genes, only Xist will be significantly overexpressed. If a gentle threshold needs to be applied to this data, authors should at least justify the reasons behind doing so. Same for Fig2D.
3. In Figure 2B, the T931del allele was recovered in the blastocyst population with a very high frequency, even higher than the male WT group (T931del: 10; WT: 3). This observation suggests that the T931del allele did not significantly affect blastocyst survival. Further clarification or additional experiments might be necessary to understand the implications of this finding on early developmental stages.
4. Similarly, in Figure 2G, there is an apparent higher expression of TE expression in the T931A/Y embryos group than in the T931del/Y group, which combined with the higher frequency of blastocyst generated in this latest group it may indicate a deeper molecular consequence after the deletion of the T931. A comparison of the transcriptome between these two cell lines help to address this possibility. Also, the authors should compare the O-GlcNAc levels of WT, T931A, and T931del mutant blastocysts by immunostaining, similar to what was done in Figure S5F.
5. In Boulard et al. 2019 O-GlcNAcylation was shown to be sufficient to modulate expression of DNA methylation-dependent TEs. It would be interesting to know (or at least discuss) if the changes in TE expression observed in OGT-mutant embryos in this study involve changes in DNA methylation. Ideally, some DNA methylation measurement optimized for low input numbers of cells would be useful.
6. The data related with the OGT-degron system in MEs seem disconnected with the rest of the manuscript. While the developmental models (blastocyst, etc) elegantly assess the contribution of O-GlcNAcylation to the control of cell survival and gene expression through the use of different OGT mutants, the degron system is a system of graded depletion that unfortunately was only possible to be used in MEFs (instead of embryos). Thus, the results obtained with the degron system in MEFs are difficult to intersect with the data from the use of OGT-mutants in embryos. Even though there are obvious interesting questions that one may want to know about this OGT degron MEF system, none of them would demonstrate a direct role for O-GlcNAcylation in cellular function, the major point addressed in the developmental system. Using the degron system in embryonic stem cells might have provided a more parallel comparison. The authors should discuss this point in more detail and either use ESC instead of MEFs or provide a stronger justification for the use of MEFs over ESC.

Minor:

1. In Fig 2C the color and shape codes are confusing to understand - there are some colors/shapes that are not represented in the PCA plot. The same in Fig 3H, where in the PCA plot there are pink triangles that do not match with the code legends.
2. In the figure legends of Figures 2D, 2E, 2F, and 2H, the notation should be corrected from "OgtT931A/Y" to "OgtT931del/Y".

Significance

To investigate the function of OGT at specific developmental stages, the authors perturbed OGT's function in vivo by creating a murine allelic series featuring four single amino acid substitutions that variably reduced OGT's catalytic activity. The goal was to identify the direct effect of O-GlcNAcylation, using a sophisticated collection of genetic mutants to evaluate in vivo the role of this modification at early stages of development. Overall, the severity of embryonic lethality correlated with the extent of catalytic impairment of OGT, demonstrating that the O-GlcNAc modification is essential for early development.

use higher up in product description

Reviewer #2 (Public Review):

Pyoverdines, siderophores produced by many Pseudomonads, are one of the most diverse groups of specialized metabolites and frequently used as model systems. Thousands of Pseudomonas genomes are available, but large scale analyses of pyoverdines are hampered by the biosynthetic gene clusters (BGCs) being spread across multiple genomic loci and existing tools' inability to accurately predict amino acid substrates of the biosynthetic adenylation (A) domains. The authors present a bioinformatics pipeline that identifies pyoverdine BGCs and predicts the A domain substrates with high accuracy. They tackled a second challenging problem by developing an algorithm to differentiate between outer membrane receptor selectivity for pyoverdines versus other siderophores and substrates. The authors applied their dataset to thousands of Pseudomonas strains, producing the first comprehensive overview of pyoverdines and their receptors and predicting many new structural variants.

The A domain substrate prediction is impressive, including the correction of entries in the MIBiG database. Their high accuracy came from a relatively small training dataset of A domains from 13 pyoverdine BGCs. The authors acknowledge that this small dataset does not include all substrates, and correctly point out that new sequence/structure pairs can be added to the training set to refine the prediction algorithm. The workflow unfortunately cannot differentiate between different variants of Asp and OHOrn. To validate their predictions, they elucidated structures of several new pyoverdines, and their predictions performed well. The authors tested their workflow on Burkholderiales A domains and had good results, suggesting it can be used on other taxa. Skimming through the source code and data, the algorithm itself appears to be sound and a clear improvement over existing tools for pyoverdine BGC annotation.

Predicting outer membrane receptor specificity is likewise a challenging problem and the authors have made a promising achievement by finding specific gene regions that differentiate the pyoverdine receptor FpvA from FpvB and other receptor families. Their predictions were not tested experimentally, but the finding that only predicted FpvA receptors were proximate to the biosynthesis genes lends credence to the predictive power of the workflow. The authors find predicted pyoverdine receptors across an impressive 468 genera, an exciting finding for expanding the role of pyoverdines as public goods beyond Pseudomonas. However, whether or not these receptors can actually recognize pyoverdines (and if so, which structures!) remains to be investigated.

In all, the authors have assembled a rich dataset that will enable large scale comparative genomic analyses. This dataset could be used by a variety of researchers, including those studying natural product evolution, public good eco/evo dynamics, and NRPS engineering.

Author response:

The following is the authors’ response to the original reviews.

Public Review: 

This study used ATAC-Seq to characterize chromatin accessibility during stages of GABAergic neuron development in induced pluripotent stem cells (iPSCs) derived from both Dravet Syndrome (DS) patients and healthy donors. The authors report accelerated GABAergic maturation to a point, followed by further differentiation into a perturbed chromatin profile, in the cells from patients. In a preliminary analysis, valproic acid, an anti-seizure medication commonly used in patients with DS, increased open chromatin in both patient and control iPSCs in a nonspecific manner, and to different degrees in cultures derived from different patients. These findings provide new information about DS-associated changes in chromatin, and provide further evidence for developmental abnormalities in interneurons with DS. 

Strengths:

This is a novel study that aims to investigate the epigenetic changes that occur in a sodium channel model of epilepsy; these changes are often ignored but may be an interesting area for future therapeutics. In general, the flow of the paper is good, and the figures are well-designed.  Reply: Thank you for your positive feedback about our work. 

Weaknesses:

The most substantial weakness relates to the observation that DS is often viewed as a monogenic form of epilepsy. It is directly linked to SCN1A gene haploinsufficiency (Yu et al, 2006; Ogiwara et al, 2007). The gene product is Nav1.1, the alpha subunit of voltage-gated sodium channel type I that regulates neuronal excitability. Yet, analysis was conducted at time points of GABAergic interneuron differentiation in which SCN1A is likely not expressed. The paper would be strengthened if SCN1A expression and Nav1.1 protein were examined across the experimental time course. If SCN1A is not yet expressed, this would complicate any explanation of how the observed epigenetic changes might arise. It also seems counterintuitive that the absence of a sodium channel can accelerate differentiation, when, a priori, one might expect the opposite (a 'less neuronal' signal). 

Thanks, this is an important point!  In our revised manuscript, we have incorporated data on the expression of SCN1A at d19 and d65 of GABAergic development in both the control and patient groups. We first retrieved data from our previous RNA-Seq analysis, showing SCN1A gene expression in our cells at both d19 and d65. We have now updated our text on the SCN1A gene expression in the revised manuscript (Revised Supplementary Figure 1A, revised text Line 108-109). Second, we confirmed the dynamics of SCN1A expression by real-time quantitative RT/PCR analysis at four time-pionts of GABAergic development (d0, d19, d35 and d65). Notably, expression of SCN1A was detected by qRT-PCR from d19 and the expression increased with differentiation. We have now included this information in the revised manuscript (Revised Supplementary Figure 1B, revised text Line 112). 

Related to this, another important limitation of the study is that the controls are cells derived from healthy individuals and not from isogenic lines. The usage of isogenic lines is extremely relevant for every study in which iPSC-derived somatic cells are used to model a disease, but specifically in diseases like DS, in which the genetic background has an ascertained impact on disease phenotype (Cetica et al, 2017 and others). This serious limitation should be considered.

Yes, we fully agree that isogenic and edited patient-derived iPSC would have been the ideal controls. At an early stage we therefore invested considerable time and efforts in order to generate isogenic lines from patientderived iPSC. However, editing of the SCN1A variants in patient-derived iPSC turned out unsuccessful after several trials and modifications so we finally turned to iPSC from healthy donors. This is now discussed together with other limitations of our study in the revised manuscript (end of discussion section, lines 499-506).

In addition, the authors should provide data on variability across cell lines and differentiations to help convince the reader that the results can be attributed to genetic defects, rather than variability across individuals. 

This is a valuable point. In the revised manuscript, we have now added plots and IF staining from individual samples to give the readers a complete picture on how they are distributed (Revised Supplementary Figure 1C, Revised Supplementary Figure 2, and Revised Supplementary Figure 4).

In the revised manuscript, we incorporated an explanation on the strategy used to compare the two groups (cases vs. controls) in more detail. In our analysis, we first compared the dynamic changes of chromatin accessibility cell line by cell line across differentiation. We then extracted the common changes from different cell lines at each time point (Revised text line 152-155, line 226-228). Using this strategy, we extracted the common changes confined to the control and patient groups, respectively. With this approach we avoid to capture the variability across individuals.

Additionally, the authors acknowledge the variability of the differentiations and cell lines, which is commendable, and they attribute this to "possibly reflecting cell line specific and endogenous differences reported previously", but could also have to do with cell death. This is a large confounding factor for ATAC-seq. Certainly, Sup Fig 1C shows lower FrIP scores, consistent with cell death, and there seems to be a lot of death in the representative images. Moreover, the iGABA neurons are very difficult to keep alive, especially to 65 days, without co-culturing with glia and/or glutamatergic neurons. The authors should comment on how much these factors may have influenced their results. 

With this point in mind, we re-examined QC of our ATAC-Seq across all samples: As shown in revised

Supplementary Figure 2C and Supplementary Figure 4C, our cutoff for FRiP is 15%, and all of samples have an FrIP of more than 15%. At the later time points (d35 or d65), we did not observe a FRiP <15%. We therefore feel confident that the quality of ATAC-Seq is good enough for downstream analysis and data interpretation.  

Regarding the differentiation protocol, we are following a directed protocol of iPSC towards interneurons. The protocol is described in detail by Maroof et al (reference 34) and slightly modified in our lab (described in reference 13). With our modified protocol, GABAergic cells are viable beyond day 65 without the need of co-cultures with astrocyte or microglia. This is also reflected by the electrophysiological activity of interneurons at d65 and at later time points (reference 13). Additionally, our ambition was to obtain a homogeneous cell population for further analysis. Adding other cell types to the cultures would have interfered with downstream processes and a need for cell sorting. Using our protocol, we obtain viable GABA interneurons after up to 100 days in culture. To assess the viability of our cells at the point of sampling (other than by morphological assessment), we used Trypan blue staining and an automated cell counter. Only samples with a viability >90% were processed for ATAC seq. which is a commonly used cut-off for cell viability. We have now modified the method section in the revised version to describe the GABAergic differentiation and sampling (line 519-529).

Finally, changes in gene expression are only inferred, as no RNA levels were measured. If RNA-seq was not possible it would have been good to see at least some of the key genes/findings corroborated with RNA/protein levels vs chromatin accessibility alone, particularly given that these molecular readouts do not always correlate. 

In our revised manuscript, we include our recently published RNA-seq performed at d19 and d65. We also correlated the RNAseq and ATACseq data obtained from the same samples.  The Pearson correlations between gene expression and chromatin accessibility were within the range 0.49-0.57 (Revised Supplementary Figure 2G, Revised supplementary Figure 4G), which is acceptable according to standard criteria. The results confirmed that the quality of ATAC-Seq is good enough for analysis of expression levels and chromatin openness in key genes. We also added gene expression levels from RNA-seq (d19 and d65) in our revised manuscript (Revised Figure 1G, Revised Figure 2G). Finally, we performed qRT-PCR analysis of key genes in each cluster and the results are now included in the revised version (Revised Supplementary Figure 3E, Revised Supplementary Figure 5E)

Additional Points:

(1) Representative images for cell-identity markers for only D65 are shown, and not D0, D19, and D35 though it is stated in the text that this was performed. At a minimum, these representative images should be shown for all lines. 

As suggested, we have now added images for cell identity markers of all iPSC lines in the revised version (Revised Supplementary Figure 1C).

(2) What QC was performed on iPSC lines, i.e. karyotype/CNV analysis and confirmation of genotypes?

All iPSC lines used in this study have been fully characterized according to standard and state-of-the art procedures: Expression of pluripotency and stemness genes has been shown by immunostaining, flow cytometry and scorecard analysis; integrity of the genome has been assessed by karyotyping using g-banding; differentiation capacity was characterized using an embryoid body assay in combination with scorecard analysis; and genotypes were verified by Sanger sequencing. Please, see the following publications for full datasets: Schuster et all, Neurobiol Dis 2019, Schuster et al Stem Cell Res 2019, Sobol et al Stem Cells and Development 2015. In our lab, the integrity of iPSC lines are routinely verified using flow cytometry (expression for TRA-1-60 and SSEA4), immunostaining (expression of NANOG, SOX2 and OCT4), Sanger sequencing (targeting variants in SCN1A gene), cell morphology analysis and analysis of mycoplasma by MycoAlert® (Lonza).

(3) Were all experiments performed on a single differentiation? Or multiples? Were the differentiations performed with the same type? If not, was batch considered in the analysis? 

Thank you for raising this question. The text Material and Methods has been modified as follows, to better describe the differentiation and sampling procedure:

“GABAergic interneuron differentiation from iPSCs was performed as previously described (reference 13). The protocol utilizes DUAL SMAD inhibition to induce neurogenesis towards neural stem cells for 10 days, followed by patterning with high levels of sonic hedgehog for nine days towards cortically fated neuronal progenitor cells (NPC) and subsequent maturation for 46 days, i.e. a total of 65 days (Figure 1A). Neuronal cells at day 65 and onwards are healthy and viable as judged by morphological assessment by light microscopy. Differentiation was performed at least 3 times per cell line.  

Cell cultures were sampled at days 0 (D0), D19, D35 and D65, respectively, by harvesting cells with TryplE and centrifugation (300 x g, 3 min). Harvested cells were counted and assessed for viability using trypan blue staining and an automated EVE cell counter (Nano Entek). Samples with a viability of >90% were chosen for ATAC-Seq library preparation (see below).”.  

I also assume that technical replicates were merged, and then all three biological replicates were kept for each analysis and outliers were not removed, e.g. Control_D19_8F seems like an example of an outlier. 

This is a valuable point. We agree on that there is variability across three health donors and patients, respevtively, but the quality of ATAC-Seq is good after multiple assessment of QC (Revised Supplementary Figure 2B-D). The color code in Supplementary Figure 1C may be mis-leading as the Pearsson correlation of all samples was displayed. Overall, the correlation from all ATAC-seq among replicates are over 0.8. At the same time, we observed that samples at d0 are clustered together, but not at the later time points. We interpret this as related to the cell-line specific plasticity of chromatin dynamic during differentiation. The observation agrees with our results from PCA (Revised Supplementary Figure 2F).  

(4) In Figure 1C, it is intriguing that the ATACseq signal gets stronger in imN. One might expect it to be strongest in the iPSCs which are undifferentiated and have the highest levels of open chromatin. Is this a function of sequencing depth, or are all the Y-axes normalized across all time points? 

This is another valuable point. Figure 1C present the average chromatin openness for clusters specific regions- not of chromatin openness from the entire genome, which is a reason for why the chromatin openness at

D35 is higher than at other time-points. The genome-wide chromatin openness is presented in revised

Supplementary Figure 2D and we have now updated the figure legend to avoid any potential misunderstanding. 

The sequencing depth for each sample is extracted in a similar range. To give the readers a complete picture, we also present the depth of sequencing reads for each sample (Revised Supplementary Figure 2A and Revised Supplementary Figure 4A). The Y-axes of genome browser tracks were normalized, and we added the normalized value in the figures. 

(5) In Figure 1F, are these all enriched terms, or were they prioritized somehow? 

Yes, the enriched terms are prioritized based on biological meanings, and we have now clarified this in the updated legend of the manuscript. In addition, all enriched terms are now included in revised Supplementary Table 2 and Supplementary Table 4. 

(6) In Figure 1G (also the same plots in Fig 2/3), are all these images normalized i.e. there is no scale bar for each track, and do they represent and aggregate BAM/bigwig?

Yes, the genome browser tracks were normalized and we have now revised the figures by adding scale bars.

It would be good to show in supplement the variability across cell lines/diffs - particularly given the variability in the heatmap/PCA - and demonstrate the rigor/reproducibility of these results. This comment applies to all these plots across the 3 figures, particularly as in some instances the samples appear to cluster by individual first and then time point (Sup Fig 3B). 

Thanks. We have now revised the figure with plots showing individual samples. 

How confident are the authors that these effects are driven by genotype and not a single cell line? In the Fig 3D representation of NANOG, it is very difficult to see any difference between patient and control. 

In Figure 3D, we showed common chromatin dynamics in the control and patient groups. To avoid any misunderstanding, we have now updated our legend in the revised manuscript. 

(7) For the changes in occupancy annotation (UTR/exon/intron etc), are these differences still significant after correcting for variability from cell line to cell line at each time point? I.e. rather than average across all three samples, what is the range?  Reply: Revised accordingly. 

(8) The VPA timepoint is not well-justified. Given that VPA would be administered in patients with fully mature inhibitory neurons, it is difficult to determine the biological relevance. I appreciate that this is a limitation of the model, but this should at least be addressed in the manuscript. 

We agree on that our model system of GABAergic interneuron development has limitations and that cells may not fully recapitulate the development and physiology in vivo. Obvious factors to consider in our system are the directed protocol to enrich for GABAergic interneurons and the differentiation time-line restricted to 65d. This is now discussed (lines 499-506).

Recommendations for the authors:

(1) The term 'mutation' has been replaced with the term ' pathogenic variant' or likely pathogenic variant depending on the context, please see PMID: 25741868 

Thank you for pointing this out. We have replaced all instances of “mutation” with “pathogenic variant” throughout the manuscript.

(2) It is unclear what the nomenclature for sample labelling is in Supplementary Figure 1, e.g. 7C, 8F, 1B.  

We apologize for this confusion. There are cell lines names. We labeled all data and images according to cell line name, i.e. control lines: Ctl1B, Ctl7C and Ctl8F; patient lines: DD1C, DD4A, DD5A. To avoid any potential confusion, we have added a note in the revised legend of Supplementary Figure 1B.

(3) Can the authors confirm that the Deseq2 FDR values are Benjamini-Hochberg procedure corrected per default settings? If so, this should ideally be added to methods or legend for clarity 

Yes, default settings were used in Deseq2 FDR values, which is added in the method part of revised manuscript. 

(4) While it makes sense that the authors present the data in the order of Figure 1, and Figure 2, this actually makes it quite difficult to compare the two datasets, especially for the functional enrichment in the "F" figures. It may be helpful to consider re-organizing the figure order. For instance, for the long-term potentiation signal in the DS-iPSCs, what does this mean in terms of biological relevance? Or maybe Figure 2 needs to be supplementary given that Figure 3 is a more direct comparison.  

Thank you for the suggestions. We attempted to reorganize during our revision. We still believe it is easier for the audience to grasp the main message if we organize it according to our current workflow—first presenting an individual differential landscape for controls and patients, and then comparing the common and unique aspects among them.

A version of this preprint has been published in the Open Access journal GigaScience (see paper https://doi.org/10.1093/gigascience/giae040), where the paper and peer reviews are published openly under a CC-BY 4.0 license. These peer reviews were as follows:

Reviewer 3: Luca Beltrame

Lavikka and coworkers present an interesting visualization framework and associated application for genomics visualization. The challenges outlined by the authors in finding appropriate visualization tools for large-scale genomics data were also experienced by this reviewer, and thus better and improved tools are always welcome.

The manuscript is well laid out, presenting the key facts in a proper manner. The use of GPU rendering for graphs is an excellent move, and I expect to be extremely useful even for machines with lower-end GPUs. The code looks reasonably written and commented (being an application, this too is important for a review). I have also tested the examples, and indeed the software is very useful (the documentation should, however, point out that some issues regarding saving the canvas still exist). One may argue that the use of JSON for the graph grammar can be awkward, but at the same time other file formats may be more problematic and/or require specialized parsers (which open yet another can of worms).

Documentation is also logically organized. As a minor suggestion, the authors may want to add some form of search to their documentation page.

There are is an open questions that the authors may want to answer: they explicitly mention GISTIC 1.0 for the G-score plots. Is there a specific reason why they chose 1.0? The 2.0 algorithm is far more robust and produces more reliable results.

I asked Claude: Can you expand on what's happening here?

Claude said:

Certainly! This code demonstrates two important Python features: argument unpacking with the * and ** operators. Let's break it down:

First, we define a function show_spread that takes three parameters: left, middle, and right.

Then, we have two examples of calling this function using different forms of argument unpacking:

a) List unpacking with *: python all_in_list = [1, 2, 3] show_spread(*all_in_list) Here, *all_in_list unpacks the list into individual arguments. It's equivalent to calling show_spread(1, 2, 3).

b) Dictionary unpacking with **: python all_in_dict = {"right": 30, "left": 10, "middle": 20} show_spread(**all_in_dict) Here, **all_in_dict unpacks the dictionary into keyword arguments. It's equivalent to calling show_spread(left=10, middle=20, right=30).

The key points are:

• unpacks an iterable (like a list or tuple) into positional arguments.
• unpacks a dictionary into keyword arguments. When using **, the keys in the dictionary must match the parameter names in the function.

This technique is powerful for making function calls more flexible and for passing collections of values as separate arguments.

The song's criticism on mass media is mainly related to sensationalism.

"Good" things are usually not sensational. They do not demand attention, hence why the code of known/unknown based on selectors for attention filters it out.

Reference Hans-Georg Moeller's explanations of Luhmann's mass media theory based on functionally differentiated systems theory.

Can also compare to Simone Weil's thoughts on collectives and opinion; organizations (thus most part of mass media) should not be allowed to form opinions as this is an act of the intellect, only residing in the individual. Opinion of any form meant to spread lies or parts of the truth rather than the whole truth should be disallowed according to her because truth is a foundational, even the most sacred, need for the soul.

People must be protected against misinformation.

This isn't code, so it shouldn't be in monospace. Underline or italicize it if you want to set it apart, or put it in quotes.

This isn't code, so it shouldn't be in monospace. Underline or italicize it if you want to set it apart, or put it in quotes.

Agreed. Lepiter creates a pretty joyful experience for what I would call "story driven development" and/or data narratives. It makes pretty fluent the "story/argument first driven" workflow that we already had with Grafoscopio and that is pretty usual in social sciences and humanities, in contrast with the "test first driven" approach of development cultures.

In my experience it is easier to introduce non developers to this kind of mindset, as we don't fight against some established tradition. Of course, different traditions can prioritize the starting/main point in differently.

Being able to refer from the code inside a document to the document that contains it, was also a need we felt with Grafoscopio. For that, We used thisNotebook, as can be seen in our republication of the Data Journalism Handbook (in Spanish) or in this screenshot from its repository:

In our case, because my approach was not to convert the document inside Pharo, but leveraging Pandoc, thisNotebook allows us to provide PDF export options on the Markdown version of the notebook, to produce high quality PDFs besides the HTML export. As the screenshot above shows, the options where stored inside the document itself, internalizing what would be an external shell command, increasing document reproducibility also in the publication front (what added to the data reproducibility front). This was years before similar approaches like the Jupyter Book or Quarto and I still think that Pharo based tools can have leaner reproducible documentation workflows that their counterparts in other languages.

Having support for similar ideas later in Lepiter, implemented by more experienced programmers, in the form thisSnippet and being able to compose it with page and database has been a real time saver in migrating some lessons from Grafoscopio to Lepiter.

Reviewer #1 (Public Review):

Summary:

The authors report a study aimed at understanding the brain's representations of viewed actions, with a particular aim to distinguish regions that encode observed body movements, from those that encode the effects of actions on objects. They adopt a cross-decoding multivariate fMRI approach, scanning adult observers who viewed full-cue actions, pantomimes of those actions, minimal skeletal depictions of those actions, and abstract animations that captured analogous effects to those actions. Decoding across different pairs of these actions allowed the authors to pull out the contributions of different action features in a given region's representation. The main hypothesis, which was largely confirmed, was that the superior parietal lobe (SPL) more strongly encodes movements of the body, whereas the anterior inferior parietal lobe (aIPL) codes for action effects of outcomes. Specifically, region of interest analyses showed dissociations in the successful cross-decoding of action category across full-cue and skeletal or abstract depictions. Their analyses also highlight the importance of the lateral occipito-temporal cortex (LOTC) in coding action effects. They also find some preliminary evidence about the organisation of action kinds in the regions examined.

Strengths:

The paper is well-written, and it addresses a topic of emerging interest where social vision and intuitive physics intersect. The use of cross-decoding to examine actions and their effects across four different stimulus formats is a strength of the study. Likewise, the a priori identification of regions of interest (supplemented by additional full-brain analyses) is a strength.

Weaknesses:

I found that the main limitation of the article was in the underpinning theoretical reasoning. The authors appeal to the idea of "action effect structures (AES)", as an abstract representation of the consequences of an action that does not specify (as I understand it) the exact means by which that effect is caused, nor the specific objects involved. This concept has some face validity, but it is not developed very fully in the paper, rather simply asserted. The authors make the claim that "The identification of action effect structure representations in aIPL has implications for theories of action understanding" but it would have been nice to hear more about what those theoretical implications are. More generally, I was not very clear on the direction of the claim here. Is there independent evidence for AES (if so, what is it?) and this study tests the following prediction, that AES should be associated with a specific brain region that does not also code other action properties such as body movements? Or, is the idea that this finding -- that there is a brain region that is sensitive to outcomes more than movements -- is the key new evidence for AES?

On a more specific but still important point, I was not always clear that the significant, but numerically rather small, decoding effects are sufficient to support strong claims about what is encoded or represented in a region. This concern of course applies to many multivariate decoding neuroimaging studies. In this instance, I wondered specifically whether the decoding effects necessarily reflected fully five-way distinction amongst the action kinds, or instead (for example) a significantly different pattern evoked by one action compared to all of the other four (which in turn might be similar). This concern is partly increased by the confusion matrices that are presented in the supplementary materials, which don't necessarily convey a strong classification amongst action kinds. The cluster analyses are interesting and appear to be somewhat regular over the different regions, which helps. However: it is hard to assess these findings statistically, and it may be that similar clusters would be found in early visual areas too.

Reviewer #2 (Public Review):

Summary:

This study uses an elegant design, using cross-decoding of multivariate fMRI patterns across different types of stimuli, to convincingly show a functional dissociation between two sub-regions of the parietal cortex, the anterior inferior parietal lobe (aIPL) and superior parietal lobe (SPL) in visually processing actions. Specifically, aIPL is found to be sensitive to the causal effects of observed actions (e.g. whether an action causes an object to compress or to break into two parts), and SPL to the motion patterns of the body in executing those actions.

To show this, the authors assess how well linear classifiers trained to distinguish fMRI patterns of response to actions in one stimulus type can generalize to another stimulus type. They choose stimulus types that abstract away specific dimensions of interest. To reveal sensitivity to the causal effects of actions, regardless of low-level details or motion patterns, they use abstract animations that depict a particular kind of object manipulation: e.g. breaking, hitting, or squashing an object. To reveal sensitivity to motion patterns, independently of causal effects on objects, they use point-light displays (PLDs) of figures performing the same actions. Finally, full videos of actors performing actions are used as the stimuli providing the most complete, and naturalistic information. Pantomime videos, with actors mimicking the execution of an action without visible objects, are used as an intermediate condition providing more cues than PLDs but less than real action videos (e.g. the hands are visible, unlike in PLDs, but the object is absent and has to be inferred). By training classifiers on animations, and testing their generalization to full-action videos, the classifiers' sensitivity to the causal effect of actions, independently of visual appearance, can be assessed. By training them on PLDs and testing them on videos, their sensitivity to motion patterns, independent of the causal effect of actions, can be assessed, as PLDs contain no information about an action's effect on objects.

These analyses reveal that aIPL can generalize between animations and videos, indicating that it is sensitive to action effects. Conversely, SPL is found to generalize between PLDs and videos, showing that it is more sensitive to motion patterns. A searchlight analysis confirms this pattern of results, particularly showing that action-animation decoding is specific to right aIPL, and revealing an additional cluster in LOTC, which is included in subsequent analyses. Action-PLD decoding is more widespread across the whole action observation network.

This study provides a valuable contribution to the understanding of functional specialization in the action observation network. It uses an original and robust experimental design to provide convincing evidence that understanding the causal effects of actions is a meaningful component of visual action processing and that it is specifically localized in aIPL and LOTC.

Strengths:

The authors cleverly managed to isolate specific aspects of real-world actions (causal effects, motion patterns) in an elegant experimental design, and by testing generalization across different stimulus types rather than within-category decoding performance, they show results that are convincing and readily interpretable. Moreover, they clearly took great care to eliminate potential confounds in their experimental design (for example, by carefully ordering scanning sessions by increasing realism, such that the participants could not associate animation with the corresponding real-world action), and to increase stimulus diversity for different stimulus types. They also carefully examine their own analysis pipeline, and transparently expose it to the reader (for example, by showing asymmetries across decoding directions in Figure S3). Overall, this is an extremely careful and robust paper.

Weaknesses:

I list several ways in which the paper could be improved below. More than 'weaknesses', these are either ambiguities in the exact claims made, or points that could be strengthened by additional analyses. I don't believe any of the claims or analyses presented in the paper show any strong weaknesses, problematic confounds, or anything that requires revising the claims substantially.

(1) Functional specialization claims: throughout the paper, it is not clear what the exact claims of functional specialization are. While, as can be seen in Figure 3A, the difference between action-animation cross-decoding is significantly higher in aIPL, decoding performance is also above chance in right SPL, although this is not a strong effect. More importantly, action-PLD cross-decoding is robustly above chance in both right and left aIPL, implying that this region is sensitive to motion patterns as well as causal effects. I am not questioning that the difference between the two ROIs exists - that is very convincingly shown. But sentences such as "distinct neural systems for the processing of observed body movements in SPL and the effect they induce in aIPL" (lines 111-112, Introduction) and "aIPL encodes abstract representations of action effect structures independently of motion and object identity" (lines 127-128, Introduction) do not seem fully justified when action-PLD cross-decoding is overall stronger than action-animation cross-decoding in aIPL. Is the claim, then, that in addition to being sensitive to motion patterns, aIPL contains a neural code for abstracted causal effects, e.g. involving a separate neural subpopulation or a different coding scheme? Moreover, if sensitivity to motion patterns is not specific to SPL, but can be found in a broad network of areas (including aIPL itself), can it really be claimed that this area plays a specific role, similar to the specific role of aIPL in encoding causal effects? There is indeed, as can be seen in Figure 3A, a difference between action-PLD decoding in SPL and aIPL, but based on the searchlight map shown in Figure 3B I would guess that a similar difference would be found by comparing aIPL to several other regions. The authors should clarify these ambiguities.

(2) Causal effect information in PLDs: the reasoning behind the use of PLD stimuli is to have a condition that isolates motion patterns from the causal effects of actions. However, it is not clear whether PLDs really contain as little information about action effects as claimed. Cross-decoding between animations and PLDs is significant in both aIPL and LOTC, as shown in Figure 4. This indicates that PLDs do contain some information about action effects. This could also be tested behaviorally by asking participants to assign PLDs to the correct action category. In general, disentangling the roles of motion patterns and implied causal effects in driving action-PLD cross-decoding (which is the main dependent variable in the paper) would strengthen the paper's message. For example, it is possible that the strong action-PLD cross-decoding observed in aIPL relies on a substantially different encoding from, say, SPL, an encoding that perhaps reflects causal effects more than motion patterns. One way to exploratively assess this would be to integrate the clustering analysis shown in Figure S1 with a more complete picture, including animation-PLD and action-PLD decoding in aIPL.

(3) Nature of the motion representations: it is not clear what the nature of the putatively motion-driven representation driving action-PLD cross-decoding is. While, as you note in the Introduction, other regions such as the superior temporal sulcus have been extensively studied, with the understanding that they are part of a feedforward network of areas analyzing increasingly complex motion patterns (e.g. Riese & Poggio, Nature Reviews Neuroscience 2003), it doesn't seem like the way in which SPL represents these stimuli are similarly well-understood. While the action-PLD cross-decoding shown here is a convincing additional piece of evidence for a motion-based representation in SPL, an interesting additional analysis would be to compare, for example, RDMs of different actions in this region with explicit computational models. These could be, for example, classic motion energy models inspired by the response characteristics of regions such as V5/MT, which have been shown to predict cortical responses and psychophysical performance both for natural videos (e.g. Nishimoto et al., Current Biology 2011) and PLDs (Casile & Giese Journal of Vision 2005). A similar cross-decoding analysis between videos and PLDs as that conducted on the fMRI patterns could be done on these models' features, obtaining RDMs that could directly be compared with those from SPL. This would be a very informative analysis that could enrich our knowledge of a relatively unexplored region in action recognition. Please note, however, that action recognition is not my field of expertise, so it is possible that there are practical difficulties in conducting such an analysis that I am not aware of. In this case, I kindly ask the authors to explain what these difficulties could be.

(4) Clustering analysis: I found the clustering analysis shown in Figure S1 very clever and informative. However, there are two things that I think the authors should clarify. First, it's not clear whether the three categories of object change were inferred post-hoc from the data or determined beforehand. It is completely fine if these were just inferred post-hoc, I just believe this ambiguity should be clarified explicitly. Second, while action-anim decoding in aIPL and LOTC looks like it is consistently clustered, the clustering of action-PLD decoding in SPL and LOTC looks less reliable. The authors interpret this clustering as corresponding to the manual vs. bimanual distinction, but for example "drink" (a unimanual action) is grouped with "break" and "squash" (bimanual actions) in left SPL and grouped entirely separately from the unimanual and bimanual clusters in left LOTC. Statistically testing the robustness of these clusters would help clarify whether it is the case that action-PLD in SPL and LOTC has no semantically interpretable organizing principle, as might be the case for a representation based entirely on motion pattern, or rather that it is a different organizing principle from action-anim, such as the manual vs. bimanual distinction proposed by the authors. I don't have much experience with statistical testing of clustering analyses, but I think a permutation-based approach, wherein a measure of cluster robustness, such as the Silhouette score, is computed for the clusters found in the data and compared to a null distribution of such measures obtained by permuting the data labels, should be feasible. In a quick literature search, I have found several papers describing similar approaches: e.g. Hennig (2007), "Cluster-wise assessment of cluster stability"; Tibshirani et al. (2001) "Estimating the Number of Clusters in a Data Set Via the Gap Statistic". These are just pointers to potentially useful approaches, the authors are much better qualified to pick the most appropriate and convenient method. However, I do think such a statistical test would strengthen the clustering analysis shown here. With this statistical test, and the more exhaustive exposition of results I suggested in point 2 above (e.g. including animation-PLD and action-PLD decoding in aIPL), I believe the clustering analysis could even be moved to the main text and occupy a more prominent position in the paper.

(5) ROI selection: this is a minor point, related to the method used for assigning voxels to a specific ROI. In the description in the Methods (page 16, lines 514-24), the authors mention using the MNI coordinates of the center locations of Brodmann areas. Does this mean that then they extracted a sphere around this location, or did they use a mask based on the entire Brodmann area? The latter approach is what I'm most familiar with, so if the authors chose to use a sphere instead, could they clarify why? Or, if they did use the entire Brodmann area as a mask, and not just its center coordinates, this should be made clearer in the text.

Author response:

The following is the authors’ response to the previous reviews.

eLife assessment:

This study presents valuable framework and findings to our understanding of the brain as a fractal object by observing the stability of its shape property within 11 primate species and by highlighting an application to the effects of aging on the human brain. The evidence provided is solid but the link between brain shape and the underlying anatomy remains unclear. This study will be of interest to neuroscientists interested in brain morphology, whether from an evolutionary, fundamental or pathological point of view, and to physicists and mathematicians interested in modeling the shapes of complex objects.

We now clarified the outstanding questions regarding if our model outputs can be related to actual primate brain anatomy, which we believe was mainly based on comments regarding the validity of our output of apparently thicker cortices than nature can produce.

We address this point in more detail in the point-by-point response below, but want to address this misunderstanding directly here: Our algorithm does not produce thicker cortices with increasing coarse-graining scales; in fact, the cortical thickness never exceeds the actual cortical thickness in our outputs, but rather thins with each coarse-graining scale. In other words, we believe that our outputs are fully in line with neuroanatomy across species.

Reviewer #2 (Public Review): 

In this manuscript, the authors analyze the shapes of cerebral cortices from several primate species, including subgroups of young and old humans, to characterize commonalities in patterns of gyrification, cortical thickness, and cortical surface area. The authors state that the observed scaling law shares properties with fractals, where shape properties are similar across several spatial scales. One way the authors assess this is to perform a "cortical melting" operation that they have devised on surface models obtained from several primate species. The authors also explore differences in shape properties between brains of young (~20 year old) and old (~80) humans. A challenge the authors acknowledge struggling with in reviewing the manuscript is merging "complex mathematical concepts and a perplexing biological phenomenon." This reviewer remains a bit skeptical about whether the complexity of the mathematical concepts being drawn from are justified by the advances made in our ability to infer new things about the shape of the cerebral cortex. 

To allow scientists from all backgrounds to adopt these complex ideas, we have made our code to “melt” the brains and for further downstream analysis publicly available. We have now also provided a graphical user interface, to allow users without substantial coding experience to run the analysis. We also believe that the algorithmic concepts are easy to understand due to the similarity to the coarse-graining procedures found in long-standing and well-accepted box-counting algorithms.

Beyond the theoretical insight of the fractal nature of cortices and providing an explicit and crucial link between vastly different brains that are gyrified and those that are not, we believe that the advance gained by our methods for future applications is clearly demonstrated in our proof-of-principle with a four-fold increase in effect size. For reference, an effect size of 8 would translate to an almost perfect separation of groups, i.e. an ideal biomarker with near 100% sensitivity and specificity.

(1) The series of operations to coarse-grain the cortex illustrated in Figure 1 produces image segmentations that do not resemble real brains.

As re-iterated in our Methods and Discussion: “Note, of course, that the coarse-grained brain surfaces are an output of our algorithm alone and are not to be directly/naively likened to actual brain surfaces, e.g. in terms of the location or shape of the folds. Our comparisons here between coarse-grained brains and actual brains is purely on the level of morphometrics across the whole cortex.”

Fig. 1 therefore serves as an explanation to the reader on the algorithmic outputs, but each melted brain is not supposed to be directly/visually compared to actual brains. Similar to algorithms measuring the fractal dimension, or the exposed surface area of a given brain, the intermediate outputs of these algorithms are not supposed to represent any biologically observed brain structures, but rather serve as an abstraction to obtain meaningful morphometrics.

We additionally added a note to the caption of Fig. 1 to clarify this point:

“Note that the actual size of the brains for analysis are rescaled (see Methods and Fig. 3); we display all brains scaled at an equal size here for the ease of visualisation of the method.”

Finally, we also edited the entire paper for terminology to clearly distinguish the terms of (1) the cortex as a 3D object, (2) coarse-grained and voxelised versions thereof, and (3) summary morphological measures derived from the former. When we invite comparisons in our paper between real brains and coarse-grained brains, this is always at the level of summary morphological measures, not at the level of the 3D objects/voxelisations themselves.

The process to assign voxels in downsampled images to cortex and white matter is biased towards the former, as only 4 corners of a given voxel are needed to intersect the original pial surface, but all 8 corners are needed to be assigned a white matter voxel. The reason for introducing this bias (and to the extent that it is present in the authors' implementation) is not provided.

This detail was in the Supplementary, and we have now added additional clarification on this specific point to our Supplementary:

“In detail, we assign all voxels in the grid with at least four corners inside the original pial surface to the pial voxelization. This process allows the exposed surface to remain approximately constant with increasing voxel sizes. A constant exposed surface is desirable, as we only want to gradually ‘melt’ and fuse the gyri, but not grow the bounding/exposed surface as well. We want the extrinsic area to remain approximately constant as we decrease the intrinsic area via coarse-graining; it is like generating iterates of a Koch curve in reverse, from more to less detailed, by increasing the length of smallest line segment.

We then assign voxels with all eight corners inside the original white matter surface to the white matter voxelization. This is to ensure integrity of the white matter, as otherwise white matter voxels in gyri may become detached from the core white matter, and thus artificially increase white matter surface area. Indeed, the main results of the paper are not very sensitive to this decision using all eight corners, vs. e.g. only four corners, as we do not directly use white matter surface area for the scaling law measurements. However, we still maintained this choice in case future work wants to make use of the white matter voxelisations or derivative measures.”

Note on the point of white matter integrity that if both grey and white matter voxelisations require all 8 corner to be inside the respective mesh, there will be voxels not assigned to either at the grey/white matter interface, causing potential downstream issues.

We further acknowledge:

“Of course, our proposed procedure is not the only conceivable way to erase shape details below a given scale; and we are actively working on related algorithms that are also computationally cheaper. Nevertheless, the current version requires no fine-tuning, is computationally feasible and conceptually simple, thus making it a natural choice for introducing the methodology and approach.”

The authors provide an intuitive explanation of why thickness relates to folding characteristics, but ultimately an issue for this reviewer is, e.g., for the right-most panel in Figure 2b, the cortex consists of several 4.9-sided voxels and thus a >2 cm thick cortex. A structure with these morphological properties is not consistent with the anatomical organization of typical mammalian neocortex. 

We assume the reviewer refers to Fig. 1B with the panel on scale=4.9mm. We would like to point out that Fig. 1 serves as an explanation of the voxelisation method. For the actual analysis and Results, we are using re-scaled brains (see Fig. 2 with the ever decreasing brain sizes). The rescaling procedure is now expanded as below:

“Morphological properties, such as cortical thicknesses measured in our ‘melted’ brains are to be understood as a thickness relative to the size of the brain. Therefore, to analyse the scaling behaviour of the different coarse-grained realisations of the same brain, we apply an isometric rescaling process that leaves all dimensionless shape properties unaffected (more details in Suppl. S3.1). Conceptually, this process fixes the voxel size, and instead resizes the surfaces relative to the voxel size, which ensures that we can compare the coarse-grained realisations to the original cortices, and test if the former, like the latter, also scale according to Eqn. (1). Resizing, or more precisely, shrinking the cortical surface is mathematically equivalent to increasing the box size in our coarse-graining method. Both achieved an erasure of folding details below a certain threshold. After rescaling, as an example, the cortical thickness also shrinks with increasing levels of coarse-graining, and never exceeds the thickness measured at native scale.”

We additionally added a note to the caption of Fig. 1 to clarify this point:

“Note that the actual size of the brains for analysis are rescaled (see Methods and Fig. 3); we display all brains scaled at an equal size here for the ease of visualisation of the method.”

Finally, we also edited the entire paper for terminology to clearly distinguish the terms of (1) the cortex as a 3D object, (2) coarse-grained versions thereof, and (3) summary morphological measures derived from the former. When we invite comparisons in our paper between real brains and coarse-grained brains, this is always at the level of summary morphological measures, not at the level of the 3D objects themselves and their detailed anatomical features.

(2) For the comparison between 20-year-old and 80-year-old brains, a well-documented difference is that the older age group possesses more cerebral spinal fluid due to tissue atrophy, and the distances between the walls of gyri becomes greater. This difference is born out in the left column of Figure 4b. It seems this additional spacing between gyri in 80 year olds requires more extensive down-sampling (larger scale values in Figure 4a) to achieve a similar shape parameter K as for the 20 year olds. The authors assert that K provides a more sensitive measure (associated with a large effect size) than currently used ones for distinguishing brains of young vs. old people. A more explicit, or elaborate, interpretation of the numbers produced in this manuscript, in terms of brain shape, might make this analysis more appealing to researchers in the aging field.

We have removed the main results relating to K and aging from our last revision already to avoid confusion. This is now only in the supplementary analysis, and our claim of K being a more sensitive measure for age and ageing – whilst still true – will be presented in more detail in a series of upcoming papers.

(3) In the Discussion, it is stated that self-similarity, operating on all length scales, should be used as a test for existing and future models of gyrification mechanisms. Given the lack of association between the abstract mathematical parameters described in this study and explicit properties of brain tissue and its constituents, it is difficult to envision how the coarse-graining operation can be used to guide development of "models of cortical gyrification."

We have clarified in more detail what we meant originally in Discussion:

“Finally, this dual universality is also a more stringent test for existing and future models of cortical gyrification mechanisms at relevant scales, and one that moreover is applicable to individual cortices. For example, any models that explicitly simulate a cortical surface as an output could be directly coarse-grained with our method and the morphological trajectories can be compared with those of actual human and primate cortices. The simulated cortices would only be ‘valid’ in terms of the dual universality, if it also produces the same morphological trajectories.”

However, we agree with the reviewer that our paper could be misread as demanding direct comparisons of each coarse-grained brain with an actual brain, and we have now added the following text to clarify that this is not our intention for the proposed method or outputs.

“Note, we do not suggest to directly compare coarse-grained brain surfaces with actual biological brain surfaces. As we noted earlier, the coarse-grained brain surfaces are an output of our algorithm alone and not to be directly/naively likened to actual brain surfaces, e.g. in terms of the location or shape of the folds. Our comparisons here between coarse-grained brains and actual brains is purely on the level of morphometrics across the whole cortex.”

Indeed, the dual universality imposes restrictive constraints on the possible shapes of real cortices, but do not fully specify them. Presumably, the location of individual folds in different individuals and species will depend on their respective evolutionary histories, so there is no reason to expect a match in fold location between the ‘melted’ cortices of more gyrified species, on one hand, and the cortex of a less-gyrified one, on the other,  even if their global morphological parameters and global mechanism of folding coincide.

(4) There are several who advocate for analyzing cortical mid-thickness surfaces, as the pial surface over-represents gyral tips compared to the bottoms of sulci in the surface area. The authors indicate that analyses of mid-thickness representations will be taken on in future work, but this seems to be a relevant control for accepting the conclusions of this manuscript.

In the context of some applications and methods, we agree that the mid-surface is a meaningful surface to analyse. However, in our work, the mid-surface is not. The fractal estimation rests on the assumption that the exposed area hugs the object of interest (hence convex hull of the pial surface), as the relationship between the extrinsic and intrinsic areas across scales determine the fractal relationship (Eq. 2). If we used the mid-surface instead of the pial surface for all estimation, this would not represent the actual object of interest, and it is separated from the convex hull. Estimating a new convex hull based on the mid surface would be the equivalent of asking for the fractal dimension of the mid-surface, not of the cortical ribbon. In other words, it would be a different question, bound to yield a different answer.

Hence, we indicated in our original response that we only have a provisional answer, but more work beyond the scope of this paper is required to answer this question, as it is a separate question. The mid-surface, as a morphological structure in its own right, will have its own scaling properties, and our provisional understanding is that these also yield a scaling law parallel to those of the cortical ribbon with the same or a similar fractal dimension. But more systematic work is required to investigate this question at native scale and across scales.

Reviewer #3 (Public Review):

Summary: Through a rigorous methodology, the authors demonstrated that within 11 different primates, the shape of the brain followed a universal scaling law with fractal properties. They enhanced the universality of this result by showing the concordance of their results with a previous study investigating 70 mammalian brains, and the discordance of their results with other folded objects that are not brains. They incidentally illustrated potential applications of this fractal property of the brain by observing a scale-dependant effect of aging on the human brain. 

Strengths: 

- New hierarchical way of expressing cortical shapes at different scales derived from previous report through implementation of a coarse-graining procedure 

- Investigation of 11 primate brains and contextualisation with other mammals based on prior literature 

- Proposition of tool to analyse cortical morphology requiring no fine tuning and computationally achievable 

- Positioning of results in comparison to previous works reinforcing the validity of the observation. 

- Illustration of scale-dependance of effects of brain aging in the human. 

Weaknesses: 

- The notion of cortical shape, while being central to the article, is not really defined, leaving some interpretation to the reader 

- The organization of the manuscript is unconventional, leading to mixed contents in different sections (sections mixing introduction and method, methods and results, results and discussion...). As a result, the reader discovers the content of the article along the way, it is not obvious at what stages the methods are introduced, and the results are sometimes presented and argued in the same section, hindering objectivity. 

To improve the document, I would suggest a modification and restructuring of the article such that: 1) by the end of the introduction the reader understands clearly what question is addressed and the value it holds for the community, 2) by the end of the methods the reader understands clearly all the tools that will be used to answer that question (not just the new method), 3) by the end of the results the reader holds the objective results obtained by applying these tools on the available data (without subjective interpretations and justifications), and 4) by the end of the discussion the reader understands the interpretation and contextualisation of the study, and clearly grasps the potential of the method depicted for the better understanding of brain folding mechanisms and properties. 

We thank this reviewer again for their attention to detail and constructive comments. We have followed the detailed suggestions provided by us in the Recommendations For The Authors, and summarise the main changes here:

- We have restructured all sections to be more clearly following Introduction, Methods, Results, and Discussion; by using subsections, we believe the structure is now more accessible to readers.

-  We have now clarified the concept of “cortical shape”, as we use it in our paper in several places, by distinguishing clearly the object of study, and the morphological properties measured from it.

Recommendations for the authors: 

Reviewer #2 (Recommendations For The Authors): None 

Reviewer #3 (Recommendations For The Authors): 

I once again compliment the authors for their elegant work. I am happy with the way they covered my first feedback. My second review takes into account some comments made by other reviewers with which I agree. 

We thank this reviewer again for their attention to detail and constructive comments.

Recommendations for clarifications: 

General comments: The purpose of the article could be made clearer in the introduction. When I differentiate results from discussion, I think of results as objective measures or observations, while discussion will relate to the interpretation of these results (including comparison with previous literature, in most cases). 

We have restructured all sections to be more clearly following Introduction, Methods, Results, and Discussion; by using subsection, we believe the structure is now more accessible to readers.

- l.39: define or discuss "cortical shape" 

We have gone through the entire paper and corrected for any ambiguities. We specifically distinguish between the cortex as a structure overall, shape measures derived from this structure, and coarse-grained versions of the structure.

- l.48-74: this would match either an introduction or a discussion rather than a methods section. 

Done

- l.98-106: this would match a discussion rather than a methods section. 

Done

- l.111: here could be a good spot to discuss the 4 vs 8 corners for inclusion of pial vs white matter voxelization 

We have discussed this in the more detailed Supplementary section now, as after restructuring, this appears to be the more suitable place.

- l.140-180: it feels that this section mixes methods, results and discussion of the results 

We agree and we have resolved this by removing sentences and re-arranging sections.

- l.183-217: mix of results and discussion 

We agree and we have resolved this by removing sentences and re-arranging sections.

Small cosmetic suggestions: 

- l.44: conservation of 'some' quantities: vague 

Changed to conservation of morphological relationships across evolution

- l.66: order of citations ([24, 22,23]) 

Will be fixed at proof stage depending on format of references.

- l.77: delete space between citation and period 

Done

- l.77: I would delete 'say' 

Done

- l.86: 'but to also analyse' -> 'to analyse' 

Done

- l.105: remove 'we are encouraged that' 

Done

- l.111: 'also see' -> 'see also' 

Done

- l.164: 'remarkable': subjective 

Done

- l.189: define approx. abbreviation 

Done

- l.190: 'approx' -> 'approx.' 

Revised

- l.195: 'dramatic': subjective 

removed

-l. 246: 'much' -> vague 

explained

for - adjacency - ecology of communications - Nora Bateson -:indyweb - Deep Humanity

Summary - A good summary of the common thread of an ecology of communication between 4 systems thinkers

adjacency - between - ecology of communications - Nora Bateson -:Indyweb - Deep Humanity - adjacency relationship - The author summarised the salient points of a Nate Hagen Great Simplification interview with Nora Bateson on the subject of an ecology of communications - It addresses the need to use language to speak on to multiple contexts of the conversants. - The epistemologically-foundational ideas of - people centered and - interpersonal information - of the indyweb / Indranet architecture are based on the Deep Humanity ideas of - individual / collective gestalt - each individual's unique lebenswelts - the multi-meaningverse inherent in any group - symmathesetic fingerprint - perspectival knowing - salience mismatch inherent in communication due to - encoding meaning from one unique meaningverse/ lebenswelts to common language code - deciding meaning from another unique meaningverse / lebenswelt

would you be willing to annotate inline in the text where each of the scripts that does one of these actions is available? there's some information I'd be interested in hunting down for certain peptides, and having a pointer to the code that does each specific thing would be helpful to accomplish this

Author response:

The following is the authors’ response to the original reviews.

We would like to thank you very much for reviewing our manuscript and express our sincere appreciation for the valuable and thoughtful comments that led us to significantly improve the manuscript on Fshr-ZsGreen reporter mice. We have seriously taken your comments to make a major revision of the manuscript, and here is a summary of the revision:

(1) New data on Fshr expression are input to the revised Manuscript:

a. Fshr expression in the testis and adipose tissues (WAT and BAT) of B6 mice;

b. Fshr expression in the testis of B6 by RNA-smFISH;

c. Comparison of Fshr expression in the testis and ovary between Fshr-ZsGreen and B8 mice by ddRT-PCR to prove Fshr expression without interruptions by insertion of P2A-ZsGreen vector;

d. Reduction of Fshr expression in osteocytes within the femoral sections from DMP1-CreERT2:Fshrfl/fl mice;

e. Fshr expression in an established Leydig cell line-TM3 by immunofluorescence and ddRT-PCR, also show Fshr located in the nuclei of TM3 cells;

f. Fshr expression at scRNA-seq level from 5 public single cell portals as Supplementary Data 3 to support our findings of the widespread expression pattern of Fshr, particularly in Leydig cells.

(2) Re-organization of Figure 2 with a new legend.

(3) A new paragraph is added to the Discussion Section of the revised MS to explain the function of P2A peptide in generation of GFP reporter mice and why Fshr express is not interrupted by the P2A-ZsGreen insertion in Fshr-ZsGreen reporter.

(4) Deletion of Figure 1-D-c, as it is not necessary.

(5) Replace of Figure 8-A (the left panel) with a reduced exposure time image.

(6) Amended parts of the revised MS are labeled in red.

A point by point response to the Reviewers’ comments:

Reviewer 1:

One of the shocking observations in this manuscript is the expression of FSHR in Leydig cells. Other observations are in the osteoblasts and endothelial cells as well as epithelial cells in different organs. The expression of ZsGreen in these tissues seems high and one shall start questioning if there are other mechanisms at play here.

First, the turnover of fluorescent proteins is long, longer than 48h, which means that they accumulate at a different speed than the endogenous FSHR This means that ZsGreen will accumulate in time while the FSHR receptor might be degraded almost immediately. This correlated with mRNA expression (by the authors) but does not with the results of other studies in single-cell sequencing (see below).

The expression of ZsGreen in Leydig cells seems much higher than in Sertoli cells, this is "disturbing" to put it mildly. This is visible in both the ZsGreen expression and the FISH assay (Figure 2 B-D).

Thank you for this valuable comments. We added new data on Fshr expression to prove the presence of Fshr in Leydig cells in B6 detected by immunofluorescence staining, RNA-smFISH and ddRT-PCR, as well as in TM3 cells-isolated Leydig cells from a male mice in the revise MS (Fig 2E, F and G), that demonstrate no interruptions of normal Fshr expression by insertion of P2A-ZsGreen vector into a locus located between exon10 and stop code. We use ZsGreen as an indicator for active Fshr promoter status, rather than a method to measure Fshr expression, which is done by ddRT-PCR. These data are shown in Figure 2G of the revised MS

In addition, we provide scRNA-seq based evidence on Fshr expression in human Leydig cells from two single cell portals (DISCO and BioGPS) as shown in Supplementary Data 3 in the revised MS. We also cited a recent report on scRNA-seq analysis of Fshr expression in Hu sheep in the revised MS as Reference 65 (PMID: 37541020) 1, which also clearly showed Fshr expression in Leydig cells at single cell level in Hu Sheep.

We believe that the lack of Fshr expression in some single cell databases may be due to the degradation of Fshr transcript in cells during the process of single cell populations. In our laboratory, we spent more than 6 months to optimize methods and reagents to perverse mRNA integrity more than 8 for RAN-seq.

The expression in WAT and BAT is also questionable as the expression of ZsGreen is high everywhere. That makes it difficult to believe that the images are truly informative. For example, the stainings of aorta show the ZsGreen expression where elastin and collagen fibres are - these are not "cells" and therefore are not expressing ZsGreen.

FISH expression (for FSHR) in WT mice is missing.

Also, the tissue sections were stained with the IgG only (neg control) but in practice both the KI and the WT tissues should be stained with the primary and secondary antibodies. The only control that I could think of to truly get a sense of this would be a tagged receptor (N-terminal) that could then be analysed by immunohistochemistry.

Reply 2 and 3: Thank you for these comments. New data on Fshr expression in WAT and BAT of B6 mice by immunofluorescence staining and in the testis of B6 mice by immunofluorescence staining and RNA-smFISH are added to the revised MS (Fig.2D and E, and Fig. 4G), showing similar patterns to that of Fshr-ZsGreen mice. Furthermore, we provide more evidences as Supplementary Data 3 on Fshr expression obtained from 4 public single cell portables, showing FSHR expression in a widespread organs and tissues (including different fractions of adipose cells) of human, mice and rat at single cell levels. Please also check Fshr expression pattern in adipose tissues by immunostaining for Fshr in previous reports (Fig. 3a of PMID: 28538730 and Fig. 2 of PMID: 25754247) 2 3, which showed a similar expression pattern to our finding. These data should address your concerns on Fshr expression in WAT and BAT and other organs/tissues.

Regard of “For example, the stainings of aorta show the ZsGreen expression where elastin and collagen fibres are - these are not "cells" and therefore are not expressing ZsGreen.” We believe that you referred to the image of the aorta in Supplementary Data2. However, Please take a look at the images of the aorta in Figure 5-C, which shows positively stained the layer of ‘elastin and collagen fibres’ for EMCN and a-SMA colocalized with Fshr expression with stained DAPI at a 1000X magnification, indicating endothelial cells and the cellular membrane presented in this layer, not just ‘elastin and collagen’.

The authors also claim:

To functionally prove the presence of FSHR in osteoblasts/osteocytes, we also deleted FSHR in osteocytes using an inducible model. The conditional knockout of FSHR triggered a much more profound increase in bone mass and decrease in fat mass than blockade by FSHR antibodies (unpublished data).

This would be a good control for all their images. I think it is necessary to make the large claim of extragonadal expression, as well as intragonadal such as Leydig cells.

Thank you for this very encouraging comment. As you suggested, we did add a result of reduced Fshr expression in osteocytes from DMP1-CreERT2+:Fshrfl/fl mice treated with tamoxifen to the revise MS, as shown in Figure 3D, demonstrating Fshr present in osteocytes and the specificity of Fshr antibody. Furthermore, we incorporated your advice on making ‘ large claim of extrogonadal and intragonadal expression of Fshr’ into the revised MS in red.

Claiming that the under-developed Leydig cells in FSHR KO animals are due to a direct effect of the FSHR, and not via a cross-talk between Sertoli and Leydig cells, is too much of a claim. It might be speculated to some degree but as written at the moment it suggests this is "proven".

Thank you for pointing out this incorrect claim and we apologized for it. In the revised MS, we deleted this claim.

We also do not know if this FSHR expressed is a spliced form that would also result in the expression of ZsGreen but in a non-functional FSHR, or whether the FSHR is immediately degraded after expression. The insertion of the ZsGreen might have disturbed the epigenetics, transcription, or biosynthesis of the mRNA regulation.

Thanks for this comment. In the revised MS, we added a new section to explain the function of P2A peptide in generation of a GFP reporter by sgRNA-guilded site specific knockin of P2A ZsGreen vector through CRISPRA/cas9 and provided a new result on comparison of Fshr expression in the testes and ovaries from Fshr-ZsGreen and B6 mice, showing equivalent Fshr expression between Fshr-ZsGreen and B6 mice (Figure 2G), which indicates no interruptions of Fshr expression by the insertion of P2A vector.

The authors should go through single-cell data of WT mice to show the existence of the FSHR transcript(s).<br /> For example here:<br /> https://www.nature.com/articles/sdata2018192

Thank you so much for the valuable comment. Yes, we took you critical advice to check Fshr expression through 4 single cell portals, including DISCO, GTEx, BioGPS and Human single cell portal, and present the collected data as Supplementary Data 3 in the revised MS, that strongly support our findings of the wider Fshr expression. Particularly, Fshr expression in Leydig cells is proved by scRNA-seq studies of human cells from DISCO and BioGPS, as well as a recent study in Hu sheep (PMID: 37541020) 1 and we cited it in the revised MS.

Reviewer 2:

Is the FSHR expression pattern affected by the knockin mice (no side-by-side comparison between wt and GSGreen mice, using in situ hybridization and ddRTPCR, at least in the gonads, is provided)?

Thanks for the comment. In the revised MS, we provided a set of new data on Fshr expression in the testis, ovary, WAT and BAT of B6 mice by immunofluorescence staining and by RNA-smFISH for Fshr expression, showing similar expression patterns. Additionally, we also performed ddRT-PCT to compare Fshr expression in the testes and ovaries between Fshr-ZsGreen and B6 mice, demonstrating equivalent expression of Fshr expression between Fshr-ZsGreen and B6 mice. Interestingly, we also observed an significantly higher Fshr expression in the testis than that in the ovary (more than 30 folds).

Is the splicing pattern of the FSHR affected in the knockin compared to wt mice, at least in the gonads?

Thanks for the question. Please see our reply to the Reviewer 1 for the function of P2A peptide used for generation of GFP reporters.  Although we didn’t directly assess the splicing pattern, we provide a result of comparison of Fshr expression in Figure 2F in the revised MS, indirectly showing no changes of the splicing pattern. We will assess the splicing pattern of Fshr in the future that has been neglected in the field.

Are there any additional off-target insertions of GSGreen in these mice?” and “Are similar results observed in separate founder mice?

Thanks for the questions. As we describe it in the method section  in detail in the MS, Fshr-ZsGreen reporter was produced by the a site-specific long ssDNA recombination of the P2A-ZsGreen targeting vector to the locus between Exon10 and stop code by CRIPRA/cas9, which was guided by site-specific single guide RNA (sgRNA). We showed the results of Southern blot, DNA sequencing and site-specific PCR, proving the site-specific insertion of P2A-ZsGreen as shown in Figure 1. Because of the site-specific recombination, professionally, only one funder line is required for the study and there are no additional off-target insertions.

How long is GSGreen half-life? Could a very long half-life be a major reason for the extremely large expression pattern observed?

Thanks for the question. The half life of ZsGreen, also called ZsGreen1, is at least 26 h in mammalian cells or slightly longer due to its tetrameric structure, in contrast with the monomeric configuration of other well-known fluorescent proteins (PMID: 17510373) 4. The rationale for using this GFP protein is that ZsGreen is an exceptionally bright green fluorescent protein, which is up to 4X brighter than EGFP—and is ideally suited for whole-cell labelling, promoter-reporter studies, considering of the higher turnover and rapid degradation of Fshr transcript. In this study, we used ZsGreen as a monitor or an indicator of the active Fshr endogenous promoter, rather than a means for measuring the promoter activity. Therefore, regardless of its accumulation or not, ZsGreen driven by Fshr promoter, indicates the presence of active Fshr promoter in the defined cells. In stead, we used ddRT-PCR to measure Fshr expression degrees in this study. In addition, we also provide single cell sequence-based evidence from 4 public single cell portables to support our findings of the wide Fshr expression. Please see Supplementary Data 3 in the revised MS.

References:

(1) Su J, Song Y, Yang Y, et al. Study on the changes of LHR, FSHR and AR with the development of testis cells in Hu sheep. Anim Reprod Sci. Sep 2023;256:107306. doi:10.1016/j.anireprosci.2023.107306

(2) Liu P, Ji Y, Yuen T, et al. Blocking FSH induces thermogenic adipose tissue and reduces body fat. Nature. Jun 1 2017;546(7656):107-112. doi:10.1038/nature22342

(3) Liu XM, Chan HC, Ding GL, et al. FSH regulates fat accumulation and redistribution in aging through the Galphai/Ca(2+)/CREB pathway. Aging Cell. Jun 2015;14(3):409-20. doi:10.1111/acel.12331

(4) Bell P, Vandenberghe LH, Wu D, Johnston J, Limberis M, Wilson JM. A comparative analysis of novel fluorescent proteins as reporters for gene transfer studies. J Histochem Cytochem. Sep 2007;55(9):931-9. doi:10.1369/jhc.7A7180.2007

On responding to the first round of reviews, the authors have nicely adjusted their wording and fairly describe the results of their study. Certain markers were identified for further investigation. Yet, an overall non-obvious relationship between immune markers and HIV reservoirs has been shown previously, and despite the attempt to leverage powerful ML algorithms, they are not magical and cannot reveal strong relationships that fundamentally do not exist. In addition, categorical classification is for now hard to interpret and the more powerful ML algorithms do not seem to outperform more classic regression methods. Therefore, it remains relatively hard to evaluate the utility of this kind of study.

Initial summary:

Semenova et al. have studied a large cross-sectional cohort of people living with HIV on suppressive ART, N=115, and performed high dimensional flow-cytometry to then search for associations between immunological and clinical parameters and intact/total HIV DNA levels.

A number of interesting data science/ML approaches were explored on the data and the project seems a serious undertaking. However, like many other studies that have looked for these kinds of associations, there was not a very strong signal. Of course the goal of unsupervised learning is to find new hypotheses that aren't obvious to human eyes, but I felt in that context, there were (1) results slightly oversold, (2) some questions about methodology in terms mostly of reservoir levels, and (3) results were not sufficiently translated back into meaning in terms of clinical outcomes.

Strengths:

The study is evidently a large and impressive undertaking and combines many cutting edge statistical techniques with a comprehensive experimental cohort of people living with HIV, notably inclusive of populations underrepresented in HIV science. A number of intriguing hypotheses are put forward that could be explored further. Data will be shared and could be a useful repository for more specific analyses.

Weaknesses:

Despite the detailed experiments and methods, there was not a very strong signal for variable(s) predicting HIV reservoir size. The spearman coefficients are ~0.3, (somewhat weak, and acknowledged as such) and predictive models reach 70-80% prediction levels, though of sometimes categorical variables that are challenging to interpret.

There are some questions about methodology, as well as some conclusions that are not completely supported by results, or at minimum not sufficiently contextualized in terms of clinical significance. Edit, authors have substantially revised the text.

On associations: the false discovery rate correction was set at 5%, but data appear underdetermined with fewer observations than variables (144vars > 115ppts), and it isn't always clear if/when variables are related (e.g inverses of one another, for instance %CD4 and %CD8).

The modeling of reservoir size was unusual, typically intact and defective HIV DNA are analyzed on a log10 scale (both for decays and predicting rebound). Also sometimes in this analysis levels are normalized (presumably to max/min?, e.g. S5), and given the large within-host variation of level we see in other works, it is not trivial to predict any downstream impact of normalization across population vs within person. Edit, fixed.

Also, the qualitative characterization of low/high reservoir is not standard, and naturally will split by early/later ART if done as above/below median. Given the continuous nature of these data it seems throughout that predicting above/below median is a little hard to translate into clinical meaning.

Lastly, work is comprehensive and appears solid, but the code was not shared to see how calculations were performed. Edit, fixed.

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

Semenova et al. have studied a large cross-sectional cohort of people living with HIV on suppressive ART, N=115, and performed high dimensional flow cytometry to then search for associations between immunological and clinical parameters and intact/total HIV DNA levels.

A number of interesting data science/ML approaches were explored on the data and the project seems a serious undertaking. However, like many other studies that have looked for these kinds of associations, there was not a very strong signal. Of course, the goal of unsupervised learning is to find new hypotheses that aren't obvious to human eyes, but I felt in that context, there were (1) results slightly oversold, (2) some questions about methodology in terms mostly of reservoir levels, and (3) results were not sufficiently translated back into meaning in terms of clinical outcomes.

We appreciate the reviewer’s perspective.  In our revised version of the manuscript, we have attempted to address these concerns by more adequately explaining the limitations of the study and by more thoroughly discussing the context of the findings.  We are not able to associate the findings with specific clinical outcomes for individual study participants but we speculate about the overall biological meaning of these associations across the cohort.  We cannot disagree with the reviewer, but we find the associations statistically significant, potentially reflecting real biological associations, and forming the basis for future hypothesis testing research. 

Strengths:

The study is evidently a large and impressive undertaking and combines many cutting-edge statistical techniques with a comprehensive experimental cohort of people living with HIV, notably inclusive of populations underrepresented in HIV science. A number of intriguing hypotheses are put forward that could be explored further. Sharing the data could create a useful repository for more specific analyses.

We thank the reviewer for this assessment.

Weaknesses:

Despite the detailed experiments and methods, there was not a very strong signal for the variable(s) predicting HIV reservoir size. The Spearman coefficients are ~0.3, (somewhat weak, and acknowledged as such) and predictive models reach 70-80% prediction levels, though sometimes categorical variables are challenging to interpret.

We agree with the reviewer that individual parameters are only weakly correlated with the HIV reservoir, likely reflecting the complex and multi-factorial nature of reservoir/immune cell interactions.  Nevertheless, these associations are statistically significant and form the basis for functional testing in viral persistence.

There are some questions about methodology, as well as some conclusions that are not completely supported by results, or at minimum not sufficiently contextualized in terms of clinical significance.  On associations: the false discovery rate correction was set at 5%, but data appear underdetermined with fewer observations than variables (144vars > 115ppts), and it isn't always clear if/when variables are related (e.g inverses of one another, for instance, %CD4 and %CD8).

When deriving a list of cell populations whose frequency would be correlated with the reservoir, we focused on well-defined cell types for which functional validation exists in the literature to consider them as distinct cell types.  For many of the populations, gating based on combinations of multiple markers leads to recovery of very few cells, and so we excluded some potential combinations from the analysis.  We are also making our raw data available for others to examine and find associations not considered by our manuscript.

The modeling of reservoir size was unusual, typically intact and defective HIV DNA are analyzed on a log10 scale (both for decays and predicting rebound). Also, sometimes in this analysis levels are normalized (presumably to max/min?, e.g. S5), and given the large within-host variation of level we see in other works, it is not trivial to predict any downstream impact of normalization across population vs within-person.

We have repeated the analysis using log10 transformed data and the new figures are shown in Figure 1 and S2-S5.

Also, the qualitative characterization of low/high reservoir is not standard and naturally will split by early/later ART if done as above/below median. Given the continuous nature of these data, it seems throughout that predicting above/below median is a little hard to translate into clinical meaning.

Our ML models included time before ART as a variable in the analysis, and this was not found to be a significant driver of the reservoir size associations, except for the percentage of intact proviruses (see Figure 2C). Furthermore, we analyzed whether any of the reservoir correlated immune variables were associated with time on ART and found that, although some immune variables are associated with time on therapy, this was not the case for most of them (Table S4). We agree that it is challenging to translate above or below median into clinical meaning for this cohort, but we emphasize that this study is primarily a hypothesis generating approach requiring additional validation for the associations observed.  We attempted to predict reservoir size as a continuous variable using the data and this approach was not successful (Figure S13). We believe that a significantly larger cohort will likely be required to generate a ML model that can accurately predict the reservoir as a continuous variable.  We have added additional discussion of this to the manuscript.

Lastly, the work is comprehensive and appears solid, but the code was not shared to see how calculations were performed.

We now provide a link to the code used to perform the analyses in the manuscript, https://github.com/lesiasemenova/ML_HIV_reservoir.

Reviewer #2 (Public Review):

Summary:

Semenova et. al., performed a cross-sectional analysis of host immunophenotypes (using flow cytometry) and the peripheral CD4+ T cell HIV reservoir size (using the Intact Proviral DNA Assay, IPDA) from 115 people with HIV (PWH) on ART. The study mostly highlights the machine learning methods applied to these host and viral reservoir datasets but fails to interpret these complex analyses into (clinically, biologically) interpretable findings. For these reasons, the direct translational take-home message from this work is lost amidst a large list of findings (shown as clusters of associated markers) and sentences such as "this study highlights the utility of machine learning approaches to identify otherwise imperceptible global patterns" - lead to overinterpretation of their data.

We have addressed the reviewer’s concern by modifications to the manuscript that enhance the interpretation of the findings in a clinical and biological context.

Strengths:

Measurement of host immunophenotyping measures (multiparameter flow cytometry) and peripheral HIV reservoir size (IPDA) from 115 PWH on ART.

Major Weaknesses:

(1) Overall, there is little to no interpretability of their machine learning analyses; findings appear as a "laundry list" of parameters with no interpretation of the estimated effect size and directionality of the observed associations. For example, Figure 2 might actually give an interpretation of each X increase in immunophenotyping parameter, we saw a Y increase/decrease in HIV reservoir measure.

We have added additional text to the manuscript in which we attempt to provide more immunological and clinical interpretation of the associations.  We also have emphasized that these associations are still speculative and will require additional validation.  Nevertheless, our data should provide a rich source of new hypotheses regarding immune system/reservoir interaction that could be tested in future work.

(2) The correlations all appear to be relatively weak, with most Spearman R in the 0.30 range or so.

We agree with the review that the associations are mostly weak, consistent with previous studies in this area.  This likely is an inherent feature of the underlying biology – the reservoir is likely associated with the immune system in complex ways and involves stochastic processes that will limit the predictability of reservoir size using any single immune parameter. We have added additional text to the manuscript to make this point clearer.

(3) The Discussion needs further work to help guide the reader. The sentence: "The correlative results from this present study corroborate many of these studies, and provide additional insights" is broad. The authors should spend some time here to clearly describe the prior literature (e.g., describe the strength and direction of the association observed in prior work linking PD-1 and HIV reservoir size, as well as specify which type of HIV reservoir measures were analyzed in these earlier studies, etc.) and how the current findings add to or are in contrast to those prior findings.

We have added additional text to the manuscript to help guide the readers through the possible biological significance of the findings and the context with respect to prior literature.

(4) The most interesting finding is buried on page 12 in the Discussion: "Uniquely, however, CD127 expression on CD4 T cells was significantly inversely associated with intact reservoir frequency." The authors should highlight this in the abstract, and title, and move this up in the Discussion. The paper describes a very high dimensional analysis and the key takeaways are not clear; the more the author can point the reader to the take-home points, the better their findings can have translatability to future follow-up mechanistic and/or validation studies.

We appreciate the reviewer’s comment.  We have increased the emphasis on this finding in the revised version of the manuscript.

(5) The authors should avoid overinterpretation of these results. For example in the Discussion on page 13 "The existence of two distinct clusters of PWH with different immune features and reservoir characteristics could have important implications for HIV cure strategies - these two groups may respond differently to a given approach, and cluster membership may need to be considered to optimize a given strategy." It is highly unlikely that future studies will be performing the breadth of parameters resulting here and then use these directly for optimizing therapy.

Our analyses indicate that membership of study participants in cluster1 or cluster 2 can be fairly accurately determined by a small number of individual parameters (KLRG1 etc, Figure 4F), and measuring the cells of PWH with the degree of breadth used in this paper would not be necessary to classify PWH into these clusters.  As such, we feel that it is not unrealistic to speculate that this finding could turn out to be clinically useful, if it becomes clear that the clusters are biologically meaningful.

(6) There are only TWO limitations listed here: cross-sectional study design and the use of peripheral blood samples. (The subsequent paragraph notes an additional weakness which is misclassification of intact sequences by IPDA). This is a very limited discussion and highlights the need to more critically evaluate their study for potential weaknesses.

We have expanded on the list of limitations discussed in the manuscript. In particular, we now address the size of the cohort, the composition with respect to different genders and demographics, lack of information for the timing of ART and the lack of information regarding intracellular transcriptional pathways.

(7) A major clinical predictor of HIV reservoir size and decay is the timing of ART initiation. The authors should include these (as well as other clinical covariate data - see #12 below) in their analyses and/or describe as limitations of their study.

All of the participants that make up our cohort were treated during chronic infection, and the precise timing of ART initiation is unclear in most of these cases.  We have added additional information to explain this in the manuscript and include this in the list of limitations.

Reviewer #3 (Public Review):

Summary:

This valuable study by Semenova and colleagues describes a large cross-sectional cohort of 115 individuals on ART. Participants contributed a single blood sample which underwent IPDA, and 25-color flow with various markers (pre and post-stimulation). The authors then used clustering, decision tree analyses, and machine learning to look for correlations between these immunophenotypic markers and several measures of HIV reservoir volume. They identified two distinct clusters that can be somewhat differentiated based on total HIV DNA level, intact HIV DNA level, and multiple T cell cellular markers of activation and exhaustion.

The conclusions of the paper are supported by the data but the relationships between independent and dependent variables in the models are correlative with no mechanistic work to determine causality. It is unclear in most cases whether confounding variables could explain these correlations. If there is causality, then the data is not sufficient to infer directionality (ie does the immune environment impact the HIV reservoir or vice versa or both?). In addition, even with sophisticated and appropriate machine learning approaches, the models are not terribly predictive or highly correlated. For these reasons, the study is very much hypothesis-generating and will not impact cure strategies or HIV reservoir measurement strategies in the short term.

We appreciate the reviewer’s comments regarding the value of our study.  We fully acknowledge that the causal nature and directionality of these associations are not yet clear and agree that the study is primarily hypothesis generating in nature.  Nevertheless, we feel that the hypotheses generated will be valuable to the field.  We have added additional text to the manuscript to emphasize the hypothesis generating nature of this paper.

Strengths:

The study cohort is large and diverse in terms of key input variables such as age, gender, and duration of ART. Selection of immune assays is appropriate. The authors used a wide array of bioinformatic approaches to examine correlations in the data. The paper was generally well-written and appropriately referenced.

Weaknesses:

(1) The major limitation of this work is that it is highly exploratory and not hypothesis-driven. While some interesting correlations are identified, these are clearly hypothesis-generating based on the observational study design.

We agree that the major goal of this study was hypothesis generating and that our work is exploratory in nature. Performing experiments with mechanism testing goals in human participants with HIV is challenging.  Additionally, before such mechanistic studies can be undertaken, one must have hypotheses to test. As such we feel our study will be useful for the field in helping to identify hypotheses that could potentially be tested.

(2) The study's cross-sectional nature limits the ability to make mechanistic inferences about reservoir persistence. For instance, it would be very interesting to know whether the reservoir cluster is a feature of an individual throughout ART, or whether this outcome is dynamic over time.

We agree with the reviewer’s comment. Longitudinal studies are challenging to carry out with a study cohort of this size, and addressing questions such as the one raised by the reviewer would be of great interest. We believe our study nevertheless has value in identifying hypotheses that could be tested in a longitudinal study.

(3) A fundamental issue is that I am concerned that binarizing the 3 reservoir metrics in a 50/50 fashion is for statistical convenience. First, by converting a continuous outcome into a simple binary outcome, the authors lose significant amounts of quantitative information. Second, the low and high reservoir outcomes are not actually demonstrated to be clinically meaningful: I presume that both contain many (?all) data points above levels where rebound would be expected soon after interruption of ART. Reservoir levels would also have no apparent outcome on the selection of cure approaches. Overall, dividing at the median seems biologically arbitrary to me.

The reviewer raises a valid point that the clinical significance of above or below median reservoir metrics is unclear, and that the size of the reservoir has potentially little relation to rebound and cure approaches.  In the manuscript, we attempted to generate models that can predict reservoir size as a continuous variable in Figure S13 and find that this approach performs poorly, while a binarized approach was more successful. As such we have included both approaches in the manuscript.  It is possible that future studies with larger sample sizes and more detailed measurements will perform better for continuous variable prediction.  While this is a fairly large study (n=115) by the standards of HIV reservoir analyses, it is a small study by the standards of the machine learning field, and accurate predictive ML models for reservoir size as a continuous variable will likely require a much larger set of samples/participants.  Nevertheless, we feel our work has value as a template for ML approaches that may be informative for understanding HIV/immune interactions and generates novel hypotheses that could be validated by subsequent studies.

(4) The two reservoir clusters are of potential interest as high total and intact with low % intact are discriminated somewhat by immune activation and exhaustion. This was the most interesting finding to me, but it is difficult to know whether this clustering is due to age, time on ART, other co-morbidity, ART adherence, or other possible unmeasured confounding variables.

We agree that this finding is one of the more interesting outcomes of the study. We examined a number of these variables for association with cluster membership, and these data are reported in Figure S8A-D.  Age, years of ART and CD4 Nadir were all clearly different between the clusters.   The striking feature of this clustering, however, is the clear separation between the two groups of participants, as opposed to a continuous gradient of phenotypes.  This could reflect a bifurcation of outcomes for people with HIV, dynamic changes in the reservoir immune interactions over time, or different levels of untreated infection.  It is certainly possible that some other unmeasured confounding variables contribute to this outcome and we have attempted to make this limitation clearer.

(5) At the individual level, there is substantial overlap between clusters according to total, intact, and % intact between the clusters. Therefore, the claim in the discussion that these 2 cluster phenotypes may require different therapeutic approaches seems rather speculative. That said, the discussion is very thoughtful about how these 2 clusters may develop with consideration of the initial insult of untreated infection and / or differences in immune recovery.

We agree with the reviewer that this claim is speculative, and we have attempted to moderate the language of the text in the revised version.

(6) The authors state that the machine learning algorithms allow for reasonable prediction of reservoir volume. It is subjective, but to me, 70% accuracy is very low. This is not a disappointing finding per se. The authors did their best with the available data. It is informative that the machine learning algorithms cannot reliably discriminate reservoir volume despite substantial amounts of input data. This implies that either key explanatory variables were not included in the models (such as viral genotype, host immune phenotype, and comorbidities) or that the outcome for testing the models is not meaningful (which may be possible with an arbitrary 50/50 split in the data relative to median HIV DNA volumes: see above).

We acknowledge that the predictive power of the models generated from these data is modest and we have clarified this point in the revised manuscript. As the reviewer indicates, this may result from the influence of unmeasured variables and possible stochastic processes.  The data may thus demonstrate a limit to the predictability of reservoir size which may be inherent to the underlying biology.  As we mention above, this study size (n-115) is fairly small for the application of ML methods, and an increased sample size will likely improve the accuracy of the models. At this stage, the models we describe are not yet useful as predictive clinical tools, but are still nonetheless useful as tools to describe the structure of the data and identify reservoir associated immune cell types.

(7) The decision tree is innovative and a useful addition, but does not provide enough discriminatory information to imply causality, mechanism, or directionality in terms of whether the immune phenotype is impacting the reservoir or vice versa or both. Tree accuracy of 80% is marginal for a decision tool.

The reviewer is correct about these points.  In the revised manuscript, we have attempted to make it clear that we are not yet advocating using this approach as a decision tool, but simply a way to visualize the data and understand the structure of the dataset.  As we discuss above, the models will likely need to be trained on a larger dataset and achieve higher accuracy before use as a decision tool.

(8) Figure 2: this is not a weakness of the analysis but I have a question about interpretation. If total HIV DNA is more predictive of immune phenotype than intact HIV DNA, does this potentially implicate a prior high burden of viral replication (high viral load &/or more prolonged time off ART) rather than ongoing reservoir stimulation as a contributor to immune phenotype? A similar thought could be applied to the fact that clustering could only be detected when applied to total HIV DNA-associated features. Many investigators do not consider defective HIV DNA to be "part of the reservoir" so it is interesting to speculate why these defective viruses appear to have more correlation with immunophenotype than intact viruses.

We agree with the reviewer that this observation could reflect prior viral burden and we have added additional text to make this clearer.  Even so, we cannot rule out a model in which defective viral DNA is engaged in ongoing stimulation of the immune system during ART, leading to the stronger association between total DNA and the immune cell phenotypes. We hypothesize that the defective proviruses could potentially be triggering innate immune pattern recognition receptors via viral RNA or DNA, and a higher burden of the total reservoir leads to a stronger apparent association with the immune phenotype.  We have included text in the discussion about this hypothesis.

(9) Overall, the authors need to do an even more careful job of emphasizing that these are all just correlations. For instance, HIV DNA cannot be proven to have a causal effect on the immunophenotype of the host with this study design. Similarly, immunophenotype may be affecting HIV DNA or the correlations between the two variables could be entirely due to a separate confounding variable

We have revised the text of the manuscript to emphasize this point, and we acknowledge that any causal relationships are, at this point, simply speculation. 

(10) In general, in the intro, when the authors refer to the immune system, they do not consistently differentiate whether they are referring to the anti-HIV immune response, the reservoir itself, or both. More specifically, the sentence in the introduction listing various causes of immune activation should have citations. (To my knowledge, there is no study to date that definitively links proviral expression from reservoir cells in vivo to immune activation as it is next to impossible to remove the confounding possible imprint of previous HIV replication.) Similarly, it is worth mentioning that the depletion of intact proviruses is quite slow such that provial expression can only be stimulating the immune system at a low level. Similarly, the statement "Viral protein expression during therapy likely maintains antigen-specific cells of the adaptive immune system" seems hard to dissociate from the persistence of immune cells that were reactive to viremia.

We updated the text of the manuscript to address these points and have added additional citations as per the reviewer’s suggestion.

(11) Given the many limitations of the study design and the inability of the models to discriminate reservoir volume and phenotype, the limitations section of the discussion seems rather brief.

We have now expanded the limitations section of the discussion and added additional considerations. We now include a discussion of the study cohort size, composition and the detail provided by the assays.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

A few specific comments:

"This pattern is likely indicative of a more profound association of total HIV DNA with host immunophenotype relative to intact HIV DNA."

Most studies I have seen (e.g. single cell from Lictherfeld/Yu group) show intact proviruses are generally more activated/detectable/susceptible to immune selection, so I have a hard time thinking defective proviruses are actually more affected by immunotype.

We hypothesize that this association is actually occurring in the opposite direction – that the defective provirus are having a greater impact on the immune phenotype, due to their greater number and potential ability to engage innate or adaptive immune receptors. We have clarified this point in the manuscript

"The existence of two distinct clusters of PWH with different immune features and reservoir characteristics could have important implications for HIV cure strategies - these two groups may respond differently to a given approach, and cluster membership may need to be considered to optimize a given strategy."

I find this a bit of a reach, given that the definition of 2 categories depended on the total size.

We have modified the language of this section to reduce the level of speculation.

"This study is cross-sectional in nature and is primarily observational, so caution should be used interpreting findings associated with time on therapy".

I found this an interesting statement because ultimately time on ART shows up throughout the analysis as a significant predictor, do you mean something about how time on ART could indicate other confounding variables like ART regimen or something?

We have rephrased this comment to avoid confusion.  We were simply trying to make the point that we should avoid speculating about longitudinal dynamics from cross sectional data.

"As expected, the plots showed no significant correlation for intact HIV DNA versus years of ART (Figure 1B), while total reservoir size was positively correlated with the time of ART (Figure 1A, Spearman r = 0.31)."<br />  Is this expected? Studies with longitudinal data almost uniformly show intact decay, at least for the first 10 or so years of ART, and defective/total stability (or slight decay). Also probably "time on ART" to not confuse with the duration of infection before ART.

We have updated the language of this section to address this comment.  We have avoided comparing our data with respect to time on ART to longitudinal studies for reasons given above.

On dimensionality reduction, as this PaCMAP seems a relatively new technique (vs tSNE and UMAP which are more standard, but absolutely have their weaknesses), it does seem important to contextualize. I think it would still be useful to show PCA and asses the % variance of each additional dimension to assess the effective dimensionality, it would be helpful to show a plot of % variance by # components to see if there is a cutoff somewhere, and if PaCMAP is really picking this up to determine the 2 dimensions/2 clusters is ideal. Figure 4B ultimately shows a lot of low/high across those clusters, and since low/high is defined categorically it's hard to know which of those dots are very close to the other categories.

We have added this analysis to the manuscript – found in Figure S9. The PCA plot indicates that members of the two clusters also separate on PCA although this separation is not as clear as for the PaCMAP plot.

Minor comments on writing etc:

Intro

-Needs some references on immune activation sequelae paragraph.

We have added some additional references to this section.

-"promote the entry of recently infected cells into the reservoir" -- that is only one possible mechanistic explanation, it's not unreasonable but it seems important to keep options open until we have more precise data that can illuminate the mechanism of the overabundance.

We have modified the text to discuss additional hypotheses.

-You might also reference Pankau et al Ppath for viral seeding near the time of ART.

We have added this reference.

-"Viral protein expression during therapy likely maintains antigen-specific cells of the adaptive immune system" - this was unclear to me, do you mean HIV-specific cells that act against HIV during ART? I think most studies show immunity against HIV (CD8 and CD4) wanes over time during ART.

The Goonetilleke lab has recently generated data indicating that antiviral T cell responses are remarkably stable over time on ART, but we agree with the reviewer that the idea that ongoing antigen expression in the reservoir maintains these cells is speculative.  We have modified the text to make this point clearer.

-Overall I think the introduction lacked a little bit of definitional precision: i.e. is the reservoir intact vs replication competent vs all HIV DNA and whether we are talking about PWH on long-term ART and how long we should be imagining? The first years of ART are certainly different than later, in terms of dynamics. The ultimate implications are likely specific for some of these categorizations.

-"persistent sequelae of the massive disruptions to T cell homeostasis and lymphoid structures that occur during untreated HIV infection" needs a lot more context/referencing. For instance, Peter Hunt showed a decrease in activation after ART a long time ago.

-Heather Best et al show T cell clonality stays perturbed after ART.

We have updated the text of the introduction and added references to address the reviewer’s comments.

Results

-It would be important to mention the race of participants and any information about expected clades of acquired viruses, this gets mentioned eventually with reference to the Table but the breakdown would be helpful right away.

We have added this information to the results section.

-"performed Spearman correlations", may be calculated or tested?

We have corrected the language for this sentence.

Comments on figures:

-Figure 1 data on linear scale (re discussion above) -- hard to even tell if there is a decay (to match with all we know from various long-term ART studies).

-Figure 4 data is shown on ln (log_e) scale, which is hard to interpret for most people.

-Figures 4 C,D, and E should have box plots to visually assess the significance.

-Figure 4B legend says purple/pink but I think the colors are different in the plot, could be about transparency

-Figure 5 it is now not clear if log_e(?).

-Figure 6 "HIV reservoir characteristics" might be better to make this more explicit. Do you mean for instance in the 6B title Total HIV DNA per million CD4+ T cells I think?

We have made these modifications.

Reviewer #2 (Recommendations For The Authors):

Minor Weaknesses:

(1) The Introduction is too long and much of the text is not directly related to the study's research question and design.

We have streamlined the introduction in the revised manuscript.

(2) While no differences were seen by age or race, according to the authors, this is unlikely to be useful since the numbers are so small in some of these subcategories. Results from sensitivity analyses (e.g., excluding these individuals) may be more informative/useful.

We agree that the lower numbers of participants for some subgroupings makes it challenging to know for sure if there are any differences based on these variables.  Have added text to clarify this. We have added age, race and gender to the LOCO analysis and to the variable inflation importance analysis (Table S5).

(3) For Figure 4, based on what was described in the Results section of the manuscript, the authors should clarify that the figures show results for TOTAL HIV DNA only (not intact DNA): "Dimension reduction machine learning approaches identified two robust clusters of PWH when using total HIV DNA reservoir-associated immune cell frequencies (Figure 4A), but not for intact or percentage intact HIV DNA (Figure 4B and 4C)".

We have added this information.

(4) The statement on page 5, first paragraph, "Interestingly, when we examined a plot of percent intact proviruses versus time on therapy (Figure 1C), we observed a biphasic decay pattern," is not new (Peluso JCI Insight 2020, Gandhi JID 2023, McMyn JCI 2023). Prior studies have clearly demonstrated this biphasic pattern and should be cited here, and the sentence should be reworded with something like "consistent with prior work", etc.

We have added citations to these studies and rephrased this comment.

(5) The Cohort and sample collection sections are somewhat thin. Further details on the cohort details should include at the very minimum some description of the timing of ART initiation (is this mostly a chronic-treated cohort?) and important covariate data such as nadir CD4+ T cell count, pre-ART viral load, duration of ART suppression, etc.

The cohort was treated during chronic infection, and we have clarified this in the manuscript.  Information regarding CD4 nadir and years on ART are included in Table 1.  Unfortunately, pre-ART viral load was not available for most members of this cohort, so we did not use it for analyses. The partial pre-ART viral load data is included with the dataset we are making publicly available.

Reviewer #3 (Recommendations For The Authors):

Minor points:

(1) What is meant by CD4 nadir? Is this during primary infection or the time before ART initiation?

We have clarified this description in the manuscript.  This term refers to the lowest CD4 count recorded during untreated infection.

(2) The authors claim that determinants of reservoir size are starting to emerge but other than the timing of ART, I am not sure what studies they are referring to.

We have updated the language of this section.  We intended to refer to studies looking at correlates of reservoir size, and feel that this is a more appropriate term that ‘determinants’

(3) The discussion does not tie in the model-generated hypotheses with the known mechanisms that sustain the reservoir: clonal proliferation balanced by death and subset differentiation. It would be interesting to tie in the proposed reservoir clusters with these known mechanisms.

We have added additional text to the manuscript to address these mechanisms.

(4) Figure 1: Total should be listed as total HIV DNA.

We have updated this in the manuscript.

(5) Figure 1C: Worth mentioning the paper by Reeves et al which raises the possibility that the flattening of intact HIV DNA at 9 years may be spurious due to small levels of misclassification of defective as intact.

We have added this reference.

(6) "Total reservoir frequency" should be "total HIV DNA concentration"

We respectfully feel that “frequency” is a more accurate term than “concentration”, since we are expressing the reservoir as a fraction of the CD4 T cells, while “concentration” suggests a denominator of volume.

(7) Figure S2-5: label y-axis total HIV DNA.

We have updated this figure.

Reviewer #1 (Public Review):

Summary:

Building upon their famous tool for the deconvolution of human transcriptomics data (EPIC), Gabriel et al. implemented a new methodology for the quantification of the cellular composition of samples profiled with Assay for Transposase-Accessible Chromatin sequencing (ATAC-seq). To build a signature for ATAC-seq deconvolution, they first created a compendium of ATAC-seq data and derived chromatin accessibility marker peaks and reference profiles for 12 cell types, encompassing immune cells, endothelial cells, and fibroblasts. Then, they coupled this novel signature with the EPIC deconvolution framework based on constrained least-square regression to derive a dedicated tool called EPIC-ATAC. The method was then assessed using real and pseudo-bulk ATAC-seq data from human peripheral blood mononuclear cells (PBMC) and, finally, applied to ATAC-seq data from breast cancer tumors to show it accurately quantifies their immune contexture.

Strengths:

Overall, the work is of very high quality. The proposed tool is timely; its implementation, characterization, and validation are based on rigorous methodologies and results in robust estimates. The newly-generated, validation data and the code are publicly available and well-documented. Therefore, I believe this work and the associated resources will greatly benefit the scientific community.

Weaknesses:

In the benchmarking analysis, EPIC-ATAC was compared also to deconvolution methods that were originally developed for transcriptomics and not for ATAC-seq data. However, the authors described in detail the specific settings used to analyze this different data modality as robustly as possible, and they discussed possible limitations and ideas for future improvement.

for - neuroscience - example - informal diversity - increases work efficacy - via sharing diverse and novel perspectives

kind of value that combines data and the code that operates on that data into a single unit.

set up the digit limit for double number

Welcome back, and in this lesson, I want to talk about AWS CloudFormation. I'm going to be brief because learning CloudFormation is something which will happen throughout the course, as we'll be using it to automate certain things. Before we dive in, I'll introduce the concepts you'll need and give you a chance to experience a simple practical example.

CloudFormation is a tool which lets you create, update, and delete infrastructure in AWS in a consistent and repeatable way using templates. Rather than creating and updating resources manually, you create a template, and CloudFormation will do the rest on your behalf.

At its base, CloudFormation uses templates. You can use a template to create AWS infrastructure using CloudFormation. You can also update a template and reapply it, which causes CloudFormation to update the infrastructure, and eventually, you can use CloudFormation to delete that same infrastructure. A CloudFormation template is written either in YAML or JSON. Depending on your experience, you might be familiar with one or both of these. If you haven't touched YAML or JSON before, don't worry. They achieve the same thing, and it's easy to convert between them. You might get to pick which one to use when writing templates, or your business might have a preference. It's mostly a matter of personal preference. Most people in the AWS space like one and dislike the other, though very few people like both. I am one of those who likes both. I started my AWS career using JSON but have come to appreciate the extra functionality that YAML offers. However, YAML can be easier to make mistakes with because it uses white spaces to indicate which parts belong to which other parts. Since spaces are not always visible, it can be a problem for less experienced engineers or architects. If I have to pick one, I'll use YAML. So for the rest of this lesson, I'll focus on YAML.

I want to quickly step through what makes a template, the components of a template, and then discuss the architecture of CloudFormation before moving on to a demo. All templates have a list of resources, at least one. The resources section of a CloudFormation template tells CloudFormation what to do. If resources are added, CloudFormation creates them. If resources are updated, CloudFormation updates them. If resources are removed from a template and that template is reapplied, then physical resources are removed. The resources section of a template is the only mandatory part of a CloudFormation template, which makes sense because without resources, the template wouldn't do anything. The simple template that we'll use in the demo lesson immediately following this one has resources defined in it, and we'll step through those and evaluate exactly what they do.

Next is the description section. This is a free text field that lets the author of the template add a description, as the name suggests. Generally, you would use this to provide details about what the template does, what resources get changed, and the cost of the template. Anything that you want users to know can be included in the description. The only restriction to be aware of is if you have both a description and an AWSTemplateFormatVersion, then the description needs to immediately follow the template format version. The template format version isn't mandatory, but if you use both, the description must directly follow the template format version. This has been used as a trick question in many AWS exams, so it pays to be aware of this restriction. The template format version allows AWS to extend standards over time. If it's omitted, the value is assumed.

The metadata in the template is the next part I want to discuss. It has many functions, including some advanced ones. For example, metadata can control how different elements in the CloudFormation template are presented through the console UI. You can specify groupings, control the order, and add descriptions and labels, which helps in managing how the UI presents the template. Generally, the bigger your template and the wider the audience, the more likely it is to have a metadata section. Metadata serves other purposes, which I'll cover later in the course.

The parameters section of a template allows you to add fields that prompt the user for more information. When applying the template from the console UI, you'll see boxes to type in or select from dropdowns. This can be used to specify things like the size of the instance to create, the name of something, or the number of availability zones to use. Parameters can have settings for valid entries and default values. You'll gain more experience with this as we progress through the course and use CloudFormation templates.

The next section is mappings, which is another optional section of the CloudFormation template and something we won't use as much, especially when starting with CloudFormation. It allows you to create lookup tables. For example, you can create a mappings table called RegionAndInstanceTypeToAMI, which selects a specific Amazon Machine Image based on the region and environment type (e.g., test or prod). This is something you'll get experience with as the course continues, but I wanted to introduce it at this point.

Next, let's talk about conditions. Conditions allow decision-making in the template, enabling certain things to occur only if a condition is met. Using conditions involves a two-step process. Step one is to create the condition. For instance, if a parameter is equal to "prod" (i.e., if the template is being used to create prod resources), then you create a condition called CreateProdResources. If the parameter "environment type" is set to "prod," the condition CreateProdResources will be true. Step two is using this condition within resources in the CloudFormation template. For example, a resource called Prodcatgifserver will only be created if the condition CreateProdResources is true. This will only be true if the "environment type" parameter is set to "prod" rather than "test." If it's set to "test," that resource won't be created.

Finally, outputs are a way for the template to present outputs based on what's being created, updated, or deleted once the template is finished. For example, outputs might return the instance ID of an EC2 instance that's been created, or if the template creates a WordPress blog, it could return the admin or setup address for that blog.

So, how exactly does CloudFormation use templates? CloudFormation starts with a template. A template contains resources and other elements you'll become familiar with as we use CloudFormation more. Let's take a simple example—a template that creates an EC2 instance. Resources inside a CloudFormation template are called logical resources. In this case, the logical resource is called "instance," with a type of AWS::EC2::Instance. The type tells CloudFormation what to create. Logical resources generally also have properties that CloudFormation uses to configure the resources in a specific way.

When you provide a template to CloudFormation, it creates a stack, which contains all the logical resources defined in the template. A stack is a living and active representation of a template. One template could create one stack, or several stacks, or anywhere in between. A stack is created when you tell CloudFormation to do something with that template.

For any logical resources in the stack, CloudFormation makes a corresponding physical resource in your AWS account. For example, if the stack contains a logical resource called "instance," which defines an EC2 instance, the physical resource is the actual EC2 instance created by CloudFormation. It's CloudFormation's job to keep the logical and physical resources in sync. When you use a template to create a stack, CloudFormation scans the template, creates a stack with logical resources, and then creates matching physical resources.

You can also update a template and use it to update the stack. When you do this, the stack's logical resources will change—new ones may be added, existing ones updated or deleted. CloudFormation performs the same actions on the physical resources, adding, updating, or removing them as necessary. If you delete a stack, its logical resources are deleted, leading CloudFormation to delete the matching physical resources.

CloudFormation is a powerful tool that allows you to automate infrastructure. For instance, if you host WordPress blogs, you can use one template to create multiple deployments rather than setting up each site individually. CloudFormation can also be part of change management, allowing you to store templates in source code repositories, make changes, get approval, and apply them as needed. It can also be used for one-off deployments.

Throughout this course, I'll be using CloudFormation to help you implement various things in demo lessons. If a demo lesson requires certain products to function, I might provide a CloudFormation template to set up the base infrastructure. Alternatively, you can use the template to implement the entire demo end-to-end. CloudFormation is super powerful, and you'll get plenty of exposure to it throughout the course.

Now, that's all the theory I wanted to cover. The next lesson will be a demo where you'll use CloudFormation to create an EC2 instance. Remember in the EC2 demo lesson, where you created an EC2 instance? In the next demo lesson, you'll create a similar EC2 instance using CloudFormation, demonstrating how much quicker and easier it is to automate infrastructure tasks with CloudFormation. So go ahead, complete this video, and when you're ready, join me in the next lesson where we'll demo CloudFormation.

à modifier

Reviewer #1 (Public Review):

Summary:

The authors applied a domain adaptation method using the principal of optimal transport (OT) to superimpose read count data onto each other. While the title suggests that the presented method is independent from and performs better than other methods of bias correction, the presented work uses a self-implemented version of GC bias correction apart of the OT domain adaptation. Performance comparisons were done both on normalized read counts as well as on copy number profiles which is already the complete set of presented use cases. Results involving copy number profiles from iChorCNA were also subjected to the bias correction measures implemented there. It is not clear at many points which correction method actually causes the observed performance.

Strengths:

The quality of superimposing distributions of normalized read counts (and copy number profiles) was sufficiently shown using uniformly distributed p-values in the interval of 0 to 1 for healthy controls D7 and D8 which differed in the choice of library preparation kit.

The ability to select a sample from the source domain for samples in the target domain was demonstrated.

Weaknesses:

Experiment Design:

The chosen bias correction methods are not explicitly designed for nor aimed at domain adaptation. The benchmark against GC bias correction while doing GC bias correction during the OT procedure is probably the most striking flaw of the entire work. GC bias correction has the purpose of correction GC biases, wherever present, NOT correcting categorical pre-analytical variables of undefined character. A more thorough examination of the presented results should address why plain iChor CNA is the best performing "domain adaptation" in some cases. Also, the extent to which the implemented GC bias correction is contributing to the performance increase independent of the OT procedure should be assessed separately in each case.<br /> Moreover, the center-and-scale standardization is probably not the most relevant contestant in domain adaptation that is out there.

Comparison of cohorts (domains) - especially healthy from D7 and D8 - it is not described which type of ChIP analysis was done for the healthy controls of the D7 domain. The utilized library preparation kit implies that D7 represents a subset of available cfDNA in a plasma sample by precipitating only certain cfDNA fragments to which undisclosed type of protein was bound. Even if the type of protein turns out to be histones, the extracted subset of cfDNA should not be regarded as coming from the same distribution of cfNDAs. For example, fragments with sub-mononucleosomal length would be depleted in the ChIP-seq data set while these could be extracted in an untargeted cfDNA sequencing data set. It needs to be clarified why the authors deem D7 and D8 healthy controls to be identical with regards to SCNA analysis. Best start with the protein targets of D7 ChIP-seq samples.

From the Illumina TruSeq ChIP product description page:<br /> "TruSeq ChIP Libary Preparation Kits provide a simple, cost-effective solution for generating chromatin immunoprecipitation sequencing (ChIP-Seq) libraries from ChIP-derived DNA. ChIP-seq leverages next-generation sequencing (NGS) to quickly and efficiently determine the distribution and abundance of DNA-bound protein targets of interest across the genome."

Redundancy:

Some parts throughout the results and discussion part reappear in the methods. The description of the methodology should be concentrated in the method section and only reiterated in a summarizing fashion where absolutely necessary.<br /> Unnecessary repetition inflate the presented work which is not appealing to the reader. Rather include more details of the utilized materials and methods in the corresponding section.

Transparency:

At the time point of review, the code was not available under the provided link.<br /> A part of the healthy controls from D8 is not contained under the provided accession (367 healthy samples are available in the data base vs. sum of D7 and D8 healthy controls is 499)

Neither in the paper nor in reference 4 is an explanation of what was targeted with the ChIP-seq approach.

Consistency:

It is not evident why a ChIP-seq library prep kit was used (sample cohorts designated as D7). The DNA isolation procedure was not presented as having an immunoprecipitation step. Furthermore, it is not clear which DNA bound proteins were targeted during ChIP seq, if such an immunoprecipitation was actually carried out.The authors self-implemented a GC bias correction procedure although they already mentioned other procedures earlier like LIQUORICE. Also, there already exist tools that can be used to correct GC bias, like deepTools (github.com/deeptools/deepTools). Other GC bias correction algorithms designed specifically for cfDNA would be Griffin (github.com/adoebley/Griffin) and GCparagon (github.com/BGSpiegl/GCparagon). When benchmarking against state-of-the-art cfDNA GC bias correction, these algorithms should appear in a relevant scientific work, somewhere other than the introduction, preferably in the results section. It should be shown that the chosen GC bias correction method is performing best under the given circumstances.

Accuracy:

Use clear labels for each group of samples. The domain number is not sufficient to effectively distinguish sample groups. Already the source name plus a simple enumeration would improve the clarity at some points.

The healthy controls of D7 and D8 are described but the numbers do not add up (257 healthy controls in line 227 vs. 260 healthy controls in line 389). Please double check this and use representative sample cohort labels in the materials description for improved clarity!

Avoid statements like "the rest" when talking about a mixed set of samples. It is not clear how many samples from which domain are addressed.

For optimal transport, knowledge about the destination is required ("where do I want to transport to?") and, thus, the proposed method can never be unsupervised. It is always necessary to know the label of both the source and target domains. In practice, this is not often the case and users might fall prey to the error of superimposing data that is actually separated by valid differences in some experimental variables.

Seemingly arbitrary cutoff values are mentioned. For example, it is not clear if choosing "the cutoff that produced the highest MCCs" is meant across methods or for each method separately (are the results for each method reported that also resulted in the highest MCC for that method?).

The Euclidean metric for assessing the similarity of (normalized) read counts is questionable for a high dimensional space: read counts are assessed for 1 Mb genomic intervals which yields around 3000 intervals (dimensions), depending on the number of excluded intervals (which was not described in more detail). There might be more appropriate measures in this high dimensional space.

It is sometimes not clear what data actually is presented. An example would be the caption of Figure 2, (C): it is suggested that all (320) ovarian cancer cases are shown in one copy number profile.

Furthermore, the authors do not make a distinction between male and female samples. A clarification is needed why the authors think SCNAs of ovarian cancer samples should be called against a reference set that contains male controls.<br /> The procedure would likely benefit from a strict separation of male and female cases which would also allow for chrX (and chrY) being included in downstream analysis.

The GC bias and mappability correction implicitly done by iChorCNA for the SCNA profile comparison is presented as "no correction" which is highly misleading. (for clarification, this is also deemed inappropriate, not just inaccurate))

The majority of interpretations presented procedure does not give any significant improvement regarding the similarity of copy number profiles are off and in many instances favor the OT procedure in an unscientific and highly inappropriate manner.

Apart of duplicate marking (which is not specified any further - provide the command(s)!), there is no information on which read (pairs) were used (primary, secondary, supplementary, mapped in a proper pair, fragment length restrictions, clipping restrictions, etc.). The authors should explain why base quality score re-calibration was done as this might be an unnecessary step if the base quality values are not used later on.

The adaptation method presented as "center-and-scale standardization" is inappropriate for unbalanced cancer profiles since it assumes the presence of identical SCNAs in all samples belonging to the same cancer entity.<br /> Please explain why normalizing 1 Mb genomic intervals to the average copy number across different cancer samples should be valid or use another domain adaptation method for performance comparison.

Statements like in line 83 (unsupervised DA) are plain wrong because transport from one domain to another requires the selection of a target domain based on a label, e.g., based on health status, cancer entity, or similar.

Relevance and Appropriateness:

Many of the presented results are not relevant or details of the procedure were incomprehensible or incomplete: the results presented in table 2 - sample assignment. The Euclidean metric seems to be inappropriate for high dimensional data. Also the selection of the cutoff based on Euclidean distance seems to enable the optimization in favor of the OT procedure. It is hypothesized that there might exist other cutoff values for which the selection of samples form the source domain would also work for other correction methods but this is not further described. It could simply be the case that OT can assign a relationship between domains

The statement that there are no continuous pre-analytical variables is wrong (304). The effect of target depth-of-coverage (DoC) was not analyzed although this represents one of the most common (continuous) and difficult to control variables in NGS data analysis. The inclusion of multiple samples from a single patient in a cohort likely represents introduction of a confounding factor ["contamination"] to the model training procedure: the temporal difference that lies between the taken samples of that patient represents leakage of information. As far as can be told from the presented data, this potential bias has not been ruled out (e.g., exclusion of all samples beyond the first from each patient or alternatively: picking all samples of a patient either for the training set or the test set).

Conscientiousness:

Statements like "good"/"best" on their own should be avoided. A clear description of why a certain procedure/methodology/algorithm performs better should be preferred in scientific writing (e.g., "highest MCC values" instead of "best MCC values").<br /> Otherwise, such statements represent mere opinions of the author rather than an unbiased evaluation of the results.<br /> The domain D8 of healthy controls seems to contain samples from multiple sources (some published other in-house). Contrary to the data availability statement (533), not all healthy control samples of the HEMA data set are available from ArrayExpress

Other Major Concerns:

Potential Irrelevance:

The manuscript represents a mere performance assessment of the proposed sWGS per-bin-read-count fitting procedure and, thus, a verification in its character, not a validation (although the model training itself was "validated" - but this is to be viewed separately from the validity of the achieved correction in a biological context). A proper (biological) validation is missing.

It is of utmost importance that parameters of the adapted (transported) samples -that lie outside of what has been optimized to be highly similar- are checked to actually validate the procedure. Especially biological signals and genome-wide parameters (GC content distribution before/after transport) need to be addressed also in hindsight of the rampant criticism towards GC bias correction by the authors. At no point in the manuscript was GC bias addressed properly, i.e., how much of an improvement is expected from GC bias correction if there is no significant GC bias?

The (potential - not clear so far) ability of making ChIP-seq data look like cfDNA data (even if only the copy number profiles SCNAs appear highly similar) raises the concern of potential future users of the tool to superimpose domains that should not be superimposed form a biological point of view because the true domain the superimposed cohorts belong to are different. The ability to superimpose anything onto anything s troubling. There is no control mechanism that allows for failure in cases where the superposition is invalid.

Chromosome X was excluded which could be avoided if data sets were split according to biological sex.

The difference between the distributions was never attributed to GC bias, hence, the benchmark against GC bias correction tools might not be relevant in the first place.

Stability of OT data transformation:

The authors state that the straight forward choice of lambda resulted in many occasions where disruptions (of unspecified nature and amplitude) are introduced in the copy number profiles of transformed data. It is not evident from the proposed work to which extent this behavior was removed from the procedure and if it can occur and how the user could resolve such a problem on their own.

In summary, the presented work needs considerable adaptation and additions before it can actually be considered a valuable contribution to the liquid biopsy field.