It's interesting to see he buys properties and flips it for profit.

Everyone's a winner?

In the classic children’s book Alice’s Adventures in Wonderland, a very chaotic race takes place. When it finishes, no one knows who has won. To clear up the confusion, one of the characters proposes a startling solution: ‘EVERYBODY has won the race, and all must have prizes’. Obviously, the author wrote this scene for its comic effect. However, today, there is a growing movement in several countries to organise children’s sport so that there are no winners or losers. In Canada, for example, a regional football organisation has decided that in the future, in matches between under-12 teams, no one will keep score. Although this decision might seem strange, there is some interesting thinking behind it.

Supporters of the idea maintain that competitive sport puts some children off exercise forever because of the intense pressure to win. Sport for children, they argue, should be fun and not about winning or losing. Another problem occurs with children who aren’t very sporty. They end up losing most of the time and feel they have let the rest of the team down. There is a real possibility that these children will develop a negative self-image which will possibly stay with them the rest of their lives. Competitive sport can also encourage kids to think of their classmates as ‘winners’ or ‘losers’ in general. These are clearly not the values we want to communicate to young children. Finally, when beating your opponent becomes the main objective in sport, there is always a danger that some children are going to want to win at any cost and will cheat.

Not everyone, of course, is in favour of sport without winners and losers. Many people maintain that losing actually builds character because it encourages you to get over disappointments and try harder. It’s also true that an element of competition is present in many aspects of life, such as doing well in exams or getting a job, and competitive sport prepares young people for these challenges. Sport is also more exciting and challenging when there is a risk of losing. In addition, for children who don’t do well in other school subjects, sport can be their one opportunity to be really good at something. Do we really want to take this opportunity away from them?

In Canada, and in other countries, more and more organisations are experimenting with non-competitive sport and it appears to be taking off. However, not all the kids are crazy about it. For example, in games which are in theory non-competitive, players often shout out the score when the ball goes in the back of the net and they celebrate victories on the pitch at the end of the match!

In the end, it’s quite tricky to come to a clear conclusion about this new version of children’s sport. Kids are different, so non-competitive sport will work well with some and not with others. Perhaps a bigger question is: Can we really talk about ‘sport’ when there is no element of competition? Isn’t sport without competition just exercise?

This sentence underscores the fundamental importance of inclusive design, reminding us that overlooking diverse user needs can lead to products that unintentionally marginalize entire groups. It highlights why diversity in design teams isn’t just a buzzword—it’s essential to creating technology that truly serves a broad spectrum of society.

It is almost worrisome how advanced technology is becoming. Some people that I know have switched majors because Ai could take over the job they are looking for.

our case is different because it was a CHATROOM-- explicit purpose is to yap-- not private, it's literally communicating. same result could've been leaked by an individual in the chatroom. just happened to be leaked by isp.

This made me think about how I’ve encountered situational disabilities in my own life. For example, trying to use a smartphone in bright sunlight when the screen becomes unreadable is a form of situational disability. Similarly, being in a loud space where I can't hear a conversation well might resemble the experience of someone with hearing loss, even if it's only temporary. It’s a reminder that disability is fluid and context-dependent, not just a fixed identity that applies to a specific group of people.

I used to use index cards when preparing speeches. I used to write everything out and down then I would break it all down into the points, and ideas. It eventually got easier to build a skeleton so that I could give the speech and not forget what I needed to say. Just took a ton of practice at least for me.

I didn't know CWI offered so many recourses of information for study and research. I think it's amazing how technology has allowed us to just be able to pull something up and be able to use it in that moment.

Author response:

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

Public Reviews:  

Reviewer #1 (Public Review):  

Summary:  

The study by Pudlowski et al. investigates how the intricate structure of centrioles is formed by studying the role of a complex formed by delta- and epsilon-tubulin and the TEDC1 and TEDC2 proteins. For this, they employ knockout cell lines, EM, and ultrastructure expansion microscopy as well as pull-downs. Previous work has indicated a role of delta- and epsilon-tubulin in triplet microtubule formation. Without triplet microtubules centriolar cylinders can still form, but are unstable, resulting in futile rounds of de novo centriole assembly during S phase and disassembly during mitosis. Here the authors show that all four proteins function as a complex and knockout of any of the four proteins results in the same phenotype. They further find that mutant centrioles lack inner scaffold proteins and contain an extended proximal end including markers such as SAS6 and CEP135, suggesting that triplet microtubule formation is linked to limiting proximal end extension and formation of the central region that contains the inner scaffold. Finally, they show that mutant centrioles seem to undergo elongation during early mitosis before disassembly, although it is not clear if this may also be due to prolonged mitotic duration in mutants.  

Strengths:  

Overall this is a well-performed study, well presented, with conclusions mostly supported by the data. The use of knockout cell lines and rescue experiments is convincing.  

Weaknesses:  

In some cases, additional controls and quantification would be needed, in particular regarding cell cycle and centriole elongation stages, to make the data and conclusions more robust. 

We thank the reviewer for these comments and have improved our analyses of these as detailed below.

Reviewer #2 (Public Review):  

Summary:  

In this article, the authors study the function of TEDC1 and TEDC2, two proteins previously reported to interact with TUBD1 and TUBE1. Previous work by the same group had shown that TUBD1 and TUBE1 are required for centriole assembly and that human cells lacking these proteins form abnormal centrioles that only have singlet microtubules that disintegrate in mitosis. In this new work, the authors demonstrate that TEDC1 and TEDC2 depletion results in the same phenotype with abnormal centrioles that also disintegrate into mitosis. In addition, they were able to localize these proteins to the proximal end of the centriole, a result not previously achieved with TUBD1 and TUBE1, providing a better understanding of where and when the complex is involved in centriole growth.  

Strengths:  

The results are very convincing, particularly the phenotype, which is the same as previously observed for TUBD1 and TUBE1. The U-ExM localization is also convincing:

despite a signal that's not very homogeneous, it's clear that the complex is in the proximal region of the centriole and procentriole. The phenotype observed in U-ExM on the elongation of the cartwheel is also spectacular and opens the question of the regulation of the size of this structure. The authors also report convincing results on direct interactions between TUBD1, TUBE1, TEDC1, and TEDC2, and an intriguing structural prediction suggesting that TEDC1 and TEDC2 form a heterodimer that interacts with the TUBD1- TUBE1 heterodimer.  

Weaknesses:  

The phenotypes observed in U-ExM on cartwheel elongation merit further quantification, enabling the field to appreciate better what is happening at the level of this structure.  

We thank the reviewer for these comments and have improved our analyses of cartwheel elongation as detailed below.

Reviewer #3 (Public Review):  

Summary:  

Human cells deficient in delta-tubulin or epsilon-tubulin form unstable centrioles, which lack triplet microtubules and undergo a futile formation and disintegration cycle. In this study, the authors show that human cells lacking the associated proteins TEDC1 or TEDC2 have these identical phenotypes. They use genetics to knockout TEDC1 or TEDC2 in p53negative RPE-1 cells and expansion microscopy to structurally characterize mutant centrioles. Biochemical methods and AlphaFold-multimer prediction software are used to investigate interactions between tubulins and TEDC1 and TEDC2.  

The study shows that mutant centrioles are built only of A tubules, which elongate and extend their proximal region, fail to incorporate structural components, and finally disintegrate in mitosis. In addition, they demonstrate that delta-tubulin or epsilon-tubulin and TEDC1 and TEDC2 form one complex and that TEDC1 TEDC2 can interact independently of tubulins. Finally, they show that the localization of four proteins is mutually dependent.  

Strengths:  

The results presented here are mostly convincing, the study is exciting and important, and the manuscript is well-written. The study shows that delta-tubulin, epsilon-tubulin, TEDC1, and TEDC2 function together to build a stable and functional centriole, significantly contributing to the field and our understanding of the centriole assembly process.  

Weaknesses:  

The ultrastructural characterization of TEDC1 and TEDC2 obtained by U-ExM is inconclusive. Improving the quality of the signals is paramount for this manuscript.  

We thank the reviewer for these comments and have improved our imaging of TEDC1 and TEDC2 localization, as detailed below.

Recommendations for the authors:

Reviewing Editor (Recommendations For The Authors):  

The reviewers agreed that the conclusions are largely supported by solid evidence, but felt that improving the following aspects would make some of the data more convincing:  

(1) The UExM localizations of TEDC1/2 are not very convincing and the reviewers suggest to complement these with alternative super-resolution approaches (e.g. SIM) and/or different labeling techniques such as pre-expansion labeling using STAR red/orange secondaries (also robust for SIM and STED), use of the Halo tag, different tag antibodies, etc 

We thank the reviewers for these recommendations and have adapted two of these strategies to improve our imaging of TEDC1 and TEDC2 localization. First, we used an alternative super-resolution approach, a Yokogawa CSU-W1 SoRA confocal scanner (resolution = 120 nm) and imaged cells grown on coverslips (not expanded). We found that TEDC1 and TEDC2 localize to procentrioles and the proximal end of parental centrioles (Fig 2 – Supplementary Figure 1a, b). Second, we used a recently described expansion gel chemistry (Kong et al., Methods Mol Biol 2024) combined with Abberior Star red and orange secondary antibodies. This technique resulted in robust signal at centrosomes and in the cytoplasm and indicated that TEDC1 and TEDC2 localize near the centriole walls of procentrioles and the proximal region of parental centrioles, near CEP44 (Fig 2 – Supplementary Figure 1c, d). These results complement and support our initial observations (Fig 2C, D) and we have edited the text to reflect this (lines 157-163). We also note that these Flag tag and V5 tag primary antibodies are specific and have little background signal in all applications (Fig 2 – Supplementary Fig 1E-J), while other commercially available antibodies against these tags did exhibit non-specific signal. 

(2) The cell cycle classifications of centrioles would strongly benefit, apart from a better description, from adding quantifications of average centriole length at a given stage based on tubulin staining (not acTub). 

We thank the reviewers for these recommendations. We have added an improved description of our cell cycle analyses (lines 234-237). We have also added new analyses for centriole length as measured by staining with alpha-tubulin (Fig 4 – Supp 3 and Fig 4 – Supp 4). We find that in all mutants, acetylated tubulin elongates along with alpha-tubulin in a similar way as control centrioles.

Reviewer #1 (Recommendations For The Authors):  

Specific points:  

(1) The introduction is a bit oddly structured. About halfway through it summarizes what is going to be presented in the study, giving the impression that it is about to conclude, but then continues with additional, detailed introduction paragraphs. Overall, the authors may also want to consider making it more concise.

We thank the reviewer for these suggestions and have shortened and restructured the introduction for clarity and conciseness.

(2) The text should explain to the non-expert reader why endogenous proteins are not detected and why exogenously expressed, tagged versions are used. Related to this, the authors state overexpression, but what is this assessment based on? Does expression at the endogenous level also rescue? At least by western blot, these questions should be addressed. 

In the text, we have added clarification about why endogenous proteins were not detected for immunofluorescence (lines 149-151). To quantify the overexpression, we have added Western blots of TEDC1 and TEDC2 to Fig 1 – Supplementary Figure 1E,F. We note that endogenous levels of both proteins are very low, and the rescue constructs are overexpressed 20 to 70 fold above endogenous levels.  

(3) The figures should clearly indicate when tagged proteins are used and detected.

Currently, this info is only found in the legends but should be in the figure panels as well. 

We have made these changes to the figure panels in Fig 2, Fig 2 – Supp 1, and Fig 3.

(4)  I could not find a description and reference to Figure 2 Supplement 2 and 3. 

We have replaced these supplements with new supplementary figures for TEDC1 and TEDC2 localization (Fig 2 – Supp 1).

(5) The multiple bands including unspecific (?) bands should be labeled to guide the reader in the western blots. 

We have labeled nonspecific bands in our Western blots with asterisks (Fig 1 – Supp 1, Fig 3)

(6) The alphafold prediction suggests that TUBD1 can bind to the TED complex in the absence of TUBE1 can this be shown? This would be a nice validation of the predicted architecture of the complex. I also missed a bit of a discussion of the predicted architecture. How could it be linked to triplet microtubule formation? Is the latest alphafold version 3 adding anything to this analysis? 

In our pulldown experiments, we found that TUBD1 cannot bind to TEDC1 or TEDC2 in the absence of TUBE1 (Fig 3C, D, IB: TUBD1). We performed this experiment with three biological replicates and found the same result. It is possible that TUBD1 and TUBE1 form an intact heterodimer, similar to alpha-tubulin and beta-tubulin, and this will be an exciting area of future research.

We have added new analysis from AlphaFold3 (Fig 3 – Supp 1B). AlphaFold3 predicts a similar structure as AlphaFold Multimer.

We have also added additional discussion about the AlphaFold prediction to the text (lines 220-222, 365-367). Thanks to the reviewer for pointing out this oversight.

(7) I suggest briefly explaining in the text how cells and centrioles at different cell cycle stages were identified. I found some info in the legend of Figure 1, but no info for other figures or in the text. Related to this, how are procentrioles defined in de novo formation? There is no parental centriole to serve as a reference. 

We have added a brief explanation of the synchronization and identification in lines 234237. We have also clarified the text regarding de novo centrioles, and now term these “de novo centrioles in the first cell cycle after their formation” (lines 271-272).

(8) Related to point 7: using acetylated tubulin as a universal length and width marker seems unreliable since it is a PTM. The authors should use general tubulin staining to estimate centriole dimensions, or at least establish that acetylated tubulin correlates well with the overall tubulin signal in all mutants. 

We have added two supplementary data figures (Fig 4 – supp 3 and Fig 4 – supp 4) in which we co-stain control and mutant centrioles with alpha-tubulin. We found that acetylated tubulin marked mutant centrioles well and as alpha-tubulin length increased, acetylated tubulin length also increased. 

(9) Presence and absence of various centriolar proteins. These analyses lack a clear reference for the precise centriole elongation stage. This is particularly problematic for proteins that are recruited at specific later stages (such as inner scaffold proteins). The staining should be correlated with centriole length measurements, ideally using general tubulin staining.  

As described for point 8, we have added two supplementary data figures in which we costain control and mutant centrioles with alpha-tubulin and found that acetylated tubulin also increases as overall tubulin length increases in all mutants. We note that inner scaffold proteins are absent in all our mutant centrioles at all stages of the cell and centriole cycle, as also previously reported for POC5 in Wang et al., 2017.

Reviewer #2 (Recommendations For The Authors):  

Here's a list of points I think could be improved:  

-  As the authors previously published, the centriole appears to have a smaller internal diameter than mature centrioles. Could the authors measure to see if the phenotype is identical? Is the centriole blocked in the bloom phase (Laporte et al. 2024)? 

We have added an additional supplementary figure (Fig 4 – supp 5) to show that mutant centrioles have smaller diameters than mature centrioles, as we previously reported for the delta-tubulin and epsilon-tubulin mutant centrioles by EM. We thank the reviewers for the additional question of the bloom phase. Given the comparatively smaller number of centrioles we analyzed in this paper compared to Laporte et al (50 to 80 centrioles per condition here, versus 800 centrioles in Laporte et al), it is difficult to definitively conclude whether there is a block in bloom phase. This would be an interesting area for future research.  

-  The images of the centrioles in EM are beautiful. Would it be possible to apply a symmetrisation on it to better see the centriolar structures? For example, is the A-C linker present? 

We thank the reviewer for this excellent suggestion. Using centrioleJ, we find that the A-C linker is absent from mutant centrioles. The symmetrized images have been added to Fig 1 – Supplementary Fig 2, and additional discussion has been added to the text (line 143-144, line 368-374).  

-  How many EM images were taken? Did the centrioles have 100% A-microtubule only or sometimes with B-MT? 

For TEM, we focused on centrioles that were positioned to give perfect cross-section images of the centriolar microtubules, and thus did not take images of off-angle or rotated centrioles. Given the difficulty of this experiment (centrioles are small structures within the cell, centrosomes are single-copy organelles, and off-angle centrioles were not imaged), we were lucky to image 3 centrioles that were in perfect cross-section – 2 for Tedc1^-/- and 1 for Tedc2^-/-. Our images indicate that these centrioles only have A-tubules (Fig 1 – Supp Fig

2).

-  In Figure 2 - it would be preferable to write TEDC2-flag or TEDC1-flag and not TEDC2/1. 

We have made this change

-  It seems that Figures 2C and D aren't cited, and some of the data in the supplemental data are not described in the main text. 

We have replaced these supplements with new supplementary figures for TEDC1 and TEDC2 localization (Fig 2 – Supp 1).

-  The signal in U-ExM with the anti-Flag antibody is heterogeneous. Did the authors test several anti-FLAG antibodies in U-ExM? 

We tested several anti-Flag and anti-V5 antibodies for our analyses, and chose these because they have little background signal in all applications (Fig 2 – Supplementary Fig 1E-J). Other commercially available antibodies against these tags did exhibit non-specific signal.

-  The AlphaFold prediction is difficult to interpret, the authors should provide more views and the PDB file. 

We have added 2 additional views of the AlphaFold prediction in Fig 3 – Supp 1A.

-  In general, but particularly for Figure 4: the length doesn't seem to be divided by the expansion factor, it is therefore difficult to compare with known EM dimensions. Can the authors correct the scale bars? 

We have corrected the scale bars for all figures to account for the expansion factor.

-  Concerning Gamma-tubulin that is "recruited to the lumen of centrioles by the inner scaffold, had localization defects in mutant centrioles. However, we were unable to reliably detect gamma-tubulin within the lumen of control or de novo-formed centrioles in S or G2-phase (Figure 4 - Supplement 1E), and thus were unable to test this hypothesis". In Laporte et al 2024, Gamma-tubulin arrives later than the inner scaffold and only on mature centrioles, so this result appears to be in line with previous observation. However, the authors should be able to detect a proximal signal under the microtubules of the procentriole, is this the case? 

We agree that this is an exciting question. However, in our expansion microscopy staining, we frequently observe that gamma-tubulin surrounds centrioles, corresponding to its role in the pericentriolar material (PCM). In our hands, we find it difficult to distinguish between centriolar gamma-tubulin at the base of the A-tubule from gamma-tubulin within the PCM.  

-  In the signal elongation of SAS-6, STIL, CEP135, CPAP, and CEP44, would it be possible to quantify the length of these signals (with dimensions divided by the expansion factor for comparison with known TEM distances)? 

We have quantified the lengths of SAS-6 and CEP135 in new Fig 4 – Supp 3 and Fig 4 – Supp 4.  

-  The authors observe that centrin is present, but only as a SFI1 dot-like localization (which is another protein that would be interesting to look at), and not an inner scaffold localization. Can the authors elaborate? These results suggest that the distal part is correctly formed with only a microtubule singlet. 

We agree with the reviewer’s interpretation that the centriole distal tip is likely correctly formed with only singlet microtubules, as both distal centrin and CP110 are present. We have added this point to the discussion (line 415).

-The authors observe that CPAP is elongated, but CPAP has two locations, proximal and distal. Is it distal or proximal elongation? Is the proximal signal of CPAP longer than that of CEP44 in the mutants? The authors discuss that the elongation could come from overexpression of CPAP, but here it seems that the centriole is not overlong, just the structures around the cartwheel. 

We thank the reviewer for this point. It is difficult for us to conclude whether the proximal or distal region is extended in the mutants, as our mutant centrioles lacks a visible separation between these two regions. It would be interesting to probe this question in the future by testing whether subdomains of CPAP may be differentially regulated in our mutants.

Reviewer #3 (Recommendations For The Authors):  

It isn't apparent to me what was counted in Figure 1C. Were all centrioles (mother centrioles and procentrioles) counted? Where is the 40% in control cells coming from? Can this set of data be presented differently? 

We apologize for the confusion. In this figure, all centrioles were counted. We have updated the figure legend for clarity. We performed this analysis in a similar way as in Wang et al., 2017 to better compare phenotypes.  

Figure 2C. and the text lines 182-187: The ultrastructural characterization of TEDC1 and TEDC2 suffers from the low quality of the TEDC1 and TEDC2 signals obtained postexpansion. In comparison with robust low-resolution immunosignal, it appears that most of the signal cannot be recovered after expansion. Another sub-resolution imaging method to re-analyze TEDC1 and TEDC22 localization would be essential. The same concern applies to Figures 2 - Supplement 2 and 3. Also, Figure 2 - Supplement 2 and Supplement 3 do not seem to be cited. 

We thank the reviewer for these recommendations. As also mentioned above, we used an alternative super-resolution approach, a Yokogawa CSU-W1 SoRA confocal scanner (resolution = 120 nm), and found that TEDC1 and TEDC2 localize to procentrioles and the proximal end of parental centrioles (Fig 2 – Supplementary Figure 1a, b). Second, we used a recently described expansion gel chemistry (Kong et al., Methods Mol Biol 2024) combined with Abberior Star red and orange secondary antibodies. This technique resulted in robust signal at centrosomes and in the cytoplasm and indicated that TEDC1 and TEDC2 localize near the centriole walls of procentrioles and the proximal region of parental centrioles, near CEP44 (Fig 2 – Supplementary Figure 1c, d). These stainings complement and support our initial observations (Fig 2C, D) and we have edited the text to reflect this (lines 157-163). We have also removed the supplementary figures that were uncited in the text.

TUBD1 and TUBE1 form a dimer and TEDC2 and TEDC1 can interact. Any speculation as to why TEDC2 does not pull down both TUBE1 and TUBD1? 

We apologize for the confusion. TEDC2 does pull down both TUBE1 and TUBD1 (Fig 3D, pull-down, second column, Tedc2-V5-APEX2 rescuing the Tedc2^-/- cells pulls down TUBD1, TUBE1, and TEDC1).  

Figure 4A and B. The authors use acetylated tubulin to determine the length of procentrioles in the S and G2 phases. However, procentrioles are not acetylated on their distal ends in these cell phase phases (as the authors also mention further in the text). Why has alpha tubulin not been used since it works well in U-ExM? The average size of the control, G2 procentrioles, seems too small in Figure 4A and not consistent with other imaging data (for instance, in Figure 4 - Supplement 1 C, Cp110, and CPAP staining). There is no statistical analysis in F4A.  

We have added two supplementary data figures (Fig 4 – supp 3 and Fig 4 – supp 4) in which we co-stain control and mutant centrioles with alpha-tubulin. We found that acetylated tubulin correlates well with overall tubulin signal in all mutants. We have added statistical analysis to the figure legend of Fig 4A.

Lines 260 - 262: "These results indicate that centrioles with singlet microtubules can elongate to the same length as controls, and therefore that triplet microtubules are not essential for regulating centriole length." It is hard to agree with this statement. Mutant procentrioles show aberrantly elongated proximal signals of several tested proteins. In addition, in lines 326 - 328, the authors state that "Together, these results indicate that centrioles lacking compound microtubules are unable to properly regulate the length of the proximal end."  

We thank the reviewer and have clarified the statement to state that these results indicate that centrioles with singlet microtubules can elongate to the same overall length as control centrioles in G2 phase.  

Line 353: The authors suggest that elongated procentriole structure in mitosis may represent intermediates in centriole disassembly. Another interpretation, more in line with the EM data from Wang et al., 2017, would be that these mutant procentrioles first additionally elongate before they disassemble in late mitosis. The aberrant intermediate structure concept would need further exploration. For instance, anti-alpha/beta-tubulin antibodies could be used to investigate centriole microtubules.  

We apologize for the confusion and have edited this section for clarity (lines 341-343): “We conclude that in our mutant cells, centrioles elongate in early mitosis to form an aberrant intermediate structure, followed by fragmentation in late mitosis.”

References need to be included in lines 122, 277, 279. 

We have added these references

Line 281: Add references PMID: 30559430 and PMID: 32526902.  

We have added these references (lines 265-266).

Line 289: "Moreover, our results suggest that centriole glutamylation is a multistep process, in which long glutamate side chains are added later during centriole maturation." This does not seem like an original observation. For instance, see PMID: 32526902.  

We have added this reference (lines 273-274).

Milton expresses sorrow not just for Lycidas but for the loss of a fellow poet. When he asks, "Who would not sing for Lycidas?", it’s not just rhetorical, it shows his own deep sense of obligation to memorialize his friend through poetry. This suggests that writing the elegy is both a duty and a way to process his own grief.

I think this idea is important because it shows that literacy isn’t just about reading and writing—it’s also shaped by people’s backgrounds, experiences, and communities. Learning doesn’t happen in isolation; it’s influenced by culture, language, and daily life. This makes me think that schools should recognize different ways students engage with literacy instead of just focusing on traditional reading and writing skills. Understanding literacy as a social practice could help make education more inclusive and meaningful.

It's not a very important point of observation, but I just found "Fort Orange" to be an odd name. Apparently, it's named after the royal family of the Netherlands (House of Orange). I assume that French settlers deciding on this name ties into their goals to expand trade in the region, as discussed in the previous chapter.

Just because a lot of things have been normalized, it doesn’t mean that they are right. Being overweight is normal as long as it doesn’t affect your health and life. On the other hand, having high levels of cholesterols is a risk factor for a wide variety of diseases that can lead to death. Therefore high cholesterol levels should be taken seriously and interventions should be done. I understand that some people might prefer a non-pharmaceutical approach which is perfectly fine as long as it’s followed correctly. Exercise, balanced diets and avoiding other dangerous behaviours are just some of the non-pharmaceutical interventions that can have an impact on decreasing these levels. This is assuming that the cholesterol level being spoken about is LDL.

This passage highlights how the Black struggle for human rights has not only advanced Black communities but has also contributed to the broader humanization of the U.S. As a Latino, I see parallels in our own fight for civil rights and recognition. Just like Black activism shaped the country, Latinx movements—like the Chicano movement or immigrant rights struggles—have pushed for justice and inclusion. It’s important to revisit past victories, but also to reflect on the challenges that remain.

This paragraph shows how important it is for teachers to support bilingual students instead of forcing them to stick to just one language. Ms. López understands that language is not just a rule to follow but a tool for learning. By letting Yamaira use Spanish, she makes history more accessible and meaningful. I also think it’s great that this doesn’t just help Yamaira—it changes the whole class. It proves that when students are allowed to use their full language skills, they can actually contribute more, not less.

The thought just never comes into his head that maybe people don’t like to stare at people because they’re also shy and there’s a bunch of new people around -or that it’s normal unless you engage someone for them to not pay attention to you, no matter how you look. his ability to rationalize was so low I think because his fear of shame was so intense, his worry that his fears of inferiority were really true. he just so easily collapsed into vulnerability.

He has such a brutal pattern of not socializing, not putting in effort, avoiding stuff, and then using the subpar result to humiliate himself with…

He was trapped in this self-destructive loop because his false self didn’t allow him to acknowledge his own role in his suffering. Vulnerable narcissism creates a deep split: on one hand, he saw himself as special and deserving of effortless admiration, but on the other, he felt completely worthless and humiliated when that admiration didn’t come. To admit that his isolation was partly his own doing would mean facing the painful reality that he wasn’t superior or owed automatic acceptance, something his false self couldn’t tolerate. Instead, he stayed in a cycle where he avoided effort, set himself up for failure, and then used that failure as proof that the world was cruel and he was a hopeless victim. What was missing was emotional responsibility the ability to see that his pain wasn’t just inflicted on him by others, but also reinforced by his own avoidance and expectations. But his narcissism likely blocked that realization, because accepting it would have shattered the fantasy that he was meant for greatness but unfairly denied it.

His only identity at that point was that he was “ it’s not that I failed, I was robbed.” if he truly accepted he wasn’t “robbed” he would have had to face the unbearable “truth”—that he wasn’t inherently special, that no “cosmic injustice” had singled him out (even the universe didn’t notice him or need him for anything, utterly worthless), and that his isolation was, at least in part, a result of his own actions (or inaction). This would have shattered his entire identity, leaving him with nothing to cling to. Without the belief that he was a victim of an unfair world, he would have had to confront the terrifying reality that he was just ordinary—that he lacked social skills, had put in no effort, and wasn’t owed admiration or love just for existing. For someone whose entire self-worth and sense of existence depended on feeling superior, this realization would have been annihilating. Instead of processing his role in isolation, his mind rejected it completely, reinforcing the victim narrative to preserve what little sense of self he had left.

Edit: no- this horrific reality of worthlessness is not the “truth.” The truth is Elliot had a true self that should’ve developed and he could’ve uncovered parts of it that would be very unique and very worthy of attention and love. The reason he felt like “nothingness” is because the false self had never allowed him to develop.

Even these days, I keep falling back into this old mentality, and I have to wrench myself out of it. The truth is, I don’t put myself out there socially because I’m scared or I don’t know how and I don’t want to risk humiliation, so I sit and then I get bitter and even darkly vengeful and hateful when I see other people bonding or living life. “I hate everyone for not recognizing me.” (if I get deep enough into the mentality, I suddenly forget the fact that the reason is I don’t put myself out there socially and i just feel blindly enraged that people don’t notice me when they should. the false self dissociates me from the social issue so then I’m just bitter in a void.) I think to myself “they would never do that with me.” I switch from thinking “ It’s because I’m nothing I have nothing to look at” to “ They beneath me, they’re so boring and normal, and they could never comprehend a being like me. I’m beyond human” and I flip-flop between these feelings. But all of this certainty is a way to ward off the fear of a more personal humiliation if I actually try. So I’ve been working in therapy on healthy ways to not immediately assign situations to my sense of identity or worth- even if they go bad. However, if I DONT immediately assign new situations to my identity, I feel like I have NO identity and experience existential collapse because the false self only “allows” me to have an identity if it’s externally based… It’s such a mess…

(The false self only allows you to form an external identity because it’s controllable, editable and shiftable, and it can be manipulated to get specific reactions from people. All of this feels very safe. Allow allowing the self to be built out of authentic pieces makes it too rigid to get “positive” responses in “every” situation, (you’ll be less pleasing to more people) and it also feels incredibly dangerous because of the severe trauma that happened the last time I was authentic as an infant. The false self basically has a PTSD response when it thinks anything authentic is being exposed. It’s also scary feeling so undeveloped and empty underneath you feel like you could never capture attention or stand out or meet any needs)

Giving in = weak. This mentality could explain why Elliot felt if he ever backed down from his tirade once it started, he would have to face the ultimate fear that he was utterly worthless. It’s true that fighting to prevail is strength, but he couldn’t see he was fighting the wrong battle, a battle that would lead him nowhere when he could’ve been spending his energy fighting the real fight. Of course someone would have to had explained to him what was happening to him.

Oh my god.. this is literally the exact thought process I had. So I wanted to be a famous popstar, but as time went on and I got older I realized that window was closing, so I dropped the fantasy because I wouldn’t have been able to have the adoration of being YOUNG, desired and talented and it felt pointless. I didn’t want to be old and just making music in some café. It’s also interesting how vulnerable narcissists seems so much closer to reality and so much more aware of the work that it would take and all the things that could go wrong, but that awareness also is what causes them to freeze and keep dropping their ideas and doubting themselves. Grandiose narcissists are so blindly certain their future will work out that some of them actually end up pulling something off from sheer confidence even if their success is usually short-lived because their disorder ruins it.

HAHAHAHA oh my godddd. OK, every narcissist I’ve known has struggled with this even me, but I ended up biting the bullet and getting a retail job and it actually taught me so much. I’m laughing because it’s just so damn tragic and stupid and egotistical but yet there’s so much pain, what a mess. On the surface, it looks like just entitlement. But it’s a lot deeper. What ended up helping me not feel humiliated for working retail was to know that even the great people had to climb the ladder to reach the heights, it doesn’t reflect their worth at all. Even the greatest people with the most intelligence have to put in work and even menial labor to get opportunity.

A huge distortion in NPD is taking any immediate new “part” of your life and having it immediately define your entire identity. Since your identity is external, anything that comes into your world and is connected to “you” that seems subpar can make you instantly feel worthless. It can also cause huge fears that this is all you will ever be worth, all you could ever amount to, even with your best effort is an ordinary life where you barely made an impact. you will always be outshone by brighter stars, always overlooked for better people. Both humiliating and terrifying. If you can’t make a mark on the world, you can’t control your environment and if you can’t control your environment, you can’t feel safe. If you can’t stand out, how can you be seen in the crowd? How can you ever make up for all of your starvation? How will you ever get your needs met?

What we’re about to witness is one of the saddest things. Elliot will fluctuate between hope and despair for quite a while, trying fantastical solutions that don’t work and aren’t connected to reality until he finally loses all hope without any real help.

He will then snap into malignancy where he just wants to cause suffering and destroy everything because he was never allowed to be a part of the world, and in his mind, no one notices him or cares he’s dying as they flaunt and tease him daily with everything he desperately needs. It feels like you’re starving to death watching everyone else eating food and no one cares. That’s enough to cause utter hatred and coldness. Every human has basic needs- belonging, emotional connection, love, being seen and accepted, safety, self esteem. NPD starves you of all.

It’s not to say that Elliot’s tragedy was the responsibility of the people around - nor should they have been expected to go out of their way to cater to him. It was the responsibility of the mental health system to help him integrate into society, help him understand what was happening to him and give him any fighting chance.

I relate so deeply… I’ve just dissociated from it and am waiting for my next life honestly. But this is my life, even right now writing this. I just watch everyone else live and get to enjoy everything that I can’t have. It’s been like this almost my whole life. (Luckily, I am in therapy so hopefully I will figure something out).

This is a great description of vNPD, especially the contemplation and rumination. People always wonder why Elliot didn’t just stop obsessing over couples and sex and focus on hobbies or a life goal. But once again, the false self is the only self Elliot has access to. And all it does is obsess over affirmation or how to get affirmation to feed itself and make you feel real or exist. Even if you just sit there without any supply, it’s not like the true self suddenly starts showing itself, and you find all these authentic things to want and think about. You’re usually devoid of authentic interests and opinions about the world -because the false self blocks anything from forming. All that’s left to do is obsess about how to “get your needs met.” It’s the only thing that has any sort of life to it.

Here we can also see Elliot starting to slip into a malignant narcissistic thought pattern. mNPD usually starts when the narcissist feels NPD (gaining safety and existence through admiration) isn’t working to keep them safe and the person shifts towards power.

Enter grandiosity as a defense with worsening trauma. He’s coping with existential terror and meaninglessness and a lot of shit I so deeply relate to and it looks like this defense finally swoops in and saves him a bit. He must have felt relief finally thinking it’s the WORLD that’s got it wrong and he’s actually the worthy one, above them all and here to change the world to something fair and “how it should be.” I can sense the relief.

People always focus on how entitled Elliot seemed for demanding sex or wanting to legit destroy those that have it- and how “other people get rejected or stay a virgin for a while but THEY don’t do this.”

But they don’t understand the deeper pathology, which needs help. What Elliot REALLY wanted was a return to a world that was simple enough where he could control it, and receive positive outcomes. He wanted to be able to feel a part of the world, and the only time he’d been able to was before social dynamics got more complicated, including sex. He felt exiled from the world and unable to participate or know how to be seen and so he was basically starving while watching everyone else eat. His only identity inside was a false self, which is a hollow construct that only feels real with social affirmation. That meant that if he couldn’t get a functioning mask and be fed socially, he would essentially feel like he didn’t exist at all. The false self covers a core emptiness that has no uniqueness or spontaneity. That means unless the person takes others’ traits, they feel they have nothing special to offer. If those traits aren’t seen and validated, the person literally feels like an empty ghost (think No Face from Spirited Away) who is forced to watch everyone else get seen and admired because they have nothing to show, nothing that stands out. You feel like you literally have nothing for anyone to look at or love. You feel utterly replaceable and humiliated. It’s a profound, all-encompassing worthlessness that can only be experienced with such an empty core.

Elliot feels like he’s been denied identity and worth and even existence by being denied sex because sex is final proof you are “superior” and finally real, included, cool, admired, adored. If you can’t stand out and shine the brightest, you’ll have to constantly fear being replaced, discarded for something better and humiliated, and left to fall apart existentially.

It’s like he see sex as the key to being able to be a part life safely, or even to exist in any meaningful way.

Elliot’s NPD ensures he CANT feel like he exists or has any worth at all unless he gets these external things and gets proof of his own existence. If he tried to draw upon any internal interests, hobbies, or thoughts, he’d likely find emptiness as the false self is entirely obsessed with getting supply and it’s the ONLY self accessible. So getting this external stuff feels like life or death. I hope this makes sense why sex is so dire for Elliot. It’s a mental health issue. When supply is cut off, the narcissist feels like they are disintegrating. This explains why Elliot’s reaction to rejection was so extreme, it wasn’t just disappointment; it felt like an existential annihilation and removal of any basic needs.

This is totally valid and Elliot is right- his life was unfair. But he can’t connect that it’s due to severe mental health issues and social skill issues. If there was a system in place to spit and help people like us, we could have been reintegrated into society.

This could easily be mistaken as PTSD, anxiety, or even avoidant personality disorder, but what you need to understand is Elliot was hiding away in the fantasy world because it was the only place he could feel superior, powerful, and admired, and actually control how he was perceived, living vicariously through characters. His narcissistic pathology could’ve been easily missed if he had been taken to a psychiatrist for analysis. taking his childhood history into account, all he has ever done for hobbies is try to adopt the popular kids hobbies. He has clear issues having any identity of his own. but since he says stuff like he didn’t care about appearances anymore, people would not have clocked his NPD because it’s vulnerable NPD.

I see Elliot’s parents are trying here, and it made me sad. so it’s more likely that they just weren’t aware that they were being emotionally neglectful. They did try and help him. This reminds me of my own parents. They had such good intentions, but they just didn’t know how to be supportive emotionally.

I think Elliot’s mentally young age made these traumas hit a lot harder and caused a lot more damage. As I said, this all eventually culminated in utter lack of empathy for humanity. And of course, he didn’t have a healthy ego or any identity to be able to take blows healthily. I always say that the reason why rejection hits cluster bs so hard and shatters them so deeply -while other kids have trauma but don’t escalate to such extremes - is BECAUSE there’s no foundation to the self, you only exist through the eyes of others and you already have early traumas that made the world seem way too scary and made you feel fundamentally worthless -to your core. It’s also understated how hard it is to know how to navigate the world without a self-concept, you don’t know how to feel seen or safe unless you’re able to constantly analyze and control your environment and stay “on top”. Elliot had trouble analyzing and navigating his environment, which would’ve caused his fear to be tenfold and his need for control to eventually increase to malignant levels.

sure, but this paints it to be sincere: it's not. its just so people can continue to offer their data, engagement/ money etc... a millionaire doesnt care: the anti-thesis: youre trying to do revolutonary work on a site that is owned by the ruling class, in which does not wish to have such revolution: clash of class interests

Thinking of Konmari and socks. Values embedded in an anthropomorphized relationships can extend beyond harm principle to the sock – it can encompass environmental concerns, attitudes toward repair.

This chapter gave a lot of insight into the complexities of privacy in our digital world. I found the section discussing how much of our "private" information is actually not as secure as we might think both eye-opening and alarming. The example about Facebook storing Instagram passwords in plain text stood out to me because it shows how companies often fail at safeguarding our most sensitive data, even when we're trusting them with it. It's easy to forget the risks we take when we freely share personal information on social media or other platforms, assuming that companies will protect us. This chapter serves as a wake-up call for individuals to take their privacy seriously and consider how much control they’re willing to give up in the digital age.

Summary of DevTools FM Podcast with Juan Capa on Membrane.io

Introduction and Background

• Juan Capa is the creator of Membrane.io, a still-in-development platform for simplifying API automation and internal tooling.

"Juan is the creator of membrane.io, a still-on-development platform for simplifying API Automation and internal tooling."_

• He has a background in game development, having spent over a decade working on console, mobile, and web games.

"I have a background in game development. I spent about 10 years a little bit more than 10 years working in game development."_

• Worked at Vercel on the CDN team after being hired through Twitter, then briefly returned to Zynga before joining Mighty under a program that allowed him to work part-time on Membrane.

"I saw a tweet by Guillermo Rauch ... He hired me to work for Vercel ... I spent two years there as the lead in the CDN team."_

"Then I guess my last last thing I did was join Mighty ... working on my startup but also working three days for them."_

• Now focusing on Membrane full-time and looking to onboard users soon.

"So yeah now I'm a member in 100 and yeah hoping that I can show to the world and onboard some users in the coming week or two."_

Membrane: Concept and Vision

• Membrane was inspired by game engines, where every entity is programmable and data is universally accessible.

"In game development, you’re dealing with this Engine with this universe, and this universe is completely programmable."_

• Aimed at simplifying API automation and small-scale applications, particularly for personal automation.

"It’s a place to write programs to build personal automation ... optimized for personal automation programs."_

• Membrane provides an abstraction over APIs, allowing users to interact with data and automate workflows through a graph-based system.

"The key to Membrane is this whole concept of a graph that is the main thing that programs use to manipulate the world."_

• Designed to be highly accessible by integrating with Visual Studio Code and leveraging JavaScript/TypeScript.

"The entire thing is built inside of Visual Studio Code ... The most used IDE is Visual Studio Code and the most used language is JavaScript."_

Durability & Orthogonal Persistence

• Membrane implements "orthogonal persistence," ensuring program state is always durable.

"I decided to start building what is sometimes called orthogonal persistence, which is this concept of a durable program."_

• Every Membrane program is an SQLite database, meaning all messages, state, and execution history are stored persistently.

"Every member program is actually just one SQLite database."_

• Programs execute with an event-sourcing model, where all inputs and outputs are first logged in SQLite before execution.

"Every message that it receives, it first goes in the database and then it's processed."_

• Uses Linux’s soft dirty pages for memory tracking, making it highly efficient in persisting only changed memory states.

"I use quickjs ... and there’s a constant in the Linux kernel called Soft Dirty Pages ... only serialize the pages that actually change."_

• Future improvements include optimizing serialization using WebAssembly’s linear memory model.

"I’m saving more data than I should, so there’s even more optimizations I can do."_

Observability & Debugging

• Membrane prioritizes perfect observability, logging every event to enable full program introspection and debugging.

"If it’s not in the logs, it didn’t happen."_

• Allows time-travel debugging, replaying past states and executions.

"You can go back to when that message was received and then run the code that was available back then."_

• Aims to support snapshot-based time travel for enhanced debugging.

"The first version I’m gonna have of that type of time travel is going to be with a snapshot that is taken every hour."_

Membrane’s Graph Model

• Membrane’s "graph" serves as a type-safe, unified interface for APIs.

"Everything is a node, which you can think of as an object or a scalar (string, number, JSON type)."_

• Drivers enable API connectivity, converting external APIs into Membrane’s schema and providing a consistent interface.

"The GitHub driver has a schema ... basically it mirrors the GitHub API as a Membrane schema."_

• Pagination is abstracted away, making API traversal seamless.

"With Membrane, you have this object that’s a one-page, and a page has a reference to the next page."_

• Users can mount different programs' graphs into their own, dynamically expanding their automation environment.

"Your graph is basically the combination of all the graphs of all your programs."_

Chrome Extension & API Interfacing

• Membrane includes a Chrome extension that recognizes API entities on webpages.

"What it does is it asks Membrane, ‘Hey, do any of the programs under Juan’s account recognize anything on this page?’"_

• Future improvements will allow automatic driver installation when encountering unrecognized APIs.

"Eventually, I can just offer you the option to install that driver with a click from the Chrome extension."_

• Currently requires users to provide their own API keys, but OAuth-based authentication is planned.

"Right now, you have to bring your own keys."_

Cron & Automation Features

• Membrane features built-in cron-like timers, which are stored in SQLite and visualized in the UI.

"The SQLite database has a table called timers, and that table holds all scheduled actions."_

• Users can visually track when timers will execute and manually trigger actions for testing.

"From Visual Studio Code, you can just hover on each timer and see how long until it fires."_

• Logs every timer execution, ensuring full transparency in automation workflows.

"If it’s not in the logs, it didn’t happen."_

Potential for Expansion & Future Vision

• Membrane’s approach is inspired by game development tooling, where objects and behaviors are always inspectable.

"In game engines, you’re dealing with objects where you can see all their properties and control them."_

• Aims to provide a seamless developer experience, where APIs become interactable entities without custom adapters.

"If you wanted to automate something with Twitter, you shouldn’t have to pre-install a driver."_

• Exploring self-hosting and open-source models to improve privacy and decentralization.

"Self-hosting membrane is going to be a thing ... I think I want to make it open-source."_

• Could enable mobile implementations, particularly for interacting with on-device automation.

"You could just access your Membrane graph from your phone."_

• Possibility of auto-generating API drivers from HAR files or OpenAPI specs.

"There are ways to generate API specs from network traffic ... from that API spec, you can generate the driver."_

Conclusion

• Membrane is a powerful tool aimed at making personal automation and API interaction seamless, leveraging game engine principles for maximum programmability.
• It provides persistent execution, deep observability, and a graph-based API abstraction layer that simplifies working with external services.
• With a focus on usability, it integrates tightly with VS Code and JavaScript while also offering innovative features like event sourcing, time travel debugging, and drag-and-drop API connections.
• The future of Membrane includes open-source possibilities, mobile integrations, and potentially eliminating the need for manually defining API adapters.
• It represents a new paradigm in developer tooling, where programs are durable, transparent, and universally programmable.

Juliet De Leon: We can only understand or analyze a study within the context of collective knowing, means that our understanding of research or findings is shaped by the knowledge that already exists in society or within a particular field. In other words, new information is interpreted based on what we already know as a group.

Let’s take memory and eyewitness testimony as an example. For a long time, like most people, I believed in the absolute validity of eyewitness testimony. The idea was simple: if you saw someone commit a crime, then they must be guilty. This belief was rooted in the assumption that our memories are accurate and reliable reflections of the past.

However, over time, our understanding of memory has evolved. We’ve learned that memory isn’t as reliable as we once thought. In fact, it’s incredibly malleable and prone to error. Studies on memory have shown that our recollections can be distorted by a number of factors, such as stress, the passage of time, or even leading questions. Eyewitnesses can remember things inaccurately, sometimes unknowingly, and these false memories can feel just as vivid and real as actual events.

Juliet De Leon: Authority is a big deal because, as kids, we’re taught to listen to our elders, and that tends to carry over when it comes to leaders. We often accept what they say without question. But the truth is, people in power are just as human as anyone else, and humans are flawed. There are tons of examples where leaders were just flat-out wrong—like how Tesla’s ideas were dismissed, the practice of slavery, and the whole Flat Earth Theory.

Now, we live in a time where we’re hit with information all day long from social media, news outlets, podcasts, blogs, and more. AI-generated content is so realistic these days that it’s hard to tell what’s real and what’s fake. So, more than ever, we need to be on our toes, questioning, assessing, and verifying everything we come across.

This reading talks about different ways to design things, like making them simple, new, powerful, hidden, fair, or useful for everyone. I agree that powerful designs can be both helpful and harmful, like how saved searches on Twitter can be used for good or bad. It made me think about how designers need to be careful about how their work affects people, not just how easy or exciting it is to use.

I wanted to touch on this specific section because I agreed with the approach, but more importantly, I believe this is the essence of what we have been learning to do in this class. Understanding what exactly is needed in a specific circumstance, and context I think is what drives a good solution. This is why sometimes simple solutions such as the Macca Scoop literally dominate in there specified problems (For those who don't know the Macca Scoop is used to place a specific amount of chips, or fries in the container which improves efficiency and produces cost exponentially which is hilarious since it's just a simple plastic thing). This is why I believe designing and research go hand and hand because you need to understand what the problem needs for the context. Especially once budgets and production get involved.

Wicked Reading: I wanted to focus on Design and Technology on page 19. Specifically the section where he expresses systematic thinking. I think this insight is beyond useful since we as a society have begun to incorrectly define things we interact with daily such as technology. The issue with these definitions is that they affect how we approach and interact with technology. Although it seems small I can see how these definitions tend to cause impacts on design, and overall expansion of ideas since they focus on the wrong things such as product, instead of valuing how technology-based design all holds very important principles that should be maintained, and I agree with this since it reminds me of one of my favorite insights on how the methodology is usually more important than the content. It is about how you do it.

I really like the idea of balancing criticism with positive feedback. It makes a lot of sense because people are more open to suggestions when they don’t feel like they’re being attacked. I’ve definitely been in situations where harsh feedback made me shut down instead of actually listening. This approach makes it easier to hear the tough parts while still feeling encouraged. It’s a great reminder that critique should be about helping someone improve, not just pointing out what’s wrong.

in the words of my therapist "shoulda, woulda, coulda." we cannot change the past. If we could have we would of. All these "well if we did this" or "if only the person in power did this thing." THEY DIDNT! and we cant do shit about it. We must build. Its never to late to start anything you just have to start you have to work and build now. so what you didnt do something in the past? you have NOW.

This makes me feel like she is tatsing more than just the fruit, like she is tasting the history and past that comes along with the fruit.

This is a great mindset to have in the workplace. As someone who has experienced having a ship-sinker on her team, I speak from experience when I say that mentality changes everything. Healthcare is stressful. Radiology departments have never had more work to do. Orders pile up, and someone who wants to just be on their own boat with their own problems will sink the department. You HAVE to lift others up and be a helper. In our jobs, people's lives can and do depend on it.

I totally agree that giving constructive feedback is not an easy thing. This is because giving a useful critique might need professional knowledge somehow and also some feedback like it's good and not great cannot let people realize how to improve their current solutions and indeed know what others' thoughts about it and test it if it's user-friendly enough. I changed my perspective because at first, I hadn't taken this course I might just give vague feedback to others but now I comprehend the importance of giving feedback.

The point about vague feedback like “It’s good” or “Could be improved” resonates with me because such comments lack depth and don’t provide any room for growth. I find this perspective useful because it highlights the importance of offering constructive, specific feedback that fosters creativity and problem-solving. It also reminds me to be more intentional in my own feedback, focusing on generating ideas rather than just critiquing.

Within healthcare, it isn't just about one person but the entire team. Imaging departments will work together to achieve smooth workflow for one another. For example if XR has a patient with CT orders, someone will call and ask if they want them after. This allows great team work for a more effective imaging time. At times it may feel like you might be the only one doing things, but always remember if it wasn't for the entire team then things wouldn't be done as fast. This is why our departments are so big, and they need us all to work together. As Patrick Mahomes would say, its not a few players that got us that win but the entire team.

Yes, I see this happening in my life a lot. Some people are already good at what I’m trying to do. Some may even be younger than me, and I often wonder if they’re smarter. But no, it’s simply that they have more practice. I might take five hours to complete a project while they only need one, but I’ve realized I’m just making up for the time they’ve already put in. They’ve already invested those extra four hours at some point, and now I’m playing catch-up. In the end, though, we’re all going to reach the same destination

I think this point was pretty interesting. This reminds me of how in class, we talked about how when brainstorming ideas, we need to unfilter out our ideas and let them flow out, because if we filter out our ideas, we may lose out on interesting ideas that can contribute to the bigger picture of how we want the project to look like. Plus, taking a little bit and looking back at the idea later can also add interesting insights that make it useful as opposed to just saying it's a dumb idea and forgetting about it.

This part of the reading is interesting because it shows how personal experience can drive creativity. I agree that students, having sat through countless lectures, are in a great position to suggest meaningful improvements. It’s a good reminder that our frustrations and experiences can lead to valuable ideas, even in areas we might not think we’re experts in.

Exhaling deeply while grading my 47th essay of the evening, coffee gone cold beside me

Oh, Blackboard... bitter laugh

Let me tell you about Blackboard through the fragments of my daily struggle, through the prism of 4/4 teaching load spread across three campuses just to make rent:

Each semester begins the same— Login attempts like scattered prayers Dashboard a maze of broken promises Features that mock with their corporate sheen While I upload syllabi at midnight Again and again and again

They sold us dreams of streamlined workflows But my grades still vanish into digital void Support tickets float unanswered Like autumn leaves in administrative wind While students message: "Professor, I can't find..." And I drown in workarounds

The cost? Oh, the cost... Not from my adjunct's pittance But I watch department meetings Where deans speak of budget constraints Yet somehow there's always money For another Blackboard module Another upgrade Another promise

Canvas beckons from across the quad Where my tenure-track colleagues reside In their technical paradise While we contingent faculty Navigate this labyrinth of legacy code Because migration costs too much For our satellite campus

Do you know what it's like To build a course shell from scratch Four times a year Because "course copy" fails While grading deadline looms? To explain to students Why their mobile app won't load?

Rubbing temples, reaching for cold coffee

But tomorrow I'll log in again Because what choice do we have? When you're paid by the course You dance to the tune they play Even when the music stutters Even when the platform breaks

Don't talk to me about "bad actors" Talk to me about survival About making do About teaching despite, not because of These digital walls we're given

...I should get back to grading. These essays won't grade themselves, and the Blackboard SpeedGrader is down. Again.

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

Summary: It has been known for many years that some peroxisomal proteins are imported by the major peroxisomal protein import receptor Pex5, which recognises the C terminal targeting signal PTS1, despite either lacking a PTS1 or if the PTS1 is blocked. Some proteins are also able to 'piggyback' into peroxisomes by binding to a partner which possesses a PTS. Eci1, the subject of this study is such a protein. This manuscript identified a PTS1-independent, non-canonical interaction interface between S. cerevisiae PEX5 and imported protein Eci1. Confocal imaging was used to observe the PTS1-independent import of Eci1 into peroxisomes and to establish dependence of Pex5 even in the absence of its piggyback partner Dci1. The authors purified the Pex5-Eci1 complex and used Cryo-EM to provide a structure of the purified PEX5-Eci1 complex. In general, this manuscript is well written and easy to read.

Major points

Most of the experiments presented are well-designed and accompanied with appropriate controls. However, please mention how many times the experiments have been repeated and how many biological samples were used in the analysis.The authors should also consider the following suggestions substantiate their conclusions:

Figure 1A: Include full-length Eci1 with an N-terminal fluorophore, Eci1 PTS1-deletion with N-terminal fluorophore, and the PTS1 deletion with a C-terminal fluorophore, to control for any disturbance of targeting by the C terminal NG tag.

Figure 1C: Confirm the Eci1 and Dci1 levels (if an antibody is available for the latter) by western blot. It is difficult to compare expression levels when comparing just a small number of cells in the microscope. Western blot would give a more robust evaluation of protein levels and help corroborate the claim that Eci1 expression is decreased in the absence of Dci1 if the authors wish to stand by this conclusion.

Figure 2: confirm the deletion and overexpression of PEX9, PEX5, and PEX7 by western blot of the relevant strains. The production of these strains is not described in the manuscript. If they have been previously described this should be referenced if not it should be included.

Figure 2: Validate these strains by checking import of a canonical PTS1 and canonical PTS2 and pex9 dependent protein to ensure they function as they should, unless these strains have been published elsewhere in which case their characterisation can be referenced.

Figure 3: The gel should include a standard of a known amount of the lysate used in the pull down to enable a semi-quantitative estimation of the amount of Eci1 protein captured by PEX5 with and without its PTS1. Also include Eci1 with a C-terminal fluorophore to be comparable with the in vivo data in Figs 1 and 2. A control with no pex5 for background would be useful. A full Coomassie-blue stained gel (not western blot) is required to demonstrate the direct interaction as with the western blot it cannot be excluded that other proteins bridge the interaction since this is a pull down from lysate not purified proteins. OPTIONAL:Interestingly the surface on Eci1 which binds pex5 is where CoA binds in the active enzyme. Would CoA compete for binding to Pex5? (could add it into the pull down expt?)

Figure S2: The complex between pex5 and eci1 is solved by cryo EM. Eci1 is hexameric usually 1 but sometimes 2 or 3 pex5s are bound to the complex. The size-exclusion chromatography figure with calculated molecular weight is required to support the stoichiometry. A native gel to show the complex, as well as a denaturing gel (using the complex) to show the individual proteins will be beneficial.

Figure S9: Would Eci1 compete with Dci1 to bind to Pex5 since they share highly conserved interfaces? If so, why did the deletion of Dci1 impair Eci1 location? Or is this just reduced expression in the dci1 deletion background? (See point 2) This seems counterintuitive/contradictory so please comment.

OPTIONAL: As the authors acknowledge this work is in vitro. It would have been interesting to examine the role of this interface in vivo by mutating one or more of the residues in Eci 1 identified as being important for the interaction. Granted that mutation can affect the folding of the protein, but the binding region is on the surface so it may not, and this can be readily checked e.g by enzyme activity or limited proteolysis.

OPTIONAL: Similarly, it would have been interesting to see if mutating the residues of PEX5 involved in the interface affect the import of other cargoes than eci1 or if reciprocal mutations in pex5 and Eci1 e.g switching charges could restore an import defect.

OPTIONAL If 8 & 9 isn't possible could a co-evolutionary analysis of the interface residues provide further independent evidence for their functional importance? They have looked at conservation of residues in Eci1 but this could be extended to a co-evolution analysis.

Minor points

Figure 1C and throughout the manuscript state clearly whether the same confocal settings are used when comparing fluorescence intensity of different images/samples.

Figure S2B: Please use different colours for PEX5 and Eci1 for clarity.

Figure 4A: please indicate the PTS1 for the other 5 molecules of Eci1. Are they buried? Or not seen? Please add explanation.

Figure 4B, C, and D: please colour the circled helix in PEX5 so that it can be more easily seen.

Please indicate the EBI-mediated interaction in Figure 4C. The relationship between 4C and 4D could be explained better as they are not viewed from the same direction

Figure S3: As the authors indicated, Pex5 binds with multiple conformations and forms a variable interface with an Eci1 subunit. Does this mean different types of non-canonical interface are possible? Please discuss this.

Figure 5A and B: they should be labelled as PEX5 TPR domain

Figure S8 is very helpful in understanding the interface and could be included in Figure 5.

Significance

While cargo recognition by Pex5-PTS1 is well understood in molecular detail there are proteins which either lack a PTS1 or have a nonessential PTS1 that still require Pex5 for import into peroxisomes. This study provides a structural view of interaction between Pex5 and its cargo Eci1, a protein that does have a PTS1 but which is not essential for import. It's not the first example of a PEX5-cargo structure to show a non-canonical binding interface and the results are compared to the human pex5-AGT structure. It is an important addition to understanding how so-called PTS context dependent or non1 non2 proteins can be imported. Is this the first structure showing Pex5 bound to an oligomer cargo? Previous work is appropriately cited in the manuscript.

The study will be of interest to audiences interested in protein-protein interaction and in protein targeting to organelles. This manuscript presents additional knowledge on how an oligomeric PTS1-independent protein can be imported into peroxisomes. The potential of other proteins using the similar importing mechanism can be tested to understand how one receptor can use apparently multiple binding modes to import a wide range of different proteins.

If we ever use AI as a support tool for out essays, do we have to use references for them? I just wanted to make sure that all the help that I used would be shown.

Reviewer #2 (Public review):

Summary:

The authors perform a remarkably comprehensive, rigorous, and extensive investigation into the spatiotemporal dynamics between ribosomal accumulation, nucleoid segregation, and cell division. Using detailed experimental characterization and rigorous physical models, they offer a compelling argument that nucleoid segregation rates are determined at least in part by the accumulation of ribosomes in the center of the cell, exerting a steric force to drive nucleoid segregation prior to cell division. This evolutionarily ingenious mechanism means cells can rely on ribosomal biogenesis as the sole determinant for the growth rate and cell division rate, avoiding the need for two separate 'sensors,' which would require careful coupling.

Strengths:

In terms of strengths; the paper is very well written, the data are of extremely high quality, and the work is of fundamental importance to the field of cell growth and division. This is an important and innovative discovery enabled through a combination of rigorous experimental work and innovative conceptual, statistical, and physical modeling.

Weaknesses:

In terms of weaknesses, I have three specific thoughts.

Firstly, my biggest question (and this may or may not be a bona fide weakness) is how unambiguously the authors can be sure their ribosomal labeling is reporting on polysomes, specifically. My reading of the work is that the loss of spatial density upon rifampicin treatment is used to infer that spatial density corresponds to polysomes, yet this feels like a relatively indirect way to get at this question, given rifampicin targets RNA polymerase and not translation. It would be good if a more direct way to confirm polysome dependence were possible.

Second, the authors invoke a phase separation model to explain the data, yet it is unclear whether there is any particular evidence supporting such a model, whether they can exclude simpler models of entanglement/local diffusion (and/or perhaps this is what is meant by phase separation?) and it's not clear if claiming phase separation offers any additional insight/predictive power/utility. I am OK with this being proposed as a hypothesis/idea/working model, and I agree the model is consistent with the data, BUT I also feel other models are consistent with the data. I also very much do not think that this specific aspect of the paper has any bearing on the paper's impact and importance.

Finally, the writing and the figures are of extremely high quality, but the sheer volume of data here is potentially overwhelming. I wonder if there is any way for the authors to consider stripping down the text/figures to streamline things a bit? I also think it would be useful to include visually consistent schematics of the question/hypothesis/idea each of the figures is addressing to help keep readers on the same page as to what is going on in each figure. Again, there was no figure or section I felt was particularly unclear, but the sheer volume of text/data made reading this quite the mental endurance sport! I am completely guilty of this myself, so I don't think I have any super strong suggestions for how to fix this, but just something to consider.

A key moment in The antipodes is when Sandy promises the group they won’t work late or on weekends. He tells them they don’t need to be perfect, just creative. This makes the job seem relaxed and easy at first. But as the play goes on, this promise fades. The gtoup loses track of time, pressure builds, and work takes over their lives. This moment is important because it shows how things can quickly sprial out of control.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

The manuscript is dedicated heavily to cell type mapping and identification of sub-type markers in the human testis but does not present enough results from cross-investigation between NOA cases versus control. Their findings are mostly based on transcriptome and the authors do not make enough use of the scATAC-seq data in their analyses as they put forward in the title. Overall, the authors should do more to include the differential profile of NOA cases at the molecular level - specific gene expression, chromatin accessibility, TF binding, pathway, and signaling that are perturbed in NOA patients that may be associated with azoospermia.

Strengths:

(1) The establishment of single-cell data (both RNA and ATAC) from the human testicular tissues is noteworthy.

(2) The manuscript includes extensive mapping of sub-cell populations with some claimed as novel, and reports marker gene expression.

(3) The authors present inter-cellular cross-talks in human testicular tissues that may be important in adequate sperm cell differentiation.

Weaknesses:

(1) A low sample size (2 OA and 3 NOA cases). There are no control samples from healthy individuals.

Thank you for your comments. We recognize that the small sample size in this study somewhat limits its generalizability. However, in transcriptomic research, limited sample sizes are a common issue due to the complexities involved in acquiring samples, particularly in studies about the reproductive system. Healthy testicular tissue samples are difficult to obtain, and studies (doi: 10.18632/aging.203675) have used obstructive azoospermia as a control group in which spermatogenesis and development are normal.

(2) Their argument about interactions between germ and Sertoli cells is not based on statistical testing.

Thank you for your comments. Due to limited funding, we have not yet fully and deeply conducted validation experiments, but we plan to carry out related experiments in the later stage. We hope that the publication of this study will help to obtain more financial support to further investigate the interactions between germ cells and Sertoli cells.

(3) Rationale/logic of the study. This study, in its present form, seems to be more about the role of sub-Sertoli population interactions in sperm cell development and does not provide enough insights about NOA.

Thank you for your comments. In Figure 6, we conducted an in-depth analysis and comparison of the differences between the Sertoli cell subtypes and the germ cell subtypes involved in spermatogenesis in the OA and NOA groups. The results revealed that in the NOA group, especially in the NOA3 group, which has a lower sperm count compared to NOA2 and NOA1, there is a significant loss of Sertoli cell subtypes including SC3, SC4, SC5, SC6, and SC8. The NOA1 group, with a sperm count close to that of the OA group, also had a Sertoli cell profile similar to the OA group. The NOA2 group, with a sperm count between that of NOA1 and NOA3, also exhibited an intermediate profile of Sertoli cell subtypes. Therefore, we suggest that change in Sertoli cell subtypes is a key factor affecting sperm count, rather than just the total number of Sertoli cells. We believe that through these analyses, we can provide in-depth insights into NOA, and we hope that the publication of this study will help obtain more funding support to further validate and expand on these findings.

(4) The authors do not make full use of the scATAC-seq data.

Thank you for your comments.We have added analysis of the scATAC-seq data and shown in the revised manuscript.

Reviewer #2 (Public Review):

Summary:

Shimin Wang et al. investigated the role of Sertoli cells in mediating spermatogenesis disorders in non-obstructive azoospermia (NOA) through stage-specific communications. The authors utilized scRNA-seq and scATAC-seq to analyze the molecular and epigenetic profiles of germ cells and Sertoli cells at different stages of spermatogenesis.

Strengths:

By understanding the gene expression patterns and chromatin accessibility changes in Sertoli cells, the authors sought to uncover key regulatory mechanisms underlying male infertility and identify potential targets for therapeutic interventions. They emphasized that the absence of the SC3 subtype would be a major factor contributing to NOA.

Weaknesses:

Although the authors used cutting-edge techniques to support their arguments, it is difficult to find conceptual and scientific advances compared to Zeng S et al.'s paper (Zeng S, Chen L, Liu X, Tang H, Wu H, and Liu C (2023) Single-cell multi-omics analysis reveals dysfunctional Wnt signaling of spermatogonia in non-obstructive azoospermia. Front. Endocrinol. 14:1138386.). Overall, the authors need to improve their manuscript to demonstrate the novelty of their findings in a more logical way.

Thank you for your detailed review of our work. We greatly appreciate your feedback and have made revisions to our manuscript accordingly.

Regarding the novelty of our research, we believe our study offers conceptual and scientific advances in several ways:

We have systematically revealed the stage-specific roles of Sertoli cell subtypes in different stages of spermatogenesis, particularly emphasizing the crucial role of the SC3 subtype in non-obstructive azoospermia (NOA). Additionally, we identified that other Sertoli cell subtypes (SC1, SC2, SC3...SC8, etc.) also collaborate in a stage-specific manner with different subpopulations of spermatogenic cells (SSC0, SSC1/SSC2/Diffed, Pa...SPT3). These findings provide new insights into the understanding of spermatogenesis disorders.

Compared to the study by Zeng S et al., our research not only focuses on the functional alterations in Sertoli cells but also comprehensively analyzes the interaction patterns between Sertoli cells and spermatogenic cells using scRNA-seq and scATAC-seq technologies. We uncovered several novel regulatory networks that could serve as potential targets for the diagnosis and treatment of NOA.

We sincerely appreciate your constructive comments and will continue to explore this area further, aiming to make a more significant contribution to the understanding of NOA mechanisms.

Reviewer #3 (Public Review):

Summary:

This study profiled the single-cell transcriptome of human spermatogenesis and provided many potential molecular markers for developing testicular puncture-specific marker kits for NOA patients.

Strengths:

Perform single-cell RNA sequencing (scRNA-seq) and single-cell assay for transposase-accessible chromatin sequencing (scATAC-seq) on testicular tissues from two OA patients and three NOA patients.

Weaknesses:

Most results are analytical and lack specific experiments to support these analytical results and hypotheses.

Thank you for your thorough review of our work. We highly value your feedback and have made revisions to our manuscript accordingly. Indeed, we have conducted immunofluorescence (IF) experiments to validate the data obtained from single-cell sequencing and have expanded the sample size to enhance the reliability of our results. To better present these validation experiments, we have reorganized and renamed the sample information, making it easier for you to understand which samples were used in the specific experiments. Following the publication of this paper, we plan to secure additional funding to deepen our research, particularly in the area of experimental validation. We sincerely appreciate your support and insightful suggestions, which have greatly helped guide our future research directions.

Reviewer #1 (Recommendations For The Authors):

(1) The authors should include results from cross-investigation comparing NOA/OA patients versus controls.

Thank you for your comments. In this study, OA was the control group. Healthy testicular tissue samples are difficult to obtain, and studies (doi: 10.18632/aging.203675) have used OA as a control group in which spermatogenesis and development are normal.

(2) In Table S1, the authors should also include the metric for scATAC-seq, and do more to show the findings the authors obtained in RNA is replicated with chromatin accessibility.

Thank you for your comments. We have added Table S2, which includes the metric for scATAC-seq.

(3) A single sample from each OA and NOA group may not be enough to confirm colocalization. The authors should include results from all available samples and use quantitative measures.

Thank you for your comments. I apologize that the sample size in this study was less than three and we could not conduct quantitative analysis. We will increase the sample size and conduct corresponding experiments in subsequent research.

(4) The Methods section does not include enough description to follow how the analyses were carried out, and is missing information on some of the key procedures such as velocity and cell cycle analyses.

Thank you for your comments. The method about velocity and cell cycle analyses was added in the revised manuscript. The description is as follows:

“Velocity analysis

RNA velocity analysis was conducted using scVelo's (version 0.2.1) generalized dynamical model. The spliced and unspliced mRNA was quantified by Velocity (version 0.17.17).”

“Cell cycle analysis

To quantify the cell cycle phases for individual cell, we employed the CellCycleScoring function from the Seurat package. This function computes cell cycle scores using established marker genes for cell cycle phases as described in a previous study by Nestorowa et al. (2016). Cells showing a strong expression of G2/M-phase or S-phase markers were designated as G2/M-phase or S-phase cells, respectively. Cells that did not exhibit significant expression of markers from either category were classified as G1-phase cells.”

(5) For the purpose of transparency, the authors should upload codes used for analyses so that each figure can be reproduced. All raw and processed data should be made publicly available.

Thank you for your comments. We have deposited scRNA-seq and scATAC-seq data in NCBI. ScRNA-seq data have been deposited in the NCBI Gene Expression Omnibus with the accession number GSE202647, and scATAC-seq data have been deposited in the NCBI database with the accession number PRJNA1177103.

Reviewer #2 (Recommendations For The Authors):

The detailed points the authors need to improve are attached below.

The results presented in the study have several weaknesses:

In Figure 1A, it's required to show HE staining results of all patients who underwent single-cell analysis were provided.

Thank you very much for your valuable suggestions. In Figure 1, we present the HE staining results paired with the single-cell data, covering all patients involved in the single-cell analysis.

- Saying "identification of novel potential molecular markers for distinct cell types" seems unsupported by the data.

Thank you for your comments. I'm sorry for the inaccuracy of my description. We have revised this sentence. The description is as follows: These findings indicate that the scRNA-seq data from this study can serve for cellular classification.

- The methods suggest an integrated analysis of scRNA-seq and scATAC-seq, but from the figures, it seems like separate analyses were performed. It's necessary to have data showing the integrated analysis.

Thank you for your comments. We have added an integrated analysis of scRNA-seq and scATAC-seq. The results were shown in Figure S2.

Figure 2 does not seem to well cover the diversity of germ cell subtypes. The main content appears to be about the differentiation process, and it seems more focused on SSCs (stem cell types), but the intended message is not clearly conveyed.

Thank you for your comments. Figure S1 revealed the diversity of germ cell subtypes. The second part of the results described the integrated findings from Figures 2 and S1.

- In Figure 2B, pseudotime could be shown, and I wonder if the pseudotime in this analysis shows a similar pattern as in Figure 2D.

Thank you for your comments. Figure 2B revealed the pseudotime analysis of 12 germ cell subpopulation. Figure 2D revealed RNA velocity of 12 germ cell subpopulation. The two methods are both used for cell trajectory analysis. The pseudotime in Figure 2B showed a similar pattern as in Figure 2D.

- While staining occurs within one tissue, saying they are co-expressed seems inaccurate as the staining locations are clearly distinct. For example, the staining patterns of A2M and DDX4 (a classical marker) are quite different, so it's hard to claim A2M as a new potential marker just because it's expressed. Also, TSSK6 was separately described as having a similar expression pattern to DDX4, but from the IF results, it doesn't seem similar.

Thank you for your comments. We have revised the Figure.

- It was described that A2M (expressed in SSC0-1), and ASB9 (expressed in SSC2) have open promoter sites in SSC0, SSC2, and Diffing_SPG, but it doesn't seem like they are only open in the promoters of those cell types. For example, there doesn't seem to be a peak in Diffing for either gene. The promoter region of the tracks is not very clear, so overall figure modification seems necessary.

Thank you for your comments. We have revised the Figure.

- The ATAC signal scale for each genomic region should be included, and clear markings for the TSS location and direction of the genes are needed.

Thank you for your comments. We have revised the figure and shown in the revised manuscript.

Figure 3A mostly shows the SSC2 in the G2/M phase, so it seems questionable to call SSC0/1 quiescent. Also, I wonder if the expression of EOMES and GFRA1 is well distinguished in the SSC subtypes as expected.

Thank you for your comments. We will validate in subsequent experiments whether the expression of EOMES and GFRA1 is clearly distinguished in the SSC subtypes.

- In Figure 3C, it would be good to have labels indicating what the x and y axes represent. The figure seems complex, and the description does not seem to fully support it.

Thank you for your comments. We have added labels indicating what the x and y axes represent in the Figure 3C. The x and y axes represent spliced and unspliced mRNA ratios, respectively.

- While TFs are the central focus, it's disappointing that scATAC-seq was not used.

Thank you for your comments. TFs analysis using scATAC-seq will be carried out in the future.

Figure 4: It would be good to have a more detailed discussion of the differences between subtypes, such as through GO analysis. The track images need modification like marking the peaks of interest and focusing more on the promoter region, similar to the previous figures.

Thank you for your comments. GO analysis results were put in Figure S5. The description is as follows:

As shown in Figure S5, SC1 were mainly involved in cell differentiation, cell adhesion and cell communication; SC2 were involved in cell migration, and cell adhesion; SC3 were involved in spermatogenesis, and meiotic cell cycle; SC4 were involved in meiotic cell cycle, and positive regulation of stem cell proliferation; SC5 were involved in cell cycle, and cell division; SC6 were involved in obsolete oxidation−reduction process, and glutathione derivative biosynthetic process; SC7 were involved in viral transcription and translational initiation; SC8 were involved in spermatogenesis and sperm capacitation.

In Figure 5, it would be good to have criteria for the novel Sertoli cell subtype presented. CCDC62 is presented as a representative marker for the SC8 cluster, but from Figure 4C, it seems to be quite expressed in the SC3 cluster as well. Therefore, in Figure 5E's protein-level check, it's unclear if this truly represents a novel SC8 subtype.

Thank you for your comments. CCDC62 expression was higher in SC8 cluster than in SC3. Since some molecular markers were not commercially available in the market, CCDC62 was selected as SC8 marker for immunofluorescence verification. Immunofluorescence results showed that CCDC62 is a novel SC8 marker.

- It might have been more meaningful to use SOX9 as a control and show that markers in the same subtype are expressed in the same location.

Thank you for your comments. To determine PRAP1, BST2, and CCDC62 as new markers for the SC subtype, we co-stained them with SOX9 (a well-known SC marker).

- Figures 4 and 5 could potentially be combined into one figure.

Thank you for your comments. Since combining Figures 4 and 5 into a single image would cause the image to be unclear, two images are used to show it.

In Figure 6, it would be good to support the results with more NOA patient data.

Thank you for your comments. Patient clinical and laboratory characteristics has been presented in Table 1.

- Rather than claiming the importance of SC3 based on 3 single-cell patient data, it would be better to validate using public data with SC3 signature genes (e.g., showing the correlation between germ cell and SC3 ratios).

Thank you for your comments. I'm sorry I didn't find public data with SC3 signature genes. In the future, we will verify the importance of SC3 through in vivo and in vitro experiments.

- 462: It seems to be referring to Figure 6G, not 6D.

Thank you for your comments. We have revised it. The description is as follows: As shown in Figure 6G, State 1 SC3/4/5 were tended to associated with PreLep, SSC0/1/2, and Diffing and Diffed-SPG sperm cells (R > 0.72).

In Figure 7, the spermatogenesis process is basically well-known, so it would be better to emphasize what novel content is being conveyed here. Additionally, emphasizing the importance of SC3 in the overall process based on GO results leaves room for a better approach.

Thank you for your valuable suggestions. Regarding Figure 7, we recognize that the spermatogenesis process is well-known, and we will focus on highlighting the novel content, particularly the role and significance of the SC3 subtype in spermatogenesis disorders. As for the importance of SC3 in the overall process based on GO results, we have validated this in Figure 8 through co-staining experiments between Sertoli cells and spermatogenic cells in OA and NOA groups. The results demonstrate a significant correlation between the number of SC3-positive cells and SPT3 spermatogenic cells, particularly in the NOA5-P8 group, where both SC3 and SPT3 cell counts are notably lower than in the NOA4-P7 group. This further supports the critical role of SC3 in the spermatogenesis process. Your suggestions have prompted us to refine our data presentation and more clearly emphasize the novel aspects of our research. We will continue to strive to ensure that every part of our research contributes meaningfully to the academic community. Thank you again for your guidance.

In Figure 8, only the contents of the IF-stained proteins are listed, which seems slightly insufficient to constitute a subsection on its own. It might have been better to conclude by emphasizing some subtypes.

Thank you for your comments. We have combined this part of the results with other results into one section. The description is as follows:

“Co-localization of subpopulations of Sertoli cells and germ cells

To determine the interaction between Sertoli cells and spermatogenesis, we applied Cell-PhoneDB to infer cellular interactions according to ligand-receptor signalling database. As shown in Figure 6G, compared with other cell types, germ cells were mainly interacted with Sertoli cells. We futher performed Spearman correlation analysis to determine the relationship between Sertoli cells and germ cells. As shown in Figure 6H, State 1 SC3/4/5 were tended to be associated with PreLep, SSC0/1/2, and Diffing and Diffed-SPG sperm cells (R > 0.72). Interestingly, SC3 was significantly positively correlated with all sperm subpopulations (R > 0.5), suggesting an important role for SC3 in spermatogenesis and that SC3 is involved in the entire process of spermatogenesis. Subsequently, to understand whether the functions of germ cells and Sertoli cells correspond to each other, GO term enrichment analysis of germ cells and sertoli cells was carried out (Figure S3, S4). We found that the functions could be divided into 8 categories, namely, material energy metabolism, cell cycle activity, the final stage of sperm cell formation, chemical reaction, signal communication, cell adhesion and migration, stem cells and sex differentiation activity, and stress reaction. These different events were labeled with different colors in order to quickly capture the important events occurring in the cells at each stage. As shown in Figure S3, we discovered that SSC0/1/2 was involved in SRP-dependent cotranslational protein targeting to membrane, and cytoplasmic translation; Diffing SPG was involved in cell division and cell cycle; Diffied SPG was involved in cell cycle and RNA splicing; Pre-Leptotene was involved in cell cycle and meiotic cell cycle; Leptotene_Zygotene was involved in cell cycle and meiotic cell cycle; Pachytene was involved in cilium assembly and spermatogenesis; Diplotene was involved in spermatogenesis and cilium assembly; SPT1 was involved in cilium assembly and flagellated sperm motility; SPT2 was involved in spermatid development and flagellated sperm motility; SPT3 was involved in spermatid development and spermatogenesis. As shown in Figure S4, SC1 were mainly involved in cell differentiation, cell adhesion and cell communication; SC2 were involved in cell migration, and cell adhesion; SC3 were involved in spermatogenesis, and meiotic cell cycle; SC4 were involved in meiotic cell cycle, and positive regulation of stem cell proliferation; SC5 were involved in cell cycle, and cell division; SC6 were involved in obsolete oxidation−reduction process, and glutathione derivative biosynthetic process; SC7 were involved in viral transcription and translational initiation; SC8 were involved in spermatogenesis and sperm capacitation. The above analysis indicated that the functions of 8 Sertoli cell subtypes and 12 germ cell subtypes were closely related.

To further verify that Sertoli cell subtypes have "stage specificity" for each stage of sperm development, we firstly performed HE staining using testicular tissues from OA3-P6, NOA4-P7 and NOA5-P8 samples. The results showed that the OA3-P6 group showed some sperm, with reduced spermatogenesis, thickened basement membranes, and a high number of sertoli cells without spermatogenic cells. The NOA4-P7 group had no sperm initially, but a few malformed sperm were observed after sampling, leading to the removal of affected seminiferous tubules. The NOA5-P8 group showed no sperm in situ (Figure 7A). Immunofluorescence staining in Figure 7B was performed using these tissues for validation. ASB9 (SSC2) was primarily expressed in a wreath-like pattern around the basement membrane of testicular tissue, particularly in the OA group, while ASB9 was barely detectable in the NOA group. SOX2 (SC2) was scattered around SSC2 (ASB9), with nuclear staining, while TF (SC1) expression was not prominent. In NOA patients, SPATS1 (SC3) expression was significantly reduced. C9orf57 (Pa) showed nuclear expression in testicular tissues, primarily extending along the basement membrane toward the spermatogenic center, and was positioned closer to the center than DDX4, suggesting its involvement in germ cell development or differentiation. BEND4, identified as a marker fo SC5, showed a developmental trajectory from the basement membrane toward the spermatogenic center. ST3GAL4 was expressed in the nucleus, forming a circular pattern around the basement membrane, similar to A2M (SSC1), though A2M was more concentrated around the outer edge of the basement membrane, creating a more distinct wreath-like arrangement. In cases of impaired spermatogenesis, this arrangement becomes disorganized and loses its original structure. SMCP (SC6) was concentrated in the midpiece region of the bright blue sperm cell tail. In the OA group, SSC1 (A2M) was sparsely arranged in a rosette pattern around the basement membrane, but in the NOA group, it appeared more scattered. SSC2 (ASB9) expression was not prominent. BST2 (SC7) was a transmembrane protein primarily localized on the cell membrane. In the OA group, A2M (SSC1) was distinctly arranged in a wreath-like pattern around the basement membrane, with expression levels significantly higher than ASB9 (SSC2). TSSK6 (SPT3) was primarily expressed in OA3-P6, while CCDC62 (SC8) was more abundantly expressed in NOA4-P7, with ASB9 (SCC2) showing minimal expression. Taken together, germ cells of a particular stage tended to co-localize with Sertoli cells of the corresponding stages. Germ cells and sertoli cells at each differentiation stage were functionally heterogeneous and stage-specific (Figure 8). This suggests that each stage of sperm development requires the assistance of sertoli cells to complete the corresponding stage of sperm development.”

Reviewer #3 (Recommendations For The Authors):

The authors revealed 11 germ cell subtypes and 8 Sertoli cell subtypes through single-cell analysis of two OA patients and three NOA patients. And found that the Sertoli cell SC3 subtype (marked by SPATS1) plays an important role in spermatogenesis. It also suggests that Notch1/2/3 signaling and integrins are involved in germ cell-Stotoli cell interactions. This is an interesting and useful article that at least gives us a comprehensive understanding of human spermatogenesis. It provides a powerful tool for further research on NOA. However, there are still some issues and questions that need to be addressed.<br /> (1) How to collect testicular tissue, please explain in detail. Extract which part of testicular tissue. It's better to make a schematic diagram.

Thank you for your comments. The process is as follows: Testicular tissues were obtained from two OA patients (OA1-P1 and OA2-P2) and three NOA patients (NOA1-P3, NOA2-P4, NOA3-P5) using micro-dissection of testicular sperm extraction separately.

(2) Whether the tissues of these patients are extracted simultaneously or separately, separated into single cells, and stored, and then single cell analysis is performed simultaneously. Please be specific.

Thank you for your comments. The testicular tissues of these patients were extracted separately, then separated into single cells, and single cell analysis was performed simultaneously.

(3) When performing single-cell analysis, cells from two OA patients were analyzed individually or combined. The same problem occurred in the cells of three NOA patients.

Thank you for your comments. Cells from two OA patients and three NOA patients were analyzed individually.

(4) Can you specifically point out the histological differences between OA and NOA in Figure 1A? This makes it easier for readers to understand the structure change between OA and NOA. Please also label representative supporting cells.

Thank you for your comments. We have revised the description and it was shown in the revised manuscript.

(5) The authors demonstrate that "We speculate that this lack of differentiation may be due to the intense morphological changes occurring in the sperm cells during this period, resulting in relatively minor differences in gene expression." Please provide some verification of this hypothesis? For example, use immunofluorescence staining to observe morphological changes in sperm cells.

Thank you for your comments. Due to limited funds, we will verified this hypothesis in future studies.

(6) The authors demonstrate that " As shown in Figure 5E, we discovered that PRAP1, BST2, and CCDC62 were co-expressed with SOX9 in testes tissues." The staining in Figure 5D is unclear, and it is difficult to explain that SOX9 is co-expressed with PRAP1 BST2 CCDC62 based on the current staining results. The staining patterns of SOX9 (green) and SOX9 (red) are also different. (SOX9 (red) appears as dots, while the background for SOX9 (green) is too dark to tell whether its staining is also in the form of dots.) In summary, increasing the clarity of the staining makes it more convincing. Alternatively, use high magnification to display these results.

Thank you for your comments. I have redyed and updated this part of the immunofluorescence staining results. Please refer to the files named Figure 1, Figure 2, Figure 5, and Figure 8.

(7) In Figure 8, the author emphasized the co-localization of Sertoli cells and Germ cells at corresponding stages and did a lot of staining, but it was difficult to distinguish the specific locations of co-localization, which was similar to Figure 5E. If possible, please mark specific colocalizations with arrows or use high magnification to display these results, in order to facilitate readers to better understand.

Thank you for your comments. We have re-stained and updated this part of the data. Please refer to the immunofluorescence staining data in the updated Figure 8.

(8) The authors emphasize that macrophages may play an important role in spermatogenesis. Therefore, adding relevant macrophage staining to observe the differences in macrophage expression between NOA and OA should better support this idea.

Thank you for your comments. Macrophage-related experiments will be further explored in the future.

(9) Notch1/2/3 signaling and integrin were discovered to be involved in germ cell-Sertoli cell interaction. However there are currently no concrete experiments to support this hypothesis. At least simple verification experiments are needed.

Thank you for your comments. Due to limited funding, studies will be carried out in the future.

(10) Data availability statements should not be limited to the corresponding author, especially for big data analysis. This is crucial to the credibility of this data (Have the scRNA-seq and scATAC-seq in this study been deposited in GEO or other databases, and when will they be released to the public?) The data for such big data analysis needs to be saved in GEO or other databases in advance so that more research can use it.

Thank you for your comments. We have deposited scRNA-seq and scATAC-seq data in NCBI. “ScRNA-seq data have been deposited in the NCBI Gene Expression Omnibus with the accession number GSE202647, and scATAC-seq data have been deposited in the NCBI database with the accession number PRJNA1177103.”

Author response:

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

Reviewer #1 (Public review):

This study by Wu et al. provides valuable computational insights into PROTAC-related protein complexes, focusing on linker roles, protein-protein interaction stability, and lysine residue accessibility. The findings are significant for PROTAC development in cancer treatment, particularly breast and prostate cancers.

The authors' claims about the role of PROTAC linkers and protein-protein interaction stability are generally supported by their computational data. However, the conclusions regarding lysine accessibility could be strengthened with more in-depth analysis. The use of the term "protein functional dynamics" is not fully justified by the presented work, which focuses primarily on structural dynamics rather than functional aspects.

Strengths:

(1) Comprehensive computational analysis of PROTAC-related protein complexes.

(2) Focus on critical aspects: linker role, protein-protein interaction stability, and lysine accessibility.

Weaknesses:

(1) Limited examination of lysine accessibility despite its stated importance.

(2) Use of RMSD as the primary metric for conformational assessment, which may overlook important local structural changes.

Reviewer #1 (Recommendations for the authors):

(1) The authors' claims about the role of PROTAC linkers and protein-protein interaction stability are generally supported by their computational data. However, the conclusions regarding lysine accessibility could be strengthened with more in-depth analysis. Expand the analysis of lysine accessibility, potentially correlating it with other structural features such as linker length.

We thank the reviewers for the suggestions! We performed time dependent correlation analysis to correlate the dihedral angles of the PROTACs and the Lys-Gly distance (Figures 6 and S17). We included detailed explanation on page 16:

“To further examine the correlation between PROTAC rotation and the Lys-Gly interaction, we performed a time-dependent correlation analysis. This analysis showed that PROTAC rotation translates motion over time, leading to the Lys-Gly interaction, with a correlation peak around 60-85 ns, marking the time of the interaction (Figure 6 and Figure S17). In addition, the pseudo dihedral angles also showed a high correlation (0.85 in the case of dBET1) with Lys-Gly distance. This indicated that degradation complex undergoes structural rearrangement and drives the Lys-Gly interaction.”

(2) The use of the term "protein functional dynamics" is not fully justified by the presented work, which focuses primarily on structural dynamics rather than functional aspects. Consider changing "protein functional dynamics" to "protein dynamics" to more accurately reflect the scope of the study.

Thanks to the reviewer for the suggestion to use the more accurate terminology! We agreed with the reviewer that if we keep “protein functional dynamics” in the title, we should focus on how the “overall protein dynamic” links to the “function” – The function is directly related to PROTAC-induced structural dynamics which is commonly seen in “protein-structural-function” relationship, but it is not our main focus. Therefore, we changed the title to replace “functional” by “structural”.  

(3) Incorporate more local and specific characterization methods in addition to RMSD for a more comprehensive conformational assessment.

We thank the reviewer for the suggestion. We performed time dependent correlation analysis to understand how the rotation of PROTACs can translate to the Lys-Gly interaction. In addition, we performed dihedral entropies analysis for each dihedral angle in the linker of the PROTACs to better examine the flexibility of each PROTAC.

We included detailed explanation at page 18: “Our dihedral entropies analysis showed that dBET57 has ~0.3 kcal/mol lower entropies than the other three linkers, suggesting dBET57 is less flexible than other PROTACs (Figure S18).”

Reviewer #2 (Public review):

Summary:

The manuscript reports the computational study of the dynamics of PROTAC-induced degradation complexes. The research investigates how different linkers within PROTACs affect the formation and stability of ternary complexes between the target protein BRD4BD1 and Cereblon E3 ligase, and the degradation machinery. Using computational modeling, docking, and molecular dynamics simulations, the study demonstrates that although all PROTACs form ternary complexes, the linkers significantly influence the dynamics and efficacy of protein degradation. The findings highlight that the flexibility and positioning of Lys residues are crucial for successful ubiquitination. The results also discussed the correlated motions between the PROTAC linker and the complex.

Strengths:

The field of PROTAC discovery and design, characterized by its limited research, distinguishes itself from traditional binary ligand-protein interactions by forming a ternary complex involving two proteins. The current understanding of how the structure of PROTAC influences its degradation efficacy remains insufficient. This study investigated the atomic-level dynamics of the degradation complex, offering potentially valuable insights for future research into PROTAC degradability.

Reviewer #2 (Recommendations for the authors):

(1) Regarding the modeling of the ternary complex, the BRD4 structure (3MXF) is from humans, whereas the CRBN structure in 4CI3 is derived from Gallus gallus. Is there a specific reason for not using structures from the same species, especially considering that human CRBN structures are available in the Protein Data Bank (e.g., 8OIZ, 4TZ4)?

We appreciate the reviewer’s insightful comment regarding the choice of crystal structures of BRD4 and CRBN structures from two species. Our initial selection of 4CI3 for CRBN structure was based on its high resolution and publication in Nature journal. Furthermore, the Gallus gallus CRBN structure shares high degree of sequence and structural similarity with Homo sapiens CRBN, especially in the ligand binding region. At the time of our study, we were aware of 4TZ4 as Homo sapiens CRBN, however, we did not use this structure since no publication or detailed experimental was associated with it. Additionally, PDB 8OIZ, was not publicly available yet for other researchers to use at the time.

(2) Based on the crystal structure (PDB ID: 6BNB) discussed in Reference 6, the ternary complex of dBET57 exhibits a conformation distinct from other PROTACs, with CRBN adopting an "open" conformation. Using the same CRBN structure for dBET57 as for other PROTACs might result in inaccurate docking outcomes.

Thank you for the reviewer’s comment! As noted by the authors in Reference 6, the observed open conformation of CRBN in the dBET57 ternary complex may result from the high salt crystallization conditions, which could drive structural rearrangement, and crystal contacts that may induce this conformation. The authors also mentioned that this open conformation could, in part, reflect CRBN’s intrinsic plasticity. However, they acknowledged that further studies are needed to determine whether this conformational flexibility is a characteristic feature of CRBN that enables it to accommodate a variety of substrates. Despite these observations, we believe that the compatibility of the observed BRD4^BD1 binding conformation with both open and closed CRBN states suggests that these conformational changes are all possible. Therefore, we believe using the same initial CRBN structure for dBET57 as for other PROTACs can still reasonably reveal the dynamic nature of the ternary complex and would not significantly affect the accuracy of our docking outcomes either.

(3) Figure 2 displays only a single frame from the simulations, which might not provide a comprehensive representation. Could a contact frequency heatmap of PROTAC with the proteins be included to offer a more detailed view?

We thank the reviewer for the suggestion! We performed the contact map analysis to observe the average distance between PROTACs and BRD4^BD1 over 400ns of MD simulation (new Figure S4 added).

We included detailed explanation at page 8 and 9: “The residues contact map throughout the 400ns MD simulation also showed different pattern of protein-protein interactions, indicating that the linkers were able to adopt different conformations (Figure S4).”

(4) The conclusions in Figure 3 and S11 are based on a single 400 ns trajectory. The reproducibility of these results is therefore uncertain.

We thank the reviewer for the suggestion! We added one more random seed MD simulation for each PROTAC to ensure the reproducibility of the results. The Result is shown in Figure S21 and the details for each MD run are updated in Table 1.

(5) Figure 4 indicates significant differences between the first and last 100 ns of the simulations. Does this suggest that the simulations have not converged? If so, how can the statistical analysis presented in this paper be considered reliable?

We thank the reviewers for the question. The simulation was initiated with a 10-15A gap between BRD4 and Ub to monitor the movement of degradation machinery and Lys-Gly interaction. The significant changes in pseudo dihedral in Figure 4 shows that the large-scale movement of the degradation complex can initiate the Lys-Gly binding. It does not relate to unstable sampling because the system remains very stable when BRD4 comes close to Ub.

(6) In Figure 5, the dihedral angle of dBET57_#9MD1 is marked on a peptide bond. Shouldn't this angle have a high energy barrier for rotation?

We thank the reviewers for catching the error! Indeed, it was an error that the dihedral angles were marked on the peptide bond. We reworked the figure and double checked our dihedral correlation analysis. The updated correlate dihedral angle selection and the correlation coefficient is shown in Figure 5.

(7) Given that crystal structures for dBET 70, 23, and 57 are available, why is there a need to model the complex using protein-protein docking?

We thank the reviewer for the feedback. Only dBET23 has the ternary complex available in a crystal structure, which has the PROTAC and both proteins, while dBET1, dBET57 and dBET70 are not completed as ternary complexes. Although dBET70 has a crystal structure, its PROTAC’s conformation is not resolved, and thus we decided to still perform protein-protein docking with dBET70. 

We includeed the explanation at page 8: “Only dBET23 crystal structure is available with the PROTAC and both proteins, while the experimentally determined ternary complexes of dBET1, dBET57 and dBET70 are not available. “

(8) On page 9, it is mentioned that "only one of the 12 PDB files had CRBN bound to DDB1 (PDB ID 4TZ4)." However, there are numerous structures of the DDB1-CRBN complex available, including those used for docking like 4CI3, as well as 4CI1, 4CI2, 8OIZ, etc.

We thank the reviewers for the comment! We acknowledged the existence of several DDB1-CRBN complex crystal structures, such as PDB IDs 4CI1, 4CI2, 4CI3, and the more recent 8OIZ. For our study, we chose to use 4TZ4 to maintain consistency in complex construction and to align with the methodology established in a previously published JBC paper (https://doi.org/10.1016/j.jbc.2022.101653), which successfully utilized the same structure for a similar construct. At the time our study was conducted, the 8OIZ structure had not yet been released. We appreciate your suggestion and will consider incorporating alternative structures in future studies to further investigate our findings.

(9) Table 2 is first referenced on page 8, while Table 1 is mentioned first on page 10. The numbering of these tables should be reversed to reflect their order of appearance in the text.

We thank the reviewer for catching the error! We switched the order of Table 1 and Table 2.

Reviewer #3 (Public review):

The authors offer an interesting computational study on the dynamics of PROTAC-driven protein degradation. They employed a combination of protein-protein docking, structural alignment, atomistic MD simulations, and post-analysis to model a series of CRBN-dBET-BRD4 ternary complexes, as well as the entire degradation machinery complex. These degraders, with different linker properties, were all capable of forming stable ternary complexes but had been shown experimentally to exhibit different degradation capabilities. While in the initial models of the degradation machinery complex, no surface Lys residue(s) of BRD4 were exposed sufficiently for the crucial ubiquitination step, MD simulations illustrated protein functional dynamics of the entire complex and local side-chain arrangements to bring Lys residue(s) to the catalytic pocket of E2/Ub for reactions. Using these simulations, the authors were able to present a hypothesis as to how linker property affects degradation potency. They were able to roughly correlate the distance of Lys residues to the catalytic pocket of E2/Ub with observed DC50/5h values. This is an interesting and timely study that presents interesting tools that could be used to guide future PROTAC design or optimization.

Reviewer #3 (Recommendations for the authors):

(1) My most important comment refers to the MM/PBSA analysis, the results of which are shown in Figure S9: binding affinities of -40 to -50 kcal/mol are unrealistic. This would correspond to a dissociation constant of 10^-37 M. This analysis needs to be removed or corrected.

We thank the reviewer for the comment! MM/PBSA analysis indeed cannot give realistic binding free energy. It does not include the configurational entropy loss which should be a large positive value. In addition, while the implicit PBSA solvent model computes solvation free energy, the absolute values may not be very accurate. However, because this is a commonly used energy calculation, and some readers may like to see quantitative values to ensure that the systems have stable intermolecular attractions, we kept the analysis in SI. We edited the figure legend, moved the Figure S10 in SI page 19, and added sentences to clearly state that the calculations did not include configuration entropy loss “Note that the energy calculations focus on non-bonded intermolecular interactions and solvation free energy calculations using MM/PBSA, where the configuration entropy loss during protein binding was not explicitly included. “.

(2) I think that the analysis of what in the different dBETx makes them cause different degradation potency is underdeveloped. The dihedral angle analysis (Figure 4B) did not explain the observed behavior in my opinion. Please add additional, clearer analysis as to what structural differences in the dBETx make them sample very different conformations.

We thank the reviewer for the suggestions! Based on the suggestion, we further performed dihedral entropy analysis for each dihedral angle in the linker part of the PROTAC to examine the flexibility of each PROTAC. Because each PROTAC has a different linker, we now clearly label them in a new Figure S18 in SI page 27. Low dihedral entropies indicate a more rigid structure and thus less flexibility to make a PROTAC more difficult to rearrange and facilitate the protein structural dynamic necessary for ubiquitination.

We added detailed explanation on page 18: “Our dihedral entropy analysis showed that dBET57 has ~0.3 kcal/mol lower configuration entropies than the other dBETs with three different linkers, suggesting that dBET57 is less flexible than the other PROTACs (Figure S18).”

(3) "The movement of the degradation machinery correlated with rotations of specific dihedrals of the linker region in dBETs (Figure 5).": this is not sufficiently clear from the figure. Definitely not in a quantitative way.

We thank the reviewers for the suggestions! To further understand the correlation between PROTACs dihedral angles and the movement of degradation machinery, we performed time dependent correlation analysis to correlate the dihedral angles of the PROTACs and the Lys-Gly distance (Figures 6 and S17).

We included detailed explanation on page 16:

“To further examine the correlation between PROTAC rotation and the Lys-Gly interaction, we performed a time-dependent correlation analysis. This analysis showed that PROTAC rotation translates motion over time, leading to the Lys-Gly interaction, with a correlation peak around 60-85 ns, marking the time of the interaction (Figure 6 and Figure S17). In addition, the pseudo dihedral angles also showed a high correlation (0.85 in the case of dBET1) with Lys-Gly distance. This indicated that degradation complex undergoes structural rearrangement and drives the Lys-Gly interaction.

(4) Cartoons are needed at multiple stages throughout the paper to enhance the clarity of what the modeled complexes looked like (e.g. which subunits they contained).

We thank the reviewers for the suggestions. We added and remade several Figures with cartoons to better represent the stages. We also used higher resolution and included clearer labels for each protein system.

(5) The difference between CRL4A E3 ligase and CRBN E3 ligase is not clear to the non-expert reader.

Thanks for the reviewer’s comment! To clarify the terms "CRL4A E3 ligase" and "CRBN E3 ligase", which refer to different levels of description for the protein complexes, we added a couple of sentences in the Figure 1 legend. As a result, the non-expert readers can clearly know the differences.

As illustrated in Figure 1,

• CRL4A E3 ligase refers to the full E3 ligase complex, which includes all protein components such as CRBN, DDB1, CUL4A, and RBX1.

• CRBN E3 ligase, on the other hand, is a more colloquial term typically used to describe just the CRBN protein, often in isolation from the full CRL4A complex.

(6) Figure 1, legend: unclear why it's E3 in A and E2 in B.

We thank the reviewer for the question! E3 ligase in Figure 1A refers to CRBN E3 ligase, where researchers also simply term it CRBN. We have added a sentence to specify that CRBN E3 ligase is also termed CRBN for simplicity. In Figure 1B, E2 was unclear in the sentences. The full name of E2 should be E2 ubiquitin-conjugating enzyme. Because the name is a bit long, researchers also call it E2 enzyme. We have corrected it and used E2 enzyme to make it clearer. 

(7) "Although the protein-protein binding affinities were similar, other degraders such as dBET1 and dBET57 had a DC50/5h of about 500 nM". It's unclear what experimental data supports the assertion that the protein-protein binding affinities are similar.

We thank reviewer for the question. Indeed, the statement is unclear.

We corrected the sentence in page 6: “Although utilizing the exact same warheads, other degraders such as dBET1 and dBET57 had a DC_50/5h of about 500 nM.”

(8) Was the construction of the degradation machinery complex guided by experimental data (maybe cryo-EM or tomography)? If not, what is the accuracy of the starting complex for MD? This may impact the reliability of the obtained results.

Thank you for your insightful comments! Yes, the construction of the degradation machinery complex was guided by available high-resolution crystal structures, which was selected to maintain consistency and align with the methodology established in a previously published JBC paper (https://doi.org/10.1016/j.jbc.2022.101653).

We acknowledged that static crystal structures represent only a single snapshot of the system and may not capture the full conformational flexibility of the complex. To address this limitation, we performed MD simulations using multiple starting structures. This approach allowed us to explore a broader conformational landscape and reduced the dependence on any single starting configuration, thereby enhancing the reliability of the results.

We hope this clarifies the robustness of our methodology and the steps taken to ensure accuracy in our simulations.

(9) "With quantitative data, we revealed the mechanism underlying dBETx-induced degradation machinery": I think this may be too strong of an assertion. The authors may have developed a mechanistic hypothesis that can be tested experimentally in the future.

We thank the reviewer for the suggestion. This is indeed a strong assertion and needs to be modified. We edited the sentence in page 7: “With quantitative data, we revealed the importance of the structural dynamics of dBETx-induced motions, which arrange positions of surface lysine residues of BRD4^BD1 and the entire degradation machinery.”

(10) Figure S2: are the RMSDs calculated over all residues? Or just the BRD4 residues? Given that the structures are aligned with respect to CRBN, the reported RMSD numbers might be artificially low since there are many more CRBN residues than there are BRD4 residues. Also, why weren't the crystal structures used for dBET 23 and 70 for the modeling? Wouldn't you want to use the most accurate possible structures? Simulations were run for 23. Why not for 70?

We thank the reviewer for the suggestion. We added a sentence to more clearly explain the RMSD calculations in Figure S2: “The structural superposition is performed based on the backbone of CRBN and RMSD calculation is conducted based on the backbone of BRD4^BD1.”

Although dBET70 has crystal structure, its PROTAC structure is not resolved, and thus we decided to still perform protein-protein docking with dBET70.  dBET1 and dBET57 do not have a crystal structure for the ternary complexes.

We included the explanation at page 8: “Only dBET23 crystal structure is available with the PROTACs and both proteins, while the experimentally determined ternary complexes of dBET1, PROTACs of dBET57 and dBET70 are not available. “

a. And there are no crystal structures available for 1 and 57? If so, please clearly say that. Otherwise please report the RMSD.

We thank the reviewer for the suggestion. We included the explanation at page 8: “Only dBET23 crystal structure is available with the PROTACs and both proteins, while the experimentally determined ternary complexes of dBET1, PROTACs of dBET57 and dBET70 are not available.”

(11) Table 2 is referenced before Table 1.

We thank the reviewer for catching the error! We switched the order for Table 1 and Table 2.

(12) Figure S3 is not referenced in the main paper.

We thank the reviewer for catching the error! We now referred Figure S3 on page7.

(13) Minor comments on grammar and sentence structure:

a. It should be "binding of a ternary complex"

b. "Our shows the importance": word missing.

c. "...providing insights into potential orientations for ubiquitination. observe whether the preferred conformations are pre-organized for ubiquitination." Word or words missing.

We thank reviewer for catching the errors! We corrected grammatical errors and unclear sentences throughout the entire paper and revised the sentences to make them easily understandable for non-expert readers.

Introduction and Purpose

“I would like to tell you why fulcro is awesome and why it's much easier to learn than you might believe so we will look at what fulcro is and what it can do for you and why is it interesting...”

• Emphasizes that the talk aims to introduce Fulcro, explain its ease of learning, and highlight its benefits.

Speaker Background

“So first of all who is… I've been doing back-end development since 2006 and front-end development since 2014 on and off…”

• Establishes the speaker’s credibility with extensive development experience.

“...I built learning materials for Fulcro beginners and I pair program with and mentor my private clients on their first Fulcro project...”

• Demonstrates the speaker’s active role in teaching Fulcro to newcomers.

Motivation for Fulcro

“When I create web applications I want to be productive and I want to have fun… I don't want to have to manually track whether the data started loading or finished or failed…”

• Highlights the desire to reduce boilerplate and tedious manual tasks.

“I don't want to write tons of boilerplate and especially not to do that and again and again for every new type data in my application…”

• Stresses that Fulcro removes repetitive coding patterns, enhancing developer efficiency.

Choosing a Full-Stack Framework

“Now there are simpler Frameworks… or you can pick a full stack framework that has all the parts you need…”

• Explains how Fulcro’s integrated approach can be preferable to patching together multiple libraries.

“...malleable web framework designed for sustainable development of real world full stack web applications...”

• Defines Fulcro as a flexible system that supports complex, long-lived applications.

Key Fulcro Capabilities

“It can render data in the UI and it uses React so it's wraps React for that…”

• Confirms that Fulcro uses React under the hood for rendering.

“It can manage state… it keeps the state for you at some place… re-render the UI so it reflects that state…”

• Describes automatic state management and reactive re-rendering.

“It makes it easy to load data from the backend… you have full control...”

• Emphasizes the fine-grained control over data fetching.

“Fulcro also caches the data for you automatically and it does so in normalized form…”

• Highlights how normalized data storage simplifies updates across the UI.

“Fulcro has excellent developer experience for multiple reasons… the biggest is locality and navigability…”

• Points out how Fulcro keeps relevant code together, making it easier to navigate and maintain.

Core Principles

1. Graph API / EQL (Edn Query Language)

“...we use graph API instead of rest API which means that we have just a single endpoint and it's the front end which asks the back end for what data it wants by sending over a query…”

• Simplifies data retrieval by letting the client specify exactly what it needs.

• UI as Pure Function of State

“UI is pure function of state… components only ever get the data they need from their parent…”

• Removes side effects from the rendering flow.

• Locality

“...to understand the UI component I shouldn't be forced to jump over four different files… so in Fulcro a component doesn’t have only a body but also a configuration map…”

• Co-locates component queries, rendering, and logic in one place.

• Normalized Client-Side State

“...it stores that data normalized in a simple tabular form where entities contain other entities replaced with references…”

• Ensures any update in one place is reflected throughout the UI.

Architecture Overview

“...it's a full stack web framework so it has the front end and back end part… front end is Fulcro proper… the back end is Fulcro’s library Pathom…”

• Describes the division between the Fulcro client and the Pathom-based server.

“On the front end… we have client DB… we have a transactional subsystem… to the back end we have Pathom… as kind of adapter between the tree of data the UI wants and whatever data sources there are.”

• Clarifies how Fulcro’s client and server components communicate via EQL queries and mutations.

UI Rendering Process

“...UI is a tree of components and for each component we have a query… these queries are composed up so that the root component’s query is the query for the whole page.”

• Outlines how each component declares its data needs, culminating in a single root query.

“...Fulcro takes this query, combines it with the client DB, and forms a tree of data that matches the query shape, then hands it off to the root to render.”

• Demonstrates the round-trip from query to final rendered UI.

Component Example

“Here we can see how a Fulcro component looks in code… The most important part here is the query…”

• Provides a code snippet showing query co-location with the component.

“...the component also includes the queries of its child components so the parent can pass down just the needed data.”

• Reinforces that data flows naturally down the component tree.

Learning Fulcro

“People have this assumption or believe that Fulcro is hard to learn but it's not…”

• Dispels the notion of steep difficulty.

“There are simpler frameworks that do just one thing… but you need to handle a number of tasks and that you need to work across both front end and back end…”

• Explains why novices might find full-stack solutions initially overwhelming.

“You need to rewire your brain… if you come in expecting that things just work the way you expect you will be running into walls…”

• Advises a mindset shift for those accustomed to different paradigms.

Recommended Beginner Resources

“...the Fulcro Developer's Guide… it describes everything in great detail but it can be overwhelming…”

• Mentions the official documentation’s comprehensive nature.

“...start with the do it yourself Fulcro Workshop… play with the concepts in practice and see how they work...”

• Suggests hands-on learning as the best first step.

“...there's this minimalist Fulcro tutorial… tries to teach you the absolute minimum amount of things you need to know…”

• Recommends a focused tutorial that avoids overload.

Simplicity Through Principles

“Fulcro doesn't do any magic… its operation is straightforward and very much possible to understand…”

• Emphasizes that Fulcro’s complexity is principled, not opaque.

“...UI is pure function of data, standard input of data is the graph API, standard output of side effects is the transaction subsystem, and data is data, meaning queries and mutations are just data.”

• Summarizes how Fulcro simplifies data handling, state management, and side effects uniformly.

Demo Highlights

“So let's have a demo… a simple Fulcro application showing todo list…”

• Introduces a working demonstration of a to-do list in Fulcro.

“...every side effect goes through transaction subsystem so I should see data here and I do, I see that they are loading them…”

• Illustrates how Fulcro logs and displays all transactions for debugging.

“I can also see the response… the data mirrors the query… if I ask for something that doesn't exist I get back empty data…”

• Demonstrates the transparency of EQL-based queries and responses.

Conclusion and Key Takeaways

“Takeaways… that full stack frameworks are really useful and especially that Fulcro is really worth looking into and learning is not hard if you are a little smart about it…”

• Concludes that Fulcro offers an approachable path to building maintainable full-stack ClojureScript applications.

“Here are some awesome resources especially the Fulcro Community guide where you find the workshop and tutorial…”

• Reiterates the availability of community-driven materials to support new learners.

Summary of the Tech Talk on Software Development Leverage

Speaker's Background & Context

• The speaker has experience with nine startups, with four successes (defined as acquired or still operational).

  "I've been involved in nine startups, four successes so far, success defined as either bought by somebody else or still exists." - Core interests include minimal degradation over time, maximum architectural clarity, and minimal boilerplate.

  "I want to build systems that have a minimal amount of that maximum architecture clarity... I want a small number of Core Concepts and I also want minimal boilerplate." - Prefers Clojure and ClojureScript due to Lisp features, a REPL, macros, full-stack capabilities, and immutable data.

  "The main things are that it's a Lisp, I've got a REPL, I've got macros, I've got full stack language immutable data and literals."

Concept of Software Development Leverage

• Defines leverage in software as maximizing efficiency while minimizing incidental complexity.

  "What’s the minimal amount of code I can write to build these things?" - Software generally consists of forms and reports, and optimizing these elements reduces complexity.

  "A lot of what we write are forms or reports essentially." - Critiques past attempts at UI and form abstraction (e.g., Informix 4GL, Visual Basic, Rails, Java Enterprise) as insufficient or overly complex.

  "Every kind of library on the planet trying to do the same sort of thing." - Identifies challenges in leverage: short levers, fragile systems, opposing mindsets, and complex structures.

  "You can have too short of a lever, the object that we're trying to move could be too big for the lever, or my strength... I could have a crowd of people who are just philosophically opposed to levers."

Key Approaches to Leverage

• Minimal Incidental Complexity: Reducing unnecessary complexity that accumulates over time.

  "We love minimal incidental complexity... other communities don’t even think about that." - Functional & Immutable Data Models: Advocates for a pure functional approach to state management and UI rendering.

  "The state of the world is some immutable thing, initialized somehow, then I walk from step to step running some pure function." - Generalized Pure Functions: Aiming for functional purity while acknowledging that some dynamism is needed.

  "To me, you're starting by breaking the ideal. You're saying, 'I’m not really going to use pure functions for that.'" - Component-Based Rendering: Prefers data-driven UI, minimizing reliance on React’s event-based state management.

  "A pure function, a render of some sort of transform of the world."

Core Abstractions for Software Leverage

1. Entity-Attribute-Value (EAV) Model: A flexible, normalized data structure for representing application state.

   "The first one is just the power of entity attribute value." 2. Idents (Universal Entity Identifiers): Unique tuples ([type id]) for referencing entities.

   "The kind allows you to prevent collisions... useful semantic information." 3. Graph Queries: Uses EDN-like queries to efficiently pull and update data.

   "Attach logic to graph queries that say when you get the result of this query, here's how you normalize it." 4. Full-Stack Datified Mutations: CQRS-like abstractions over side effects and state transitions.

   "CQRS kind of idea... I’m going to make an abstract thing that says what I want to do."

Emergent Benefits of This Approach

• Normalized State Representation: Enables automatic merging of data, reducing complexity in state updates.

  "This gives me on my world, my immutable World in that diagram of kind of our idealized application." - Minimizing UI Boilerplate: Using annotated queries and data-driven components reduces manual UI code.

  "A UI location-specific way to annotate my UI... initial state is just a mirror of that." - Abstracting Side Effects: Remote calls and transactions become well-structured, reducing ad-hoc state management.

  "Transact things... processing system talks to remotes for side effects, talks to the database for local changes, and triggers renders."

State Machines for Process Control

• Advocates state machines for handling application logic, avoiding scattered imperative code.

  "Very often, process is just peppered around everywhere... having a state machine that abstracts over this is powerful." - Uses state charts (Harel state machines) for complex workflows like authentication.

  "State charts are way better when your state machine gets large."

Fulcro & RAD (Rapid Application Development)

• Fulcro: A ClojureScript-based framework built on these principles.

  "How do I simplify F? How do I get these core pieces generic enough to reuse?" - RAD: Built to automate UI and backend generation, minimizing redundant work.

  "I really wanted to minimize the boilerplate right... tired of handwriting schema." - Plugins for Databases, Forms, Reports, and APIs: Reduces custom implementation for common application patterns.

  "Datomic support gives me my network API and integration with Datomic in 1900 lines of code."

Key Takeaways

• Graph-based, normalized application state leads to better leverage and scalability.
• Functional purity where possible, and controlled side effects when necessary.
• Automatic UI and backend generation through metadata and introspection.
• Composable, small-core abstractions allow flexibility without unnecessary complexity.

"A very small number of Core Concepts... it's pluggable, you can escape from everything... it's just an annotated data model."

This approach significantly reduces the long-term maintenance cost of applications by emphasizing reusability, composition, and functional principles.

I've seen this myself. Whenever my mom goes back to her home country, Bangladesh, to see her family, she brings all sorts of goods from NYC. I was under the impression that these goods didn't exist at all in Bangladesh, but it turns that was not the case. For example, she bought a decent amount of drugstore makeup for them and mentioned that they requested some of it. She also said that they have most of these same (drugstore) makeup brands in Bangladesh; it's just way more expensive and therefore not worth paying for the quality, so they ask her to bring it for them when she visits.

Reflection: This article caught my attention because my partner just got an Oura Ring after we talked about it over the holidays. We were both curious about how well it actually tracks sleep and if it’s worth the hype. I liked how the article broke down the features in a way that made sense, especially the focus on comfort and how it’s less bulky than other trackers. The idea that it measures things like heart rate and body temperature to give a full picture of sleep is really interesting. My partner has already started using it, and we have been looking at the data together, trying to see if it actually helps with better sleep habits. I also liked how the article explained the app in a simple way, since some tech reviews can feel too complicated. The battery life seems like a big plus too, since constant charging can be annoying. Reading this made me think about how good writing makes even technical topics easy to understand, which is something I want to work on in my own writing.

This is one of the most irritating things in the English language to me. Studying Japanese, one of the first things you're taught is that there's no stress on any syllable more than any other when saying a word. It's a Mora-timed language, where English is a stress-timed language. Mora-timed languages don't put stress on certain syllables the way we do in English. By changing the way we are using a word, and thereby changing our intonation, the language just keeps getting more confusing; stress one syllable wrong and everyone in a three mile radius will go, "Why did you say that like that?"

I had no idea that either endings came from different languages. I guess I just assumed that they were already apart of the English language.

I agree with this idea because it shows that people aren't just passively living through history instead they are also making their own stories about it. It makes sense that people can be shaped by history and at the same time try to make sense of it through their own perspective. But this can cause confusion because it’s hard to separate what’s really happening in history from the way we tell our own personal stories about it.

This particular quote's reference of "secret things" is not in regard to actual secrets but to the truth or origin of life. By using such examples of "lore of all the learned, the seed of all which springs" is an explanation that all that exists is because of and within us all. This shows that all of nature owes it's existence and origin to God but there is a bit of that God (or divine) in each of us showing that God is not just around us but within us all as well. "All objects, both manifest and unmanifest, including nature are created out of God at the beginning of the kalpa and after completing their life cycles all objects dissolve into God at the end of the kalpa (creation)". This shows that the cycle of life ends where it begins. By being born and created from a God and ending that cycle by becoming a part of that God's divinity. Source: Mohanty, Susil Kumar. International Journal of Science and Research (IJSR). "Scientific Analysis of “The Bhagavad Gita” on God Reflecting Ancient Indian Culture". Vol. 13. 2024.

This highlights the ethical dilemma of social media’s data-driven business model. While targeted ads can be useful and even beneficial to consumers, they also create risks when used to exploit vulnerabilities—such as addictive behaviors or manipulative content. The pursuit of engagement at all costs raises concerns about user autonomy, mental health, and the broader societal impact of algorithm-driven influence. It’s a reminder that the power these platforms hold isn’t just about profit—it shapes culture, behavior, and even democracy.

What makes him view these words as useless? I am going to college for animation and feel the words living, dead, and human are all great descriptors for how one's art can look. He makes a good point about it being an opinion of how the art looks, but all art is subjective. One person pours their views of the world into a visual piece, and people may interpret in many different words. I do think that you should include what makes the art seem alive or dead, not just use those words as sole descriptors. e.g. "The outstanding feature of Mr. X's work is it's ability to imitate living quality through it's _ and _. As I write this, I can understand how it could be seen as a meaningless word... interesting.

(I just reverse-engineered the naming convention of the instrument the Euphonium... Eu/phon/ium--literally just "The good sound -er") On another note. Where is the dividing line, Orwell? does the use of this word here not fall into the category of "Pretentious Diction" as you laid out before, given that it's a Greek based word?

...Okay I can't find the flaw. I can feel it certainly, it's so wordy I can hardly follow, but... I don't know---this isn't the example of "Bad Writing" I'd have expected to find in an English class...

There isn't anything completley wrong with using words like fragmentory I think it just has to do with the mindset of the one writing it. Orwell seems to have a problem with people writing who don't actually do so mindfully. To me this is still about the overuse of these words but it's because they tend to come from people who don't absorb their own ideas.

I was seriously thinking the same thing. There were some snippets that invoked some sort of imagery, but they never felt engaging. The other point I totally agree. It's just a bunch of fluff that doesn't really mean anything.

The film depicts how in times of true desperation to control their own reproductive choices women came together to create something bigger than themselves. Despite the constant threat of legal repercussions, they created a support network that prioritized women’s health and autonomy.

The "Jane" network wasn't just about providing a medical service it was about community, trust, and the fight to ensure women didn’t have to navigate their most personal decisions in isolation. In today’s world, that sense of solidarity is just as important. Whether it’s the voices of activists, healthcare providers, or everyday women demanding better protections for reproductive rights, the power of women coming together to fight for change is the heart of the movement.

This reminds me of the discourse that took place with The Little Mermaid live action movie. It is so bizarre that where people draw the line with a mythical and magical character like a mermaid is their race. It's okay if they have every magical power but being Black is just too unrealistic. This discourse is cyclical.

I dont know if it's just me, but the idea of having a movie that profits off a story about the abuse of a black girl feels... wrong? Critics are talking about it like some show, and while it is a film, it is one that should provoke thought and dicussion rather than hype. I dont know if that makes sense

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

This study uses single nucleus multiomics to profile the transcriptome and chromatin accessibility of mouse XX and XY primordial germ cells (PGCs) at three time-points spanning PGC sexual differentiation and entry of XX PGCs into meiosis (embryonic days 11.5-13.5). They find that PGCs can be clustered into sub-populations at each time point, with higher heterogeneity among XX PGCs and more switch-like developmental transitions evident in XY PGCs. In addition, they identify several transcription factors that appear to regulate sex-specific pathways as well as cell-cell communication pathways that may be involved in regulating XX vs XY PGC fate transitions. The findings are important and overall rigorous. The study could be further improved by a better connection to the biological system, including the addition of experiments to validate the 'omics-based findings in vivo and putting the transcriptional heterogeneity of XX PGCs in the context of findings that meiotic entry is spatially asynchronous in the fetal ovary. Overall, this study represents an advance in germ cell regulatory biology and will be a highly used resource in the field of germ cell development.

Strengths:

(1) The multiomics data is mostly rigorously collected and carefully interpreted.

(2) The dataset is extremely valuable and helps to answer many long-standing questions in the field.

(3) In general, the conclusions are well anchored in the biology of the germ line in mammals.

Weaknesses:

(1) The nature of replicates in the data and how they are used in the analysis are not clearly presented in the main text or methods. To interpret the results, it is important to know how replicates were designed and how they were used. Two "technical" replicates are cited but it is not clear what this means.

The two independent technical replicates comprised different pools of paired gonads. This sentence was added to the methods section of the revised manuscript.

(2) Transcriptional heterogeneity among XX PGCs is mentioned several times (e.g., lines 321-323) and is a major conclusion of the paper. It has been known for a long time that XX PGCs initiate meiosis in an anterior-to-posterior wave in the fetal ovary starting around E13.5. Some heterogeneity in the XX PGC populations could be explained by spatial position in the ovary without having to invoke novel subpopulations.

We thank the reviewer for pointing out this important biological phenomenon. We also recognize that transcriptional heterogeneity among XX PGCs is likely due to the anterior-to-posterior wave of meiotic initiation in E13.5 ovaries and highlight this possibility in our manuscript. However, since our study utilizes single-nucleus RNA-sequencing and not spatial transcriptomics, we are not able to capture the spatial location of the XX PGCs analyzed in our dataset. As such, our analysis applied clustering tools to classify the populations of XX PGCs captured in our dataset. 

(3) There is essentially no validation of any of the conclusions. Heterogeneity in the expression of a given marker could be assessed by immunofluorescence or RNAscope.

In our revised manuscript, we included immunofluorescence staining of potential candidate factors involved in PGC sex determination, such as PORCN and TFAP2C. Testing and optimizing antibodies for the targets identified in this study are ongoing efforts in our lab and we look forward to sharing our results with the research community.

(4) The paper sometimes suffers from a problem common to large resource papers, which is that the discussion of specific genes or pathways seems incomplete. An example here is from the analysis of the regulation of the Bnc2 locus, which seems superficial. Relatedly, although many genes and pathways are nominated for important PGC functions, there is no strong major conclusion from the paper overall.

In this manuscript, we set out to identify candidate factors, some already known and many others unknown, involved in the developmental pathways of PGC sex determination using computational tools. Our goal, as a research group and with future collaborators, is to screen these interesting candidates and discover their function in the primordial germ cell. Our research, presented in this study, represents a launching pad for which to identify future projects that will investigate these factors in further detail.

Reviewer #2 (Public Review):

Summary:

This manuscript by Alexander et al describes a careful and rigorous application of multiomics to mouse primordial germ cells (PGCs) and their surrounding gonadal cells during the period of sex differentiation.

Strengths:

In thoughtfully designed figures, the authors identify both known and new candidate gene regulatory networks in differentiating XX and XY PGCs and sex-specific interactions of PGCs with supporting cells. In XY germ cells, novel findings include the predicted set of TFs regulating Bnc2, which is known to promote mitotic arrest, as well as the TFs POU6F1/2 and FOXK2 and their predicted targets that function in mitosis and signal transduction. In XX germ cells, the authors deconstruct the regulation of the premeiotic replication regulator Stra8, which reveals TFs involved in meiosis, retinoic acid signaling, pluripotency, and epigenetics among predictions; this finding, along with evidence supporting the regulatory potential of retinoic acid receptors in meiotic gene expression is an important addition to the debate over the necessity of retinoic acid in XX meiotic initiation. In addition, a self-regulatory network of other TFs is hypothesized in XX differentiating PGCs, including TFAP2c, TCF5, ZFX, MGA, and NR6A1, which is predicted to turn on meiotic and Wnt signaling targets. Finally, analysis of PGC-support cell interactions during sex differentiation reveals more interactions in XX, via WNTs and BMPs, as well as some new signaling pathways that predominate in XY PGCs including ephrins, CADM1, Desert Hedgehog, and matrix metalloproteases. This dataset will be an excellent resource for the community, motivating functional studies and serving as a discovery platform.

Weaknesses:

My one major concern is that the conclusion that PGC sex differentiation (as read out by transcription) involves chromatin priming is overstated. The evidence presented in the figures includes a select handful of genes including Porcn, Rimbp1, Stra8, and Bnc2 for which chromatin accessibility precedes expression. Given that the authors performed all of their comparisons between XX versus XY datasets at each timepoint, have they missed an important comparison that would be a more direct test of chromatin priming: between timepoints for each sex? Furthermore, it remains possible that common mechanisms of differentiation to XX and XY could be missing from this analysis that focused on sexspecific differences.

We thank the reviewer for their thoughtful assessment and suggestions, as stated here. We note that chromatin priming in PGCs prior to sex determination is a well-documented research finding (see references below), that is further supported by our single-nucleus multiomics data. To support these findings previously stated in the scientific literature, we included data demonstrating the asynchronous correlation between chromatin accessibility and gene expression during PGC sex determination. Specifically, we investigated the associations of differentially accessible chromatin peaks with differentially expressed gene expression for each PGC type (between sexes and across embryonic stages) using computational tools and methods that are well-established and applied by the research community. In our manuscript, we note that the patterns we identified support the potential role of chromatin priming in PGC sex determination. Nevertheless, we further highlight that a comprehensive profile of 3D chromatin structure and enhancer-promoter contacts in differentiating PGCs is needed to fully understand how changes to chromatin facilitate PGC sex determination.

References:

(1) Chen, M., et al. Integration of single-cell transcriptome and chromatin accessibility of early gonads development among goats, pigs, macaques, and humans. Cell Reports 41 (2022).

(2) Huang, T.-C. et al. Sex-specific chromatin remodelling safeguards transcription in germ cells. Nature 600, 737–742 (2021).

Reviewer #3 (Public Review):

Summary:

Alexander et al. reported the gene-regulatory networks underpinning sex determination of murine primordial germ cells (PGCs) through single-nucleus multiomics, offering a detailed chromatin accessibility and gene expression map across three embryonic stages in both male (XY) and female (XX) mice. It highlights how regulatory element accessibility may precede gene expression, pointing to chromatin accessibility as a primer for lineage commitment before differentiation. Sexual dimorphism in these elements and gene expression increases over time, and the study maps transcription factors regulating sexually dimorphic genes in PGCs, identifying sex-specific enrichment in various transcription factors. Strengths:

The study includes step-wise multiomic analysis with some computational approach to identify candidate TFs regulating XX and XY PGC gene expression, providing a detailed timeline of chromatin accessibility and gene expression during PGC development, which identifies previously unknown PGC subpopulations and offers a multimodal reference atlas of differentiating PGC clusters. Furthermore, the study maps a complex network of transcription factors associated with sex determination in PGCs, adding depth to our understanding of these processes.

Weaknesses:

While the multiomics approach is powerful, it primarily offers correlational insights between chromatin accessibility, gene expression, and transcription factor activity, without direct functional validation of identified regulatory networks.

As stated in our response above to a similar concern, we note that our research study represents a launching pad for which to identify future projects that will investigate candidates that may be involved in PGC sex determination, in further detail. With this rich dataset in hand, our goal in future research projects is to screen these candidates and discover their function in PGCs. 

Response to Recommendations

Reviewer #1 (Recommendations For The Authors):

(1) Clarify at first introduction how combined ATAC-seq/RNA-seq mulitomics libraries were prepared, including if ATAC and RNA-seq data are from the same cell.

This information was added to the introduction of the revised manuscript.

(2) Clarify what the two technical replicates represent. Are they two libraries from the same gonad or the same pool of gonads? Are they from 2 different gonads?

The two independent technical replicates comprised different pools of paired gonads. This sentence was added to the methods section of the revised manuscript.

(3) In Supplemental Figure 1, there is substantial variation in the number of unique snATAC-seq fragments between some conditions. Could this create a systematic bias that affects clustering?

We recognize the concern that substantial variation in the number of unique snATAC-seq fragments between conditions could potentially create a systematic bias that affects clustering. However, we analyzed our snATAC-seq dataset with Signac, which performs term frequency-inverse document frequency (TF-IDF) normalization. This is a process that normalizes across cells to correct for differences in cellular sequencing depth. Given that sequencing depth was taken into account in our normalization and clustering procedures, and that the unbiased clustering of PGCs also reflects the sex and embryonic stage of PGCs, we are confident that the clustering of the snATAC-seq datasets closely reflects the biological variability present in the PGCs collected.

References:

Signac Website:  https://stuartlab.org/signac/articles/pbmc_vignette

Stuart, T., Srivastava, A., Madad, S., Lareau, C. A., & Satija, R. (2021). Single-cell chromatin state analysis with Signac. Nature methods, 18(11), 1333-1341.

(4) In Figures 2a, 2e, 3a, and 3e, the visualization scheme is very difficult to follow. It's very hard to see the colors corresponding to average expression for many genes because the circles are so small. In addition, the yellow color is hard to see and makes it hard to estimate the size of the circle since the boundaries can be indistinct. I recommend using a different visualization scheme and/or set of size scales be used.

In Figures 2a, 2e, 3a, and 3e, we chose this color palette to be inclusive of viewers who are colorblind. The chosen colors are visible on both a computer screen and on printed paper. We also included a legend of the color scale and dot size representing the average expression and percent of cells expressing the gene, respectively. If the color cannot be seen, it is because the cell population is not expressing the gene.

(5) Perform in vivo validation (immunofluorescence or RNAscope) of at least some targets implicated in PGC development by this study.

Such validations (immunofluorescence staining of PORCN and TFAP2C) are now included in Figure 4 and the supplement.

(6) In line 351, the authors state that "we observed a strong demarcation between XX and XY PGCs at E12.5-E13.5." But in Figure 1j it looks like a reasonably high fraction of both XX and XY E12.5 cells are in cluster 1, which should mean that there is some overlap.

While it is true that Figure 1j shows overlap of both XX and XY E12.5 cells in cluster 1, we were commenting on the separation of E12.5 XX (clusters 4 and 5) and E12.5 XY (clusters 8 and 9) PGCs. We have modified the sentence beginning at line 351 to state that the separation between XX and XY PGCs occurs at E13.5.

(7) In lines 404-405: "We first linked snATAC-seq peaks to XY PGC functional genes". It is important to know how the peaks were linked to genes.

We added the following sentence to address this comment: “Peak-to-gene linkages were determined using Signac functionalities and were derived from the correlation between peak accessibility and the intensity of gene expression.”

(8) In Supplemental Figure 5c, the XX E11.5 condition has a substantially higher fraction of ATAC peaks at promoter regions compared to the others. Does this have statistical and biological significance?

This is an interesting observation beyond the scope of our manuscript. Many interesting questions arise from this study and it is our plan to investigate further in the future. 

(9) Line 885: "The increased number of DA peaks at E13.5 may be the result of changes to chromatin structure as XX PGCs enter meiotic prophase I"; but in Figure 4b, there's only a modest increase in DAP number from E12.5 to E13.5 in XX PGCs, compared to a massive gain in XY PGCs.

In our manuscript, we comment on both phenomena: the doubling of differentially accessible peaks in XX PGCs from E12.5 to E13.5 and the massive increase in differentially accessible peaks in XY PGCs from E12.5 to E13.5. In our description of these results, we propose several hypotheses leading to these increases in differentially accessible peaks. As such, it cannot be ruled out that the changes to chromatin structure that occur during meiotic prophase I contribute to the gain in differentially accessible peaks in XX PGCs at E13.5, and we included this statement in the manuscript accordingly.

Reviewer #2 (Recommendations For The Authors):

(1) The methods state at line 141 that nuclei with mitochondrial reads of more than 25% were removed, however our understanding from the Bioconductor manual and companion manuscript (Amezquita, R.A., Lun, A.T.L., Becht, E. et al. Orchestrating single-cell analysis with Bioconductor. Nat Methods 17, 137-145 (2020). https://doi.org/10.1038/s41592-019-0654-x) is that snRNA-seq approaches remove mitochondrial transcripts entirely and datasets containing mitochondrial transcripts are thought to feature incompletely stripped nuclei. It is thought that mitochondrial transcripts participating in nuclear import may remain hanging on to the nuclear envelope and get encapsulated into GEMs. If the mitochondrial read cutoff of 25% was used intentionally to keep this potentially contaminating signal, please justify why this was done for this dataset.

We agree with the reviewer that the presence of mitochondrial transcripts may be potentially contaminating signal. In our preprocessing steps, we removed the mitochondrial genes and transcripts from our datasets so that they would not influence or affect our analyses. The following sentence was added to the methods section on snRNA-seq data processing: “Mitochondrial genes and transcripts were removed from the snRNA-seq datasets to eliminate any potentially contaminating signal.”

(2) Methods line 227: please include log2fold change and p-adjusted value cutoffs for GO enrichment.

We used clusterprofiler for our GO enrichment analysis. Our GO enrichment analysis did not include a log2fold change analysis and the p-adjusted value cutoff is stated in the methods.

(3) Results line 310: the claim that "At E12.5-E13.5, XY PGCs converged onto a single distinct population (cluster 7), indicating less transcriptional diversity among E12.5-E13.5 XY PGCs when compared to E12.5E13.5 XX PGCs (Fig1d)" would be strengthened if the authors quantified transcriptional distance with distance metrics such as euclidean or cosine distance.

We used a clustering approach to gain insights into the transcriptional diversity of PGC populations. Using an additional metric, such as Euclidean or cosine distance, would not provide meaningful information not already achieved by clustering or change the conclusions presented in the manuscript.

(4) Results line 317: the authors allude to Lars2 defining clusters 2 & 3 as a marker gene, but it is not clear why this is highlighted until the reader reaches the discussion, which alludes to the published role of Lars2 in reproduction. Please consider moving this sentence to the results section for clarity and perhaps expanding the discussion on the meaning.

To provide clarity, we added the statement “genes with reported roles in reproduction” to the results section.

(5) In Figure 2a, why do the authors choose to focus on Zkscan5 in XY PGCs when it is expressed by such a small portion of cells (<25%)? Do they assume that this is due to dropouts?

We chose to focus on Zkscan5 as an example because of its enriched and differential expression in male PGCs, the motif for Zkscan5 is not enriched in female PGCs, and the reported roles of Zkscan5 in regulating cellular proliferation and growth. Zkscan5 is an example of how candidate genes can be identified for further investigation.

(6) Line 461: "the population of E13.5 XX PGCs displaying the strongest Stra8 expression levels corresponded to the same population of XX PGCs with the highest module score of early meiotic prophase I genes (Figure 3c; Supplementary Fig. 3a-b)". However did the authors also consider examining the Stra8+ XX PGCs that do not robustly express meiotic genes to understand more about their differentiation potential?

We are thankful to the reviewer for this suggestion. However, this research question is beyond the scope of the manuscript. We plan to investigate further in future research studies.

(7) Line 505: "when we searched for the presence of RA receptor motifs in peaks linked to genes related to meiosis and female sex determination, we found that Stra8, Rec8, Rnf2, Sycp1, Sycp2, Ccnb3, and Zglp1 contain the RA receptor motifs in their regulatory sequences (Supplementary Figure 4g)." My read of the text is that the authors are not taking a side on the RA and meiosis controversy, but rather trying to reveal what the data can tell us, and the answer is that there is a strong signature linking RA to meiotic genes, which supports this as a valid biological pathway. But what is the strength of the RA>meiosis pathway compared to other mechanisms (which must be functioning in the triple receptor KO)? Perhaps the authors could take this analysis further with the following questions: (1) ask whether meiotic genes are more enriched in RA motifs compared to other expressed genes or other motifs (2) compare the strength of peak-gene correlations for all peaks containing RA receptor motifs vs. those with peaks for Zglp1, Rnf2, etc binding. The strengths of these correlations could provide clues to how much gene expression varies in response to RA exposure vs. modulation of these other factors and thus tell us something about how much RA is playing a role.

We agree with the reviewer that this is a very interesting and important question. We also thank the reviewer for their thoughtful suggestions on the types of bioinformatics analyses that could answer this question. However, the section on RA signaling during PGC sex determination is only a small part of the manuscript and would be better analyzed in greater detail in a future research study or publication.

(8) The shift from promoters in E11.5 XX PGCs to distal intergenic regions is fascinating. What can we learn about epigenetic reprogramming/methylation changes across gene bodies? 

We agree with the reviewer that this is an interesting question about gene regulation in E11.5 XX PGCs. However, we prefer to analyze the epigenetic reprogramming changes across gene bodies in this cell population in additional research studies. Our purpose and goal for this section was to link differentially accessible chromatin peaks with differentially expressed genes to identify putative gene regulatory networks.

(9) Line 581: why did the authors choose to highlight and validate PORCN1 in PGCs? Please elaborate.

As stated in the manuscript, we chose to highlight and validate PORCN1 in PGCs because of its role in WNT signaling and because of the visibly strong correlation between chromatin accessibility at the XXenriched DAP in Fig. 4c (dashed box) and and gene expression of PORCN1.

(10) Figure 5f would be easier to interpret if presented as two columns rather than a circle; show one line of the proteins and the other line with the transcripts so that each is on the same line and there are connections between them.

This comment is related to stylistic preferences. The purpose of Fig. 5f is to demonstrate that the candidate transcription factors may regulate the expression of other enriched transcription factors. Figure 5f figure accomplishes this goal.

(11) Line 640: "The predicted target genes of TCFL5 totaled 74% (367/494) of all DEGs with peak-to-gene linkages in XX PGCs". This seems like a high number and a lot of work for just TCFL5; given the overlap between other TFs and target genes, how many of these 367 target genes overlap with other TFs?

We agree with the reviewer that this is an important declaration to make. We added the following sentence to the results section on TCFL5: “A large majority of the predicted target genes of TCFL5 were also predicted to be the target genes of the enriched TFs presented in Fig. 5e, e.g., the predicted target genes of these TFs overlapped with 4%-100% of the predicted target genes of TCFL5.”

(12) The presentation of TCFL5 in the results section would make more sense with the additional mention of reproductive phenotypes already known (currently in the discussion Lines 914-917). I would furthermore suggest that the discussion goes into more depth on the difference between the regulatory network of TCFL5 in XX meiosis vs XY.

We thank the reviewer for this comment, however, we already state in the results section that TCFL5 is known to influence XX PGC sex determination.

(13) In the Methods, please state more clearly for those not familiar that the genetic background of mice is mixed.

We described the mice with their official names, which provides the context of their genetic backgrounds.

(14) Please specify which morphologic criteria were used to verify the stage of embryos in the methods.

We added the following text to the methods section of the revised manuscript: “Plug date was used to determine the stage of embryos collected for single-nucleus RNA-seq and ATAC-seq. The stage of E11.5 embryos was confirmed by counting somites. The stage of embryos collected at E12.5 was confirmed by the morphological presence of the vessel and cords of the testes collected from XY embryos. Similarly, we confirmed the stage of embryos collected at E13.5 by the size of the gonads, the presence of more distinct cords in the testes of XY embryos, and the elongation of the ovaries of XX embryos.”

(15) The total number of cells and PGCs that passed QC and are included in UMAPS should be stated.

The requested information was added to the legend for Fig. 1 of the revised manuscript: “The number of PGCs per sex and embryonic stage are: 375 E11.5 XX PGCs; 1,106 E12.5 XX PGCs; 750 E13.5 XX PGCs; 110 E11.5 XY PGCs; 465 E12.5 XY PGCs; and 348 E13.5 XY PGCs.”

(16) The order of timepoints changes between figures, and this is not for any obvious reason. Please make it consistent. Figures 1 and 6 list XX 11.5, 12.5, 13.5, and the same for XY, but Figures 2, 3, and 4 use the reverse order: XY E13.5, E12.5, E11.5, and then XX. 

We thank the reviewer for this comment. However, we chose this order for each of the figures to match the coordinates of the graphs and where we would expect the reader to begin reading the graph first. For example, in Figure 3a, XX E11.5 is closest to the x-axis and would be expected to be read first.   

(17) In Figure S2 the colors of clusters are hard to distinguish, and it is suggested that the cluster numbers should be listed above each colored bar to avoid frustration.

We made the suggested correction to Figure S2.

(18) In Figures 2e and 3e: what do the dashed boxes indicate?

The dashed boxes are to guide the reader’s eyes to the fact that the order of transcription factors/genes under the Cistrome DB regulatory potential score and gene expression plots are the same.

(19) In Figure 5a: break panels into i-iv so that the in-text call-outs are not all the same.

We made the suggested correction to Figure 5a and modified the in-text call-outs.

(20) Please indicate XX in Figure 5e and XY in Figure 5l.

We made the suggested correction to Figure 5e and 5l.

(21) In Figure S5c: Please reorganize DA chromatin peak charts so that columns are XX and XY with rows at the same timepoint.

We made the suggested correction to Figure S5c.

(22) In Figure S7a: please make images larger so that the overlapping expression of PORCN and TRA98 is more visible, and consider adding a more magnified panel.

This image is now included in the main text, with expanded panels.

(23) Line 742-754: this seems like a long introduction for the results section; please consider tightening it up.

We believe this text is important and necessary to provide context to the bioinformatics analyses of cell signaling pathways in PGCs. Not all readers will be familiar with the ligand-receptor signals between gonadal support cells and PGCs, and this text provides details on which signaling pathways are known to direct sex determination of PGCs.

(24) For UMAP plots in Figures 2c, 3c, S3b, and S4b, the text overlaid with the timepoints and sexes onto the UMAP plots is misleading, as it allows the reader to presume that the entire group of cells for a given sex/timepoint is located in the location of the text overlay. However, from the UMAP plots in Figure 1i-j, it is clear that the cells from a given sex/timepoint are actually spread across multiple identified clusters. Thus, the overlaid text obscures the important heterogeneity detected. To better represent the actual locations on the UMAP plot of cells from each sex/timepoint, it would be better to show inset density plots alongside these UMAP plots so the reader can locate the cells for themselves. 

We thank the reviewer for this comment. However, we chose this formatting to offer simplicity and ease of understanding to our UMAPs in addition to highlighting the general biological patterns of gene expression. If the reader is interested in discerning more of the heterogeneity of the UMAPs, they may refer back to Figure 1.

Reviewer #3 (recommendations for the authors):

There are some errors or places that need clarification or corrections:

(1) Figure 1f, according to the graph, it should be 8 clusters, not 9.

There are 9 clusters because the numbering for the clusters start at ‘0’.

(2) Why did cluster 8 have so many different states of cells from both sexes?

The identification of cluster 8 is likely an artifact of sequencing, and would require several different analyses to figure out why cluster 8 has many different states of cells from both sexes. While this will address a technical issue associated with the dataset, this will not change any major conclusions of the study.

(3) Figure 1i, shouldn't that be ten instead of eleven?

There are 11 clusters because the numbering for the clusters start at ‘0’.

(4) Figure 2a, zkscan expression level comparison was not so obvious as the bubble size was small. How many folds of differences from xx pgc?

There is a 1.5 fold increase in the expression of Zkscan5 between XY and XX PGCs at E13.5. We included this information in the revised manuscript.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

This paper introduces a new approach to modeling human behavioral responses using image-computable models. They create a model (VAM) that is a combination of a standard CNN coupled with a standard evidence accumulation model (EAM). The combined model is then trained directly on image-level data using human behavioral responses. This approach is original and can have wide applicability. However, many of the specific findings reported are less compelling.

Strengths:

(1) The manuscript presents an original approach to fitting an image-computable model to human behavioral data. This type of approach is sorely needed in the field.

(2) The analyses are very technically sophisticated.

(3) The behavioral data are large both in terms of sample size (N=75) and in terms of trials per subject.

Weaknesses:

Major

(1) The manuscript appears to suggest that it is the first to combine CNNs with evidence accumulation models (EAMs). However, this was done in a 2022 preprint

(https://www.biorxiv.org/content/10.1101/2022.08.23.505015v1) that introduced a network called RTNet. This preprint is cited here, but never really discussed. Further, the two unique features of the current approach discussed in lines 55-60 are both present to some extent in RTNet. Given the strong conceptual similarity in approach, it seems that a detailed discussion of similarities and differences (of which there are many) should feature in the Introduction.

Thanks for pointing this out—we agree that the novel contributions of our model (the VAM) with respect to prior related models (including RTNet) should be clarified, and have revised the Introduction accordingly. We include the following clarifications in the Introduction:

“The key feature of the VAM that distinguishes it from prior models is that the CNN and EAM parameters are jointly fitted to the RT, choice, and visual stimulus data from individual participants in a unified Bayesian framework. Thus, both the visual representations learned by the CNN and the EAM parameters are directly constrained by behavioral data. In contrast, prior models first optimize the CNN to perform the behavioral task, then separately fit a minimal set of high-level CNN parameters [RTNet, Rafiei et al., 2024] and/or the EAM parameters to behavioral data [Annis et al., 2021; Holmes et al., 2020; Trueblood et al., 2021]. As we will show, fitting the CNN with human data—rather than optimizing the model to perform a task—has significant consequences for the representations learned by the model.”

E.g. in the case of RTNet, the variability of the Bayesian CNN weight distribution, the decision threshold, and the magnitude of the noise added to the images are adjusted to match the average human accuracy (separately for each task condition). RTNet is an interesting and useful model that we believe has complementary strengths to our own work.

Since there are several other existing models in addition to the VAM and RTNet that use CNNs to generate RTs or RT proxies (by our count, at least six that we cite earlier in the Introduction), we felt it was inappropriate to preferentially include a detailed comparison of the VAM and RTNet beyond the passage quoted above.

(2) In the approach here, a given stimulus is always processed in the same way through the core CNN to produce activations v_k. These v_k's are then corrupted by Gaussian noise to produce drift rates d_k, which can differ from trial to trial even for the same stimulus. In other words, the assumption built into VAM appears to be that the drift rate variability stems entirely from post-sensory (decisional) noise. In contrast, the typical interpretation of EAMs is that the variability in drift rates is sensory. This is also the assumption built into RTNet where the core CNN produces noisy evidence. Can the authors comment on the plausibility of VAM's assumption that the noise is post-sensory?

In our view, the VAM is compatible with a model in which the drift rate variability for a given stimulus is due to sensory noise, since we do not specify the origin of the Gaussian noise added to the drift rates. As the reviewer notes, the CNN component of the VAM processes a given stimulus deterministically, yielding the mean drift rates. This does not preclude us from imagining an additional (unmodeled) sensory process that adds variability to the drift rates. The VAM simply represents this and other hypothetical sources of variability as additive Gaussian noise. We agree however that it is worthwhile to think about the origin of the drift rate variability, though it is not a focus of our work.

(3) Figure 2 plots how well VAM explains different behavioral features. It would be very useful if the authors could also fit simple EAMs to the data to clarify which of these features are explainable by EAMs only and which are not.

In our view, fitting simple EAMs to the data would not be especially informative and poses a number of challenges for the particular task we study (LIM) that are neatly avoided by using the VAM. In particular, as we show in Figure 2, the stimuli vary along several dimensions that all appear to influence behavior: horizontal position, vertical position, layout, target direction, and flanker direction. Since the VAM is stimulus-computable, fitting the VAM automatically discovers how all of these stimulus features influence behavior (via their effect on the drift rates outputted by the CNN). In contrast, fitting a simple EAM (e.g. the LBA model) necessitates choosing a particular parameterization that specifies the relationship between all of the stimulus features and the EAM model parameters. This raises a number of practical questions. For example, should we attempt to fit a separate EAM for each stimulus feature, or model all stimulus features simultaneously?

Moreover, while we could in principle navigate these issues and fit simple EAMs to the data, we do not intend to claim that simple EAMs fail to explain the relationship between stimulus features and behavior as well as the VAM. Rather, the key strength of the VAM relative to simple EAMs is that it includes a detailed and biologically plausible model of human vision. The majority of the paper capitalizes on this strength by showing how behavioral effects of interest (namely congruency effects) can be explained in terms of the VAM’s visual representations.

(4) VAM is tested in two different ways behaviorally. First, it is tested to what extent it captures individual differences (Figure 2B-E). Second, it is tested to what extent it captures average subject data (Figure 2F-J). It wasn't clear to me why for some metrics only individual differences are examined and for other metrics only average human data is examined. I think that it will be much more informative if separate figures examine average human data and individual difference data. I think that it's especially important to clarify whether VAM can capture individual differences for the quantities plotted in Figures 2F-J.

We would like to clarify that Fig. 2J in fact already shows how well the VAM captures individual differences for the average subject data shown in Fig. 2H (stimulus layout) and Fig. 2I (stimulus position). For a given participant and stimulus feature, we calculated the Pearson's r between model/participant mean RTs across each stimulus feature value. Fig. 2J shows the distribution of these Pearson’s r values across all participants for stimulus layout and horizontal/vertical position.

Fig. 2G also already shows how well the VAM captures individual differences in behavior. Specifically, this panel shows individual differences in mean RT attributable to differences in age. For Fig. 2F, which shows how the model drift rates differ on congruent vs. incongruent trials, there is no sensible way to compare the models to the participants at any level of analysis (since the participants do not have drift rates). 

(5) The authors look inside VAM and perform many exploratory analyses. I found many of these difficult to follow since there was little guidance about why each analysis was conducted. This also made it difficult to assess the likelihood that any given result is robust and replicable. More importantly, it was unclear which results are hypothesized to depend on the VAM architecture and training, and which results would be expected in performance-optimized CNNs. The authors train and examine performance-optimized CNNs later, but it would be useful to compare those results to the VAM results immediately when each VAM result is first introduced.

Thanks for pointing this out—we apologize for any confusion caused by our presentation of the CNN analyses. We have added in additional motivating statements, methodological clarifications, and relevant references to our Results, particularly for Figure 3 in which we first introduce the analyses of the CNN representations/activity. In general, each analysis is prefaced by a guiding question or specific rationale, e.g. “How do the models' visual representations enable target selectivity for stimuli that vary along several irrelevant dimensions?” We also provide numerous references in which these analysis techniques have been used to address similar questions in CNNs or the primate visual cortex.

We chose to maintain the current organization of our results in which the comparison between the VAM and the task-optimized models are presented in a separate figure. We felt that including analyses of both the VAM and task-optimized models in the initial analyses of the CNN representations would be overwhelming for many readers. As the reviewer acknowledges, some readers may already find these results challenging to follow. 

(6) The authors don't examine how the task-optimized models would produce RTs. They say in lines 371-2 that they "could not examine the RT congruency effect since the task-optimized models do not generate RTs." CNNs alone don't generate RTs, but RTs can easily be generated from them using the same EAM add-on that is part of VAM. Given that the CNNs are already trained, I can't see a reason why the authors can't train EAMs on top of the already trained CNNs and generate RTs, so these can provide a better comparison to VAM.

We appreciate this suggestion, but we judge the suggestion to “train EAMs on top of the already trained CNNs and generate RTs” to be a significant expansion of the scope of the paper with multiple possible roads forward. In particular, one must specify how the outputs of the task-optimized CNN (logits for each possible response) relate to drift rates, and there is no widely-accepted or standard way to do this. Previously proposed methods include transforming representation distances in the last layer to drift rates (https://doi.org/10.1037/xlm0000968), fitting additional subject-specific parameters that map the logits to drift rates

(https://doi.org/10.1007/s42113-019-00042-1), or using the softmax-scored model outputs as drift rates directly (https://doi.org/10.1038/s41562-024-01914-8), though in the latter case the RTs are not on the same scale as human data. In our view, evaluating these different methods is beyond the scope of this paper. An advantage of the VAM is that one does not have to fit two separate models (a CNN and a EAM) to generate RTs.

Nonetheless, we agree that it would be informative to examine something like RTs in the task-optimized models. Our revised Results section now includes an analysis of the confidence of the task-optimized models’ decisions, which we use a proxy for RTs:   

“Since the task-optimized models do not generate RTs, it is not possible to directly measure RT congruency effects in these models without making additional assumptions about how the CNN's classification decisions relate to RTs. However, as a coarse proxy for RT, we can examine the confidence of the CNN's decisions, defined as the softmax-scored logit (probability) of the most probable direction in the final CNN layer. This choice of RT proxy is motivated by some prior studies that have combined CNNs with EAMs [Annis et al., 2021; Holmes et al., 2020; Trueblood et al., 2021]. These studies explicitly or implicitly derive a measure of decision confidence from the activity of the last CNN layer. The confidence measure is then mapped to the EAM drift rates, such that greater decision confidence generally corresponds to higher drift rates (and therefore shorter RTs).

We calculated the average confidence of each task-optimized CNN separately for congruent vs. incongruent trials. On average, the task-optimized models showed higher confidence on congruent vs. incongruent trials (W = 21.0, p < 1e-3, Wilcoxon signed-rank test; Cohen's d = 0.99; n = 75 models). These analyses therefore provide some evidence that task-optimized CNNs have the capacity to exhibit congruency effects, though an explicit comparison of the magnitude of these effects with human data requires additional modeling assumptions (e.g., fitting a separate EAM).”

(7) The Discussion felt very long and mostly a summary of the Results. I also couldn't shake the feeling that it had many just-so stories related to the variety of findings reported. I think that the section should be condensed and the authors should be clearer about which explanations are speculations and which are air-tight arguments based on the data.

We have shortened the Discussion modestly and we have added in some clarifying language to help clarify which arguments are more speculative vs. directly supported by our data.

Specifically, we added in the phrase “we speculate that…” for two suggestions in the Discussion (paragraphs 3 and 5), and we ensured that any other more speculative suggestions contain such clarifying language. We have also added in subheadings in the Discussion to help readers navigate this section. 

(8) In one of the control analyses, the authors train different VAMs on each RT quantile. I don't understand how it can be claimed that this approach can serve as a model of an individual's sensory processing. Which of the 5 sets of weights (5 VAMs) captures a given subject's visual processing? Are the authors saying that the visual system of a given subject changes based on the expected RT for a stimulus? I feel like I'm missing something about how the authors think about these results.

We agree that these particular analyses may cause confusion and have removed them from our revised manuscript.

Reviewer #2 (Public Review):

In an image-computable model of speeded decision-making, the authors introduce and fit a combined CCN-EAM (a 'VAM') to flanker-task-like data. They show that the VAM can fit mean RTs and accuracies as well as the congruency effect that is present in the data, and subsequently analyze the VAM in terms of where in the network congruency effects arise.

Overall, combining DNNs and EAMs appears to be a promising avenue to seriously model the visual system in decision-making tasks compared to the current practice in EAMs. Some variants have been proposed or used before (e.g., doi.org/10.1016/j.neuroimage.2017.12.078 , doi.org/10.1007/s42113-019-00042-1), but always in the context of using task-trained models, rather than models trained on behavioral data. However, I was surprised to read that the authors developed their model in the context of a conflict task, rather than a simpler perceptual decision-making task. Conflict effects in human behavior are particularly complex, and thereby, the authors set a high goal for themselves in terms of the to-be-explained human behavior. Unfortunately, the proposed VAM does not appear to provide a great account of conflict effects that are considered fundamental features of human behavior, like the shape of response time distributions, and specifically, delta plots (doi.org/10.1037/0096-1523.20.4.731). The authors argue that it is beyond the scope of the presented paper to analyze delta plots, but as these are central to studies of human conflict behavior, models that aim to explain conflict behavior will need to be able to fit and explain delta plots.

Theories on conflict often suggest that negative/positive-trending delta plots arise through the relative timing of response activation related to relevant and irrelevant information.

Accumulation for relevant and irrelevant information would, as a result, either start at different points in time or the rates vary over time. The current VAM, as a feedforward neural network model, does not appear to be able to capture such effects, and perhaps fundamentally not so: accumulation for each choice option is forced to start at the same time, and rates are a static output of the CNN.

The proposed solution of fitting five separate VAMs (one for each of five RT quantiles) is not satisfactory: it does not explain how delta plots result from the model, for the same reason that fitting five evidence accumulation models (one per RT quantile) does not explain how response time distributions arise. If, for example, one would want to make a prediction about someone's response time and choice based on a given stimulus, one would first have to decide which of the five VAMs to use, which is circular. But more importantly, this way of fitting multiple models does not explain the latent mechanism that underlies the shape of the delta plots.

As such, the extensive analyses on the VAM layers and the resulting conclusions that conflict effects arise due to changing representations across layers (e.g., "the selection of task-relevant information occurs through the orthogonalization of relevant and irrelevant representations") - while inspiring, they remain hard to weigh, as they are contingent on the assumption that the VAM can capture human behavior in the conflict task, which it struggles with. That said, the promise of combining CNNs and EAMs is clearly there. A way forward could be to either adjust the proposed model so that it can explain delta plots, which would potentially require temporal dynamics and time-varying evidence accumulation rates, or perhaps to start simpler and combine CCNs-EAMs that are able to fit more standard perceptual decision-making tasks without conflict effects.

We thank the reviewer for their thoughtful comments on our work. However, we note that the

VAM does in fact capture the positive-trending RT delta plot observed in the participant data (Fig. S4A), though the intercepts for models/participants differ somewhat. On the other hand, the conditional accuracy functions (Fig. S4B) reveal a more pronounced difference between model and participant behavior. As the reviewer points out, capturing these effects is likely to require a model that can produce time-varying drift rates, whereas our model produces a fixed drift rate for a given stimulus. We also agree that fitting a separate VAM to each RT quantile is not a satisfactory means of addressing this limitation and have removed these analyses from our revised manuscript.

However, while we agree that accurately capturing these dynamic effects is a laudable goal, it is in our view also worthwhile to consider explanations for the mean behavioral effect (i.e. the accuracy congruency effect), which can occur independently of any consideration of dynamics. One of our main findings is that across-model variability in accuracy congruency effects is better attributed to variation in representation geometry (target/flanker subspace alignment) vs.

variation in the degree of flanker suppression. This finding does not require any consideration of dynamics to be valid at the level of explanation we pursue (across-user variability in congruency effects), but also does not preclude additional dynamic processes that could give rise to more specific error patterns. Our revised discussion now includes a section where we summarize and elaborate on these ideas:

“It is not difficult to imagine how the orthogonalization mechanism described above, which explains variability in accuracy congruency effects across individuals, could act in concert with other dynamic processes that explain variability in congruency effects within individuals (e.g., as a function of RT). In general, any process that dynamically gates the influence of irrelevant sensory information on behavioral outputs could accomplish this, for example ramping inhibition of incorrect response activation [https://doi.org/10.3389/fnhum.2010.00222], a shrinking attention spotlight [https://doi.org/10.1016/j.cogpsych.2011.08.001], or dynamics in neural population-level geometry [https://doi.org/10.1038/nn.3643]. To pursue these ideas, future work may aim to incorporate dynamics into the visual component and decision component of the VAM with recurrent CNNs [https://doi.org/10.48550/arXiv.1807.00053, https://doi.org/10.48550/arXiv.2306.11582] and the task-DyVA model [https://doi.org/10.1038/s41562-022-01510-8], respectively.”

Reviewer #3 (Public Review):

Summary:

In this article, the authors combine a well-established choice-response time (RT) model (the Linear Ballistic Accumulator) with a CNN model of visual processing to model image-based decisions (referred to as the Visual Accumulator Model - VAM). While this is not the first effort to combine these modeling frameworks, it uses this combination of approaches uniquely.

Specifically, the authors attempt to better understand the structure of human information representations by fitting this model to behavioral (choice-RT) data from a classic flanker task. This objective is made possible by using a very large (by psychological modeling standards) industry data set to jointly fit both components of this VAM model to individual-level data. Using this approach, they illustrate (among other results) (1) how the interaction between target and flanker representations influence the presence and strength of congruency effects, (2) how the structure of representations changes (distributed versus more localized) with depth in the CNN model component, and (3) how different model training paradigms change the nature of information representations. This work contributes to the ML literature by demonstrating the value of training models with richer behavioral data. It also contributes to cognitive science by demonstrating how ML approaches can be integrated into cognitive modeling. Finally, it contributes to the literature on conflict modeling by illustrating how information representations may lead to some of the classic effects observed in this area of research.

Strengths:

(1) The data set used for this analysis is unique and is made publicly available as part of this article. Specifically, they have access to data for 75 participants with >25,000 trials per participant. This scale of data/individual is unusual and is the foundation on which this research rests.

(2) This is the first time, to my knowledge, that a model combining a CNN with a choice-RT model has been jointly fit to choice-RT data at the level of individual people. This type of model combination has been used before but in a more restricted context. This joint fitting, and in particular, learning a CNN through the choice-RT modeling framework, allows the authors to probe the structure of human information representations learned directly from behavioral data.

(3) The analysis approaches used in this article are state-of-the-art. The training of these models is straightforward given the data available. The interesting part of this article (opinion of course) is the way in which they probe what CNN has learned once trained. I find their analysis of how distractor and target information interfere with each other particularly compelling as well as their demonstration that training on behavioral data changes the structure of information representations when compared to training models on standard task-optimized data.

Weaknesses:

(1) Just as the data in this article is a major strength, it is also a weakness. This type of modeling would be difficult, if not impossible to do with standard laboratory data. I don't know what the data floor would be, but collecting tens of thousands of decisions for a single person is impractical in most contexts. Thus this type of work may live in the realm of industry. I do want to re-iterate that the data for this study was made publicly available though!

We suspect (but have not systematically tested) that the VAMs can be fitted with substantially less data. We use data augmentation techniques (various randomized image transformations) during training to improve the generalization capabilities of the VAMs, and these methods are likely to be particularly important when training on smaller datasets. One could consider increasing the amount of image data augmentation when working with smaller datasets, or pursuing other forms of data augmentation like resampling from estimated RT distributions (see https://doi.org/10.1038/s41562-022-01510-8 for an example of this). In general, we don’t think that prospective users of our approach should be discouraged if they have only a few hundred trials per subject (or less) - it’s worth trying!

(2) While this article uses choice-RT data it doesn't fully leverage the richness of the RT data itself. As the authors point out, this modeling framework, the LBA component in particular, does not account for some of the more nuanced but well-established RT effects in this data. This is not a big concern given the already nice contributions of this article and it leads to an opportunity for ongoing investigation.

We agree that fully capturing the more nuanced behavioral effects you mention (e.g. RT delta plots and conditional accuracy functions) is a worthwhile goal for future research—see our response to Reviewer #2 for a more detailed discussion. ----------

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

(1) The phrase in the Abstract "convolutional neural network models of visual processing and traditional EAMs are jointly fitted" made me initially believe that the two models were fitted independently. You may want to re-word to clarify.

We think that the phrase “jointly fitted” already makes it clear that both the CNN and EAM parameters are estimated simultaneously, in agreement with how this term is usually used. But we have nonetheless appended some additional clarifying language to that sentence (“in a unified Bayesian framework”).

(2) Lines 27-28: EAMs "are the most successful and widely-used computational models of decision-making." This is only true for the specific type of decision-making examined here, namely joint modeling of choice and response times. Signal detection theory is arguably more widely-used when response times are not modeled.

Thanks for pointing this out - we have revised the referenced sentence accordingly.

(3) Could the authors clarify what is plotted in Figure 2F?

Fig. 2F shows the drift rates for the target, flanker, and “other” (non-target/non-flanker) accumulators averaged over trials and models for congruent vs. incongruent trials. In case this was a source of confusion, we do not show the value of the flanker drift rates on congruent trials because the flanker and target accumulators are identical (i.e. the flanker/congruent drift rates are equivalent to the target/congruent drift rates).

(4) Lines 214-7: "The observation that single-unit information for target direction decreased between the fourth and final convolutional layers while population-level decoding remained high is especially noteworthy in that it implies a transition from representing target direction with specialized "target neurons" to a more distributed, ensemble-level code." Can the authors clarify why this is the only reasonable explanation for these results? It seems like many other explanations could be construed.

We have added additional clarification to this section and now use more tentative language:

“The observation that single-unit information for target direction decreased between the fourth and final convolutional layers indicates that the units become progressively less selective for particular target directions. Since population-level decoding remained high in these layers, this suggests a transition from representing target direction with specialized "target neurons" to a more distributed, ensemble-level code.”

(5) Lines 372-376: "Thus, simply training the model to perform the task is not sufficient to reproduce a behavioral phenomenon widely-observed in conflict tasks. This challenges a core (but often implicit) assumption of the task-optimized training paradigm, namely that to do a task well, a training model will result in model representations that are similar to those employed by humans." While I agree with the general sentiment, I feel that its application here is strange. Unless I'm missing something, in the context of the preceding sentence, the authors seem to be saying that researchers in the field expect that CNNs can produce a behavioral phenomenon (RTs) that is completely outside of their design and training. I don't think that anyone actually expects that.

We moved the discussion/analyses of RTs to the next paragraph. It should now be clear that this statement refers specifically to the absence of an accuracy congruency effect in the task-optimized models.

(6) Lines 387-389: "As a result, the VAMs may learn richer representations of the stimuli, since a variety of stimulus features-layout, stimulus position, flanker direction-influence behavior (Figure 2)." That is certainly true of tasks like this one where an optimal model would only focus on a tiny part of the image, whereas humans are distracted by many features. I'm not sure that this distractibility is the same as "richer representations". When CNNs classify images based on the background, would the authors claim that they have richer representations than humans?

We agree that “richer” may not be the best way to characterize these representations, and have changed it to “more complex”.

(7) Is it possible that drift rate d_k for each response happens to be negative on a given trial? If so, how is the decision given on such trials (since presumably none of the accumulators will ever reach the boundary)?

It is indeed possible for all of the drift rates to be negative, though we found that this occurred for a vanishingly small number of trials (mean ± s.e.m. percent trials/model: 0.080 ± 0.011%, n = 75 models), as reported in the Methods. These trials were excluded from analyses.

(8)  Can the authors comment on how they chose the CNN architecture and whether they expect that different architectures will produce similar results?

Before establishing the seven-layer CNN architecture used throughout the paper, we conducted some preliminary experiments using other architectures that differed primarily in the number of CNN layers. We found that models with significantly fewer than seven layers typically failed to reach human-level accuracy on the task while larger models achieved human-level accuracy but (unsurprisingly) took longer to train.

Reviewer #3 (Recommendations For The Authors):

- In the introduction to this paper (particularly the paragraph beginning in line 33), the authors note that EAMs have typically been used in simplified settings and that they do not provide a means to account for how people extract information from naturalistic stimuli. While I agree with this, the idea of connecting CNNs of visual processing with EAMs for a joint modeling framework has been done. I recommend looking at and referencing these two articles as well as adjusting the tenor of this part of an introduction to better reflect the current state of the literature. For full disclosure, I am one of the authors on these articles. https://link.springer.com/article/10.1007/s42113-019-00042-1 https://www.sciencedirect.com/science/article/abs/pii/S0010027721001323

We agree—thanks for pointing this out. The revised Introduction now discusses prior related models in more detail (including those referenced above) and better clarifies the novel contributions of our model. We specifically highlight that a novel contribution of the VAM is that “the CNN and EAM parameters are jointly fitted to the RT, choice, and visual stimulus data from individual participants in a unified Bayesian framework.”

- The statement in lines 56-58 implies that this is the first article to glue CNNs together with EAMs. I would edit this accordingly based on the prior comment here and references provided. I will note that the second feature of the approach in this paper is still novel and really nice, namely the fact that the CNN and the EAM are jointly fitted. In the aforementioned references, the CNN is trained on the image set, and individual level Bayesian estimation was only applied to the EAM. Thus, it may be useful to highlight the joint estimation aspect of this investigation as well as how the uniqueness of the data available makes it possible.

Agreed—see above.

- Figure 3c and associated text. I understand the MI analysis you are performing here, however it is difficult to interpret as it stands. In the figure, what does a MI of 0.1 mean?? Can you give some context to that scale? I do find the interpretation of the hunchback shape in lines 210-222 to be somewhat of a stretch. The discussion that precedes (lines 199-209) this is clear and convincing. Can this discussion be strengthened more? And more interpretability of Figure 3c would be helpful; entropic scales can be hard to interpret without some context or scale associated.

The MI analyses in Fig. 3C (and also Figs. 4C and 6E) show normalized MI, in which the raw MI has been divided by the entropy of the stimulus feature distribution. This normalization facilitates comparing the MI for different stimulus features, which is relevant for Figs. 4C and 6E. The normalized MI has a possible range of [0, 1], where 1 indicates perfect correlation between the two variables and 0 indicates complete independence. We now note in the legend of these figures that the possible normalized MI range is [0, 1], which should help with interpreting these values. Our revised results section for Fig. 3C now also includes some additional remarks on our interpretation of the hunchback shape of the MI.

- Lines 244-248 and the analyses in Figure 3 suggest a change in the behavior of the CNN around layer 4. This is just a musing, but what would happen if you just used a 4 layer CNN, or even a 3 layer? This is not just a methods question. Your analysis suggests a transition from localized to distributed information representation. Right now, the EAM only sees the output of the distributed representation. What if it saw the results the more local representations from early layers? Of course, a shallower network may just form the distributed representations earlier, but it would interesting if there were a way to tease out not just the presence of distributed vs local representations, but the utility of those to the EAM.

Thanks for this interesting suggestion. We did do some preliminary experiments in models with fewer layers, though we only examined the outputs of these models and did not assess their representations. We found that models with 3–5 layers generally failed to achieve human-level accuracy on the task. In principle, one could relate this observation to the representations of these models as a means of assessing the relative utility of distributed/local representations. However, there are confounding factors that one would ideally control for in order to compare models with different numbers of layers in this fashion (namely, the number of parameters).

- Section Line 359 (Task optimized models) - It would be helpful to clarify here what these task-optimized models are being trained to do. As I understand it, they are being trained to directly predict the target direction. But are you asking them to learn to predict the true target direction? Or are you training them to predict what each individual responds? I think it is the second (since you have 75 of these), but it's not clear. I looked at the methods and still couldn't get a clear description of this. Also, are you just stripping the LBA off of the end of the CNN and then essentially putting a softmax in its place? If so, it would be helpful to say so.

The task-optimized models were actually trained to output the true target direction in each stimulus, rather than trained to match the decisions of the human participants. We trained 75 such models since we wanted to use exactly the same stimuli as were used to train each VAM. The task-optimized CNNs were identical to those used in the VAMs, except that the outputs of the last layer were converted to softmax-scored probabilities for each direction rather than drift rates. The Results and Methods section now included additional commentary that clarifies these points.

- Line 373-376: This statement is pretty well established at this point in the similarity judgement literature. I recommend looking at and referencing https://onlinelibrary.wiley.com/doi/full/10.1111/cogs.13226 https://www.nature.com/articles/s41562-020-00951-3 https://link.springer.com/article/10.1007/s42113-020-00073-z

Thanks for pointing this out. For reference, the statement in question is “Thus, simply training the model to perform the task is not sufficient to reproduce a behavioral phenomenon widely-observed in conflict tasks. This challenges a core (but often implicit) assumption of the task-optimized training paradigm, namely that training a model to do a task well will result in model representations that are similar to those employed by humans.”

We agree that the first and third reference you mention are relevant, and we now cite them along with some other relevant work. In our view, the second reference you mention is not particularly relevant (that paper introduces a new computational model for similarity judgements that is fit to human data, but does not comment on training models to perform tasks vs. fitting to human data).

- Line 387-388: "VAMs may learn richer representations". This is a bit of a philosophical point, but I'll go ahead and mention it. The standard VAM does not necessarily learn "richer" feature representations. Rather, you are asking the VAM and task-optimized models to do different things. As a result, they learn different representations. "Better" or "richer" is in the eye of the beholder. In one view, you could view the VAM performance as sub-par since it exhibits strange artifacts (congruency effects) and the expansion of dimensionality in the VAM representations is merely a side-effect of poor performance. I'm not advocating this view, just playing devils advocate and suggesting a more nuanced discussion of the difference between the VAM and task-optimized models.

We agree—this is a great point. We have changed this statement to read “the VAMs may learn more complex [rather than richer] representations of the stimuli”.

- Lines 567-570: Here you discuss how the LBA backend of the VAM can't account for shrinking spotlight-like RT effects but that fitting models to different RT quantiles helps overcome this. I find this to be one of the weakest points of the paper (the whole process of fitting RT quantiles separately to begin with). This is just a limitation of the RT component of the model. This is a great paper but this is just a limitation inherent in the model. I don't see a need to qualify this limitation and think it would be better to just point out that this is a limitation of the LBA itself (be more clear that it is the LBA that is the limiting factor here) and that this leaves room for future research. From your last sentence of this paragraph, I agree that recurrent CNNs would be interesting. I will note that RNN choice-RT models are out there (though not with CNNs as part of the model).

We agree and have revised this section of the Discussion accordingly (see our response to Reviewer #2 for more detail). We also removed the analyses of models trained on separate RT quantiles.

Absolutely. We're so skeptical (cynical?) about new technology now. There's a lot less "let's explore this and see if it's fun" and a lot more "watch, I bet it's just hype/it's not that useful/etc."

I can understand it to an extent, because the hype train can be so annoying. But maybe that's part of the problem itself. Why are we focused on what others are saying or think about the technology rather than just trying to find out for ourselves?

I would love to see recaps make a comeback and become as successful, or even more successful, than they were before. That being said, I still feel there are many factors preventing this from happening. For example, discussions on platforms like TikTok could hinder their return. However, if recaps do come back, I don’t think they need to rely on platforms like Twitter. As this generation moves away from apps like Twitter, I believe discussions on Reddit, forums, or even casual conversations about books might overshadow recaps and potentially make them obsolete. That said, I do think it’s possible for recaps to return, though perhaps not with the current generation. And if they do gain traction with this generation, it likely won’t happen for several years.

Really though I only have three or four shows that I have to wait weekly for and most. of the shows that I know I have to wait for. I'll just wait until Netflix drops the entire show and then I'll binge it. I think if anything, this has made me realize how often I do binge shows when they drop.

You don't really see this anymore because, with Netflix and similar platforms, people are always watching shows. There will always be someone to talk to about a show, even if they're not the original audience who watched it when it first came out.

I completely disagree with Fitzgerald on this. TikTok and Instagram are filled with recappers and normal people who like to talk about a certain show. For example, outer banks when it first came out got spoiled for a lot of people by these recappers. I think it’s different in today’s time but not completely gone from our society. An app like X (formerly twitter) can let you search specific things and show the latest updates so if a show/movie comes out it gives people a place to see others thoughts and views on what you just watched.

This made me realize that I have honestly never waited for episodes of a show to drop on a weekly basis. I always just wait for it to arrive on Netflix or another streaming service. Wow.

I am hoping for a Masters in social work. I am going to become a therapist for children in non-nuclear families.

This class will help me grow my credits so i am one step closer to a degree.

It's actually #1 now in App. Let's lead with this fact, then summarize the rest. Also interesting that THIS isn't getting banned, even though it has the potential to syphon up WAY more information about people's personal lives or whatever they put into AI.

Summary of the Talk on the Future of CMS, No-Code, Low-Code, and AI-Generated Applications

Evolution of CMS and No-Code Tools

Traditional CMS:

"Back in the day it was like WordPress... the original web where we would write code and then we would just push it up to servers."

CMS emerged to allow non-developers to contribute to web content without coding.

No-Code Tools:

"No code is like your drag and drop GUI... Webflow or whatever."

Introduced drag-and-drop interfaces for broader accessibility, with pros and cons in usability.

AI in No-Code:

"Fast forward even further we've got AI coding right... now a person can just make an app."

AI models like Claude 3.5 enable app generation with minimal developer intervention.

Current No-Code/Low-Code AI Tools Landscape

Key Tools in the Market:

"Let's create the definitive list here... Cursor, Bolt, Lovable."

Cursor is for developers; Bolt and Lovable cater to non-developers with different strengths.

Strengths and Weaknesses:

"Bolt is a great boilerplate generator... Lovable is great if you want ShadCN styling."

Developers prefer Bolt for flexibility; Lovable is preferred for pre-styled design systems.

Challenges with AI-Generated Code

Integration Issues:

"It's not your existing code base... you need to use your components, design system, and backend logic."

AI-generated code often exists in isolation, making integration difficult for enterprise use.

Code Quality Concerns:

"Engineers are not going to want a pull request by a non-engineer."

Quality control and maintainability remain significant barriers.

Customization and Precision:

"Webflow is hard to use... but it gives you 100% precision control."

While AI provides convenience, fine-grained control is still preferred by professionals.

Future of AI-Driven Development

Combining AI with Structured CMS-like Workflows:

"Ideally, we have something like a headless CMS where we can make updates over API."

Future solutions should enable AI updates via APIs while maintaining design consistency.

Ideal Workflow Vision:

"In an ideal world, we can be editing with prompts and visually."

The goal is a hybrid model with AI-driven automation and manual precision controls.

AI-Based Iteration and Optimization:

"AI should listen to your customers... iterate really fast."

Faster feedback loops and continuous optimization through AI experimentation.

Technical Approaches to Solving Challenges

Meta's React Blocks:

"What React blocks let developers do is a backend dev in Python can code up a React UI."

An approach that allows dynamic UI changes without shipping new native app versions.

Mitosis Framework:

"Mitosis is a project that explores transpilation and visual manipulation."

Enables converting JSX into structured JSON for flexible rendering and AI-based updates.

Code-Driven Visual Editing:

"SwiftUI allows updating code with visual previews and vice versa."

Bidirectional code editing is a possible future solution but is still complex.

Current Limitations and Considerations

Performance and Feasibility Issues:

"When I had Google bots crawling my AI-generated site, I got a $4,000/day Anthropics bill."

Generating content in real-time is currently too expensive at scale.

Security and Compliance Risks:

"Dynamic code delivery is ripe with security challenges."

Any AI-driven solutions must consider performance, security, and governance.

Key Use Cases and Applications

Prototyping vs. Production:

"Phenomenal prototyping tools, but moving to production is challenging."

AI tools excel in concept validation but require extensive refinement for production.

Personalization Opportunities:

"The AI could automatically scale things up or down based on performance."

Future possibilities include hyper-personalized user experiences.

Conclusion and Outlook

Near-Term Expectations:

"Webflow and Framer will likely add more AI features over time."

Existing players are expected to incorporate AI capabilities gradually.

Long-Term Potential:

"AI tools will eventually iterate and personalize dynamically based on user input."

The convergence of AI, CMS, and design systems may redefine how software is built.

This summary captures the essence of the speaker's discussion, highlighting key concepts, industry trends, challenges, and possible future developments in the AI-powered CMS and no-code/low-code space.

As someone with a Hispanic background, I find the description of Ādhān al-Qāḍī, or "The Judge’s Ears," to be both evocative to the average reader and relatable to myself and others who are of a similar cultural background. In addition, this recipe serves as a history lesson in showing us the impact that the invasions of the Moors of Spain for centuries had on the nation’s culinary practices. The appreciation for a pastry that is "scrumptiously luscious and brittle to the bite" speaks to a universal love for food that delights the senses. In Hispanic culture, we have our own array of pastries that carry rich flavors and textures, such as empanadas and arepas, which also offer that satisfying crunch and warmth. The imagery of the pastry being "fit for a judge" resonates with me. It reflects the importance of quality and craft in food, much like how our traditional dishes often carry the weight of cultural significance. In Hispanic communities, cooking is not just about sustenance; it’s an art form passed down through generations, much like the care taken in preparing a delicate pastry like Ādhān al-Qāḍī. The recipe’s mention of the pastry maintaining its crispness even after freezing and reheating reminds me of how many of our beloved recipes also stand the test of time. For instance, many Hispanic desserts, like flan or tres leches cake, can be enjoyed days later without losing their charm. This speaks to the craftsmanship involved in creating foods that are meant to be savored repeatedly. Moreover, the notion of a pastry being "deliciously addictive" is something I can wholeheartedly agree with. In Hispanic culture, gatherings often revolve around food, and the irresistible nature of treats encourages sharing and connection among family and friends. It’s fascinating how something as simple as a pastry can evoke such depth and history, bridging cultural divides and celebrating the joy that food brings to our lives. Your description inspires me to seek out this pastry and explore its flavors, reminding me of the beauty found in diverse culinary traditions.

In this excerpt, Li not only brings light to a misconception of her country that she is playing a part in combatting through real life examples of the history of Chinese food, but from just these couple sentences it is inferred that misconceptions about foreign cultures are widespread and include aspects that are often not thought of in regard to the topic, such as food. The phrase “I try my best to use original recipes and raw materials” shows her passion for preserving the culinary heritage of China, which is vital for understanding the cultural and historical context of the food. By sourcing original recipes, Li not only honors traditional methods of cooking but also provides a connection to the past and how it’s been carried to the present, which allows us as readers to appreciate the complexity and richness of ancient Chinese gastronomy. The mention of “raw materials” suggests an emphasis on natural and unprocessed ingredients, which aligns with contemporary trends in cooking that prioritize sustainability and minimalism. This approach can serve as a counterpoint to modern culinary practices that often rely on convenience and industrial production, thereby inviting audiences to reflect on the value of traditional food systems that have sustained communities for centuries. The second part of the excerpt, “The purpose of my video series is to show audiences what people actually ate in ancient China,” represents an educational goal. By aiming to reveal historical eating habits, Li provides insights into the daily lives and societal norms of ancient Chinese people. This not only enriches viewers’ understanding of the culture but also carries us to a deeper appreciation for the evolution of Chinese cuisine. Overall, Li's initiative can be seen as part of a broader movement to revive and celebrate traditional culinary practices. This effort not only preserves valuable historical knowledge but also engages a new generation of food enthusiasts who seek to connect with their heritage through authentic culinary experiences. By bridging the past and present, Li’s work contributes to the ongoing dialogue about food, culture, and identity in a rapidly changing world.

It's amazing how digital platforms can bring together different cultures! By watching these food vloggers, we're not just learning recipes but we're also gaining insight into cultures that bring ancient traditions into our kitchens. Digital existence allows us to experience a slice of history through something as universal as cooking. I think it's quite fascinating how the internet can play such a crucial learning role in people's lives.

For this question, I think the authenticity in the media is diverse. For example, people often share a good life to express or achieve some purpose. But the reality is often not so good. It's just the way people protect themselves in the media. Or the authenticity of some brands. Some brand accounts are displayed on social media in a perfect way, but the truth is that some brands are not as well packaged as they are.

This effort demonstrates the economic ambitions driving colonization, beyond just settling new lands. It's a glimpse into the practical side of building a colony - trying to find ways to make it self-sustaining and profitable.

It’s interesting to think about how trust was such a fundamental part of survival in nomadic societies, and how that need for trust hasn’t gone away—it’s just evolved into more complex systems in modern life. I never really thought about it that way before, especially with things like buying food. Even though we don’t always feel like we need to trust the person behind the counter, so many systems rely on trust—like laws, regulations, and the people making sure everything runs smoothly. It makes me realize how much we depend on trust in ways we don’t always see.

Welcome back.

This is part two of this lesson.

We're going to continue immediately from the end of part one.

So let's get started.

Now let's look at another example.

This looks more complex, but we're going to use the same process to identify the correct answer.

So this is a multi-select question and we're informed that we need to pick two answers, but we're still going to follow the same process.

The first step is to check if we can eliminate any of the answers immediately.

Do any of the answers not make sense without reading the question?

Well, nothing immediately jumps out as wrong, but answer E does look strange to me.

It feels like it's not a viable solution.

I can see the word encryption mentioned and it's rare that I see lambda and encryption mentioned in the same statement.

So at this stage, let's just say that answer E is in doubt.

So it's the least preferred answer at this point.

So keep that in your mind.

It's fine to have answers which you think are not valid.

We don't know enough to immediately exclude it, but we can definitely say that we think there's something wrong with it.

Given that we need to select two answers out of the five, we don't need to worry about E as long as there are two potentially correct answers.

So let's move on.

Now, the real step one is to identify what matters in the question text.

So let's look at that.

Now, the question is actually pretty simple.

It gives you two requirements.

The first is that all data in the cloud needs to be encrypted at rest.

And the second is that any encryption keys are stored on premises.

For any answers to be correct, they need to meet both of these requirements.

So let's follow a similar process on the answer text first looking for any word fluff and then looking for keywords which can help identify either the correct answers or more answers that we can exclude.

So the first three answers, they all state server side encryption, but the remaining two answers don't.

And so the first thing that I'm going to try to do with this question is to analyze whether server side encryption means anything.

Does it exclude the answers or does it point to those answers being correct?

Well, server side encryption means that S3 performs the encryption and decryption operations.

But depending on the type of server side encryption, it means that S3 either handles the keys or the customer handles the keys.

But at this stage, using server side encryption doesn't mean that the answers are right or wrong.

You can use it or you can't use it.

That doesn't immediately point to correct versus incorrect.

What we need to do is to look at the important keywords.

Now, if we assume that we are excluding answer E for now unless we need it, then we have four different possible answers, each of which is using a different type of encryption.

So I've highlighted these.

So we've got S3 managed keys, SSE-S3, KMS managed keys, which is SSE-KMS, customer provided keys, which is SSE-C, and then using client side encryption.

Now, the first requirement of the question states encryption at rest and all of the answers A, B, C and D, they all provide encryption at rest.

But it also states that encryption keys are to be stored on premises.

Answers A and B use server side encryption where AWS handle the encryption process and the encryption keys.

So SSE-S3 and SSE-KMS both mean that AWS are handling the encryption keys.

And because of this, the keys are not stored on premises and so they don't meet the second criteria in the question.

And this means that they're both invalid and can be excluded.

Now, this leaves answers C, D and E, and we already know that we're assuming that we're ignoring answer E for now and only using it if we have to.

So we just have to evaluate if C and D are valid.

And if they are, then those are the answers that we select.

So SSE-C means that the encryption is performed by S3 with the keys that the customer provides.

So that works.

It's a valid answer.

So that means at the very least C is correct.

It can be used based on the criteria presented by the question.

Now, answer D suggests client side encryption, which means encrypting the data on the client side and just passing the encrypted data to S3.

So that also works.

So answers C and D are both potentially correct answers.

So both answers C and D do meet the requirements of the question.

And because of this, we don't have to evaluate answer E at all.

It's always been a questionable answer.

And since the question only requires us to specify two correct answers, we can go ahead and exclude E.

And that gives us the correct answers of C and D.

So that's another question answered.

And it's followed the same process that we used on the previous example.

So really answering questions within AWS is simply following the same process.

Try and eliminate any crazy answers.

So any answers that you can eliminate based just on the text of those answers, then exclude them right away because it reduces the cognitive overhead of having to pick between potentially four correct answers.

If you can eliminate the answers down to three or two, you significantly reduce the complexity of the question.

The next step is to find what really matters in the question, find the keywords in both the preamble and the question.

Then highlight and remove any question fluff.

So anything in the question which doesn't matter, eliminate any of the words which aren't relevant technically to the product or products that you select.

So this is something that comes with experience, being able to highlight what matters and what doesn't matter in questions.

And the more practice that you do, the easier this becomes.

Next, identify what really matters in the answers.

So again, this comes down to identifying any shared common words and removing those and then identifying any of the keywords that occur in the answers.

And then once you've got the keywords in the answers and the keywords in the questions, then you can eliminate any bad answers that occur in the question.

Now, ideally at this point, what remains are correct answers.

You might start off with four or five answers.

You might eliminate two or three.

The question asks for two correct answers.

And that's it.

You've finished the question.

But if you have more answers than you need to provide, then you need to quickly select between what remains and you can do that by doing this keyword matching.

So look for things which stand out.

Look for things which aren't best practice according to AWS.

Look for things which breach a timescale requirement in the question.

Look for things which can't perform at the levels that the question requires or that cost too much based on the criteria and the question.

Essentially, what you're doing is looking for that one thing that will let you eliminate any other answers and leave you with the answers that the question requires.

Generally, when I'm answering most questions, it's a mixture between the correct answer jumping out at me and eliminating incorrect answers until I'm left with the correct answers.

You can approach questions in two different ways.

Either looking for the correct answers or eliminating the incorrect ones.

Do whichever works the best for you and follow the same process throughout every question in the exam.

The one big piece of advice that I can give is don't panic.

Everybody thinks they're running out of time.

Most people do run out of time.

So follow the exam technique process that I detailed in the previous lesson to try and get you additional time, leave the really difficult questions until the end, and then just follow this logical process step by step through the exam.

Keep an eye on the remaining amount of time that you have at every point through the exam and I know that you will do well.

Most people fail the exam because of their exam technique, not their lack of technical capability.

With that being said, though, that's everything I wanted to cover in this set of lessons.

Good luck with the practice tests.

Good luck with the final exam.

And if you do follow this process, I know that you'll do really well.

With that being said, though, go ahead and complete this video and then when you're ready, I'll look forward to joining you in the next.

Welcome back and from the very start this course has been about more than just the technical side.

So this is part one of a two-part lesson set and in this lesson I'll focus on some exam technique hints and tips that you might find useful in the exam.

Now in terms of the exam itself it's going to have questions of varying levels of difficulty and this is also based on your own strengths and weaknesses.

Conceptually though understand that on average the AWS exams will generally feel like they have 25% easy questions, 50% medium questions and 25% really difficult questions.

Assuming that you've prepared well and have no major skill gaps this is the norm.

For most people this is how it feels.

The problem is the order of the difficulty is going to feel random so you could have all of your easy ones at the start or at the end or scattered between all of the other questions and this is part of the technique of the AWS exams how to handle question difficulty in the most efficient way possible.

Now I recommend conceptually that my students think of exams in three phases.

You want to spend most of your time on phase two.

So structurally in phase one I normally try to go through all of the 65 questions and identify ones that I can immediately answer.

You can use the exam tools including mark for review and just step through all of the questions on the exam answering anything that's immediately obvious.

If you can answer a question within 10 seconds or have a good idea of what the answer will be and just need to consider it for a couple more seconds this is what I term a phase one question.

Now the reason that I do these phase one questions first is that they're easy they take very little time and because you know the subject so well you have a very low chance of making a mistake.

So once you've finished all of these easy questions the phase one questions what you're left with is the medium or yellow questions and the hard or red questions.

My aim is that I want to leave the hard questions until the very end of the test.

They're going to be tough to answer anyway and so what I want to do at this stage in phase two is to go through whatever questions remain so whatever isn't easy and I'm looking to identify any red questions and mark them for review and then just skip past them.

I don't want to worry about any red questions in phase two.

What phase two is about is powering through the medium questions.

These will require some thought but they don't scare you they're not impossible.

The medium questions so the yellow questions should make up the bulk of the time inside the exam.

They should be your focus because these are the questions which will allow you to pass or fail the exam.

For most people the medium questions represent the bulk of the exam questions.

Generally your perception will be that most of the questions will be medium.

There'll be some easy and some hard so you need to focus in phase two which represents the bulk of the exam on just these medium questions.

So my suggestion generally is in phase two you've marked the hard questions for review and just skipped past them and then you focused on the medium questions.

Now after you've completed these medium questions you need to look at your remaining time and it might be that you have 40 minutes left or you might only have four minutes or even less.

In the remaining time that you have left you should be focusing on the remaining red questions the difficult questions.

If you have 40 minutes left then you can take your time.

If you have four minutes you might have to guess or even just click answers at random.

Now both of these approaches are fine because at this point you've covered the majority of the questions.

You've answered all of the easy questions and you've completed all of the medium questions.

What remains are questions that you might get wrong regardless but because you've pushed them all the way through to the end of your time allocation whether you're considering them carefully and answering them because you have 40 minutes left or whether you're just answering them at random they won't impact your process in answering the earlier questions.

So if you don't follow this approach what tends to happen is you're focusing really heavily on the hard questions at the start of the exam and that means that you run out of time towards the end but if you follow this three-stage process by this point all that you have left is a certain number of minutes and a certain set of really difficult questions and you can take your time safe in the knowledge that you've already hopefully passed to the exam based on the easy and medium questions and the hard ones as simply a bonus.

Now at a high level this process is designed to get you to answer all of the questions that you're capable of answering as quickly as possible and leave anything that causes you to doubt yourself or that you struggle with to the end.

So pick off the easy questions, focus on the medium and then finish up with the really hard questions at the end.

I know that it sounds simple but unless you focus really hard on this process or one like it then your actual exam experience could be fairly chaotic.

If you're unlucky enough to get hard questions at the start and you don't use a process like this it can really spoil your flow.

So before we finish this lesson just some final hints and tips that I've got based on my own experiences.

First if this is your first exam assume that you're going to run out of time.

Most people enter the exam not having an understanding of the structure and most people myself included with my first exam will run out of time.

The way that you don't run out of time and the way that you succeed is to be efficient, have a process.

Now assuming that you have the default amount of time you need to be aware that you have two minutes to read the question, read the answers and to make a decision.

So this sounds like a lot but it's not a lot of time to do all of those individual components.

You shouldn't be guessing on any answers until the end.

If you're guessing on a question then it should be in the hard question category and you should be tackling this at the end.

I don't want you guessing on any easy questions or any medium questions.

If you're guessing then you shouldn't be looking at it until right at the very end.

Another way of looking at this is if you are unsure about a question or you're forced to guess early on you need to be aware that a question that's later on so further on in the exam might prompt you to remember the correct answer for an earlier question.

So if you do have to guess on any questions then use the mark for review feature.

You can mark any question that you want for review as you go through the course and then at any point or right at the end you can see all the questions which are flagged for review and revisit them.

So use that feature it can be used if you're doubtful on any of the answers or you want to prompt yourself as with the hard questions to revisit them toward the end of the exam.

Now this should be logical but take all the practice tests that you can.

One of my favorite test vendors in the space is the team over at TutorialsDojo.com.

They offer a full range of practice questions for all of the major AWS exams so definitely give their site a look.

One of the benefits of the exam questions created over at TutorialsDojo is that they are more difficult than the real exam questions so they can prepare you for a much higher level of difficulty and by the time you get into the exam you should find it relatively okay.

So my usual method is to suggest that people take a course and then once they've finished the course take the practice test in the course, follow that up with the tutorials Dojo practice tests and for any questions they get wrong it can identify areas that they need additional study.

So rinse and repeat that process, perform that additional study, redo the practice tests and when you're regularly scoring above 90% on those practice tests then you're ready to do the real exam.

And at this point there are all of my suggestions for exam technique.

In the next lesson I want to focus on questions themselves because it's the questions and your efficiency during the process of answering questions which can mean the difference between success and failure.

So go ahead complete this video and in the next video when you're ready we'll look at some techniques on how you can really excel when tackling exam questions.

Welcome back and in this lesson I want to talk about a few related features of CloudFormation and those are weight conditions, creation policies and the CFN signal tool.

So let's jump in and get started straight away.

Before we look at all of those features as a refresher I want to step through what actually happens with the traditional CloudFormation provisioning process and let's assume that we're building an EC2 instance and we're using some user data to bootstrap WordPress.

Well if we do this the process starts with logical resources within the template and the template is used to create a Cloud Formation stack.

Now you know by now that it's the job of the stack to take the logical resources in a template and then create, update or delete physical resources to match them within an AWS account.

So in this case it creates an EC2 instance within an AWS account.

From CloudFormation's perspective in this example it initiates the creation of an EC2 instance so when EC2 reports back that the physical resource has completed provisioning the logical resource changes to create complete and that means everything's good right?

Well the truth is we just don't know.

With simple provisioning when the relevant system EC2 in this case tells CloudFormation that it's finished then CloudFormation has no further access to any other information beyond the fact that EC2 is telling it that that resource has completed its provisioning process.

With more complex resource provisions like this one where bootstrapping goes on beyond when the instance itself is ready then the completion state isn't really available until after the bootstrapping finishes and even then there's no built-in link to communicate back to CloudFormation whether that bootstrapping process was successful or whether it failed.

An EC2 instance will be in a create complete state long before the bootstrapping finishes and so even when it's finished if it fails the resource itself still shows create complete.

Creation policies, weight conditions and CFN signal provide a few ways that we can get around this default limitation and allow systems to provide more detailed signals on completion or not to CloudFormation.

So let's have a look at how this works.

The way that this enhanced signaling is done is via the CFN signal command which is included in the AWS CFN bootstrap package.

The principle is simple enough you configure CloudFormation to hold or pause a resource and I'll talk more about the ways that this is done next but you configure CloudFormation to wait for a certain number of success signals.

You want to make it so that resources such as EC2 instances tell CloudFormation that they're okay.

So in addition to configuring it to wait for a certain number of success signals you also configure a timeout.

This is a value in hours, minutes and seconds within which those signals can be received.

Now the maximum permitted value for this is 12 hours and once configured it means that a logical resource such as an EC2 instance will just wait.

It won't automatically move into a create complete state once the EC2 system says that it's ready.

Instead if the number of success signals that you define is received by CloudFormation within the timeout period then the status of that resource changes into create complete and the stack process continues with the knowledge that the EC2 instance really is finished and ready to go because on the instance you've configured something to explicitly send that signal or signals to CloudFormation.

CFN signal is a utility running on the instance itself actually sending a signal back to the CloudFormation service.

Now if CFN signal communicates a failure signal suggesting that the bootstrapping process didn't complete successfully then the creation of the resource in the stack fails and the stack itself fails.

So that's important to understand CFN signal can send success signals or failure signals and a failure signal explicitly fails the process.

Now another possible outcome of this is the timeout period can be reached without the required number of success signals and in this situation CloudFormation views this as an implicit failure.

The resource being created fails and then logically the stack fails the entire process that it's doing.

Now the actual thing which is being signaled using CFN signal is a logical resource specifically a resource such as EC2 or auto scaling groups which is using a creation policy or a specific type of separate resource called a weight condition resource.

Now AWS suggests that for provisioning EC2 and auto scaling groups you should use a creation policy because it's tied to that specific resource that you're handling but you might have other requirements to signal outside of a specific resource.

For example if you're integrating CloudFormation with an external IT system of some kind in that case you might choose to use a weight condition and next I want to visually step through how both of these work because it will make a lot more sense when you see the architecture visually.

Let's start with the example of an auto scaling group which uses a launch configuration to launch three EC2 instances.

These are within a template and that's used to create a stack.

Because I'm using a creation policy here a few things happen which are different to how CloudFormation normally functions.

First the creation policy here adds a signal requirement and timeout to the stack.

In this case the stack needs three signals and it has a timeout of 15 minutes to receive them.

So the EC2 instances are provisioned but because of the creation policy the auto scaling group doesn't move into a create complete state as normal.

It waits.

It can't complete until the creation policy directive is fulfilled.

The user data for the EC2 instances contains some bootstrapping and then this CFN signal statement at the bottom.

So once the bootstrapping process whatever it is has been completed and let's say that it's installing the Categorum application well the CFN signal tool signals the resource in this case the auto scaling group that it's completed the build.

So this CFN signal that's at the bottom left of your screen this is an actual utility which runs on the EC2 instance as part of the bootstrapping process.

And this causes each instance to signal once and the auto scaling group resource in the stack requires three of these signals within 15 minutes.

If it gets them all and assuming that they're all success signals then the stack moves into a create complete state.

If anything else happens so maybe a timeout happens or maybe one of the three instances has a bug then it will signal a failure and in any of those cases the stack will move into a create failed state.

Creation policies are generally used for EC2 instances or for auto scaling groups and if you do any of the advanced demo lessons in any of my courses you're going to see that I make use of this feature to ensure resources which are being provisioned are actually provisioned correctly before moving on to the next stage.

Now there are situations when you need some additional functionality maybe you want to pass data back to cloud formation or want to put general wait states into your template which can't be passed until a signal is received and that's where wait conditions come in handy.

Wait conditions operate in a similar way to creation policies.

A wait condition is a specific logical resource not something defined in an existing resource.

A wait condition can depend on other resources and other resources can also depend on a wait condition so it can be used as a more general progress gate within a template a point which can't be passed until those signals are received.

A wait condition will not proceed to create complete until it gets its signals or the timeout configured on that wait condition expires.

Now a wait condition relies on a wait handle and a wait handle is another logical resource whose sole job is to generate a pre-signed URL which can be used to send signals to.

It's pre-signed so that whatever using it doesn't need to use any AWS credentials they're included in the pre-signed URL.

So let's say that we have an EC2 instance or external server.

These are responsible for performing a process maybe some final detailed configuration or maybe they assign licensing something which has to happen after a part of the template but before the other part.

So these generate a JSON document which contains some amazing information or some amazing occurrence.

This is just an example it can be as complex or as simple as needed.

This document is passed back as the signal it has a status, a reason, a unique ID and some data.

Now what's awesome about this is that not only does this signal allow resource creation to be paused and then continued when this event has occurred but the data which has passed back can also be accessed elsewhere in the template.

We can use the get at function to query for the data attribute of the wait condition and get access to the details on the signal.

Now this allows a small amount of data exchange and processing between whatever is signaling and the cloud formation stack.

So you can inject specific data about a given event into the JSON document, send this back as a signal and then access this elsewhere in the cloud formation stack and this might be useful for certain things like licensing or to get additional status information about the event from the external system.

And that's wait conditions.

In many ways they're just like creation policies.

They have the same concept.

They allow a specific resource creation to be paused, not allowing progress until signaling is received.

Only wait conditions they're actually a separate resource and can use some more advanced data flow features like I'm demonstrating here.

AWS recommend creation policies for most situations because they're simpler to manage but as you create more complex templates you might well have need to use wait conditions as well and for the exams it's essential that you understand both creation policies and wait conditions which is why I wanted to go into detail on both.

Now that's all of the theory that I wanted to cover about creation policies and wait conditions and these are both things that you're going to get plenty of practical experience of in various demo lessons in all of my courses but I wanted to cover the theory and the architecture so that you can understand them when you come across them in those demos.

For now though thanks for watching go ahead and complete this video and when you're ready I'll look forward to you joining me in the next.

He says it was truly his passion but it’s not- he’s just now imagining another narcissistic fantasy. :(

to participate as a viewer also means to draw from your own experiences (conscious or not) in your response. there is some psychic exchange between performer, viewer, movement, music.

As an analogy, it would be hard to gain an understanding of a dish I've never had from just the raw ingredients eaten seprately. Perhaps there is needed infusing, mellowing by roasting, aromas from toasting, removal of seeds and stems, fermentation, baking, stewing that makes the experience nourishing to more than just the stomach. And perhaps a recipe is incomplete without it's original name, a visit to its place of procurement, attachment to its means of production, and the occasion for which it is prepared as a remedy for heartache, celebration, and expression of love.

I can relate to the authors experience in Brazil. I grew up Vietnamese American which exposed me to two different cultures and experiences. Even now, my family always comments on my body and face as if it’s a normal topic of conversation. I hear comments about my weight, hair, complexion every day from my mom, specifically. Sometimes they can be harsh and rarely are they positive, but I’ve gotten used to it. I’ve always told myself just Vietnamese culture. American culture is very different. I learned that from a young age. I remember I made a comment on my classmates weight when I was in the third grade and I hurt his feelings and had to speak to the teacher after class. I didn’t fully understand then the impact my words had on him, but looking back, I realized the difference between what I’m used to at home and what is the cultural norm here in the states. I’ve had many other eye-opening experiences like this throughout the years. I understand now that there are different values and beliefs between the two cultures and I must learn to respect both.

One of the most interesting things about this passage is the internal conflict within the Jenipapo-Kanindé community about embracing an "Indian" identity. It made me think about how complicated identity can be when it’s tied to social and political factors. On one hand, being recognized as indigenous brought real benefits, like better infrastructure and resources, but on the other hand, the stigma associated with the label made some people reject it. This shows how identity isn’t just personal or cultural, but is also shaped by outside pressures and perceptions. It makes me wonder: How do communities decide whether to take on an identity when it comes with both benefits and social stigma? And how much influence do outside groups, like researchers for example, have in shaping or even pressuring communities to adopt certain identities? Is it empowering, or does it create more conflict? I also think this connects to bigger questions about how marginalized groups reclaim or redefine their identities. For example, how do they balance wanting recognition for their heritage with dealing with stereotypes and discrimination? Personally, it makes me reflect on how external labels, whether cultural, ethnic, or something else, can impact how people see themselves or their place in society.

Social media brands are getting noticed for their human-like personas on the internet. There are pros such as building engagement, however it also raises ethical questions about these brands using serious issues, such as depression, to grab their target audience. There is some fear that young adults with mental health issues or isolation are turning to brands to relate to or create connections with. It's often hard to tell if there are real people or just simple marketing tactics to create more sales off of emotional engagement.

This is important just as the P in PARRS. We need to skim through and find clues if it’s relevant to the subject

High contrast. This website provides enough contrast between text and its background which is extremely beneficial for those with low vision and who do not have contrast-enhancing technology. This is an example of a good web accessibility practice as it adheres to the WCAG Contrast requirements, as it outlines that large-scale text and images have a contrast ratio of at least 4.5:1.

Perceivability Issue – Low Contrast

The text on website lacks sufficient contrast against the white or light-coloured background. This makes it harder for users with visual impairments or low contrast sensitivity to read and navigate the website effectively. Improving the contrast between the text and background would align with WCAG standards and ensure that all users can perceive the content clearly.

I chose the word "recoil" because I think it begins the showing of the might of God. Elements and people are reacting from the exposure to God's abilities and control, and the word "recoil" is a summary of that, as later lines tell of the various monuments that moved, like the ocean, mountains, and streams. When I think of the word recoil, I picture someone or something curling up in a either scared or disgusted way. It's a dramatic reaction in my opinion. The drama of using it in this context alludes to the power Milton is trying to convey. He's showing the extremity of God's power and just how forbidding it was. As a reader, once I arrive to this word, the Psalm turns darker. The beginning lines explicate the story of the Israelites and God's deliverance of them to the promised land; it has a more upbeat and positive feeling. Then you get to the lines where Milton describes how God's glory was unveiled and shown and you feel like an outsider, feeling the darkness of being on the other end of God's promise. In this way, the word "recoil" helps set up a sort of tension, displaying God's two sides, deliverer and wrathful.

We (scientists, anyway) want answers, we live for precision and correctness and repeatability. I think this gets in the way of finding productive ways of dealing with both interpersonal problems and social issues on a larger scale. In many cases (such as our innate tendencies of aggression, control over the less powerful, and focus on self-interest), it makes more sense to accept them as "givens" and seek ways to ameliorate them than it does to try understanding them.

Good: Keyboard Navigation You can navigate the BBC News website using just your keyboard, which is awesome for people who can’t use a mouse. Just hit the ‘Tab’ key, and you can move through links, buttons, and other interactive elements. Super handy for accessibility!

Good: Alt Text for Images They do a great job of adding descriptive alt text to their images, so people using screen readers can still “see” what’s there. This is a must-have for visually impaired users and something they’ve done right.

Good: Color Contrast The text stands out really well against the background, making it easy to read for everyone, including people with color blindness. Good contrast is one of those simple things that can make a huge difference.

Good: Skip to Content Link They have a “Skip to Content” link so you can jump straight to the main stuff without having to tab through all the menus. It’s a small detail, but it’s really helpful for keyboard users and screen readers.

Bad: Video Captions While most videos have captions, some are either missing them or using auto-generated captions that aren’t always accurate. This is a bit of a letdown because good captions are so important for people who are deaf or hard of hearing.

made me think of how this is still the norm/expectation for many in Nigeria, and how my generation frowns on it so much. It's not unique to us, and is actually the way that most of the world handled disputes for a large part of history. We're just in a unique time where our traditions haven't changed as much as the West's have, but we're so influenced by their culture that we're stuck somewhere in the liminal space between both cultures.

Unit 1 Socratic Seminar: Is Social Media More Beneficial or Negative to Society? Directions: Read and Annotate the readings by making comments in the document, or the margins if you’re doing it on paper on the 2 articles listed below. Use a physical highlighter or highlighter tool for quotes/ideas you want to explore more and talk about. TYPE YOUR ANSWERS IN BLUE IF DOING THIS DIGITALLY.<br /> Fill out the Summaries at the bottom of Fill out the 6 questions you will use during the Socratic to drive the conversation. Do this part LAST!

Write out the 6 Questions you will use during the Socratic Seminar. (Do this LAST IN BLUE FONT) 1. How will social media evolve in the future 2.How is social media affecting our outside interactions with others 3. 4. 5. 6.

Reading #1: Supporters Argue: Social Media Is Beneficial Overall 1a Supporters argue that social networking is a phenomenon that is beneficial overall and has changed the world for the better. Perhaps the greatest measure of social media's success, they contend, is the role it played in ousting undemocratic governments in Tunisia and Egypt. Journalist Peter Beaumont of the British newspaper the Guardian argued in 2011 that "a young woman or a young man with a smartphone" was the "defining" image of the Arab Spring. "The instantaneous nature of how social media communicate self-broadcast ideas, unlimited by publication deadlines and broadcast news slots, explains in part the speed at which these revolutions have unraveled, their almost viral spread across a region," he contended. "It explains, too, the often loose and non-hierarchical organization of the protest movements unconsciously modeled on the networks of the web." 2a Indeed, supporters argue that social media can be extremely useful in encouraging people who would not typically be politically motivated to engage in various issues or causes. While such statements are sometimes derided by critics as "hashtag activism" or "slacktivism," defenders insist that such actions really can make a difference. "What is commonly called slacktivism is not at all about 'slacking activists,'" Harvard University sociology professor Zeynep Tufekci wrote on her blog in 2012. "[R]ather it is about non-activists taking symbolic action—often in spheres traditionally engaged only by activists or professionals (governments, NGOs, international institutions.). Since these so-called 'slacktivists' were never activists to begin with, they are not in dereliction of their activist duties. On the contrary, they are acting, symbolically and in a small way, in a sphere that has traditionally been closed off to 'the masses' in any meaningful fashion." 3a Social media has many other benefits, advocates contend, including the potential to assist during times of catastrophe. During and after the terrorist attacks that rocked Paris, France, in November 2015, supporters note, people took to Facebook, Twitter, and other social media to communicate to loved ones that they were safe, or to offer refuge to people stranded in the city. "The attacks which ravaged the French capital yesterday showed how social media can also play a much more positive role," Forbes contributor Federico Guerrini wrote. "Facebook activated its Safety Check tool…to help people in areas affected by a disaster let their Facebook friends know they are safe. Twitter was also helpful: residents used the hashtag #porteouverte [open doors] to offer shelter to people stranded in the city." Advocates of social networking contend that sites like Facebook and Twitter have brought people closer together. "It has never been easier to make friends than it is right now, mainly thanks to social networking sites," writer Dave Parrack argued on the technology website MakeUseOf.com in 2012. "Just a few decades ago it was pretty tough to connect with people unless you were the overly outgoing type able to make conversation with anyone at a party. The rise of mobile phones helped change this, connecting people in a new way, but then social networks sprang up and the whole idea of friendship changed once more and forever." 4a Supporters maintain that social networking sites increasingly function as a refuge where people can relax with their friends and family. "This is where social media become a powerful social force in the modern sphere," Taso Lagos of the University of Washington wrote in the Seattle Times in 2012. "Because we live in a world of constant anxiety and stress about our lives, our careers, the planet and the fate of our families and friends, trusted sites like Facebook and Twitter are places we turn to relieve this tension and allow us to live and express our humanity." Social media, he argued, are "the community centers of the future." 5a Such sites provide many valuable benefits, defenders argue, including enhancing people's sense of self-worth. The act of taking and posting selfies, they contend, helps people exert control over their self-image and the way they are viewed. "The harshly judged practice of self-picture taking," Huffington Post contributor Molly Fosco wrote in March 2014, "while perhaps excessive or annoying at times, can actually be a really simple way to feel really good about yourself…. Although our selfies might be veiled in narcissism, self-obsession, or boastfulness I think that for many it's a genuine attempt to boost self-esteem. Seeing a close-up picture of your own face and willingly showing it to thousands of people with one click is a form of self-confidence that I don't think should be quickly dismissed." 6a Supporters of social media discount many of the fears typically raised by opponents, noting that it is common for new technology to stir criticism. In the late 19th century, they note, some observers predicted that the telephone would severely damage interpersonal relationships, just as detractors of social media do today. The telephone "was going to bring down our society," Megan Moreno of the University of Wisconsin in Madison told the New York Times in 2012. "Men would be calling women and making lascivious comments, and women would be so vulnerable, and we'd never have civilized conversations again." She added, "When a new technology comes out that is something so important, there is this initial alarmist reaction." Write out a 100-word summary of your thoughts/ideas/opinions of the strengths and weaknesses of the Beneficial Side. (TYPE IN BLUE FONT) Social media supporters argue that it is a good thing for the world and there is proof that it helps the movements like Arab Spring to go on smoothly and global peace talks to be the most constructive. It leads to a lot of people and even calling on them to fight for their cause. It is useful for giving real time help like social media platforms and info at the time of an emergency and also brings people who don't live close to each other, closer to each other. Social media can also lead to the development of good self esteem through many apps. These benefits of the media source have risks which include being heavily dependent on technology, getting wrong info, and the threat of getting into harmful sites with people despite how useful it can be sometimes .

Reading #2: Opponents Argue: Social Media Is Not Beneficial Overall 1b Opponents of social networking argue that such sites are not beneficial overall and that they gradually erode many essential aspects of communication and socialization. "The shortcomings of social media would not bother me awfully if I did not suspect that Facebook friendship and Twitter chatter are displacing real rapport and real conversation," New York Times commentator Bill Keller argued in 2011. "The things we may be unlearning, tweet by tweet—complexity, acuity, patience, wisdom, intimacy—are things that matter." 2b Indeed, critics contend, the rise of social networking has coincided with a decline in the quality of conversation. "As we ramp up the volume and velocity of online connections, we start to expect faster answers," MIT psychology professor Sherry Turkle wrote in the New York Times in 2012. "To get these, we ask one another simpler questions; we dumb down our communications, even on the most important matters." 3b Opponents argue that social media can contribute to feelings of sadness and loneliness. A study by researchers at the University of Michigan in 2013, they note, found that college-aged users felt worse the more they used Facebook. Because people's Facebook personas are often curated to make their lives seem fun or perfect, critics argue, that browsing social media can contribute to feelings of inadequacy. "When you're on a site like Facebook, you get lots of posts about what people are doing," co-author John Jonides, a cognitive neuroscientist at the Department of Psychology at the University of Michigan, told National Public Radio in 2013. "That sets up a social comparison — you maybe feel your life is not as full and rich as those people you see on Facebook." 4b Social media, critics charge, can lead people to obsess about themselves and their self-image to the point where it can be harmful. People need to look deeper for self-worth, they contend, than achieving "likes" by posting selfies on social media. "[I]if you've just spent half an hour editing a photo by blurring around your eyes with one app, adding eyelashes with another, then changing the colors with a third," Teen Vogue contributor Tiffany Perry wrote in March 2016, "chances are you're giving too much merit to how others perceive you." 5b Other critics claim that the impact of social media on political phenomena like the Arab Spring has been overstated. New Yorker columnist Malcolm Gladwell noted in 2011 that many revolutions took place throughout history before the advent of social networking. "People with a grievance will always find ways to communicate with each other," he wrote. "How they choose to do it is less interesting, in the end, than why they were driven to do it in the first place." 6b Opponents also assert that promoting political or social causes on social media has little real impact other than to make the person making the post feel good about themselves. In 2013, for example, the United Nations Children's Emergency Fund (UNICEF), a U.N. organization that raises money to help and protect children throughout the world, ran an ad campaign with a slogan that read "Like us on Facebook, and we will vaccinate zero children against polio." The point of the campaign, UNICEF explained, was not to disparage "likes" but to encourage more active support, such as contributing money to buy vaccines. "Slacktivism's inherent laziness disqualifies it as a real agent of progress because it does not possess the enthusiasm necessary for change," contributor Elias Tavaras wrote for the Hill in January 2016. "How can a post on Facebook inspire necessary action, especially when sitting down on a comfy computer chair? Indeed, the passion one may feel disappears, with a simple scroll or is drowned out by the other slacktivist posts." 7b Critics charge that social media users are in danger of having their online personas co-opted by corporations eager to collect the information users share and employ it for marketing purposes. Robert Barry of the pop culture website The Quietus argues that social media is turning people into "branded products." "Online businesses which seem to be promising something for nothing—from social networking to file sharing—are really offering you, their audience, as a readymade and fully packaged item for purchase," he argued, "be that by the ghost of advertising's future, or the investor whose faith gives that ghost substance." Write out a 100-word summary of your thoughts/ideas/opinions of the strengths and weaknesses of the Against Side. (TYPE IN BLUE FONT)

I try to emphasize that grading is not the be all, end all. It is just a tool to determine if the course content is understood (it's also a good evaluation of myself and how well I'm able to impart the knowledge and skills asked for)

It's interesting how, for those who grew up with English as their native language, the symbol for a big brown thing sticking out of the ground with green puffs is recognized as a 'tree.' In other languages, the symbols or words might be similar or completely different. As someone who is bilingual, I've noticed that the word for 'tree' in our second language is significantly different from the English term. If we were to integrate that word into the English language, it would have no meaning at all.

This shows how so many men just think it's okay to take advantage of anybody they are in higher power in for instance bosses not giving fair wages or even promotions to improve because its controlling tactic.

Right before this I was just thinking how I'm not really either one or the other and I was kind of in the middle. So I'm glad to know that it's actually a good spot and not bad to be more on one side rather than the other

Author response:

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

Reviewer #1 (Public review):

Summary:

In the manuscript the authors describe a new pipeline to measure changes in vasculature diameter upon optogenetic stimulation of neurons. The work is useful to better understand the hemodynamic response on a network /graph level.

Strengths:

The manuscript provides a pipeline that allows to detect changes in the vessel diameter as well as simultaneously allows to locate the neurons driven by stimulation.

The resulting data could provide interesting insights into the graph level mechanisms of regulating activity dependent blood flow.

Weaknesses:

(1) The manuscript contains (new) wrong statements and (still) wrong mathematical formulas.

The symbols in these formulas have been updated to disambiguate them, and the accompanying statements have been adjusted for clarity.

(2) The manuscript does not compare results to existing pipelines for vasculature segmentation (opensource or commercial). Comparing performance of the pipeline to a random forest classifier (illastik) on images that are not preprocessed (i.e. corrected for background etc.) seems not a particularly useful comparison.

We’ve now included comparisons to Imaris (a commercial) for segmentation and VesselVio (open-source) for graph extraction software.

For the ilastik comparison, the images were preprocessed prior to ilastik segmentation, specifically by doing intensity normalization.

Example segmentations utilizing Imaris have now been included. Imaris leaves gaps and discontinuities in the segmentation masks, as shown in Supplementary Figure 10. The Imaris segmentation masks also tend to be more circular in cross-section despite irregularities on the surface of the vessels observable in the raw data and identified in manual segmentation. This approach also requires days to months to generate per image stack.

“Comparison with commercial and open-source vascular analysis pipelines

To compare our results with those achievable on these data with other pipelines for segmentation and graph network extraction, we compared segmentation results qualitatively with Imaris version 9.2.1 (Bitplane) and vascular graph extraction with VesselVio [1]. For the Imaris comparison, three small volumes were annotated by hand to label vessels. Example slices of the segmentation results are shown in Supplementary Figure 10. Imaris tended to either over- or under-segment vessels, disregard fine details of the vascular boundaries, and produce jagged edges in the vascular segmentation masks. In addition to these issues with segmentation mask quality, manual segmentation of a single volume took days for a rater to annotate. To compare to VesselVio, binary segmentation masks (one before and one after photostimulation) generated with our deep learning models were loaded into VesselVio for graph extraction, as VesselVio does not have its own method for generating segmentation masks. This also facilitates a direct comparison of the benefits of our graph extraction pipeline to VesselVio. Visualizations of the two graphs are shown in Supplementary Figure 11. Vesselvio produced many hairs at both time points, and the total number of segments varied considerably between the two sequential stacks: while the baseline scan resulted in 546 vessel segments, the second scan had 642 vessel segments. These discrepancies are difficult to resolve in post-processing and preclude a direct comparison of individual vessel segments across time. As the segmentation masks we used in graph extraction derive from the union of multiple time points, we could better trace the vasculature and identify more connections in our extracted graph. Furthermore, VesselVio relies on the distance transform of the user supplied segmentation mask to estimate vascular radii; consequently, these estimates are highly susceptible to variations in the input segmentation masks.We repeatedly saw slight variations between boundary placements of all of the models we utilized (ilastik, UNet, and UNETR) and those produced by raters. Our pipeline mitigates this segmentation method bias by using intensity gradient-based boundary detection from centerlines in the image (as opposed to using the distance transform of the segmentation mask, as in VesselVio).”

(3) The manuscript does not clearly visualize performance of the segmentation pipeline (e.g. via 2d sections, highlighting also errors etc.). Thus, it is unclear how good the pipeline is, under what conditions it fails or what kind of errors to expect.

On reviewer’s comment, 2D slices have been added in the Supplementary Figure 4.

(4) The pipeline is not fully open-source due to use of matlab. Also, the pipeline code was not made available during review contrary to the authors claims (the provided link did not lead to a repository). Thus, the utility of the pipeline was difficult to judge.

All code has been uploaded to Github and is available at the following location: https://github.com/AICONSlab/novas3d

The Matlab code for skeletonization is better at preserving centerline integrity during the pruning of hairs from centerlines than the currently available open-source methods.

- Generalizability: The authors addressed the point of generalizability by applying the pipeline to other data sets. This demonstrates that their pipeline can be applied to other data sets and makes it more useful.  However, from the visualizations it's unclear to see the performance of the pipeline, where the pipelines fails etc. The 3d visualizations are not particularly helpful in this respect . In addition, the dice measure seems quite low, indicating roughly 20-40% of voxels do not overlap between inferred and ground truth. I did not notice this high discrepancy earlier. A thorough discussion of the errors appearing in the segmentation pipeline would be necessary in my view to better assess the quality of the pipeline.

2D slices from the additional datasets have been added in the Supplementary Figure 13 to aid in visualizing the models’ ability to generalize to other datasets.

The dice range we report on (0.7-0.8) is good when compared to those (0.56-86) of 3D segmentations of large datasets in microscopy [2], [3], [4], [5], [6]. Furthermore, we had two additional raters segment three images from the original training set. We found that the raters had a mean inter class correlation  of 0.73 [7]. Our model outperformed this Dice score on unseen data: Dice scores from our generalizability tests on C57 mice and Fischer rats on par or higher than this baseline.

Reviewer #2 (Public review):<br /> The authors have addressed most of my concerns sufficiently. There are still a few serious concerns I have. Primarily, the temporal resolution of the technique still makes me dubious about nearly all of the biological results. It is good that the authors have added some vessel diameter time courses generated by their model. But I still maintain that data sampling every 42 seconds - or even 21 seconds - is problematic. First, the evidence for long vascular responses is lacking. The authors cite several papers:

Alarcon-Martinez et al. 2020 show and explicitly state that their responses (stimulus-evoked) returned to baseline within 30 seconds. The responses to ischemia are long lasting but this is irrelevant to the current study using activated local neurons to drive vessel signals.

Mester et al. 2019 show responses that all seem to return to baseline by around 50 seconds post-stimulus.

In Mester et al. 2019, diffuse stimulations with blue light showed a return to baseline around 50 seconds post-stimulus (cf. Figure 1E,2C,2D). However, focal stimulations where the stimulation light is raster scanned over a small region focused in the field of view show longer-lasting responses (cf. Figure 4) that have not returned to baseline by 70 seconds post-stimulus [8]. Alarcon-Martinez et al. do report that their responses return baseline within 30 seconds; however, their physiological stimulation may lead to different neuronal and vessel response kinetics than those elicited by the optogenetic stimulations as in current work.

O'Herron et al. 2022 and Hartmann et al. 2021 use opsins expressed in vessel walls (not neurons as in the current study) and directly constrict vessels with light. So this is unrelated to neuronal activity-induced vascular signals in the current study.

We agree that optogenetic activation of vessel-associated cells is distinct from optogenetic activation of neurons, but we do expect the effects of such perturbations on the vasculature to have some commonalities.

There are other papers including Vazquez et al 2014 (PMID: 23761666) and Uhlirova et al 2016 (PMID: 27244241) and many others showing optogenetically-evoked neural activity drives vascular responses that return back to baseline within 30 seconds. The stimulation time and the cell types labeled may be different across these studies which can make a difference. But vascular responses lasting 300 seconds or more after a stimulus of a few seconds are just not common in the literature and so are very suspect - likely at least in part due to the limitations of the algorithm.

The photostimulation in Vazquez et al. 2014 used diffuse photostimulation with a fiberoptic probe similar to Mester et al. 2019 as opposed to raster scanning focal stimulation we used in this study and in the study by Mester et al. 2019  where we observed the focal photostimulation to elicited longer than a minute vascular responses. Uhlirova et al. 2016 used photostimulation powers between 0.7 and 2.8 mW, likely lower than our 4.3 mW/mm2 photostimulation. Further, even with focal photostimulation, we do see light intensity dependence of the duration of the vascular responses. Indeed, in Supplementary Figure 2, 1.1 mW/mm2 photostimulation leads to briefer dilations/constrictions than does 4.3 mW/mm2; the 1.1 mW/mm2 responses are in line, duration wise, with those in Uhlirova et al. 2016.

Critically, as per Supplementary Figure 2, the analysis of the experimental recordings acquired at 3-second temporal resolution did likewise show responses in many vessels lasting for tens of seconds and even hundreds of seconds in some vessels.

Another major issue is that the time courses provided show that the same vessel constricts at certain points and dilates later. So where in the time course the data is sampled will have a major effect on the direction and amplitude of the vascular response. In fact, I could not find how the "response" window is calculated. Is it from the first volume collected after the stimulation - or an average of some number of volumes? But clearly down-sampling the provided data to 42 or even 21 second sampling will lead to problems. If the major benefit to the field is the full volume over large regions that the model can capture and describe, there needs to be a better way to capture the vessel diameter in a meaningful way.

In the main experiment (i.e. excluding the additional experiments presented in the Supplementary Figure 2 that were collected over a limited FOV at 3s per stack), we have collected one stack every 42 seconds. The first slice of the volume starts following the photostimulation, and the last slice finishes at 42 seconds. Each slice takes ~0.44 seconds to acquire. The data analysis pipeline (as demonstrated by the Supplementary Figure 2) is not in any way limited to data acquired at this temporal resolution and - provided reasonable signal-to-noise ratio (cf. Figure 5) - is applicable, as is, to data acquired at much higher sampling rates.

It still seems possible that if responses are bi-phasic, then depth dependencies of constrictors vs dilators may just be due to where in the response the data are being captured - maybe the constriction phase is captured in deeper planes of the volume and the dilation phase more superficially. This may also explain why nearly a third of vessels are not consistent across trials - if the direction the volume was acquired is different across trials, different phases of the response might be captured.

Alternatively, like neuronal responses to physiological stimuli, the vascular responses elicited by increases in neuronal activity may themselves be variable in both space and time.

I still have concerns about other aspects of the responses but these are less strong. Particularly, these bi-phasic responses are not something typically seen and I still maintain that constrictions are not common. The authors are right that some papers do show constriction. Leaving out the direct optogenetic constriction of vessels (O'Herron 2022 & Hartmann 2021), the Alarcon-Martinez et al. 2020 paper and others such as Gonzales et al 2020 (PMID: 33051294) show different capillary branches dilating and constricting. However, these are typically found either with spontaneous fluctuations or due to highly localized application of vasoactive compounds. I am not familiar with data showing activation of a large region of tissue - as in the current study - coupled with vessel constrictions in the same region. But as the authors point out, typically only a few vessels at a time are monitored so it is possible - even if this reviewer thinks it unlikely - that this effect is real and just hasn't been seen.

Uhlirova et al. 2016 (PMID: 27244241) observed biphasic responses in the same vessel with optogenetic stimulation in anesthetized and unanesthetized animals (cf Fig 1b and Fig 2, and section “OG stimulation of INs reproduces the biphasic arteriolar response”). Devor et al. (2007) and Lindvere et al. (2013) also reported on constrictions and dilations being elicited by sensory stimuli.

I also have concerns about the spatial resolution of the data. It looks like the data in Figure 7 and Supplementary Figure 7 have a resolution of about 1 micron/pixel. It isn't stated so I may be wrong. But detecting changes of less than 1 micron, especially given the noise of an in vivo prep (brain movement and so on), might just be noise in the model. This could also explain constrictions as just spurious outputs in the model's diameter estimation. The high variability in adjacent vessel segments seen in Figure 6C could also be explained the same way, since these also seem biologically and even physically unlikely.

Thank you for your comment. To address this important issue, we performed an additional validation experiment where we placed a special order of fluorescent beads with a known diameter of 7.32 ± 0.27um, imaged them following our imaging protocol, and subsequently used our pipeline to estimate their diameter. Our analysis converged on the manufacturer-specified diameters, estimating the diameter to be 7.34 ± 0.32. The manuscript has been updated to detail this experiment, as below:

Methods section insert

“Second, our boundary detection algorithm was used to estimate the diameters of fluorescent beads of a known radius imaged under similar acquisition parameters. Polystyrene microspheres labelled with Flash Red (Bangs Laboratories, inc, CAT# FSFR007) with a nominal diameter of 7.32um and a specified range of 7.32 ± 0.27um as determined by the manufacturer using a Coulter counter were imaged on the same multiphoton fluorescence microscope set-up used in the experiment (identical light path, resonant scanner, objective, detector, excitation wavelength and nominal lateral and axial resolutions, with 5x averaging). The images of the beads had a higher SNR than our images of the vasculature, so Gaussian noise was added to the images to degrade the SNR to the same level of that of the blood vessels. The images of the beads were segmented with a threshold, centroids calculated for individual spheres, and planes with a random normal vector extracted from each bead and used to estimate the diameter of the beads. The same smoothing and PSF deconvolution steps were applied in this task. We then reported the mean and standard deviation of the distribution of the diameter estimates. A variety of planes were used to estimate the diameters.”

Results Section Insert

“Our boundary detection algorithm successfully estimated the radius of precisely specified fluorescent beads. The bead images had a signal-to-noise ratio of 6.79 ± 0.16 (about 35% higher than our in vivo images): to match their SNR to that of in vivo vessel data, following deconvolution, we added Gaussian noise with a standard deviation of 85 SU to the images, bringing the SNR down to 5.05 ± 0.15. The data processing pipeline was kept unaltered except for the bead segmentation, performed via image thresholding instead of our deep learning model (trained on vessel data). The bead boundary was computed following the same algorithm used on vessel data: i.e., by the average of the minimum intensity gradients computed along 36 radial spokes emanating from the centreline vertex in the orthogonal plane. To demonstrate an averaging-induced decrease in the uncertainty of the bead radius estimates on a scale that is finer than the nominal resolution of the imaging configuration, we tested four averaging levels in 289 beads. Three of these averaging levels were lower than that used on the vessels, and one matched that used on the vessels (36 spokes per orthogonal plane and a minimum of 10 orthogonal planes per vessel). As the amount of averaging increased, the uncertainty on the diameter of the beads decreased, and our estimate of the bead's diameter converged upon the manufacturer's Coulter counter-based specifications (7.32 ± 0.27um), as tabulated in Table 1.”

Reviewer #1 (Recommendations for the authors):

Comments to the authors replies to the reviews:

- Supplementary Figure 13:

As indicated before the 3d images + scale makes it impossible to judge the quality of the outputs.

As aforementioned, 2D slices have been added to the Supplementary Figure 13.

- Supplementary Table 3:

There is a significant increase in the Hausdorrf and Mean Surface Distance measures for the new data, why ?

A single vessel being missed by either the rater or the model would significantly affect the Hausdorff distance (HD) and by extension Mean Surface Distance: this is particularly pertinent in the LSFM image with its much larger FOV and thus a potential for much larger max distances to result from missed vessels in the prediction or ground truth data. Large Hausdorff distances may indicate a vessel was missed in either the ground truth or the segmentation mask.

Of note, a different rater annotated these additional datasets from the raters labeling the ground truth data. There is a high variability in boundary placements between raters. On a test where three raters segmented the same three images from the original dataset, we computed a ICC of 0.73 across their segmentations. Our model Dice scores on predictions in out-of-distribution data sets were on par with the inter-rater ICC on the Thy1ChR2 2PFM data.

- Supplementary Figure 2: The authors provide useful data on the time responses. However, looking at those figures, it is puzzling why certain vessels were selected as responding as there seems almost no change after stimulation. In addition, some of the responses seem to actually start several tens of seconds before the actual stimulus (particularly in A).

Only some traces in C and D (dark blue) seem to be actually responding vessels.

This is not discussed and unclear.

Supplementary Figure 2 displays the time courses of vessel calibre for all vessels in the FOV, not just those deemed responders.

The aforementioned effects are due to the loess smoothing filter having been applied to the time courses for the preliminary response, which has been rectified in the updated figures. In particular, Supplementary Figure 2 has been updated with separate loess smoothing before and after photostimulation. The (pre-stimulation) effect is gone once the loess smoothing has been separated.

- R Point 7: As indicated before and in agreement with the alternative reviewer, the quality of the results in 3d is difficult to judge. No 2d sections that compare 'ground truth' with inferred results are shown in the current manuscript which would enable a much better judgment. The provided video is still 3d and not a video going through 2d slices. Also, in the video the overlap of vasculature and raw data seems to be very good and near 100%, why is the dice measure reported earlier so low ? Is this a particularly good example ?

Some examples, indicating where the pipeline fails (and why) would be helpful to see, to judge its performance better (ideally in 2d slices).

As discussed in the public comments, the 2D slices are now included in Suppl. Fig. 4 and suppl. Fig 13 to facilitate visual assessment. The vessels are long and thin so that slight dilations or constrictions impact the Dice scores without being easily visualizable.

- Author response images 6 and 7. From the presented data the constrictions measured in the smaller vessels may be a result (at least partly) of noise. This seems to be particularly the case in Author response image 7 left top and bottom for example. It would be helpful to show the actual estimates of the vessels radii overlaid in the (raw) images. In some of the pictures the noise level seems to reach higher values than the 10-20% of noise used in the tests by the authors in the revision.

The vessel radii are estimated as averages across all vertices of the individual vessels: it is thus not possible to overlay them meaningfully in 2D slices: in Figure 2B, we do show a rendering of sample vessel-wise radii estimates.

- "We tested the centerline detection in Python, scipy (1.9.3) and Matlab. We found that the Matlab implementation performed better due to its inclusion of a branch length parameter for the identification of terminal branches, which greatly reduced the number of false branches; the Python implementation does not include this feature (in any version) and its output had many more such "hair" artifacts. Clearmap skeletonization uses an algorithm by Palagyi & Kuba(1999) to thin segmentation masks, which does not include hair removal. Vesselvio uses a parallelized version of the scipy implementation of Lee et al. (1994) algorithm which does not do hair removal based on a terminal branch length filter; instead, Vesselvio performs a threshold-based hair removal that is frequently overly aggressive (it removes true positive vessel branches), as highlighted by the authors."

This statement is wrong. The removal of small branches in skeletons is algorithmically independent of the skeletonization algorithm itself. The authors cite a reference concerned with the algorithm they are currently employing for the skeletonization. Careful assessment of that reference shows that this algorithm removes small length branches after skeletonization is performed. This feature is available in open-source packages as well, or could be easily implemented.

We appreciate that skeletonization is distinct from hair removal and have reworded this paragraph for clarity. We are currently working with SciPy developers to implement hair removal in their image processing pipeline so as to render our pipeline fully open-source.

The removal of hairs after skeletonization with length based thresholding leads to the possibility of removing parts of centerlines in the main part of vessels after branch points with hairs. The Matlab implementation does not do this and leaves the main branches intact.

This text has been updated to:

“Hair” segments shorter than 20 μm and terminal on one end were iteratively removed, starting with the shortest hairs and merging the longest hairs at junctions with 2 terminal branches with the main vessel branch to reduce false positive vascular branches and minimize the amount of centerlines removed. This iterative hair removal functionality of the skeletonization algorithm is currently unavailable in python, but is available in Matlab [9].

- "On the reviewer's comment, we did try inputting normalized images into Ilastik, but this did not improve its results." This is surprising. Reasonable standard preprocessing (e.g. background removal, equalization, and vessel enhancement) would probably restore most of illastik's performance in the indicated panel.

While the improvement may be present in a particular set of images, the generalizability of such improvement to other patches is often poor in our experience, as reflected by aforementioned results and the widespread uptake of DL approaches to image segmentation. The in vivo datasets also contain artifacts arising from eg. bleeding into the FOV that ilastik is highly sensitive to. This is an example of noise that is not easily removed by standard preprocessing.

- "Typical pre-processing/standard computer vision techniques with parameter tuning do not generalize on out-of-distribution data with different image characteristics, motivating the shift to DL-based approaches."

I disagree with this statement. DL approaches can generalize typically when trained with sufficient amount of diverse data. However, DL approaches can also fail with new out of distribution data. In that situation they only be 'rescued' via new time intensive data generation and retraining. Simple standard image pre-processing steps (e.g. to remove background or boost vessel structures) have well defined parameter that can be easily adapted to new out of distribution data as clear interpretations are available. The time to adapt those parameters is typically much smaller than retraining of DL frameworks.

We find that the standard image processing approaches with parameter tuning work robustly only if fine-tuned on individual images; i.e., the fine-tuning does not generalize across datasets. This approach thus does not scale to experiments yielding large image sizes/having high throughput experiments. While DL models may not generalize to out-of-distribution data, fine-tuning DL models with a small subset of labels generally produce superior models to parameter tuning that can be applied to entire studies. Moreover, DL fine-tuning is typically an efficient process due to very limited labelling and training time required.

- It is still unclear how the authors pipeline performs compared with other (open source or commercially) available pipelines. As indicated before, comparing to illastik, particularly when feeding non preprocessed data, does not seem to be a particularly high bar.

This question has also been raised by the other reviewer who asked to compare to commercially available pipelines.

This question was not answered by the authors, and instead the authors reply by claiming to provide an open source pipeline. In fact, the use of matlab in their pipeline does not make it fully open-source either. Moreover, as mentioned before, open-source pipelines for comparisons do exists.

As discussed above, the manuscript now includes comparisons to Imaris for segmentation and Vesselvio for graph extraction. The pipeline is on github.

-"We agree with the review that this question is interesting; however, it is not addressable using present data: activated neuronal firing will have effects on their postsynaptic neighbors, yet we have no means of measuring the spread of activation using the current experimental model."

Distances to the closest neuron in the manuscript are measured without checking if it's active. Thus, distances to the first set of n neurons could be measured in the same way, ignoring activation effects.

Shorter distances to an entire ensemble of neurons would still be (more) informative of metabolic demands.

This could indeed be done within the existing framework. The connected-components-3d can be used to extract individual occurrences of neurons in the FOV from the neuron segmentation mask. Each neuron could then have its distance calculated to each point on the vessel centerlines.

- model architecture:

It is unclear from the description if any positional encoding was used for the image patches.

It is unclear if the architecture / pipeline can handle any volume sizes or is trained on a fixed volume shapes? In the latter case how is the pipeline applied?

The model includes positional encoding, as described in Hatamizadeh et al. 2021.

The model can be applied to images of any size, as demonstrated on larger images in Supplementary Figure 9 and on smaller images in Supplementary Figure 2. The pipeline is applied in the same way. It will read in the size of an input image and output an image of the same size.

- transformer models often show better results when using a learning rate scheduler that adjust the learning rate (up and down ramps typically). Did the authors test such approaches?

We did not use a learning rate scheduler, as we found we were getting good results without using one.

- formula (4): The 95% percentile of two numbers is the max, and thus (5) is certainly not what the HD95 metric is. The formula is simply wrong as displayed.

Thank you. The formula has been updated.

- formula (5): formula 5 is certainly wrong: n_X, n_y are either integer numbers as indicated by the sum indices or sets when used in the distances, but can't be both at the same time.

Thank you for your comment. The Formula has been updated.

- The statement:

"this functionality of the skeletonization algorithm is currently unavailable in any python implementation, but is available in Matlab [56]."

is not correct (see reply above)

Please see the response above. This text has been updated to:

“Hair” segments shorter than 20 μm and terminal on one end were iteratively removed, starting with the shortest hairs and merging the longest hairs at junctions with 2 terminal branches with the main vessel branch to reduce false positive vascular branches and minimize the amount of centerlines removed. This iterative hair removal functionality of the skeletonization algorithm is currently unavailable in Python, but is available in Matlab [9].

- the centerline extraction is performed after taking the union of smoothed masks. The union operation can induce novel 'irregular' boundaries that degrade skeletonization performance. I would expect to apply smoothing after the union?

Indeed the images were smoothed via dilation after taking the union, as described in the previous set of responses to private comments.

- "The radius estimate defined the size of the Gaussian kernel that was convolved with the image to smooth the vessel: smaller vessels were thus convolved with narrower kernels."

It's unclear what image were filtered ?

We have updated this text for clarity:

The radius estimate defined the size of the Gaussian kernel that was convolved with the 2D image slice to smooth the vessel: smaller vessels were thus convolved with narrower kernels.

- Was deconvolution on the raw images applied or after Gaussian filtering ?

The deconvolution was applied before Gaussian filtering.

- ",we extracted image intensities in the orthogonal plane from the deconvolved raw registered image. A 2D Gaussian kernel with sigma equal to 80% of the estimated vessel-wise radius was used to low-pass filter the extracted orthogonal plane image and find the local signal intensity maximum searching, in 2D, from the center of the image to the radius of 10 pixels from the center."

Would it not be better to filter the 3d image before extracting a 2d plane and filter then ?

That could be done, but would incur a significant computational speed penalty. 2D convolutions are faster, and produced excellent accuracy when estimating radii in our bead experiment.

What algorithm was used to obtain the 2d images.

The 2d images were obtained using scipy.ndimage.map_coordinates.

- Figure 2: H is this the filtered image or the raw data ?

Panel H is raw data.

- It would be good to see a few examples of the raw data overlaid with the radial estimates to evaluate the approach (beyond the example in K).

Additional examples are shown in Figure 5.

- Figure 2 K: Why are boundary points greater than 2 standard deviations away from the mean excluded ?

They are excluded to account for irregularities as vessels approach junctions [10], [11] REF.

- Figure 2 L: what exactly is plotted here ? What are vertex wise changes, is that the difference between the minimum and maximum of all the detected radii for a single vertex? Why do some vessels (red) show high values consistently throughout the vessel ?

Figure 2L displays change in the radius of vertices - in this FOV- following photostimulation in relation to baseline.

- Assortativity: to calculate the assortativity, are radius changes binned in any form to account for the fact that otherwise, $e_{xy}$ and related measures will be likely be based on single data points?

Assortativity is not calculated from single data points. It can be calculated by either binning into categories or computing it on scalars i.e. average radius across a vessel segment:

See here for info on calculating assortativity from binned categories (ie classifying a vessel as a constrictor, dilator or non-responder):

https://networkx.org/documentation/stable/reference/algorithms/generated/networkx.algorithms.assortativity.attribute_assortativity_coefficient.html#networkx.algorithms.assortativity.attribute_assortativity_coefficient

And see here for calculating assortativity from scalar values:

https://networkx.org/documentation/stable/reference/algorithms/generated/networkx.algorithms.assortativity.numeric_assortativity_coefficient.html#networkx.algorithms.assortativity.numeric_assortativity_coefficient

We calculated the assortativity using scalar values.

In both cases, one uses all nodes and calculates the correlation between each node and its neighbours with an attribute that can be binned or is a scalar. Binning the value on a given node would not affect the number of nodes in a graph.

- "Ilastik tended to over-segment vessels, i.e. the model returned numerous false positives, having a high recall (0.89{plus minus}0.19) but low precision (0.37{plus minus}0.33) (Figure 3, Supplementary Table 3)."

As indicated before, and looking at Figure 4, over segmentation seems due to too high background. A suggested preprocessing step on the raw images to remove background could have avoided this.

The images were normalized in preprocessing.

- Figure 4: The 3d panels are not much easier to read in the revised version. As suggested by other reviewers, 2d sections indicating the differences and errors would be much more helpful to judge the pipelines quality more appropriately.

As discussed above, 2D sections are now available in a supplementary figure.

- Figure 3: What would be the dice score (and other measures) between two ground truths extracted by two annotations by two humans (assisted e.g. by illastik).

Two additional human rates annotated images. We observed a ICC of 0.73 across a total of three raters on the three images.

- Figure 5: The authors only provide the absolute value of SU for the sigma noise levels. This only has some meaning when compared to the mean or median SU of the images. In the text the maximal intensity of 1023 SU is mentioned, but what are those values in images with weaker / smaller vessels (as provided in the constriction examples in the revision)/

I am unclear why this validation figure should be part of the main manuscript while generalization performance is left out.

The manuscript has been updated with the mean SNR value of 5.05 ± 0.15 to provide context for the quality of our images.

Bibliography

(1) J. R. Bumgarner and R. J. Nelson, “Open-source analysis and visualization of segmented vasculature datasets with VesselVio,” Cell Rep. Methods, vol. 2, no. 4, Apr. 2022, doi: 10.1016/j.crmeth.2022.100189.

(2) G. Tetteh et al., “DeepVesselNet: Vessel Segmentation, Centerline Prediction, and Bifurcation Detection in 3-D Angiographic Volumes,” Front. Neurosci., vol. 14, Dec. 2020, doi: 10.3389/fnins.2020.592352.

(3) N. Holroyd, Z. Li, C. Walsh, E. Brown, R. Shipley, and S. Walker-Samuel, “tUbe net: a generalisable deep learning tool for 3D vessel segmentation,” Jul. 24, 2023, bioRxiv. doi: 10.1101/2023.07.24.550334.

(4) W. Tahir et al., “Anatomical Modeling of Brain Vasculature in Two-Photon Microscopy by Generalizable Deep Learning,” BME Front., vol. 2020, p. 8620932, Dec. 2020, doi: 10.34133/2020/8620932.

(5) R. Damseh, P. Delafontaine-Martel, P. Pouliot, F. Cheriet, and F. Lesage, “Laplacian Flow Dynamics on Geometric Graphs for Anatomical Modeling of Cerebrovascular Networks,” ArXiv191210003 Cs Eess Q-Bio, Dec. 2019, Accessed: Dec. 09, 2020. [Online]. Available: http://arxiv.org/abs/1912.10003

(6) T. Jerman, F. Pernuš, B. Likar, and Ž. Špiclin, “Enhancement of Vascular Structures in 3D and 2D Angiographic Images,” IEEE Trans. Med. Imaging, vol. 35, no. 9, pp. 2107–2118, Sep. 2016, doi: 10.1109/TMI.2016.2550102.

(7) T. B. Smith and N. Smith, “Agreement and reliability statistics for shapes,” PLOS ONE, vol. 13, no. 8, p. e0202087, Aug. 2018, doi: 10.1371/journal.pone.0202087.

(8) J. R. Mester et al., “In vivo neurovascular response to focused photoactivation of Channelrhodopsin-2,” NeuroImage, vol. 192, pp. 135–144, May 2019, doi: 10.1016/j.neuroimage.2019.01.036.

(9) T. C. Lee, R. L. Kashyap, and C. N. Chu, “Building Skeleton Models via 3-D Medial Surface Axis Thinning Algorithms,” CVGIP Graph. Models Image Process., vol. 56, no. 6, pp. 462–478, Nov. 1994, doi: 10.1006/cgip.1994.1042.

(10) M. Y. Rennie et al., “Vessel tortuousity and reduced vascularization in the fetoplacental arterial tree after maternal exposure to polycyclic aromatic hydrocarbons,” Am. J. Physiol.-Heart Circ. Physiol., vol. 300, no. 2, pp. H675–H684, Feb. 2011, doi: 10.1152/ajpheart.00510.2010.

(11) J. Steinman, M. M. Koletar, B. Stefanovic, and J. G. Sled, “3D morphological analysis of the mouse cerebral vasculature: Comparison of in vivo and ex vivo methods,” PLOS ONE, vol. 12, no. 10, p. e0186676, Oct. 2017, doi: 10.1371/journal.pone.0186676.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Summary:

In this work, the authors present a cornucopia of data generated using deep mutational scanning (DMS) of variants in MET kinase, a protein target implicated in many different forms of cancer. The authors conducted a heroic amount of deep mutational scanning, using computational structural models to augment the interpretation of their DMS findings.

Strengths:

This powerful combination of computational models, experimental structures in the literature, dose-response curves, and DMS enables them to identify resistance and sensitizing mutations in the MET kinase domain, as well as consider inhibitors in the context of the clinically relevant exon-14 deletion. They then try to use the existing language model ESM1b augmented by an XGBoost regressor to identify key biophysical drivers of fitness. The authors provide an incredible study that has a treasure trove of data on a clinically relevant target that will appeal to many.

We thank Reviewer 1 for their generous assessment of our manuscript!

Weaknesses:

However, the authors do not equally consider alternative possible mechanisms of resistance or sensitivity beyond the impact of mutation on binding, even though the measure used to discuss resistance and sensitivity is ultimately a resistance score derived from the increase or decrease of the presence of a variant during cell growth.

For this resistance screen, Ba/F3 was a carefully chosen cellular selection system due to its addiction to exogenously provided IL-3, undetected expression of endogenous RTKs (including MET), and dependence on kinase transgenes to promote signaling and growth under IL-3 withdrawal. Together this allows for the readout of variants that alter kinase-driven proliferation without the caveat of bypass resistance. In our previous phenotypic screen (Estevam et al., 2024, eLife), we also carefully examined the impact of all possible MET kinase domain mutations both in the presence and absence of IL-3 withdrawal, but no inhibitors. There, we identified a small group of mutations that were associated with gain-of-function behavior located at conserved regulatory motifs outside of the catalytic site, yet these mutations were largely sensitive to inhibitors within this screen.

Here, the majority of resistance mutations were located at or near the ATP-binding pocket, suggesting an impact on resistance through direct drug interactions. However, there was also a small population of distal mutations that met our statistical definitions of resistance. Within the crizotinib selection, sites such as T1293, L1272, T1261, amongst others, demonstrated resistance profiles but were located in C-lobe away from the catalytic site. While we did not experimentally validate these specific mutations, it is possible that non-direct drug binders instead promote resistance through allosteric or conformational mechanisms which preserve kinase activity and signaling. Indeed, our ML framework explicitly included conformational and stability effects as significant in improving predictions.

We would be happy to further discuss any specific alternative resistance mechanisms Reviewer 1 has in mind! Thank you for highlighting this!

There are also points of discussion and interpretation that rely heavily on docked models of kinase-inhibitor pairs without considering alternative binding modes or providing any validation of the docked pose. Lastly, the use of ESM1b is powerful but constrained heavily by the limited structural training data provided, which can lead to misleading interpretations without considering alternative conformations or poses.

The majority of our interpretations are grounded in the X-ray structures of WT MET bound to the inhibitors studied (or close analogs). The use of docked models (note - to mutant structures predicted by UMol, not ESM, that can have conformational changes) is primarily in the ML part of the manuscript. Indeed, in our models, conformational and binding mode changes are taken into account as features (see Ligand RMSD, Residue RMSD). There are certainly improved methods (AF3 variants) emerging that might have even more power to model these changes, but they come with greater computational costs and are something we will be evaluating in the future.

We added to the results section: “While our features can account for some changes in MET-mutant conformation and altered inhibitor binding pose, the prediction of these aspects can likely be improved with new methods.”

Reviewer #2 (Public review):

Summary:

This manuscript provides a comprehensive overview of potential resistance mutations within MET Receptor Tyrosine Kinase and defines how specific mutations affect different inhibitors and modes of target engagement. The goal is to identify inhibitor combinations with the lowest overlap in their sensitivity to resistant mutations and determine if certain resistance mutations/mechanisms are more prevalent for specific modes of ATP-binding site engagement. To achieve this, the authors measured the ability of ~6000 single mutants of MET's kinase domain (in the context of a cytosolic TPR fusion) to drive IL-3-independent proliferation (used as a proxy for activity) of Ba/F3 cells (deep mutational profiling) in the presence of 11 different inhibitors. The authors then used co-crystal and docked structures of inhibitor-bound MET complexes to define the mechanistic basis of resistance and applied a protein language model to develop a predictive model of inhibitor sensitivity/resistance.

Strengths:

The major strengths of this manuscript are the comprehensive nature of the study and the rigorous methods used to measure the sensitivity of ~6000 MET mutants in a pooled format. The dataset generated will be a valuable resource for researchers interested in understanding kinase inhibitor sensitivity and, more broadly, small molecule ligand/protein interactions. The structural analyses are systematic and comprehensive, providing interesting insights into resistance mechanisms. Furthermore, the use of machine learning to define inhibitor-specific fitness landscapes is a valuable addition to the narrative. Although the ESM1b protein language model is only moderately successful in identifying the underlying mechanistic basis of resistance, the authors' attempt to integrate systematic sequence/function datasets with machine learning serves as a foundation for future efforts.

We thank Reviewer 2 for their thoughtful assessment of our manuscript!

Weaknesses:

The main limitation of this study is that the authors' efforts to define general mechanisms between inhibitor classes were only moderately successful due to the challenge of uncoupling inhibitor-specific interaction effects from more general mechanisms related to the mode of ATP-binding site engagement. However, this is a minor limitation that only minimally detracts from the impressive overall scope of the study.

We agree. We have added to the discussion: “A full landscape of mutational effects can help to predict drug response and guide small molecule design to counteract acquired resistance. The ability to define molecular mechanisms towards that goal will likely require more purposefully chosen chemical inhibitors and combinatorial mutational libraries to be maximally informative.”

Reviewer #3 (Public review):

Summary:

In the manuscript 'Mapping kinase domain resistance mechanisms for the MET receptor tyrosine kinase via deep mutational scanning' by Estevam et al, deep mutational scanning is used to assess the impact of ~5,764 mutants in the MET kinase domain on the binding of 11 inhibitors. Analyses were divided by individual inhibitor and kinase inhibitor subtypes (I, II, I 1/2, and III). While a number of mutants were consistent with previous clinical reports, novel potential resistance mutants were also described. This study has implications for the development of combination therapies, namely which combination of inhibitors to avoid based on overlapping resistance mutant profiles. While one suggested pair of inhibitors with the least overlapping resistance mutation profiles was suggested, this manuscript presents a proof of concept toward a more systematic approach for improved selection of combination therapeutics. Furthermore, in a final part of this manuscript the data was used to train a machine learning model, the ESM-1b protein language model augmented with an XG Boost Regressor framework, and found that they could improve predictions of resistance mutations above the initial ESM-1b model.

Strengths:

Overall this paper is a tour-de-force of data collection and analysis to establish a more systematic approach for the design of combination therapies, especially in targeting MET and other kinases, a family of proteins significant to therapeutic intervention for a variety of diseases. The presentation of the work is mostly concise and clear with thousands of data points presented neatly and clearly. The discovery of novel resistance mutants for individual MET inhibitors, kinase inhibitor subtypes within the context of MET, and all resistance mutants across inhibitor subtypes for MET has clinical relevance. However, probably the most promising outcome of this paper is the proposal of the inhibitor combination of Crizotinib and Cabozantib as Type I and Type II inhibitors, respectively, with the least overlapping resistance mutation profiles and therefore potentially the most successful combination therapy for MET. While this specific combination is not necessarily the point, it illustrates a compelling systematic approach for deciding how to proceed in developing combination therapy schedules for kinases. In an insightful final section of this paper, the authors approach using their data to train a machine learning model, perhaps understanding that performing these experiments for every kinase for every inhibitor could be prohibitive to applying this method in practice.

We thank Reviewer 3 for their assessment of our manuscript (we are very happy to have it described as a tour-de-force!)

Weaknesses:

This paper presents a clear set of experiments with a compelling justification. The content of the paper is overall of high quality. Below are mostly regarding clarifications in presentation.

Two places could use more computational experiments and analysis, however. Both are presented as suggestions, but at least a discussion of these topics would improve the overall relevance of this work. In the first case it seems that while the analyses conducted on this dataset were chosen with care to be the most relevant to human health, further analyses of these results and their implications of our understanding of allosteric interactions and their effects on inhibitor binding would be a relevant addition. For example, for any given residue type found to be a resistance mutant are there consistent amino acid mutations to which a large or small or effect is found. For example is a mutation from alanine to phenylalanine always deleterious, though one can assume the exact location of a residue matters significantly. Some of this analysis is done in dividing resistance mutants by those that are near the inhibitor binding site and those that aren't, but more of these types of analyses could help the reader understand the large amount of data presented here. A mention at least of the existing literature in this area and the lack or presence of trends would be worthwhile. For example, is there any correlation with a simpler metric like the Grantham score to predict effects of mutations (in a way the ESM-1b model is a better version of this, so this is somewhat implicitly discussed).

Indeed we experimented with including these types of features in the XGBoost scheme (particularly residue volume change and distance) to augment the predictive power of the ESM model - see Figure 8 - figure supplement 1; however, we didn’t find them as significant. Therefore, the signal is likely very small and/or incorporated into the baseline ESM model.

Indeed, this discussion relates to the second point this manuscript could improve upon: the machine learning section. The main actionable item here is that this results section seems the least polished and could do a better job describing what was done. In the figure it looks like results for certain inhibitors were held out as test data - was this all mutants for a single inhibitor, or some other scheme? Overall I think the implications of this section could be fleshed out, potentially with more experiments.

Figure 8A and the methods section contain a very detailed explanation of test data. We have thought about it and do not have any easy path to improve the description, which we reproduce here:

“Experimental fitness scores of MET variants in the presence of DMSO and AMG458 were ignored in model training and testing since having just one set of data for a type I ½ inhibitor and DMSO leads to learning by simply memorizing the inhibitor type, without generalizability. The remaining dataset was split into training and test sets to further avoid overfitting (Figure 8A). The following data points were held out for testing - (a) all mutations in the presence of one type I (crizotinib) and one type II (glesatinib analog) inhibitor, (b) 20% of randomly chosen positions (columns) and (c) all mutations in two randomly selected amino acids (rows) (e.g. all mutations to Phe, Ser). After splitting the dataset into train and test sets, the train set was used for XGBoost hyperparameter tuning and cross-validation. For tuning the hyperparameters of each of the XGBoost models, we held out 20% of randomly sampled data points in the training set and used the remaining 80% data for Bayesian hyperparameter optimization of the models with Optuna (Akiba et al., 2019), with an objective to minimize the mean squared error between the fitness predictions on 20% held out split and the corresponding experimental fitness scores. The following hyperparameters were sampled and tuned: type of booster (booster - gbtree or dart), maximum tree depth (max_depth), number of trees (n_estimators), learning rate (eta), minimum leaf split loss (gamma), subsample ratio of columns when constructing each tree (colsample_bytree), L1 and L2 regularization terms (alpha and beta) and tree growth policy (grow_policy - depthwise or lossguide). After identifying the best combination of hyperparameters for each of the models, we performed 10-fold cross validation (with re-sampling) of the models on the full training set. The training set consists of data points corresponding to 230 positions and 18 amino acids. We split these into 10 parts such that each part corresponds to data from 23 positions and 2 amino acids. Then, at each of 10 iterations of cross-validation, models were trained on 9 of 10 parts (207 positions and 16 amino acids) and evaluated on the 1 held out part (23 positions and 2 amino acids). Through this protocol we ensure that we evaluate performance of the models with different subsets of positions and amino acids. The average Pearson correlation and mean squared error of the models from these 10 iterations were calculated and the best performing model out of 8192 models was chosen as the one with the highest cross-validation correlation. The final XGBoost models were obtained by training on the full training set and also used to obtain the fitness score predictions for the validation and test sets. These predictions were used to calculate the inhibitor-wise correlations shown in Figure 8B.“

As mentioned in the 'Strengths' section, one of the appealing aspects of this paper is indeed its potential wide applicability across kinases -- could you use this ML model to predict resistance mutants for an entirely different kinase? This doesn't seem far-fetched, and would be an extremely compelling addition to this paper to prove the value of this approach.

This is exactly where we want to go next! But as we see here, it is going to be hard and require more purposeful selection of chemicals and likely combinatorial mutations to be maximally informative (see also reviewer 2 response where we have added text)

Another area in which this paper could improve its clarity is in the description of caveats of the assay. The exact math used to define resistance mutants and its dependence on the DMSO control is interesting, it is worth discussing where the failure modes of this procedure might be. Could it be that the resistance mutants identified in this assay would differ significantly from those found in patients? That results here are consistent with those seen in the clinic is promising, but discrepancies could remain.

Thank you for pointing this out. The greatest trade-off of probing the intracellular MET kinase (juxtamembrane, kinase domain, c-tail) in the constitutively active TPR system is that while we gain cytoplasmic expression, constitutive oligomerization, and HGF-independent activation, other features like membrane-proximal effects are lost and translatability of some mutations in non-proliferative conditions may also be limited. Nevertheless, Ba/F3 allows IL-3 withdrawal to serve as an effective variant readout of transgenic kinase variant effects due to its undetectable expression of endogenous RTKs and addiction to exogenous interleukin-3 (IL-3).

In our previous study, we were also interested in comparing the phenotypic results to available patient populations in cBioPortal. We observed that our DMS captured known oncogenic MET kinase variants, in addition to a population of gain-of-function variants within clinical residue positions that have not been clinically reported. Interestingly, the population of possible novel gain-of-function mutant codons were more distant in genetic space (2-3 Hamming distance) from wild type than the clinically reported variant codon (1-2 Hamming distance).

For this inhibitor screen, we also carefully compared previously reported and validated resistance mutations across referenced publications to that of our inhibitor screen, and observed large agreement as noted in-text. While discrepancies could definitely remain, there is precedence for consistency.

Furthermore a more in depth discussion of the MetdelEx14 results is warranted. For example, why is the DMSO signature in Figure 1 - supplement 4 so different from that of Figure 1?

In our previous study (Estevam et al., 2024), we more directly compared MET and METΔExon14, and while observed several differences, especially at conserved regulatory motifs, the TPR expression system did not provide a robust differential. Therefore, we hypothesize that a membrane-bound context is likely necessary to obtain a differential that captures juxtamembrane regulatory effects for these two isoforms. For that reason, we did not place heavy emphasis on the differences between MET and METΔExon14 in this study. Nevertheless, we performed parallel analysis of the METΔExon14 inhibitor DMS and provided all source and analyzed data in our GitHub repository (https://github.com/fraser-lab/MET_kinase_Inhibitor_DMS).

In our analysis of resistance, we used Rosace to score and compare DMSO and inhibitor landscapes. We present the full distribution of raw scores in Figure 1 for each condition. However, to visually highlight resistance mutations as a heatmap, we subtracted the scores of each variant in each inhibitor condition from the raw DMSO score, making the heatmaps in Figure 1 - supplement 4 appear more “blue.”

And finally, there is a lot of emphasis put on the unexpected results of this assay for the tivantinib "type III" inhibitor - could this in fact be because the molecule "is highly selective for the inactive or unphosphorylated form of c-Met" according to Eathiraj et al JBC 2011?

The work presented by Eathiraj et al JBC 2011 is a key study we reference and is foundational to tivantinib. While the point brought up about tivantinib’s selective preference for an inactive conformation is valid, this is also true for type II kinase inhibitors. In our study, regardless of inhibitor conformational preference, tivantinib was the only one with a nearly identical landscape to DMSO and exhibited selection even in the absence of Ba/F3 MET-addiction (Figure 1E). This result is in closer agreement with MET agnostic behavior reported by Basilico et al., 2013 and Katayama et al., 2013.

While this paper is crisply written with beautiful figures, the complexity of the data warrants a bit more clarity in how the results are visualized. Namely, clearly highlighting mutants that have previously reported and those identified by this study across all figures could help significantly in understanding the more novel findings of the work.

To better compare and contrast novel mutation identified in this study to others, we compiled a list of reported resistance mutations from recent clinical and experimental studies (Pecci et al 2024; Yao et al., 2023; Bahcall et al., 2022; Recondo et al., 2020; Rotow et al ., 2020; Fujino et al., 2019), since a direct database with resistance annotations does not exist for MET, to the best of our knowledge. In total, this amounted to 31 annotated resistance mutations across crizotinib, capmatinib, tepotinib, savolitinib, cabozantinib, merestinib, and glesatinib, which we have now tabulated in a new figure (Figure 4) and commentary in the main text:

To assess the agreement between our DMS and previously annotated resistance mutations, we compiled a list of reported resistance mutations from recent clinical and experimental studies (Pecci et al 2024; Yao et al., 2023; Bahcall et al., 2022; Recondo et al., 2020; Rotow et al ., 2020; Fujino et al., 2019) (Figure 4A,B). Overall, previously discovered mutations are strongly shifted to a GOF distribution for the drugs where resistance is reported from treatment or experiment; in contrast, the distribution is centered around neutral for those sites for other drugs not reported in the literature (Figure 4C). However, even in cases such as L1195V, we observe GOF DMS scores indicative of resistance to previously reported inhibitors. Given this overall strong concordance with prior literature and clinical results, we can also provide hypotheses to clarify the role of mutations that are observed in combination with others. For example, H1094Y is a reported driver mutation that has been linked to resistance in METΔEx14 for glesatinib with either the secondary L1195V mutation or in isolation (Recodo et al., 2020). However, in our assay H1094Y demonstrated slight sensitivity to gelesatinib, suggesting that either resistance is linked to the exon14 deletion isoform, the L1195V mutation, or a cellular factor not modeled well by the BaF3 system.

Finally, the potential impacts and follow-ups of this excellent study could be communicated better - it is recommended that they advertise better this paper as a resource for the community both as a dataset and as a proof of concept. In this realm I would encourage the authors to emphasize the multiple potential uses of this dataset by others to provide answers and insights on a variety of problems.

Please see below

Related to this, the decision to include the MetdelEx14 results, but not discuss them at all is interesting, do the authors expect future analyses to lead to useful insights? Is it surprising that trends are broadly the same to the data discussed?

Our previous paper suggests that Ba/F3 isn’t a great model for measuring the differences between MET and METΔEx14, so we haven’t emphasized other than to point to our previous paper. We include the full analysis here nonetheless as a resource. Potentially where the greatest differences between resistance mutant behaviors would be observed is in the full-length, membrane-bound MET and METΔEx14 receptor isoforms. While outside of the scope of this study, there is great potential to use the resistance mutations identified in this study as a filtered group to test and map differential inhibitor sensitivities between receptor isoforms.

And finally it could be valuable to have a small addition of introspection from the authors on how this approach could be altered and/or improved in the future to facilitate the general application of this approach for combination therapies for other targets.

See also reviewer 2 response where we have added text.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

Major points of revision:

(1) It seems like much of the structural interpretation of the inhibitor binding mode, outside of crizotinib binding, appears to come from docked models of the inhibitor to the MET kinase domain. Given the potential variability of the docked structure to the kinase domain, it would be useful for the authors to consider alternative possible binding modes that their docking pipeline may have suggested. It could also be useful to provide some degree of validation or contextualization of their docking models.

All individual figures are very carefully inspected based on either existing crystal structures of the inhibitor or closely related inhibitors (ATP, 3DKC; crizotinib, 2WGJ; tepotinib, 4R1V; tivantinib, 3RHK; AMG-458, 5T3Q; NVP-BVU972, 3QTI; merestinib, 4EEV; savolitinib, 6SDE). In total, four structural interpretations were the result of docking onto reference experimental structures (capmatinib, cabozantinib, glumetinib, glesatinib). As we wrote above, different conformations and binding modes are possible in predicted mutant structures (as we did here at scale) and included in the ML analysis already.

(2) In the first section, the authors classify an inhibitor as Type Ia on docking models, but mention the conflicting literature describing it as type Ib - it would be helpful to provide a contextualization of why this distinction between Ia and Ib matters, and what difference it might make. It would also be useful to know if their docking score only suggested poses compatible with Ia or if other poses were provided as well. Validation using other method might be beneficial, especially since they acknowledge the conflicting literature for classification. Or at least recontextualization that more evidence would be needed.

Kinase inhibitors have several canonical structural definitions we use to base the classifications in this study. Specifically, type I inhibitors are classified in MET by interactions with Y1230, D1228, K1110 in addition to its conformation in the ATP-binding site. Type I inhibitors are further subdivided into type 1a in MET if it leverages interactions with the solvent front and residue G1163. In prior literature referenced, tepotinib was classified as type 1b, which would imply it does not have solvent front interactions, like savolitinib (PDB 6SDE) or NVP-BVU972 (PDB 3QTI). However, in the tepotinib experimental structure (PDB 4R1V), we observed a greater structural resemblance to other type 1a inhibitors opposed to type 1b (Figure 1 - figure supplement 1b).

(3) The measure used to discuss resistance and sensitivity is ultimately a resistance score derived from the increase or decrease of the presence of a variant during cell growth. This is not a measure of direct binding. It would be helpful if the authors discussed alternative mechanisms through which these variants may impact resistance and/or sensitivity, such as stability, protonation effects, or kinase activity. The score itself may be convolving over all these potential mechanisms to drive GOF and LOF observed behavior.

See the response to the public review. Indeed, our ML framework explicitly included conformational and stability effects as significant in improving predictions.

(4) While it is promising to try and improve the predictive properties of ESM1b, it is not exactly clear why the authors considered their structural data of 11 inhibitors a sufficient dataset with which to augment the model. It would be useful for the authors to provide some additional context for why they wished to augment ESM1b in particular with their dataset, and provide any metrics indicating that their training data of 11 inhibitors provided an adequate statistical sample.

We don’t understand what this means. Sorry!

(5) The authors use ESM-1b to predict the fitness impact of each mutation and augment it using protein structural data of drug-target interactions. However, using an XGBoost regressor on a single set of 11 kinase-inhibitor interaction pairs is an incredibly sparse dataset to train upon. It would be useful for the authors to consider the limitations of their model, as well as its extensibility in the context of alternate binding poses, alternate conformations, or changes in protonation states of ligand or inhibitor.

On the contrary - this is 11 chemicals across 3000 mutations. We have discussed alternative interpretations above.

Minor points:

(1) It would also be useful for the authors to provide more context around their choice of regressor. XGBoost is a powerful regressor but can easily overfit high dimensional data when paired with language models such as ESM-1b. This would be particularly useful since some of the features to train on were also generated using existing models such as ThermoMPNN.

Yes - we are quite concerned about overfitting and have tried to assess overfitting by careful design of test and validation sets.

(2) The authors also mention excluding their DMSO and AMG458 scores in the model training and testing due to overfitting issues - it would be useful to have an SI figure pointing to this data.

No - we exclude the DMSO because that is the reference (baseline) and AMG because it has a different binding mode. This isn’t related to overfitting.

(3) The authors mention in their docking pipeline that 5 binding modes were used for each ligand docking, but it appears that only one binding mode is considered in the main figures. It would be useful for the authors to provide additional details about what were the other binding modes used for, how different were each binding mode, and how was the "primary" mode selected (and how much better was its score than the others).

The reviewer misinterprets the difference between poses shown in figures, based on mostly crystal structures or carefully selected templates, and the use of docked models in feature engineering for the ML part of the study. Where existing crystal structures do not exist, we performed docking for capmatinib, cabozantinib, glumetinib, glesatinib onto reference structures bound to type I (2WGJ) and type II (4EEV) inhibitors. We selected one representative binding mode based on the reference inhibitor, and while not exact, at a minimum these models provide a basis for structural interpretation.

Reviewer #2 (Recommendations for the authors):

My main suggestion is for the authors to add a few sentences (in non-technical language) to the results section, specifically before the results shown in Figure 3, defining gain-of-function, loss-of-function, resistance, and sensitivity. While these definitions are present in the materials and methods section, explicitly discussing them prior to the relevant results would significantly improve the overall readability of the manuscript.

We defined “gain-of-function” and “loss-of-function” mutations as those with fitness scores statistically greater or lower than wild-type. Within the DMSO condition, gain-of-function and loss-of -function labels describe mutational perturbation to protein function, whereas within inhibitor conditions, the labels describe the difference in fitness introduced by an inhibitor.

We have also clarified these definitions where the terms are first introduced: “As expected, the DMSO control population displayed a bimodal distribution with mutations exhibiting wild-type fitness centered around 0, with a wider distribution of mutations that exhibited loss- or gain-of-function effects, as defined by fitness scores with statistically significant lower or greater scores than wild-type, respectively.”

Figure 7D. Please add a bit more detail to the legend on how fold change (y-axis) was calculated.

Here, fold change represents the number of viable cells at each inhibitor concentration relative to the TKI control, measured with the CellTiter-Glo® Luminescent Cell Viability Assay (Promega) as an end point readout. We have updated the legend of Figure 7D with calculation details: “Dose-response for each inhibitor concentration is represented as the fraction of viable cells relative to the TKI free control.”

I must admit, I did not understand what "Specific inhibitor fitness landscapes also aid in identifying mutations with potential drug sensitivity, such as R1086 and C1091 in the MET P-loop" means. These are positions where most mutations lead to greater sensitivity to crizotinib. Is the idea that there are potentially clinically-relevant MET mutations that can be targeted over wild type with crizotinib?

Thank you for highlighting this! The P-loop (phosphate-binding loop) is a glycine-rich structural motif conserved in kinase domains. This motif is located in the N-lobe, where its primary role is to gate ATP entry into the active site and stabilize the phosphate groups of ATP when bound. Therefore, the P-loop is a common target region for ATP-competitive inhibitor design, but also a site where resistance can emerge (Roumiantsev et al., 2002). The idea we’d like to convey is that identifying residues that offer the potential for drug stabilization with the added benefit of having lower risk resistance, is an attractive consideration for novel inhibitor design.

We have added to the text: “Individual inhibitor resistance landscapes also aid in identifying target residues for novel drug design by providing insights into mutability and known resistance cases. This enables the selection of vectors for chemical elaboration with potential lower risk of resistance development. Sites with mutational profiles such as R1086 and C1091, located in the common drug target P-loop of MET, could be likely candidates for crizotinib.”

Reviewer #3 (Recommendations for the authors):

(1) Suggested Improvements to the Figures:

a)  Figure 4A - T1261 seems to be mislabeled

b)  In Figure 3A it's suggested to highlight mutants determined to be resistance mutants by this scheme.

c)  In Figure 3D it would be informative to highlight which of these resistance mutants have already been previously reported and which are novel to this study

d)  Throughout figures 3A, 3D, and 4G the graphical choices on how to highlight synonymous mutations and mutations not performed in the assay needs improvement.

The Green vs Grey 'TRUE' vs 'FALSE' boxes are confusing. Just a green box indicating synonymous mutations would be sufficient. Additionally these green boxes are hard to see, and often edges of this green box are currently missing making it even more difficult to see and interpret.

* In Figure 4A mutants do not seem to be indicated by a line or plus sign, but this is not explained in the legend or the caption. Please add.

* In 3D and 4G it is not clear if the mutants not performed are indicated at all - perhaps they are indicated in white, making them indistinguishable from scores with 0. Please clarify.

T1261 and G1242 are now correctly labeled.

In text we have also highlighted reported resistance mutations for crizotinib, which are inclusive of clinical reports and in vitro characterization: “These sites, and many of the individual mutations, have been noted in prior reports, such as: D1228N/H/V/Y, Y1230C/H/N/S, G1163R.”

We have adjusted the heatmaps to improve visual clarity. Mutations with score 0 are white, as indicated by the scale bar, and mutations uncaptured by the screen are now in light yellow. The green outline distinguishing WT synonymous mutations have also been adjusted so edges are no longer cut off. In our representations, we only distinguished mutations by the score color scale bar and WT outline. What looked like a “plus” or “line” in the original figure was only the heatmap background, which now should be resolved in the updated figure and legends for Figure 3 and Figure 4.

(2) Some Minor Suggested Improvements to the Text:

a)  The abbreviation CBL for 'CBL docking site' is used without being defined.

b)  Figure 3G is referenced, but it does not exist.

c)  In the sentence 'Beyond these well characterized sites, regions with sensitivity occurred throughout the kinase, primarily in loop-regions which have the greatest mutational tolerance in DMSO, but do not provide a growth advantage in the presence of an inhibitor (Figure 1 - Figure Supplement 1; Figure 1 - Figure Supplement 2).'. It is not clear why these supplemental figures are being referenced.

d)  In the supplement section 'Enrich2 Scoring' has what seem like placeholders for citations in [brackets]

Cbl is a E3 ubiquitin ligase that plays a role in MET regulation through engagement with exon 14, specifically at Y1003 when phosphorylated. This mode of regulation was more highlighted in our previous study. However, since Cbl was only mentioned briefly in this study, we have removed reference to it to simplify the text.

In addition, we have removed the figure 3G reference and corrected the in-text range. We have also removed references to figure supplements where unnecessary and edited the “Enrich2 scoring” method section to now reference missing citations.

battery tips,

This sentence shows that categorizing change methods can help reduce things but it could also ignore critical aspects. It's just the first step to truly understand the steps. We need to look at them in greater depth and context.

I feel this very important. It's easy to disregard someone's work just because they made a mistake. It's important though to remember that all humans make mistakes even simple ones. The right thing to do would be to politely correct them but still acknowledge their work or message.

Often, a major event or even a series of smaller ones pushes young people to see how racism personally affects them. These moments aren’t just eye opening; they change how teens view themselves and their role in the world. For Black adolescents, it’s a tough but transformative step in making sense of their racial identity.

Author response:

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

Reviewer #1 (Public review): 

Summary: 

This manuscript reports the substrate-bound structure of SiaQM from F. nucleatum, which is the membrane component of a Neu5Ac-specific Tripartite ATP-dependent Periplasmic (TRAP) transporter. Until recently, there was no experimentally derived structural information regarding the membrane components of TRAP transporter, limiting our understanding of the transport mechanism. Since 2022, there have been 3 different studies reporting the structures of the membrane components of Neu5Ac-specific TRAP transporters. While it was possible to narrow down the binding site location by comparing the structures to proteins of the same fold, a structure with substrate bound has been missing. In this work, the authors report the Na+-bound state and the Na+ plus Neu5Ac state of FnSiaQM, revealing information regarding substrate coordination. In previous studies, 2 Na+ ion sites were identified. Here, the authors also tentatively assign a 3rd Na+ site. The authors reconstitute the transporter to assess the effects of mutating the binding site residues they identified in their structures. Of the 2 positions tested, only one of them appears to be critical to substrate binding.

Strengths: 

The main strength of this work is the capture of the substrate bound state of SiaQM, which provides insight into an important part of the transport cycle.

Weaknesses: 

The main weakness is the lack of experimental validation of the structural findings. The authors identified the Neu5Ac binding site, but only test 2 residues for their involvement in substrate interactions, which is quite limited. However, comparison with previous mutagenesis studies on homologues supports the location of the Neu5Ac binding site. The authors tentatively identified a 3rd Na+ binding site, which if true would be an impactful finding, but this site was not sufficiently experimentally tested for its contribution to Na+ dependent transport. This lack of experimental validation prevents the authors from unequivocally assigning this site as a Na+ binding site. However, the reporting of these new data is important as it will facilitate follow up studies by the authors or other researchers. 

Comments on revisions: 

Overall, the authors have done a good job of addressing the reviewers' comments. It's good to know that the authors are working on the characterisation of the potential metal binding site mutants - characterizing just a few of these will provide much-needed experimental support for this potential Na+ site. 

The new MD simulations provide additional support for the new Na+ site and could be included.

However, as the authors know, direct experimental characterisation of mutants is the ideal evidence of the Na+ site.

Aside from the characterisation of mutants, which seems to be held up by technical issues, the only remaining issue is the comparison of the Na+- and Na+/Neu5Ac-bound states with ASCT2. It still does not make sense to me why the authors are not directly comparing their Na+ only and Na+/Neu5Ac states with the structures of VcINDY in the Na+-only and Na+/succinate bound states. These VcINDY structures also revealed no conformational changes in the HP loops upon binding succinate, as the authors see for SiaQM. Therefore, this comparison is very supportive. It is understood that the similarity to the DASS structure is mentioned on p.17, but it is also interesting and useful to note that TRAP and DASS transporters also share a lack of substrateinduced local conformational changes, to the extent these things have been measured.

We acknowledge the summary weakness that experimental data to support the third Na binding site is critical.

Based on the reviewer’s suggestion, we added the following in the main text and a supplementary figure comparing the Na ion binding sites between VcINDY and SiaQM. Page 13.

“These two sodium ion binding sites are also conserved in the structure of VcINDY (Supplementary Figure 7) (Sauer et al., 2022). In both cases, the sodium ions are bound at the helix-loop-helix ends of HP1 and HP2. The binding sites utilize both side chains and main chain carbonyl groups. The number of main chain carbonyl interactions suggests that they are critical, and using main chain rather than side chain interactions minimizes the likelihood of point mutations affecting the binding.”

Reviewer #3 (Public review): 

The manuscript by Goyal et al report substrate-bound and substrate-free structures of a tripartite ATP independent periplasmic (TRAP) transporter from a previously uncharacterized homolog, F. nucleatum. This is one of most mechanistically fascinating transporter families, by means of its QM domain (the domain reported in his manuscript) operating as a monomeric 'elevator', and its P domain functioning as a substrate-binding 'operator' that is required to deliver the substrate to the QM domain; together, this is termed an 'elevator with an operator' mechanism.

Remarkably, previous structures had not demonstrated the substrate Neu5Ac bound. In addition, they confirm the previously reported Na+ binding sites, and report a new metal binding site in the transporter, which seems to be mechanistically relevant. Finally, they mutate the substrate binding site and use proteoliposomal uptake assays to show the mechanistic relevance of the proposed substrate binding residues.

Strengths: 

The structures are of good quality, the presentation of the structural data has improved, the functional data is robust, the text is well-written, and the authors are appropriately careful with their interpretations. Determination of a substrate bound structure is an important achievement and fills an important gap in the 'elevator with an operator' mechanism.

Weaknesses: 

Although the possibility of the third metal site is compelling, I do not feel it is appropriate to model in a publicly deposited PDB structure without directly confirming experimentally. The authors do not extensively test the binding sites due to technical limitations of producing relevant mutants; however, their model is consistent with genetic assays of previously characterized orthologs, which will be of benefit to the field. Finally, some clarifications of EM processing would be useful to readers, and it would be nice to have a figure visualizing the unmodeled lipid densities - this would be important to contextualize to their proposed mechanism.

Reviewer #3 (Recommendations for the authors): 

I appreciate the authors' responses to our critiques; the revised manuscript is much improved and has addressed most of my concerns. I look forward to seeing their follow up experiments testing mutational e=ects. I think MD simulations of ion-binding sites on their own are supportive but by themselves not su=icient to prove the existence of a functional Na+-binding site. Some clarifications in the methods/supplements would satisfy my concerns about data processing and analysis.

- Unliganded map: were the 141,272 particles used for one class of ab initio? This is unusual, usually multiple ab initio classes are used to further eliminate junk particles. The authors themselves use 6 classes for the substrate-bound dataset.

We classified the particles into multiple 3-D classes.  There was no improvement in statistics or maps on splitting these further.  Hence, we did not pursue that further. 

- Substrate-bound map: how did the four 'identical' classes independently refine? Are similar Na+/substate densities found in each separate class?

The other classes refined to worse than 4.5 Å resolution. We stopped characterizing them past that point.  We were hoping to see multiple conformations that are diLerent – and hopefully a class where only two sodium ions could be bound.  However, any interpretation at 4.5 Å would be unreliable.

- Both maps: all ab initio classes prior to final refinement should be displayed in the supplementary workflow, this is common for EM processing diagrams.

We agree it is common – however, unless there is a good reason to discuss the other classes, we are not convinced of the value of crowding the figures.

- What specific refinement package and version of Phenix are the authors using? It seems unusual that it is not possible to refine without a metal in Phenix real-space refinement, I have seen many structures where there is no issue refining without critical ions/waters. The authors should double check that they are using the appropriate scattering table for cryo-EM, which should be "electron".

Sorry for the confusion – we did not mean to say we cannot refine without a metal. If we want to add something to the density, we cannot refine it without suggesting a metal or solvent.  The site without anything added will refine without any issues but in the absence of additional verification, we cannot be sure of the identity of the ions. We are confident of the metal binding site – but not confident of the exact metal bound.  We used Sodium as our first hypothesis.

We don’t think the scattering factors will help in the identification of the ions. Servalcat as part of CCP-EM can produce diLerence maps and we believe that for identification of ions, it will require higher resolution (<2.5 Å) but at this resolution, we can say that there is a nonprotein density but not more than that. We were using “electron” (which we believe is default with phenix.real_space_refine). The refinement was performed using standard protocols and appropriate scattering factors (Phenix version 1.19x), and we have previously used similar refinement protocols for other maps/models (Example -Vinothkumar KR, Arya CK, Ramanathan G, Subramanian R. 2021. Comparison of CryoEM and X-ray structures of dimethylformamidase. Progress in Biophysics and Molecular Biology, CryoEM microscopy developments and their biological applications 160:66–78. doi:10.1016/j.pbiomolbio.2020.06.008).

To convince the reviewer of the quality of the maps, we have added figures that show the model-to-map fit of all of the main secondary structural elements in both the unliganded and the Neu5Ac bound forms.

- I certainly understand the authors' reluctance to not model the entirety of protein densities; however, I think it would be useful to highlight these densities in the global context of the protein. A common way to show this is to show the density proximal to protein chains in one color, and the remaining densities in a contrasting color (Figure 1 somewhat demonstrates this but it is di=icult to tell). I think this would be a nice figure to show the presence and location of unmodeled densities.

We have modified supplementary figure 3 to include unmodelled densities in panels G and H for both structures.

- Small detail, "uniform" is misspelled as "unifrom" in supplementary Figure 3. 

Thank you.  Corrected.

As previously stated, AI is being accepted into society more and more everyday. However, a lot of people are using AI simply because it's a convenient tool to make our lives easier, not because they understand the technology and want to use it to its full potential. This kind of technology has to be used and regarded in a way that creates a balance of using AI for its intended use and maintaining the integrity of the work we produce.

This is something to be increasingly aware of. It's true that AI and it's use will continue to grow as the technology is further developed. It's also evident that no matter who opposes AI it is being accepted more and more into our everyday lives.

I didn't really think about this. There's been big talk about AI being the future but not really how we're going to get to that point or how it may effect us on the deepest level.