Add files using upload-large-folder tool

peer_reviews/supplementary_0_Peer Review File__71d5acfb69c4bdc017cb86913c9b10418bf02ea552576b7df314d44fc27eb73d/supplementary_0_Peer Review File__71d5acfb69c4bdc017cb86913c9b10418bf02ea552576b7df314d44fc27eb73d.mmd

@@ -0,0 +1,717 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Seismic evidence for uniform crustal accretion along slow- spreading ridges in the equatorial Atlantic Ocean  
+
+![](images/Figure_unknown_0.jpg)
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+Dear authors,  
+
+the manuscript is nicely written and presents a part of recently acquired wide- angle seismic data along an \(\sim 870 \text{km}\) long transect imaging oceanic crust of different ages between \(5^{\circ} \text{N}\) and \(5^{\circ} \text{S}\) . An S- wave analysis adds to the database of the area. P- wave traveltime picks are the same as in the two previous works (44, 45) and the accompanying paper. But the P- wave model was generated newly and used for newly generated S- wave tomography. The authors propose a uniform crustal accretion caused by 2D sheet- like mantle upwelling similar to fast spreading ridges to occur at the slow- spreading equatorial Mid- Atlantic Ridge (MAR) which is different to observations further north and south along the MAR. The figures are presented well and have a high quality. The data analysis and methods are presented in a good standard and good details are given to reproduce the work.  
+
+Based on observations in reasonably old oceanic crust (8- 70 Ma), the authors conclude on the style of accretion during the formation of the crust. Which is also targeted by the title. I'm missing an explanation why we see differences to the northern and southern MAR. The authors state that their findings contradict previous works at the northern and southern MAR. On the other hand, most named works observe young oceanic crust \(< 2 \text{Ma}\) , while the authors of this work present oceanic crust \(>7 \text{Ma}\) . Why do we have uniform crustal accretion along the segments at the equatorial MAR but not in the before mentioned regions north or south of \(5^{\circ}\) . Indeed, the data image old oceanic crust and a direct comparision to today's crustal accretion specifically at these 5 segments would be necessary to conclude on the source of the observed uniform crust.  
+
+The authors name three hypotheses that should explain the reason for this uniform crustal structure. Their discussion is announced, I however can hardly find it completely discussed. Hypothesis 2 has been proposed in previous works from co- authors (45, 56). While hypothesis 1 is widely discussed and even further explanations are given that accretion was uniform during crustal accretion. However, the other two hypotheses are almost not discussed and it is not explained why they should not be valid. I do not see why hypothesis 2 is ruled out (two- stage accretion). What arguments contradict a two stage crustal accretion, during the time of segment formation and during the time the aging crust is passing the adjacent ridge axis?  
+
+In my opinion additional evidence is needed to support the interpretation and conclusions and to rule out the other named hypotheses. While I still find the work interesting and worth publishing after minor changes, I think there are other journals this work would fit better, since it does not present a very novel and clear explanation or extremely novel findings. The dataset is presented in many pieces.  
+
+Detailed points:  
+
+Lines 55- 60: Give references to the both modes.  
+
+Lines 73/74: Give a reference.  
+
+Line 85: Define the area of "equatorial Atlantic Ocean". Is this according to fig. 2 it is \(+ / - 5^{\circ}\)  
+
+Line 118: Make clear that these are converted (PS) or since hydrophones were used for picking PSP converted. What is the conversion horizon? For young crust with no sediments most often it is the seafloor itself.  
+
+Line 134: Does tectonic accretion need a detachment fault? I understood that deachments develop under certain conditions regarding the magmatic budget. Too less or too much magma, no detachment development. Compare to Buck et al., 2005. M=Magmatic Budget; M<=0.5  
+
+Line 143: add Pmp --> "... reflections (Pmp)."  
+
+Lines 161- 163: It looks like you observe a seismic Moho in that segment. What does this
+
+<--- Page Split --->
+
+reflector stand for? A really seismic Moho seperating igneous crust from mantle or a serpentinisation front or any other explanation?  
+
+Line 178: "a slight thinning" <-- quantify slight. Or is this represented by 5.4 km total crustal thickness?  
+
+Line 181: Crust thins by \(\sim 600 \text{m}\) to a total crustal thickness of xx km? Does your total crustal thickness include sediments or not? In line 174 you state a total change in crustal thickness of 200 m. This contradicts line 181 with \(\sim 600 \text{m}\) change in crustal thickness. Lines 227- 231: The authors propose three hypotheses to explain the small variation in crustal thickness in old crust \((>7 \text{Ma})\) . However, in the following sections the discussion of these three hypothesis concentrates on evidences for hypothesis 1 mainly. Hypothesis 2: Stretching here refers to the segment centre after crustal formation? I see this hardly discussed in the following sections. A stretching of the segment centre only can be ruled out as long as magnetic anomalies are ridge parallel. Hypothesis 3: I don't see this discussed in the following sections. Why do you rule out this 2 stage accretion?  
+
+Line 273: Remember that you are not studying the segment during the time of crustal formation (first stage).  
+
+Line 276: Did you look at the mantle bouger anomaly at the recent segment centres of this five segments studied in this work?  
+
+Lines 295- 297: Give a reference for the "thick crust".  
+
+Lines 353- 364: This seems to be a kind of doubling to the lines 328- 331.  
+
+Line 384: \(50\%\) versus \(90\%\) magmatic accretion \(< - >\) You look at very different times of crustal accretion within the 5 segments. You might image magmatically active times of the specific segments.  
+
+Line 394: Why did you change frequency, was the signal for Pn so much improved but Pg and PmP haven't been visible any longer?  
+
+Lines 457 and followings: You observe two times converted S- Waves. Starting as P- Wave at the source, converting at or in the subsurface, travelling as S through the subsurface becoming converted to P at a boundary or seafloor and than being recorded as P- Wave to the hydrophone. Where does the conversion takes place? If at the seafloor, did you take the sedimentary cover into account? It will extremely delay the S- wave arrivals even if only a few centimeter of sediments.  
+
+Line 467: add km to "... 60 to 760 horizontal distances ..."  
+
+Line 468: change to something more clear maybe: "the lower crust has sparse ray coverage, PmP only."  
+
+Lines 476- 477: You do not have Vs there at all.  
+
+Line 486: "50 models" \(\rightarrow\) "50 calculated models" or "50 inverted models"  
+
+Line 512: "...models is 20x3 (Supp..." add "km". Did you also run a checkerboard test with inverted anomalies, to show that the resolution is independent of the polarity of the anomaly?  
+
+Lines 527- 528: "higher resolution of Vs than Vp" <-- Would be expected because of low Vs wave speed and wave length.  
+
+Lines 529- 531: This is a very weak explaination, because you do not allow for Moho depth adjustments. This does not mean that this Vp- Moho is the best possible. By this you do not allow for more variation.  
+
+Fig. 4: Panel b, segment 3N, shows models with a negative vertical velocity gradient which hints to problems in the inversion or how do you explain a negative velocity gradient in your models?  
+
+Suppl. Fig. 5: panel d, I do not see the advantage of presenting the variance of Vp/Vs, since Vp is kept constant it more or less reflects the variance of Vs.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+This manuscript presents a new Vp and Vs model for the oceanic crust across several segments in the equatorial Atlantic. The seismic line spans a large range of crustal ages. The velocity modeling shows that (1) the crust was magmatically accreted, and (2) crustal thickness is roughly constant, in contrast to other slow- spreading crust where crustal thickening is observed toward segment centers. The authors interpret this crustal thickness
+
+<--- Page Split --->
+
+observation as indicating that the crust in this part of the Atlantic accreted magmatically via 2D sheet- like mantle upwelling rather than the 3D plume- like upwelling generally hypothesized for slow- spreading ridges.  
+
+This is an interesting result, and is well presented in the manuscript. It has important implications for how we conceptualize crustal accretion and mantle upwelling at ridges. I do have some questions and comments, which are listed below. The main points that I think need addressing are these: (1) the resolution tests could do a better job of convincing readers that the uniform crustal thickness is a robust result; and (2) the manuscript makes some statements about the implications of the study that seem to equate magmatic upwelling at ridges with 2D sheet- like upwelling, although elsewhere the existence of 3D plume- like upwelling at some slow- spreading segments is discussed. I suspect the first issue is only because of the choice of testing parameters, and the second can be addressed by refining some of those over- broad statements.  
+
+I understand the desire not to speculate too much without solid evidence, but I would be interested to read more of the authors' thoughts on why magmatic crust in the equatorial Atlantic is apparently accreting differently than elsewhere along the MAR. Is it the presence of extra- long transforms and their stabilizing effects on ridge thermal structure that make this 2D upwelling possible in a slow- spreading regime? This is suggested in the discussion, but the abstract and final conclusions focus on the potential to map out 2D upwelling regions rather than ideas as to why this upwelling scenario exists. Discussion of the role of these long transforms seems especially relevant in the context of the related manuscript on the velocity structure beneath Romanche.  
+
+## Specific comments:  
+
+L18: I am not sure that "contradicting [a] previous hypothesis" is accurate here. I assume that the "previous hypothesis" is supposed to be that 2D sheet- like upwelling cannot occur at slow- spreading ridges, though this isn't explicitly stated. Either that previous hypothesis should be more clearly defined (if it is actually a previous hypothesis and not a straw man), or I would suggest rephrasing this to state that "...uniform magmatic accretion at slow spreading rates is due to a two- dimensional sheet- like mantle upwelling more similar to magmatism at fast- spreading ridges than at previously surveyed slow- spreading ridge segments."  
+
+L212- 214: This is slightly concerning - if this inversion could be missing as much as 1 km of crustal thickness variation, then it is difficult to argue that the crustal thickness here is unambiguously uniform (and, in particular, more uniform from ends to centers than other MAR crust - if your points on Fig 1 moved up 1 km on the vertical axis, they would start to cluster with the other MAR points). The final sentence of this paragraph is "These tests demonstrate that the observed uniform crust within these five ridge segments is real" (L215). Please clarify the justification for that last sentence.  
+
+Fig 6 and 7, L514- 515: Related to the previous point, why was the crustal thickness perturbation wavelength for the checkerboard tests set at 70 km? Given that the recovered crustal thickness from that perturbation is significantly more uniform than the input model, it's not a very convincing argument for this model being able to properly resolve crustal thickness, but perhaps a longer perturbation wavelength could be better resolved. A longer wavelength would also be relevant to the question of whether the crustal thickness is uniform on a segment scale.  
+
+L275 paragraph: It seems like there are two arguments here mixed together: (1) uniform crustal thickness implies uniform (non- 3D) upwelling, and (2) there are long- wavelength trends in MORB geochemistry that aren't segment- scale and don't correlate to transform faults, indicating that the presence of transforms isn't controlling mantle temperatures enough to make upwelling more 3D and diapiric. I think the geochemistry part should be a paragraph on its own. The first part is more of a restatement of the main argument of the paper.
+
+<--- Page Split --->
+
+L338 paragraph: This paragraph also seems to contain more than one big idea: (1) some kind of balance between TF enhanced melting and hydrothermal cooling at segment centers might lead to relatively flat isotherms and uniform upwelling (up to L353), and (2) the length of a transform (but not specifically the offset on the transform) might have some effects on thermal structure, with mega- transforms facilitating this 2D upwelling scenario. I would recommend separating the first part of the paragraph from the second part, and reordering the paragraph for the second part to put the main point upfront.  
+
+L386: I am not sure what "suggesting that the mode of accretion plays a more important role in defining the 2- D versus 3- D mantle upwelling" means here. Does "mode of accretion" refer to tectonic vs magmatic accretion? If this is intended to imply a connection between the (large) percentage of magmatically accreted crust in the equatorial Atlantic and the mode of upwelling, i.e. 2D upwelling at slow spreading rates is possible if most of the crust is magmatically accreted, then I'd like to see more evidence for that connection/discussion of some mechanism. Otherwise it sounds kind of like correlation implying causation.  
+
+L387- 389 (and also L19- 20): I am not convinced by this idea that the "lateral extent of magmatic accretion could be used to map the 2D sheet- like mantle upwelling regions along the global spreading ridge system." Maybe what is meant is that the lateral extent of uniform- thickness magmatic- style crust can be used to define where sheet- like upwelling has occurred? "Magmatic accretion" also encompasses crust with segment- scale thickness variations that was accreted via 3D plume- like magmatic processes at slow- spreading ridges.  
+
+Some picky comments on grammar/typos/figures:  
+
+L18: "contradicting previous hypotheses" or "contradicting the previous hypothesis" - as written, the singular "hypothesis" doesn't work.  
+
+L19: the lateral extent  
+
+L284- 287: For the sentence that starts "While north of the St. Paul fracture zone..." I think if you remove the word "While" it would make sense grammatically.  
+
+L403: 50 or 70 ms? Do you mean 50- 70 ms?  
+
+Supp. Fig 2: Why shift times by 2 seconds rather than having the time axis start at 2? It's not incorrect or anything, but I'm genuinely curious if there's a reason for doing this.  
+
+L467- 468: for the 60 to 760 "horizontal distances" is the unit km along the seismic line?  
+
+Supp. Fig 6c and 7c: Could you use more distinct colors than red and magenta for the two Moho lines? I can't tell them apart.  
+
+Fig 1: How is the boundary between uniform and segment center- focused crust defined (dashed black horizontal line)? Is it just set at the smallest known thickness variation for non- equatorial Atlantic segments? Please add a citation or provide justification (or remove the line - not sure what purpose it serves since the groups of data points are already fairly distinct).  
+
+Fig 2: It would be useful to have a figure panel showing the (half- )spreading rate at the time of formation along this seismic line, so we don't have to try and match fig 2b to the mapped isochrons in fig 2a. This could replace 2b, or be an additional panel.  
+
+Fig 3, 5, etc: Why is there a gap between segments 2 and 3N? It seems odd particularly because the gap is narrower than the unconstrained part of the Vs model in the Romanche FZ, at least in the bars across the top of Fig 3a.
+
+<--- Page Split --->
+
+Figs 1 and 7: caption for 1 says only active source- derived crustal thicknesses are used, but the caption for 7 says data come from both seismic and gravity data; the number of points is the same for the two figures. Are any of them from gravity data?  
+
+Reviewer #3 (Remarks to the Author):  
+
+What are the noteworthy results?  
+
+Noteworthy is that the authors find relatively uniform crustal thickness at 5 off- axis ridge segments in the central Atlantic. In addition, 4.5 out of 5 segments appear constructed magmatically indicating two- dimensional mantle upwelling and crustal accretion.  
+
+Will the work be of significance to the field and related fields? How does it compare to the established literature? If the work is not original, please provide relevant references.  
+
+This result is surprising because previous results in both the northern and southern Atlantic ocean show:  
+
+i. Frequent amagmatic crustal accretion at the MAR especially near major first and second order offsets. ii. Significant observed variations in crustal thickness within MAR segments.  
+
+Existing crustal thickness variations have motivated two models in the established literature: (i) three- dimensional plume- like mantle upwelling; and (ii) two- dimensional mantle upwelling with melt focusing to segment centers along the topography at the base of the lithosphere.  
+
+- The Discussion argues against both of these two models for the equatorial Atlantic.- I am under the impression that the Discussion also argues against these models more broadly in the paragraph comparing crustal thickness variations to 1st and 2nd order segment lengths (lines 315-336) for a compilation of slow spreading ridges. However, it is not quite clear what process this comparison is seeking to test. Clarify the purpose of the paragraph making this comparison.  
+
+Does the work support the conclusions and claims, or is additional evidence needed?  
+
+This work is lacking in that a new model that explains both these new results and the existing observations is not clearly given in the abstract nor the conclusions.  
+
+Of notable relevance to developing a new model are the facts that:  
+
+1. this region of the mid-ocean ridge is broken up by very long offset transform faults and is a noted mantle cold spot (though this is confusing with additional discussion of a Cobb hotspot). This is not mentioned in the abstract. 
+2. the equatorial Atlantic is a cold spot and this appears crucial to the final interpretation in the very last (concluding?) paragraph (line 374-375, line 378). But adding this at the very end of the manuscript seems like a suddenly introduced new hypothesis. Specifically, this fact was not mentioned in the abstract and introduction. 
+3. existing observations seems to support more drastic crustal thickness variations along active ridge segments than off axis. This is mentioned and models related to this observation are rejected, however, an explanation for this observation is not provided.  
+
+In the Discussion, paragraph starting at line 275, a convincing argument is made that in the equatorial Atlantic the new crustal thickness measurements and existing geochemical observations support a 2D style of mantle upwelling and crustal accretion.  
+
+The model that the authors appear to favor is buried in the Discussion (lines 362- 364): "this suggests the mega- transform could facilitate a stable 2- D sheet- like mantel upwelling
+
+<--- Page Split --->
+
+and a relatively uniform crustal accretion". This seems to be the authors main conclusion and needs to be included in the Abstract. An outstanding question that I have about this process is whether the relatively cold mantle temperature in the region is adding to the effect of mega- transforms in controlling the mantle thermal structure.  
+
+Instead what the structure of the paper conveys to be more important is the poorly developed idea that: "the lateral extent[d]T of magmatic accretion could be used to define the sheet- like mantle upwelling regions beneath the global ridge system". It is simply stated at the end of the Abstract (note that when reading it I could not understand what "lateral extent" referred to). The next mention is in the very last sentence of the manuscript: "and the lateral extent of the magmatic accretion could be used to map the 2- D sheet- like mantle upwelling regions along the global spreading ridge system". Why is this important? This point does not seem as relevant to me as featuring the main interpretation/model in the Abstract.  
+
+The final paragraph (lines 386- 387) also states: "that the mode of accretion plays an important role in defining the 2- D versus 3- D mantle upwelling". What exactly is the thinking here? Does the mode of accretion control whether mantle upwelling is 2- D versus 3- D, i.e. is the control top- down? Or is the mode accretion a consequence of the pattern of mantle upwelling, i.e. the control is bottom- up? I believe the second is being argued here - using language that is more clear than "defining" might help clarify the intent of this statement.  
+
+Are there any flaws in the data analysis, interpretation and conclusions? Do these prohibit publication or require revision?  
+
+- The seismic data shown in the supplement looks good. There are very nice S wave arrivals - a consequence of collecting this data off-axis where acoustic waves can convert into S waves in the seafloor sediments at the ray entry point.- Lacking is a specific test of whether a model with along axis variations in crustal thickness is precluded by the data. This is required to support the interpretation that crustal thickness is more or less uniform along the 5 ridge segments. Specifically, the variable ability to recover the Moho checkerboard pattern raises this question (lines 212-214 & Supplemental Figures 6-8). A discussion the relative weight of travel times for direct P phases versus those for reflected phases on the results is not included.- The effect of varying the regularization parameters on the velocity structure and Moho topography is not explored in the Monte Carlo analysis.- To support the suggestion that colder mantle is playing a role in generating an average crustal thickness of 5.5 km (lines 379-381) the paper needs to include a calculation that shows whether or not a 150°C reduction in mantle is consistent with a reduction in crustal thickness/melt production of 500 m.  
+
+Is the methodology sound? Does the work meet the expected standards in your field? The seismic analysis seems sound and uses a well- established method  
+
+Is there enough detail provided in the methods for the work to be reproduced?- Yes. However, the choice of regularization parameters for the preferred model are not given.  
+
+## Detailed comments:  
+
+- Abstract: The two existing models for crustal thickness variations at slow spreading ridges are incorrectly merged together in the phrase (lines 11-12) "due a three-dimensional plume-like mantle upwelling with melt focusing to segment centres".
+
+<--- Page Split --->
+
+- Introduction 
+- issues with the description of crustal accretion at fast spreading ridges: o Line 32: 
+- Delete NTOs: Second order offsets are OSCs at fast spreading ridges and NTOs at slow spreading ridges 
+- This statement ignores the literature that crust is not thin beneath OSCs at the EPR (e.g. Canales et al., 2003). o Lines 33-34: Incorrect: fast spreading ridges are fed by individual mantle upwellings (e.g. Toomey 2007) and are not just two-dimensional sheet-like mantle upwellings.  
+
+- Supplemental Figures 6 &7: Caption reads that this is for Vp model, figures are labelled as being S models  
+
+In summary, I recommend major revisions so that the manuscript more clearly argues for the main processes that the authors infer from these new observations.
+
+<--- Page Split --->
+
+Reviewers' comments are shown in black; authors' response is shown in blue.  
+
+## Reviewer #1:  
+
+The manuscript is nicely written and presents a part of recently acquired wide- angle seismic data along an \(\sim 870 \mathrm{km}\) long transect imaging oceanic crust of different ages between \(5^{\circ}\mathrm{N}\) and \(5^{\circ}\mathrm{S}\) . An S- wave analysis adds to the database of the area. P- wave traveltime picks are the same as in the two previous works (44, 45) and the accompanying paper. But the P- wave model was generated newly and used for newly generated S- wave tomography. The authors propose a uniform crustal accretion caused by 2D sheet- like mantle upwelling similar to fast spreading ridges to occur at the slow- spreading equatorial Mid- Atlantic Ridge (MAR) which is different to observations further north and south along the MAR. The figures are presented well and have a high quality. The data analysis and methods are presented in a good standard and good details are given to reproduce the work.  
+
+Based on observations in reasonably old oceanic crust (8- 70 Ma), the authors conclude on the style of accretion during the formation of the crust, which is also targeted by the title. I'm missing an explanation why we see differences to the northern and southern MAR. The authors state that their findings contradict previous works at the northern and southern MAR. On the other hand, most named works observe young oceanic crust \(< 2 \mathrm{Ma}\) , while the authors of this work present oceanic crust \(>7 \mathrm{Ma}\) . Why do we have uniform crustal accretion along the segments at the equatorial MAR but not in the before mentioned regions north or south of \(5^{\circ}\) . Indeed, the data image old oceanic crust and a direct comparison to today's crustal accretion specifically at these 5 segments would be necessary to conclude on the source of the observed uniform crust.  
+
+In the revised manuscript, we have proposed two mechanisms that are likely facilitate a 2- D mantle upwelling beneath the slow- spreading ridges in the equatorial Atlantic Ocean: (1) en échelon large oceanic transform faults (TFs) and (2) higher \(\mathrm{CO_2}\) and \(\mathrm{H}_2\mathrm{O}\) concentrations in the primitive mantle melt, observed along the equatorial MAR.  
+
+The spreading ridge in the equatorial Atlantic Ocean is offset by several large oceanic TFs. The Romanche TF is the largest oceanic TF on the Earth, and the St. Paul and Chain TFs are much longer than most TFs in the North and South Atlantic Ocean [Wolfson- Schwehr and Boettcher, 2019] (Supplementary Fig. 13). By considering a more realistic brittle mantle weakening, Behn et al. [2007] argued that the thermal structure of the transform zone is much warmer than that predicted from the half- space cooling model, which can better fit the depth of seismicity on the oceanic TFs. Their modelling also demonstrates an enhanced mantle upwelling along the transform zone and a much thinner lithosphere at the transform zone, especially at the centre of the transform, than estimated in previous studies using simplified rheologic laws, as shown by Wang et al. (2022, Supplementary Fig.
+
+<--- Page Split --->
+
+10). The en échelon large TFs could significantly enhance the mantle upwelling in the equatorial Atlantic Ocean.  
+
+The 2- D sheet- like mantle upwelling and the formation of uniform crust could also be facilitated by the relatively higher \(\mathrm{CO_2}\) and \(\mathrm{H}_2\mathrm{O}\) concentrations in the melt in the equatorial Atlantic Ocean region [Le Voyer et al., 2019] (Fig. 9 in main text) as compared to the North and South Atlantic Ocean where large crustal thickness variations are observed. Large amount of these volatiles in the mantle will decrease the mantle solidus and increase the depth extent of the melting regime [Keller and Katz, 2016], leading to the enhanced production of melt beneath the spreading centres. Melt in the mantle can decrease the viscosity of the mantle [Whitehead et al., 1984], which could facilitate the mantle to flow. The presence of a large amount of volatiles ( \(\mathrm{CO_2}\) and \(\mathrm{H}_2\mathrm{O}\) ) would also decrease the density of the melt and mantle, and hence the mantle would be more buoyant, leading to a more 2D sheet- like upwelling.  
+
+It would be great to have the crustal thickness estimation along the ridge axis, parallel to our profile, but that would be mega- project of its own, and is beyond the scope of this paper. Furthermore, most of the existing seismic studies of the axis of slow- spreading ridges have been limited to the length of the segments, not beyond the transform, and hence has poor constrains on the transform system.  
+
+The authors name three hypotheses that should explain the reason for this uniform crustal structure. Their discussion is announced, I however can hardly find it completely discussed. Hypothesis 2 has been proposed in previous works from co- authors (45, 56). While hypothesis 1 is widely discussed and even further explanations are given that accretion was uniform during crustal accretion. However, the other two hypotheses are almost not discussed and it is not explained why they should not be valid. I do not see why hypothesis 2 is ruled out (two- stage accretion). What arguments contradict a two stage crustal accretion, during the time of segment formation and during the time the aging crust is passing the adjacent ridge axis?  
+
+We proposed three hypotheses for the uniform crust of magmatic origin observed within the five segments in the equatorial Atlantic Ocean: the crust has either (1) been modified after crustal formation by tectonic extension and stretching during the magmatic period, or (2) by a second- stage crustal accretion at RTIs, or (3) was originally formed uniformly at the ridge axis. We rule out the hypotheses 1 and 2 by the arguments in the second and third paragraphs in the 'Discussions' section.  
+
+The tectonic extension and stretching through normal faulting account for \(\sim 10\%\) of plate separation at the slow- spreading ridges [Combier et al., 2015; Escartin et al., 1999]. The spacing and heave of normal faults are generally larger at segment ends than those at segment centres of the slow- spreading ridges, indicating that more tectonic extension occurs at segment ends due to the decreasing magma
+
+<--- Page Split --->
+
+supply [Shaw, 1992; Shaw and Lin, 1993]. Since the tectonic extension and stretching could thin the oceanic crust [Combier et al., 2015; Escartín et al., 1999; Escartín and Lin, 1995], more tectonic extension and stretching towards the segment ends will enhance the along- axis variation in crustal thickness [Detrick et al., 1995; Shaw, 1992], rather than making the crust uniform at segment scale, validating that the tectonic extension and stretching cannot explain the uniform crust observed along our profile, ruling out the first hypothesis.  
+
+Marjanović et al. [2020] and Grevemeyer et al. [2021] proposed that the oceanic transform fault (TF) crust can get augmented by a second- stage of magmatic accretion at the RTI, which accounts for the relatively thicker crust and relatively shallower seafloor at the oceanic fracture zone (FZ) than at the TF valley. However, the gravity data reveal a \(\sim 5\) km- thick crust within the Chain transform zone [Harmon et al., 2018] comparable with the average \(\sim 5.1\) km- thick crust within the Chain FZ along our profile, suggesting that the second- stage of magmatic accretion may have limited augmentation to the TF crust at RTIs. Grevemeyer et al. [2021] argued that this second- stage of magmatic accretion at RTI forms the J- shaped ridges observed on the bathymetry data, which can extend across the RTIs and often terminate in the older plate. However, the J- shaped ridges have very short extensions (generally \(< 10 \mathrm{km}\) ) over the old ocean floor [Grevemeyer et al., 2021], suggesting that the second- stage of crustal augmentation occurs within a limited region in the vicinity of the RTI. Since the spreading centre between two major TFs is \(\geq 200 \mathrm{km}\) long in our study region, the second- stage of magmatic accretion at RTI cannot change the segment- scale variations in the crustal thickness. Therefore we rule out the hypothesis 2.  
+
+We have augmented these discussions in the revised version of the paper.  
+
+In my opinion additional evidence is needed to support the interpretation and conclusions and to rule out the other named hypotheses. While I still find the work interesting and worth publishing after minor changes, I think there are other journals this work would fit better, since it does not present a very novel and clear explanation or extremely novel findings. The dataset is presented in many pieces.  
+
+Detailed points:  
+
+Lines 55- 60: Give references to the both modes.  
+
+We cited Escartín et al. [2008] and Searle [2013] for the magmatic accretion mode and cited Cann et al. [1997] for the tectonic accretion mode.  
+
+Lines 73/74: Give a reference.  
+
+We cited Carlson and Miller [1997] as reference.
+
+<--- Page Split --->
+
+Line 85: Define the area of "equatorial Atlantic Ocean". Is this according to fig. 2 it is \(+ / - 5^{\circ}\) ?  
+
+Here we modified to 'covering \(\sim 8 - 70\) Ma old lithosphere between \(4^{\circ}\mathrm{N}\) and \(4^{\circ}\mathrm{S}\) in the equatorial Atlantic Ocean'. The equatorial Atlantic oceanic ridge can be defined as the ridge segments between the Vema TF at \(\sim 10^{\circ}\mathrm{N}\) and the Ascension TF at \(\sim 7^{\circ}\mathrm{S}\) [Uditnsev, 1996], but we didn't specify this definition in the manuscript because it is not related to the main idea of this work.  
+
+Line 118: Make clear that these are converted (PS) or since hydrophones were used for picking PSP converted. What is the conversion horizon? For young crust with no sediments most often it is the seafloor itself.  
+
+In the main text we clarified that the picked S- wave arrivals have P- to- S and S- to- P conversion modes at the interface between the igneous crust and the sedimentary layer (if present) or water. And we provided detailed analyses in Section 'Identifying the P- to- S and S- to- P conversion interfaces' in Methods and Supplementary Figs. 3- 5.  
+
+Line 134: Does tectonic accretion need a detachment fault? I understood that detachments develop under certain conditions regarding the magmatic budget. Too less or too much magma, no detachment development. Compare to Buck et al., 2005. \(\mathrm{M} =\) Magmatic Budget; \(\mathrm{M}< = 0.5\)  
+
+Here we deleted 'along detachment fault'.  
+
+Line 143: add PmP \(\rightarrow\) "... reflections (PmP)."  
+
+Modified as suggested by the reviewer.  
+
+Lines 161- 163: It looks like you observe a seismic Moho in that segment. What does this reflector stand for? A really seismic Moho seperating igneous crust from mantle or a serpentinisation front or any other explanation?  
+
+We observed a seismic Moho along the whole profile except in Segment 3- N. We interpret this seismic Moho as the boundary between crust and mantle, which is characterized by rapid increase in both P- and S- wave velocities across the Moho from the crustal to the mantle. The clear P- and S- wave reflections off the Moho demonstrate a relatively thin or sharp boundary between the crust and mantle. Though our tomographic method cannot constrain the thickness of the Moho transition zone, this seismic Moho is not likely represents a serpentinization front as the serpentinization process tends to be gradual leading to more smooth velocity variation in depth.  
+
+Line 178: "a slight thinning" \(\leftarrow\) - quantify slight. Or is this represented by 5.4 km total crustal thickness?
+
+<--- Page Split --->
+
+Here we modified to 'The oceanic crust shows a slight thinning in a narrow zone in the vicinity of the FZ, TF and pseudo- fault regions, where the thinnest crust at these tectonic discontinuities is \(< 1.5 \mathrm{km}\) thinner relative to the average of each segment'.  
+
+Line 181: Crust thins by \(\sim 600 \mathrm{m}\) to a total crustal thickness of \(\mathrm{xx km}\) ? Does your total crustal thickness include sediments or not? In line 174 you state a total change in crustal thickness of \(200 \mathrm{m}\) . This contradicts line 181 with \(\sim 600 \mathrm{m}\) change in crustal thickness.  
+
+At the beginning of Section 'Intra- segment crustal thickness variation within magmatic segments', we clarify 'the thickness of the igneous crust is defined as the thickness between the top basement and the seismically constrained Moho'. So all the crustal thicknesses discussed in this manuscript don't include the sediments.  
+
+Here we changed to 'At the southern end of Segment 1, the crust thins by \(\sim 600 \mathrm{m}\) to \(\sim 4.8 \mathrm{km}\) thick at the centre of the northern St. Paul FZ valley over \(< 5 \mathrm{km}\) distance'.  
+
+Lines 227- 231: The authors propose three hypotheses to explain the small variation in crustal thickness in old crust ( \(>7 \mathrm{Ma}\) ). However, in the following sections the discussion of these three hypotheses concentrates on evidences for hypothesis 1 mainly. Hypothesis 2: Stretching here refers to the segment centre after crustal formation? I see this hardly discussed in the following sections. A stretching of the segment centre only can be ruled out as long as magnetic anomalies are ridge parallel. Hypothesis 3: I don't see this discussed in the following sections. Why do you rule out this 2 stage accretion?  
+
+As mentioned above, we have discussed the three hypotheses in the 'Discussions' section more in- depth. We have elaborated on all the three hypotheses, rejected two as detailed in the response to the second question of reviewer and in the discussion section in the main text.  
+
+Line 273: Remember that you are not studying the segment during the time of crustal formation (first stage).  
+
+Here we changed to 'a 3- D diapiric mantle upwelling would produce a large variation in crustal thickness within a ridge segment [Lin and Morgan, 1992; Lin et al., 1990], which is inconsistent with our observations of the five crustal segments studied here.'  
+
+Line 276: Did you look at the mantle bouger anomaly at the recent segment centres of this five segments studied in this work?  
+
+Marjanović et al. [2020] processed the high- resolution gravity data collocated with our seismic profile collected during the ILAB- SPARC experiment. Overall, the MBA displays short- wavelength artefacts, not long- wavelength variations with low MBA at segment centres and high MBA at segment ends.
+
+<--- Page Split --->
+
+To support our interpretation of presence of uniform crust in the study region, we also compared the along- axis variation of seafloor depth and MBA along several sections of the ridge axis that show focused (Lucky Strike and \(\mathrm{H}_2\mathrm{O}\) ) and uniform crustal (Equatorial MAR and MAR at \(30^{\circ}\mathrm{S}\) ) variations but here we show two extreme examples for the Lucky Strike segment at \(37^{\circ}\mathrm{N}\) and the segment between the St. Paul and Romanche TFs (Supplementary Fig. 14). The Lucky Strike segment shows enhanced and focused magmatic accretion at the centre of the segment [Seher et al., 2010; Singh et al., 2006], leading to large variations in seafloor depth (1.3- 2.8 km) and MBA ( \(\sim 30\mathrm{mGal}\) ) from the centre to the distal ends (Supplementary Fig.14 a,b). In contrast, the segment between the St. Paul and Romanche TFs shows \(< 0.7 - 1.0\mathrm{km}\) variation in seafloor depth and \(< 15 - 20\mathrm{mGal}\) variation in the MBA (Supplementary Fig.14 c,d), much smaller than those observed along the Lucky Strike segment but comparable with variations along the fast- spreading ridges (200- 700 m and \(10 - 20\mathrm{mGal}\) , respectively; [Lin and Morgan, 1992]), supporting the presence of relatively uniform crust.  
+
+Lines 295- 297: Give a reference for the "thick crust".  
+
+Modified as suggested by the reviewer.  
+
+Lines 353- 364: This seems to be a kind of doubling to the lines 328- 331.  
+
+In Line 328- 331, we discussed the relation between the along- axis crustal thickness variation within a second- order ridge segment and the length of the second- order ridge segment. In Line 353- 364, we discussed the relation between the along- axis crustal variation and the length of the adjacent oceanic TF. So it's not doubling. However, we have removed the discussion of the relationship between crustal thickness variation and the length of second- order ridge segment in the current manuscript.  
+
+Line 384: \(50\%\) versus \(90\%\) magmatic accretion \(< - >\) You look at very different times of crustal accretion within the 5 segments. You might image magmatically active times of the specific segments.  
+
+In this work, we study the thickness of crust within five crustal segments formed 8, 24, 40 and \(70\mathrm{Ma}\) , respectively. We don't believe it's a coincidence that we sampled five segment formed during magmatically active times, though we have no strong evidence to rule out this possibility. However, all the five ridge segments seem to be magmatically robust, which is supported by the uniform magmatic crust (Fig. 5) and the well- defined Moho boundary (Figs. 3 and 4a).  
+
+Line 394: Why did you change frequency, was the signal for Pn so much improved but Pg and PmP havn't been visible any longer?  
+
+Filtering the seismic data from 4- 20 Hz is sufficient for picking the near- offset ( \(< 100\mathrm{km}\) ) mantle refractions (Pn) but is difficult for picking Pn arrivals at far offset up to \(700\mathrm{km}\) [Wang et al., 2022]. To pick the Pn arrivals up to \(700\mathrm{km}\) offset, we filtered the data to a maximum frequency of \(15\mathrm{Hz}\) . To
+
+<--- Page Split --->
+
+keep the consistency of the frequency band, we filtered the data from 4 to 15 Hz for picking Pn arrivals at both near and far offsets.  
+
+Lines 457 and followings: You observe two times converted S- Waves. Starting as P- Wave at the source, converting at or in the subsurface, travelling as S through the subsurface becoming converted to P at a boundary or seafloor and than being recorded as P- Wave to the hydrophone. Where does the conversion takes place? If at the seafloor, did you take the sedimentary cover into account? It will extremely delay the S- wave arrivals even if only a few centimeter of sediments.  
+
+We clarified in the main text that the picked S- wave arrivals have P- to- S and S- to- P conversions at the interface between igneous crust and the sediment (if present) or water. We have also provided detailed analyses in Section 'Identifying the P- to- S and S- to- P conversion interfaces' in Methods and in Supplementary Figs. 3- 5.  
+
+Line 467: add km to "... 60 to 760 horizontal distances ..."  
+
+Modified as suggested by the reviewer.  
+
+Line 468: change to something more clear maybe: "the lower crust has sparse ray coverage, PnP only."  
+
+We rewrote to clarify this point.  
+
+Lines 476- 477: You do not have Vs there at all.  
+
+We have clarified this point.  
+
+Line 486: "50 models" --> "50 calculated models" or "50 inverted models"  
+
+Modified as suggested by the reviewer.  
+
+Line 512: "...models is 20x3 (Supp...") add "km". Did you also run a checkerboard test with inverted anomalies, to show that the resolution is independent of the polarity of the anomaly?  
+
+As suggested by the reviewer, we performed checkerboard tests with inverted velocity anomalies and also with inverted variation in Moho depth (see Methods). The results are shown in Supplementary Figs. 9- 11. Our tests suggest the resolution of the used tomography method is independent on the polarity of the anomaly.  
+
+Lines 529- 531: This is a very weak explaination, because you do not allow for Moho depth adjustments. This does not mean that this Vp- Moho is the best possible. By this you do not allow for more variation.
+
+<--- Page Split --->
+
+We deleted this sentence in Lines 529- 531.  
+
+Fig. 4: Panel b, segment 3N, shows models with a negative vertical velocity gradient which hints to problems in the inversion or how do you explain a negative velocity gradient in your models?  
+
+The negative vertical velocity gradient is observed in the 1- D velocity profile because the lower crustal velocity in this region is not constrained by PmP arrivals. The negative vertical velocity gradient is produced by the smoothing effect of tomography. In this work, we only discuss the crustal velocity structure within \(\sim 2.2 \mathrm{km}\) sub- basement depth for the segment 3- N, which is constrained by the dense Pg arrivals.  
+
+Suppl. Fig. 5: panel d, I do not see the advantage of presenting the variance of Vp/Vs, since Vp is kept constant it more or less reflects the variance of Vs.  
+
+We removed the plot of the standard deviation of Vp/Vs.  
+
+## Reviewer #2:  
+
+This manuscript presents a new Vp and Vs model for the oceanic crust across several segments in the equatorial Atlantic. The seismic line spans a large range of crustal ages. The velocity modeling shows that (1) the crust was magmatically accreted, and (2) crustal thickness is roughly constant, in contrast to other slow- spreading crust where crustal thickening is observed toward segment centers. The authors interpret this crustal thickness observation as indicating that the crust in this part of the Atlantic accreted magmatically via 2D sheet- like mantle upwelling rather than the 3D plume- like upwelling generally hypothesized for slow- spreading ridges.  
+
+This is an interesting result, and is well presented in the manuscript. It has important implications for how we conceptualize crustal accretion and mantle upwelling at ridges. I do have some questions and comments, which are listed below. The main points that I think need addressing are these: (1) the resolution tests could do a better job of convincing readers that the uniform crustal thickness is a robust result; and (2) the manuscript makes some statements about the implications of the study that seem to equate magmatic upwelling at ridges with 2D sheet- like upwelling, although elsewhere the existence of 3D plume- like upwelling at some slow- spreading segments is discussed. I suspect the first issue is only because of the choice of testing parameters, and the second can be addressed by refining some of those over- broad statements.
+
+<--- Page Split --->
+
+We have provided extensive resolution tests as pointed out by the reviewer and have revised the manuscript to address the second point.  
+
+I understand the desire not to speculate too much without solid evidence, but I would be interested to read more of the authors' thoughts on why magmatic crust in the equatorial Atlantic is apparently accreting differently than elsewhere along the MAR. Is it the presence of extra- long transforms and their stabilizing effects on ridge thermal structure that make this 2D upwelling possible in a slow- spreading regime? This is suggested in the discussion, but the abstract and final conclusions focus on the potential to map out 2D upwelling regions rather than ideas as to why this upwelling scenario exists. Discussion of the role of these long transforms seems especially relevant in the context of the related manuscript on the velocity structure beneath Romanche.  
+
+In the revised manuscript, we proposed two mechanisms that likely could facilitate a 2- D mantle upwelling beneath slow- spreading ridges in the equatorial Atlantic Ocean: (1) en échelon large oceanic transform faults (TFs) and (2) higher \(\mathrm{CO_2}\) and \(\mathrm{H}_2\mathrm{O}\) concentrations in the mantle melt in the equatorial Atlantic region. Please refer to the response of the first question from reviewer 1 and the discussion section in main text for the details.  
+
+Specific comments:  
+
+L18: I am not sure that "contradicting [a] previous hypothesis" is accurate here. I assume that the "previous hypothesis" is supposed to be that 2D sheet- like upwelling cannot occur at slow- spreading ridges, though this isn't explicitly stated. Either that previous hypothesis should be more clearly defined (if it is actually a previous hypothesis and not a straw man), or I would suggest rephrasing this to state that "...uniform magmatic accretion at slow spreading rates is due to a two- dimensional sheet- like mantle upwelling more similar to magmatism at fast- spreading ridges than at previously surveyed slow- spreading ridge segments."  
+
+We thank reviewer for the above remark and suggestion. We have modified the text accordingly.  
+
+L212- 214: This is slightly concerning – if this inversion could be missing as much as 1 km of crustal thickness variation, then it is difficult to argue that the crustal thickness here is unambiguously uniform (and, in particular, more uniform from ends to centers than other MAR crust – if your points on Fig 1 moved up 1 km on the vertical axis, they would start to cluster with the other MAR points). The final sentence of this paragraph is "These tests demonstrate that the observed uniform crust within these five ridge segments is real" (L215). Please clarify the justification for that last sentence.  
+
+Here we performed the checkerboard tests using checkerboard pattern with a velocity anomaly of \(10\%\) and Moho depth perturbation of different half- wavelengths (50, 100 and 200 km, respectively). In
+
+<--- Page Split --->
+
+these tests, the variations in Moho depth in the checkerboard models are \(\sim 2.5 - 3.0 \mathrm{km}\) . The final inverted results (Supplementary Figs. 9- 11) show that the used travel time tomography method can recover the Moho depth and its lateral variation for most portions along our seismic profile. So if large crustal thickness variations exist within the five crustal segments we have studied here, we should observe large crustal thickness variations in the tomographic model. However, the tomographic crustal structure shown relatively uniform thickness, which in turn, demonstrates no large crustal thickness variations within the five segments.  
+
+Fig 6 and 7, L514- 515: Related to the previous point, why was the crustal thickness perturbation wavelength for the checkerboard tests set at \(70 \mathrm{km}\) ? Given that the recovered crustal thickness from that perturbation is significantly more uniform than the input model, it's not a very convincing argument for this model being able to properly resolve crustal thickness, but perhaps a longer perturbation wavelength could be better resolved. A longer wavelength would also be relevant to the question of whether the crustal thickness is uniform on a segment scale.  
+
+As suggested by the reviewer, we performed the checkerboard tests using models with variation of different lateral half- wavelengths in Moho depth (50, 100 and \(200 \mathrm{km}\) ). The inverted results are shown in Supplementary Figs. 9- 11. These tests demonstrate that the used travel time tomography method can almost recover the Moho depth and its lateral variation for most portions along our seismic profile.  
+
+L275 paragraph: It seems like there are two arguments here mixed together: (1) uniform crustal thickness implies uniform (non- 3D) upwelling, and (2) there are long- wavelength trends in MORB geochemistry that aren't segment- scale and don't correlate to transform faults, indicating that the presence of transforms isn't controlling mantle temperatures enough to make upwelling more 3D and diapiric. I think the geochemistry part should be a paragraph on its own. The first part is more of a restatement of the main argument of the paper.  
+
+We rephrased this paragraph as suggested by reviewer. In the revised manuscript, we first concluded that the relatively uniform crustal thickness observed in the equatorial Atlantic Ocean is due to a relatively uniform (nearly 2- D) mantle upwelling beneath the ridges at the time of crustal accretion. And in the following paragraph, we used the results of long- wavelength trends from geochemistry as an evidence to support our conclusion of a nearly 2- D mantle upwelling in the study region.  
+
+L338 paragraph: This paragraph also seems to contain more than one big idea: (1) some kind of balance between TF enhanced melting and hydrothermal cooling at segment centers might lead to relatively flat isotherms and uniform upwelling (up to L353), and (2) the length of a transform (but not specifically the offset on the transform) might have some effects on thermal structure, with megatransforms facilitating this 2D upwelling scenario. I would recommend separating the first part of the
+
+<--- Page Split --->
+
+paragraph from the second part, and reordering the paragraph for the second part to put the main point upfront.  
+
+We thank the reviewer for the suggestion. We have now separated these two ideas. The balance between transform fault (TF) enhanced melting and hydrothermal cooling at segment centres could lead to a relatively flat isotherm. And we used this to argue that the melt is not necessarily focused to the segment centres at the base of the lithosphere as suggested by Magde and Sparks [1997] and some later works. This is further supported by the geochemistry study which suggests that the melt does not focus to a magma chamber but erupts rapidly vertically.  
+
+In the revised manuscript, we propose that the en échelon large oceanic TFs facilitate the uniform mantle upwelling in the equatorial Atlantic region, using a compilation of segment length and transform offset to support this idea.  
+
+L386: I am not sure what "suggesting that the mode of accretion plays a more important role in defining the 2- D versus 3- D mantle upwelling" means here. Does "mode of accretion" refer to tectonic vs magmatic accretion? If this is intended to imply a connection between the (large) percentage of magmatically accreted crust in the equatorial Atlantic and the mode of upwelling, i.e. 2D upwelling at slow spreading rates is possible if most of the crust is magmatically accreted, then I'd like to see more evidence for that connection/discussion of some mechanism. Otherwise it sounds kind of like correlation implying causation.  
+
+We have deleted this sentence to avoid confusion.  
+
+L387- 389 (and also L19- 20): I am not convinced by this idea that the "lateral extent of magmatic accretion could be used to map the 2D sheet- like mantle upwelling regions along the global spreading ridge system." Maybe what is meant is that the lateral extent of uniform- thickness magmatic- style crust can be used to define where sheet- like upwelling has occurred? "Magmatic accretion" also encompasses crust with segment- scale thickness variations that was accreted via 3D plume- like magmatic processes at slow- spreading ridges.  
+
+We have deleted this sentence as this idea is not well developed and is not really helpful to the main conclusion of this paper.  
+
+Some picky comments on grammar/typos/figures:  
+
+L18: "contradicting previous hypotheses" or "contradicting the previous hypothesis" - as written, the singular "hypothesis" doesn't work.  
+
+Modified as suggested by the reviewer.
+
+<--- Page Split --->
+
+L19: the lateral extent  
+
+Modified as suggested by the reviewer.  
+
+L284- 287: For the sentence that starts "While north of the St. Paul fracture zone..." I think if you remove the word "While" it would make sense grammatically.  
+
+Modified as suggested by the reviewer.  
+
+L403: 50 or 70 ms? Do you mean 50- 70 ms?  
+
+The picking uncertainty of the PmP arrivals is 50 ms or 70 ms; we have clarified this point.  
+
+Supp. Fig 2: Why shift times by 2 seconds rather than having the time axis start at 2? It's not incorrect or anything, but I'm genuinely curious if there's a reason for doing this.  
+
+We shifted the travel time by 2 s to simultaneously show the P- wave first arrivals and the S- wave arrivals in the same plot, otherwise the S- wave first arrivals with offset \(>10\mathrm{km}\) would appear at negative time when the reduction velocity is \(4\mathrm{km / s}\) . So the 2 s shift is only for display purpose. We have clarified this point in the figure caption.  
+
+L467- 468: for the 60 to 760 "horizontal distances" is the unit km along the seismic line?  
+
+Modified as suggested by the reviewer.  
+
+Supp. Fig 6c and 7c: Could you use more distinct colors than red and magenta for the two Moho lines? I can't tell them apart.  
+
+We have modified the figures to address the above point.  
+
+Fig 1: How is the boundary between uniform and segment center- focused crust defined (dashed black horizontal line)? Is it just set at the smallest known thickness variation for non- equatorial Atlantic segments? Please add a citation or provide justification (or remove the line – not sure what purpose it serves since the groups of data points are already fairly distinct).  
+
+We have removed the bold dashed line showing \(2.8\mathrm{km}\) crustal thickness difference from Fig. 1.  
+
+Fig 2: It would be useful to have a figure panel showing the (half- )spreading rate at the time of formation along this seismic line, so we don't have to try and match fig 2b to the mapped isochrons in fig 2a. This could replace 2b, or be an additional panel.  
+
+As suggested by the reviewer, we have plotted the half- spreading rate of the ridge at the time the studied five segments were formed (see Fig. 2b in main text).
+
+<--- Page Split --->
+
+Fig 3, 5, etc: Why is there a gap between segments 2 and 3N? It seems odd particularly because the gap is narrower than the unconstrained part of the Vs model in the Romanche FZ, at least in the bars across the top of Fig 3a.  
+
+In our discussion, the Segment 2 extends from the centre of the Romanche transform valley to the centre of the northern valley of the Saint Paul fracture zone. For Segment 3- N, we could define it from \(\sim 340 \mathrm{km}\) horizontal distance to the centre of the Romanche transform valley. Gregory et al. [2021] argued that the crust below the Romanche transform valley is primarily composed of mafic rocks. In this case, we defined the northern end of the Segment 3- N at the southern bound wall of the Romanche transform valley. This doesn't influence our interpretations.  
+
+Figs 1 and 7: caption for 1 says only active source- derived crustal thicknesses are used, but the caption for 7 says data come from both seismic and gravity data; the number of points is the same for the two figures. Are any of them from gravity data?  
+
+The maximum crustal thickness variation between the Pico Offset- Oceanographer TFs [Detrick et al., 1995] and the Atlantis TF- Kane TFs [Lin et al., 1990] are obtained from gravity data. All the other estimates are from active- source seismic studies. We clarified this point in Supplementary Table 3.  
+
+## Reviewer #3:  
+
+What are the noteworthy results? Noteworthy is that the authors find relatively uniform crustal thickness at 5 off- axis ridge segments in the central Atlantic. In addition, 4.5 out of 5 segments appear constructed magmatically indicating two- dimensional mantle upwelling and crustal accretion.  
+
+Will the work be of significance to the field and related fields? How does it compare to the established literature? If the work is not original, please provide relevant references.  
+
+This result is surprising because previous results in both the northern and southern Atlantic Ocean show: i. Frequent magmatic crustal accretion at the MAR especially near major first and second order offsets. ii. Significant observed variations in crustal thickness within MAR segments.  
+
+Existing crustal thickness variations have motivated two models in the established literature: (i) three- dimensional plume- like mantle upwelling; and (ii) two- dimensional mantle upwelling with melt focusing to segment centers along the topography at the base of the lithosphere.
+
+<--- Page Split --->
+
+- The Discussion argues against both of these two models for the equatorial Atlantic. 
+- I am under the impression that the Discussion also argues against these models more broadly in the paragraph comparing crustal thickness variations to 1st and 2nd order segment lengths (lines 315-336) for a compilation of slow spreading ridges. However, it is not quite clear what process this comparison is seeking to test. Clarify the purpose of the paragraph making this comparison.  
+
+We have rephrased this paragraph to focus on discussing whether the patterns of mantle upwelling and crustal accretion at slow-spreading ridges are related to the first-order ridge segment length. Christeson et al. [2020] proposed that a longer first-order ridge segment can facilitate larger mantle upwelling. Our compilation (Fig. 7) demonstrates that the length of first-order ridge segment has no influence on the crustal accretion process, hence on the mantle upwelling pattern.  
+
+Does the work support the conclusions and claims, or is additional evidence needed? This work is lacking in that a new model that explains both these new results and the existing observations is not clearly given in the abstract or the conclusions. Of notable relevance to developing a new model are the facts that: 1. this region of the mid-ocean ridge is broken up by very long offset transform faults and is a noted mantle cold spot (though this is confusing with additional discussion of a Cobb hotspot). This is not mentioned in the abstract. 2. the equatorial Atlantic is a cold spot and this appears crucial to the final interpretation in the very last (concluding?) paragraph (line 374-375, line 378). But adding this at the very end of the manuscript seems like a suddenly introduced new hypothesis. Specifically, this fact was not mentioned in the abstract and introduction. 3. existing observations seems to support more drastic crustal thickness variations along active ridge segments than off axis. This is mentioned and models related to this observation are rejected, however, an explanation for this observation is not provided.  
+
+In the revised manuscript, we proposed two mechanisms that could facilitate a 2- D mantle upwelling beneath slow- spreading ridges in the equatorial Atlantic Ocean: (1) en échelon large oceanic transform faults (TFs) and (2) higher \(\mathrm{CO_2}\) and \(\mathrm{H}_2\mathrm{O}\) concentrations in the mantle melt. And we also highlight this in the abstract. Please refer to the response to the first question from reviewer 1 and the discussion section in main text for the details.  
+
+In the Discussion, paragraph starting at line 275, a convincing argument is made that in the equatorial Atlantic the new crustal thickness measurements and existing geochemical observations support a 2D style of mantle upwelling and crustal accretion.  
+
+The model that the authors appear to favor is buried in the Discussion (lines 362- 364): "this suggests the mega- transform could facilitate a stable 2- D sheet- like mantle upwelling and a relatively uniform crustal accretion". This seems to be the authors main conclusion and needs to be included in the
+
+<--- Page Split --->
+
+Abstract. An outstanding question that I have about this process is whether the relatively cold mantle temperature in the region is adding to the effect of mega- transforms in controlling the mantle thermal structure.  
+
+In the revised manuscript, we proposed two mechanisms that could facilitate a 2- D mantle upwelling beneath slow- spreading ridges in the equatorial Atlantic Ocean: (1) large oceanic transform faults and (2) higher \(\mathrm{CO_2}\) and \(\mathrm{H}_2\mathrm{O}\) concentrations in the melt. And we highlight this in the abstract. From our observations, we cannot conclude whether the relatively cold mantle temperature in the region is adding to the effect of mega- transforms in controlling the mantle thermal structure.  
+
+Instead what the structure of the paper conveys to be more important is the poorly developed idea that: "the lateral extent[d]T of magmatic accretion could be used to define the sheet- like mantle upwelling regions beneath the global ridge system". It is simply stated at the end of the Abstract (note that when reading it I could not understand what "lateral extent" referred to). The next mention is in the very last sentence of the manuscript: "and the lateral extent of the magmatic accretion could be used to map the 2- D sheet- like mantle upwelling regions along the global spreading ridge system". Why is this important? This point does not seem as relevant to me as featuring the main interpretation/model in the Abstract.  
+
+## Here we deleted this sentence.  
+
+The final paragraph (lines 386- 387) also states: "that the mode of accretion plays an important role in defining the 2- D versus 3- D mantle upwelling". What exactly is the thinking here? Does the mode of accretion control whether mantle upwelling is 2- D versus 3- D, i.e. is the control top- down? Or is the mode accretion a consequence of the pattern of mantle upwelling, i.e. the control is bottom- up? I believe the second is being argued here – using language that is more clear than "defining" might help clarify the intent of this statement.  
+
+## Here we deleted this sentence.  
+
+Are there any flaws in the data analysis, interpretation and conclusions? Do these prohibit publication or require revision?  
+
+- The seismic data shown in the supplement looks good. There are very nice S wave arrivals – a consequence of collecting this data off-axis where acoustic waves can convert into S waves in the seafloor sediments at the ray entry point.  
+
+- Lacking is a specific test of whether a model with along axis variations in crustal thickness is precluded by the data. This is required to support the interpretation that crustal thickness is more or less uniform along the 5 ridge segments. Specifically, the variable ability to recover the Moho
+
+<--- Page Split --->
+
+checkerboard pattern raises this question (lines 212- 214 & Supplemental Figures 6- 8). A discussion of the relative weight of travel times for direct P phases versus those for reflected phases on the results is not included.  
+
+Firstly, we performed the checkerboard tests using checkerboard pattern with size of \(20 \times 2 \mathrm{km}\) and \(10\%\) velocity anomaly. And we added Moho depth perturbation of different half- wavelengths (50, 100 and \(200 \mathrm{km}\) ) in the checkerboard model. In these tests, the variations in Moho depth in the checkerboard models are \(\sim 2.5 - 3.0 \mathrm{km}\) . The final inverted results (Supplementary Figs. 9- 11) show that the used travel time tomography method can recover the Moho depth and its lateral variation for most portions along our seismic profile. And the resolvability of the used tomography method is independent on the polarities of the velocity anomaly and the Moho depth perturbation. This means if large along- axis crustal thickness variations do exist, our traveltime method can recover it.  
+
+Secondly, the results from the Monte- Carlo analysis [Korenaga et al., 2000] by taking the variations of regularization parameters into account show similar crustal Vp structure with relatively uniform crust, which suggests the inversion of the picked Pg and PmP travel times is robust. The standard deviation of the Moho depth is \(< 400 \mathrm{m}\) . This further supports that the uniform crust along our seismic profile is real. Here we presented two inverted results which obtained using parameters allowing large updating in Moho depth (see Fig. A2 below). For these two models, the Moho boundary shows more variability, but the crust still shows little variations in thickness, where the standard deviation of average crustal thickness is \(< 0.4 \mathrm{km}\) .  
+
+![PLACEHOLDER_23_0]
+
+
+<--- Page Split --->
+![PLACEHOLDER_24_0]
+
+<center>Figure A2: (a,b) Final inverted results obtained using parameters allowing large updating in Moho depth. (c) and (d) show the crustal thickness variations (in black) and average crustal thickness (in red) of each segment for final inverted model (a) and (b), respectively. The values in red in plots (c,d) represent the average crustal thickness and the standard deviations. </center>  
+
+- The effect of varying the regularization parameters on the velocity structure and Moho topography is not explored in the Monte Carlo analysis.  
+
+As suggested by the reviewer, we performed the Monte Carlo analysis for crustal P- wave velocity structure by considering the variation of the regularization parameters on the velocity and Moho structure. (see Section 'Monte- Carlo analysis' in Methods). The results are shown in Supplementary Fig. 8. The variance in the final crustal Vp model is less than \(0.1 \mathrm{km / s}\) in the upper crust and is less than \(0.3 \mathrm{km / s}\) in the lower crust (Supplementary Fig. 8a). The maximum standard deviation of the Moho depth is \(\sim 400 \mathrm{m}\) (Supplementary Fig. 8b). The preferred Moho (Fig. 3a) falls in the standard deviation of the average Moho depth from the Monte- Carlo analysis (see red curve in Supplementary Fig. 8b). Similar Monte- Carlo analyses are performed to assess the variance in the crustal Vs. Supplementary Fig. 8c shows the variance of the crustal Vs calculated using 50 final inverted models.
+
+<--- Page Split --->
+
+For most portion of the crust, the crustal Vs has a variance \(< 0.1 \mathrm{km / s}\) , and large variances \(>0.1 \mathrm{km / s}\) are observed around the TF and FZs and at the southern and northern extremity of the model.  
+
+- To support the suggestion that colder mantle is playing a role in generating an average crustal thickness of \(5.5 \mathrm{km}\) (lines 379-381) the paper needs to include a calculation that shows whether or not a \(150^{\circ}\mathrm{C}\) reduction in mantle is consistent with a reduction in crustal thickness/melt production of 500 m.  
+
+Schilling et al. [1995] estimated that the minimum melting temperature is \(\sim 1300^{\circ}\mathrm{C}\) in the equatorial Atlantic Ocean. The numerical modelling of Behn and Grove [2015] demonstrated that a \(\sim 5 \mathrm{km}\) thick crust can be formed when the half-spreading rate is \(\sim 20 \mathrm{mm / year}\) and the mantle potential temperatures is \(1300^{\circ}\mathrm{C}\) , which is \(500 \mathrm{m}\) thinner than the estimates \((\sim 5.5 \mathrm{km})\) we obtained in this paper. The large amount of volatiles \((\mathrm{CO}_{2}\) and \(\mathrm{H}_{2}\mathrm{O})\) in the mantle in the equatorial Atlantic Ocean will decrease the mantle solidus and increase the depth extent of the melting regime, leading to the enhanced production of melt. The enhanced melt supply could increase the crustal thickness. In the revised manuscript, we delete that convincing expression of \(150^{\circ}\mathrm{C}\) temperature reduction in the equatorial Atlantic Ocean.  
+
+Is the methodology sound? Does the work meet the expected standards in your field? - The seismic analysis seems sound and uses a well- established method  
+
+Is there enough detail provided in the methods for the work to be reproduced? - Yes. However, the choice of regularization parameters for the preferred model are not given.  
+
+We have provided more details of the traveltime tomography in the revised manuscript. In our traveltime tomography, both first- and second- order velocity regularizations are imposed to obtain a smooth velocity model [Van Avendonk et al., 2004]. The weight given to the horizontal derivatives is 4 times of that given to the vertical derivative, following Van Avendonk et al. [2004] and Roland et al. [2012]. The regularization parameters are tested and selected in each iteration step to avoid the introduction of artefacts. We use the standard \(\times^{2}\) value [Van Avendonk et al., 2004] to measure the mismatch between the modelled and manually picked travel times. Large regulation values are used at the early stage of tomography and the regulation values are reduced when \(\times^{2}\) value approaches 1. These points have been added in the ‘Methods’ section.  
+
+Detailed comments:  
+
+- Abstract: The two existing models for crustal thickness variations at slow spreading ridges are incorrectly merged together in the phrase (lines 11-12) “due a three-dimensional plume-like mantle upwelling with melt focusing to segment centres”.
+
+<--- Page Split --->
+
+We have modified this sentence.  
+
+- Introduction – issues with the description of crustal accretion at fast spreading ridges: o Line 32:- Delete NTOs: Second order offsets are OSCs at fast spreading ridges and NTOs at slow spreading ridges  
+
+Modified as suggested by the reviewer.  
+
+- This statement ignores the literature that crust is not thin beneath OSCs at the EPR (e.g. Canales et al., 2003).  
+
+Modified as suggested by the reviewer.  
+
+o Lines 33- 34: Incorrect: fast spreading ridges are fed by individual mantle upwellings (e.g. Toomey 2007) and are not just two- dimensional sheet- like mantle upwellings.  
+
+At fast- spreading ridge, the relatively uniform crust is generally interpreted as due to a uniform, 2- D, sheet- like mantle upwelling beneath ridge axis [Lin and Morgan, 1992]. Here the sheet- like mantle upwelling refers to the mantle upwelling within a second- order ridge segments, not the first- order ridge segments between two major transform faults. We agree that the overall mantle upwelling between two major transform faults is not perfectly sheet- like due to the offset of overlapping spreading centres. We went through the paper from Toomey et al. [2007] in Nature. They proposed the mantle upwelling beneath the fast- spreading East- Pacific Rise could be skewed. But Singh and Macdonald [2009] pointed out that the results from Toomey et al. [2007] are not reliable.  
+
+- Supplemental Figures 6 &7: Caption reads that this is for Vp model, figures are labelled as being S models  
+
+Modified as suggested by the reviewer. In these figures, 'S' means south direction. We replotted the figures to avoid confusions.  
+
+In summary, I recommend major revisions so that the manuscript more clearly argues for the main processes that the authors infer from these new observations.
+
+<--- Page Split --->
+
+## References  
+
+Behn, M. D., M. S. Boettcher, and G. Hirth (2007), Thermal structure of oceanic transform faults, Geology, 35(4), 307- 310.  
+
+Behn, M. D., and T. L. Grove (2015), Melting systematics in mid- ocean ridge basalts: Application of a plagioclase- spinel melting model to global variations in major element chemistry and crustal thickness, J. Geophys. Res., 120(7), 4863- 4886.  
+
+Cann, J. R., et al. (1997), Corrugated slip surfaces formed at ridge- transform intersections on the Mid- Atlantic Ridge, Nature, 385(6614), 329- 332.  
+
+Carlson, R. L., and D. J. Miller (1997), A new assessment of the abundance of serpentinite in the oceanic crust, Geophys. Res. Lett., 24(4), 457- 460.  
+
+Christeson, G. L., et al. (2020), South Atlantic Transect: Variations in Oceanic Crustal Structure at \(31^{\circ}\mathrm{S}\) , Geochem. Geophys. Geosyst., 21(7), e2020GC009017.  
+
+Combier, V., et al. (2015), Three- dimensional geometry of axial magma chamber roof and faults at Lucky Strike volcano on the Mid- Atlantic Ridge, J. Geophys. Res., 120(8), 5379- 5400.  
+
+Detrick, R. S., H. D. Needham, and V. Renard (1995), Gravity anomalies and crustal thickness variations along the Mid- Atlantic Ridge between \(33^{\circ}\mathrm{N}\) and \(40^{\circ}\mathrm{N}\) , J. Geophys. Res., 100(B3), 3767- 3787.  
+
+Escartin, J., et al. (1999), Quantifying tectonic strain and magmatic accretion at a slow spreading ridge segment, Mid- Atlantic Ridge, \(29^{\circ}\mathrm{N}\) , J. Geophys. Res., 104(B5), 10421- 10437.  
+
+Escartin, J., and J. Lin (1995), Ridge offsets, normal faulting, and gravity anomalies of slow spreading ridges, J. Geophys. Res., 100(B4), 6163- 6177.  
+
+Escartin, J., et al. (2008), Central role of detachment faults in accretion of slow- spreading oceanic lithosphere, Nature, 455(7214), 790- 794.  
+
+Gregory, E. P. M., S. C. Singh, M. Marjanović, and Z. Wang (2021), Serpentinized peridotite versus thick mafic crust at the Romanche oceanic transform fault, Geology, 49(9), 1132- 1136.  
+
+Grevemeyer, I., et al. (2021), Extensional tectonics and two- stage crustal accretion at oceanic transform faults, Nature, 591(7850), 402- 407.  
+
+Harmon, N., et al. (2018), Marine Geophysical Investigation of the Chain Fracture Zone in the Equatorial Atlantic From the PI- LAB Experiment, J. Geophys. Res., 123(12), 11016- 11030.  
+
+Keller, T., and R. F. Katz (2016), The Role of Volatiles in Reactive Melt Transport in the Asthenosphere, J. Petrol., 57(6), 1073- 1108.  
+
+Korenaga, J., et al. (2000), Crustal structure of the southeast Greenland margin from joint refraction and reflection seismic tomography, J. Geophys. Res., 105(B9), 21591- 21614.  
+
+Le Voyer, M., et al. (2019), Carbon Fluxes and Primary Magma CO2 Contents Along the Global Mid- Ocean Ridge System, Geochem. Geophys. Geosyst., 20(3), 1387- 1424.  
+
+Lin, J., and J. P. Morgan (1992), The spreading rate dependence of three- dimensional mid- ocean ridge gravity structure, Geophys. Res. Lett., 19(1), 13- 16.
+
+<--- Page Split --->
+
+Lin, J., et al. (1990), Evidence from gravity data for focused magmatic accretion along the Mid- Atlantic Ridge, Nature, 344(6267), 627- 632.  
+
+Magde, L. S., and D. W. Sparks (1997), Three- dimensional mantle upwelling, melt generation, and melt migration beneath segment slow spreading ridges, J. Geophys. Res., 102(B9), 20571- 20583.  
+
+Marjanović, M., et al. (2020), Seismic Crustal Structure and Morpho- tectonic Features Associated with the Chain Fracture Zone and their Role in the Evolution of the Equatorial Atlantic Region, J. Geophys. Res., 125, e2020JB020275.  
+
+Roland, E., D. Lizarralde, J. J. McGuire, and J. A. Collins (2012), Seismic velocity constraints on the material properties that control earthquake behavior at the Quebrada- Discovery- Gofar transform faults, East Pacific Rise, J. Geophys. Res., 117(B11), B11102.  
+
+Schilling, J.- G., et al. (1995), Thermal structure of the mantle beneath the equatorial Mid- Atlantic Ridge: Inferences from the spatial variation of dredged basalt glass compositions, J. Geophys. Res., 100(B6), 10057- 10076.  
+
+Searle, R. (2013), Mid- Ocean Ridges, Cambridge University Press, Cambridge.  
+
+Seher, T., et al. (2010), Crustal velocity structure of the Lucky Strike segment of the Mid- Atlantic Ridge at 37°N from seismic refraction measurements, J. Geophys. Res., 115(B3), B03103.  
+
+Shaw, P. R. (1992), Ridge segmentation, faulting and crustal thickness in the Atlantic Ocean, Nature, 358(6386), 490- 493.  
+
+Shaw, P. R., and J. Lin (1993), Causes and consequences of variations in faulting style at the Mid- Atlantic Ridge, J. Geophys. Res., 98(B12), 21839- 21851.  
+
+Singh, S. C., et al. (2006), Discovery of a magma chamber and faults beneath a Mid- Atlantic Ridge hydrothermal field, Nature, 442(7106), 1029- 1032.  
+
+Singh, S. C., and K. C. Macdonald (2009), Mantle skewness and ridge segmentation, Nature, 458(7241), E11- E12.  
+
+Toomey, D. R., et al. (2007), Skew of mantle upwelling beneath the East Pacific Rise governs segmentation, Nature, 446(7134), 409- 414.  
+
+Uditsev, G. (1996), Equatorial segment of the mid- Atlantic ridge, Technical Series. Intergovernmental Oceanographic Commission= Serie technique.  
+
+Van Avendonk, H. J. A., D. J. Shillington, W. S. Holbrook, and M. J. Hornbach (2004), Inferring crustal structure in the Aleutian island arc from a sparse wide- angle seismic data set, Geochem. Geophys. Geosyst., 5(8), Q08008.  
+
+Wang, Z., et al. (2022), Deep hydration and lithospheric thinning at oceanic transform plate boundaries, Nature Geosci., 15, 741- 746.  
+
+Whitehead, J. A., H. J. B. Dick, and H. Schouten (1984), A mechanism for magmatic accretion under spreading centres, Nature, 312(5990), 146- 148.  
+
+Wolfson- Schwehr, M., and M. S. Boettcher (2019), Chapter 2 - Global Characteristics of Oceanic Transform Fault Structure and Seismicity, in Transform Plate Boundaries and Fracture Zones, edited by J. C. Duarte, pp. 21- 59, Elsevier, doi:https://doi.org/10.1016/B978- 0- 12- 812064- 4.00002- 5.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors reacted on every single remark that reviewers gave for the manuscript on uniform crustal accretion along slow- spreading ridges in the equatorial Atlantic Ocean. In my opinion unclear points have been explained, changed or taken away. Changes to text and figures have been made and I would recommend the manuscript for publication in its current state from the point of science.  
+
+Best regards  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors have addressed my initial review thoroughly, and I appreciate the time and effort they put into revisions. I think the revised discussion is much clearer, and shifting the focus from mapping out different types of crustal accretion to more consideration of the mechanisms that might cause this distinctive mode of magmatic accretion in the equatorial Atlantic made the paper more interesting to me, at least.  
+
+I have one tiny comment on wording: line 291 of the revised manuscript says that the crust is "not thick enough, and hence not hot enough" for ductile lower crust to be a factor - I would guess the intended meaning is that thin (magmatic) crust indicates there wasn't excessive magma production, so the mantle temperature wasn't anomalously high and the lower crust is unlikely to have been unusually hot, but the phrasing is not entirely clear.  
+
+Other than that picky note on phrasing, I think the manuscript is in good shape and suitable for publication. The work is significant, the observations support the conclusions, and the discussion is thoughtful.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The revised manuscript has been substantially improved to address the reviewers' comments. In particular, the interpretation and of the results and reasoning much more clearly explains the significance of the results: that the equatorial Atlantic is different from the north and south Atlantic likely due to the effect of long transform faults on the mantle thermal structure and/or the presence of high water and CO2 in the mantle source.  
+
+Two minor revisions remain to be made:  
+
+1. The last sentence of the manuscript (lines 424-425) concludes that 2D upwelling is the norm at slow spreading ridges. However, this is not what the paper argues. Instead the paper argues that there is 2D upwelling in the equatorial Atlantic and while there is significant along axis variation in crustal thickness in the northern and southern Atlantic. Rephrase to be consistent with the reasoning in the text. The same for the text on lines 363-366.  
+
+2. Paragraph on tectonic extension (lines 247-256): The argument here is focused on tectonic modification of the segment ends. I am of the same opinion as Reviewer 1 that the option of tectonic modification and crustal thinning of the segment centers should also be discussed in order to fully address the proposed hypotheses.  
+
+3. I am attaching a marked up PDF with many small English grammar edits for the main body of the text and the figure captions.
+
+<--- Page Split --->
+
+Reviewers' comments are shown in black; authors' response is shown in blue.  
+
+## Reviewer #2:  
+
+The authors have addressed my initial review thoroughly, and I appreciate the time and effort they put into revisions. I think the revised discussion is much clearer, and shifting the focus from mapping out different types of crustal accretion to more consideration of the mechanisms that might cause this distinctive mode of magmatic accretion in the equatorial Atlantic made the paper more interesting to me, at least.  
+
+I have one tiny comment on wording: line 291 of the revised manuscript says that the crust is "not thick enough, and hence not hot enough" for ductile lower crust to be a factor - I would guess the intended meaning is that thin (magmatic) crust indicates there wasn't excessive magma production, so the mantle temperature wasn't anomalously high and the lower crust is unlikely to have been unusually hot, but the phrasing is not entirely clear. Other than that picky note on phrasing, I think the manuscript is in good shape and suitable for publication. The work is significant, the observations support the conclusions, and the discussion is thoughtful.  
+
+We have rephrased this sentence to 'the crust formed at the MAR in the equatorial Atlantic Ocean is much thinner (5.5- 5.6 km) than that formed at the Reykjanes Ridge, indicating that the mantle is colder, and therefore the lower crust is not hot enough to enable rapid ductile flow within the lower crust.'  
+
+## Reviewer #3:  
+
+The revised manuscript has been substantially improved to address the reviewers' comments. In particular, the interpretation and of the results and reasoning much more clearly explains the significance of the results: that the equatorial Atlantic is different from the north and south Atlantic likely due to the effect of long transform faults on the mantle thermal structure and/or the presence of high water and CO2 in the mantle source. Two minor revisions remain to be made:  
+
+1. The last sentence of the manuscript (lines 424-425) concludes that 2D upwelling is the norm at slow spreading ridges. However, this is not what the paper argues. Instead the paper argues that there is 2D upwelling in the equatorial Atlantic and while there is significant along axis variation in crustal thickness in the northern and southern Atlantic. Rephrase to be consistent with the reasoning in the text. The same for the text on lines 363-366.  
+
+We have deleted the sentence on lines 424- 425 and have rephrased the texts on lines 363- 366.
+
+<--- Page Split --->
+
+2. Paragraph on tectonic extension (lines 247-256): The argument here is focused on tectonic modification of the segment ends. I am of the same opinion as Reviewer 1 that the option of tectonic modification and crustal thinning of the segment centers should also be discussed in order to fully address the proposed hypotheses.  
+
+In this paragraph, we discussed the difference in the amounts of tectonic extension and stretching between segment centres and segment ends by comparing the spacing, heave and throw of normal faults. The spacing, heave and throw of normal faults are generally larger at segment ends than at segment centres, indicating that more tectonic extension occurs at segment ends [Shaw, 1992; Shaw and Lin, 1993]. This conclusion can be made without detailing the features of tectonic faulting at the segment centre and ends. Since the tectonic extension thins the oceanic crust [Combier et al., 2015; Escartin et al., 1999; Escartin and Lin, 1995], more tectonic extension means more thinning of oceanic crust would occur at the segment ends. So, the overall influence of tectonic extension on crustal thickness variation would enhance the along-axis crustal thickness variation, contrary to our observation of the uniform crust along the profile.  
+
+3. I am attaching a marked up PDF with many small English grammar edits for the main body of the text and the figure captions.  
+
+We deeply appreciate reviewer for the careful check of English grammar. We have modified accordingly.  
+
+## References  
+
+Combier, V., et al. (2015), Three- dimensional geometry of axial magma chamber roof and faults at Lucky Strike volcano on the Mid- Atlantic Ridge, J. Geophys. Res., 120(8), 5379- 5400.  
+
+Escartin, J., et al. (1999), Quantifying tectonic strain and magmatic accretion at a slow spreading ridge segment, Mid- Atlantic Ridge, \(29^{\circ}\mathrm{N}\) , J. Geophys. Res., 104(B5), 10421- 10437.  
+
+Escartin, J., and J. Lin (1995), Ridge offsets, normal faulting, and gravity anomalies of slow spreading ridges, J. Geophys. Res., 100(B4), 6163- 6177.  
+
+Shaw, P. R. (1992), Ridge segmentation, faulting and crustal thickness in the Atlantic Ocean, Nature, 358(6386), 490- 493.  
+
+Shaw, P. R., and J. Lin (1993), Causes and consequences of variations in faulting style at the Mid- Atlantic Ridge, J. Geophys. Res., 98(B12), 21839- 21851.  
+
+Shaw, P. R. (1992), Ridge segmentation, faulting and crustal thickness in the Atlantic Ocean, Nature, 358(6386), 490- 493.  
+
+Shaw, P. R., and J. Lin (1993), Causes and consequences of variations in faulting style at the Mid- Atlantic Ridge, J. Geophys. Res., 98(B12), 21839- 21851.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__71db9c463f8c5446731402d5aef7b63bae38535e361ac2d5c86ef745bbe875aa/images_list.json

@@ -0,0 +1,453 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "The following is our revisions to the manuscript on Page 12 of revised manuscript:",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "The following is our revisions to the manuscript on Page 14 of revised manuscript:",
+    "footnote": [],
+    "bbox": [
+      [
+        156,
+        245,
+        842,
+        840
+      ]
+    ],
+    "page_idx": 4
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5. The in vivo anticancer performance of DBG under US irradiation. a) Schematic illustration of 4T1 tumor-bearing mouse model establishment and treatment procedures in vivo. b, c) In vivo retentions of 2-DG and BFO in different formulations in different periods post subcutaneous injection (days 0, 3, 5, and 7) into 4T1 tumor-bearing BALB/c mice. d) Body weight of mice in different groups ( \\(n = 5\\) mice per group) in the 28-day treatment period. e) Average volume changes of the tumor of mice in different groups. f) Average tumor weight of mice in different groups ( \\(n = 5\\) mice per group) at the ends of treatments. g) Morbidity-free survival of different groups of mice after different treatments ( \\(n = 5\\) mice per group). h) Tumor volume change curves of each mouse after various treatments ( \\(n = 5\\) mice per group). i) Representative tumor immunofluorescence staining images of ROS, Ki67, and TUNEL after different treatments. j, k) Western blot analysis of the expressions of GRP78 in tumor tissues after different treatments and corresponding quantitative analysis. l-m) The GO enrichment and further MF analysis between the control and DBG + US groups. Data were expressed as means \\(\\pm \\mathrm{SD}\\) in d-f, and k. \\(*P< 0.01\\) , \\(**P< 0.001\\) , \\(***P< 0.0001\\) . Source data are provided as a Source Data file.",
+    "footnote": [],
+    "bbox": [
+      [
+        176,
+        90,
+        816,
+        558
+      ]
+    ],
+    "page_idx": 5
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Figure 6. Immune activation evaluations in vivo. FC analyses and quantification results in vivo of (a, b) M1 macrophages (gated on CD11b+/F4/80+ macrophages) and (c, d) mature DCs (gated on CD11c+ DC cells) and (e, f) CD4+/CD8+ T cells (gated on CD3+ T cells) in tumors of mice after different treatments ( \\(n = 5\\) biologically independent samples). g) tSNE visualization and quantification of all immune cells in the tumor by cytometry from 4T1 tumor-bearing BALB/c mice after different treatments, h-1) Expressions of various cytokines in tumor tissues of mice after different treatments measured by ELISA. Data were expressed as means \\(\\pm\\) SD in b, d, and f. \\(*P< 0.01\\) , \\(**P< 0.001\\) , \\(***P< 0.0001\\) . Source data are provided as a Source Data file.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        87,
+        840,
+        612
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Figure S13. Immunofluorescence images of proliferated cytotoxic T lymphocytes (CD3+/CD8+) and polarized M1 TAMs (F4/80+/CD86+) in 4T1 tumor tissue slices.",
+    "footnote": [],
+    "bbox": [
+      [
+        149,
+        90,
+        847,
+        243
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Figure S1. The change curves of the absorption intensity at \\(370\\mathrm{nm}\\) of TMB solutions containing equal BFO and \\(\\mathrm{BaTiO_3}\\) US irradiation for varied durations. \\(\\mathrm{n} = 3\\) independent experiments and data are presented as mean \\(\\pm \\mathrm{SD}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        328,
+        305,
+        670,
+        511
+      ]
+    ],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3. The in vitro anticancer and ICD-triggering performances of BFO nanosheets. a) CLSM microscopic images of 4T1 cancer cells stained with Calcein-AM/PI after treatments under different conditions. b) tSNE visualization and quantification of apoptotic 4T1 cells by cytometry after different treatments. c) Frequency of PHA-L binding to 4T1 cancer cells treated with different concentrations of DBG plus US irradiation ( \\(n = 5\\) biologically independent samples). d, e) Western blot analysis and corresponding quantifications of 4T1 cancer cell lysates showing expression of the ER stress protein GRP78 upon treatment with the indicated drugs ( \\(n = 3\\) biologically independent samples). f) FC analysis of CRT expression of 4T1 cancer cells after different treatments. g) Quantifications of CRT, HMGB1, and HSP70 fluorescence intensity in 4T1 cancer cells after different treatments ( \\(n = 3\\) biologically independent samples). h) Immunofluorescence images of CRT, HMGB1, and HSP70 of 4T1 cancer cells after different treatments. Data were expressed as means \\(\\pm \\mathrm{SD}\\) in c, e, and g. \\(*P< 0.01\\) , \\(**P< 0.001\\) , \\(***P< 0.0001\\) . Source data are provided as a Source Data file.",
+    "footnote": [],
+    "bbox": [
+      [
+        145,
+        88,
+        850,
+        636
+      ]
+    ],
+    "page_idx": 14
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5. The in vivo anticancer performance of DBG under US irradiation. a) Schematic illustration of 4T1 tumor-bearing mouse model establishment and treatment procedures in vivo. b, c) In vivo retentions of 2-DG and BFO in different formulations in different periods post subcutaneous injection (days 0, 3, 5, and 7) into 4T1 tumor-bearing BALB/c mice. d) Body weight of mice in different groups ( \\(n = 5\\) mice per group) in the 28-day treatment period. e) Average volume changes of the tumor of mice in different groups. f) Average tumor weight of mice in different groups ( \\(n = 5\\) mice per group) at the ends of treatments. g) Morbidity-free survival of different groups of mice after different treatments ( \\(n = 5\\) mice per group). h) Tumor volume change curves of each mouse after various treatments ( \\(n = 5\\) mice per group). i) Representative tumor immunofluorescence staining images of ROS, Ki67, and TUNEL after different treatments. j, k) Western blot analysis of the expressions of GRP78 in tumor tissues after different treatments and corresponding quantitative analysis. l-m) The GO enrichment and further MF analysis between the control and DBG + US groups. Data were expressed as means \\(\\pm \\mathrm{SD}\\) in d-f, and k. \\(*P< 0.01\\) , \\(**P< 0.001\\) , \\(***P< 0.0001\\) . Source data are provided as a Source Data file.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        88,
+        838,
+        595
+      ]
+    ],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_7.jpg",
+    "caption": "Figure 7. Effectiveness in combating distant tumors. a) Schematic illustration of 4T1 bilateral tumor inoculation and treatment procedures in vivo. b) Body weight variations of mice in different groups ( \\(n = 5\\) mice per group) in the 28-day treatment period. Average volume and weight changes of (c, d) primary tumors and (e, f) distant tumors of mice in different groups ( \\(n = 5\\) mice per group) at the ends of treatments. g) Volume change curves of distant tumors in each mouse after different treatments ( \\(n = 5\\) mice per group). h) tSNE visualization and quantification of all immune cells in distant tumors by cytometry from 4T1 tumor-bearing BALB/c mice after different treatments. i) Immunofluorescence images of proliferated cytotoxic T lymphocytes ( \\(\\mathrm{CD3^{+} / CD8^{+}}\\) ) and polarized M1 TAMs ( \\(\\mathrm{F4 / 80^{+} / CD86^{+}}\\) ) in 4T1 tumor tissue slices of tumor. Data were expressed as means \\(\\pm\\) SD in b-f. \\(*P< 0.01\\) , \\(**P< 0.001\\) , \\(***P< 0.0001\\) . Source data are provided as a Source Data file.",
+    "footnote": [],
+    "bbox": [
+      [
+        160,
+        85,
+        830,
+        675
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "Figure S8. a) Schematic illustration of the orthotopic osteosarcoma-bearing mouse model",
+    "footnote": [],
+    "bbox": [
+      [
+        152,
+        530,
+        848,
+        880
+      ]
+    ],
+    "page_idx": 18
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_5.jpg",
+    "caption": "Figure S9. Levels of MDA in tumor tissues of mice after different treatments. Analysis of the excised main organs and tumor slices of 4T1-bearing mice after different treatments by H&E staining. Data were expressed as means \\(\\pm\\) SD \\((n = 5)\\) . \\(*P< 0.01\\) , \\(**P< 0.001\\) , \\(***P< 0.0001\\) . Source data are provided as a Source Data file.",
+    "footnote": [],
+    "bbox": [
+      [
+        330,
+        85,
+        666,
+        295
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3. The in vitro anticancer and ICD-triggering performances of BFO nanosheets. a) CLSM microscopic images of 4T1 cancer cells stained with Calcein-AM/PI after treatments under different conditions. b) tSNE visualization and quantification of apoptotic 4T1 cells by cytometry after different treatments. c) Frequencies of PHA-L binding to 4T1 cancer cells treated with different concentrations of DBG plus US irradiation ( \\(n = 5\\) biologically independent samples). d, e) Western blot analysis and corresponding quantifications of 4T1 cancer cell lysates showing the expressions of the ER stress protein GRP78 upon treatments with the indicated drugs ( \\(n = 3\\) biologically independent samples). f) FC analysis of CRT expression of 4T1 cancer cells after different treatments. g) Intracellular ATP levels of 4T1 cells after different treatments ( \\(n = 5\\) biologically independent samples). h) Immunofluorescence images of CRT, HMGB1, and HSP70 of 4T1 cancer",
+    "footnote": [],
+    "bbox": [
+      [
+        160,
+        88,
+        838,
+        625
+      ]
+    ],
+    "page_idx": 21
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_6.jpg",
+    "caption": "Figure S11. Immunofluorescence analysis of the expressions of CRT, and HSP70, as well as the release of HMGB1 from tumor tissues of a) babl/c mice and b) NOG mice after different treatments.",
+    "footnote": [],
+    "bbox": [
+      [
+        170,
+        158,
+        833,
+        610
+      ]
+    ],
+    "page_idx": 28
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_7.jpg",
+    "caption": "Figure S6. Western blot analysis and corresponding quantifications of 4T1 cancer cell lysates showing expression of the ER stress protein IRE1-a, XBP1, P-eIF2A, and PERK upon treatments with the indicated drugs.",
+    "footnote": [],
+    "bbox": [
+      [
+        180,
+        697,
+        797,
+        825
+      ]
+    ],
+    "page_idx": 29
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_8.jpg",
+    "caption": "Figure S4. In vitro cell viabilities (a) H22 cells and (b) CT26 cells by different treatments (n = 5 independent samples).",
+    "footnote": [],
+    "bbox": [
+      [
+        249,
+        545,
+        745,
+        700
+      ]
+    ],
+    "page_idx": 29
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_9.jpg",
+    "caption": "Figure S12. The body weight, tumor volumes and weights of different types of tumor-bearing mice at the ends of various treatments, including (a-c) H-22 tumor-bearing mice model and (d-f) CT26 tumor-bearing mice model ( \\(n = 5\\) mice per group).",
+    "footnote": [],
+    "bbox": [
+      [
+        186,
+        95,
+        810,
+        350
+      ]
+    ],
+    "page_idx": 30
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_10.jpg",
+    "caption": "Figure S10. (a) Phagocytosis ratio of 4T1 cancer cells determined by FC. (b-d) Levels of various cytokines in treated BMDCs suspensions measured by ELISA method ( \\(n = 3\\) biologically independent samples).",
+    "footnote": [],
+    "bbox": [
+      [
+        228,
+        99,
+        772,
+        333
+      ]
+    ],
+    "page_idx": 31
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_11.jpg",
+    "caption": "Figure S20. (a) Gating strategies for isolating F4/80<sup>+</sup>/CD86<sup>+</sup> M1 macrophages from tumor tissue.",
+    "footnote": [],
+    "bbox": [
+      [
+        297,
+        90,
+        700,
+        481
+      ]
+    ],
+    "page_idx": 32
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_12.jpg",
+    "caption": "Figure S21. Gating strategies and tSNE visualization for different immune cells from tumor tissue.",
+    "footnote": [],
+    "bbox": [
+      [
+        258,
+        88,
+        737,
+        450
+      ]
+    ],
+    "page_idx": 33
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_13.jpg",
+    "caption": "Figure S35. Gating strategies for CD62L-/CD44+ \\(\\mathrm{T_{CM}}\\) and CD62L-/CD44+ \\(\\mathrm{T_{EM}}\\) cells from spleen tissue of mice.",
+    "footnote": [],
+    "bbox": [
+      [
+        207,
+        494,
+        789,
+        661
+      ]
+    ],
+    "page_idx": 34
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_14.jpg",
+    "caption": "Figure S31. Average volume changes of the rechallenged tumors of C57 mice in different groups \\((n = 5\\) mice per group). b) Volume change curves of rechallenged tumors in each mouse after different treatments.",
+    "footnote": [],
+    "bbox": [
+      [
+        186,
+        653,
+        790,
+        795
+      ]
+    ],
+    "page_idx": 34
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_15.jpg",
+    "caption": "Figure S32. Average volume changes of the tumors of nude mice in different groups (n = 5 mice per group). b) Volume change curves of tumors in each mouse after different treatments.",
+    "footnote": [],
+    "bbox": [
+      [
+        192,
+        90,
+        789,
+        205
+      ]
+    ],
+    "page_idx": 35
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_16.jpg",
+    "caption": "Figure S25. Flow cytometry analysis of in vivo a) CD206+ macrophages (gated on CD11b+/F4/80+ macrophages), b) CD86+ macrophages (gated on CD11b+/F4/80+ macrophages), c) CD4+ T cells (gated on CD3+ T cells), and d) CD4+ T cells (gated on CD3+ T cells) in tumors of mice after different treatments.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_17.jpg",
+    "caption": "Figure S7. Quantifications of extracellular lactate for 4T1 cells treated with 2DG at the concentration of \\(2\\mathrm{mM}\\) for different time intervals ( \\(\\mathrm{n} = 3\\) independent samples).",
+    "footnote": [],
+    "bbox": [
+      [
+        384,
+        161,
+        616,
+        308
+      ]
+    ],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_18.jpg",
+    "caption": "Figure S4. In vitro cell viabilities (a) H22 cells and (b) CT26 cells by different treatments (n = 5 independent samples).",
+    "footnote": [],
+    "bbox": [
+      [
+        220,
+        95,
+        775,
+        264
+      ]
+    ],
+    "page_idx": 38
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_19.jpg",
+    "caption": "Figure S12. The body weight, tumor volumes and weights of different types of tumor-bearing mice at the ends of various treatments, including (a-c) H-22 tumor-bearing mice model and (d-f) CT26 tumor-bearing mice model ( \\(n = 5\\) mice per group).",
+    "footnote": [],
+    "bbox": [
+      [
+        180,
+        445,
+        808,
+        703
+      ]
+    ],
+    "page_idx": 41
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_20.jpg",
+    "caption": "Following are our revisions to the revised manuscript on Page 11 and Page 19:",
+    "footnote": [],
+    "bbox": [
+      [
+        162,
+        250,
+        812,
+        383
+      ]
+    ],
+    "page_idx": 42
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_21.jpg",
+    "caption": "Figure S26. Quantification of the number of target immune cells in tumor tissues of mice in different groups.",
+    "footnote": [],
+    "bbox": [
+      [
+        231,
+        92,
+        764,
+        260
+      ]
+    ],
+    "page_idx": 42
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_22.jpg",
+    "caption": "Figure S11. (a) Images of the excised main organs and (b) tumor slices of 4T1-bearing mice after different treatments by H&E staining.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        95,
+        846,
+        655
+      ]
+    ],
+    "page_idx": 43
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_23.jpg",
+    "caption": "Figure S27. (a) H&E and (b) Ki67 staining of lung tissue from different groups of 4T1 tumor-bearing mice.",
+    "footnote": [],
+    "bbox": [
+      [
+        278,
+        90,
+        760,
+        393
+      ]
+    ],
+    "page_idx": 44
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_24.jpg",
+    "caption": "Figure S23. (a, b) Western blot analysis and corresponding quantifications of tumor tissues showing the expressions of the TNF-α, IL6, and TLR4 proteins before and after DBG + US treatment. (c) Immunofluorescence staining of TNF-α, IL6, and TLR4 in tumour tissues before and after DBG + US treatment.",
+    "footnote": [],
+    "bbox": [
+      [
+        198,
+        97,
+        789,
+        412
+      ]
+    ],
+    "page_idx": 45
+  }
+]

peer_reviews/supplementary_0_Peer Review File__71db9c463f8c5446731402d5aef7b63bae38535e361ac2d5c86ef745bbe875aa/supplementary_0_Peer Review File__71db9c463f8c5446731402d5aef7b63bae38535e361ac2d5c86ef745bbe875aa.mmd

@@ -0,0 +1,705 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Ultrasound- triggered and Glycosylation Inhibition- enhanced Tumor Piezocatalytic Immunotherapy  
+
+![](images/Figure_unknown_0.jpg)
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+This research reported a reactive oxygen species (ROS)- sensitive scaffold (DBG), comprised of the synergistic components of piezocatalytic bismuth ferrite (BiFeO3) nanosheets and glucose/mannose analog 2- deoxy- d- glucose (2- DG), for enhanced cancer piezocatalytic immunotherapy. Based on four mouse cancer models, the authors demonstrated the synergy effects of ultrasound- triggered piezo effects and glycosylation inhibition with an enhanced tumor piezocatalytic immunotherapy. This work provides a promising piezocatalytic immunotherapy for malignant solid tumors featuring both low immunogenicity and high levels of N-glycosylation. Thus, it can be considered for publication after addressing the following issues.  
+
+1. There are many immunosuppressive factors in the tumor microenvironment. Why did the author choose this one? What is the important role of Glycosylation Inhibition? Which type of tumor is a priority for Glycosylation Inhibition?  
+
+2. The quantitative analysis of immunofluorescence intensity data for Fig3h should be provided.  
+
+3. The fluorescent markers represented by the different colors in Fig 4b should be labeled directly in the figure so that the reader can better understand the cell types represented by the colors.  
+
+4. According to the article, the deformation of the BFO is caused by the pressure generated by ultrasound. How much pressure can ultrasonic waves generate?  
+
+5. What is the power of ultrasound irradiation during in vivo treatment? This parameter should be specified in the experimental protocol.  
+
+6. This work showed many interesting flow cytometry results, so it was recommended that the experimental procedure be described in more detail in the "Methods" allow subsequent researchers to learn and repeat.  
+
+7. Some figures are too small to distinguish. Please enlarge these figures for a clearer presentation.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors show the potency of a Nanocatalytic immunotherapy in which a nanosheet is used for change local tumor environment by interfering with reactive oxygen species (ROS)-  
+
+sensitive scaffold (DBG) and and glucose/mannose analog 2- deoxy- d- glucose  
+
+(2- DG), to alter glycosylation, excited with ultrasound irradiation. The nanosheets have great impact to control tumor growth in various tumor models.
+
+<--- Page Split --->
+
+However the paper is insufficient in the immunological read outs of macrophages, dendritic T cell subsets. Also the impact on glycosylation is not investigated or shown. Thus the therapeutic mode of their nanosheet is very effective the proofs that indeed Myeloid cells are more inflammatory (M1) is not measured in the way the authors describe in Figure 4 with single markers.  
+
+Also no tumor specific CD8 T cells are measured. The immunological readouts are quite simplistic and could be more informative if more markers and also tumor specificity could be addressed to really show that the immune subsets are altered. Also on their glycosylation effects no proofs or checks have been performed that they really change N- Glycosylations.  
+
+Reviewer #3 (Remarks to the Author):  
+
+In this study, the authors design a ROS- sensitive scaffold (DBG) comprised of the synergistic components of piezocatalytic BiFeO3 nanosheets and 2- deoxy- d- glucose (2- DG) for enhanced cancer piezocatalytic immunotherapy. After the implantation in the tumors, BiFeO3 generates ROS and piezoelectric signals under ultrasound irradiation to activate anti- tumor immune responses, and the release of 2- DG disrupts the N- glycans synthesis to further overcome the immunosuppression of tumor. Overall, this study is interesting and organized, but some issues should be addressed before publication:  
+
+1. ROS have strong cytotoxicity, please modify the expression that 'the generating ROS from piezocatalytic BFO could simultaneously and significantly promote the activation of cytotoxic T lymphocytes and M1 polarization of tumor-associated macrophages'.  
+
+2. Do the authors compare the piezocatalytic effects of BiFeO3 with other piezoelectric materials like BaTiO3? Please state the advantages of using BiFeO3 rather than others in this system.  
+
+3. In Fig. 3A, the normal 4T1 cells should not be circular morphology. Please repeat this experiment. Detailed information of stained markers should be marked in Fig. 4b&c. The sample numbers (n) should be presented in many Figures like Fig. 3e&g, 4g, 6b&d&f and several figures in SI.  
+
+4. This ROS-sensitive hydrogel scaffold would consume the BFO-generated ROS under ultrasound irradiation. Thus, what percentage of generating ROS is used for cancer therapy? How to balance the two functions?  
+
+5. From Fig. S8, the figure orders are wrong, please correct them. In page 25, 'Fig. S179' also need to be corrected.  
+
+6. In Fig. 5g&h and Fig. 7e&g, please present the tumor growth curve with long time rather than 14d, especially BG+US and DBG+US groups. Only 14d observation period is hard to obtain the significant difference between these two groups.  
+
+7. Deep tissue penetration capability (on the order of centimeters) is the main advantage of ultrasound. The authors should further investigate the piezocatalytic therapy of DBG on deep-sited tumors (e.g., osteosarcoma or orthotopic liver tumors).  
+
+8. Do the authors notice the BiFeO3-induced ferroptosis during the cancer therapy? This process would also promote the generation of ROS.
+
+<--- Page Split --->
+
+## Response to reviewer I.  
+
+## Comments from reviewer I:  
+
+This research reported a reactive oxygen species (ROS)- sensitive scaffold (DBG), comprised of the synergistic components of piezocatalytic bismuth ferrite (BiFeO3) nanosheets and glucose/mannose analog 2- deoxy- d- glucose (2- DG), for enhanced cancer piezocatalytic immunotherapy. Based on four mouse cancer models, the authors demonstrated the synergy effects of ultrasound- triggered piezo effects and glycosylation inhibition with an enhanced tumor piezocatalytic immunotherapy. This work provides a promising piezocatalytic immunotherapy for malignant solid tumors featuring both low immunogenicity and high levels of N- glycosylation. Thus, it can be considered for publication after addressing the following issues.  
+
+Response: Thank you very much for the insightful comments and suggestions. We have revised the manuscript carefully and provided detailed responses below in a point- by- point manner according to your suggestions.  
+
+1. There are many immunosuppressive factors in the tumor microenvironment. Why did the author choose this one? What is the important role of Glycosylation Inhibition? Which type of tumor is a priority for Glycosylation Inhibition?  
+
+Response: Thank you for the professional comment. As you stated and as described in the manuscript, many factors contribute to immunosuppression, including overexpression of immune checkpoints, hypoxia, and abnormalities of amino acid metabolism in tumors, which help tumors to achieve immune escape by impairing the activity of immune effector cells. Tumor growth is accompanied by cellular glycosylation, which would deteriorate the immune anticancer responses by masking neo- epitopes onto immune cells or disrupting their functions, ultimately resulting in the immunosuppression of tumors. Inhibition of tumor glycosylation can destroy the protective shield of cancer cells and re- expose them to the attack by immune effector cells, which can effectively enhance the immunotherapeutic efficacy. Studies on this scheme have rarely been reported. In addition, glycosylation is one of the most frequently occurring protein modifications (Sci. Transl. Med. 2022, 14, eabg3072). Therefore, we try to enhance the effect of piezocatalytic immunotherapy by inhibiting tumor glycosylation. The most important role of glycosylation inhibition is to amplify the immune antitumor effect under the activation by piezocatalytic therapy. Extremely malignant proliferating tumors such as breast cancer, pancreatic cancer, and melanoma may be hypersensitive to glycosylation inhibition due to their higher levels of glucose metabolism than others.  
+
+2. The quantitative analysis of immunofluorescence intensity data for Fig3h should be provided.  
+
+Response: Thank you for the professional comment. We have provided the quantitative analysis of immunofluorescence intensity data for Fig. 3h in Fig. 3g of the revised manuscript.
+
+<--- Page Split --->
+![](images/Figure_unknown_1.jpg)
+
+<center>The following is our revisions to the manuscript on Page 12 of revised manuscript: </center>  
+
+Figure 3. The in vitro anticancer and ICD- triggering performances of BFO nanosheets. a) CLSM microscopic images of 4T1 cancer cells stained with Calcein- AM/PI after treatments under different conditions. b) tSNE visualization and quantification of apoptotic 4T1 cells by cytometry after different treatments. c) Frequency of PHA- L binding to 4T1 cancer cells treated with different concentrations of DBG plus US irradiation ( \(n = 5\) biologically independent samples). d, e) Western blot analysis and corresponding quantifications of 4T1 cancer cell lysates showing expression of the ER stress protein GRP78 upon treatment with the indicated drugs ( \(n = 3\) biologically independent samples). f) FC analysis of CRT expression of 4T1 cancer cells after different treatments. g) Quantifications of CRT, HMGB1, and HSP70 fluorescence intensity in 4T1 cancer cells after different treatments ( \(n = 3\) biologically independent samples). h) Immunofluorescence images of CRT, HMGB1, and HSP70 of 4T1 cancer cells after different treatments. Data were expressed as means \(\pm \mathrm{SD}\) in c, e, and g. \(*P< 0.01\) , \(**P< 0.0001\) , \(***P< 0.001\) . Source data are provided as a Source Data file.
+
+<--- Page Split --->
+
+3. The fluorescent markers represented by the different colors in Fig 4b should be labeled directly in the figure so that the reader can better understand the cell types represented by the colors.  
+
+Response: We appreciate your careful review. As suggested, the fluorescent markers represented by the different colors in Fig. 4b have been labeled directly.  
+
+![](images/Figure_5.jpg)
+
+<center>The following is our revisions to the manuscript on Page 14 of revised manuscript: </center>  
+
+<center>Figure 4. The evaluation of in vitro immune response activation by DBG and phagocytic behavior of macrophages towards tumor cells subjected to different treatments. a) Schematics of the assay to study immune cell responses in vitro. BMDCs and BMMs were cultured in the lower </center>
+
+<--- Page Split --->
+
+chamber, whereas pretreated 4T1 cells and their residues were added in the upper chamber. The CLSM images of b) BMDCs stained by PE- Anti CD11c and FITC- Anti CD86 and c) BMMs stained by PE- Anti F4/80 and FITC- Anti CD86. Photo- microscopy image of d) BMDCs and e) BMMs after different treatments. f) Schematic illustration of the design of in vitro phagocytosis assay. g, h) Phagocytosis ratio of 4T1 cancer cells determined by FC \((n = 3\) biologically independent samples). i) Representative CLSM images of CFSE- stained 4T1 cancer cells phagocytosed by PE- Anti CD11b- stained RAW264.7 cells. Data were expressed as means \(\pm \mathrm{SD}\) in g. \(^{*}P< 0.01\) , \(^{**}P< 0.001\) , \(^{***}P< 0.0001\) . Source data are provided as a Source Data file.  
+
+4. According to the article, the deformation of the BFO is caused by the pressure generated by ultrasound. How much pressure can ultrasonic waves generate?  
+
+Response: We appreciate your careful review and kind question. When US is irradiated into the water, gas bubbles will be generated owing to the cavitation effect. As the generated bubble bursts, a shock wave is generated. In general, it is known that the pressure of the shock wave produced by the US at 1 MHz or less is approximately 50 MPa (Adv. Mater. 2023, 35(18), 2300437; Ultrason. Sonochem. 2011, 18, 864).  
+
+5. What is the power of ultrasound irradiation during in vivo treatment? This parameter should be specified in the experimental protocol.  
+
+Response: We appreciate your careful review and kind suggestion. The power density of ultrasound irradiation during in vivo treatment was \(1\mathrm{Wcm}^{-2}\) . As suggested, this parameter had been specified in the experimental protocol.  
+
+## Following are our revisions to the manuscript on Page 31:  
+
+Tumor models and treatment. An orthotopic breast cancer model was established in female BALB/c mice by injecting 4T1 cells into the mammary fat pad of mice \((5 \times 10^{7}\) cells/mL). 4T1 tumor- bearing mice were randomly divided into five groups \((n = 16)\) , including (1) Control, (2) US, (3) DBG, (4) BG + US, and (5) DBG + US. All of the US irradiations (1 MHz, \(1\mathrm{Wcm}^{-2}\) , \(50\%\) duty cycle, 5 min) were performed on days 1 and 3 after scaffold injection.  
+
+6. This work showed many interesting flow cytometry results, so it was recommended that the experimental procedure be described in more detail in the "Methods" allow subsequent researchers to learn and repeat.  
+
+Response: We appreciate your valuable comments and have supplemented the details of the method with the changes being highlighted.  
+
+## Following are our revisions to the manuscript on Page 33:  
+
+Flow cytometry analysis. Single- cell suspensions obtained from the lymph node, tumor, spleen, and bone marrow of BALB/c mice were prepared by gentle washing and passing through a \(70\mu \mathrm{m}\) mesh cell strainer. Subsequently, the cells were incubated with
+
+<--- Page Split --->
+
+Fc receptor-blocking reagent (BD Biosciences) for \(15\mathrm{min}\) at \(4^{\circ}\mathrm{C}\) and then stained with the following fluorochrome-conjugated antibodies staining protocol: CD45- PE- Cy5, CD11c- Pacific Blue, CD80- Alexa Fluor 647, CD86- APC for analyzing maturation of DCs; CD45- PE- Cy5, CD11b- Alexa Fluor 594, CD- F4/80- Pacific Blue, CD206- Alexa Fluor 647, CD86- APC for analyzing polarization of TAMs; CD45- PE- Cy5, CD3- PE, CD8- PerCP- Cy5.5, CD4- FITC for analyzing proliferation of \(\mathrm{CD8^{+} / CD4^{+} T}\) cells; CD45- PE- Cy5, CD44- Alexa Fluor 647, CD4- FITC, CD8- PerCP- Cy5.5, and CD62L- APC for analyzing activation of immune memory cells; CD45- PE- Cy5, CD19- Alexa Fluor 647 for analyzing activation of B cells; CD45- PE- Cy5, CD49- Alexa Fluor 675 for analyzing activation of NK cells. After 30 minutes of staining, the single- cell suspension was washed twice with PBS/ \(2\%\) FBS and then stained with PI. Finally, the phenotype of different immune cells was analyzed by FCM. All data were processed using the flow cytometry analysis software (FlowJo, version 10). All antibodies were purchased from BioLegend, eBioscience, and R&D Systems.  
+
+7. Some figures are too small to distinguish. Please enlarge these figures for a clearer presentation.  
+
+Response: Thank you very much for your constructive suggestion. We are sorry that these figures are not clear enough, and some of them are indeed too small because these figures have to be inserted into the manuscript template for submission. We will upload the clear original images for the editor and reviewers to review.
+
+<--- Page Split --->
+
+## Response to reviewer II.  
+
+## Comments from reviewer II:  
+
+The authors show the potency of a Nanocatalytic immunotherapy in which a nanosheet is used for change local tumor environment by interfering with reactive oxygen species (ROS)- sensitive scaffold (DBG) and and glucose/mannose analog 2- deoxy- d- glucose (2- DG), to alter glycosylation, excited with ultrasound irradiation. The nanosheets have great impact to control tumor growth in various tumor models. However the paper is insufficient in the immunological read outs of macrophages, dendritic cell subsets. Also the impact on glycosylation is not investigated or shown. Thus the therapeutic mode of their nanosheet is very effective the proofs that indeed Myeloid cells are more inflammatory (M1) is not measured in the way the authors describe in Figure 4 with single markers. Also no tumor specific CD8 T cells are measured. The immunological readouts are quite simplistic and could be more informative if more markers and also tumor specificity could be addressed to really show that the immune subsets are altered. Also on their glycosylation effects no proofs or checks have been performed that they really change N- Glycosylations.  
+
+Response: Thank you very much for the insightful comments and suggestions. We have revised the manuscript carefully and provided detailed responses below according to your comments. As suggested, we have systematically measured the changes of tumor- specific immune cell subsets after different treatments, including macrophages, DCs, and effector T cells, by using FC where the DCs were marked by AntiCD45/CD11c/CD80/CD86 antibodies, macrophages were marked by AntiCD45/CD11b/F4/80/CD206/CD86 antibodies, and effector T cells were marked by Anti- CD45/CD3/CD4/CD8 antibodies, respectively.  
+
+From the FC results in Figs. 6a, b, PCT treatment (BG + US or DBG + US) resulted in a significant M1 polarization of TAMs in the mouse tumors. The maturation of DCs in draining lymph nodes is essential for presenting antigens to T cells and activating the T cell effector. Due to the pronounced ICD in the tumor tissue of the mice after the PCT treatment, the percentage of mature DCs in the BG + US and DBG + US groups reached \(21\%\) and \(26\%\) , respectively, which were significantly higher than those in the other groups (Figs. 6c, d). Correspondingly, significantly increased percentages of tumor- specific CD4 \(^+\) T and CD8 \(^+\) T cells were detected in the tumors from the PCT- treated mice (Figs. 6e, f). In addition, the infiltration of immune cells into the mouse tumors was evaluated 3 days after the last treatment. As expected, in the US irradiation- only and DBG injection- only treatments, no significant changes in the intratumoral immune cell populations (e.g., TAMs, DCs, NK cells, B cells, and T cells) can be found, compared with the injection of PBS, exhibiting a "cold" immune status of tumor (Fig. 6g). However, the treatment with BG under US irradiation significantly increased the TAMs, DCs, NK cells, B cells, and T cells infiltrations to 5.6, 4.9, 4.6, 9.2, and \(9.6\%\) , respectively, compared to 0.7, 1.8, 1, 0.8, 0.9, and \(1.6\%\) in the control group. Notably, the immune cell infiltration in the tumor was further elevated in the mice once treated with US- excited DBG due to the additional effects of N- glycosylation inhibition by the released 2- DG. The comprehensive activation of the immune system by the DBG + US
+
+<--- Page Split --->
+
+therapeutic approach was further evidenced by the substantial increases in the populations of infiltrated F4/80+/CD86+ macrophages and \(\mathrm{CD3^{+} / CD8^{+}}\) T cells in tumor tissues, as observed by immunofluorescence (Fig. S13).  
+
+We have tried to evaluate the ability of DBG to inhibit tumor N- glycosylation in vivo by western blot and liquid chromatography tandem- mass spectrometry (LC- MS/MS) analysis. From the western blot analysis (Figs. 5j, k), the treatment of 4T1 tumors with DBG + US exhibited a marked up- regulation of the ER stress marker GRP78 induced by the inhibition of N- glycosylation. The BG + US treatments also slightly upregulated the GRP78 expression of the tumor, possibly due to the ROS- induced ER stress. The quantitative results of tumor N- glycosylation levels also showed that DBG+US treatment did inhibit tumor N- glycosylation levels, with a total of 865 proteins being involved in altering N- glycosylation levels after DBG+US treatment, of which 456 were significantly downregulated (Figs. S10). The results of GO enrichment and further MF analysis between the control and DBG + US groups are presented in Figs 5l, m. The enrichment of BPs was mainly involved in cellular and biological processes, MFs were associated with protein activity, especially "protein binding," "protein complex binding," and "identical protein binding". Therefore, DBG + US treatment can alter the functions of associated proteins on the surface of tumor cells by inhibiting N- glycosylation, thereby enhancing their mediated immunotherapeutic effects.
+
+<--- Page Split --->
+![](images/Figure_6.jpg)
+
+<center>Figure 5. The in vivo anticancer performance of DBG under US irradiation. a) Schematic illustration of 4T1 tumor-bearing mouse model establishment and treatment procedures in vivo. b, c) In vivo retentions of 2-DG and BFO in different formulations in different periods post subcutaneous injection (days 0, 3, 5, and 7) into 4T1 tumor-bearing BALB/c mice. d) Body weight of mice in different groups ( \(n = 5\) mice per group) in the 28-day treatment period. e) Average volume changes of the tumor of mice in different groups. f) Average tumor weight of mice in different groups ( \(n = 5\) mice per group) at the ends of treatments. g) Morbidity-free survival of different groups of mice after different treatments ( \(n = 5\) mice per group). h) Tumor volume change curves of each mouse after various treatments ( \(n = 5\) mice per group). i) Representative tumor immunofluorescence staining images of ROS, Ki67, and TUNEL after different treatments. j, k) Western blot analysis of the expressions of GRP78 in tumor tissues after different treatments and corresponding quantitative analysis. l-m) The GO enrichment and further MF analysis between the control and DBG + US groups. Data were expressed as means \(\pm \mathrm{SD}\) in d-f, and k. \(*P< 0.01\) , \(**P< 0.001\) , \(***P< 0.0001\) . Source data are provided as a Source Data file. </center>
+
+<--- Page Split --->
+![](images/Figure_unknown_2.jpg)
+
+<center>Figure 6. Immune activation evaluations in vivo. FC analyses and quantification results in vivo of (a, b) M1 macrophages (gated on CD11b+/F4/80+ macrophages) and (c, d) mature DCs (gated on CD11c+ DC cells) and (e, f) CD4+/CD8+ T cells (gated on CD3+ T cells) in tumors of mice after different treatments ( \(n = 5\) biologically independent samples). g) tSNE visualization and quantification of all immune cells in the tumor by cytometry from 4T1 tumor-bearing BALB/c mice after different treatments, h-1) Expressions of various cytokines in tumor tissues of mice after different treatments measured by ELISA. Data were expressed as means \(\pm\) SD in b, d, and f. \(*P< 0.01\) , \(**P< 0.001\) , \(***P< 0.0001\) . Source data are provided as a Source Data file. </center>
+
+<--- Page Split --->
+![](images/Figure_unknown_3.jpg)
+
+<center>Figure S13. Immunofluorescence images of proliferated cytotoxic T lymphocytes (CD3+/CD8+) and polarized M1 TAMs (F4/80+/CD86+) in 4T1 tumor tissue slices. </center>
+
+<--- Page Split --->
+
+## Response to reviewer III.  
+
+## Comments from reviewer III:  
+
+In this study, the authors design a ROS- sensitive scaffold (DBG) comprised of the synergistic components of piezocatalytic \(\mathrm{BiFeO_3}\) nanosheets and 2- deoxy- d- glucose (2- DG) for enhanced cancer piezocatalytic immunotherapy. After the implantation in the tumors, \(\mathrm{BiFeO_3}\) generates ROS and piezoelectric signals under ultrasound irradiation to activate anti- tumor immune responses, and the release of 2- DG disrupts the N- glycans synthesis to further overcome the immunosuppression of tumor. Overall, this study is interesting and organized, but some issues should be addressed before publication:  
+
+Response: Thank you very much for the insightful comments and suggestions. We have revised the manuscript carefully and provided detailed responses below in a point- by- point manner according to your suggestions.  
+
+1. ROS have strong cytotoxicity, please modify the expression that 'the generating ROS from piezocatalytic BFO could simultaneously and significantly promote the activation of cytotoxic T lymphocytes and M1 polarization of tumor-associated macrophages'.  
+
+Response: Thank you very much for your careful review and professional suggestions. As suggested, the expression had been modified to: 'We demonstrated that the BFO- loaded hydrogel scaffold could simultaneously and significantly promote the activation of cytotoxic T lymphocytes and M1 polarization of tumor- associated macrophages under ultrasound irradiation by strong ROS cytotoxicity- induced ICD of tumor cells and local electric stimulations' on page 5 of revised manuscript.  
+
+2. Do the authors compare the piezocatalytic effects of \(\mathrm{BiFeO_3}\) with other piezoelectric materials like \(\mathrm{BaTiO_3}\) ? Please state the advantages of using \(\mathrm{BiFeO_3}\) rather than others in this system.  
+
+Response: Thank you very much for your comments. We compared the piezocatalytic effects of \(\mathrm{BiFeO_3}\) with \(\mathrm{BaTiO_3}\) according to your suggestion by performing the TMB chromogenic assay. As shown in Fig. S1, the increase of TMB absorption intensity at 370 nm by an equal amount of \(\mathrm{BiFeO_3}\) under US irradiation is more pronounced than that of \(\mathrm{BaTiO_3}\) . Such results reveal that the piezocatalytic effect of \(\mathrm{BiFeO_3}\) is stronger than \(\mathrm{BaTiO_3}\) , which is capable of generating more ROS after being exposed to ultrasound irradiation. The two main advantages of using \(\mathrm{BiFeO_3}\) nanosheets in this system rather than other materials such as \(\mathrm{BaTiO_3}\) are: (1) The sheet- like \(\mathrm{BiFeO_3}\) is able to perform greater deformation under ultrasound- mediated pressure than the regular- shaped nanoparticles, which can generate enlarged potential difference favorable for the redox reaction. (2) The band gap of \(\mathrm{BiFeO_3}\) is smaller than that of \(\mathrm{BaTiO_3}\) (2.2 eV VS 3.2 eV), and the separation of electrons and holes is more likely to occur, which benefits the generation of ROS. In addition, we have added a description in the revised manuscript and provided the corresponding data in SI.
+
+<--- Page Split --->
+
+## The following is our revisions to the revised manuscript on Page 10:  
+
+A typical colorimetric assay was performed to investigate the BFO- mediated ROS generation, where the classical piezocatalyst \(\mathrm{BaTiO_3}\) was used as a control. The colorless \(3,3',5,5'\) - tetramethylbenzidine (TMB) was used as an indicator, which could be oxidized by ROS to chromogenic TMB exhibiting a characteristic absorption at \(370\mathrm{nm}^{39}\) . BFO nanosheets show higher piezocatalytic activity than \(\mathrm{BaTiO_3}\) , which may be due to its shorter band gap and higher deformation. As Fig. S1 illustrated, a quicker increase in the absorption intensity of TMB at \(370\mathrm{nm}\) can be observed in the solution containing BFO than in the solution containing \(\mathrm{BaTiO_3}\) under US irradiation.  
+
+![](images/Figure_3.jpg)
+
+<center>Figure S1. The change curves of the absorption intensity at \(370\mathrm{nm}\) of TMB solutions containing equal BFO and \(\mathrm{BaTiO_3}\) US irradiation for varied durations. \(\mathrm{n} = 3\) independent experiments and data are presented as mean \(\pm \mathrm{SD}\) . </center>  
+
+3. In Fig. 3A, the normal 4T1 cells should not be circular morphology. Please repeat this experiment. Detailed information of stained markers should be marked in Fig. 4b&c. The sample numbers (n) should be presented in many Figures like Fig. 3e&g, 4g, 6b&d&f and several figures in SI.  
+
+Response: We truly appreciate your insightful comments. In the alive/dead double staining assay, the cells after different treatments were collected by trypsin digestion and then re- dispersed in PBS for observation, thus the normal 4T1 cells in the figure showed circular morphology. In addition, we repeated this experiment as you suggested, these cells in control groups by direct CLSM observation show a normal pike- shaped morphology. The new images are provided in Fig. 3a of the revised manuscript. The detailed information of stained markers have been marked in Fig. 4b&c. In addition, the sample numbers (n) had been presented in all Figures in the revised manuscript and SI.  
+
+The following is our revisions to the manuscript on Page 12 of revised manuscript:
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Figure 3. The in vitro anticancer and ICD-triggering performances of BFO nanosheets. a) CLSM microscopic images of 4T1 cancer cells stained with Calcein-AM/PI after treatments under different conditions. b) tSNE visualization and quantification of apoptotic 4T1 cells by cytometry after different treatments. c) Frequency of PHA-L binding to 4T1 cancer cells treated with different concentrations of DBG plus US irradiation ( \(n = 5\) biologically independent samples). d, e) Western blot analysis and corresponding quantifications of 4T1 cancer cell lysates showing expression of the ER stress protein GRP78 upon treatment with the indicated drugs ( \(n = 3\) biologically independent samples). f) FC analysis of CRT expression of 4T1 cancer cells after different treatments. g) Quantifications of CRT, HMGB1, and HSP70 fluorescence intensity in 4T1 cancer cells after different treatments ( \(n = 3\) biologically independent samples). h) Immunofluorescence images of CRT, HMGB1, and HSP70 of 4T1 cancer cells after different treatments. Data were expressed as means \(\pm \mathrm{SD}\) in c, e, and g. \(*P< 0.01\) , \(**P< 0.001\) , \(***P< 0.0001\) . Source data are provided as a Source Data file. </center>
+
+<--- Page Split --->
+
+4. This ROS-sensitive hydrogel scaffold would consume the BFO-generated ROS under ultrasound irradiation. Thus, what percentage of generating ROS is used for cancer therapy? How to balance the two functions?  
+
+Response: Thank you very much for your insightful comments. It is well known that ROS are a kind of transiently produced and rapidly diffusing radical species, which cannot be accurately quantified. It could be seen from the semi-quantitative EPR spectra of equal amounts of BFO and DBG under ultrasonic irradiation (Fig. 2k, l), in which no significant difference can be found in the amount of ROS produced by them, suggesting that only a small portion of the BFO- generated ROS has been consumed by the hydrogel scaffold. Thus, most of the BFO- generated ROS can be utilized for cancer therapy, which is also confirmed by the in vitro and in vivo excellent anticancer therapeutic outcomes of DBG + US treatment. The loading amount of BFO in DBG, the duration of US irradiation, and the power density of US are the main factors affecting the activity of DBG in the production of ROS. Therefore, the two functions of DBG- generated ROS for triggering hydrogel degradation and antitumor can be balanced by adjusting the BFO loading amount, US irradiation time, and power density of US.  
+
+5. From Fig. S8, the figure orders are wrong, please correct them. In page 25, 'Fig. S179' also need to be corrected.  
+
+Response: We are very grateful to you for helping us to point out these mistakes. We have corrected these errors in the revised manuscript and double- checked the manuscript carefully.  
+
+6. In Fig. 5g&h and Fig. 7e&g, please present the tumor growth curve with long time rather than 14d, especially BG + US and DBG + US groups. Only 14d observation period is hard to obtain the significant difference between these two groups.  
+
+Response: Thank you for your valuable suggestion. According to your suggestion, we have repeated the in vivo therapeutic experiments where the tumor growth curve was prolonged to 28 days. As expected, at the end of the treatment, a more significant difference in tumor suppression was observed between the BG + US and DBG + US groups. The new results were provided in the revised manuscript.  
+
+The following is our revisions to Fig. 5g&h and Fig. 7e&g of the Revised manuscript:
+
+<--- Page Split --->
+![](images/Figure_7.jpg)
+
+<center>Figure 5. The in vivo anticancer performance of DBG under US irradiation. a) Schematic illustration of 4T1 tumor-bearing mouse model establishment and treatment procedures in vivo. b, c) In vivo retentions of 2-DG and BFO in different formulations in different periods post subcutaneous injection (days 0, 3, 5, and 7) into 4T1 tumor-bearing BALB/c mice. d) Body weight of mice in different groups ( \(n = 5\) mice per group) in the 28-day treatment period. e) Average volume changes of the tumor of mice in different groups. f) Average tumor weight of mice in different groups ( \(n = 5\) mice per group) at the ends of treatments. g) Morbidity-free survival of different groups of mice after different treatments ( \(n = 5\) mice per group). h) Tumor volume change curves of each mouse after various treatments ( \(n = 5\) mice per group). i) Representative tumor immunofluorescence staining images of ROS, Ki67, and TUNEL after different treatments. j, k) Western blot analysis of the expressions of GRP78 in tumor tissues after different treatments and corresponding quantitative analysis. l-m) The GO enrichment and further MF analysis between the control and DBG + US groups. Data were expressed as means \(\pm \mathrm{SD}\) in d-f, and k. \(*P< 0.01\) , \(**P< 0.001\) , \(***P< 0.0001\) . Source data are provided as a Source Data file. </center>
+
+<--- Page Split --->
+![](images/Figure_unknown_4.jpg)
+
+<center>Figure 7. Effectiveness in combating distant tumors. a) Schematic illustration of 4T1 bilateral tumor inoculation and treatment procedures in vivo. b) Body weight variations of mice in different groups ( \(n = 5\) mice per group) in the 28-day treatment period. Average volume and weight changes of (c, d) primary tumors and (e, f) distant tumors of mice in different groups ( \(n = 5\) mice per group) at the ends of treatments. g) Volume change curves of distant tumors in each mouse after different treatments ( \(n = 5\) mice per group). h) tSNE visualization and quantification of all immune cells in distant tumors by cytometry from 4T1 tumor-bearing BALB/c mice after different treatments. i) Immunofluorescence images of proliferated cytotoxic T lymphocytes ( \(\mathrm{CD3^{+} / CD8^{+}}\) ) and polarized M1 TAMs ( \(\mathrm{F4 / 80^{+} / CD86^{+}}\) ) in 4T1 tumor tissue slices of tumor. Data were expressed as means \(\pm\) SD in b-f. \(*P< 0.01\) , \(**P< 0.001\) , \(***P< 0.0001\) . Source data are provided as a Source Data file. </center>
+
+<--- Page Split --->
+
+7. Deep tissue penetration capability (on the order of centimeters) is the main advantage of ultrasound. The authors should further investigate the piezocatalytic therapy of DBG on deep-sited tumors (e.g., osteosarcoma or orthotopic liver tumors).  
+
+Response: Thank you for your professional comment and valuable suggestion. According to your suggestion, we investigated the piezocatalytic therapy of DBG on the orthotopic osteosarcoma-bearing mouse model. As Fig. S7 showed, DBG + US treatment has triggered the strongest tumor-suppressing effects compared to other groups. Such results suggest that ultrasound is indeed able to reach deep tumor tissues thereby to trigger DBG-mediated piezocatalytic therapeutic effects. We have added a description in the revised manuscript and provided the corresponding data in SI.  
+
+## The following is our revisions to the manuscript on Page 18:  
+
+For tumor volume, the treatment with solely US irradiation or DBG injection failed to inhibit the growth of tumors in mice, compared to the control group (Fig. 5e, f, h). Comparably, BG + US treatment triggered strong tumor-suppressing effects ( \(94.99\%\) inhibition of tumor growth), attributed to generating a great amount of ROS by the piezocatalytic effect of BFO. Notably, the best inhibition effect was shown by the DBG + US treated group, which may be further enhanced by the release of 2- DG during PCT. Moreover, similar tumor- suppressing effects were observed on the orthotopic osteosarcoma- bearing mouse model thanks to the deep tissue penetration capability (on the order of centimeters) of ultrasound (Fig. S7). Once the N- glycosylation of tumor cells was effectively inhibited, resulting in the strengthened killing effect of the activated immune cells.  
+
+![](images/Figure_unknown_5.jpg)
+
+<center>Figure S8. a) Schematic illustration of the orthotopic osteosarcoma-bearing mouse model </center>
+
+<--- Page Split --->
+
+establishment and treatment procedures in vivo. The K7M2 WT cells \((5 \times 10^{5}\) cells) were injected into the right proximal lateral tibia of the mice using a 27- gauge needle. b) Average volume changes of the tumors in different groups. c) Representative tumor immunofluorescence staining images of H&E, Ki67, and TUNEL after different treatments. \(n = 3\) mice per group, data were expressed as means \(\pm \mathrm{SD}\) , \(*P < 0.01\) , \(**P < 0.001\) , \(***P < 0.0001\) . Source data are provided as a Source Data file.  
+
+8. Do the authors notice the BiFeO₃- induced ferroptosis during the cancer therapy? This process would also promote the generation of ROS.  
+
+Response: Thank you for your professional comment. We agree with you that ferroptosis can also promote the generation of ROS, thus we further evaluated the capability of DBG to induce tumor ferroptosis. We measured the malondialdehyde (MDA) levels, a marker of ferroptosis, in tumor tissues of mice after different treatments. As Fig. S9 shows, the levels of ferroptosis- related markers MDA in the DBG + US group remained almost the same as those of PBS and other groups, illustrating that DBG could not induce significant ferroptosis. Therefore, the intramural ROS originated from the BFO- mediated piezocatalytic process rather than from the BFO- induced ferroptosis of tumor cells is responsible for immune response activation and ICD. We have added a description in the revised manuscript and provided the corresponding data in SI.  
+
+## The following is our revisions to the manuscript on Page 19:  
+
+while very weak fluorescence was observed in the other groups, the ROS generated during the piezoelectric catalysis is responsible for the tumor cell- killing effect in vivo. Considering that ferroptosis can also promote the generation of ROS, the capability of DBG to induce ferroptosis of tumors was evaluated<sup>46</sup>. After different treatments, the malondialdehyde (MDA) levels, a marker of ferroptosis, were measured in tumor tissues of mice using an MDA kit. The levels of ferroptosis- related markers MDA in the DBG + US group remained almost the same as the those of PBS and other groups (Fig. S), illustrating that DBG could not induce significant ferroptosis. Therefore, the intramural ROS originated from the BFO- mediated piezocatalytic process rather than from the BFO- induced ferroptosis of tumor cells is responsible for immune response activation and ICD.
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure S9. Levels of MDA in tumor tissues of mice after different treatments. Analysis of the excised main organs and tumor slices of 4T1-bearing mice after different treatments by H&E staining. Data were expressed as means \(\pm\) SD \((n = 5)\) . \(*P< 0.01\) , \(**P< 0.001\) , \(***P< 0.0001\) . Source data are provided as a Source Data file. </center>
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Piezocatalytic immunotherapy):  
+
+The authors have carefully revised the manuscript according to the reviewers'comments. The quality and novelty of this manuscript are greatly improved, and I fully agree to publish it.  
+
+Reviewer #3 (Metalloimmunotherapy, metal oxide NPs, nanomedicine):  
+
+The authors have addressed all of the issues, this paper is suitable for publication.  
+
+Reviewer #4 (immunotherapy, ICD)  
+
+The manuscript by Wu W et al reported the ICD induction and anti- tumor effect by a US- activatable nanocatalytic therapy, via the activation of ROS and piezoelectric signals. The study indicates a synergetic effect of these novel cancer- treatment approaches, as well as the priming of adaptive immune response in animal models. The study is designed regarding the strategy for the optimization of nanocatalytic therapy, but the characterization of immune activation, including ICD induction and immune profiling in vivo is not yet well established with the current data set of such quality and the methods details provided. Please refer to the comments below.  
+
+## Major concerns:  
+
+1) The authors should improve the methods that were employed to detect the ICD hallmarks in vitro. The CRT and HMGB1 staining seem very unclassical (Figure 4d), CRT and Hsp70 are abundant in the ER and translocated onto the cell membrane during ICD, but this is not observed from the images provided. Especially considering that all the cells were fixed with PFA, in which case all cells should have visible CRT and HSP70 staining in the ER. For precise evaluation of CRT/Hsp70 exposure, it is important to stain unfixed/unpermeabilized cells for imaging, or perform a flow cytometry surface-CRT / Hsp70 detection and exclude dead cells while quantifying surface CRT. HMGB1 is located in the nuclei which should be majorly colocalized with DAPI in non-treated conditions, and released from the nuclei upon ICD. In the image provided here, it seems both signals, including the DAPI, diminished in the DBG+US condition. In general, HMGB1 release is accompanied by cell death which leads to shrunk nuclei and brighter DAPI staining. The authors should carefully check the images and quantify the subcellular localization of those markers more precisely, not just general intensity change. The same issue can be detected in Figure S8 which needs to be addressed. For the evaluation of ER stress (Figure 3 d,e), WB detection or immunofluorescent staining of eIF2A phosphorylation and XBP1 splicing should be preferred. The authors should also detect ATP release from treated cells which is yet another important marker for ICD, and this can be easily achieved.
+
+<--- Page Split --->
+
+2) Considering the impact and level of Nature Communications, as well as the immune-priming nature of the proposed therapy, the study should at least include two cancer models/cell lines for all the experiments to avoid the cell type specificity of the treatments.  
+
+3) The in vitro experiments should focus more on BMDCs than macrophages as DCs are absolutely more central in ICD processing.  
+
+4) Figure 6-8, it is important to provide the complete gating strategy for the flow cytometry data, as a supplementary figure for example. The authors displayed many types of immune cells like DCs, T cells, NK, and B cells, but in the figures, there are no figures showing how those populations are defined. For example, markers of NK cells are not included anywhere. CD11c alone is not sufficient to define DCs. There are antibodies to CD44 and CD62L listed in the materials, which are for memory T cells. But these cells were not shown in any umaps, is there any statistical change in these populations? How many mice were included in those immune profiling assays displayed in Figures 6, 7, and 8? Please provide plots with all the individual points and eligible statistics, for all groups that were analyzed.  
+
+5) To confirm that the ICD induction by DBG+US, an anticancer vaccination exp should be done in immunocompetent animals, and the tumor eradication should be tested also in immunodeficient animals (nude mice or wt mice with T cell depletion) to confirm the anticancer effect is dependent on the immune system.  
+
+## Minor:  
+
+1) Some representative immunofluorescent images lack proper labeling of channels in the figures, e.g. Figures 3 and 4 should be improved as in Figure 5i  
+
+2) The authors should also test the phagocytosis of treated 4T1 cells by BMDCs (as for BMDM in figure 4 f - h), which is more important to validate that US+DBG induced ICD of cancer cells  
+
+3) Figure 4b, it is surprising to see that many CD86+ cells do not express CD11c, are those cells actually macrophages than DCs? It would be important to provide the qualification data of BMDCs to confirm that the protocol used for BMDC differentiation is proper, BTW, the detailed protocols for BMDC and BMM preparation are also missing from the methods part.  
+
+4) Please unify the labeling of statistical significance across all figures. In the current version, some are with stars, some are exact p-values  
+
+5) Figure S12, what are the x and y-axis of the FACS plots?
+
+<--- Page Split --->
+
+Reviewer #5 (tumor cell biology, glycosylation inhibition):  
+
+The manuscript by Pu at al., describes a nano- bioengineering approach comprising ROS- inducing piezocatalytic bismuth ferrite scaffold with 2- deoxy- D- glucose (DBG) for improved immunotherapy targeting solid tumors. Although the results presented suggest that DBG holds promise as a more effective anti- cancer treatment, concerns remain that the evaluation of anti- tumor activities of DBG is limited and does not provide significant insights into its underlying effects. Furthermore, some experimental strategies lack rigor leading to an unconvincing interpretation of results. Some of these deficiencies are associated with an insufficient recognition of the diversity of effects that inhibition of N- glycosylation would have on multiple cell functions and on immunotherapy. Likewise, in addition to inhibiting N- glycosylation, 2- DG itself will impact other cellular processes. Overall, while the nano- engineering aspects of BG and DBG are well described, the manuscript falls short of providing novel insights about how the piezocatalytic immunotherapy will impact tumor epithelia. Additionally, given that the goal of this work is to highlight DBG's improved immunotherapy activities in a range of solid tumors, the manuscript would benefit from broadening its scope beyond the breast cancer model.  
+
+Major concerns:  
+
+1. In the Introduction section, the significance of posttranslational modifications mediated by N-glycosylation should be described in a greater detail, considering its acknowledged effects on protein folding, secretion, targeting to intracellular and extracellular sites, intercellular adhesion, differentiation, EMT, and adaptive immunity.  
+
+2. Given that inhibitory roles of 2-DG span beyond N-glycosylation and include inhibition of glycolysis and protein synthesis, along with induction of autophagy, among others, these diverse effects of 2-DG should be considered.  
+
+3. In vitro and in vivo studies are based on one 4T1 cell line, a model of stage IV human breast cancer. It would be good to interrogate at least two additional cell lines representative of other tumor subtypes to evaluate more generalized anti-tumor effects of DBG.  
+
+4. Immunoblot analyses in Figures 3d and 5j evaluating the induction of ER stress rely on one marker, GRP78, only. Additional markers of ER stress should be assessed and quantified, such as PERK, IRE1-a, and BIP.  
+
+5. In Figure 3h and 7i IF images lack markers of cellular architecture and, even when enlarged, they are not sufficiently clear to allow interpretation of changes in cell shape. The IF images should be quantified.  
+
+6. Figure 5 a & b should include images of dissected tumors and their corresponding H&Es. While Figure S17 includes H&Es of lung metastases, including a small fraction of metastases in DBG treated mice, these H&Es should be of higher resolution and accompanied by either IF or IHC analyses of these tissues.  
+
+7. The RNA-seq results presented in Figure 6 h - j are promising but G0 and KEGG analyses should be validated by quantitative IFs and immunoblot analyses.
+
+<--- Page Split --->
+
+8. The Materials and Methods section should include a detailed description of RNA-seq analyses with accompanying statistics. Although the RNA-seq study is included in the Supplementary Methods section, the description is sketchy and limited in scope.  
+
+9. The Discussion section should be more comprehensive and include an in-depth interpretation of anti-tumorigenic significance of DBG in the context of its targeted activities.
+
+<--- Page Split --->
+
+## Response to reviewer #1.  
+
+The authors have carefully revised the manuscript according to the reviewers' comments. The quality and novelty of this manuscript are greatly improved, and I fully agree to publish it.  
+
+Response: Thank you very much for the positive comment and kind recommendation.  
+
+## Response to reviewer #3.  
+
+The authors have addressed all of the issues, this paper is suitable for publication.  
+
+Response: Thank you very much for the positive comment and kind recommendation.  
+
+## Response to reviewer #4.  
+
+## Comments from reviewer #4:  
+
+The manuscript by Wu W et al reported the ICD induction and anti- tumor effect by a US- activatable nanocatalytic therapy, via the activation of ROS and piezoelectric signals. The study indicates a synergetic effect of these novel cancer- treatment approaches, as well as the priming of adaptive immune response in animal models. The study is designed regarding the strategy for the optimization of nanocatalytic therapy, but the characterization of immune activation, including ICD induction and immune profiling in vivo is not yet well established with the current data set of such quality and the methods details provided. Please refer to the comments below.  
+
+Response: Thank you very much for the insightful comments and suggestions. We have revised the manuscript carefully and provided detailed responses below in a point- by- point manner according to your suggestions.  
+
+Major concerns:  
+
+1. The authors should improve the methods that were employed to detect the ICD hallmarks in vitro. The CRT and HMGB1 staining seem very unclassical (Figure 4d), CRT and Hsp70 are abundant in the ER and translocated onto the cell membrane during ICD, but this is not observed from the images provided. Especially considering that all the cells were fixed with PFA, in which case all cells should have visible CRT and
+
+<--- Page Split --->
+
+HSP70 staining in the ER. For precise evaluation of CRT/Hsp70 exposure, it is important to stain unfixed/unpermeabilized cells for imaging, or perform a flow cytometry surface- CRT/Hsp70 detection and exclude dead cells while quantifying surface CRT. HMGB1 is located in the nuclei which should be majorly colocalized with DAPI in non- treated conditions, and released from the nuclei upon ICD. In the image provided here, it seems both signals, including the DAPI, diminished in the DBG+US condition. In general, HMGB1 release is accompanied by cell death which leads to shrunk nuclei and brighter DAPI staining. The authors should carefully check the images and quantify the subcellular localization of those markers more precisely, not just general intensity change. The same issue can be detected in Figure S8 which needs to be addressed. For the evaluation of ER stress (Figure 3 d,e), WB detection or immunofluorescent staining of eIF2A phosphorylation and XBP1 splicing should be preferred. The authors should also detect ATP release from treated cells which is yet another important marker for ICD, and this can be easily achieved.  
+
+Response: Thank you very much for the careful review and insightful comments. We have carefully checked the images and stained the unfixed/unpermeabilized cells for imaging for the CRT/Hsp70 exposure and flow cytometry detection. The HMGB1 release behavior of tumor cells received different treatments was re- imaged. In addition, we also detected ATP release from treated cells according to your suggestions (Fig. 3g), which further confirmed the ICD of DBG + US- treated 4T1 cells. And, the same issue in Figure S8 also has been addressed as Figure S11. For a more precise assessment of ER stress, the eIF2A phosphorylation and XBP1 splicing expression in 4T1 cells with different treatments were also examined by WB as suggested, and the results were similar to those of GRP78 (Fig. S6).
+
+<--- Page Split --->
+![](images/Figure_unknown_6.jpg)
+
+<center>Figure 3. The in vitro anticancer and ICD-triggering performances of BFO nanosheets. a) CLSM microscopic images of 4T1 cancer cells stained with Calcein-AM/PI after treatments under different conditions. b) tSNE visualization and quantification of apoptotic 4T1 cells by cytometry after different treatments. c) Frequencies of PHA-L binding to 4T1 cancer cells treated with different concentrations of DBG plus US irradiation ( \(n = 5\) biologically independent samples). d, e) Western blot analysis and corresponding quantifications of 4T1 cancer cell lysates showing the expressions of the ER stress protein GRP78 upon treatments with the indicated drugs ( \(n = 3\) biologically independent samples). f) FC analysis of CRT expression of 4T1 cancer cells after different treatments. g) Intracellular ATP levels of 4T1 cells after different treatments ( \(n = 5\) biologically independent samples). h) Immunofluorescence images of CRT, HMGB1, and HSP70 of 4T1 cancer </center>
+
+<--- Page Split --->
+
+cells after different treatments. Data were expressed as means \(\pm \mathrm{SD}\) in c, e, and g. \(*P< 0.05\) , \(**P< 0.01\) , \(***P< 0.001\) , \(****P< 0.0001\) . Source data are provided as a Source Data file.  
+
+![](images/Figure_unknown_7.jpg)
+
+<center>Figure S11. Immunofluorescence analysis of the expressions of CRT, and HSP70, as well as the release of HMGB1 from tumor tissues of a) babl/c mice and b) NOG mice after different treatments. </center>  
+
+![](images/Figure_unknown_8.jpg)
+
+<center>Figure S6. Western blot analysis and corresponding quantifications of 4T1 cancer cell lysates showing expression of the ER stress protein IRE1-a, XBP1, P-eIF2A, and PERK upon treatments with the indicated drugs. </center>
+
+<--- Page Split --->
+
+2. Considering the impact and level of Nature Communications, as well as the immune- priming nature of the proposed therapy, the study should at least include two cancer models/cell lines for all the experiments to avoid the cell type specificity of the treatments.  
+
+Response: Thank you for the professional comment. We appreciate your careful review and kind suggestion. Following your suggestion, we evaluated the in vitro anticancer activity of DBG-mediated PCT therapy on H- 22 mouse hepatocellular carcinoma cells and CT26 mouse colon cancer cells using the classical CCK- 8 assay. In addition, we also evaluated the in vivo antitumor efficacies of DBG on H- 22 and CT26 tumor- bearing mouse models. The results (Fig. S4 and Fig. S12) suggest that DBG is also highly effective on H- 22 and CT26 tumors.  
+
+Following are our revisions to the revised manuscript on Page 11 and Page 19:  
+
+Moreover, the similar effective anti- tumor effects of DBG- mediated PCT were also observed on mouse liver cancer cells (H- 22 cells) and mouse colon cells (CT26 cells) (Fig. S4).  
+
+![](images/Figure_unknown_9.jpg)
+
+<center>Figure S4. In vitro cell viabilities (a) H22 cells and (b) CT26 cells by different treatments (n = 5 independent samples). </center>  
+
+The piezocatalytic immunotherapy effect of DBG with US irradiation was also performed on breast orthotopic 4T1, H- 22 and CT26 tumor- bearing mouse models.
+
+<--- Page Split --->
+![](images/Figure_unknown_10.jpg)
+
+<center>Figure S12. The body weight, tumor volumes and weights of different types of tumor-bearing mice at the ends of various treatments, including (a-c) H-22 tumor-bearing mice model and (d-f) CT26 tumor-bearing mice model ( \(n = 5\) mice per group). </center>  
+
+3. The in vitro experiments should focus more on BMDCs than macrophages as DCs are absolutely more central in ICD processing.  
+
+Response: We appreciate your careful review and professional comments. Given the central role of DCs in the ICD process, we further investigated the phagocytosis of treated 4T1 cells by BMDCs and their cytokine secretion in vitro. As shown in Fig. S10, DBG + US-treated cancer cells were most efficiently phagocytosed by DCs, which prompted DCs to secrete a large amount of pro-inflammatory cytokines (e.g. IL1, IL18, TNF- \(\alpha\) ) for further initiation of cellular ICD.  
+
+The following is our revisions to the revised manuscript on page 16:
+
+<--- Page Split --->
+![](images/Figure_unknown_11.jpg)
+
+<center>Figure S10. (a) Phagocytosis ratio of 4T1 cancer cells determined by FC. (b-d) Levels of various cytokines in treated BMDCs suspensions measured by ELISA method ( \(n = 3\) biologically independent samples). </center>  
+
+4. Figure 6-8, it is important to provide the complete gating strategy for the flow cytometry data, as a supplementary figure for example. The authors displayed many types of immune cells like DCs, T cells, NK, and B cells, but in the figures, there are no figures showing how those populations are defined. For example, markers of NK cells are not included anywhere. CD11c alone is not sufficient to define DCs. There are antibodies to CD44 and CD62L listed in the materials, which are for memory T cells. But these cells were not shown in any umaps, is there any statistical change in these populations? How many mice were included in those immune profiling assays displayed in Figures 6, 7, and 8? Please provide plots with all the individual points and eligible statistics, for all groups that were analyzed.  
+
+Response: We appreciate your careful review and kind question. Based on your suggestions, we have provided the complete gating strategy for the flow cytometry data of Figure 6- 8 as follows:
+
+<--- Page Split --->
+![](images/Figure_unknown_12.jpg)
+
+<center>Figure S20. (a) Gating strategies for isolating F4/80<sup>+</sup>/CD86<sup>+</sup> M1 macrophages from tumor tissue. </center>  
+
+(b) Gating strategies for isolating CD80<sup>+</sup>/CD86<sup>+</sup> mature DCs from tumor tissue. (c) Gating strategies for isolating CD4<sup>+</sup> and CD8<sup>+</sup> T cells from tumor tissue. (d) Gating strategies and tSNE visualization for different immune cells from tumor tissue.
+
+<--- Page Split --->
+![](images/Figure_unknown_13.jpg)
+
+<center>Figure S21. Gating strategies and tSNE visualization for different immune cells from tumor tissue. </center>  
+
+![](images/Figure_unknown_14.jpg)
+
+<center>Figure S35. Gating strategies for CD62L-/CD44+ \(\mathrm{T_{CM}}\) and CD62L-/CD44+ \(\mathrm{T_{EM}}\) cells from spleen tissue of mice. </center>  
+
+In addition, as shown in the above gating strategies, the mature DCs were defined as \(\mathrm{CD45^{+} / CD11c^{+} / CD80^{+} / CD86^{+}}\) cells, not just CD11c+ cells. After carefully checking the Figures in the manuscript, we have found that the vertical coordinate of Figure 8k was written incorrectly, and it should be CD62L instead of CD862, which is for memory T cells. Thank you very much for the kind reminding, which is highly appreciated. The percentage of memory T cells in the spleen of the therapeutic group was significantly
+
+<--- Page Split --->
+
+increased compared to the control group. We randomly selected 3 mice in each group for immune profiling assays, and have provided plots with all the individual points and eligible statistics, for all groups that were analyzed, as suggested.  
+
+5. To confirm that the ICD induction by DBG+US, an anticancer vaccination exp should be done in immunocompetent animals, and the tumor eradication should be tested also in immunodeficient animals (nude mice or wt mice with T cell depletion) to confirm the anticancer effect is dependent on the immune system.  
+
+Response: We appreciate your careful review and kind suggestion. In this research, the anticancer vaccination exp was carried out on the Babl/c mice which is the immunocompetent animal model. To further demonstrate the immune anticancer effect induced by DBG + US treatment, we performed another vaccination experiment on a wild-type C57 mouse model, which also exhibited a strong immune memory antitumor effect (Fig. S31). In addition, we tested the anticancer effects of DBG + US with tumor-bearing nude mice to validate the role played by the immune system in tumor eradication. The tumor growth curves show that although tumors decreases rapidly in size after DBG + US treatment, they begin to grow quickly again beyond 10 days of treatment, suggesting that tumor eradication is indeed dependent on the immune system (Fig. S32).  
+
+## The related results is as follows:  
+
+![](images/Figure_unknown_15.jpg)
+
+<center>Figure S31. Average volume changes of the rechallenged tumors of C57 mice in different groups \((n = 5\) mice per group). b) Volume change curves of rechallenged tumors in each mouse after different treatments. </center>
+
+<--- Page Split --->
+![](images/Figure_unknown_16.jpg)
+
+<center>Figure S32. Average volume changes of the tumors of nude mice in different groups (n = 5 mice per group). b) Volume change curves of tumors in each mouse after different treatments. </center>  
+
+## Minor:  
+
+1) Some representative immunofluorescent images lack proper labeling of channels in the figures, e.g. Figures 3 and 4 should be improved as in Figure 5i  
+
+Response: We appreciate your careful review and kind reminding. We have improved the labeling of the channels in Figures 3 and 4 according to your suggestion.  
+
+2) The authors should also test the phagocytosis of treated 4T1 cells by BMDCs (as for BMDM in figure 4 f - h), which is more important to validate that US+DBG induced ICD of cancer cells  
+
+Response: We appreciate your careful review and kind reminding. We have tested the phagocytosis of treated 4T1 cells by BMDCs. As shown in Fig. S10a, DBG + US- treated cancer cells were most efficiently phagocytosed by DCs.  
+
+![](images/Figure_unknown_17.jpg)
+
+
+<--- Page Split --->
+
+Figure S10. (a) Phagocytosis ratio of 4T1 cancer cells determined by FC. (b- d) Levels of various cytokines in treated BMDCs suspensions measured by the ELISA method ( \(n = 3\) biologically independent samples).  
+
+3) Figure 4b, it is surprising to see that many \(\mathrm{CD86^{+}}\) cells do not express CD11c, are those cells actually macrophages than DCs? It would be important to provide the qualification data of BMDCs to confirm that the protocol used for BMDC differentiation is proper, BTW, the detailed protocols for BMDC and BMM preparation are also missing from the methods part.  
+
+Response: We appreciate your careful review and kind suggestions. These \(\mathrm{CD86^{+}}\) cells are indeed mature DCs with unique tentacles that are not found in macrophages. Probably due to operation issues, a number of cells were not successfully stained by the CD11c antibody. We have provided the qualification images of BMDCs for BMDCs differentiation. In addition, we have provided the detailed protocols for BMDC and BMMs preparation in the supplementary methods part.  
+
+The detailed protocols for BMDCs and BMMs preparation are as follows:  
+
+BMMs were isolated from the femurs and tibias of BALB/c mice (female, 7 weeks old). Briefly, after euthanizing the mice, the femurs and tibia of the hind legs were collected, and the bone marrow cells were gently flushed with precooled RIPA 1640. After centrifugation, cells were resuspended in complete Dulbecco's modified Eagle's medium containing \(2\mathrm{mM}\) 1-glutamine, \(10\%\) FBS, macrophage colony-stimulating factor (M- CSF, \(20\mathrm{ng / ml}\) ), penicillin ( \(50\mathrm{U / ml}\) ), and streptomycin ( \(50\mathrm{ng / ml}\) ) and seeded into sterileplastic petri dishes ( \(10\mathrm{ml}\) ) at a density of \(5\times 10^{6}\) cells per dish. The medium was replaced on day 3, and on day 7, the BMMs cells were harvested and used for further experiments. BMDCs were also obtained by the above method, where the macrophage colony-stimulating factor was replaced with DCs colony-stimulating factor (M- CSF, \(20\mathrm{ng / mL}\) and IL- 4 \(10\mathrm{ng / mL}\) ).  
+
+4) Please unify the labeling of statistical significance across all figures. In the current version, some are with stars, some are exact p-values.
+
+<--- Page Split --->
+
+Response: We appreciate your careful review and kind comment. We have unified the labeling of statistical significance with stars across all Figures.  
+
+5) Figure S12, what are the x and y-axis of the FACS plots?  
+
+Response: We appreciate your careful review. We have labeled the x and y- axis of Figure S12 as follows:  
+
+![](images/Figure_unknown_18.jpg)
+
+<center>Figure S25. Flow cytometry analysis of in vivo a) CD206+ macrophages (gated on CD11b+/F4/80+ macrophages), b) CD86+ macrophages (gated on CD11b+/F4/80+ macrophages), c) CD4+ T cells (gated on CD3+ T cells), and d) CD4+ T cells (gated on CD3+ T cells) in tumors of mice after different treatments. </center>
+
+<--- Page Split --->
+
+## Response to reviewer #5.  
+
+## Comments from reviewer #5:  
+
+The manuscript by Pu at al., describes a nano- bioengineering approach comprising ROS- inducing piezocatalytic bismuth ferrite scaffold with 2- deoxy- D- glucose (DBG) for improved immunotherapy targeting solid tumors. Although the results presented suggest that DBG holds promise as a more effective anti- cancer treatment, concerns remain that the evaluation of anti- tumor activities of DBG is limited and does not provide significant insights into its underlying effects. Furthermore, some experimental strategies lack rigor leading to an unconvincing interpretation of results. Some of these deficiencies are associated with an insufficient recognition of the diversity of effects that inhibition of N- glycosylation would have on multiple cell functions and on immunotherapy. Likewise, in addition to inhibiting N- glycosylation, 2- DG itself will impact other cellular processes. Overall, while the nano- engineering aspects of BG and DBG are well described, the manuscript falls short of providing novel insights about how the piezocatalytic immunotherapy will impact tumor epithelia. Additionally, given that the goal of this work is to highlight DBG's improved immunotherapeutic activities in a range of solid tumors, the manuscript would benefit from broadening its scope beyond the breast cancer model.  
+
+Response: Thank you very much for the insightful comments and suggestions. We have revised the manuscript carefully and provided detailed responses below in a point- by- point manner according to your suggestions.  
+
+Major concerns:  
+
+1. In the Introduction section, the significance of posttranslational modifications mediated by N-glycosylation should be described in a greater detail, considering its acknowledged effects on protein folding, secretion, targeting to intracellular and extracellular sites, intercellular adhesion, differentiation, EMT, and adaptive immunity.  
+
+Response: Thank you for the professional suggestion. Based on your suggestions, we have described the effects of post-translational modifications mediated by N
+
+<--- Page Split --->
+
+glycosylation on immune escape in more detail in the Introduction section of the revised manuscript.  
+
+## Following are our revisions to the revised manuscript on Page 5:  
+
+This coating not only impairs the immune anti- cancer response by masking new epitopes of immune cells or disrupting their function, but also interferes the intracellular and extracellular targeting recognition of cancer cells by immune cells, ultimately resulting in the immunosuppression of tumors<sup>30</sup>. In addition, cancer cells are able to maintain ER protein homeostasis by upregulating the level of N- glycosylation of related proteins in order to increase ER folding capacity and ER- associated degradation (ERAD), which deactivates the immunologically derived toxic proteins such as granzymes<sup>31</sup>.  
+
+2. Given that inhibitory roles of 2-DG span beyond N-glycosylation and include inhibition of glycolysis and protein synthesis, along with induction of autophagy, among others, these diverse effects of 2-DG should be considered.  
+
+Response: Thank you for the professional comments. As you noted, 2-DG is mainly known for its capacity to block glycolysis through hexokinase and phosphoglucose isomerase inhibition. However, recent studies demonstrate that 2-DG in a certain low concentration range (e.g., 2-4 mM) primarily can change the glycosylation levels of cancer cells without causing obvious inhibition on glycolysis or other effects (Sci. Transl. Med. 2022, 14, eabg3072). The DBG used in our study corresponded to a 2-DG concentration of as low as \(\sim 1.2\) mM, thus the other effects caused by DBG on cells could be negligible. To confirm this, we investigated the glycolytic behavior of 4T1 cells directly treated with low concentrations of 2-DG for different time periods by measuring lactate production in 4T1 cells. As shown in Figure S7, the glycolytic flux of cells did not change significantly in 12 or 24 h of treatment with 2 mM 2-DG.  
+
+Following are our revisions to the revised manuscript on Page 13:
+
+<--- Page Split --->
+
+And, the low- dose released 2- DG were less effective in driving glycolytic flux as measured by cellular lactate production (Fig. S7).  
+
+![](images/Figure_unknown_19.jpg)
+
+<center>Figure S7. Quantifications of extracellular lactate for 4T1 cells treated with 2DG at the concentration of \(2\mathrm{mM}\) for different time intervals ( \(\mathrm{n} = 3\) independent samples). </center>  
+
+3. In vitro and in vivo studies are based on one 4T1 cell line, a model of stage IV human breast cancer. It would be good to interrogate at least two additional cell lines representative of other tumor subtypes to evaluate more generalized anti-tumor effects of DBG.  
+
+Response: We appreciate your careful review and kind suggestion. Following your suggestion, we evaluated the in vitro anticancer activity of DBG- mediated PCT therapy on H- 22 mouse hepatocellular carcinoma cells and CT26 mouse colon cancer cells using the classical CCK- 8 assay. In addition, we also evaluated the in vivo antitumor effects of DBG on H- 22 and CT26 tumor- bearing mouse models. The results (Fig. S4 and Fig. S12) suggest that DBG is also highly effective on these cancers.  
+
+## Following are our revisions to the revised manuscript on Page 11 and Page 19:  
+
+Moreover, the similar effective anti- tumor effects of DBG- mediated PCT were also observed on mouse liver cancer cells (H- 22 cells) and mouse colon cells (CT26 cells) (Fig. S4).
+
+<--- Page Split --->
+![](images/Figure_unknown_20.jpg)
+
+<center>Figure S4. In vitro cell viabilities (a) H22 cells and (b) CT26 cells by different treatments (n = 5 independent samples). </center>  
+
+The piezocatalytic immunotherapy effect of DBG with US irradiation was also performed on breast orthotopic 4T1, H- 22 and CT26 tumor- bearing mouse models (Fig. S12).  
+
+![](images/Figure_unknown_21.jpg)
+
+<center>Figure S12. The body weight, tumor volumes and weights of different types of tumor-bearing mice at the ends of various treatments, including (a-c) H-22 tumor-bearing mice model and (d-f) CT26 tumor-bearing mice model ( \(n = 5\) mice per group). </center>  
+
+4. Immunoblot analyses in Figures 3d and 5j evaluating the induction of ER stress rely on one marker, GRP78, only. Additional markers of ER stress should be assessed and quantified, such as PERK, IRE1-a, and BIP.
+
+<--- Page Split --->
+
+Response: We appreciate your careful review and professional suggestion. Following your suggestion, we assessed and quantified the other types of ER stress markers PERK, IRE1- a, and BIP (also termed GRP78) in tumor cells and tumor tissues after different treatments, which show similar trends to GRP78 (Fig. S6).  
+
+![](images/Figure_unknown_22.jpg)
+
+<center>Following are our revisions to the revised manuscript on Page 11 and Page 19: </center>  
+
+Figure S6. Western blot analysis and corresponding quantifications of 4T1 cancer cell lysates showing expression of the ER stress protein IRE1- a, XBP1, P- eIF2A, and PERK upon treatment with the indicated drugs.  
+
+5. In Figure 3h and 7i IF images lack markers of cellular architecture and, even when enlarged, they are not sufficiently clear to allow interpretation of changes in cell shape. The IF images should be quantified.  
+
+Response: We appreciate your careful review and kind comments. For precise evaluation of CRT/Hsp70 exposure, we directly stained and imaged unfixed/unpermeabilized cells, and which indicates that the positive signals are mainly distributed on the surface of the cell membrane, which is consistent with the characteristics of cellular ICD. In addition, to better quantify the number of immune cells, we enlarged Figure 7i and quantified the target immune cells. The relative quantitative results were also added in the revised SI.  
+
+Following are our revisions to the revised manuscript and SI:
+
+<--- Page Split --->
+![](images/Figure_unknown_23.jpg)
+
+<center>Figure S26. Quantification of the number of target immune cells in tumor tissues of mice in different groups. </center>  
+
+6. Figure 5 a&b should include images of dissected tumors and their corresponding H&Es. While Figure S17 includes H&Es of lung metastases, including a small fraction of metastases in DBG treated mice, these H&Es should be of higher resolution and accompanied by either IF or IHC analyses of these tissues.  
+
+Response: We appreciate your professional comments and kind suggestions. According to your your suggestion, we have provided the images of dissected tumors and their corresponding H&Es after different treatments (Fig. S11). And the H&Es of lung metastases with high resolution and their IF analyses (Ki67) are also added in Figure S27.  
+
+Following are our revisions to the revised SI:
+
+<--- Page Split --->
+![](images/Figure_unknown_24.jpg)
+
+<center>Figure S11. (a) Images of the excised main organs and (b) tumor slices of 4T1-bearing mice after different treatments by H&E staining. </center>
+
+<--- Page Split --->
+![PLACEHOLDER_46_0]
+
+<center>Figure S27. (a) H&E and (b) Ki67 staining of lung tissue from different groups of 4T1 tumor-bearing mice. </center>  
+
+7. The RNA-seq results presented in Figure 6 h-j are promising but G0 and KEGG analyses should be validated by quantitative IFs and immunoblot analyses.  
+
+Response: We appreciate your valuable comments. Following your suggestion, we validated three typical immune activation-related signaling pathways, such as Toll-like receptor 4, TNF- \(\alpha\) and IL6 signaling pathways, by quantitative IF and WB analyses. The results show obvious upregulation of Toll-like receptor 4, TNF- \(\alpha\) and IL6 proteins upon DBG + US treatment, demonstrating the activation of related signaling pathways.
+
+<--- Page Split --->
+![PLACEHOLDER_47_0]
+
+<center>Figure S23. (a, b) Western blot analysis and corresponding quantifications of tumor tissues showing the expressions of the TNF-α, IL6, and TLR4 proteins before and after DBG + US treatment. (c) Immunofluorescence staining of TNF-α, IL6, and TLR4 in tumour tissues before and after DBG + US treatment. </center>  
+
+8. The Materials and Methods section should include a detailed description of RNA-seq analyses with accompanying statistics. Although the RNA-seq study is included in the Supplementary Methods section, the description is sketchy and limited in scope.  
+
+Response: We appreciate your careful review. As suggested, a detailed description of the RNA-seq analysis has been provided in the Materials and Methods section.  
+
+Following are our revisions to the revised SI:  
+
+## RNA sequencing.  
+
+1. RNA extraction: Tumors were isolated from mice after 3 days of different treatments, and total RNA was extracted from the tissue using TRIzol® Reagent according to the manufacturer’s instructions (Invitrogen), and genomic DNA was removed using DNase I (TaKara). Then RNA quality was determined by 2100 Bioanalyser (Agilent) and
+
+<--- Page Split --->
+
+quantified using the ND- 2000 (NanoDrop Technologies). Only high- quality RNA sample(OD260/280=1.8\~2.2, OD260/230≥2.0, RIN≥6.5, 28S:18S≥1.0, >1μg) was used to construct sequencing library.  
+
+2. Library preparation, and Illumina Hiseq xten/Nova seq 6000 Sequencing: RNA-seq transcriptome library was prepared following TruSeqTM RNA sample preparation Kitfrom Illumina (San Diego, CA) using 1μg of total RNA. Shortly, messenger RNA was isolated according to polyA selection method by oligo(dT) beads and then fragmented by fragmentationbuffer firstly. Secondly double-stranded cDNA was synthesized using a SuperScriptdouble-stranded cDNA synthesis kit (Invitrogen, CA) with random hexamer primers (Illumina). Then the synthesized cDNA was subjected to end-repair, phosphorylation and ‘A’ base additionaccording to Illumina’s library construction protocol. Libraries were size selected for cDNA targetfragments of 300 bp on 2% Low Range Ultra Agarose followed by PCR amplified using PhusionDNA polymerase (NEB) for 15 PCR cycles. After quantified by TBS380, paired-end RNA-seq sequencing library was sequenced with the Illumina HiSeq xten/NovaSeq 6000 sequencer (2 ×150bp read length).  
+
+3. Read mapping: The raw paired end reads were trimmed and quality controlled by SeqPrep (https://github.com/jstjohn/SeqPrep) and Sickle (https://github.com/najoshi/sickle) with default parameters. Then clean reads were separately aligned to reference genome with orientation mode using HISAT2 (http://ccb.jhu.edu/software/hisat2/index.shtml) software. The mapped reads of each sample were assembled by StringTie(https://ccb.jhu.edu/software/stringtie/index.shtml?t=example) in a reference-based approach.  
+
+4. Differential expression analysis and Functional enrichment: To identify DEGs (differential expression genes) between two different samples, the expression level of each transcript was calculated according to the transcripts per million reads (TPM) method. RSEM (http://deweylab.biostat.wisc.edu/rsem/) was used to quantify gene
+
+<--- Page Split --->
+
+abundances. Essentially, differential expression analysis was performed using the DESeq2/DEGseq/EdgeRwith Q value \(\leq 0.05\) , DEGs with \(|\log 2\mathrm{FC}| > 1\) and Q value \(\leq = 0.05\) (DESeq2 or EdgeR)/Qvalue \(\leq = 0.001\) (DEGseq) were considered to be significantly different expressed genes). In addition, functional- enrichment analysis including GO and KEGG were performed to identify which DEGs were significantly enriched in GO terms and metabolic pathways at Bonferroni- corrected P- value \(\leq 0.05\) compared with the whole- transcriptome background. GO functional enrichment and KEGG pathway analysis were carried out by Gootools (https://github.com/tanghaibao/Gootools) and KOBAS (http://kobas.cbi.pku.edu.cn/home.do).  
+
+5. Alternative Splice events Identification: All the alternative splice events that occurred in our sample were identified by using recently releases program rMATS(http://rnaseq-mats.sourceforge.net/index.html). Only the isoforms that were similar to the reference or comprised novel splice junctions were considered, and the splicing differences were detected as exon inclusion, exclusion, alternative 5', 3', and intron retention events.  
+
+9. The Discussion section should be more comprehensive and include an in-depth interpretation of anti-tumorigenic significance of DBG in the context of its targeted activities.  
+
+Response: Thank you for your careful review and valuable suggestions. As suggested, we have further provided the in-depth interpretation of anti-tumorigenic significance of DBG in the context of its targeted activities in the revised manuscript.  
+
+## Following are our revisions to the revised manuscript on page 28:  
+
+It is noted that the systemic administration of immunotherapeutic agents may destroy the homeostatic function of immune cells in non- target tissues, therefore a local delivery vehicle could allow for modulation and sustained release of the payload, which thereby would not only minimize off- target related side effects, but also enhance effective drug bioavailability. Based on this consideration, in this paper, we
+
+<--- Page Split --->
+
+constructed an injectable ROS- sensitive in situ hydrogel bio- piezoelectric scaffold (DBG) for local synergistic delivery of piezocatalysts and 2- DG to maximize anticancer efficacy, in which 2- DG- mediated inhibition of glycosylation could reverse the immunosuppressive niche of tumor and subsequently promote anti- tumor immune responses activated by piezocatalytic therapy. Such a tumor treatment strategy featuring in situ targeting of activated immunity and inhibition of tumor glycosylation to remodel immunosuppression is expected to provide a new paradigm in interventional oncology.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+## Reviewer #4:  
+
+The authors have addressed most of the major concerns from the referees and the revised version is highly improved. I would like to suggest a few points:  
+
+1) For the WB detection of ER stress, the spliced XPB1 (XBP1s) should be detected rather than the total protein. Only XBP1s upregulation indicates ER stress occurring (refer to 10.1038/s41418-017-0044-9); p-Eifa2 should be accompanied with total EIF2a, especially considering the apparent upregulation of total PERK and IRE-a, instead of their phosphorylation during ER stress... the WB results here show a rather general increase of protein translation, but not a specific upregulation of ER-stress pathway, which need to be explained. The general upregulation of protein synthesis is surprising, especially considering the role of 2-DG in inhibition of glycolysis and protein synthesis (as mentioned by Reviewer #5).  
+
+2) Seems all the flowcytometry gating lacks a cell viability control dye, which restricted the precision of proper cell population defining.  
+
+3) Methods for BMM and BMDC differentiation is wrong, BMDCs should be differentiated with GM-CSF -/+ IL4, but not M-CSF, this might be a typo or wrong methods that need to be carefully treated.  
+
+## Reviewer #5:  
+
+For the most part, the authors have responded to many of my concerns in a satisfactory way.
+
+<--- Page Split --->
+
+## Response to reviewer #4.  
+
+## Comments from reviewer #4:  
+
+The authors have addressed most of the major concerns from the referees and the revised version is highly improved. I would like to suggest a few points:  
+
+Response: Thank you very much for the positive comments and careful reviews. We have revised the manuscript carefully and provided detailed responses below in a point- by- point manner according to your suggestions.  
+
+1) For the WB detection of ER stress, the spliced XPB1 (XBP1s) should be detected rather than the total protein. Only XBP1s upregulation indicates ER stress occurring (refer to 10.1038/s41418-017-0044-9); p-Eifa2 should be accompanied with total EIF2a, especially considering the apparent upregulation of total PERK and IRE-a, instead of their phosphorylation during ER stress... the WB results here show a rather general increase of protein translation, but not a specific upregulation of ER-stress pathway, which need to be explained. The general upregulation of protein synthesis is surprising, especially considering the role of 2-DG in inhibition of glycolysis and protein synthesis (as mentioned by Reviewer #5).  
+
+Response: Thank you very much for your professional comments. To more accurately indicate ER stress, we have detected the expression level of spliced XPB1 (XBP1s) instead of the one of XBP1 of tumor cells treated with 2-DG as well as the total EIF2a, according to your suggestion. The relative data were added in revised supporting information (Fig. S6). PERK is a type I transmembrane protein located in the endoplasmic reticulum membrane, belonging to the elf2a upstream kinase family, and IRE1α is the most prominent and evolutionarily conserved unfolded protein response (UPR) signal transducer during endoplasmic reticulum functional upset, of both of which would be significantly activated and upregulated during the early stage of ER stress of cells. As you and Reviewer #5 noted, 2-DG is mainly known for its capacity to block glycolysis through hexokinase and phosphoglucose isomerase inhibition. However, recent studies demonstrate that 2-DG in a certain low concentration range
+
+<--- Page Split --->
+
+(e.g., 2- 4 mM) primarily can change the glycosylation levels of cancer cells without causing obvious inhibition on glycolysis or other effects (Sci. Transl. Med. 2022, 14, eabg3072). The DBG used in our study corresponded to a 2- DG concentration of as low as \(\sim 1.2 \mathrm{mM}\) , thus the other effects caused by DBG on cells could be negligible. To confirm this, we investigated the glycolytic behavior of 4T1 cells directly treated with low concentrations of 2- DG for different time periods by measuring lactate production in 4T1 cells. As shown in Figure S7, the glycolytic flux of cells did not change significantly in 12 or 24 h of treatment with \(2 \mathrm{mM} 2 - \mathrm{DG}\) .  
+
+2) Seems all the flowcytometry gating lacks a cell viability control dye, which restricted the precision of proper cell population defining.  
+
+Response: We appreciate your careful review. We have performed live/dead staining with PI to exclude dead cells during FC analysis. This is a common gating operation, so we did not show it in the gating strategy. We have readded this gating strategy in the revised manuscript based on your suggestion.  
+
+3) Methods for BMM and BMDC differentiation is wrong, BMDCs should be differentiated with GM-CSF \(-/ + \mathrm{IL4}\) , but not M-CSF, this might be a typo or wrong methods that need to be carefully treated.  
+
+Response: We appreciate your careful review. We have corrected the wrong type of GM-CSF instead of M-CSF in the revised supporting information.  
+
+## Response to reviewer #5.  
+
+## Comments from reviewer #5:  
+
+For the most part, the authors have responded to many of my concerns in a satisfactory way.  
+
+Response: Thank you very much for the positive comment.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #4 (Remarks to the Author):  
+
+The latest revised manuscript has addressed most of the questions from my previous review report  
+
+Reviewer #5 (Remarks to the Author):  
+
+The authors responded to the remaining few concerns in a satisfactory way and this manuscript is now acceptable for publication.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__72029e53925d7ad4b9d13bca370bd70820e0aa98858b14f70cb0c95ec2ff75c6/images_list.json

@@ -0,0 +1,100 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1. Previously reported supramolecular containers based on \\(\\alpha ,\\gamma\\) -cyclic peptides. Top: cyclic octapeptide (blue) topped with a porphyrin moiety (green) used in the recognition of \\(4,4^{\\prime}\\) -bipyridines. Center: smaller alternatives derived from dimer-forming \\(N\\) -propargylated cyclic hexapeptides (CP1, grey) through Sonogashira cross-couplings with Iodopyridines or copper-catalyzed azide-alkyne cycloaddition (CuAAC) that have been used as ion transporters.[27] Bottom: cartoon model of supramolecular capsule derived from CP1 and a tris-azide derivative described in this work.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2. Synthetic strategy used for the preparation of capsules D2 and D4 and initially proposed encapsulation model for the recognition of anions (2X<D2 and 2X<D4).",
+    "footnote": [],
+    "bbox": [
+      [
+        163,
+        95,
+        816,
+        415
+      ]
+    ],
+    "page_idx": 30
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3. Side (a and c) and top (b and d) views of the crystal structures of dimeric supramolecular capsules D2 (top) and D4 (bottom), respectively. The molecules of acetonitrile entrapped in the cavity are represented in CPK models. The nitrogen of nitrile groups is pointing towards one of the caps close to the triazole protons with shorter distance in the Ach-based capsule (2.66 Å, bottom) than in the Acp derivative (2.79-2.68 Å, top). For clarity only triazole and amide protons are shown. The yellow dashed lines highlight the hydrogen bonds.",
+    "footnote": [],
+    "bbox": [
+      [
+        142,
+        99,
+        857,
+        589
+      ]
+    ],
+    "page_idx": 31
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4. NMR spectra of pure CP2 (bottom) and after the addition of different equivalents of fluoride (TBAF). In blue colour are highlighted the signals corresponding to the new species formed after the addition of the fluoride, light blue denotes the signals corresponding to the tetrabutylammonium counterion.",
+    "footnote": [],
+    "bbox": [
+      [
+        168,
+        95,
+        828,
+        390
+      ]
+    ],
+    "page_idx": 32
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5. Experiments of heterodimer (D2-3 and D2-4) formation followed by anion recognition, top view. Bottom a) NMR spectra corresponding to these studies in which the characteristic signals of each component are highlighted with specific colours; orange, green and lavender for homodimers D2, D3 and D4, respectively, dark blue for CP2 interacting with fluoride, and plum and teal green for heterodimers D2-3 and D2-4, respectively.",
+    "footnote": [],
+    "bbox": [
+      [
+        144,
+        95,
+        856,
+        666
+      ]
+    ],
+    "page_idx": 33
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Figure 6. Side (a) and top (b) view of the crystal structure of water-fluoride cluster entrapped in supramolecular capsule D2 (3F.8H₂O<2CP2) as twin complexes. In these complexes three fluoride ions (light green) and eight water molecules are hydrogen-bonded to the amide protons of two cyclic peptides at different planes (the hydrogen bond network (yellow dashed lines) in the cluster is only shown for one of the complexes). Side (c) and top (d) view of the crystal structure of encapsulated chloride-water cluster between two CP2 (3Cl.4H₂O<2CP2). The three chloride ions and water molecules are occupying six equivalent chemical (and crystallographic) positions forming a hexagonal structure, where the two subunits are aligned forming a trigonal bipyramid shaped (d) capsule.",
+    "footnote": [],
+    "bbox": [
+      [
+        140,
+        128,
+        855,
+        551
+      ]
+    ],
+    "page_idx": 34
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_7.jpg",
+    "caption": "Figure 7. Chloride transport experiments using capsule D2 in liposomes containing lucigenin (A) or HPTS (B-D).",
+    "footnote": [],
+    "bbox": [
+      [
+        256,
+        95,
+        740,
+        831
+      ]
+    ],
+    "page_idx": 35
+  }
+]

peer_reviews/supplementary_0_Peer Review File__72029e53925d7ad4b9d13bca370bd70820e0aa98858b14f70cb0c95ec2ff75c6/supplementary_0_Peer Review File__72029e53925d7ad4b9d13bca370bd70820e0aa98858b14f70cb0c95ec2ff75c6.mmd

@@ -0,0 +1,483 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Recognition of anion- water clusters by peptide- based supramolecular capsules  
+
+![](images/Figure_1.jpg)
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+This manuscript describes the synthesis and characterization of a supramolecular dimer which can bind anions. The manuscript is well- written and the methods use to characterize the compounds (including X- ray diffraction, NMR, transmembrane transport experiments) are fairly thorough. The results showing how the macromolecules bind the anions, and the solvating water molecules is interesting and relevant to the larger topic of selective binding of ions. I would recommend this paper for publication without revision.  
+
+Reviewer #2 (Remarks to the Author):  
+
+Granja and coworkers demonstrated supramolecular dimerization of a capped cup- shaped cyclic peptide and its ability to recognize different anions in the form of hydrated clusters. The formation of supramolecular dimers or anion binding and transport have been explored by several groups using different systems. What is interesting about the present study is the importance of water in the binding. Although the anion binding is studied in Dichloromethane and Acetonitrile mixture, a trace amount of water present in the solvents/chemicals is essential for binding. The removal of water traces by adding molecular sieves to the host- guest complex resulted in the removal of anion from the guest. The effect is reversible wherein the addition of water again leads to the binding. In supramolecular chemistry, the role of solvents in a study has high significance on the behavior of a molecule/material yet rarely their role is completely understood. Hence the present study focussing on the binding of anion- water clusters is significant.  
+
+The study is experimentally rich with clean characterization and analysis, but the presentation can be improved.  
+
+Please provide integrations in the supporting figures for the 1H NMR spectra of the supramolecular dimers and the complexes.  
+
+There are multiple typos throughout the manuscript and SI for scientific and non- scientific terms (for example, RMN for NMR), please correct.  
+
+The authors demonstrate that the CP1- CP4 exists as the corresponding supramolecular dimeric forms D1- D4 in solution. In the discussion/figures, authors represented the dimers using different notations like CP, 2CP, and D leading to confusion and difficulty in understanding the content (for example, refer to Figure 5 and the corresponding discussion).
+
+<--- Page Split --->
+
+The authors conclude that the recognition process most likely involves dimeric capsules and trimeric anions. It was also mentioned that experimental evidence regarding this is provided in the case of halides F- and Cl using crystal structures and mass but not in the case of other anions (Page 13 of the manuscript). It was unclear how the authors concluded the stoichiometry of binding for the anions other than F- and Cl- without mass, crystal structure, or ITC. There is a significant increase in the size of the anions N3- , OAc- , and NO3- compared to halides. The cavity size does not seem to be able to fit three big anions. If there is no clear confirmation, I suggest mentioning the uncertainty about stoichiometry rather than generalizing that all anions could be trimers. In that case, it is suggestible to modify the title also to "anion- water clusters" instead of "trimeric anionic- water clusters"  
+
+Please provide a brief explanation of the results obtained from DFT  
+
+TOC can be improved; the current TOC does not represent the anion- water clusters clearly in the supramolecular dimer.  
+
+Reviewer #3 (Remarks to the Author):  
+
+This paper reports the synthesis of a cyclic peptide which is furnished with a cap to produce a hemispherical unit which dimerises through hydrogen bonding interactions. The two halves of the sphere are shown to open up to encapsulate fluoride and chloride in solution and in the solid state. In the crystal structure, there is clear evidence of water being involves in the formation of a specific cluster. In solution, water is required, but the structural nature of the cluster is not clearly demonstrated experimentally. Membrane transport is demonstrated, and explored using a conventional set of experiments.  
+
+My view is that this paper contains some very nice supramolecular chemistry work, and that the importance of water in the process is a highlight, though not as important as the introduction makes out. The experiments and analysis are all ably conducted and meet all the requirements of the field. It's all very well done, and will be likely used as an example of hydrated anions being encapsulated (though plenty of crystal structures exist of anion complex hydrates already) rather than built upon directly by others.  
+
+The introduction contains some statements which need work - electrostatic interactions are a noncovalent interation (top of p2), proteins are anion- selective and usually hold anions in a largely desolvated state. It's too much to call a host- guest complex a 'new state of matter.'
+
+<--- Page Split --->
+
+There should be consideration of the possibility of deprotonation by fluoride. Solvent needs to be given clearly in the main text when talking about NMR titrations etc.
+
+<--- Page Split --->
+
+The following pages provide point by point answers to each of the questions, comments or suggestions proposed by the different reviewers.  
+
+## Reviewer: 1  
+
+- Comment. "This manuscript describes the synthesis and characterization of a supramolecular dimer which can bind anions. The manuscript is well-written, and the methods used to characterize the compounds (including X-ray diffraction, NMR, transmembrane transport experiments) are fairly thorough. The results showing how the macromolecules bind the anions, and the solvating water molecules is interesting and relevant to the larger topic of selective binding of ions. I would recommend this paper for publication without revision."  
+
+Response. We greatly appreciate the positive comments about our work, how it was carried out and the characterization of the compounds and their derivatives. Thank you.  
+
+## Reviewer: 2  
+
+- Comment. "Granja and coworkers demonstrated supramolecular dimerization of a capped cup-shaped cyclic peptide and its ability to recognize different anions in the form of hydrated clusters. The formation of supramolecular dimers or anion binding and transport have been explored by several groups using different systems. What is interesting about the present study is the importance of water in the binding. Although the anion binding is studied in Dichloromethane and Acetonitrile mixture, a trace amount of water present in the solvents/chemicals is essential for binding. The removal of water traces by adding molecular sieves to the host-guest complex resulted in the removal of anion from the guest. The effect is reversible wherein the addition of water again leads to the binding. In supramolecular chemistry, the role of solvents in a study has high significance on the behavior of a molecule/material yet rarely their role is completely understood. Hence the present study focussing on the binding of anion-water clusters is significant."  
+
+The study is experimentally rich with clean characterization and analysis, but the presentation can be improved."  
+
+Response. We thank the referee for his/her interest and positive comments on our work and this manuscript. We are very pleased to hear his/her comments about the quality and significance of our work and the careful characterization and analysis of the results that we report here. Following his/her advice, the presentation was revised to improve it.
+
+<--- Page Split --->
+
+- Comment. "Please provide integrations in the supporting figures for the \(^1 H\) NMR spectra of the supramolecular dimers and the complexes."  
+
+Response. We have included all the integrations in the \(^1 H\) NMR spectra of the supramolecular dimers and complexes in the supporting information, in the section 8. Thank you very much for your kind suggestion.  
+
+- Comment. "There are multiple typos throughout the manuscript and SI for scientific and non-scientific terms (for example, RMN for NMR), please correct."  
+
+Response. We are very sorry for the inconvenience related to the quality of the written document. We have edited both the manuscript and SI to correct all the typographical errors in these documents. Non-scientific terms such as RMN were also corrected.  
+
+- Comment. "The authors demonstrate that the CP1-CP4 exists as the corresponding supramolecular dimeric forms D1-D4 in solution. In the discussion/figures, authors represented the dimers using different notations like CP, 2CP, and D leading to confusion and difficulty in understanding the content (for example, refer to Figure 5 and the corresponding discussion)."  
+
+Response. Again, we are very sorry for these errors, which cause some confusion and may make difficult to understand the content of the work. In general, we have used the term CP for the abbreviation of Cyclic Peptide generically, while CP1- CP4 are used to describe the molecular structure, in which all atoms are covalently linked, corresponding to the four cyclic peptides used in this work. The terms D1- D4 for homodimers or the corresponding heterodimers (D2- 3 or D2- 4) are intended to emphasize the supramolecular species that are present predominantly in the organic solvents used in this work (dichloromethane or acetonitrile in dichloromethane). Therefore, we preferably use Dn to describe the supramolecular species that are mainly present in solution experiments or in the crystals. The same can be said for heterodimeric species, in which two different cyclic peptides are mixed and assembled in solution forming hydrogen bonds between both molecules. Therefore, D (dimer) is used to describe the supramolecular species in which both peptide units are linked by hydrogen bonds. To clarify this possible ambiguity, we have modified the nomenclature for complexes with anions, in which the two cyclic peptides are not directly linked to each other by hydrogen bonds, using the new numeration, mA·nH₂O=2CP2, for the complex with the anion cluster in which "m" represents the number of anions entrapped between both CPs (most likely three), "A" indicates the type of anion (F-, Cl-, Br-, I-, AcO-, N₃⁻ or NO₃⁻) and "n" represents the number of water molecules that form the entrapped anion clusters.  
+
+- Comment. "The authors conclude that the recognition process most likely involves dimeric capsules and trimeric anions. It was also mentioned that experimental evidence regarding this is provided in the case of halides F- and Cl using crystal structures and mass but not in the case of other anions (Page 13 of the manuscript). It was unclear how the authors concluded the stoichiometry of binding for the anions other than F- and Cl- without mass, crystal structure, or ITC. There is a significant increase in the size of the anions N3-, OAc-, and NO3- compared to halides. The cavity size does not seem to be able to fit three big anions. If there is no clear confirmation, I suggest mentioning the
+
+<--- Page Split --->
+
+uncertainty about stoichiometry rather than generalizing that all anions could be trimers. In that case, it is suggestive to modify the title also to "anion- water clusters" instead of "trimeric anionic- water clusters".  
+
+Response. We appreciate very much this comment regarding stoichiometry with other anions different to fluoride and chloride. Unfortunately, despite having attempted several experimental techniques, including different crystallization conditions, we were unable to unequivocally confirm the proposed stoichiometry. In any case, we consider that the invariability of the C3 symmetry of the cyclic peptide, as reflected in the NMR spectra, upon the addition of the different anions support our stoichiometry hypothesis. Another clear evidence found, also derived from the NMR studies, comes from the fact that the cyclic peptide adopts, after the additions, very similar conformations to that observed for the complexes with chloride and fluoride ions. The main difference between them is the chemical shift of amide proton (NH), whose down- field shift is correlated with the strength of the hydrogen bonds with the different anions and not with other conformational changes. In any case, following his/her advice we have modified the title of our work and also the abbreviation for each complex \((\mathbf{mA}\cdot \mathbf{nH}_2\mathbf{O}\subset 2\mathbf{CP}_2)\) , in which "m" represents the number of anions (A) that form the entrapped clusters.  
+
+- Comment. "Please provide a brief explanation of the results obtained from DFT."  
+
+Response. We thank this referee for asking about the DFT studies, which mainly confirm the stability of the proposed dimers and give some additional structural insights, hydrogen bonds distances, angles, and so on, to those species that have not crystallized. The truth is that we decided not to go on describing these results, which mainly confirmed what was deciphered with other experiments, so as not to make this article too long. Following the referee demand, we have decided to include an additional section in the supporting information (Supplementary Discussion III) in which we briefly describe the most relevant DFT results.  
+
+- Comment. "TOC can be improved; the current TOC does not represent the anion-water clusters clearly in the supramolecular dimer."  
+
+Response. We thank the reviewer for his kind suggestion. Therefore, a new TOC figure is attached to better represent the anion- water clusters in the supramolecular dimer.  
+
+## Reviewer: 3  
+
+- Comment. "This paper reports the synthesis of a cyclic peptide which is furnished with a cap to produce a hemispherical unit which dimerises through hydrogen bonding interactions. The two halves of the sphere are shown to open up to encapsulate fluoride and chloride in solution and in the solid state. In the crystal structure, there is clear evidence of water being involves in the formation of a specific cluster. In solution, water is required, but the structural nature of the cluster is not clearly demonstrated experimentally. Membrane transport is demonstrated, and explored using a conventional set of experiments.
+
+<--- Page Split --->
+
+My view is that this paper contains some very nice supramolecular chemistry work, and that the importance of water in the process is a highlight, though not as important as the introduction makes out. The experiments and analysis are all ably conducted and meet all the requirements of the field. It's all very well done, and will be likely used as an example of hydrated anions being encapsulated (though plenty of crystal structures exist of anion complex hydrates already) rather than built upon directly by others."  
+
+Response. We thank the referee for the positive review of our supramolecular chemistry work, highlighting the important role played by water molecules in the recognition properties of capped \(\alpha ,\gamma\) - cyclic peptides. We are glad to find his interest and confirmation about the analysis and experimental planification and development to clearly stablish the encapsulation of hydrated clusters. It is true that there are a number of crystal structures having anion complexes hydrated, although only a few of them describe polyanionic cluster of the type here reported.  
+
+- Comment. "The introduction contains some statements which need work - electrostatic interactions are a noncovalent interaction (top of p2), proteins are anion-selective and usually hold anions in a largely desolvated state. It's too much to call a host-guest complex a 'new state of matter."  
+
+Response. We very much regret the lack of precision in the wording of some concepts mentioned in the introduction. We have modified the introduction to include the "directional" noncovalent interactions to differentiate to those pure electrostatic between cations and anions. Now we mention: "The design principles used in this development are generally dominated by directional noncovalent interactions, such as halogen, chalcogen or hydrogens bonds, among others, rather than by purely electrostatic ones, as they generally provide better selectivity."  
+
+It is true that the majority of anion- selective proteins recognize them in their desolvated state, however some authors have proposed that the selective transport of fluoride carried out by the FLUC protein, which presents a remarkable selectivity against chloride (Gomez, D. T.; Pratt, L. R.; Asthagiri, D. N. Rempe, S. B. Acc. Chem. Res. 2022, 55, 2201- 2212) or the recognition of other anions such as sulfate (K. Balamurugan, M. T. Pisabarro, ACS Omega 2021, 6, 25350- 25360) could be related to the recognition of their hydrates in a similar way to the selective transport of sodium (versus potassium) in the sodium channel proteins.  
+
+We have modified the mentioned sentence at the introduction in the following way:  
+
+"In some cases, the ion selectivity in aqueous media does not depend as much on the relative size or charge of the ions, but on the size of their corresponding hydrates.[9,10] In fact, some authors have proposed that the selectivity in fluoride transport by FLUC protein could be explained in a similar way to the \(\mathrm{K^{+} / Na^{+}}\) selectivity of sodium ion channels.[11] Additionally, the greater solubility of negatively- charged proteins may be related to stronger hydration free energies for negatively charged groups on protein surfaces than their cationic counterparts.[12]"  
+
+Regarding the term "new state of matter", we would like to make some clarifications to this concept. From our point of view, molecular capsules are much more complex molecular containers than the "classic" molecular receptors characteristic of host/guest chemistry, since
+
+<--- Page Split --->
+
+they can recognize more than one substrate, they can incorporate a variable number of solvent molecules, they have a dynamic behavior in which the entrapped molecules flow at different rates among others. With that phrase we intended to emphasize that molecular capsules, such as those reported by Fujita and other authors, generate closed spaces that are more voluminous than conventional molecular receptors, can incorporate ad hoc reactive elements, trap macromolecules or proteins, maintaining or modifying their reactivity in which the substrates pass through their molecular walls, so they can be filtered or sorted on demand among other unique properties. In any case, since it seems that we have not depicted clearly enough this concept, this sentence has been modified to indicate the following: "They can be considered to create a new kind of molecular environment in which trapped molecule(s) exist."  
+
+- Comment. "There should be consideration of the possibility of deprotonation by fluoride. Solvent needs to be given clearly in the main text when talking about NMR titrations etc."  
+
+Response. We appreciate his advice and concerns very much. Deprotonation of amide groups was something we were very worried about from the beginning of this experiment, and this was one of the reasons why we decided to carry out all the experiments in dichloromethane and not in chloroform, in which, in addition to its well- known decomposition forming HCl, the deprotonation of chloroform with fluoride generates DF (deuterium fluoride) that exchange with amide NH and consequently the integration values of this proton signal decreases with time. This is something we have never seen in our studies. We are very confident that deprotonation with fluoride is not taking place in these experiments, as confirmed by the x- ray crystallographic studies. We believe that the recognition of the anion clusters reduces the intrinsic basicity of this anion. Since we have now included the \(^1\mathrm{H}\) NMR integrals, this can be checked in the proton spectra corresponding to \(3\mathrm{F}\cdot 8\mathrm{H}_2\mathrm{O}\subset 2\mathrm{CP}_2\) , in the Supplementary Information, section 8.  
+
+Solvent conditions were added in the manuscript following the referee advise. In general, deuterochloroform was used for the NMR characterization of dimers (homo and heterodimers) and titration experiments with bis- nitrile compounds (malononitrile and succinonitrile). For those experiments in which anion were added, a mixture of acetonitrile and dichloromethane (10%) was used.  
+
+Highlighted manuscript is attached to this letter.
+
+<--- Page Split --->
+
+## Recognition of anion-water clusters by peptide-based supramolecular capsules  
+
+Victoria López- Corbalán, \(^†\) Alberto Fuertes, \(^†\) Antonio L. Llamas- Saiz, \(^‡\) Manuel Amorín, \(^†\) & Juan R. Granja \(^†*\)  
+
+\(^†\) Centro Singular de Investigación en Química Biolóxica e Materiais Moleculares (CIQUS) and Organic Chemistry Department, University of Santiago de Compostela, and \(^‡\) Unidad de Rayos X; Área de infraestructuras de Investigación, RIAIDT Edificio CACTUS, University of Santiago de Compostela, 15782, Santiago de Compostela, (Spain).  
+
+The biological and technological importance of anion- mediated processes has made the development of improved methods for the selective recognition of anions one of the most relevant research topics today. The hydration sphere of anions plays an important role in the functions performed by anions by forming a variety of cluster complexes. Here we describe a supramolecular capsule that recognizes new hydrated anion clusters. These clusters are most likely composed of three ions that form hydrated C3 symmetry complexes that are entrapped within the supramolecular capsule of the same symmetry. The capsule is made of self- assembled Acp- based cyclic peptide equipped with a tris(triazolylethyl)amine cap. To recognise the hydrated anion clusters, the hexapeptide capsule must disassemble to entrap them between its two subunits.  
+
+Anion recognition is a biologically and technologically relevant process in which ad hoc designed receptors are able to bind to a specific ion triggering a specialised function.[1- 2] In the last years, a large number of new synthetic receptors have been developed for this purpose.[3- 4] The design principles used in this development are
+
+<--- Page Split --->
+
+generally dominated by directional noncovalent interactions, such as halogen, chalcogen or hydrogens bonds, among others, rather than by purely electrostatic ones, as they generally provide better selectivity.[5,6] In this regard, nature is able to differentiate biologically relevant ions of similar properties, such as fluoride or chloride, to carry out selective cross- membrane transport with remarkable selectivity.[7,8] The strong hydration of anions is one of the main obstacles to their recognition due to the additional energetic penalty paid for their desolvation during host- guest binding. In some cases, the ion selectivity in aqueous media does not depend as much on the relative size or charge of the ions, but on the size of their corresponding hydrates.[9,10] In fact, some authors have proposed that the selectivity in fluoride transport by FLUC protein could be explained in a similar way to the \(\mathrm{K^{+} / Na^{+}}\) selectivity of sodium ion channels.[11] Additionally, the greater solubility of negatively- charged proteins may be related to stronger hydration free energies for negatively charged groups on protein sufaces than their cationic counterparts.[12] Therefore, water- anion clusters represent an important chemical complex that is not yet fully understood despite their biological relevance. Several experimental and theoretical studies have been carried out to unmask the behaviour of hydrated halide clusters or other ions.[13,14,15,16] Most of these clusters are formed by a single ion surrounded by a variable number of water molecules, although polyanionic clusters have also been reported.[17] The recognition of two or more anions is critical to many biological processes, such in chloride transport.[18] One important approach to study this type of clusters is by analysing those generated in confined spaces.[19,20] Thus, several water clusters have been characterized by studying those formed in small protein pockets or other nanostructured materials. In this work we describe a new class of hydrated polyanionic clusters, most likely three, embedded between two cyclic peptide subunits of a novel supramolecular dimeric capsule.  
+
+# Figure 1.
+
+<--- Page Split --->
+
+Supramolecular capsules are dynamic molecular assemblies composed of two or more self- complementary units held together by non- covalent interactions that, upon assembling, form an empty closed system whose internal cavities are useful tools in host- guest chemistry.[21,22] They can be considered to create a new kind of molecular environment in which trapped molecule(s) exist. A wide variety of supramolecular capsules have been prepared with complementary or opposite functions, ranging from large to small, from charged to neutral or from open to closed systems. They exhibit unique properties, such as remarkable spectrochemical, electrochemical or magnetic effects. Therefore, they are useful tools for the molecular recognition of a wide variety of anionic, cationic or neutral guests (including gases) and are also able to catalyse a number of reactions leading to unusual products in a remarkably efficient way.[23,24,25,26] Recently, we have shown that flat- shaped cyclic peptides[27,28] equipped with a molecular cap (porphyrin moiety) provide a hydrogen- bonded dimeric receptor that was able to entrap long linear 4,4'- bipyridyl derivatives (Figure 1).[29] Here, we have created a smaller variant topped with a tris- triazolylamine component that is able to recognize a new class of water- anion clusters. To do this, the amide protons involved in the hydrogen bonds that supramolecularly bind both receptor subunits participate in the recognition of the hydrated anion. These clusters sandwiched between the two cyclic peptide components resembles a “molecular burger”, in which the anionic complexes are intercalated, like meat patty and cheese, between the two subunits that play the role of a “molecular bun”. The isobutyl pendants of Leu side chain are cross- interdigitated providing a hydrophobic supramolecular complex that can facilitate ion transport across lipid membranes.  
+
+Results and Discussion. Recently, we have prepared the cyclic hexapeptide bearing three properly chains attached to the \(\gamma\) - residues (CP1) that are pointing opposite to the dimerization plane (D1).[30,31,32] These groups not only prevent stacking of cyclic peptides across that face, thereby restricting the assembling process to discrete
+
+<--- Page Split --->
+
+dimers rather than nanotubes, but can also act as valuable reactive element for the incorporation of other functional moieties at the entrance of the dimer (Figure 1). For example, Sonogashira cross- coupling provided derivatives bearing ortho- or meta- oriented pyridine moieties.[31] The resulting peptides were able not only to encapsulate Xe atoms in the tubular structure of stacked dimers, but also to efficiently transport ion pairs in model membranes.[30] Additionally, the acetylene moiety was used in a copper- catalysed alkyne- azide cycloaddition reaction to incorporate a variety of oligoethylene glycol pendants to generate novel membrane spanners.[32] Here, we describe the preparation of self- assembled cavitands (Figure 2) covered by a tris- triazolyl motif attached to a central core that not only plays the role of a molecular cap but also acts as a binding site for the recognition of small anions.[33,34,35,36,37,38]  
+
+## Figure 2.  
+
+Taking in account the aforementioned objectives, the cyclic peptide CP1 was synthesized using the protocol described previously (Figure 2).[30] For the cap incorporation, CP1 was subjected to a triple click reaction using the previously optimised method in which catalytic amounts of air- sensitive copper complex [Cu(MeCN)4PF6] (10 mol% per alkyne) were used in presence of DIEA and TBTA ligand.[32,39] Under these conditions the expected cavitand CP2 was obtained, although in low yield (15%). After screening for improving reaction conditions, a new protocol was found using copper bromide and DBU in toluene, providing CP2 in 48% yield.[40]  
+
+## Figure 3.  
+
+This peptide retained the self- assembling properties of its precursor, CP1, adopting the proposed flat conformation and forming the C3- symmetric dimer (D2), as denoted by NMR experiments (Supplementary Fig. 1).[41,42] Clear evidence of dimer formation was derived from the down- field shift of the amide proton (8.04 ppm) whose
+
+<--- Page Split --->
+
+chemical shift did not change with CP dilution up to \(200~\mu \mathrm{M}\) in \(\mathrm{CD}_2\mathrm{Cl}_2\) , confirming the large dimerization constant ( \(\mathrm{Ka} > 10^{4} \mathrm{M}^{- 1}\) ) similar to the non- capped derivatives.[29] Further evidence of the dimeric structure derived from the MS (2045.25 [ \(\mathrm{M}_{\mathrm{D2}} + \mathrm{Na}]^+\) , Supplementary Fig. 2) and FTIR spectroscopy, with amide A band a \(3286~\mathrm{cm^{- 1}}\) and carbonyl vibrations at \(1623\) and \(1532~\mathrm{cm^{- 1}}\) , characteristic of the antiparallel- type \(\beta\) - sheet structure (Supplementary Fig. 3).[43,44,45] The final evidence was obtained from the X- ray crystallography analysis of a crystal obtained from a solution of \(\mathbf{D2}\) in \(10\%\) \(\mathrm{CD}_3\mathrm{CN}\) in \(\mathrm{CD}_2\mathrm{Cl}_2\) from which a dimeric supramolecular capsule held by six hydrogen bonds between both subunits was found (Figure 3a- b). Interestingly one acetonitrile moiety is entrapped in the inner cavity of this dimer, with its nitrile group pointing towards the protons of the triazolyl moieties of one of the subunits. On the other side, one of the triazole moieties is rotated, arranging two of the nitrogen atoms of the heterocycle towards the internal cavity, perhaps with the goal of partially filling it due to the lack of a suitable guess. We checked if bis- nitrile derivatives, i.e., malononitrile (MN), could interact simultaneously with both tris- triazole motives improving its encapsulation (Supplementary Figs. 4, 5 and 6). We found out that it was necessary to add up to 30 equivalents of malononitrile (Supplementary Fig. 5) to a dichloromethane ( \(\mathrm{CD}_2\mathrm{Cl}_2\) ) solution of \(\mathbf{D2}\) to get a new species (MN \(\mathbf{\sigma} = \mathbf{D2}\) ), in a 4 to 1 ratio with respect to the empty capsule. This new species was assigned to the encapsulated complex due to the down- field shift of triazole proton ( \(\Delta \delta = 0.06\) ppm). 2D- NMR experiments confirmed that \(\mathrm{H}\beta_{\mathrm{Acp}}\) (cis- oriented with carboxy and amino groups) suffered down- field shift (Supplementary Fig. 6), which is consistent with the incorporation of a malononitrile molecule in the cavity. The weak interaction is most likely due to the \(109^{\circ}\) angle between the two nitriles, which possibly hinders the simultaneous formation of strong hydrogen bonds with both tris(triazolylethyl)amine moieties. Succinonitrile (SN) was also evaluated, considering its additional methylene moiety allows both cyano groups to be arranged at a \(180^{\circ}\) angle, although we were concerned that its length would exceed
+
+<--- Page Split --->
+
+the size of the cavity. Fortunately, the addition of small amounts of SN ( \(\sim 0.5\) equiv) over a \(\mathrm{CD}_2\mathrm{Cl}_2\) solution of D2 (6.4 mM) already gave rise to a new set of signals corresponding to \(\mathbf{SN}\subset \mathbf{D2}\) (Supplementary Figs. 4 and 7), suggesting a higher affinity than MN. After the addition of 7 equiv, the encapsulated complex ( \(\mathbf{SN}\subset \mathbf{D2}\) ) is already the main component in the mixture, although in a 1.4:1 ratio with D2 (Supplementary Fig. 7A). Whereas for the MN recognition there is almost no change in the chemical shift of amide protons, for the SN the signal is up- field shifted (7.84 ppm) suggesting a tight binding that reduced the conformational freedom (signals are sharper than those of the free capsule) and stress the hydrogen bond network. The new singlet at 3.30 ppm corresponds to the methylene groups from the entrapped SN molecule as indicated the 2D- NMR experiments (Figure S7B- C). DFT geometry optimizations confirm the stability of both complexes (Supplementary Fig. 8, see Supplementary Discussion III for further information). The main difference between both encapsulated bis- nitrile complexes was the length of hydrogen bonds, which become longer as the size of the encapsulated nitrile increases.  
+
+## Figure 4.  
+
+Once the ability of D2 to form hydrogen- bonded supramolecular inclusion complexes with appropriate guests and the capability of proton of tris- triazole cap to form hydrogen bonds with the entrapped molecule was confirmed, the recognition of small anions was evaluated. Initially, we assessed halide recognition using NMR titrations (specifications regarding \(^1\mathrm{H}\) NMR titrations are detailed in Supplementary Discussion I). Clearly, addition of different portions of tetrabutylammonium fluoride to a solution of CP2 (7.3 mM) in \(10\%\) \(\mathrm{CD}_3\mathrm{CN} / \mathrm{CD}_2\mathrm{Cl}_2\) gave rise to a new set of signals that were attributed to the recognition of the fluoride anion (Figure 4). In any case, it was necessary to add ca. 3 equivalents of TBAF to achieve the complete disappearance of the signals corresponding to D2. Similar results were obtained with the addition of
+
+<--- Page Split --->
+
+TBACI (Supplementary Fig. 9) under similar conditions (10% \(\mathrm{CD_3CN / CD_2Cl_2}\) ), but in this case it was necessary to add more than 15 equivalents of chloride to fully shift the equilibrium towards the new species, suggesting a weaker interaction (Table 1). No changes were observed for the addition of larger halides (bromide and iodide) or nitrate. Careful analysis of the \(^1\mathrm{H}\) NMR spectra of both experiments reveal that, as opposed to what it was found in anion recognition with tris(triazolyl) receptors,[33- 38] the heterocyclic protons are up- field shifted (7.47 ppm). This suggests that these protons must not be participating in halide recognition. On the other hand, the amide proton signals are down- field shifted (from 8.00 to 10.00 ppm for fluoride and to 8.62 ppm for chloride), implying their participation in stronger hydrogen bonds. Conformational changes derived from the interaction with the anions are clearly evidenced by the splitting of the signal of the methylene linker (singlet at 4.59 ppm for D2) between the triazole rings and the cyclic peptide backbone into two doublets at 4.90 and 4.04 ppm. In addition, conformational changes at the CP backbone suggest changes in the flat conformation characteristic of the sheet structure. For example, \(\mathrm{H\alpha_{Leu}}\) , initially at 4.81 ppm, undergoes a remarkable up- field shift to 4.15 ppm for fluoride (4.19 for chloride), while the \(\mathrm{H\gamma_{Acp}}\) , initially at 4.73 ppm, undergoes an opposite, albeit milder, down- field shift to 4.90 ppm (4.83 ppm for chloride). Two- dimensional NMR experiments provided evidence of conformational changes, in which nOe cross- peaks that are not present in the empty capsule appear after addition of the halide (Supplementary Fig. 10), such as the one between \(\mathrm{H\gamma_{Acp}}\) and one of the protons of \(\mathrm{H\beta_{Leu}}\) upon recognition of fluoride. The disappearance, upon complexation, of the strong nOe cross- peaks between \(\mathrm{H_{triazole}}\) and \(\mathrm{H\alpha_{Leu}}\) and of the methylene linker and the \(\mathrm{H\beta_{Leu}}\) also seems relevant. Further evidence of the halide and CP interactions were found at ESI- MS, in which several ion peaks corresponding to complexes between both components are the main detectable species (Supplementary Fig. 11).  
+
+Table 1.
+
+<--- Page Split --->
+
+In a search for improving the binding, we discovered that water molecules play a key role in the supramolecular recognition process. The addition of 4Å molecular sieves, with the aim of drying the solution of TBACl and D2, gives rise to the recovery of the starting dimeric capsule. Filtration of the resulting solution and addition of 10 μL of water provoke the recovery of the chloride-capsule complex (3Cl.nH2O<2CP2) spectrum (Supplementary Fig. 12A). Inspired by this finding, new studies with the other larger and less hygroscopic anions were reviewed to evaluate the role of water molecules in their recognition. The addition of a few drops of water (7 μL) to the 10% CD3CN/CD2Cl2 solution of D2 (4.7 mM), resulted in only a small down-field shift of the triazole proton signal, presumably due to water encapsulation (Supplementary Fig. 13); but successive additions of increasing amounts of TBAB yielded similar changes to those already described for chloride additions (28 equivalents added and \(\delta_{\mathrm{NH}} 8.37 \mathrm{ppm}\) , Table 1) (Supplementary Fig. 14). Remarkably, even recognition of iodide was also observed, although more equivalents of iodide (86 equiv, Table 1) were required, and the down-field shift of NH signal (8.07 ppm) was even smaller (Supplementary Fig. 15). This definitely underlines the importance of water molecules in the anion recognition process, most likely due to the ability of the capsule to recognize the hydration spheres of the anions.[9] Next, other anions (see Table 1, Supplementary Figs. 16, 17, 18 and 19) were evaluated and a similar behaviour was found; those that were recognised, such as nitrate, acetate or azide ions, only did so in the presence of some water molecules. Larger ions, such as tribromide or hexafluorophosphate did not interact even in the presence of different amounts of water. These results confirm the remarkable ability of D2 to recognise the hydration spheres of different anions. We decided to further study the recognition of acetate to stress out the importance of water molecules in the complexation process by drying this solution. After incubating a solution of D2 containing 66 equivalents of TBAA with molecular sieves for 48 hours, the \(^1 \mathrm{H}\) NMR signals corresponding to D2 were recovered, while those assigned to the original
+
+<--- Page Split --->
+
+complex disappeared. The subsequent addition of few drops of water (10 μL) immediately re- established the original complex (Supplementary Fig. 12B). Interestingly, while overnight incubation was sufficient to dehydrate the encapsulated chloride complex with molecular sieves, drying the acetate mixture required two days for the full recovery of D2, suggesting a tighter binding. Apart of the key role paid by water molecules, the chemical shift and number of equivalents of the different anions tested suggest that size, shape, and basicity are important parameters in the recognition process. To confirm the role played by the capsule and the tris(triazolylethyl)amine cap, CP1 was evaluated for the recognition of anions. TBAF additions over a solution of D1 in dichloromethane with a 10% of acetonitrile did not provide any change in its \(^1\mathrm{H}\) NMR spectrum, confirming the relevance of the cap in the anion/water cluster recognition (Supplementary Fig. 20). The role of the tetrabutylammonium counterion was also evaluated, for this purpose we decided to make use of the known affinity of crown ethers for alkali cations. \(^{[46]}\) The 15- crown- 5 is known for being the one whose radius fits better with \(\mathrm{Na}^+\) . \(^{[47]}\) Therefore, we were able to solubilize NaOAc in deuterated acetonitrile using this crown ether and this solution was used in new titration experiments with D2 in dichloromethane. Similar results were obtained, although larger amounts of the sodium acetate/crown ether acetonitrile solution were required to shift the equilibrium, most likely due to the lower solubility of the complex under these conditions (Supplementary Fig. 21A). In fact, the NMR spectrum of the mixture two weeks after titration showed a clear increase of the acetate complex with respect to the initial conditions, going from 1.4:1 ratio to 3:1 (Supplementary Fig. 21B). This can be attributed to the slow solubilization of the acetate salt triggered by the formation of the complex. ROESY spectra showed a similar cross- peaks pattern to those obtained using tetrabutylammonium as counterion (Supplementary Fig. 22).  
+
+The strong down- field shift of amide protons upon anion addition suggests their involvement in its recognition, consequently, further experiments were carried out to
+
+<--- Page Split --->
+
+understand this behaviour. It is well stablished that cyclic peptides made by five- membered ring \(\gamma\) - amino acids (Acp) form heterodimeric complexes with CPs containing six- membered ring \(\gamma\) - amino acids, i.e. CP3,[42,48,49] which are more stable than the corresponding homodimers. Therefore, we decided to evaluate the importance of capsule integrity in the cluster recognition. Consequently, CP3, which was prepared following a similar strategy to the one used in the preparation of CP2 but starting from 3- aminocyclohexanecarboxylic acid (Ach, Supplementary Fig. 23), also forms the corresponding dimer D3 in dichloromethane solution. CP3 was mixed with a solution of D2, and the resulting mixture showed in the NMR spectra the appearance of a new set of signals that were assigned to the supramolecular aggregate D2- 3 (Figure 5 and Supplementary Fig. 24). Further evidence of the heterodimeric structure derived from MS experiments (Supplementary Fig. 25) in which the 1840.18 ion peak corresponds to the heterodimeric complex. This confirmed, once again, the higher stability of the heterodimeric aggregates because of the better van der Walls fitting between both cycloalkyl rings.[50] The formation of this heterodimer allowed us to evaluate if dimers containing only one tris(triazolylethyl)amine cap (cavitand D2- 3) could still recognize anions. Interestingly, after the addition of small portions of fluoride or chloride (TBAF or TBACl) on the \(\mathrm{CD_2Cl_2}\) solution of heterodimer D2- 3, simultaneous recovery of homodimer D3 and the formation of the corresponding complexes of CP2 with the anions (3F- nH2O<2CP2 and 3Cl- nH2O<2CP2) were observed. Three equivalents of fluoride were also required to shift the equilibrium to the homodimeric components. This confirms that the interaction with hydrated anions is even more favourable than that of the heterodimer and that both capped subunits are necessary in this process.  
+
+## Figure 5.  
+
+Cyclic peptide CP3 was then transformed into the tris(triazolylethyl)amine capped Ach derivative (CP4) using similar conditions in \(50\%\) yield (Figure 2).
+
+<--- Page Split --->
+
+Therefore, the recognition properties of the new Ach- based capsule topped with the tris(triazolylethyl)amine motif were also studied. These cavitand self- dimerized in dichloromethane solution to form D4, as denoted by NMR, MS and FTIR (see Supplementary Information section 7 and 8). Once again, single crystal suitable for X- ray diffraction of this compound also confirmed the capsule formation that also entrap one acetonitrile molecule in its cavity (Figure 3C- D). D4 has all the triazole protons pointing towards the internal cavity making its structure more symmetric than the Acp- based one (figure 3a- b), with similar length in all the interpeptide hydrogen bonds. These are generally longer (2.24 Å) than those of D2, which range from 1.91 to 2.35 Å (2.10 Å in average), although the capsule D4 is slightly more compact with a shorter distance between the two nitrogen atoms of the tris(triazolylethyl) caps (14.30 Å versus 14.85 Å). In contrast to the recognition properties of D2, additions of more than twenty equivalents of TBAF or TBACl do not result in any change in the NMR spectra of D4 (Supplementary Fig. 26). This indicates that the Ach- based capsule is not capable of recognising anions, most likely due to the greater rigidity of the six- membered ring of this \(\gamma\) - amino acid that prevents the CP from adopting the appropriate conformation for the recognition of such species.  
+
+Both cavitands (CP2 and CP4) also assemble into the heterodimer D2- 4 (Figure 5 and Supplementary Fig. 27A) when an equimolar mixture of both compounds in nonpolar solvents (CD₂Cl₂) is prepared. The appearance of new signals in the \(^1\mathrm{H}\) NMR spectra that do not correspond to any of the homodimers confirms the formation of D2- 4. For example, the broad signal at 4.95 ppm belonging to the HαLeu and the one at 4.22 ppm corresponding to one the methylene of the tris(triazolylethyl)amine cap are signals that belong to CP2 of the heterodimer (Supplementary Fig. 27B). Moreover, the MS also confirms the formation of D2- 4 (Supplementary Fig. 28). Once again, addition of fluoride (TBAF) to this dichloromethane (CD₂Cl₂) mixture prompts the splitting of the heterodimer D2- 4 into the corresponding homodimer D4 and the complex of hydrated
+
+<--- Page Split --->
+
+fluoride with cavitand \(\mathbf{CP2}\) \((3\mathrm{F}\cdot \mathrm{nH}_2\mathrm{O}\subset 2\mathrm{CP2})\) . In all cases, the addition of more than three equivalents of fluoride was necessary to achieve complete dissociation of the heterodimers. DFT geometry optimization of both heterodimeric species, D2- 3 and D2- 4, are shown in Supplementary Fig. 29, which once again confirmed the stability of the mentioned dimers and provide further information about the hydrogen bonds length and angles (see also Supplementary Discussion III for further information).  
+
+## Figure 6.  
+
+Crystals suitable for X- ray diffraction were obtained from solutions containing the supramolecular capsules D2 and the tetrabutylammonium halides (chloride and fluoride, Figure 6). To our delight, in both cases the cyclic peptide capsules entrap a new kind of hydrated halide clusters made by three ions, as confirmed by the number of tetrabutylammonium ions that co- crystalized with the peptide capsule. Four and eight water molecules, for chloride and fluoride, respectively, are forming these clusters, confirming the higher tendency of the latter to have larger hydration shells. In both cases the three halide ions are distributed into six equivalent chemical positions that are shared with another three water molecules. Although for the chloride crystal all the ion positions are forming a hexagonal structure with all the positions placed at the same layer, for fluoride cluster, the six positions are placed at two different levels forming two triangular structures that are \(60^{\circ}\) rotated with respect to each other. To entrap these clusters, cyclic peptide dimers dissociate to allow amide protons to hydrogen- bond with halide ions and water molecules.[51] It is notorious that in the solid state all the carbonyls are oriented towards the opposite side in which the interaction with the anionic cluster occurs, which could explain the observed up- field shift of \(\mathrm{H}\alpha_{\mathrm{Leu}}\) in the NMR spectrum. This conformational change is due to the geometrical variations of the \(\alpha\) - amino acids that go from the characteristic \(\beta\) - sheet conformation of flat disc- shaped CPs to a turn-
+
+<--- Page Split --->
+
+like structure. Interpeptide distance is slightly larger for the encapsulated fluoride cluster than that for chloride despite the larger size of the latter.  
+
+With respect to the fluoride- capsule complex (Figure 6a- b), the unit cell has two non- equivalent complexes \((3\mathrm{F}\cdot 8\mathrm{H}_{2}\mathrm{O}\subset 2\mathrm{CP}2)\) . In each complex, in addition to the three water molecules exchangeable by fluoride ions and hydrogen bonded to the amide proton, there are five other crystallographic positions preferably occupied by water molecules, although fluoride could also partially occupy any of these positions, since it is not possible to unambiguously differentiate both atoms due to their similar electron densities. In any case, to fulfil the hydrogen acceptor capability of the fluoride ion, we assume that these ions must be located in the positions in which they are bonded to the amide protons of the same cavitand and surrounded by three water molecules forming the first hydration shell of each fluoride (Supplementary Fig. 30). The fluoride occupancy in the two \(3\mathrm{F}\cdot 8\mathrm{H}_{2}\mathrm{O}\subset 2\mathrm{CP}2\) complex of each unit cell is not exactly the same (Supplementary Fig. 31), even though both complexes present analogue disposition. With respect to the rest of water molecules, there are two that are axially placed, forming hydrogen bonds with the fluoride- water exchangeable positions, while the other three are in the cluster equatorial perimeter forming hydrogen bonds with the exchangeable fluoride- water positions (Supplementary Fig. 32). These water molecules are placed at the position in which capsule is not fully closed forming a window. The Leu side chains are facing each other to create a hydrophobic oval- shaped aggregate that entrap the halide cluster in a non- polar environment.  
+
+Concerning the chloride complex (Figure 6c- d) the six equivalent positions are in the same plane and each chloride ion is hydrogen- bonded to one amide proton with N...Cl distance of 3.21 Å. In this case, the coating with the leucine side chains is less compact, leaving a wider window as compared to the fluoride complex. Furthermore, electron density can only be attributed to a maximum of four water molecules, one of
+
+<--- Page Split --->
+
+them located at \(50\%\) occupancy at the top and bottom of the hexagonal bipyramid and more deeply buried in the cavity of the supramolecular capsule, remaining partially hydrogen- bonded to the three chloride ions. In this crystal structure, the complex of the capsule with the chloride cluster is co- crystallized with the dimer D3, forming columns in which D3 is alternated with the encapsulated trichloride cluster (3CI.4H2O<2CP2). Within D3 there is a dioxane molecule occupying three equivalent positions around the ternary symmetry axis of the dimer (Supplementary Fig. 33).  
+
+Unfortunately, it was no possible to obtain crystal structures of the complexes formed with the other anions (bromide, acetate, azide, iodide and so on) that would allow unequivocal confirmation of the formation of similar structures with these ions. In any case, the previous characterizations suggest the formation of clusters that are embedded in the equatorial cleft generated by the two CP subunits. To confirm this further, we carried out a detailed analysis comparing the NMR data of the different complexes and the X- ray diffraction data (Supplementary Discussion II), from which we concluded that the recognition process most likely involves the trapping by two CP2 units of clusters composed of three anions, although we do not have a conclusive evidence of such stoichiometry, surrounded by several water molecules depending on the type of anion and its solvation. To this complex we have used the coding mA. \(\mathbf{nH}_2\mathbf{O}\mathbf{C}2\mathbf{CP}2\) for the complex with the anion cluster in which "m" represents the number of anions entrapped between both CPs (most likely three), "A" indicates the type of anion and "n" represents the number of water molecules that form each anion cluster.  
+
+After confirmation of the anion recognition ability of the supramolecular capsule D2, transmembrane transport experiments were carried out.[1] For that purpose, lucigenin- trapped liposomes (LG<LUV) were prepared with which the intravesicular delivery of chloride ions was clearly stablished (Figure 7A).[52] The transport is slow
+
+<--- Page Split --->
+
+compared with previously published chloride transporters and high concentration of capsule is required \((\mathrm{EC}_{50} \sim 100 \mu \mathrm{M})\) . Additionally, we decided to examine the chloride transport ability of D4 and D1 using the lucigenin assay. As we expected from the previous NMR findings, neither D1 nor D4 were capable of transporting chloride (Supplementary Fig. 34), confirming that the recognition of the ions was necessary to be able to mediate its transport. Apart from that, further variations of the D2 lucigenin assay, revealed that the transport efficiency did not change with the counterion used \((\mathrm{Na}^{+}, \mathrm{Li}^{+}, \mathrm{K}^{+} \text{or} \mathrm{Cs}^{+})\) , suggesting that cation is not involved in the transport process (Figure 7A3). To confirm the potential antiport transport of nitrate, experiments in which nitrate was substituted for the more hydrophilic sulphate, whose transmembrane transport is extremely difficult, were carried out (Supplementary Fig. 35). [53,54] These experiments did not show any significant reduction in chloride transport rates, suggesting that chloride/nitrate exchange must not be involved in the transport mechanism. Therefore, the symport \((\mathrm{H}^{+} / \mathrm{Cl}^{- })\) or antiport \((\mathrm{OH}^{- } / \mathrm{Cl}^{- })\) must be associated to this migration. To confirm association of chloride transport with change in the pH, HPTS loaded vesicles (HPTS \(\equiv\) LUV) were used (Figure 7B). [55] Thus, vesicles basification promoted by D2, denoted by a fluorescence increase, would be associated to the co- transport of chloride. Unambiguous and fast enhancement of dye emission was found upon creating a pH gradient of almost one unit after the addition of a sodium hydroxide solution to the extravesicular medium, yielding an enhanced activity with \(\mathrm{EC}_{50} = 3 \mu \mathrm{M}\) , suggesting that anion transport was the rate limiting step and not the \(\mathrm{H}^{+}\) or \(\mathrm{OH}^{- }\) co- transport. To clarify this finding, HPTS studies in the presence of a proton transporter (FCCP) were carried out (Figure 7D). [56] As expected no changes in transport rates were found, confirming proton transport was not the limiting step. DLS measurements were then performed, which confirmed both the homogeneity and integrity of the vesicles throughout these experiments and, consequently, the observed
+
+<--- Page Split --->
+
+fluorescence changes are not due to bleaching or membrane disruption (Supplementary Fig. 36).  
+
+Finally, to evaluate the relative rates in anion transport, competitive experiments were carried out using HPTSCLUV (Figure 7C).[52,57] For this purpose, vesicles whose internal buffer contained sodium chloride (100 mM) at pH 7 were placed in a variety of isosmotic buffer solutions with different sodium salts with other counterions. In this type of experiments, a pH gradient is generated if the transporter (D2) facilitates dominant ion influx or efflux depending on anion selectivity. These permeability differences give rise to a membrane potential that drives net proton transport. Vesicle acidification occurs when ion influx is faster than chloride efflux, while ions that are transported more slowly than chloride cause an increase in intravesicular pH. We found that acetate, fluoride, and azide provided vesicle basification, while bromide and iodide were transported faster than chloride, being acetate the slower influxed anion and iodide the faster one. Therefore, D2 showed Hofmeister pattern \((\mathrm{I}^{-} > \mathrm{Br}^{-} > \mathrm{NO}_{3}^{-} \sim \mathrm{Cl}^{-} > \mathrm{N}_{3}^{-} > \mathrm{F}^{-} > \mathrm{OAc}^{-})\) with the exception of nitrate that is almost as fast as chloride.[58] The strongly hydrated, salting- out ions, are those with slower transport rates while previous ion recognition experiments showed the need of water molecules in the recognition of anion cluster and the strong binding for fluoride or acetate compared with iodide or bromide. Therefore, it seems that ions exchange \((\mathrm{K}_{\mathrm{on}} / \mathrm{K}_{\mathrm{off}})\) at the interfaces must play an important role in the transport process and not so much the selectivity of the binding itself. All the details regarding the transport assays are carefully addressed in Supplementary Discussion IV.  
+
+## Figure 7.  
+
+In conclusion, we have presented the design, synthesis, and anion recognition properties of a cyclic \(\alpha , \gamma\) - hexapeptide equipped with a tris(triazolylethyl)amine cap. We
+
+<--- Page Split --->
+
+have shown that, in presence of some anions, the self- assembled CP is able to reaccommodate its backbone conformation from a self- dimerizing flat circular- shaped structure to a triangular one that creates a binding pocket suitable for hosting hydrated anions. Moreover, we have rigorously identified and studied the different aspects involved in the recognition. First, the essential role played by the water molecules that are involved not only in the formation of anion- water clusters, forming the hydration shell of the different anions, and helping them to take the appropriate size and shape, but also in the interaction and entrapment inside the cavity formed by the two peptide hemicapsules. In addition, two key structural components of the cyclic peptide were identified: the five- member ring \(\gamma\) - amino acid and the tris(triazolylethyl)amine cap, without which the anion accommodation is not possible. Furthermore, we have shown that the cationic counterion does not play any relevant role in the supramolecular recognition, which supports our theory of the three main actors: capped peptide, water and anion.  
+
+Finally, the anion transporting properties were also explored. Our studies have shown that D2 can transport different anions across model lipid membranes, with a tendency to favour the transport of anions with weaker hydration spheres, which are generally weaker recognized. Transport is associated with the modification the pH of the intravesicular medium, most likely through a \(\mathrm{Cl}^-\) / \(\mathrm{H}^+\) symport mechanism, although the \(\mathrm{Cl}^-\) / \(\mathrm{OH}^-\) antiport exchange could not be ruled out, the rate- limiting step being the migration of the anions. The peptide composition, design simplicity and anion transporting properties linked to pH regulation activity open- up several avenues for the therapeutic use of these novel self- assembled systems.[9]  
+
+## Methods  
+
+General \(^1\mathrm{H}\) - NMR titrations protocol
+
+<--- Page Split --->
+
+Stock solutions of internal standard, either dioxane (2.5 mM) or TMSS (0.15 mM), in a mixture of \(\mathrm{CD}_3\mathrm{CN}\) in \(\mathrm{CD}_2\mathrm{Cl}_2\) (10% v/v) were prepared. These solutions were used to prepare all the samples needed in the titration experiments, in order to keep the concentration of the internal standard constant throughout the experiments. The signals from both standards, dioxane (s, 3.6 ppm in \(\mathrm{CD}_2\mathrm{Cl}_2\) ) or TMSS (s, 0.18 ppm in \(\mathrm{CD}_2\mathrm{Cl}_2\) ), were used to calculate the concentration of all the components along the titration experiments.  
+
+Generally, \(450~\mu \mathrm{L}\) of the solution containing the corresponding peptide capsule (D2, D3 or D4, ca. 5mM) was placed in the NMR tube. After recording the \(^1\mathrm{H}\) NMR spectra of the starting sample, successive additions of the corresponding titrant were made via micropipette and the corresponding spectrum was taken.  
+
+## General procedure for vesicles preparation  
+
+Vesicles were prepared, under argon, in a round bottom flask by slow evaporation of a solution of EYPC in \(\mathrm{CHCl}_3\) (25 mg/mL, 1 mL) to form a thin and homogeneous film on the flask surface. The film was dried overnight under high vacuum and then carefully hydrated with the intravesicular aqueous media. The resulting mixture was tumbled for 1 hour in the rotavapor at 180 r.p.m. but at atmospheric pressure. Every 15 minutes, the rotavapor rotation angle was changed (60°, 50°, 45° and 35°) in order to get and homogeneous dispersion. After that, the milky sample was subjected to 11 freeze- thaw cycles ( \(\mathrm{N}_2\) (l) \(\rightarrow 40^{\circ}\mathrm{C}\) water), and the resulting suspension was extruded (25 times) across polycarbonate membrane (200 nm pore size). Finally, the suspension was passed through a size exclusion column (Sephadex G- 25) previously equilibrated with the isosmotic extravesicular medium.[56] The resulting vesicle suspension was taken up in a total volume of 5 mL, giving an approximate lipid concentration 6.6 mM. Size and particle number consistency between different vesicles batches were checked using DLS.
+
+<--- Page Split --->
+
+## General procedure for transport measurement  
+
+General procedure for transport measurementIn a plastic cuvette containing a \(4\mathrm{mm}\) diameter stirring bar, the previously prepared vesicle suspensions (50 \(\mu \mathrm{L}\) ) were dispersed in the extravesicular media (1950 \(\mu \mathrm{L}\) ). The cuvette was placed with moderate stirring in the fluorometer equipped with a module that allows the successive additions of titrant during the measurement in the dark. Data (fluorescence emission band at \(535\mathrm{nm}\) for lucigenin essays or \(510\mathrm{nm}\) for HPTS experiments) were collected for thirty minutes every second. One minute after the experiment started, aqueous solutions of NaCl (25 \(\mu \mathrm{L}\) , 2 M, lucigenin assay) or NaOH (25 \(\mu \mathrm{L}\) , 0.5 M, HPTS assay) were added. After an additional minute (minute 2 of the measurement), a solution of the CP at different concentrations (25 \(\mu \mathrm{L}\) , in iPrOH for the lucigenin assay or in DMSO for the HPTS) were added. Finally, after 27 minutes an aqueous solution of Triton (50 \(\mu \mathrm{L}\) , \(10\%\) v/v) was added to provoke liposome lysis. After three more minutes of stabilization, the resulting signal was used for the data normalization.  
+
+## DLS measurements  
+
+DLS measurementsFor each batch of vesicle prepared, some control measurements were performed before the transport experiments to check the homogeneity between batches. The size, polydispersity index and particle concentration were measured using the same volumes as in the fluorescence assay. Four samples were collected and checked for each essay. Regarding the lucigenin experiments (Supplementary Fig. 36), the first sample, containing only the vesicles dispersed in the extravesicular media (blue line), the second sample, which corresponds to the mixture after the addition of the aqueous solution of NaCl (green line), the third measurement was carried out after the incorporation of CP (pink line), and finally, the fourth sample was done after the addition of Triton X- 100 (red line). The four samples were stirred for 25 minutes, and after a 5 min lag resting
+
+<--- Page Split --->
+
+perioded the DLS were recorded. For the HPTS assay (See Supplementary Fig. 36) in HEPES buffer (10 mM, NaCl 100 mM, pH 7), the analyzed mixtures correspond to the initial conditions (blue line), and after successive additions of NaOH (green line), cyclic peptide (pink line) and Triton (red line).  
+
+Notice that samples measurement indicates monodispersity, in view of the sharp slope of the correlation curves, which was maintained in all samples except after the addition of the surfactant, Triton X- 100, when the vesicles undergo lysis. After lysis, the faster decay of the correlation coefficient is a signal of the smaller particle size. Also, the slope, less sharp, indicates higher polydispersity. We measured the particle concentration, which also indicated homogeneity between batches and confirmed the stability of the vesicles until their lysis.  
+
+## DFT geometry optimizations  
+
+Computational studies were carried out using the sources from Centro de Supercomputación de Galicia (CESGA). All calculations were performed with the Gaussian 16 rev. C01 package. The geometries used in the calculations were based on the crystal structures derived from this study. Calculations were performed in vacuum. We carried out DFT geometry optimizations using B3LYP with the moderate- size basis set 6- 31G (d, p). We also included GD3BJ as empirical dispersion.  
+
+Supplementary Information accompanies the paper on https://www.nature.com/nchem/. Detailed descriptions of the synthesis and characterization of key compounds, including NMR spectra ( \(^1\mathrm{H}\) and \(^{13}\mathrm{C}\) , NOESY and/or ROESY) and FTIR spectra of peptides CP2, CP3 and CP4. CCDC- 2311116 to CCDC- 2311119 contains the supplementary crystallographic data for this paper. The data can be obtained free of charge from The Cambridge Crystallographic Data Centre via https://www.ccdc.cam.ac.uk/structures.
+
+<--- Page Split --->
+
+Acknowledgements This work was supported by the Spanish Agencia Estatal de Investigación (AEI) (PID2019- 111126RB- 100 and PID2022- 142440NB- 100), the Xunta de Galicia (ED431C 2021/21 and Centro singular de Investigación de Galicia accreditation 2019- 2022, ED431G 2019/03), and the European Union (European Regional Development Fund - ERDF). We also thank the ORFEO- CINCA network and Mineco (RED2022- 134287- T). V.L.- C. thanks the Xunta de Galicia for her research contract (ED481A- 2019/117). All calculations were carried out at the Centro de Supercomputación de Galicia (CESGA).  
+
+Correspondence and requests for materials should be addressed to Juan R. Granja (juanr.granja@usc.es).
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 1. Previously reported supramolecular containers based on \(\alpha ,\gamma\) -cyclic peptides. Top: cyclic octapeptide (blue) topped with a porphyrin moiety (green) used in the recognition of \(4,4^{\prime}\) -bipyridines. Center: smaller alternatives derived from dimer-forming \(N\) -propargylated cyclic hexapeptides (CP1, grey) through Sonogashira cross-couplings with Iodopyridines or copper-catalyzed azide-alkyne cycloaddition (CuAAC) that have been used as ion transporters.[27] Bottom: cartoon model of supramolecular capsule derived from CP1 and a tris-azide derivative described in this work. </center>
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 2. Synthetic strategy used for the preparation of capsules D2 and D4 and initially proposed encapsulation model for the recognition of anions (2X<D2 and 2X<D4). </center>
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Figure 3. Side (a and c) and top (b and d) views of the crystal structures of dimeric supramolecular capsules D2 (top) and D4 (bottom), respectively. The molecules of acetonitrile entrapped in the cavity are represented in CPK models. The nitrogen of nitrile groups is pointing towards one of the caps close to the triazole protons with shorter distance in the Ach-based capsule (2.66 Å, bottom) than in the Acp derivative (2.79-2.68 Å, top). For clarity only triazole and amide protons are shown. The yellow dashed lines highlight the hydrogen bonds. </center>
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Figure 4. NMR spectra of pure CP2 (bottom) and after the addition of different equivalents of fluoride (TBAF). In blue colour are highlighted the signals corresponding to the new species formed after the addition of the fluoride, light blue denotes the signals corresponding to the tetrabutylammonium counterion. </center>
+
+<--- Page Split --->
+![](images/Figure_6.jpg)
+
+<center>Figure 5. Experiments of heterodimer (D2-3 and D2-4) formation followed by anion recognition, top view. Bottom a) NMR spectra corresponding to these studies in which the characteristic signals of each component are highlighted with specific colours; orange, green and lavender for homodimers D2, D3 and D4, respectively, dark blue for CP2 interacting with fluoride, and plum and teal green for heterodimers D2-3 and D2-4, respectively. </center>
+
+<--- Page Split --->
+![](images/Figure_7.jpg)
+
+<center>Figure 6. Side (a) and top (b) view of the crystal structure of water-fluoride cluster entrapped in supramolecular capsule D2 (3F.8H₂O<2CP2) as twin complexes. In these complexes three fluoride ions (light green) and eight water molecules are hydrogen-bonded to the amide protons of two cyclic peptides at different planes (the hydrogen bond network (yellow dashed lines) in the cluster is only shown for one of the complexes). Side (c) and top (d) view of the crystal structure of encapsulated chloride-water cluster between two CP2 (3Cl.4H₂O<2CP2). The three chloride ions and water molecules are occupying six equivalent chemical (and crystallographic) positions forming a hexagonal structure, where the two subunits are aligned forming a trigonal bipyramid shaped (d) capsule. </center>
+
+<--- Page Split --->
+![PLACEHOLDER_36_0]
+
+<center>Figure 7. Chloride transport experiments using capsule D2 in liposomes containing lucigenin (A) or HPTS (B-D). </center>
+
+<--- Page Split --->
+
+
+<table><tr><td>Dimer</td><td>Anion</td><td>Equiv[α]</td><td>Molar Fraction 3A·nH2Oc2CP2[β]</td><td>NH shift (ppm)</td><td>Htriazole shift (ppm)</td><td>pka</td></tr><tr><td>D2</td><td>F-</td><td>3.13</td><td>1</td><td>10</td><td>7.38</td><td>3.2</td></tr><tr><td>D2</td><td>Cl-</td><td>24.7</td><td>0.90</td><td>8.62</td><td>7.29</td><td>-8.0</td></tr><tr><td>D2</td><td>Br[α]</td><td>28.1</td><td>0.92</td><td>8.35</td><td>7.28</td><td>-9.0</td></tr><tr><td>D2</td><td>I[α]</td><td>86</td><td>0.71</td><td>8.07</td><td>7.27</td><td>-</td></tr><tr><td>D2</td><td>N3[α]</td><td>35</td><td>0.67</td><td>8.55</td><td>7.29</td><td>4.75</td></tr><tr><td>D2</td><td>OAc-</td><td>66</td><td>0.88</td><td>8.97</td><td>7.27</td><td>4.7</td></tr><tr><td>D2</td><td>NO3[α]</td><td>60</td><td>0.77</td><td>8.19</td><td>7.30</td><td>-1.4</td></tr><tr><td>D4</td><td>F-</td><td>8</td><td>0</td><td>-</td><td>-</td><td>3.2</td></tr><tr><td>D4</td><td>Cl-</td><td>36</td><td>0</td><td>-</td><td>-</td><td>-8.0</td></tr></table>  
+
+Table 1. Key features of the anion encapsulation by tris- triazolyl modified CPs (D2 and D4); On the right: illustrative anion binding experiments derived by titrations carried out by NMR experiments. The dashed lines are used to indicate the anions in which extra water was added to facilitate complex formation. [a] equivalent number are given with respect to CP2 concentration, [b] Molar fraction was calculated at the mentioned maximum number of equivalents of the corresponding anions, [c] 7 \(\mu \mathrm{L}\) of water were added before the additions of anion solution.  
+
+[1] Davis, J. T.; Gale, P. A. & Quesada, R. Advances in anion transport and supramolecular medicinal chemistry. Chem. Soc. Rev. 49, 6056- 6086 (2020).  
+
+[2] Rowe, S. M.; Miller, S. & Sorscher, E. J. Cystic fibrosis. N. Engl. J. Med. 352, 1992- 2001 (2005).  
+
+[3] McNaughton, D. A.; Fares, M.; Picci, G.; Gale & P. A. Caltagirone, C. Advances in fluorescent and colorimetric sensors for anionic species. Coord. Chem. Rev. 427, 213573 (2021).  
+
+[4] Chen, L.; Berry, S. N.; Wu, X.; Howe, E. N. W. & Gale, P. A. Advances in anion receptor chemistry. Chem 6, 61- 141 (2020).  
+
+[5] Molina, P.; Zapata, F. & Caballero, A. Anion recognition strategies based on combined noncovalent interactions. Chem. Rev. 117, 9907- 9972 (2017).
+
+<--- Page Split --->
+
+[6] Liu, Y.; Sengupta, A.; Raghavachari, K. & Flood, A. H. Anion binding in solution: Beyond the electrostatic regime. Chem 3, 411- 427 (2017).  
+
+[7] Stockbridge, R.; Kolmakova- Partensky, L.; Shane, T.; Koide, A.; Koide, S.; Miller, C. & Newstead, S. Crystal structures of a double- barrelled fluoride ion channel. Nature 525, 548- 551 (2015).  
+
+[8] Wu, X.; Wang, P.; Turner, P.; Lewis, W.; Catal, O.; Thomas, D. S. & Gale, P. A. Tetraurea macrocycles: Aggregation- driven binding of chloride in aqueous solutions. Chem 5, 1210- 1222 (2019).  
+
+[9] Gomez, D. T.; Pratt, L. R.; Asthagiri, D. N. & Rempe, S. B. Hydrated anions: From clusters to bulk solution with quasi- chemical theory. Acc. Chem. Res. 55, 2201- 2212 (2022).  
+
+[10] Balamurugan, K. & Pisabarro, M. T. Stabilizing Role of Water Solvation on Anion- π Interactions in Proteins. ACS Omega 6, 25350- 25360 (2021).  
+
+[11] Varma, S.; Sabo, D. & Rempe, S. B. K+/Na+ Selectivity in K Channels and Valinomycin: Over coordination Versus Cavity- size constraints. J. Mol. Biol. 376, 13- 22 (2008).  
+
+[12] Yao, W.; Wang, K.; Wu, A.; Reed, W. F. & Gibb, B. C. Anion binding to ubiquitin and its relevance to the Hofmeister effects. Chem. Sci. 12, 320- 330 (2021).  
+
+[13] Castleman, A. W. & Keesee, R. G. Ionic clusters. Chem, Rev. 86, 589- 618 (1986).  
+
+[14] Jungwirth, P. & Cremer, P. Beyond Hofmeister. Nat. Chem. 6, 261- 263 (2014).  
+
+[15] Marcus, Y. Effect of ions on the structure of water: Structure making and breaking. Chem. Rev. 109, 1346- 1370 (2009).  
+
+[16] Cremer, P. S.; Flood, A. H.; Gibb, B. C. & Mobley, D. L. Collaborative routes to clarifying the murky waters of aqueous supramolecular chemistry. Nat. Chem. 10, 8- 16 (2018).  
+
+[17] He, Q.; Tu, P. & Sessler, J. L. Supramolecular chemistry of anionic dimers, trimers, tetramers, and clusters. Chem 4, 46- 93 (2018).
+
+<--- Page Split --->
+
+[18] Harris, A.; Saita, M.; Resler, T.; Hughes-Visentin, A.; Maia, R.; Pranga- Sellnau, F.: Bondar, A.- N.; Heberle, J. & Brown, L. S. Molecular details of the unique mechanism of chloride transport by a cyanobacterial rhodopsin. Phys. Chem. Chem. Phys. 20, 3184- 3199 (2018).  
+
+[19] Sokkalingam, P.; Shraberg, J.; Rick, S. W. & Gibb, B. C. Binding hydrated anions with hydrophobic pockets. J. Am. Chem. Soc. 138, 48- 51 (2016).  
+
+[20] Chakraborty, S.; Dutta, R.; Wong, B. M. & Ghosh, P. Anion directed conformational diversities of an arene based hexa- amide receptor and recognition of the [F4(H2O)6]4+ cluster. RSC Adv. 4, 62689- 62693 (2014).  
+
+[21] Dariush, A. Self- assembled molecular capsule. In Synergy in Supramolecular Chemistry; Tatsuya, N., Ed.; CRC Press: Boca Raton, FL, 2015; pp 133- 148.  
+
+[22] Solomonov, A. & Shimanovich, U. Self- assembly in protein- based bionanomaterials. Israel J. Chem. 60, 1152- 1170 (2020).  
+
+[23] Saha, R.; Mondal, B. & Mukherjee, P. S. Molecular cavity for catalysis and formation of metal nanoparticles for use in catalysis. Chem. Rev. 122, 12244- 12307 (2022).  
+
+[24] McTernan, C. T.; Davies, J. A. & Nitschke, J. R. Beyond platonic: How to build metal- organic polyhedra capable of binding low- symmetry, information- rich molecular cargoes. Chem. Rev. 122, 10393- 10437 (2022).  
+
+[25] Bowman- James, K. Supramolecular cages trap pesky anions. Science 365, 124- 125 (2019).  
+
+[26] Rizzuto, F. J.; von Krbek, L. K. S. & Nitschke, J. R. Strategies for binding multiple guests in metal- organic cages. Nat. Rev. Chem. 3, 204- 222 (2019).  
+
+[27] Rodríguez- Vázquez, N.; Amorín, M. & Granja, J. R. Recent advances in controlling the internal and external properties of self- assembling cyclic peptide nanotubes and dimers. Org. Biomol. Chem. 15, 4490- 4505 (2017).
+
+<--- Page Split --->
+
+[28] Song, Q.; Cheng, Z.; Kariuki, M.; Hall, S.C.L.; Hill, S.K.; Rho, J.Y. & Perrier, S. Molecular self- assembly and supramolecular chemistry of cyclic peptides. Chem. Rev. 121, 13936- 13995 (2021).  
+
+[29] Ozores, H. L.; Amorin, M. & Granja, J. R. Self- assembling molecular capsules based on \(\alpha ,\gamma\) - cyclic peptides. J. Am. Chem. Soc. 139, 776- 784 (2017).  
+
+[30] Pizzi, A.; Ozores, L. H.; Calvelo, M.; García- Fandiño, R.; Amorin, M.; Demitri, N.; Terraneo, G.; Bracco, S.; Comotti, A.; Sozzani, P.; Bezuidenhout, C. X.; Metrangolo, P. & Granja, J. R. Tight Xenon confinement in a crystalline sandwich- like hydrogen- bonded dimeric capsule of a cyclic peptide. Angew. Chem. Int. Ed. 58, 14472- 14476 (2019).  
+
+[31] Fuertes, A.; Amorin, M. & Granja, J. R. Versatile symport transporters based on cyclic peptide dimers. Chem. Commun. 56, 46- 49 (2020).  
+
+[32] Fuertes, A.; Amorin, M. & Granja, J. R. Membrane spanners based on a dimeric \(\alpha ,\beta\) - cyclic peptide core. Supramol. Chem. 32, 239- 246 (2020).  
+
+[33] Lee, C.- H.; Lee, S.; Yoon, H. & Jang, W.- D. Strong binding affinity of a Zinc- porphyrin- based receptor for halides through the cooperative effects of quadruple C- H hydrogen bonds and axial ligation. Chem. Eur. J. 17, 13898- 13903 (2011).  
+
+[34] Parks, F.C.; Liu, Y.; Debnath, S.; Stutsman, S. R.; Raghavachari, K. & Flood, A. H. Allosteric control of photofoldamers for selecting between anion regulation and double- to- single helix switching. J. Am. Chem. Soc. 140, 17711- 17723 (2018).  
+
+[35] Kim, H.; Hong, K.- I.; Lee, J. H.; Kang, P.; Choi, M.- G. & Jang, W.- D. Triazole- bearing calixpyrroles: strong halide binding affinities through multiple N- H and C- H hydrogen bonds. Chem. Commun. 54, 10863- 10865 (2018).  
+
+[36] Liu, Y.; Zhao, W.; Chen, C.- H.& Flood, A. H. Chloride capture using a C- H hydrogen- bonding cage. Science 365, 159- 161 (2019).
+
+<--- Page Split --->
+
+[37] Liu, Y.; Parks, F. C.; Sheetz, E. G.; Chen, C.- H. & Flood, A. H. Polarity-tolerant chloride binding in foldamer capsules by programmed solvent-exclusion. J. Am. Chem. Soc. 143, 3191–3204 (2021).  
+
+[38] Mondal, D.; Ahmad, M.; Panwaria, P.; Upadhyay, A. & Talukdar, P. Anion recognition through multivalent C–H hydrogen bonds: Anion-induced foldamer formation and transport across phospholipid membranes. J. Org. Chem. 87, 10–17 (2022).  
+
+[39] Fuertes, A.; Ozores, H. L.; Amorín, M. & Granja, J. R. Self-assembling Venturi-like peptide nanotubes. Nanoscale 9, 748–753 (2017).  
+
+[40] Bock, V. D.; Perciaccante, R.; Jansen, T. P.; Hiemstra, H. & van Maarseveen, J. H. Click Chemistry as a Route to Cyclic Tetrapeptide Analogues: Synthesis of cyclo- [Pro- Val- \(\psi\) (triazole)- Pro- Tyr]. Org. Lett. 8, 919–922 (2006).  
+
+[41] Amorín, M.; Castedo, L. & Granja, J. R. New cyclic peptide assemblies with hydrophobic cavities: The structural and thermodynamic basis of a new class of peptide nanotubes. J. Am. Chem. Soc. 125, 2844–2845 (2003).  
+
+[42] Brea, R. J.; Amorín, M.; Castedo, L. & Granja, J. R. Methyl- blocked dimeric \(\alpha ,\gamma\) - peptide nanotube segments: Formation of a peptide heterodimer through backbone- backbone interactions. Angew. Chem. Int. Ed. 44, 5710–5713 (2005).  
+
+[43] Haris, P. I. & Chapman, D. The conformational analysis of peptides using Fourier Transform IR spectroscopy. Biopolymers (Pept. Sci.) 37, 251–263 (1995).  
+
+[44] Krimm, S. & Bandekar, J. in Advances in Protein Chemistry, ed. C. B. Anfinsen, J. T. Edsall and F. M. Richards, Academic Press, Orlando, FL, 1986, pp. 181–364.  
+
+[45] Barth, A. & Zscherp, C. What vibrations tell us about proteins. Q. Rev. Biophys. 35, 369–430 (2002).  
+
+[46] In- Hou, C.; Hong, Z. & Dearden V, David. Macrocyclic chemistry in the gas phase: intrinsic cation affinities and complexation rates for alkali metal cation complexes of crown ethers and glymes. J. Am. Chem. Soc. 115, 5736–5744 (1993).
+
+<--- Page Split --->
+
+[47] Buchanan, W. G.; Gerzain, M. & Bensimon, C. 1,4,7,10,13- Pentaoxacyclopentadecane (15- crown- 5) sodium iodide complex, \(\mathrm{C_{10}H_{20}O_5\cdot NaI}\) . Acta. Cryst. 50, 1016- 1019 (1994).  
+
+[48] Brea, R. J.; Pérez- Alvite, M. J.; Panciera, M. Mosquera, M.; Castedo, L. & Granja, J. R. Highly efficient and directional homo- and heterodimeric energy transfer materials based on fluorescently derivatized \(\alpha ,\gamma\) - cyclic octapeptides. Chem. Asian J. 6, 110- 121 (2011).  
+
+[49] Aragay, G.; Ventura, B.; Guerra, A.; Pintre, I.; Chiorboli, C.; García- Fandiño, R.; Flamigni, L.; Granja, J. R. & Ballester, P. Self- sorting of cyclic peptide homodimers into a heterodimeric assembly featuring an efficient photoinduced intramolecular electron- transfer process. Chem. Eur. J. 20, 3427- 3438 (2014).  
+
+[50] García- Fandiño, R.; Castedo, L.; Granja, J. R. & Vázquez, S. A. Interaction and dimerization energies in methyl- blocked \(\alpha ,\gamma\) - peptide nanotube segments. J. Phys. Chem. B 114, 4973- 4983 (2010).  
+
+[51] Dutta, R. & Ghosh, P. Recent developments in anion induced capsular self- assemblies. Chem. Commun. 50, 10538- 10554 (2014).  
+
+[52] Wu, X. & Gale, P. A. Measuring anion transport selectivity: a cautionary tale. Chem. Commun. 57, 3979- 3982 (2021).  
+
+[53] Marcus, Y. Thermodynamics of solvation of ions. Part 5. Gibbs free energy of hydration at 298.15 K. J. Chem. Soc., Faraday Trans. 87, 2995- 2999 (1991).  
+
+[54] Tong, C. C.; Quesada, R.; Sessler, J. L. & Gale, P. A. meso- Octamethylcalix[4]pyrrole: an old yet new transmembrane ion- pair transporter. Chem. Commun. 6321- 6323 (2008).  
+
+[55] Kano, K. & Fendler, J. H. Pyranine as a sensitive pH probe for liposome interiors and surfaces. pH gradients across phospholipid vesicles. Biochim. Biophys. Acta 509, 289- 299 (1978).  
+
+[56] Wu, X.; Small, J. R.; Cataldo, A.; Withecombe, A. M.; Turner P. & Gale, P. A. Voltage- switchable HCl transport enabled by lipid headgroup- transporter interactions. Angew. Chem. Int. Ed. 58, 15142- 15147 (2019).
+
+<--- Page Split --->
+
+[57] Madhavan, N.; Robert, E. C. & Gin, M. S. A highly active anion-selective aminocyclodextrin ion channel. Angew. Chem. Int. Ed. 44, 7584–7587 (2005).  
+
+[58] Kang, B.; Tang, H.; Zhao, Z. & Song, S. Hofmeister series: Insights of ion specificity from amphiphilic assembly and interface property. ACS Omega 5, 6229–6239 (2020).  
+
+[59] Jowett, L. A.; Howe, E. N. W.; Soto-Cerrato, V.; Van Rossom, W.; Pérez-Tomás, R. & Gale, P. A. Indole-based perenosins as highly potent HCl transporters and potential anti-cancer agents. Sci. Rep. 7, 9397 (2017).  
+
+[60] Gilchrist, A. M. et al. Supramolecular methods: the 8- hydroxypyrene- 1,3,6- trisulfonic acid (HPTS) transport assay. Supramol. Chem. 33, 325–344 (2021).
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+Reviewer #1 was asked to look over the response given to Reviewer #2]  
+
+I feel the authors have successfully responded to the reviewers' comments and the manuscript is ready for publication.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__72029e53925d7ad4b9d13bca370bd70820e0aa98858b14f70cb0c95ec2ff75c6/supplementary_0_Peer Review File__72029e53925d7ad4b9d13bca370bd70820e0aa98858b14f70cb0c95ec2ff75c6_det.mmd

@@ -0,0 +1,628 @@
+<|ref|>title<|/ref|><|det|>[[61, 41, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[69, 111, 361, 140]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[70, 162, 806, 218]]<|/det|>
+Recognition of anion- water clusters by peptide- based supramolecular capsules  
+
+<|ref|>image<|/ref|><|det|>[[57, 732, 240, 782]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[250, 732, 911, 785]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 91, 300, 106]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 147, 402, 163]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 175, 880, 284]]<|/det|>
+This manuscript describes the synthesis and characterization of a supramolecular dimer which can bind anions. The manuscript is well- written and the methods use to characterize the compounds (including X- ray diffraction, NMR, transmembrane transport experiments) are fairly thorough. The results showing how the macromolecules bind the anions, and the solvating water molecules is interesting and relevant to the larger topic of selective binding of ions. I would recommend this paper for publication without revision.  
+
+<|ref|>text<|/ref|><|det|>[[116, 323, 402, 339]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 350, 878, 551]]<|/det|>
+Granja and coworkers demonstrated supramolecular dimerization of a capped cup- shaped cyclic peptide and its ability to recognize different anions in the form of hydrated clusters. The formation of supramolecular dimers or anion binding and transport have been explored by several groups using different systems. What is interesting about the present study is the importance of water in the binding. Although the anion binding is studied in Dichloromethane and Acetonitrile mixture, a trace amount of water present in the solvents/chemicals is essential for binding. The removal of water traces by adding molecular sieves to the host- guest complex resulted in the removal of anion from the guest. The effect is reversible wherein the addition of water again leads to the binding. In supramolecular chemistry, the role of solvents in a study has high significance on the behavior of a molecule/material yet rarely their role is completely understood. Hence the present study focussing on the binding of anion- water clusters is significant.  
+
+<|ref|>text<|/ref|><|det|>[[115, 590, 870, 625]]<|/det|>
+The study is experimentally rich with clean characterization and analysis, but the presentation can be improved.  
+
+<|ref|>text<|/ref|><|det|>[[115, 665, 872, 700]]<|/det|>
+Please provide integrations in the supporting figures for the 1H NMR spectra of the supramolecular dimers and the complexes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 740, 861, 775]]<|/det|>
+There are multiple typos throughout the manuscript and SI for scientific and non- scientific terms (for example, RMN for NMR), please correct.  
+
+<|ref|>text<|/ref|><|det|>[[115, 816, 866, 888]]<|/det|>
+The authors demonstrate that the CP1- CP4 exists as the corresponding supramolecular dimeric forms D1- D4 in solution. In the discussion/figures, authors represented the dimers using different notations like CP, 2CP, and D leading to confusion and difficulty in understanding the content (for example, refer to Figure 5 and the corresponding discussion).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 117, 883, 300]]<|/det|>
+The authors conclude that the recognition process most likely involves dimeric capsules and trimeric anions. It was also mentioned that experimental evidence regarding this is provided in the case of halides F- and Cl using crystal structures and mass but not in the case of other anions (Page 13 of the manuscript). It was unclear how the authors concluded the stoichiometry of binding for the anions other than F- and Cl- without mass, crystal structure, or ITC. There is a significant increase in the size of the anions N3- , OAc- , and NO3- compared to halides. The cavity size does not seem to be able to fit three big anions. If there is no clear confirmation, I suggest mentioning the uncertainty about stoichiometry rather than generalizing that all anions could be trimers. In that case, it is suggestible to modify the title also to "anion- water clusters" instead of "trimeric anionic- water clusters"  
+
+<|ref|>text<|/ref|><|det|>[[115, 339, 627, 356]]<|/det|>
+Please provide a brief explanation of the results obtained from DFT  
+
+<|ref|>text<|/ref|><|det|>[[115, 395, 860, 431]]<|/det|>
+TOC can be improved; the current TOC does not represent the anion- water clusters clearly in the supramolecular dimer.  
+
+<|ref|>text<|/ref|><|det|>[[116, 499, 402, 515]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 527, 878, 655]]<|/det|>
+This paper reports the synthesis of a cyclic peptide which is furnished with a cap to produce a hemispherical unit which dimerises through hydrogen bonding interactions. The two halves of the sphere are shown to open up to encapsulate fluoride and chloride in solution and in the solid state. In the crystal structure, there is clear evidence of water being involves in the formation of a specific cluster. In solution, water is required, but the structural nature of the cluster is not clearly demonstrated experimentally. Membrane transport is demonstrated, and explored using a conventional set of experiments.  
+
+<|ref|>text<|/ref|><|det|>[[114, 694, 867, 804]]<|/det|>
+My view is that this paper contains some very nice supramolecular chemistry work, and that the importance of water in the process is a highlight, though not as important as the introduction makes out. The experiments and analysis are all ably conducted and meet all the requirements of the field. It's all very well done, and will be likely used as an example of hydrated anions being encapsulated (though plenty of crystal structures exist of anion complex hydrates already) rather than built upon directly by others.  
+
+<|ref|>text<|/ref|><|det|>[[115, 842, 871, 897]]<|/det|>
+The introduction contains some statements which need work - electrostatic interactions are a noncovalent interation (top of p2), proteins are anion- selective and usually hold anions in a largely desolvated state. It's too much to call a host- guest complex a 'new state of matter.'
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 860, 126]]<|/det|>
+There should be consideration of the possibility of deprotonation by fluoride. Solvent needs to be given clearly in the main text when talking about NMR titrations etc.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[130, 189, 902, 223]]<|/det|>
+The following pages provide point by point answers to each of the questions, comments or suggestions proposed by the different reviewers.  
+
+<|ref|>sub_title<|/ref|><|det|>[[131, 308, 231, 323]]<|/det|>
+## Reviewer: 1  
+
+<|ref|>text<|/ref|><|det|>[[130, 330, 903, 448]]<|/det|>
+- Comment. "This manuscript describes the synthesis and characterization of a supramolecular dimer which can bind anions. The manuscript is well-written, and the methods used to characterize the compounds (including X-ray diffraction, NMR, transmembrane transport experiments) are fairly thorough. The results showing how the macromolecules bind the anions, and the solvating water molecules is interesting and relevant to the larger topic of selective binding of ions. I would recommend this paper for publication without revision."  
+
+<|ref|>text<|/ref|><|det|>[[130, 462, 902, 496]]<|/det|>
+Response. We greatly appreciate the positive comments about our work, how it was carried out and the characterization of the compounds and their derivatives. Thank you.  
+
+<|ref|>sub_title<|/ref|><|det|>[[131, 528, 231, 543]]<|/det|>
+## Reviewer: 2  
+
+<|ref|>text<|/ref|><|det|>[[130, 550, 903, 768]]<|/det|>
+- Comment. "Granja and coworkers demonstrated supramolecular dimerization of a capped cup-shaped cyclic peptide and its ability to recognize different anions in the form of hydrated clusters. The formation of supramolecular dimers or anion binding and transport have been explored by several groups using different systems. What is interesting about the present study is the importance of water in the binding. Although the anion binding is studied in Dichloromethane and Acetonitrile mixture, a trace amount of water present in the solvents/chemicals is essential for binding. The removal of water traces by adding molecular sieves to the host-guest complex resulted in the removal of anion from the guest. The effect is reversible wherein the addition of water again leads to the binding. In supramolecular chemistry, the role of solvents in a study has high significance on the behavior of a molecule/material yet rarely their role is completely understood. Hence the present study focussing on the binding of anion-water clusters is significant."  
+
+<|ref|>text<|/ref|><|det|>[[130, 774, 902, 808]]<|/det|>
+The study is experimentally rich with clean characterization and analysis, but the presentation can be improved."  
+
+<|ref|>text<|/ref|><|det|>[[131, 821, 902, 888]]<|/det|>
+Response. We thank the referee for his/her interest and positive comments on our work and this manuscript. We are very pleased to hear his/her comments about the quality and significance of our work and the careful characterization and analysis of the results that we report here. Following his/her advice, the presentation was revised to improve it.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[130, 146, 904, 181]]<|/det|>
+- Comment. "Please provide integrations in the supporting figures for the \(^1 H\) NMR spectra of the supramolecular dimers and the complexes."  
+
+<|ref|>text<|/ref|><|det|>[[131, 194, 904, 244]]<|/det|>
+Response. We have included all the integrations in the \(^1 H\) NMR spectra of the supramolecular dimers and complexes in the supporting information, in the section 8. Thank you very much for your kind suggestion.  
+
+<|ref|>text<|/ref|><|det|>[[130, 250, 904, 285]]<|/det|>
+- Comment. "There are multiple typos throughout the manuscript and SI for scientific and non-scientific terms (for example, RMN for NMR), please correct."  
+
+<|ref|>text<|/ref|><|det|>[[131, 299, 904, 349]]<|/det|>
+Response. We are very sorry for the inconvenience related to the quality of the written document. We have edited both the manuscript and SI to correct all the typographical errors in these documents. Non-scientific terms such as RMN were also corrected.  
+
+<|ref|>text<|/ref|><|det|>[[131, 355, 904, 439]]<|/det|>
+- Comment. "The authors demonstrate that the CP1-CP4 exists as the corresponding supramolecular dimeric forms D1-D4 in solution. In the discussion/figures, authors represented the dimers using different notations like CP, 2CP, and D leading to confusion and difficulty in understanding the content (for example, refer to Figure 5 and the corresponding discussion)."  
+
+<|ref|>text<|/ref|><|det|>[[130, 454, 904, 753]]<|/det|>
+Response. Again, we are very sorry for these errors, which cause some confusion and may make difficult to understand the content of the work. In general, we have used the term CP for the abbreviation of Cyclic Peptide generically, while CP1- CP4 are used to describe the molecular structure, in which all atoms are covalently linked, corresponding to the four cyclic peptides used in this work. The terms D1- D4 for homodimers or the corresponding heterodimers (D2- 3 or D2- 4) are intended to emphasize the supramolecular species that are present predominantly in the organic solvents used in this work (dichloromethane or acetonitrile in dichloromethane). Therefore, we preferably use Dn to describe the supramolecular species that are mainly present in solution experiments or in the crystals. The same can be said for heterodimeric species, in which two different cyclic peptides are mixed and assembled in solution forming hydrogen bonds between both molecules. Therefore, D (dimer) is used to describe the supramolecular species in which both peptide units are linked by hydrogen bonds. To clarify this possible ambiguity, we have modified the nomenclature for complexes with anions, in which the two cyclic peptides are not directly linked to each other by hydrogen bonds, using the new numeration, mA·nH₂O=2CP2, for the complex with the anion cluster in which "m" represents the number of anions entrapped between both CPs (most likely three), "A" indicates the type of anion (F-, Cl-, Br-, I-, AcO-, N₃⁻ or NO₃⁻) and "n" represents the number of water molecules that form the entrapped anion clusters.  
+
+<|ref|>text<|/ref|><|det|>[[130, 760, 904, 894]]<|/det|>
+- Comment. "The authors conclude that the recognition process most likely involves dimeric capsules and trimeric anions. It was also mentioned that experimental evidence regarding this is provided in the case of halides F- and Cl using crystal structures and mass but not in the case of other anions (Page 13 of the manuscript). It was unclear how the authors concluded the stoichiometry of binding for the anions other than F- and Cl- without mass, crystal structure, or ITC. There is a significant increase in the size of the anions N3-, OAc-, and NO3- compared to halides. The cavity size does not seem to be able to fit three big anions. If there is no clear confirmation, I suggest mentioning the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[131, 146, 904, 196]]<|/det|>
+uncertainty about stoichiometry rather than generalizing that all anions could be trimers. In that case, it is suggestive to modify the title also to "anion- water clusters" instead of "trimeric anionic- water clusters".  
+
+<|ref|>text<|/ref|><|det|>[[130, 211, 904, 444]]<|/det|>
+Response. We appreciate very much this comment regarding stoichiometry with other anions different to fluoride and chloride. Unfortunately, despite having attempted several experimental techniques, including different crystallization conditions, we were unable to unequivocally confirm the proposed stoichiometry. In any case, we consider that the invariability of the C3 symmetry of the cyclic peptide, as reflected in the NMR spectra, upon the addition of the different anions support our stoichiometry hypothesis. Another clear evidence found, also derived from the NMR studies, comes from the fact that the cyclic peptide adopts, after the additions, very similar conformations to that observed for the complexes with chloride and fluoride ions. The main difference between them is the chemical shift of amide proton (NH), whose down- field shift is correlated with the strength of the hydrogen bonds with the different anions and not with other conformational changes. In any case, following his/her advice we have modified the title of our work and also the abbreviation for each complex \((\mathbf{mA}\cdot \mathbf{nH}_2\mathbf{O}\subset 2\mathbf{CP}_2)\) , in which "m" represents the number of anions (A) that form the entrapped clusters.  
+
+<|ref|>text<|/ref|><|det|>[[130, 450, 837, 468]]<|/det|>
+- Comment. "Please provide a brief explanation of the results obtained from DFT."  
+
+<|ref|>text<|/ref|><|det|>[[130, 482, 904, 599]]<|/det|>
+Response. We thank this referee for asking about the DFT studies, which mainly confirm the stability of the proposed dimers and give some additional structural insights, hydrogen bonds distances, angles, and so on, to those species that have not crystallized. The truth is that we decided not to go on describing these results, which mainly confirmed what was deciphered with other experiments, so as not to make this article too long. Following the referee demand, we have decided to include an additional section in the supporting information (Supplementary Discussion III) in which we briefly describe the most relevant DFT results.  
+
+<|ref|>text<|/ref|><|det|>[[130, 605, 904, 639]]<|/det|>
+- Comment. "TOC can be improved; the current TOC does not represent the anion-water clusters clearly in the supramolecular dimer."  
+
+<|ref|>text<|/ref|><|det|>[[130, 653, 904, 686]]<|/det|>
+Response. We thank the reviewer for his kind suggestion. Therefore, a new TOC figure is attached to better represent the anion- water clusters in the supramolecular dimer.  
+
+<|ref|>sub_title<|/ref|><|det|>[[131, 719, 231, 733]]<|/det|>
+## Reviewer: 3  
+
+<|ref|>text<|/ref|><|det|>[[130, 742, 904, 858]]<|/det|>
+- Comment. "This paper reports the synthesis of a cyclic peptide which is furnished with a cap to produce a hemispherical unit which dimerises through hydrogen bonding interactions. The two halves of the sphere are shown to open up to encapsulate fluoride and chloride in solution and in the solid state. In the crystal structure, there is clear evidence of water being involves in the formation of a specific cluster. In solution, water is required, but the structural nature of the cluster is not clearly demonstrated experimentally. Membrane transport is demonstrated, and explored using a conventional set of experiments.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[131, 146, 904, 247]]<|/det|>
+My view is that this paper contains some very nice supramolecular chemistry work, and that the importance of water in the process is a highlight, though not as important as the introduction makes out. The experiments and analysis are all ably conducted and meet all the requirements of the field. It's all very well done, and will be likely used as an example of hydrated anions being encapsulated (though plenty of crystal structures exist of anion complex hydrates already) rather than built upon directly by others."  
+
+<|ref|>text<|/ref|><|det|>[[131, 260, 904, 378]]<|/det|>
+Response. We thank the referee for the positive review of our supramolecular chemistry work, highlighting the important role played by water molecules in the recognition properties of capped \(\alpha ,\gamma\) - cyclic peptides. We are glad to find his interest and confirmation about the analysis and experimental planification and development to clearly stablish the encapsulation of hydrated clusters. It is true that there are a number of crystal structures having anion complexes hydrated, although only a few of them describe polyanionic cluster of the type here reported.  
+
+<|ref|>text<|/ref|><|det|>[[131, 384, 904, 450]]<|/det|>
+- Comment. "The introduction contains some statements which need work - electrostatic interactions are a noncovalent interaction (top of p2), proteins are anion-selective and usually hold anions in a largely desolvated state. It's too much to call a host-guest complex a 'new state of matter."  
+
+<|ref|>text<|/ref|><|det|>[[131, 465, 904, 581]]<|/det|>
+Response. We very much regret the lack of precision in the wording of some concepts mentioned in the introduction. We have modified the introduction to include the "directional" noncovalent interactions to differentiate to those pure electrostatic between cations and anions. Now we mention: "The design principles used in this development are generally dominated by directional noncovalent interactions, such as halogen, chalcogen or hydrogens bonds, among others, rather than by purely electrostatic ones, as they generally provide better selectivity."  
+
+<|ref|>text<|/ref|><|det|>[[131, 588, 904, 705]]<|/det|>
+It is true that the majority of anion- selective proteins recognize them in their desolvated state, however some authors have proposed that the selective transport of fluoride carried out by the FLUC protein, which presents a remarkable selectivity against chloride (Gomez, D. T.; Pratt, L. R.; Asthagiri, D. N. Rempe, S. B. Acc. Chem. Res. 2022, 55, 2201- 2212) or the recognition of other anions such as sulfate (K. Balamurugan, M. T. Pisabarro, ACS Omega 2021, 6, 25350- 25360) could be related to the recognition of their hydrates in a similar way to the selective transport of sodium (versus potassium) in the sodium channel proteins.  
+
+<|ref|>text<|/ref|><|det|>[[131, 712, 815, 728]]<|/det|>
+We have modified the mentioned sentence at the introduction in the following way:  
+
+<|ref|>text<|/ref|><|det|>[[131, 735, 904, 836]]<|/det|>
+"In some cases, the ion selectivity in aqueous media does not depend as much on the relative size or charge of the ions, but on the size of their corresponding hydrates.[9,10] In fact, some authors have proposed that the selectivity in fluoride transport by FLUC protein could be explained in a similar way to the \(\mathrm{K^{+} / Na^{+}}\) selectivity of sodium ion channels.[11] Additionally, the greater solubility of negatively- charged proteins may be related to stronger hydration free energies for negatively charged groups on protein surfaces than their cationic counterparts.[12]"  
+
+<|ref|>text<|/ref|><|det|>[[131, 843, 904, 894]]<|/det|>
+Regarding the term "new state of matter", we would like to make some clarifications to this concept. From our point of view, molecular capsules are much more complex molecular containers than the "classic" molecular receptors characteristic of host/guest chemistry, since
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[131, 147, 904, 323]]<|/det|>
+they can recognize more than one substrate, they can incorporate a variable number of solvent molecules, they have a dynamic behavior in which the entrapped molecules flow at different rates among others. With that phrase we intended to emphasize that molecular capsules, such as those reported by Fujita and other authors, generate closed spaces that are more voluminous than conventional molecular receptors, can incorporate ad hoc reactive elements, trap macromolecules or proteins, maintaining or modifying their reactivity in which the substrates pass through their molecular walls, so they can be filtered or sorted on demand among other unique properties. In any case, since it seems that we have not depicted clearly enough this concept, this sentence has been modified to indicate the following: "They can be considered to create a new kind of molecular environment in which trapped molecule(s) exist."  
+
+<|ref|>text<|/ref|><|det|>[[131, 348, 901, 384]]<|/det|>
+- Comment. "There should be consideration of the possibility of deprotonation by fluoride. Solvent needs to be given clearly in the main text when talking about NMR titrations etc."  
+
+<|ref|>text<|/ref|><|det|>[[131, 390, 904, 600]]<|/det|>
+Response. We appreciate his advice and concerns very much. Deprotonation of amide groups was something we were very worried about from the beginning of this experiment, and this was one of the reasons why we decided to carry out all the experiments in dichloromethane and not in chloroform, in which, in addition to its well- known decomposition forming HCl, the deprotonation of chloroform with fluoride generates DF (deuterium fluoride) that exchange with amide NH and consequently the integration values of this proton signal decreases with time. This is something we have never seen in our studies. We are very confident that deprotonation with fluoride is not taking place in these experiments, as confirmed by the x- ray crystallographic studies. We believe that the recognition of the anion clusters reduces the intrinsic basicity of this anion. Since we have now included the \(^1\mathrm{H}\) NMR integrals, this can be checked in the proton spectra corresponding to \(3\mathrm{F}\cdot 8\mathrm{H}_2\mathrm{O}\subset 2\mathrm{CP}_2\) , in the Supplementary Information, section 8.  
+
+<|ref|>text<|/ref|><|det|>[[131, 620, 904, 708]]<|/det|>
+Solvent conditions were added in the manuscript following the referee advise. In general, deuterochloroform was used for the NMR characterization of dimers (homo and heterodimers) and titration experiments with bis- nitrile compounds (malononitrile and succinonitrile). For those experiments in which anion were added, a mixture of acetonitrile and dichloromethane (10%) was used.  
+
+<|ref|>text<|/ref|><|det|>[[131, 806, 541, 823]]<|/det|>
+Highlighted manuscript is attached to this letter.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[140, 106, 752, 158]]<|/det|>
+## Recognition of anion-water clusters by peptide-based supramolecular capsules  
+
+<|ref|>text<|/ref|><|det|>[[140, 179, 780, 228]]<|/det|>
+Victoria López- Corbalán, \(^†\) Alberto Fuertes, \(^†\) Antonio L. Llamas- Saiz, \(^‡\) Manuel Amorín, \(^†\) & Juan R. Granja \(^†*\)  
+
+<|ref|>text<|/ref|><|det|>[[139, 252, 838, 388]]<|/det|>
+\(^†\) Centro Singular de Investigación en Química Biolóxica e Materiais Moleculares (CIQUS) and Organic Chemistry Department, University of Santiago de Compostela, and \(^‡\) Unidad de Rayos X; Área de infraestructuras de Investigación, RIAIDT Edificio CACTUS, University of Santiago de Compostela, 15782, Santiago de Compostela, (Spain).  
+
+<|ref|>text<|/ref|><|det|>[[139, 453, 861, 757]]<|/det|>
+The biological and technological importance of anion- mediated processes has made the development of improved methods for the selective recognition of anions one of the most relevant research topics today. The hydration sphere of anions plays an important role in the functions performed by anions by forming a variety of cluster complexes. Here we describe a supramolecular capsule that recognizes new hydrated anion clusters. These clusters are most likely composed of three ions that form hydrated C3 symmetry complexes that are entrapped within the supramolecular capsule of the same symmetry. The capsule is made of self- assembled Acp- based cyclic peptide equipped with a tris(triazolylethyl)amine cap. To recognise the hydrated anion clusters, the hexapeptide capsule must disassemble to entrap them between its two subunits.  
+
+<|ref|>text<|/ref|><|det|>[[139, 780, 860, 886]]<|/det|>
+Anion recognition is a biologically and technologically relevant process in which ad hoc designed receptors are able to bind to a specific ion triggering a specialised function.[1- 2] In the last years, a large number of new synthetic receptors have been developed for this purpose.[3- 4] The design principles used in this development are
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 110, 861, 817]]<|/det|>
+generally dominated by directional noncovalent interactions, such as halogen, chalcogen or hydrogens bonds, among others, rather than by purely electrostatic ones, as they generally provide better selectivity.[5,6] In this regard, nature is able to differentiate biologically relevant ions of similar properties, such as fluoride or chloride, to carry out selective cross- membrane transport with remarkable selectivity.[7,8] The strong hydration of anions is one of the main obstacles to their recognition due to the additional energetic penalty paid for their desolvation during host- guest binding. In some cases, the ion selectivity in aqueous media does not depend as much on the relative size or charge of the ions, but on the size of their corresponding hydrates.[9,10] In fact, some authors have proposed that the selectivity in fluoride transport by FLUC protein could be explained in a similar way to the \(\mathrm{K^{+} / Na^{+}}\) selectivity of sodium ion channels.[11] Additionally, the greater solubility of negatively- charged proteins may be related to stronger hydration free energies for negatively charged groups on protein sufaces than their cationic counterparts.[12] Therefore, water- anion clusters represent an important chemical complex that is not yet fully understood despite their biological relevance. Several experimental and theoretical studies have been carried out to unmask the behaviour of hydrated halide clusters or other ions.[13,14,15,16] Most of these clusters are formed by a single ion surrounded by a variable number of water molecules, although polyanionic clusters have also been reported.[17] The recognition of two or more anions is critical to many biological processes, such in chloride transport.[18] One important approach to study this type of clusters is by analysing those generated in confined spaces.[19,20] Thus, several water clusters have been characterized by studying those formed in small protein pockets or other nanostructured materials. In this work we describe a new class of hydrated polyanionic clusters, most likely three, embedded between two cyclic peptide subunits of a novel supramolecular dimeric capsule.  
+
+<|ref|>title<|/ref|><|det|>[[459, 879, 536, 896]]<|/det|>
+# Figure 1.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 110, 861, 760]]<|/det|>
+Supramolecular capsules are dynamic molecular assemblies composed of two or more self- complementary units held together by non- covalent interactions that, upon assembling, form an empty closed system whose internal cavities are useful tools in host- guest chemistry.[21,22] They can be considered to create a new kind of molecular environment in which trapped molecule(s) exist. A wide variety of supramolecular capsules have been prepared with complementary or opposite functions, ranging from large to small, from charged to neutral or from open to closed systems. They exhibit unique properties, such as remarkable spectrochemical, electrochemical or magnetic effects. Therefore, they are useful tools for the molecular recognition of a wide variety of anionic, cationic or neutral guests (including gases) and are also able to catalyse a number of reactions leading to unusual products in a remarkably efficient way.[23,24,25,26] Recently, we have shown that flat- shaped cyclic peptides[27,28] equipped with a molecular cap (porphyrin moiety) provide a hydrogen- bonded dimeric receptor that was able to entrap long linear 4,4'- bipyridyl derivatives (Figure 1).[29] Here, we have created a smaller variant topped with a tris- triazolylamine component that is able to recognize a new class of water- anion clusters. To do this, the amide protons involved in the hydrogen bonds that supramolecularly bind both receptor subunits participate in the recognition of the hydrated anion. These clusters sandwiched between the two cyclic peptide components resembles a “molecular burger”, in which the anionic complexes are intercalated, like meat patty and cheese, between the two subunits that play the role of a “molecular bun”. The isobutyl pendants of Leu side chain are cross- interdigitated providing a hydrophobic supramolecular complex that can facilitate ion transport across lipid membranes.  
+
+<|ref|>text<|/ref|><|det|>[[139, 787, 861, 893]]<|/det|>
+Results and Discussion. Recently, we have prepared the cyclic hexapeptide bearing three properly chains attached to the \(\gamma\) - residues (CP1) that are pointing opposite to the dimerization plane (D1).[30,31,32] These groups not only prevent stacking of cyclic peptides across that face, thereby restricting the assembling process to discrete
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[138, 110, 861, 415]]<|/det|>
+dimers rather than nanotubes, but can also act as valuable reactive element for the incorporation of other functional moieties at the entrance of the dimer (Figure 1). For example, Sonogashira cross- coupling provided derivatives bearing ortho- or meta- oriented pyridine moieties.[31] The resulting peptides were able not only to encapsulate Xe atoms in the tubular structure of stacked dimers, but also to efficiently transport ion pairs in model membranes.[30] Additionally, the acetylene moiety was used in a copper- catalysed alkyne- azide cycloaddition reaction to incorporate a variety of oligoethylene glycol pendants to generate novel membrane spanners.[32] Here, we describe the preparation of self- assembled cavitands (Figure 2) covered by a tris- triazolyl motif attached to a central core that not only plays the role of a molecular cap but also acts as a binding site for the recognition of small anions.[33,34,35,36,37,38]  
+
+<|ref|>sub_title<|/ref|><|det|>[[458, 446, 536, 464]]<|/det|>
+## Figure 2.  
+
+<|ref|>text<|/ref|><|det|>[[138, 488, 861, 708]]<|/det|>
+Taking in account the aforementioned objectives, the cyclic peptide CP1 was synthesized using the protocol described previously (Figure 2).[30] For the cap incorporation, CP1 was subjected to a triple click reaction using the previously optimised method in which catalytic amounts of air- sensitive copper complex [Cu(MeCN)4PF6] (10 mol% per alkyne) were used in presence of DIEA and TBTA ligand.[32,39] Under these conditions the expected cavitand CP2 was obtained, although in low yield (15%). After screening for improving reaction conditions, a new protocol was found using copper bromide and DBU in toluene, providing CP2 in 48% yield.[40]  
+
+<|ref|>sub_title<|/ref|><|det|>[[458, 738, 536, 756]]<|/det|>
+## Figure 3.  
+
+<|ref|>text<|/ref|><|det|>[[139, 787, 861, 892]]<|/det|>
+This peptide retained the self- assembling properties of its precursor, CP1, adopting the proposed flat conformation and forming the C3- symmetric dimer (D2), as denoted by NMR experiments (Supplementary Fig. 1).[41,42] Clear evidence of dimer formation was derived from the down- field shift of the amide proton (8.04 ppm) whose
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 110, 861, 875]]<|/det|>
+chemical shift did not change with CP dilution up to \(200~\mu \mathrm{M}\) in \(\mathrm{CD}_2\mathrm{Cl}_2\) , confirming the large dimerization constant ( \(\mathrm{Ka} > 10^{4} \mathrm{M}^{- 1}\) ) similar to the non- capped derivatives.[29] Further evidence of the dimeric structure derived from the MS (2045.25 [ \(\mathrm{M}_{\mathrm{D2}} + \mathrm{Na}]^+\) , Supplementary Fig. 2) and FTIR spectroscopy, with amide A band a \(3286~\mathrm{cm^{- 1}}\) and carbonyl vibrations at \(1623\) and \(1532~\mathrm{cm^{- 1}}\) , characteristic of the antiparallel- type \(\beta\) - sheet structure (Supplementary Fig. 3).[43,44,45] The final evidence was obtained from the X- ray crystallography analysis of a crystal obtained from a solution of \(\mathbf{D2}\) in \(10\%\) \(\mathrm{CD}_3\mathrm{CN}\) in \(\mathrm{CD}_2\mathrm{Cl}_2\) from which a dimeric supramolecular capsule held by six hydrogen bonds between both subunits was found (Figure 3a- b). Interestingly one acetonitrile moiety is entrapped in the inner cavity of this dimer, with its nitrile group pointing towards the protons of the triazolyl moieties of one of the subunits. On the other side, one of the triazole moieties is rotated, arranging two of the nitrogen atoms of the heterocycle towards the internal cavity, perhaps with the goal of partially filling it due to the lack of a suitable guess. We checked if bis- nitrile derivatives, i.e., malononitrile (MN), could interact simultaneously with both tris- triazole motives improving its encapsulation (Supplementary Figs. 4, 5 and 6). We found out that it was necessary to add up to 30 equivalents of malononitrile (Supplementary Fig. 5) to a dichloromethane ( \(\mathrm{CD}_2\mathrm{Cl}_2\) ) solution of \(\mathbf{D2}\) to get a new species (MN \(\mathbf{\sigma} = \mathbf{D2}\) ), in a 4 to 1 ratio with respect to the empty capsule. This new species was assigned to the encapsulated complex due to the down- field shift of triazole proton ( \(\Delta \delta = 0.06\) ppm). 2D- NMR experiments confirmed that \(\mathrm{H}\beta_{\mathrm{Acp}}\) (cis- oriented with carboxy and amino groups) suffered down- field shift (Supplementary Fig. 6), which is consistent with the incorporation of a malononitrile molecule in the cavity. The weak interaction is most likely due to the \(109^{\circ}\) angle between the two nitriles, which possibly hinders the simultaneous formation of strong hydrogen bonds with both tris(triazolylethyl)amine moieties. Succinonitrile (SN) was also evaluated, considering its additional methylene moiety allows both cyano groups to be arranged at a \(180^{\circ}\) angle, although we were concerned that its length would exceed
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 110, 863, 530]]<|/det|>
+the size of the cavity. Fortunately, the addition of small amounts of SN ( \(\sim 0.5\) equiv) over a \(\mathrm{CD}_2\mathrm{Cl}_2\) solution of D2 (6.4 mM) already gave rise to a new set of signals corresponding to \(\mathbf{SN}\subset \mathbf{D2}\) (Supplementary Figs. 4 and 7), suggesting a higher affinity than MN. After the addition of 7 equiv, the encapsulated complex ( \(\mathbf{SN}\subset \mathbf{D2}\) ) is already the main component in the mixture, although in a 1.4:1 ratio with D2 (Supplementary Fig. 7A). Whereas for the MN recognition there is almost no change in the chemical shift of amide protons, for the SN the signal is up- field shifted (7.84 ppm) suggesting a tight binding that reduced the conformational freedom (signals are sharper than those of the free capsule) and stress the hydrogen bond network. The new singlet at 3.30 ppm corresponds to the methylene groups from the entrapped SN molecule as indicated the 2D- NMR experiments (Figure S7B- C). DFT geometry optimizations confirm the stability of both complexes (Supplementary Fig. 8, see Supplementary Discussion III for further information). The main difference between both encapsulated bis- nitrile complexes was the length of hydrogen bonds, which become longer as the size of the encapsulated nitrile increases.  
+
+<|ref|>sub_title<|/ref|><|det|>[[458, 562, 536, 579]]<|/det|>
+## Figure 4.  
+
+<|ref|>text<|/ref|><|det|>[[138, 602, 863, 879]]<|/det|>
+Once the ability of D2 to form hydrogen- bonded supramolecular inclusion complexes with appropriate guests and the capability of proton of tris- triazole cap to form hydrogen bonds with the entrapped molecule was confirmed, the recognition of small anions was evaluated. Initially, we assessed halide recognition using NMR titrations (specifications regarding \(^1\mathrm{H}\) NMR titrations are detailed in Supplementary Discussion I). Clearly, addition of different portions of tetrabutylammonium fluoride to a solution of CP2 (7.3 mM) in \(10\%\) \(\mathrm{CD}_3\mathrm{CN} / \mathrm{CD}_2\mathrm{Cl}_2\) gave rise to a new set of signals that were attributed to the recognition of the fluoride anion (Figure 4). In any case, it was necessary to add ca. 3 equivalents of TBAF to achieve the complete disappearance of the signals corresponding to D2. Similar results were obtained with the addition of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 110, 861, 848]]<|/det|>
+TBACI (Supplementary Fig. 9) under similar conditions (10% \(\mathrm{CD_3CN / CD_2Cl_2}\) ), but in this case it was necessary to add more than 15 equivalents of chloride to fully shift the equilibrium towards the new species, suggesting a weaker interaction (Table 1). No changes were observed for the addition of larger halides (bromide and iodide) or nitrate. Careful analysis of the \(^1\mathrm{H}\) NMR spectra of both experiments reveal that, as opposed to what it was found in anion recognition with tris(triazolyl) receptors,[33- 38] the heterocyclic protons are up- field shifted (7.47 ppm). This suggests that these protons must not be participating in halide recognition. On the other hand, the amide proton signals are down- field shifted (from 8.00 to 10.00 ppm for fluoride and to 8.62 ppm for chloride), implying their participation in stronger hydrogen bonds. Conformational changes derived from the interaction with the anions are clearly evidenced by the splitting of the signal of the methylene linker (singlet at 4.59 ppm for D2) between the triazole rings and the cyclic peptide backbone into two doublets at 4.90 and 4.04 ppm. In addition, conformational changes at the CP backbone suggest changes in the flat conformation characteristic of the sheet structure. For example, \(\mathrm{H\alpha_{Leu}}\) , initially at 4.81 ppm, undergoes a remarkable up- field shift to 4.15 ppm for fluoride (4.19 for chloride), while the \(\mathrm{H\gamma_{Acp}}\) , initially at 4.73 ppm, undergoes an opposite, albeit milder, down- field shift to 4.90 ppm (4.83 ppm for chloride). Two- dimensional NMR experiments provided evidence of conformational changes, in which nOe cross- peaks that are not present in the empty capsule appear after addition of the halide (Supplementary Fig. 10), such as the one between \(\mathrm{H\gamma_{Acp}}\) and one of the protons of \(\mathrm{H\beta_{Leu}}\) upon recognition of fluoride. The disappearance, upon complexation, of the strong nOe cross- peaks between \(\mathrm{H_{triazole}}\) and \(\mathrm{H\alpha_{Leu}}\) and of the methylene linker and the \(\mathrm{H\beta_{Leu}}\) also seems relevant. Further evidence of the halide and CP interactions were found at ESI- MS, in which several ion peaks corresponding to complexes between both components are the main detectable species (Supplementary Fig. 11).  
+
+<|ref|>text<|/ref|><|det|>[[463, 875, 533, 891]]<|/det|>
+Table 1.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 110, 861, 875]]<|/det|>
+In a search for improving the binding, we discovered that water molecules play a key role in the supramolecular recognition process. The addition of 4Å molecular sieves, with the aim of drying the solution of TBACl and D2, gives rise to the recovery of the starting dimeric capsule. Filtration of the resulting solution and addition of 10 μL of water provoke the recovery of the chloride-capsule complex (3Cl.nH2O<2CP2) spectrum (Supplementary Fig. 12A). Inspired by this finding, new studies with the other larger and less hygroscopic anions were reviewed to evaluate the role of water molecules in their recognition. The addition of a few drops of water (7 μL) to the 10% CD3CN/CD2Cl2 solution of D2 (4.7 mM), resulted in only a small down-field shift of the triazole proton signal, presumably due to water encapsulation (Supplementary Fig. 13); but successive additions of increasing amounts of TBAB yielded similar changes to those already described for chloride additions (28 equivalents added and \(\delta_{\mathrm{NH}} 8.37 \mathrm{ppm}\) , Table 1) (Supplementary Fig. 14). Remarkably, even recognition of iodide was also observed, although more equivalents of iodide (86 equiv, Table 1) were required, and the down-field shift of NH signal (8.07 ppm) was even smaller (Supplementary Fig. 15). This definitely underlines the importance of water molecules in the anion recognition process, most likely due to the ability of the capsule to recognize the hydration spheres of the anions.[9] Next, other anions (see Table 1, Supplementary Figs. 16, 17, 18 and 19) were evaluated and a similar behaviour was found; those that were recognised, such as nitrate, acetate or azide ions, only did so in the presence of some water molecules. Larger ions, such as tribromide or hexafluorophosphate did not interact even in the presence of different amounts of water. These results confirm the remarkable ability of D2 to recognise the hydration spheres of different anions. We decided to further study the recognition of acetate to stress out the importance of water molecules in the complexation process by drying this solution. After incubating a solution of D2 containing 66 equivalents of TBAA with molecular sieves for 48 hours, the \(^1 \mathrm{H}\) NMR signals corresponding to D2 were recovered, while those assigned to the original
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 110, 861, 820]]<|/det|>
+complex disappeared. The subsequent addition of few drops of water (10 μL) immediately re- established the original complex (Supplementary Fig. 12B). Interestingly, while overnight incubation was sufficient to dehydrate the encapsulated chloride complex with molecular sieves, drying the acetate mixture required two days for the full recovery of D2, suggesting a tighter binding. Apart of the key role paid by water molecules, the chemical shift and number of equivalents of the different anions tested suggest that size, shape, and basicity are important parameters in the recognition process. To confirm the role played by the capsule and the tris(triazolylethyl)amine cap, CP1 was evaluated for the recognition of anions. TBAF additions over a solution of D1 in dichloromethane with a 10% of acetonitrile did not provide any change in its \(^1\mathrm{H}\) NMR spectrum, confirming the relevance of the cap in the anion/water cluster recognition (Supplementary Fig. 20). The role of the tetrabutylammonium counterion was also evaluated, for this purpose we decided to make use of the known affinity of crown ethers for alkali cations. \(^{[46]}\) The 15- crown- 5 is known for being the one whose radius fits better with \(\mathrm{Na}^+\) . \(^{[47]}\) Therefore, we were able to solubilize NaOAc in deuterated acetonitrile using this crown ether and this solution was used in new titration experiments with D2 in dichloromethane. Similar results were obtained, although larger amounts of the sodium acetate/crown ether acetonitrile solution were required to shift the equilibrium, most likely due to the lower solubility of the complex under these conditions (Supplementary Fig. 21A). In fact, the NMR spectrum of the mixture two weeks after titration showed a clear increase of the acetate complex with respect to the initial conditions, going from 1.4:1 ratio to 3:1 (Supplementary Fig. 21B). This can be attributed to the slow solubilization of the acetate salt triggered by the formation of the complex. ROESY spectra showed a similar cross- peaks pattern to those obtained using tetrabutylammonium as counterion (Supplementary Fig. 22).  
+
+<|ref|>text<|/ref|><|det|>[[140, 844, 860, 892]]<|/det|>
+The strong down- field shift of amide protons upon anion addition suggests their involvement in its recognition, consequently, further experiments were carried out to
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 111, 863, 760]]<|/det|>
+understand this behaviour. It is well stablished that cyclic peptides made by five- membered ring \(\gamma\) - amino acids (Acp) form heterodimeric complexes with CPs containing six- membered ring \(\gamma\) - amino acids, i.e. CP3,[42,48,49] which are more stable than the corresponding homodimers. Therefore, we decided to evaluate the importance of capsule integrity in the cluster recognition. Consequently, CP3, which was prepared following a similar strategy to the one used in the preparation of CP2 but starting from 3- aminocyclohexanecarboxylic acid (Ach, Supplementary Fig. 23), also forms the corresponding dimer D3 in dichloromethane solution. CP3 was mixed with a solution of D2, and the resulting mixture showed in the NMR spectra the appearance of a new set of signals that were assigned to the supramolecular aggregate D2- 3 (Figure 5 and Supplementary Fig. 24). Further evidence of the heterodimeric structure derived from MS experiments (Supplementary Fig. 25) in which the 1840.18 ion peak corresponds to the heterodimeric complex. This confirmed, once again, the higher stability of the heterodimeric aggregates because of the better van der Walls fitting between both cycloalkyl rings.[50] The formation of this heterodimer allowed us to evaluate if dimers containing only one tris(triazolylethyl)amine cap (cavitand D2- 3) could still recognize anions. Interestingly, after the addition of small portions of fluoride or chloride (TBAF or TBACl) on the \(\mathrm{CD_2Cl_2}\) solution of heterodimer D2- 3, simultaneous recovery of homodimer D3 and the formation of the corresponding complexes of CP2 with the anions (3F- nH2O<2CP2 and 3Cl- nH2O<2CP2) were observed. Three equivalents of fluoride were also required to shift the equilibrium to the homodimeric components. This confirms that the interaction with hydrated anions is even more favourable than that of the heterodimer and that both capped subunits are necessary in this process.  
+
+<|ref|>sub_title<|/ref|><|det|>[[458, 789, 537, 806]]<|/det|>
+## Figure 5.  
+
+<|ref|>text<|/ref|><|det|>[[140, 836, 860, 885]]<|/det|>
+Cyclic peptide CP3 was then transformed into the tris(triazolylethyl)amine capped Ach derivative (CP4) using similar conditions in \(50\%\) yield (Figure 2).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 110, 861, 587]]<|/det|>
+Therefore, the recognition properties of the new Ach- based capsule topped with the tris(triazolylethyl)amine motif were also studied. These cavitand self- dimerized in dichloromethane solution to form D4, as denoted by NMR, MS and FTIR (see Supplementary Information section 7 and 8). Once again, single crystal suitable for X- ray diffraction of this compound also confirmed the capsule formation that also entrap one acetonitrile molecule in its cavity (Figure 3C- D). D4 has all the triazole protons pointing towards the internal cavity making its structure more symmetric than the Acp- based one (figure 3a- b), with similar length in all the interpeptide hydrogen bonds. These are generally longer (2.24 Å) than those of D2, which range from 1.91 to 2.35 Å (2.10 Å in average), although the capsule D4 is slightly more compact with a shorter distance between the two nitrogen atoms of the tris(triazolylethyl) caps (14.30 Å versus 14.85 Å). In contrast to the recognition properties of D2, additions of more than twenty equivalents of TBAF or TBACl do not result in any change in the NMR spectra of D4 (Supplementary Fig. 26). This indicates that the Ach- based capsule is not capable of recognising anions, most likely due to the greater rigidity of the six- membered ring of this \(\gamma\) - amino acid that prevents the CP from adopting the appropriate conformation for the recognition of such species.  
+
+<|ref|>text<|/ref|><|det|>[[137, 616, 861, 893]]<|/det|>
+Both cavitands (CP2 and CP4) also assemble into the heterodimer D2- 4 (Figure 5 and Supplementary Fig. 27A) when an equimolar mixture of both compounds in nonpolar solvents (CD₂Cl₂) is prepared. The appearance of new signals in the \(^1\mathrm{H}\) NMR spectra that do not correspond to any of the homodimers confirms the formation of D2- 4. For example, the broad signal at 4.95 ppm belonging to the HαLeu and the one at 4.22 ppm corresponding to one the methylene of the tris(triazolylethyl)amine cap are signals that belong to CP2 of the heterodimer (Supplementary Fig. 27B). Moreover, the MS also confirms the formation of D2- 4 (Supplementary Fig. 28). Once again, addition of fluoride (TBAF) to this dichloromethane (CD₂Cl₂) mixture prompts the splitting of the heterodimer D2- 4 into the corresponding homodimer D4 and the complex of hydrated
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[138, 109, 861, 273]]<|/det|>
+fluoride with cavitand \(\mathbf{CP2}\) \((3\mathrm{F}\cdot \mathrm{nH}_2\mathrm{O}\subset 2\mathrm{CP2})\) . In all cases, the addition of more than three equivalents of fluoride was necessary to achieve complete dissociation of the heterodimers. DFT geometry optimization of both heterodimeric species, D2- 3 and D2- 4, are shown in Supplementary Fig. 29, which once again confirmed the stability of the mentioned dimers and provide further information about the hydrogen bonds length and angles (see also Supplementary Discussion III for further information).  
+
+<|ref|>sub_title<|/ref|><|det|>[[458, 304, 536, 321]]<|/det|>
+## Figure 6.  
+
+<|ref|>text<|/ref|><|det|>[[138, 350, 861, 859]]<|/det|>
+Crystals suitable for X- ray diffraction were obtained from solutions containing the supramolecular capsules D2 and the tetrabutylammonium halides (chloride and fluoride, Figure 6). To our delight, in both cases the cyclic peptide capsules entrap a new kind of hydrated halide clusters made by three ions, as confirmed by the number of tetrabutylammonium ions that co- crystalized with the peptide capsule. Four and eight water molecules, for chloride and fluoride, respectively, are forming these clusters, confirming the higher tendency of the latter to have larger hydration shells. In both cases the three halide ions are distributed into six equivalent chemical positions that are shared with another three water molecules. Although for the chloride crystal all the ion positions are forming a hexagonal structure with all the positions placed at the same layer, for fluoride cluster, the six positions are placed at two different levels forming two triangular structures that are \(60^{\circ}\) rotated with respect to each other. To entrap these clusters, cyclic peptide dimers dissociate to allow amide protons to hydrogen- bond with halide ions and water molecules.[51] It is notorious that in the solid state all the carbonyls are oriented towards the opposite side in which the interaction with the anionic cluster occurs, which could explain the observed up- field shift of \(\mathrm{H}\alpha_{\mathrm{Leu}}\) in the NMR spectrum. This conformational change is due to the geometrical variations of the \(\alpha\) - amino acids that go from the characteristic \(\beta\) - sheet conformation of flat disc- shaped CPs to a turn-
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[140, 111, 858, 159]]<|/det|>
+like structure. Interpeptide distance is slightly larger for the encapsulated fluoride cluster than that for chloride despite the larger size of the latter.  
+
+<|ref|>text<|/ref|><|det|>[[138, 189, 861, 724]]<|/det|>
+With respect to the fluoride- capsule complex (Figure 6a- b), the unit cell has two non- equivalent complexes \((3\mathrm{F}\cdot 8\mathrm{H}_{2}\mathrm{O}\subset 2\mathrm{CP}2)\) . In each complex, in addition to the three water molecules exchangeable by fluoride ions and hydrogen bonded to the amide proton, there are five other crystallographic positions preferably occupied by water molecules, although fluoride could also partially occupy any of these positions, since it is not possible to unambiguously differentiate both atoms due to their similar electron densities. In any case, to fulfil the hydrogen acceptor capability of the fluoride ion, we assume that these ions must be located in the positions in which they are bonded to the amide protons of the same cavitand and surrounded by three water molecules forming the first hydration shell of each fluoride (Supplementary Fig. 30). The fluoride occupancy in the two \(3\mathrm{F}\cdot 8\mathrm{H}_{2}\mathrm{O}\subset 2\mathrm{CP}2\) complex of each unit cell is not exactly the same (Supplementary Fig. 31), even though both complexes present analogue disposition. With respect to the rest of water molecules, there are two that are axially placed, forming hydrogen bonds with the fluoride- water exchangeable positions, while the other three are in the cluster equatorial perimeter forming hydrogen bonds with the exchangeable fluoride- water positions (Supplementary Fig. 32). These water molecules are placed at the position in which capsule is not fully closed forming a window. The Leu side chains are facing each other to create a hydrophobic oval- shaped aggregate that entrap the halide cluster in a non- polar environment.  
+
+<|ref|>text<|/ref|><|det|>[[139, 752, 861, 885]]<|/det|>
+Concerning the chloride complex (Figure 6c- d) the six equivalent positions are in the same plane and each chloride ion is hydrogen- bonded to one amide proton with N...Cl distance of 3.21 Å. In this case, the coating with the leucine side chains is less compact, leaving a wider window as compared to the fluoride complex. Furthermore, electron density can only be attributed to a maximum of four water molecules, one of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[139, 110, 861, 302]]<|/det|>
+them located at \(50\%\) occupancy at the top and bottom of the hexagonal bipyramid and more deeply buried in the cavity of the supramolecular capsule, remaining partially hydrogen- bonded to the three chloride ions. In this crystal structure, the complex of the capsule with the chloride cluster is co- crystallized with the dimer D3, forming columns in which D3 is alternated with the encapsulated trichloride cluster (3CI.4H2O<2CP2). Within D3 there is a dioxane molecule occupying three equivalent positions around the ternary symmetry axis of the dimer (Supplementary Fig. 33).  
+
+<|ref|>text<|/ref|><|det|>[[139, 333, 861, 750]]<|/det|>
+Unfortunately, it was no possible to obtain crystal structures of the complexes formed with the other anions (bromide, acetate, azide, iodide and so on) that would allow unequivocal confirmation of the formation of similar structures with these ions. In any case, the previous characterizations suggest the formation of clusters that are embedded in the equatorial cleft generated by the two CP subunits. To confirm this further, we carried out a detailed analysis comparing the NMR data of the different complexes and the X- ray diffraction data (Supplementary Discussion II), from which we concluded that the recognition process most likely involves the trapping by two CP2 units of clusters composed of three anions, although we do not have a conclusive evidence of such stoichiometry, surrounded by several water molecules depending on the type of anion and its solvation. To this complex we have used the coding mA. \(\mathbf{nH}_2\mathbf{O}\mathbf{C}2\mathbf{CP}2\) for the complex with the anion cluster in which "m" represents the number of anions entrapped between both CPs (most likely three), "A" indicates the type of anion and "n" represents the number of water molecules that form each anion cluster.  
+
+<|ref|>text<|/ref|><|det|>[[139, 780, 860, 885]]<|/det|>
+After confirmation of the anion recognition ability of the supramolecular capsule D2, transmembrane transport experiments were carried out.[1] For that purpose, lucigenin- trapped liposomes (LG<LUV) were prepared with which the intravesicular delivery of chloride ions was clearly stablished (Figure 7A).[52] The transport is slow
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 111, 861, 848]]<|/det|>
+compared with previously published chloride transporters and high concentration of capsule is required \((\mathrm{EC}_{50} \sim 100 \mu \mathrm{M})\) . Additionally, we decided to examine the chloride transport ability of D4 and D1 using the lucigenin assay. As we expected from the previous NMR findings, neither D1 nor D4 were capable of transporting chloride (Supplementary Fig. 34), confirming that the recognition of the ions was necessary to be able to mediate its transport. Apart from that, further variations of the D2 lucigenin assay, revealed that the transport efficiency did not change with the counterion used \((\mathrm{Na}^{+}, \mathrm{Li}^{+}, \mathrm{K}^{+} \text{or} \mathrm{Cs}^{+})\) , suggesting that cation is not involved in the transport process (Figure 7A3). To confirm the potential antiport transport of nitrate, experiments in which nitrate was substituted for the more hydrophilic sulphate, whose transmembrane transport is extremely difficult, were carried out (Supplementary Fig. 35). [53,54] These experiments did not show any significant reduction in chloride transport rates, suggesting that chloride/nitrate exchange must not be involved in the transport mechanism. Therefore, the symport \((\mathrm{H}^{+} / \mathrm{Cl}^{- })\) or antiport \((\mathrm{OH}^{- } / \mathrm{Cl}^{- })\) must be associated to this migration. To confirm association of chloride transport with change in the pH, HPTS loaded vesicles (HPTS \(\equiv\) LUV) were used (Figure 7B). [55] Thus, vesicles basification promoted by D2, denoted by a fluorescence increase, would be associated to the co- transport of chloride. Unambiguous and fast enhancement of dye emission was found upon creating a pH gradient of almost one unit after the addition of a sodium hydroxide solution to the extravesicular medium, yielding an enhanced activity with \(\mathrm{EC}_{50} = 3 \mu \mathrm{M}\) , suggesting that anion transport was the rate limiting step and not the \(\mathrm{H}^{+}\) or \(\mathrm{OH}^{- }\) co- transport. To clarify this finding, HPTS studies in the presence of a proton transporter (FCCP) were carried out (Figure 7D). [56] As expected no changes in transport rates were found, confirming proton transport was not the limiting step. DLS measurements were then performed, which confirmed both the homogeneity and integrity of the vesicles throughout these experiments and, consequently, the observed
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[139, 111, 858, 159]]<|/det|>
+fluorescence changes are not due to bleaching or membrane disruption (Supplementary Fig. 36).  
+
+<|ref|>text<|/ref|><|det|>[[137, 189, 862, 753]]<|/det|>
+Finally, to evaluate the relative rates in anion transport, competitive experiments were carried out using HPTSCLUV (Figure 7C).[52,57] For this purpose, vesicles whose internal buffer contained sodium chloride (100 mM) at pH 7 were placed in a variety of isosmotic buffer solutions with different sodium salts with other counterions. In this type of experiments, a pH gradient is generated if the transporter (D2) facilitates dominant ion influx or efflux depending on anion selectivity. These permeability differences give rise to a membrane potential that drives net proton transport. Vesicle acidification occurs when ion influx is faster than chloride efflux, while ions that are transported more slowly than chloride cause an increase in intravesicular pH. We found that acetate, fluoride, and azide provided vesicle basification, while bromide and iodide were transported faster than chloride, being acetate the slower influxed anion and iodide the faster one. Therefore, D2 showed Hofmeister pattern \((\mathrm{I}^{-} > \mathrm{Br}^{-} > \mathrm{NO}_{3}^{-} \sim \mathrm{Cl}^{-} > \mathrm{N}_{3}^{-} > \mathrm{F}^{-} > \mathrm{OAc}^{-})\) with the exception of nitrate that is almost as fast as chloride.[58] The strongly hydrated, salting- out ions, are those with slower transport rates while previous ion recognition experiments showed the need of water molecules in the recognition of anion cluster and the strong binding for fluoride or acetate compared with iodide or bromide. Therefore, it seems that ions exchange \((\mathrm{K}_{\mathrm{on}} / \mathrm{K}_{\mathrm{off}})\) at the interfaces must play an important role in the transport process and not so much the selectivity of the binding itself. All the details regarding the transport assays are carefully addressed in Supplementary Discussion IV.  
+
+<|ref|>sub_title<|/ref|><|det|>[[457, 782, 538, 799]]<|/det|>
+## Figure 7.  
+
+<|ref|>text<|/ref|><|det|>[[139, 823, 860, 871]]<|/det|>
+In conclusion, we have presented the design, synthesis, and anion recognition properties of a cyclic \(\alpha , \gamma\) - hexapeptide equipped with a tris(triazolylethyl)amine cap. We
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[138, 110, 861, 501]]<|/det|>
+have shown that, in presence of some anions, the self- assembled CP is able to reaccommodate its backbone conformation from a self- dimerizing flat circular- shaped structure to a triangular one that creates a binding pocket suitable for hosting hydrated anions. Moreover, we have rigorously identified and studied the different aspects involved in the recognition. First, the essential role played by the water molecules that are involved not only in the formation of anion- water clusters, forming the hydration shell of the different anions, and helping them to take the appropriate size and shape, but also in the interaction and entrapment inside the cavity formed by the two peptide hemicapsules. In addition, two key structural components of the cyclic peptide were identified: the five- member ring \(\gamma\) - amino acid and the tris(triazolylethyl)amine cap, without which the anion accommodation is not possible. Furthermore, we have shown that the cationic counterion does not play any relevant role in the supramolecular recognition, which supports our theory of the three main actors: capped peptide, water and anion.  
+
+<|ref|>text<|/ref|><|det|>[[138, 510, 861, 757]]<|/det|>
+Finally, the anion transporting properties were also explored. Our studies have shown that D2 can transport different anions across model lipid membranes, with a tendency to favour the transport of anions with weaker hydration spheres, which are generally weaker recognized. Transport is associated with the modification the pH of the intravesicular medium, most likely through a \(\mathrm{Cl}^-\) / \(\mathrm{H}^+\) symport mechanism, although the \(\mathrm{Cl}^-\) / \(\mathrm{OH}^-\) antiport exchange could not be ruled out, the rate- limiting step being the migration of the anions. The peptide composition, design simplicity and anion transporting properties linked to pH regulation activity open- up several avenues for the therapeutic use of these novel self- assembled systems.[9]  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 796, 218, 812]]<|/det|>
+## Methods  
+
+<|ref|>text<|/ref|><|det|>[[140, 824, 435, 841]]<|/det|>
+General \(^1\mathrm{H}\) - NMR titrations protocol
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[139, 111, 861, 300]]<|/det|>
+Stock solutions of internal standard, either dioxane (2.5 mM) or TMSS (0.15 mM), in a mixture of \(\mathrm{CD}_3\mathrm{CN}\) in \(\mathrm{CD}_2\mathrm{Cl}_2\) (10% v/v) were prepared. These solutions were used to prepare all the samples needed in the titration experiments, in order to keep the concentration of the internal standard constant throughout the experiments. The signals from both standards, dioxane (s, 3.6 ppm in \(\mathrm{CD}_2\mathrm{Cl}_2\) ) or TMSS (s, 0.18 ppm in \(\mathrm{CD}_2\mathrm{Cl}_2\) ), were used to calculate the concentration of all the components along the titration experiments.  
+
+<|ref|>text<|/ref|><|det|>[[139, 310, 860, 415]]<|/det|>
+Generally, \(450~\mu \mathrm{L}\) of the solution containing the corresponding peptide capsule (D2, D3 or D4, ca. 5mM) was placed in the NMR tube. After recording the \(^1\mathrm{H}\) NMR spectra of the starting sample, successive additions of the corresponding titrant were made via micropipette and the corresponding spectrum was taken.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 454, 488, 471]]<|/det|>
+## General procedure for vesicles preparation  
+
+<|ref|>text<|/ref|><|det|>[[138, 482, 861, 871]]<|/det|>
+Vesicles were prepared, under argon, in a round bottom flask by slow evaporation of a solution of EYPC in \(\mathrm{CHCl}_3\) (25 mg/mL, 1 mL) to form a thin and homogeneous film on the flask surface. The film was dried overnight under high vacuum and then carefully hydrated with the intravesicular aqueous media. The resulting mixture was tumbled for 1 hour in the rotavapor at 180 r.p.m. but at atmospheric pressure. Every 15 minutes, the rotavapor rotation angle was changed (60°, 50°, 45° and 35°) in order to get and homogeneous dispersion. After that, the milky sample was subjected to 11 freeze- thaw cycles ( \(\mathrm{N}_2\) (l) \(\rightarrow 40^{\circ}\mathrm{C}\) water), and the resulting suspension was extruded (25 times) across polycarbonate membrane (200 nm pore size). Finally, the suspension was passed through a size exclusion column (Sephadex G- 25) previously equilibrated with the isosmotic extravesicular medium.[56] The resulting vesicle suspension was taken up in a total volume of 5 mL, giving an approximate lipid concentration 6.6 mM. Size and particle number consistency between different vesicles batches were checked using DLS.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[140, 141, 512, 158]]<|/det|>
+## General procedure for transport measurement  
+
+<|ref|>text<|/ref|><|det|>[[138, 169, 863, 530]]<|/det|>
+General procedure for transport measurementIn a plastic cuvette containing a \(4\mathrm{mm}\) diameter stirring bar, the previously prepared vesicle suspensions (50 \(\mu \mathrm{L}\) ) were dispersed in the extravesicular media (1950 \(\mu \mathrm{L}\) ). The cuvette was placed with moderate stirring in the fluorometer equipped with a module that allows the successive additions of titrant during the measurement in the dark. Data (fluorescence emission band at \(535\mathrm{nm}\) for lucigenin essays or \(510\mathrm{nm}\) for HPTS experiments) were collected for thirty minutes every second. One minute after the experiment started, aqueous solutions of NaCl (25 \(\mu \mathrm{L}\) , 2 M, lucigenin assay) or NaOH (25 \(\mu \mathrm{L}\) , 0.5 M, HPTS assay) were added. After an additional minute (minute 2 of the measurement), a solution of the CP at different concentrations (25 \(\mu \mathrm{L}\) , in iPrOH for the lucigenin assay or in DMSO for the HPTS) were added. Finally, after 27 minutes an aqueous solution of Triton (50 \(\mu \mathrm{L}\) , \(10\%\) v/v) was added to provoke liposome lysis. After three more minutes of stabilization, the resulting signal was used for the data normalization.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 568, 301, 585]]<|/det|>
+## DLS measurements  
+
+<|ref|>text<|/ref|><|det|>[[138, 595, 863, 871]]<|/det|>
+DLS measurementsFor each batch of vesicle prepared, some control measurements were performed before the transport experiments to check the homogeneity between batches. The size, polydispersity index and particle concentration were measured using the same volumes as in the fluorescence assay. Four samples were collected and checked for each essay. Regarding the lucigenin experiments (Supplementary Fig. 36), the first sample, containing only the vesicles dispersed in the extravesicular media (blue line), the second sample, which corresponds to the mixture after the addition of the aqueous solution of NaCl (green line), the third measurement was carried out after the incorporation of CP (pink line), and finally, the fourth sample was done after the addition of Triton X- 100 (red line). The four samples were stirred for 25 minutes, and after a 5 min lag resting
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[139, 110, 860, 216]]<|/det|>
+perioded the DLS were recorded. For the HPTS assay (See Supplementary Fig. 36) in HEPES buffer (10 mM, NaCl 100 mM, pH 7), the analyzed mixtures correspond to the initial conditions (blue line), and after successive additions of NaOH (green line), cyclic peptide (pink line) and Triton (red line).  
+
+<|ref|>text<|/ref|><|det|>[[139, 225, 861, 415]]<|/det|>
+Notice that samples measurement indicates monodispersity, in view of the sharp slope of the correlation curves, which was maintained in all samples except after the addition of the surfactant, Triton X- 100, when the vesicles undergo lysis. After lysis, the faster decay of the correlation coefficient is a signal of the smaller particle size. Also, the slope, less sharp, indicates higher polydispersity. We measured the particle concentration, which also indicated homogeneity between batches and confirmed the stability of the vesicles until their lysis.  
+
+<|ref|>sub_title<|/ref|><|det|>[[141, 454, 378, 471]]<|/det|>
+## DFT geometry optimizations  
+
+<|ref|>text<|/ref|><|det|>[[139, 481, 861, 644]]<|/det|>
+Computational studies were carried out using the sources from Centro de Supercomputación de Galicia (CESGA). All calculations were performed with the Gaussian 16 rev. C01 package. The geometries used in the calculations were based on the crystal structures derived from this study. Calculations were performed in vacuum. We carried out DFT geometry optimizations using B3LYP with the moderate- size basis set 6- 31G (d, p). We also included GD3BJ as empirical dispersion.  
+
+<|ref|>text<|/ref|><|det|>[[139, 703, 857, 836]]<|/det|>
+Supplementary Information accompanies the paper on https://www.nature.com/nchem/. Detailed descriptions of the synthesis and characterization of key compounds, including NMR spectra ( \(^1\mathrm{H}\) and \(^{13}\mathrm{C}\) , NOESY and/or ROESY) and FTIR spectra of peptides CP2, CP3 and CP4. CCDC- 2311116 to CCDC- 2311119 contains the supplementary crystallographic data for this paper. The data can be obtained free of charge from The Cambridge Crystallographic Data Centre via https://www.ccdc.cam.ac.uk/structures.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[139, 112, 857, 301]]<|/det|>
+Acknowledgements This work was supported by the Spanish Agencia Estatal de Investigación (AEI) (PID2019- 111126RB- 100 and PID2022- 142440NB- 100), the Xunta de Galicia (ED431C 2021/21 and Centro singular de Investigación de Galicia accreditation 2019- 2022, ED431G 2019/03), and the European Union (European Regional Development Fund - ERDF). We also thank the ORFEO- CINCA network and Mineco (RED2022- 134287- T). V.L.- C. thanks the Xunta de Galicia for her research contract (ED481A- 2019/117). All calculations were carried out at the Centro de Supercomputación de Galicia (CESGA).  
+
+<|ref|>text<|/ref|><|det|>[[140, 328, 847, 344]]<|/det|>
+Correspondence and requests for materials should be addressed to Juan R. Granja (juanr.granja@usc.es).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[140, 98, 860, 500]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[139, 529, 861, 718]]<|/det|>
+<center>Figure 1. Previously reported supramolecular containers based on \(\alpha ,\gamma\) -cyclic peptides. Top: cyclic octapeptide (blue) topped with a porphyrin moiety (green) used in the recognition of \(4,4^{\prime}\) -bipyridines. Center: smaller alternatives derived from dimer-forming \(N\) -propargylated cyclic hexapeptides (CP1, grey) through Sonogashira cross-couplings with Iodopyridines or copper-catalyzed azide-alkyne cycloaddition (CuAAC) that have been used as ion transporters.[27] Bottom: cartoon model of supramolecular capsule derived from CP1 and a tris-azide derivative described in this work. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[163, 95, 816, 415]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[139, 440, 860, 515]]<|/det|>
+<center>Figure 2. Synthetic strategy used for the preparation of capsules D2 and D4 and initially proposed encapsulation model for the recognition of anions (2X<D2 and 2X<D4). </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[142, 99, 857, 589]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[139, 613, 861, 805]]<|/det|>
+<center>Figure 3. Side (a and c) and top (b and d) views of the crystal structures of dimeric supramolecular capsules D2 (top) and D4 (bottom), respectively. The molecules of acetonitrile entrapped in the cavity are represented in CPK models. The nitrogen of nitrile groups is pointing towards one of the caps close to the triazole protons with shorter distance in the Ach-based capsule (2.66 Å, bottom) than in the Acp derivative (2.79-2.68 Å, top). For clarity only triazole and amide protons are shown. The yellow dashed lines highlight the hydrogen bonds. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[168, 95, 828, 390]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[139, 410, 860, 515]]<|/det|>
+<center>Figure 4. NMR spectra of pure CP2 (bottom) and after the addition of different equivalents of fluoride (TBAF). In blue colour are highlighted the signals corresponding to the new species formed after the addition of the fluoride, light blue denotes the signals corresponding to the tetrabutylammonium counterion. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[144, 95, 856, 666]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[139, 690, 861, 823]]<|/det|>
+<center>Figure 5. Experiments of heterodimer (D2-3 and D2-4) formation followed by anion recognition, top view. Bottom a) NMR spectra corresponding to these studies in which the characteristic signals of each component are highlighted with specific colours; orange, green and lavender for homodimers D2, D3 and D4, respectively, dark blue for CP2 interacting with fluoride, and plum and teal green for heterodimers D2-3 and D2-4, respectively. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[140, 128, 855, 551]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[138, 576, 861, 853]]<|/det|>
+<center>Figure 6. Side (a) and top (b) view of the crystal structure of water-fluoride cluster entrapped in supramolecular capsule D2 (3F.8H₂O<2CP2) as twin complexes. In these complexes three fluoride ions (light green) and eight water molecules are hydrogen-bonded to the amide protons of two cyclic peptides at different planes (the hydrogen bond network (yellow dashed lines) in the cluster is only shown for one of the complexes). Side (c) and top (d) view of the crystal structure of encapsulated chloride-water cluster between two CP2 (3Cl.4H₂O<2CP2). The three chloride ions and water molecules are occupying six equivalent chemical (and crystallographic) positions forming a hexagonal structure, where the two subunits are aligned forming a trigonal bipyramid shaped (d) capsule. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[256, 95, 740, 831]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[139, 849, 819, 896]]<|/det|>
+<center>Figure 7. Chloride transport experiments using capsule D2 in liposomes containing lucigenin (A) or HPTS (B-D). </center>
+
+<--- Page Split --->
+<|ref|>table<|/ref|><|det|>[[137, 97, 880, 314]]<|/det|>
+
+<table><tr><td>Dimer</td><td>Anion</td><td>Equiv[α]</td><td>Molar Fraction 3A·nH2Oc2CP2[β]</td><td>NH shift (ppm)</td><td>Htriazole shift (ppm)</td><td>pka</td></tr><tr><td>D2</td><td>F-</td><td>3.13</td><td>1</td><td>10</td><td>7.38</td><td>3.2</td></tr><tr><td>D2</td><td>Cl-</td><td>24.7</td><td>0.90</td><td>8.62</td><td>7.29</td><td>-8.0</td></tr><tr><td>D2</td><td>Br[α]</td><td>28.1</td><td>0.92</td><td>8.35</td><td>7.28</td><td>-9.0</td></tr><tr><td>D2</td><td>I[α]</td><td>86</td><td>0.71</td><td>8.07</td><td>7.27</td><td>-</td></tr><tr><td>D2</td><td>N3[α]</td><td>35</td><td>0.67</td><td>8.55</td><td>7.29</td><td>4.75</td></tr><tr><td>D2</td><td>OAc-</td><td>66</td><td>0.88</td><td>8.97</td><td>7.27</td><td>4.7</td></tr><tr><td>D2</td><td>NO3[α]</td><td>60</td><td>0.77</td><td>8.19</td><td>7.30</td><td>-1.4</td></tr><tr><td>D4</td><td>F-</td><td>8</td><td>0</td><td>-</td><td>-</td><td>3.2</td></tr><tr><td>D4</td><td>Cl-</td><td>36</td><td>0</td><td>-</td><td>-</td><td>-8.0</td></tr></table>  
+
+<|ref|>text<|/ref|><|det|>[[137, 323, 860, 516]]<|/det|>
+Table 1. Key features of the anion encapsulation by tris- triazolyl modified CPs (D2 and D4); On the right: illustrative anion binding experiments derived by titrations carried out by NMR experiments. The dashed lines are used to indicate the anions in which extra water was added to facilitate complex formation. [a] equivalent number are given with respect to CP2 concentration, [b] Molar fraction was calculated at the mentioned maximum number of equivalents of the corresponding anions, [c] 7 \(\mu \mathrm{L}\) of water were added before the additions of anion solution.  
+
+<|ref|>text<|/ref|><|det|>[[137, 612, 860, 660]]<|/det|>
+[1] Davis, J. T.; Gale, P. A. & Quesada, R. Advances in anion transport and supramolecular medicinal chemistry. Chem. Soc. Rev. 49, 6056- 6086 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[137, 677, 833, 695]]<|/det|>
+[2] Rowe, S. M.; Miller, S. & Sorscher, E. J. Cystic fibrosis. N. Engl. J. Med. 352, 1992- 2001 (2005).  
+
+<|ref|>text<|/ref|><|det|>[[138, 712, 860, 758]]<|/det|>
+[3] McNaughton, D. A.; Fares, M.; Picci, G.; Gale & P. A. Caltagirone, C. Advances in fluorescent and colorimetric sensors for anionic species. Coord. Chem. Rev. 427, 213573 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[138, 776, 860, 821]]<|/det|>
+[4] Chen, L.; Berry, S. N.; Wu, X.; Howe, E. N. W. & Gale, P. A. Advances in anion receptor chemistry. Chem 6, 61- 141 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[138, 840, 860, 885]]<|/det|>
+[5] Molina, P.; Zapata, F. & Caballero, A. Anion recognition strategies based on combined noncovalent interactions. Chem. Rev. 117, 9907- 9972 (2017).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 141, 860, 188]]<|/det|>
+[6] Liu, Y.; Sengupta, A.; Raghavachari, K. & Flood, A. H. Anion binding in solution: Beyond the electrostatic regime. Chem 3, 411- 427 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[137, 206, 860, 252]]<|/det|>
+[7] Stockbridge, R.; Kolmakova- Partensky, L.; Shane, T.; Koide, A.; Koide, S.; Miller, C. & Newstead, S. Crystal structures of a double- barrelled fluoride ion channel. Nature 525, 548- 551 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[138, 270, 860, 316]]<|/det|>
+[8] Wu, X.; Wang, P.; Turner, P.; Lewis, W.; Catal, O.; Thomas, D. S. & Gale, P. A. Tetraurea macrocycles: Aggregation- driven binding of chloride in aqueous solutions. Chem 5, 1210- 1222 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[138, 334, 860, 380]]<|/det|>
+[9] Gomez, D. T.; Pratt, L. R.; Asthagiri, D. N. & Rempe, S. B. Hydrated anions: From clusters to bulk solution with quasi- chemical theory. Acc. Chem. Res. 55, 2201- 2212 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[138, 398, 850, 444]]<|/det|>
+[10] Balamurugan, K. & Pisabarro, M. T. Stabilizing Role of Water Solvation on Anion- π Interactions in Proteins. ACS Omega 6, 25350- 25360 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[138, 462, 816, 508]]<|/det|>
+[11] Varma, S.; Sabo, D. & Rempe, S. B. K+/Na+ Selectivity in K Channels and Valinomycin: Over coordination Versus Cavity- size constraints. J. Mol. Biol. 376, 13- 22 (2008).  
+
+<|ref|>text<|/ref|><|det|>[[138, 527, 848, 573]]<|/det|>
+[12] Yao, W.; Wang, K.; Wu, A.; Reed, W. F. & Gibb, B. C. Anion binding to ubiquitin and its relevance to the Hofmeister effects. Chem. Sci. 12, 320- 330 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[138, 591, 724, 608]]<|/det|>
+[13] Castleman, A. W. & Keesee, R. G. Ionic clusters. Chem, Rev. 86, 589- 618 (1986).  
+
+<|ref|>text<|/ref|><|det|>[[138, 627, 707, 644]]<|/det|>
+[14] Jungwirth, P. & Cremer, P. Beyond Hofmeister. Nat. Chem. 6, 261- 263 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[138, 662, 860, 707]]<|/det|>
+[15] Marcus, Y. Effect of ions on the structure of water: Structure making and breaking. Chem. Rev. 109, 1346- 1370 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[138, 726, 860, 772]]<|/det|>
+[16] Cremer, P. S.; Flood, A. H.; Gibb, B. C. & Mobley, D. L. Collaborative routes to clarifying the murky waters of aqueous supramolecular chemistry. Nat. Chem. 10, 8- 16 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[138, 790, 860, 836]]<|/det|>
+[17] He, Q.; Tu, P. & Sessler, J. L. Supramolecular chemistry of anionic dimers, trimers, tetramers, and clusters. Chem 4, 46- 93 (2018).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 142, 860, 217]]<|/det|>
+[18] Harris, A.; Saita, M.; Resler, T.; Hughes-Visentin, A.; Maia, R.; Pranga- Sellnau, F.: Bondar, A.- N.; Heberle, J. & Brown, L. S. Molecular details of the unique mechanism of chloride transport by a cyanobacterial rhodopsin. Phys. Chem. Chem. Phys. 20, 3184- 3199 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[138, 235, 860, 281]]<|/det|>
+[19] Sokkalingam, P.; Shraberg, J.; Rick, S. W. & Gibb, B. C. Binding hydrated anions with hydrophobic pockets. J. Am. Chem. Soc. 138, 48- 51 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[138, 299, 860, 373]]<|/det|>
+[20] Chakraborty, S.; Dutta, R.; Wong, B. M. & Ghosh, P. Anion directed conformational diversities of an arene based hexa- amide receptor and recognition of the [F4(H2O)6]4+ cluster. RSC Adv. 4, 62689- 62693 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[138, 392, 860, 438]]<|/det|>
+[21] Dariush, A. Self- assembled molecular capsule. In Synergy in Supramolecular Chemistry; Tatsuya, N., Ed.; CRC Press: Boca Raton, FL, 2015; pp 133- 148.  
+
+<|ref|>text<|/ref|><|det|>[[138, 456, 858, 502]]<|/det|>
+[22] Solomonov, A. & Shimanovich, U. Self- assembly in protein- based bionanomaterials. Israel J. Chem. 60, 1152- 1170 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[138, 520, 860, 566]]<|/det|>
+[23] Saha, R.; Mondal, B. & Mukherjee, P. S. Molecular cavity for catalysis and formation of metal nanoparticles for use in catalysis. Chem. Rev. 122, 12244- 12307 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[138, 584, 860, 658]]<|/det|>
+[24] McTernan, C. T.; Davies, J. A. & Nitschke, J. R. Beyond platonic: How to build metal- organic polyhedra capable of binding low- symmetry, information- rich molecular cargoes. Chem. Rev. 122, 10393- 10437 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[138, 677, 784, 694]]<|/det|>
+[25] Bowman- James, K. Supramolecular cages trap pesky anions. Science 365, 124- 125 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[138, 712, 860, 758]]<|/det|>
+[26] Rizzuto, F. J.; von Krbek, L. K. S. & Nitschke, J. R. Strategies for binding multiple guests in metal- organic cages. Nat. Rev. Chem. 3, 204- 222 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[138, 776, 860, 850]]<|/det|>
+[27] Rodríguez- Vázquez, N.; Amorín, M. & Granja, J. R. Recent advances in controlling the internal and external properties of self- assembling cyclic peptide nanotubes and dimers. Org. Biomol. Chem. 15, 4490- 4505 (2017).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 141, 860, 189]]<|/det|>
+[28] Song, Q.; Cheng, Z.; Kariuki, M.; Hall, S.C.L.; Hill, S.K.; Rho, J.Y. & Perrier, S. Molecular self- assembly and supramolecular chemistry of cyclic peptides. Chem. Rev. 121, 13936- 13995 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[138, 206, 860, 253]]<|/det|>
+[29] Ozores, H. L.; Amorin, M. & Granja, J. R. Self- assembling molecular capsules based on \(\alpha ,\gamma\) - cyclic peptides. J. Am. Chem. Soc. 139, 776- 784 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[138, 270, 860, 373]]<|/det|>
+[30] Pizzi, A.; Ozores, L. H.; Calvelo, M.; García- Fandiño, R.; Amorin, M.; Demitri, N.; Terraneo, G.; Bracco, S.; Comotti, A.; Sozzani, P.; Bezuidenhout, C. X.; Metrangolo, P. & Granja, J. R. Tight Xenon confinement in a crystalline sandwich- like hydrogen- bonded dimeric capsule of a cyclic peptide. Angew. Chem. Int. Ed. 58, 14472- 14476 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[138, 392, 860, 438]]<|/det|>
+[31] Fuertes, A.; Amorin, M. & Granja, J. R. Versatile symport transporters based on cyclic peptide dimers. Chem. Commun. 56, 46- 49 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[138, 456, 860, 502]]<|/det|>
+[32] Fuertes, A.; Amorin, M. & Granja, J. R. Membrane spanners based on a dimeric \(\alpha ,\beta\) - cyclic peptide core. Supramol. Chem. 32, 239- 246 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[138, 520, 860, 594]]<|/det|>
+[33] Lee, C.- H.; Lee, S.; Yoon, H. & Jang, W.- D. Strong binding affinity of a Zinc- porphyrin- based receptor for halides through the cooperative effects of quadruple C- H hydrogen bonds and axial ligation. Chem. Eur. J. 17, 13898- 13903 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[138, 612, 860, 686]]<|/det|>
+[34] Parks, F.C.; Liu, Y.; Debnath, S.; Stutsman, S. R.; Raghavachari, K. & Flood, A. H. Allosteric control of photofoldamers for selecting between anion regulation and double- to- single helix switching. J. Am. Chem. Soc. 140, 17711- 17723 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[138, 705, 860, 779]]<|/det|>
+[35] Kim, H.; Hong, K.- I.; Lee, J. H.; Kang, P.; Choi, M.- G. & Jang, W.- D. Triazole- bearing calixpyrroles: strong halide binding affinities through multiple N- H and C- H hydrogen bonds. Chem. Commun. 54, 10863- 10865 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[138, 797, 860, 844]]<|/det|>
+[36] Liu, Y.; Zhao, W.; Chen, C.- H.& Flood, A. H. Chloride capture using a C- H hydrogen- bonding cage. Science 365, 159- 161 (2019).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 141, 860, 189]]<|/det|>
+[37] Liu, Y.; Parks, F. C.; Sheetz, E. G.; Chen, C.- H. & Flood, A. H. Polarity-tolerant chloride binding in foldamer capsules by programmed solvent-exclusion. J. Am. Chem. Soc. 143, 3191–3204 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[138, 206, 860, 281]]<|/det|>
+[38] Mondal, D.; Ahmad, M.; Panwaria, P.; Upadhyay, A. & Talukdar, P. Anion recognition through multivalent C–H hydrogen bonds: Anion-induced foldamer formation and transport across phospholipid membranes. J. Org. Chem. 87, 10–17 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[138, 299, 860, 345]]<|/det|>
+[39] Fuertes, A.; Ozores, H. L.; Amorín, M. & Granja, J. R. Self-assembling Venturi-like peptide nanotubes. Nanoscale 9, 748–753 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[138, 363, 860, 438]]<|/det|>
+[40] Bock, V. D.; Perciaccante, R.; Jansen, T. P.; Hiemstra, H. & van Maarseveen, J. H. Click Chemistry as a Route to Cyclic Tetrapeptide Analogues: Synthesis of cyclo- [Pro- Val- \(\psi\) (triazole)- Pro- Tyr]. Org. Lett. 8, 919–922 (2006).  
+
+<|ref|>text<|/ref|><|det|>[[138, 456, 860, 530]]<|/det|>
+[41] Amorín, M.; Castedo, L. & Granja, J. R. New cyclic peptide assemblies with hydrophobic cavities: The structural and thermodynamic basis of a new class of peptide nanotubes. J. Am. Chem. Soc. 125, 2844–2845 (2003).  
+
+<|ref|>text<|/ref|><|det|>[[138, 548, 860, 623]]<|/det|>
+[42] Brea, R. J.; Amorín, M.; Castedo, L. & Granja, J. R. Methyl- blocked dimeric \(\alpha ,\gamma\) - peptide nanotube segments: Formation of a peptide heterodimer through backbone- backbone interactions. Angew. Chem. Int. Ed. 44, 5710–5713 (2005).  
+
+<|ref|>text<|/ref|><|det|>[[138, 641, 860, 687]]<|/det|>
+[43] Haris, P. I. & Chapman, D. The conformational analysis of peptides using Fourier Transform IR spectroscopy. Biopolymers (Pept. Sci.) 37, 251–263 (1995).  
+
+<|ref|>text<|/ref|><|det|>[[138, 705, 860, 751]]<|/det|>
+[44] Krimm, S. & Bandekar, J. in Advances in Protein Chemistry, ed. C. B. Anfinsen, J. T. Edsall and F. M. Richards, Academic Press, Orlando, FL, 1986, pp. 181–364.  
+
+<|ref|>text<|/ref|><|det|>[[138, 770, 850, 787]]<|/det|>
+[45] Barth, A. & Zscherp, C. What vibrations tell us about proteins. Q. Rev. Biophys. 35, 369–430 (2002).  
+
+<|ref|>text<|/ref|><|det|>[[138, 804, 860, 882]]<|/det|>
+[46] In- Hou, C.; Hong, Z. & Dearden V, David. Macrocyclic chemistry in the gas phase: intrinsic cation affinities and complexation rates for alkali metal cation complexes of crown ethers and glymes. J. Am. Chem. Soc. 115, 5736–5744 (1993).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 144, 860, 191]]<|/det|>
+[47] Buchanan, W. G.; Gerzain, M. & Bensimon, C. 1,4,7,10,13- Pentaoxacyclopentadecane (15- crown- 5) sodium iodide complex, \(\mathrm{C_{10}H_{20}O_5\cdot NaI}\) . Acta. Cryst. 50, 1016- 1019 (1994).  
+
+<|ref|>text<|/ref|><|det|>[[138, 205, 860, 275]]<|/det|>
+[48] Brea, R. J.; Pérez- Alvite, M. J.; Panciera, M. Mosquera, M.; Castedo, L. & Granja, J. R. Highly efficient and directional homo- and heterodimeric energy transfer materials based on fluorescently derivatized \(\alpha ,\gamma\) - cyclic octapeptides. Chem. Asian J. 6, 110- 121 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[138, 291, 860, 392]]<|/det|>
+[49] Aragay, G.; Ventura, B.; Guerra, A.; Pintre, I.; Chiorboli, C.; García- Fandiño, R.; Flamigni, L.; Granja, J. R. & Ballester, P. Self- sorting of cyclic peptide homodimers into a heterodimeric assembly featuring an efficient photoinduced intramolecular electron- transfer process. Chem. Eur. J. 20, 3427- 3438 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[138, 408, 860, 454]]<|/det|>
+[50] García- Fandiño, R.; Castedo, L.; Granja, J. R. & Vázquez, S. A. Interaction and dimerization energies in methyl- blocked \(\alpha ,\gamma\) - peptide nanotube segments. J. Phys. Chem. B 114, 4973- 4983 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[138, 471, 858, 517]]<|/det|>
+[51] Dutta, R. & Ghosh, P. Recent developments in anion induced capsular self- assemblies. Chem. Commun. 50, 10538- 10554 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[138, 535, 860, 580]]<|/det|>
+[52] Wu, X. & Gale, P. A. Measuring anion transport selectivity: a cautionary tale. Chem. Commun. 57, 3979- 3982 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[138, 600, 860, 645]]<|/det|>
+[53] Marcus, Y. Thermodynamics of solvation of ions. Part 5. Gibbs free energy of hydration at 298.15 K. J. Chem. Soc., Faraday Trans. 87, 2995- 2999 (1991).  
+
+<|ref|>text<|/ref|><|det|>[[138, 664, 860, 709]]<|/det|>
+[54] Tong, C. C.; Quesada, R.; Sessler, J. L. & Gale, P. A. meso- Octamethylcalix[4]pyrrole: an old yet new transmembrane ion- pair transporter. Chem. Commun. 6321- 6323 (2008).  
+
+<|ref|>text<|/ref|><|det|>[[138, 728, 860, 773]]<|/det|>
+[55] Kano, K. & Fendler, J. H. Pyranine as a sensitive pH probe for liposome interiors and surfaces. pH gradients across phospholipid vesicles. Biochim. Biophys. Acta 509, 289- 299 (1978).  
+
+<|ref|>text<|/ref|><|det|>[[138, 792, 860, 865]]<|/det|>
+[56] Wu, X.; Small, J. R.; Cataldo, A.; Withecombe, A. M.; Turner P. & Gale, P. A. Voltage- switchable HCl transport enabled by lipid headgroup- transporter interactions. Angew. Chem. Int. Ed. 58, 15142- 15147 (2019).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 142, 860, 191]]<|/det|>
+[57] Madhavan, N.; Robert, E. C. & Gin, M. S. A highly active anion-selective aminocyclodextrin ion channel. Angew. Chem. Int. Ed. 44, 7584–7587 (2005).  
+
+<|ref|>text<|/ref|><|det|>[[138, 207, 860, 254]]<|/det|>
+[58] Kang, B.; Tang, H.; Zhao, Z. & Song, S. Hofmeister series: Insights of ion specificity from amphiphilic assembly and interface property. ACS Omega 5, 6229–6239 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[138, 271, 837, 345]]<|/det|>
+[59] Jowett, L. A.; Howe, E. N. W.; Soto-Cerrato, V.; Van Rossom, W.; Pérez-Tomás, R. & Gale, P. A. Indole-based perenosins as highly potent HCl transporters and potential anti-cancer agents. Sci. Rep. 7, 9397 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[138, 363, 834, 410]]<|/det|>
+[60] Gilchrist, A. M. et al. Supramolecular methods: the 8- hydroxypyrene- 1,3,6- trisulfonic acid (HPTS) transport assay. Supramol. Chem. 33, 325–344 (2021).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 88, 321, 105]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[115, 118, 404, 134]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 144, 657, 160]]<|/det|>
+Reviewer #1 was asked to look over the response given to Reviewer #2]  
+
+<|ref|>text<|/ref|><|det|>[[112, 173, 840, 204]]<|/det|>
+I feel the authors have successfully responded to the reviewers' comments and the manuscript is ready for publication.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__72153a72369383baa71c48c5c9d80804cc76b376f3713b4c8e8b0481f17c1880/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__72153a72369383baa71c48c5c9d80804cc76b376f3713b4c8e8b0481f17c1880/supplementary_0_Peer Review File__72153a72369383baa71c48c5c9d80804cc76b376f3713b4c8e8b0481f17c1880.mmd

@@ -0,0 +1,470 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+RUFY3 and RUFY4 are ARL8 effectors that promote coupling of endolysosomes to dynein- dynactin  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+In this study, Keren- Kaplan et al identify 2 novel effectors of the small GTPase Arl8, namely RUFY3 and RUFY4. They use proximity labeling coupled with mitochondrial targeting (Mito ID) to identify these ARL8 interactors. Using imaging and biochemical approaches, they demonstrate that RUFY3 and 4 localize to LAMP1 and ARL8 positive vesicles and drive perinuclear clustering of these organelles. They also show using in vitro experiments that RUFY3 and 4 interact with dynein- dynactin complex. Using deletion constructs they identify the CC2 domain on RUFY3 and 4 to be responsible for ARL8 binding and driving the perinuclear clustering. RUFY3 also appears to localize to a subset of axonal LAMP1- postive organelles, in culture primary neurons. Lastly, RUFY 3 / 4 appears to be required for the juxtanuclear movement of lysosomes upon serum starvation.  
+
+The study is interesting and novel in that it identifies new effectors of ARL8 GTPase and links this GTPase to retrograde transport for the first time. Through well- designed experiments, striking images and quantitative data, the authors convey most of the points mentioned above very clearly. There are some minor issues that should be addressed prior to publication:  
+
+1. Role of the different domains (CC2 and FYVE in particular):  
+
+a) In figure 3d, it appears that RUFY3 lacking RUN or CC1 domains, while still localize to lysosomes/vesicles, are much less 'tightly' clustered to perinuclear region. This could suggest a role for these domains in supporting the retrograde movement or in interaction with retrograde transport machinery? It would be helpful to see a quantification of the strength of this clustering.  
+
+b) There seems substantial recruitment of RUFY3 to vesicles even in absence of CC2 (Fig 3d) although the perinuclear clustering is clearly reduced, while the version of protein lacking CC2 and FYVE domains is largely cytosolic. Is it more accurate to so both are involved in recruitment to vesicles? Also, can the authors explain why this looks different from the mito -experiments?  
+
+## 2. RUFY3/4 function in neurons:  
+
+a) Kymographs in Fig 5e seem to suggest that RUFY4 is on both anterogradely and retrogradely moving LAMP1 vesicles while RUFY3 is largely on retrogradely moving ones. This looks a bit different from the quantification in 5h where both RUFY3/4 look to be more on retrogradely moving vesicles.  
+
+b) Likewise, while both seem to increase relative proportion of retrogradely moving LAMP1 vesicles, there is a strong reduction of total number of motile LAMP1 axonal vesicles. This data is very interesting overall but concluding that they increase retrograde transport seems a bit simplistic. Could RUFY3/4 affect Arl8 interaction/binding with SKIP and /or Kinesin? Likewise, are there changes to pauses in
+
+<--- Page Split --->
+
+LAMP1 vesicle motion, processivity, velocity? Further dissection of how RUFY3/4 over expression alters the motility will shed important information on the transport properties of these organelles.  
+
+c) As the authors point out in their discussion RUFY3/4 localizes to a subset of axonal LAMP1-positive organelles. Given the interest in these organelles, their maturation and transport in the neuronal cell biology field, it would help to further define this sub-population: are there Rab7 positive? Are the acidic (there is a strong correlation between retrogradely moving LAMP1 vesicles in axons and their acidic nature). These experiments will help determine more clearly, the nature of these RUFY3-positive endolysosomes.  
+
+d) While the authors clearly demonstrated a reduction of LAMP1 tracks, based on over expression, from their live imaging experiments (does this include stationary vesicles?), it would be better to examine the number of endogenous LAMP1 vesicles (per unit length of axon), to conclude that there are reduced number of lysosomes in axons.  
+
+## Other minor points:  
+
+In Fig 7 e, it will be good to include Raplog in the graph's X axis as it is a bit confusing without that- it appears as if the \(+ / -\) are for the siRNA.  
+
+Reviewer #2 (Remarks to the Author):  
+
+In this study, the authors reveal a new retrograde transport pathway for late endosomes/lysosomes mediated by a small GTPase Arl8b, which has thus far only been appreciated for its role in anterograde transport of the same organelles. Specifically, Keren- Kaplan et al identify RUFY3 and 4 as the first minus- end- directed transport effectors for Arl8b. The authors show that RUFY3 interacts with Arl8b using its CC2 domain, thereby allowing endosomes to be transported towards the perinuclear area in a dynein/dynactin dependent manner. The manuscript is well structured and easy to read. Most of the conclusions drawn are substantiated by the experiments presented, and the data appears reproducible and of high quality. It is important to point out however that the novelty of the findings is threatened by a recent preprint by Kumar and colleagues (DOI: 10.21203/rs.3.rs- 345822/v1). Irrespective of this, several important issues would need to be addressed prior to publication:  
+
+## Major points  
+
+1. My biggest concern is the implicit assumption by the authors that all retrograde transport mediated by RUFY3/4 results downstream of Arl8b. This approach disregards the possibility that Rab7 may (also) be directing these actions. It is well established that a subset of late endosomes/lysosomes contains both Arl8 and Rab7, and several other Arl8 effectors, including SKIP and the HOPS complex are shared by
+
+<--- Page Split --->
+
+these GTPases. The authors should address this point, for example by examining whether Rab7 (QL versus TN) can interact with RUFY3/4.  
+
+2. The authors mention that they find that binding of Arl8 to RUFY involves the CC2 domain of RUFY3. Although the truncation data indicates necessity, the authors should include evaluation of the CC2 domain (or combined with FYVE) to demonstrate sufficiency with respect to the recruitment to Arl8b.  
+
+3. It would have been nice to see that engagement of RUFY3/4 with peripheral Arl8b-positive endosomes drives their transport into the juxtanuclear region in live cells.  
+
+## Minor points  
+
+4. In figure 2g, dCC2-FLAG and dCC2dFYVE-FLAG are not pulled down with GST-Arl8b-QL, however a band is visible for the condition where they used GST-Arl8b-TN. What could be the reason for this?  
+
+5. RUFY3 siRNAs need to be deconvoluted to show that multiple RUFY3 siRNA duplexes show similar affects.  
+
+6. Figure 4h: in the legend it is not mentioned what the yellow-colored arrows refer to.  
+
+7. Figure 7e is missing the annotation +/- Rapalog.  
+
+8. Figure 6c Annotation of GST-DLIC1 and GST-DLIC-CT is swapped.  
+
+9. Some of the clustering pictures in Fig4 are not convincing enough (compared to the quantification). For example, RUN-GFP does not look as clustered as RUFY3-GFP, while the quantifications show similar average clustering.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The manuscript by Keren- Kaplan et al. is a well written study that describes identification of novel ARL8 GTPase effectors, RUN- and FYVE- domain containing proteins, RUFY3 and RUFY4 that are important for retrograde lysosome movement in cells. The authors use a combination of cell biology and biochemical methods to show that RUFY3 and RUFY4, but not other members of the RUFY family, associate with ARL8- lysosomes. The authors also map regions in RUFY3 important for an interaction with ARL8. Furthermore, the authors provide evidence that the retrograde movement of ARL8- lysosomes is mediated by RUFY3 and RUFY4 association with the dynein complex, although direct evidence of binding to dynein is only shown for RUFY3. Together with previous studies from this group and others showing that ARL8 can associate with kinesin adaptors to move lysosomes in an anterograde direction, this work provides important insights into how ARL8 regulates movement of lysosomes in cells.
+
+<--- Page Split --->
+
+The experiments are described in sufficient detail to be replicated and statistical analysis appears appropriate. I think that this study will be of high interest to the field of cellular trafficking, and I recommend its publication in the journal of Nature Communications.  
+
+I have few minor suggestions described below:  
+
+1. In section titled: "ARL8B promotes recruitment of RUFY3 and RUFY4 to a juxtanuclear cluster of vesicles" authors conclude that ARL8 promotes the recruitment of RUFY3 and RUFY4 vesicles via the CC2 domain and that FYVE domain makes additional contributions to this recruitment. However, the contributions of RUFY4 domains are never tested. I would suggest clarifying this statement to focus only on domain contributions from RUFY3.  
+
+2. The different RUFY3-GFP truncation constructs shown in inserts in figure 4e look very cytoplasmic, which is not fully consistent with images shown in figure 3d. Is there an increase in cytoplasmic localization of these constructs when co-expressed with LAMP1? If so, this would be unexpected and should be addressed in the text. It would also be helpful if the authors included full size images in the main figure or in supplementary figure.  
+
+3. There is a typo in the legend for figure 1, panel f is labeled with an e.  
+
+4. Figure 2 panel a, please add description of what is being blotted in the input.  
+
+5. Figure 2 panel e has RUN-GFP included although there is no description of this construct anywhere in the text or figure legend. Similarly, there is RUN-FLAG included in panel f and g, but again there is no description of this construct. Please add description of these constructs.  
+
+6. In general, all labels for microscopy images are very small and hard to see. Could the authors please increase font size or rearrange labels to be above or below images, so it is easier to see them?  
+
+7. Please add reference for the source of ARL8A-B-KO cells.  
+
+Reviewer #4 (Remarks to the Author):  
+
+In this manuscript, the authors identified RUFY3.1, a previously uncharacterized longer isoform of RUFY3, and RUFY4 as novel ARL8 effectors by using the MitolD method. They then found that both RUFY3.1 and RUFY4 promote retrograde transport of lysosomes through physical interaction with the dynein-dynactin motor complex and that their activity is required for accumulation of lysosomes in the juxtanuclear area of cells. Overall, the experiments are well conducted and there is good inclusion of positive and negative controls to reveal the selectivity and specificity of the observations. Thus, the study makes a significant contribution to the mechanistic understanding of ARL8-mediated lysosomal positioning that plays important roles in a wide range of cellular processes.
+
+<--- Page Split --->
+
+Specific comments:  
+
+1. In Figure S1c, the authors showed that RUFY3.2, an originally described RUFY3 isoform, is present in the cytosol in HeLa cells, strongly suggesting that RUFY3.2 is not an ARL8 effector. In the present manuscript, however, it is not clear whether RUFY3.2 binds to ARL8. The authors should test their interaction biochemically.  
+
+2. Based on the results of Figure 3d, the authors concluded that in addition to the ARL8-interacting CC2 domain, the FYVE domain also contributes to the RYFY3.1 localization to cytoplasmic vesicles. Is RUFY3.1 localized to cytoplasmic vesicles via its FYVE domain independently of ARL8? The authors need to test the localization of RUFY3.1-deltaCC2 in ARL8A/B-KO cells.  
+
+3. In Figure 5i, expression of RUFY3-GFP or RUFY4-GFP promoted retrograde transport of lysosomes in axons but also "reduced the number of moving lysosomes". Why? Please discuss the possible reason. Does expression of RUFY3-GFP or RUFY4-GFP also affect lysosome transport in "dendrites"?  
+
+4. Distinct Rab binding activities of the RUFY family members in the Discussion section is important (lines 397-404), but the information about the Rab33A binding activity of RUFY3.2 is missing.  
+
+5. Typos.  
+
+(line 40) unless; (line 357) RUFY3.3; (line 422) Once possibility; (line 683) Fig. 3c,4b should read Fig. 3b, 4b.  
+
+Figure 6c seems to be mis-labeled. The far right lane should be the GST-DLIC1 full-length protein.  
+
+Page numbers of several references are missing (e.g., lines 973,1026, 1064, 1082, 1096, and 1114).  
+
+Reviewer #5 (Remarks to the Author):  
+
+Keren- Kaplan et al. use an alternative BiolD2 assay called MitoID to identify proximal proteins of ARL8A and ARL8B. This resulted in the identification of RUFY3 and RUFY4 proteins that bind to the GTP- bound form of ARL8. The authors further demonstrate the interaction with dynein dynactin and link juxtanuclear redistribution of lysosomes under various conditions with the complex that they have described in this study.
+
+<--- Page Split --->
+
+For the proteomics part, the design of the experiment is not very clear in the current manuscript. While the authors briefly mention in the methods section that 3 biological replicates were used (with 3 'raw' in the data files), this is not clear from the results part, nor from fig. S1A or the processed tables. It would be better to make this clear to the readers in the figures and the text. Accordingly, the data processing to obtain Figures 1B and 1C and Supplementary dataset 1 should be better explained in the manuscript. BiolD and BiolD2 data have a tendency to generate a lot of background (especially upon massive overexpression as is the case here) so careful data analysis and processing is essential and should be transparent. For example, how exactly was the abundance ratio obtained and how was this transformed to fit the 0- 100 axis? Moreover, the proteomics data should be made available in a public repository so it becomes available to the community. It is not clear whether this is planned. Along this lines, it is also worthwhile to elaborate briefly on the benefits of MitolD over classical BiolD (or BiolD2) to better motivate the use of this interesting method to the readers.  
+
+Fig 4g. shows RT- qPCR analysis upon knockdown of the RUFY3. While I appreciate the other data supporting the link between RUFY3 and lysosomal localization, perturbance is important. The authors should clarify if this was a single siRNA or a pool of siRNAs. In the case of a single siRNA, it would be best to add another siRNA to eliminate the possibility of off- target effects. In the case of a pool, the pool should be deconvoluted to find (hopefully) at least 2 decent siRNAs that give the same phenotype.  
+
+## Minor comments  
+
+1. Line 110. The originally described MitolD procedure attaches BiolD and the mitochondrial targeting sequences C-terminally from the Ras family protein under investigation. Please provide some more explanation for switching the modules in this study.  
+
+2. Line 112. Provide a reference for the ARL8 mutants that lock the protein in the GDP and GTP-bound state.  
+
+3. Line 601: I assume that TECP is referring to the reducing chemical tris (2-carboxyethyl)phosphine so this should be TCEP.  
+
+4. Methods line 710-712. Transfection method and incubation time after transfection are not mentioned.  
+
+5. Methods line 718. "Following incubation, cells were washed" 1I think this should be "beads were washed"?  
+
+6. Fig 2b. On the bottom panel, many aspecific signals can be seen with very different intensities, please indicate the band sizes expected for each FLAG-tagged construct (for example in the figure legend).  
+
+7. Line 208, Fig 3c makes use of HeLa ARL8A-B KO cells, but these cells are not mentioned anywhere in the methods. Please describe their origin or how you established them, and provide evidence of successful KO if these cells have not been published elsewhere.  
+
+8. Fig 7e. Indicate which bars refer to rapalog-treated cells. Also check RUFY4 condition (NT and DHC siRNA)
+
+<--- Page Split --->
+
+We thank the reviewers for their insightful and positive comments (italics). Changes to the text are indicated in red in the attached manuscript file. Below is a point- by- point response to the reviewers' comments:  
+
+## Reviewer #1  
+
+In this study, Keren- Kaplan et al identify 2 novel effectors of the small GTPase Arl8, namely RUIFY3 and RUIFY4. They use proximity labeling coupled with mitochondrial targeting (Mito ID) to identify these ARL8 interactors. Using imaging and biochemical approaches, they demonstrate that RUIFY3 and 4 localize to LAMP1 and ARL8 positive vesicles and drive perinuclear clustering of these organelles. They also show using in vitro experiments that RUIFY3 and 4 interact with dynein- dynactin complex. Using deletion constructs they identify the CC2 domain on RUIFY3 and 4 to be responsible for ARL8 binding and driving the perinuclear clustering. RUIFY3 also appears to localize to a subset of axonal LAMP1- positive organelles, in culture primary neurons. Lastly, RUIFY3/4 appears to be required for the juxtanuclear movement of lysosomes upon serum starvation.  
+
+The study is interesting and novel in that it identifies new effectors of ARL8 GTPase and links this GTPase to retrograde transport for the first time. Through well- designed experiments, striking images and quantitative data, the authors convey most of the points mentioned above very clearly. There are some minor issues that should be addressed prior to publication:  
+
+We thank this reviewer for the positive assessment of our manuscript. We addressed the issues mentioned by this reviewer as described below.  
+
+1. Role of the different domains (CC2 and FYVE in particular):  
+
+a) In figure 3d, it appears that RUIFY3 lacking RUN or CC1 domains, while still localize to lysosomes/vesicles, are much less 'tightly' clustered to perinuclear region. This could suggest a role for these domains in supporting the retrograde movement or in interaction with retrograde transport machinery? It would be helpful to see a quantification of the strength of this clustering.  
+
+We have revised this figure to show transfected cells with more similar levels of expression, which allows for better comparison of the localization of RUIFY3- GFP deletion constructs (new Fig. 3f). Overall, we did not see significant changes in juxtanuclear clustering of RUIFY3 vesicles upon deletion of domains other than CC2 (more dispersed) (new Fig. 3h). Importantly, in the new version of the manuscript, we have additionally quantified the association of the RUIFY3- GFP deletion constructs with vesicles (new Fig. 3g). We found that deletion of the RUN domain, alone or in combination with the CC1 domain, did not significantly alter association of RUIFY3- GFP with vesicles (new Fig. 3g). Deletion of the CC1 domain alone decreased vesicle association, though to a much lesser extent than deletion of the CC2 or FYVE domains (Fig. 3g). These results clearly show that both the CC2 and FYVE domains promote membrane association of RUIFY3 vesicles, and that the CC2 domain additionally promotes juxtanuclear clustering of the vesicles. Other domains appear to be less important for these processes.  
+
+b) There seems substantial recruitment of RUIFY3 to vesicles even in absence of CC2 (Fig 3d) although
+
+<--- Page Split --->
+
+the perinuclear clustering is clearly reduced, while the version of protein lacking CC2 and FYVE domains is largely cytosolic. Is it more accurate to so both are involved in recruitment to vesicles?  
+
+The reviewer's observations are correct. The revised images in the new Fig. 3f better show the localization of different RUFY3- GFP deletion constructs with comparable levels of expression. The \(\Delta \mathrm{CC2}\) construct exhibits partially decreased association with vesicles (new Fig. 3f, g), and these are less clustered in the juxtanuclear area (new Fig. 3f, h). The \(\Delta \mathrm{FYVE}\) construct is also less associated with vesicles (new Fig. 3f, g) and the remaining vesicles are tightly clustered in the juxtanuclear area (new Fig. 3f, h). The CC2 or FYVE domains alone are completely cytosolic (new Fig. 3f, g). We have also added a new Supplementary Fig. 2a showing that ARL8 KO decreases the clustering of full- length RUFY3 and abrogates the localization of the \(\Delta \mathrm{FYVE}\) construct to the tight juxtanuclear cluster. From these experiments, we conclude that both the CC2 and FYVE domains contribute to association with vesicles, and that the CC2 domain additionally contributes to juxtanuclear clustering. These findings are now described in more detail in the Results section. We thank the reviewer for raising this issue, which helped us provide a more precise description of the roles of the CC2 and FYVE domains.  
+
+Also, can the authors explain why this looks different from the mito- experiments?  
+
+The localizations of RUFY3/4 in Figs. 1 (mito- ID with the T34N ARL8 construct) and 3 (nonmito- ID) are not so different. If there is a subtle difference, it could be due to the fact that in the experiment in Fig. 1 cells were fixed, whereas in that in Fig. 3 cells were imaged live. The difference in the membrane recruitment of the CC2 domain alone in Figs. 2d and 3f is likely due to the overexpression of the mitochondrially targeted ARL8 in Fig. 2d. This is now explained in the text.  
+
+2. RUFY3/4 function in neurons:  
+
+a) Kymographs in Fig 5e seem to suggest that RUFY4 is on both anterogradely and retrogradely moving LAMP1 vesicles while RUFY3 is largely on retrogradely moving ones. This looks a bit different from the quantification in 5h where both RUFY3/4 look to be more on retrogradely moving vesicles.  
+
+We thank the reviewer for pointing out this discrepancy. It was due to a poor choice of the representative images. We have now replaced the kymograph analyses and chosen images that are more representative of the quantification (new Fig. 6e, f). Both the images and the quantification show that RUFY3 and RUFY4 are more abundant in retrogradely- moving vesicles.  
+
+b) Likewise, while both seem to increase relative proportion of retrogradely moving LAMP1 vesicles, there is a strong reduction of total number of motile LAMP1 axonal vesicles. This data is very interesting overall but concluding that they increase retrograde transport seems a bit simplistic. Could RUFY3/4 affect Arl8 interaction/binding with SKIP and/or Kinesin?  
+
+We agree with the reviewer that RUFY3/4 could compete with SKIP and kinesins for binding to ARL8, and we now mention this caveat in the Discussion. However, the fact that RUFY3 KD disperses endolysosomes, that RUFY3/4 interact with dynein- dynactin, and that they redistribute peroxisomes to the cell center in a dynein- dynactin- dependent manner
+
+<--- Page Split --->
+
+support their role as dynein- dynactin adaptors. We think increased retrograde transport is the most sensible interpretation for the effect of RUFY3/4 on axonal LAMP1 vesicles.  
+
+Likewise, are there changes to pauses in LAMP1 vesicle motion, processivity, velocity? Further dissection of how RUFY3/4 over expression alters the motility will shed important information on the transport properties of these organelles.  
+
+New quantifications shown in Figs. 6h, i demonstrate that RUFY3/4 have little or no effect on the velocity and run length of both anterograde and retrograde LAMP1 vesicles in the axon. The main effects are a decrease in the relative number of anterograde LAMP1 tracks and the total number of LAMP1 tracks in the axon. While we appreciate the opportunity to provide these additional data, we think that further analyses of axonal transport exceed the scope of this study and our possibilities at this time.  
+
+c) As the authors point out in their discussion RUFY3/4 localizes to a subset of axonal LAMP1-positive organelles. Given the interest in these organelles, their maturation and transport in the neuronal cell biology field, it would help to further define this sub-population: are there Rab7 positive? Are the acidic (there is a strong correlation between retrogradely moving LAMP1 vesicles in axons and their acidic nature). These experiments will help determine more clearly, the nature of these RUFY3-positive endolysosomes.  
+
+The identity of LAMP1 vesicles in the axon is currently a matter of debate, and we do not aspire to resolve this important but complex problem in this study. Nevertheless, we are happy to provide additional characterization of these vesicles in the new Supplementary Fig. 4. We find that RUFY3 and RUFY4 exhibit \(100\%\) co- localization with ARL8 in axonal vesicles (new Supplementary Fig. 4d- f). We also find that \(\sim 85\%\) RUFY3 and \(\sim 40\%\) RUFY4 are associated with LysoTracker- positive vesicles (new Supplementary Fig. 4h), and that RUFY3/4- LysoTracker- positive vesicles move mostly in the retrograde direction (new Supplementary Fig. 4i, j). These data indicate that RUFY3 is largely associated, and RUFY4 partially associated, with retrogradely- moving acidic endolysosomes. In addition, RUFY4 seems to associate with another population of non- acidic vesicles, the identity of which remains to be determined.  
+
+We did not examine the co- localization with RAB7 because this is a very complex issue that deserves its own, dedicated study, and because RUFY3 and RUFY4 do not interact with RAB7 (new data in Supplementary fig. 1e, f).  
+
+d) While the authors clearly demonstrated a reduction of LAMP1 tracks, based on over expression, from their live imaging experiments (does this include stationary vesicles?), it would be better to examine the number of endogenous LAMP1 vesicles (per unit length of axon), to conclude that there are reduced number of lysosomes in axons.  
+
+In the new Supplementary fig. 4a, b, we show that expression of RUFY3 or RUFY4 decreases the number of endogenous LAMTOR4 (a bona fide endolysosomal marker) puncta per unit length of axon. We used this particular marker because the antibody gives stronger, more specific staining. This finding is consistent with the reduction of not only moving LAMP1 tracks but also the overall number of endolysosomes in the axon.
+
+<--- Page Split --->
+
+Other minor points:  
+
+In Fig 7 e, it will be good to include Raplog in the graph's X axis as it is a bit confusing without that- it appears as if the \(+ / -\) are for the siRNA.  
+
+We fixed this issue on the figure (now Fig. 8e).  
+
+## Reviewer #2  
+
+In this study, the authors reveal a new retrograde transport pathway for late endosomes/lysosomes mediated by a small GTPase Arl8b, which has thus far only been appreciated for its role in anterograde transport of the same organelles. Specifically, Keren- Kaplan et al identify RUY3 and 4 as the first minus- end- directed transport effectors for Arl8b. The authors show that RUY3 interacts with Arl8b using its CC2 domain, thereby allowing endosomes to be transported towards the perinuclear area in a dynein/dynactin dependent manner. The manuscript is well structured and easy to read. Most of the conclusions drawn are substantiated by the experiments presented, and the data appears reproducible and of high quality. It is important to point out however that the novelty of the findings is threatened by a recent preprint by Kumar and colleagues (DOI: 10.21203/rs.3.rs- 345822/v1). Irrespective of this, several important issues would need to be addressed prior to publication:  
+
+We thank this reviewer for his positive comments on our manuscript.  
+
+## Major points  
+
+1. My biggest concern is the implicit assumption by the authors that all retrograde transport mediated by RUY3/4 results downstream of Arl8b. This approach disregards the possibility that Rab7 may (also) be directing these actions. It is well established that a subset of late endosomes/lysosomes contains both Arl8 and Rab7, and several other Arl8 effectors, including SKIP and the HOPS complex are shared by these GTPases. The authors should address this point, for example by examining whether Rab7 (QL versus TN) can interact with RUY3/4.  
+
+In the new Supplementary fig. 1e, f, we show that neither Q67L nor T22N forms of mito- . RAB7A relocate RUY3 and RUY4 to mitochondria. This is in contrast to the mitochondrial relocation of a known RAB7A effector, RILP, by the Q67L construct (positive control). This demonstrates that RUY3/4 are not RAB7 effectors. In light of these results, we did not further pursue a possible relationship of RUY3/4 with RAB7A.  
+
+2. The authors mention that they find that binding of Arl8 to RUY involves the CC2 domain of RUY3. Although the truncation data indicates necessity, the authors should include evaluation of the CC2 domain (or combined with FYVE) to demonstrate sufficiency with respect to the recruitment to Arl8b.  
+
+We thank the reviewer for raising this important issue. We now provide new mito- targeting (new Fig. 2d, e) and pulldown (new Fig. 2h, i) data showing that the CC2 but not FYVE domain binds the active form of ARL8.  
+
+3. It would have been nice to see that engagement of RUY3/4 with peripheral Arl8b-positive endosomes drives their transport into the juxtanuclear region in live cells.
+
+<--- Page Split --->
+
+While we acknowledge the importance of the suggested experiment, it would be difficult to perform in non- neuronal cells because RUFY3/4 causes strong clustering of endolysosomes in the juxtanuclear area. We think that our experiments in neurons provide evidence for an increase in the ratio of retrograde vs. anterograde LAMP1 vesicles (Fig. 6e, f), consistent with RUFY3/4 promoting retrograde transport.  
+
+## Minor points  
+
+4. In figure 2g, dCC2-FLAG and dCC2dFYVE-FLAG are not pulled down with GST-Arl8b-QL, however a band is visible for the condition where they used GST-Arl8b-TN. What could be the reason for this?  
+
+This is a faint non- specific band that is seen in overexposed blots. For better appreciation of the relative intensity of this non- specific band vs. specific bands, in the new Fig. 2g we show a shorter exposure of the blots. In any case, both the long and short exposures are shown in the source data file.  
+
+5. RUFY3 siRNAs need to be deconvoluted to show that multiple RUFY3 siRNA duplexes show similar affects.  
+
+In the new Supplementary Fig. 3, we show that 3 of the 4 single siRNAs in the original SMARTpool effectively knock down RUFY3, and that the 3 effective siRNAs cause peripheral dispersal of endolysosomes.  
+
+6. Figure 4h: in the legend it is not mentioned what the yellow-colored arrows refer to.  
+
+We now indicate in the text that the yellow arrows point to LAMP1 structures at cell vertices (new Fig. 5d).  
+
+7. Figure 7e is missing the annotation +/- Rapalog.  
+
+This was fixed in the new version of Fig. 7e, now Fig. 8e.  
+
+8. Figure 6c Annotation of GST-DLIC1 and GST-DLIC-CT is swapped.  
+
+This was fixed in the new version of Fig. 6c, now Fig. 7c.  
+
+9. Some of the clustering pictures in Fig4 are not convincing enough (compared to the quantification). For example, \(\Delta\) RUN-GFP does not look as clustered as RUFY3-GFP, while the quantifications show similar average clustering.  
+
+We have replaced the image for \(\Delta\) RUN by another image that is more representative of the quantitative data (new Fig. 5a). Both the images and the quantification suggest that this construct is a bit less effective at clustering endolysosomes, although the differences do not rise to the level of statistical significance.  
+
+Reviewer #3
+
+<--- Page Split --->
+
+The manuscript by Keren- Kaplan et al. is a well written study that describes identification of novel ARL8 GTPase effectors, RUN- and FYVE- domain containing proteins, RUFY3 and RUFY4 that are important for retrograde lysosome movement in cells. The authors use a combination of cell biology and biochemical methods to show that RUFY3 and RUFY4, but not other members of the RUFY family, associate with ARL8- lysosomes. The authors also map regions in RUFY3 important for an interaction with ARL8. Furthermore, the authors provide evidence that the retrograde movement of ARL8- lysosomes is mediated by RUFY3 and RUFY4 association with the dynein complex, although direct evidence of binding to dynein is only shown for RUFY3. Together with previous studies from this group and others showing that ARL8 can associate with kinesin adaptors to move lysosomes in an anterograde direction, this work provides important insights into how ARL8 regulates movement of lysosomes in cells.  
+
+The experiments are described in sufficient detail to be replicated and statistical analysis appears appropriate. I think that this study will be of high interest to the field of cellular trafficking, and I recommend its publication in the journal of Nature Communications.  
+
+We also thank this reviewer for the positive comments on our manuscript.  
+
+I have few minor suggestions described below:  
+
+1. In section titled: "ARL8B promotes recruitment of RUFY3 and RUFY4 to a juxtanuclear cluster of vesicles" authors conclude that ARL8 promotes the recruitment of RUFY3 and RUFY4 vesicles via the CC2 domain and that FYVE domain makes additional contributions to this recruitment. However, the contributions of RUFY4 domains are never tested. I would suggest clarifying this statement to focus only on domain contributions from RUFY3.  
+
+We thank the review for pointing out this misstatement. We have modified the text to refer only to domains of RUFY3.  
+
+2. The different RUFY3-GFP truncation constructs shown in inserts in figure 4e look very cytoplasmic, which is not fully consistent with images shown in figure 3d. Is there an increase in cytoplasmic localization of these constructs when co-expressed with LAMP1? If so, this would be unexpected and should be addressed in the text. It would also be helpful if the authors included full size images in the main figure or in supplementary figure.  
+
+The images in the insets of Fig. 4a look more cytosolic because 1) the constructs are overexpressed in order to maximize their effects on endolysosomes, and 2) the cells were fixed before staining, in contrast to those in Fig. 3d (now Fig. 3f), which were imaged live. We now clarify these differences in the text and figure legends.  
+
+3. There is a typo in the legend for figure 1, panel f is labeled with an e.  
+
+Fixed.  
+
+4. Figure 2 panel a, please add description of what is being blotted in the input.  
+
+We now indicate in the figure that the input is immunoblotted for the FLAG epitope.  
+
+5. Figure 2 panel e has RUN-GFP included although there is no description of this construct anywhere in
+
+<--- Page Split --->
+
+the text or figure legend. Similarly, there is RUN- FLAG included in panel f and g, but again there is no description of this construct. Please add description of these constructs.  
+
+We now describe in the text the results with the RUN- FLAG construct in reference to Fig. 2f, 8.  
+
+6. In general, all labels for microscopy images are very small and hard to see. Could the authors please increase font size or rearrange labels to be above or below images, so it is easier to see them?  
+
+Whenever possible, we have increased the font size of the labels. In some cases, it was not possible for reasons of space.  
+
+7. Please add reference for the source of ARL8A- B- KO cells.  
+
+We added a reference for these cells (Keren- Kaplan and Bonifacino, 2021, Curr. Biol. 31, 540- 554 e545).  
+
+## Reviewer #4 (Remarks to the Author)  
+
+In this manuscript, the authors identified RUFY3.1, a previously uncharacterized longer isoform of RUFY3, and RUFY4 as novel ARL8 effectors by using the MitoID method. They then found that both RUFY3.1 and RUFY4 promote retrograde transport of lysosomes through physical interaction with the dynein- dynactin motor complex and that their activity is required for accumulation of lysosomes in the juxtanuclear area of cells. Overall, the experiments are well conducted and there is good inclusion of positive and negative controls to reveal the selectivity and specificity of the observations. Thus, the study makes a significant contribution to the mechanistic understanding of ARL8- mediated lysosomal positioning that plays important roles in a wide range of cellular processes.  
+
+We thank this reviewer for the positive comments on our study.  
+
+## Specific comments:  
+
+1. In Figure S1c, the authors showed that RUFY3.2, an originally described RUFY3 isoform, is present in the cytosol in HeLa cells, strongly suggesting that RUFY3.2 is not an ARL8 effector. In the present manuscript, however, it is not clear whether RUFY3.2 binds to ARL8. The authors should test their interaction biochemically.  
+
+We have now added mitochondrial relocalization and pulldown analyses for RUFY3.1 vs RUFY3.2 (new Supplementary fig. 1c,d,g,h). The results show that the shorter RUFY3.2 isoform, lacking the FYVE domain and part of the CC2 domain, does not interact with ARL8.  
+
+2. Based on the results of Figure 3d, the authors concluded that in addition to the ARL8-interacting CC2 domain, the FYVE domain also contributes to the RYFY3.1 localization to cytoplasmic vesicles. Is RUFY3.1 localized to cytoplasmic vesicles via its FYVE domain independently of ARL8? The authors need to test the localization of RUFY3.1-deltaCC2 in ARL8A/B-KO cells.
+
+<--- Page Split --->
+
+We thank the reviewer for suggesting this very important experiment. In the new Supplementary fig. 2a, we show that ARL8 KO reduces overall vesicular staining for RUFY3- GFP and the juxtanuclear clustering of RUFY3- GFP vesicles, does not change the already dispersed distribution of the \(\Delta \mathrm{CC2}\) construct, and abrogates the association of the \(\Delta \mathrm{FYVE}\) construct with vesicles. These findings confirm that the ARL8- binding CC2 domain is important for juxtanuclear clustering, while both the CC2 and the FYVE domain contribute to vesicle recruitment.  
+
+3. In Figure 5i, expression of RUFY3-GFP or RUFY4-GFP promoted retrograde transport of lysosomes in axons but also "reduced the number of moving lysosomes". Why? Please discuss the possible reason. Does expression of RUFY3-GFP or RUFY4-GFP also affect lysosome transport in "dendrites"?  
+
+In the new version of the manuscript, we explain more clearly that expression of RUFY3- GFP or RUFY4- GFP both reduces the total number of moving LAMP1 vesicles, and shifts the balance of the moving LAMP1 vesicles towards retrograde transport. This is consistent with RUFY3- GFP or RUFY4- GFP promoting retrograde over anterograde transport. As in non- neuronal cells, the reduction in LAMP1 tracks and LAMTOR4 puncta in the axon (new Fig. 6g and Supplementary Fig. 4a,b; see response to Reviewer #1, point d) likely results from accumulation of endolysosomes in the juxtanuclear area of the neurons. We now make this explanation clearer in the text.  
+
+RUFY3- FLAG and RUFY4- FLAG also colocalize with LAMP1- GFP and endogenous LAMTOR4 in dendrites (Fig. 6a- d). However, because dendrites have mixed microtubule polarity, it is harder to assess whether RUFY3- FLAG and RUFY4- FLAG alter the distribution or movement of lysosomes. For this reason, we limited our imaging of moving endolysosomes to axons, where microtubules are uniformly polarized and it is easier to assess anterograde vs. retrograde movement.  
+
+4. Distinct Rab binding activities of the RUFY family members in the Discussion section is important (lines 397-404), but the information about the Rab33A binding activity of RUFY3.2 is missing.  
+
+In the new version of the Discussion, we mention the fact that RUFY3.2 interacts with RAB33, and point out that the significance of this interaction for recruitment of RUFY3.2 to endolysosomes remains to be addressed.  
+
+5. Typos. (line 40) unles; (line 357) RUFY3.3; (line 422) Once possibility; (line 683) Fig. 3c,4b should read Fig. 3b, 4b.  
+
+Figure 6c seems to be mis- labeled. The far- right lane should be the GST- DLIC1 full- length protein.  
+
+Page numbers of several references are missing (e.g., lines 973,1026, 1064, 1082, 1096, and 1114).  
+
+All of these typos were fixed.  
+
+Reviewer #5:
+
+<--- Page Split --->
+
+Keren- Kaplan et al. use an alternative BioID2 assay called MitoID to identify proximal proteins of ARL8A and ARL8B. This resulted in the identification of RUFY3 and RUFY4 proteins that bind to the GTP- bound form of ARL8. The authors further demonstrate the interaction with dynein dynactin and link juxtanuclear redistribution of lysozymes under various conditions with the complex that they have described in this study.  
+
+For the proteomics part, the design of the experiment is not very clear in the current manuscript. While the authors briefly mention in the methods section that 3 biological replicates were used (with 3 'raw' in the data files), this is not clear from the results part, nor from fig. S1A or the processed tables. It would be better to make this clear to the readers in the figures and the text. Accordingly, the data processing to obtain Figures 1B and 1C and Supplementary dataset 1 should be better explained in the manuscript. BioID and BioID2 data tend to generate a lot of background (especially upon massive overexpression as is the case here) so careful data analysis and processing is essential and should be transparent. For example, how exactly was the abundance ratio obtained and how was this transformed to fit the 0- 100 axis?  
+
+We now provide a more detailed description of the MitoID and mass spectrometry procedures in the legend to Fig. 1 and the Methods sections. This includes a statement in both the Fig. 1 legend and Methods that 3 biological replicates were used.  
+
+MitoID was chosen precisely because the targeting of the bait constructs to mitochondria provides for a more uniform background of non- specific hits, irrespective of the original localization of the baits. This is now mentioned in the first paragraph of the Results section.  
+
+The default value of 100 was set as the maximum fold change allowed. For example, if the calculated ratios are 50, 80, 120 and 150 for proteins A, B, C, and D, the ratios reported by PD software are 50, 80, 100, 100 for proteins A, B, C and D. The abundance ratio was not transformed in any way. It was plotted on the graph as presented in the Excel file.  
+
+Moreover, the proteomics data should be made available in a public repository, so it becomes available to the community. It is not clear whether this is planned.  
+
+Proteomics raw data and search results were deposited in the MassIVE repository and will be available upon publication: Title: Tal Keren- Kaplan and Juan Bonifacino LFQ data, link: ftp://massive.ucds.edu/MSV000087741/.  
+
+An Excel spreadsheet with all the proteins identified in the study is also included as Supplementary Data 1.  
+
+Along these lines, it is also worthwhile to elaborate briefly on the benefits of MitoID over classical BioID (or BioID2) to better motivate the use of this interesting method to the readers.  
+
+In the Results section (under section "Identification of RUFY3 and RUFY4 as ARL8 effectors"), we now mention that one advantage of MitoID over BioID is a more uniform identification of non- specific proteins due to the targeting of all constructs to mitochondria, irrespective of what compartment they normally associate with in the absence of the mitochondrial- targeting signal (e.g., endolysosomes vs. cytosol).  
+
+Fig 4g. shows RT- qPCR analysis upon knockdown of the RUFY3. While I appreciate the other data supporting the link between RUFY3 and lysosomal localization, perturbance is important. The authors
+
+<--- Page Split --->
+
+should clarify if this was a single siRNA or a pool of siRNAs. In the case of a single siRNA, it would be best to add another siRNA to eliminate the possibility of off- target effects. In the case of a pool, the pool should be deconvoluted to find (hopefully) at least 2 decent siRNAs that give the same phenotype.  
+
+We thank the reviewer for this comment. We now indicate in the legend to the new Fig. 5 that RUY3 was knocked down using a SMARTpool. In addition, in the new Supplementary Fig. 3, we show RUY3 KD and LAMP1 redistribution data for single siRNAs from the SMARTpool. Three of the 4 siRNAs are effective in knocking down RUY3 and causing endolysosome dispersal.  
+
+## Minor comments  
+
+1. Line 110. The originally described MitoID procedure attaches BioID and the mitochondrial targeting sequences C-terminally from the Ras family protein under investigation. Please provide some more explanation for switching the modules in this study.  
+
+The RAB GTPases used in the original study are anchored to membranes via their C- terminus. In contrast, ARL GTPases, including the ARL8 used in our study, are anchored to membranes via their N-terminus, thus the different design of our constructs. We now briefly explain this in the first paragraph of the results and the corresponding section of the Methods (Recombinant DNAs).  
+
+2. Line 112. Provide a reference for the ARL8 mutants that lock the protein in the GDP and GTP-bound state.  
+
+We added a reference for these mutations.  
+
+3. Line 601: I assume that TECP is referring to the reducing chemical tris (2-carboxyethyl)phosphine so this should be TCEP.  
+
+We corrected the abbreviation to TCEP and explained its meaning (Tris(2- carboxyethyl)phosphine).  
+
+4. Methods line 710-712. Transfection method and incubation time after transfection are not mentioned.  
+
+The transfection protocol was described in the "Cell culture and treatments" section.  
+
+5. Methods line 718. "Following incubation, cells were washed" à I think this should be "beads were washed"?  
+
+Corrected.  
+
+6. Fig 2b. On the bottom panel, many aspecific signals can be seen with very different intensities, please indicate the band sizes expected for each FLAG-tagged construct (for example in the figure legend)  
+
+We added asterisks indicating the undegraded proteins and included the expected molecular masses in the figure legend.
+
+<--- Page Split --->
+
+7. Line 208, Fig 3c makes use of HeLa ARL8A-B KO cells, but these cells are not mentioned anywhere in the methods. Please describe their origin or how you established them and provide evidence of successful KO if these cells have not been published elsewhere.  
+
+The HeLa ARL8A- B KO cells were described previously (Keren- Kaplan and Bonifacino, 2021, Curr. Biol. 31, 540- 554 e545). We have added the reference to the text.  
+
+8. Fig 7e. Indicate which bars refer to rapalog-treated cells. Also check RUIFY4 condition (NT and DHC siRNA)  
+
+We modified Fig. 7e (now Fig. 8e) to indicate the rapalog- treated cells. We also corrected the labeling error for RUIFY4.
+
+<--- Page Split --->
+
+REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The study by Keren- Kaplan et al identifying 2 novel effectors of the small GTPase Arl8, namely RUFY3 and RUFY4 is interesting and novel in that it identifies these new effectors of ARL8 GTPase and links this GTPase to retrograde transport for the first time. As mentioned before in initial review the topic is important and the study itself is very well done with well- designed experiments, striking images and quantitative data. The authors have addressed the minor issues raised in initial review with additional experiments, quantitative data they have included and/or clarifications in text. I recommend publication of this interesting study.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors addressed my points  
+
+Reviewer #3 (Remarks to the Author):  
+
+The authors have addressed all my comments.  
+
+Reviewer #4 (Remarks to the Author):  
+
+In the revised manuscript, the authors properly addressed all of the concerns raised by this reviewer. Thus, I would like to recommend the revised manuscript for publication in its present form.
+
+<--- Page Split --->
+
+Reviewer #5 (Remarks to the Author):  
+
+Thank you for introducing clarifications in the manuscript. Please rephrase 'using four pooled siRNAs' to 'using a pool of four siRNAs'.  
+
+Manuscript has further improved and can be accepted now.
+
+<--- Page Split --->
+
+Responses to Reviewers  
+
+Authors' responses are indicated in red.  
+
+Reviewer #1 (Remarks to the Author):  
+
+The study by Keren- Kaplan et al identifying 2 novel effectors of the small GTPase Arl8, namely RUY3 and RUY4 is interesting and novel in that it identifies these new effectors of ARL8 GTPase and links this GTPase to retrograde transport for the first time. As mentioned before in initial review the topic is important and the study itself is very well done with well- designed experiments, striking images and quantitative data. The authors have addressed the minor issues raised in initial review with additional experiments, quantitative data they have included and/or clarifications in text. I recommend publication of this interesting study.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors addressed my points  
+
+Reviewer #3 (Remarks to the Author):  
+
+The authors have addressed all my comments.  
+
+Reviewer #4 (Remarks to the Author):  
+
+In the revised manuscript, the authors properly addressed all of the concerns raised by this reviewer. Thus, I would like to recommend the revised manuscript for publication in its present form.  
+
+We are pleased to have been able to address all the comments by Reviewers #1- 4.  
+
+Reviewer #5 (Remarks to the Author):  
+
+Thank you for introducing clarifications in the manuscript. Please rephrase 'using four pooled siRNAs' to 'using a pool of four siRNAs'.  
+
+Manuscript has further improved and can be accepted now.  
+
+The text was changed as requested by this reviewer. We thank the review for recommending acceptance.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__72153a72369383baa71c48c5c9d80804cc76b376f3713b4c8e8b0481f17c1880/supplementary_0_Peer Review File__72153a72369383baa71c48c5c9d80804cc76b376f3713b4c8e8b0481f17c1880_det.mmd

@@ -0,0 +1,663 @@
+<|ref|>title<|/ref|><|det|>[[100, 40, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[107, 110, 373, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[108, 161, 893, 220]]<|/det|>
+RUFY3 and RUFY4 are ARL8 effectors that promote coupling of endolysosomes to dynein- dynactin  
+
+<|ref|>text<|/ref|><|det|>[[272, 732, 876, 784]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 150, 393, 166]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 203, 881, 366]]<|/det|>
+In this study, Keren- Kaplan et al identify 2 novel effectors of the small GTPase Arl8, namely RUFY3 and RUFY4. They use proximity labeling coupled with mitochondrial targeting (Mito ID) to identify these ARL8 interactors. Using imaging and biochemical approaches, they demonstrate that RUFY3 and 4 localize to LAMP1 and ARL8 positive vesicles and drive perinuclear clustering of these organelles. They also show using in vitro experiments that RUFY3 and 4 interact with dynein- dynactin complex. Using deletion constructs they identify the CC2 domain on RUFY3 and 4 to be responsible for ARL8 binding and driving the perinuclear clustering. RUFY3 also appears to localize to a subset of axonal LAMP1- postive organelles, in culture primary neurons. Lastly, RUFY 3 / 4 appears to be required for the juxtanuclear movement of lysosomes upon serum starvation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 377, 880, 449]]<|/det|>
+The study is interesting and novel in that it identifies new effectors of ARL8 GTPase and links this GTPase to retrograde transport for the first time. Through well- designed experiments, striking images and quantitative data, the authors convey most of the points mentioned above very clearly. There are some minor issues that should be addressed prior to publication:  
+
+<|ref|>text<|/ref|><|det|>[[115, 460, 562, 477]]<|/det|>
+1. Role of the different domains (CC2 and FYVE in particular):  
+
+<|ref|>text<|/ref|><|det|>[[115, 488, 869, 561]]<|/det|>
+a) In figure 3d, it appears that RUFY3 lacking RUN or CC1 domains, while still localize to lysosomes/vesicles, are much less 'tightly' clustered to perinuclear region. This could suggest a role for these domains in supporting the retrograde movement or in interaction with retrograde transport machinery? It would be helpful to see a quantification of the strength of this clustering.  
+
+<|ref|>text<|/ref|><|det|>[[115, 600, 874, 672]]<|/det|>
+b) There seems substantial recruitment of RUFY3 to vesicles even in absence of CC2 (Fig 3d) although the perinuclear clustering is clearly reduced, while the version of protein lacking CC2 and FYVE domains is largely cytosolic. Is it more accurate to so both are involved in recruitment to vesicles? Also, can the authors explain why this looks different from the mito -experiments?  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 712, 350, 728]]<|/det|>
+## 2. RUFY3/4 function in neurons:  
+
+<|ref|>text<|/ref|><|det|>[[115, 739, 870, 793]]<|/det|>
+a) Kymographs in Fig 5e seem to suggest that RUFY4 is on both anterogradely and retrogradely moving LAMP1 vesicles while RUFY3 is largely on retrogradely moving ones. This looks a bit different from the quantification in 5h where both RUFY3/4 look to be more on retrogradely moving vesicles.  
+
+<|ref|>text<|/ref|><|det|>[[115, 805, 878, 877]]<|/det|>
+b) Likewise, while both seem to increase relative proportion of retrogradely moving LAMP1 vesicles, there is a strong reduction of total number of motile LAMP1 axonal vesicles. This data is very interesting overall but concluding that they increase retrograde transport seems a bit simplistic. Could RUFY3/4 affect Arl8 interaction/binding with SKIP and /or Kinesin? Likewise, are there changes to pauses in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 872, 126]]<|/det|>
+LAMP1 vesicle motion, processivity, velocity? Further dissection of how RUFY3/4 over expression alters the motility will shed important information on the transport properties of these organelles.  
+
+<|ref|>text<|/ref|><|det|>[[114, 136, 876, 245]]<|/det|>
+c) As the authors point out in their discussion RUFY3/4 localizes to a subset of axonal LAMP1-positive organelles. Given the interest in these organelles, their maturation and transport in the neuronal cell biology field, it would help to further define this sub-population: are there Rab7 positive? Are the acidic (there is a strong correlation between retrogradely moving LAMP1 vesicles in axons and their acidic nature). These experiments will help determine more clearly, the nature of these RUFY3-positive endolysosomes.  
+
+<|ref|>text<|/ref|><|det|>[[114, 255, 879, 328]]<|/det|>
+d) While the authors clearly demonstrated a reduction of LAMP1 tracks, based on over expression, from their live imaging experiments (does this include stationary vesicles?), it would be better to examine the number of endogenous LAMP1 vesicles (per unit length of axon), to conclude that there are reduced number of lysosomes in axons.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 397, 261, 412]]<|/det|>
+## Other minor points:  
+
+<|ref|>text<|/ref|><|det|>[[115, 424, 852, 460]]<|/det|>
+In Fig 7 e, it will be good to include Raplog in the graph's X axis as it is a bit confusing without that- it appears as if the \(+ / -\) are for the siRNA.  
+
+<|ref|>text<|/ref|><|det|>[[115, 527, 393, 544]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 584, 881, 766]]<|/det|>
+In this study, the authors reveal a new retrograde transport pathway for late endosomes/lysosomes mediated by a small GTPase Arl8b, which has thus far only been appreciated for its role in anterograde transport of the same organelles. Specifically, Keren- Kaplan et al identify RUFY3 and 4 as the first minus- end- directed transport effectors for Arl8b. The authors show that RUFY3 interacts with Arl8b using its CC2 domain, thereby allowing endosomes to be transported towards the perinuclear area in a dynein/dynactin dependent manner. The manuscript is well structured and easy to read. Most of the conclusions drawn are substantiated by the experiments presented, and the data appears reproducible and of high quality. It is important to point out however that the novelty of the findings is threatened by a recent preprint by Kumar and colleagues (DOI: 10.21203/rs.3.rs- 345822/v1). Irrespective of this, several important issues would need to be addressed prior to publication:  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 807, 210, 823]]<|/det|>
+## Major points  
+
+<|ref|>text<|/ref|><|det|>[[115, 834, 883, 906]]<|/det|>
+1. My biggest concern is the implicit assumption by the authors that all retrograde transport mediated by RUFY3/4 results downstream of Arl8b. This approach disregards the possibility that Rab7 may (also) be directing these actions. It is well established that a subset of late endosomes/lysosomes contains both Arl8 and Rab7, and several other Arl8 effectors, including SKIP and the HOPS complex are shared by
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 841, 125]]<|/det|>
+these GTPases. The authors should address this point, for example by examining whether Rab7 (QL versus TN) can interact with RUFY3/4.  
+
+<|ref|>text<|/ref|><|det|>[[115, 136, 859, 191]]<|/det|>
+2. The authors mention that they find that binding of Arl8 to RUFY involves the CC2 domain of RUFY3. Although the truncation data indicates necessity, the authors should include evaluation of the CC2 domain (or combined with FYVE) to demonstrate sufficiency with respect to the recruitment to Arl8b.  
+
+<|ref|>text<|/ref|><|det|>[[115, 202, 800, 237]]<|/det|>
+3. It would have been nice to see that engagement of RUFY3/4 with peripheral Arl8b-positive endosomes drives their transport into the juxtanuclear region in live cells.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 277, 212, 293]]<|/det|>
+## Minor points  
+
+<|ref|>text<|/ref|><|det|>[[115, 304, 847, 340]]<|/det|>
+4. In figure 2g, dCC2-FLAG and dCC2dFYVE-FLAG are not pulled down with GST-Arl8b-QL, however a band is visible for the condition where they used GST-Arl8b-TN. What could be the reason for this?  
+
+<|ref|>text<|/ref|><|det|>[[115, 351, 850, 386]]<|/det|>
+5. RUFY3 siRNAs need to be deconvoluted to show that multiple RUFY3 siRNA duplexes show similar affects.  
+
+<|ref|>text<|/ref|><|det|>[[115, 397, 750, 415]]<|/det|>
+6. Figure 4h: in the legend it is not mentioned what the yellow-colored arrows refer to.  
+
+<|ref|>text<|/ref|><|det|>[[115, 426, 482, 443]]<|/det|>
+7. Figure 7e is missing the annotation +/- Rapalog.  
+
+<|ref|>text<|/ref|><|det|>[[115, 454, 602, 472]]<|/det|>
+8. Figure 6c Annotation of GST-DLIC1 and GST-DLIC-CT is swapped.  
+
+<|ref|>text<|/ref|><|det|>[[115, 483, 879, 537]]<|/det|>
+9. Some of the clustering pictures in Fig4 are not convincing enough (compared to the quantification). For example, RUN-GFP does not look as clustered as RUFY3-GFP, while the quantifications show similar average clustering.  
+
+<|ref|>text<|/ref|><|det|>[[115, 605, 393, 621]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 660, 883, 844]]<|/det|>
+The manuscript by Keren- Kaplan et al. is a well written study that describes identification of novel ARL8 GTPase effectors, RUN- and FYVE- domain containing proteins, RUFY3 and RUFY4 that are important for retrograde lysosome movement in cells. The authors use a combination of cell biology and biochemical methods to show that RUFY3 and RUFY4, but not other members of the RUFY family, associate with ARL8- lysosomes. The authors also map regions in RUFY3 important for an interaction with ARL8. Furthermore, the authors provide evidence that the retrograde movement of ARL8- lysosomes is mediated by RUFY3 and RUFY4 association with the dynein complex, although direct evidence of binding to dynein is only shown for RUFY3. Together with previous studies from this group and others showing that ARL8 can associate with kinesin adaptors to move lysosomes in an anterograde direction, this work provides important insights into how ARL8 regulates movement of lysosomes in cells.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 827, 143]]<|/det|>
+The experiments are described in sufficient detail to be replicated and statistical analysis appears appropriate. I think that this study will be of high interest to the field of cellular trafficking, and I recommend its publication in the journal of Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[115, 183, 459, 199]]<|/det|>
+I have few minor suggestions described below:  
+
+<|ref|>text<|/ref|><|det|>[[115, 211, 881, 301]]<|/det|>
+1. In section titled: "ARL8B promotes recruitment of RUFY3 and RUFY4 to a juxtanuclear cluster of vesicles" authors conclude that ARL8 promotes the recruitment of RUFY3 and RUFY4 vesicles via the CC2 domain and that FYVE domain makes additional contributions to this recruitment. However, the contributions of RUFY4 domains are never tested. I would suggest clarifying this statement to focus only on domain contributions from RUFY3.  
+
+<|ref|>text<|/ref|><|det|>[[115, 313, 860, 402]]<|/det|>
+2. The different RUFY3-GFP truncation constructs shown in inserts in figure 4e look very cytoplasmic, which is not fully consistent with images shown in figure 3d. Is there an increase in cytoplasmic localization of these constructs when co-expressed with LAMP1? If so, this would be unexpected and should be addressed in the text. It would also be helpful if the authors included full size images in the main figure or in supplementary figure.  
+
+<|ref|>text<|/ref|><|det|>[[115, 414, 635, 430]]<|/det|>
+3. There is a typo in the legend for figure 1, panel f is labeled with an e.  
+
+<|ref|>text<|/ref|><|det|>[[115, 442, 693, 459]]<|/det|>
+4. Figure 2 panel a, please add description of what is being blotted in the input.  
+
+<|ref|>text<|/ref|><|det|>[[115, 471, 873, 523]]<|/det|>
+5. Figure 2 panel e has RUN-GFP included although there is no description of this construct anywhere in the text or figure legend. Similarly, there is RUN-FLAG included in panel f and g, but again there is no description of this construct. Please add description of these constructs.  
+
+<|ref|>text<|/ref|><|det|>[[115, 535, 864, 570]]<|/det|>
+6. In general, all labels for microscopy images are very small and hard to see. Could the authors please increase font size or rearrange labels to be above or below images, so it is easier to see them?  
+
+<|ref|>text<|/ref|><|det|>[[115, 582, 547, 598]]<|/det|>
+7. Please add reference for the source of ARL8A-B-KO cells.  
+
+<|ref|>text<|/ref|><|det|>[[115, 697, 393, 712]]<|/det|>
+Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 752, 868, 898]]<|/det|>
+In this manuscript, the authors identified RUFY3.1, a previously uncharacterized longer isoform of RUFY3, and RUFY4 as novel ARL8 effectors by using the MitolD method. They then found that both RUFY3.1 and RUFY4 promote retrograde transport of lysosomes through physical interaction with the dynein-dynactin motor complex and that their activity is required for accumulation of lysosomes in the juxtanuclear area of cells. Overall, the experiments are well conducted and there is good inclusion of positive and negative controls to reveal the selectivity and specificity of the observations. Thus, the study makes a significant contribution to the mechanistic understanding of ARL8-mediated lysosomal positioning that plays important roles in a wide range of cellular processes.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 118, 259, 134]]<|/det|>
+Specific comments:  
+
+<|ref|>text<|/ref|><|det|>[[115, 145, 866, 218]]<|/det|>
+1. In Figure S1c, the authors showed that RUFY3.2, an originally described RUFY3 isoform, is present in the cytosol in HeLa cells, strongly suggesting that RUFY3.2 is not an ARL8 effector. In the present manuscript, however, it is not clear whether RUFY3.2 binds to ARL8. The authors should test their interaction biochemically.  
+
+<|ref|>text<|/ref|><|det|>[[115, 257, 879, 329]]<|/det|>
+2. Based on the results of Figure 3d, the authors concluded that in addition to the ARL8-interacting CC2 domain, the FYVE domain also contributes to the RYFY3.1 localization to cytoplasmic vesicles. Is RUFY3.1 localized to cytoplasmic vesicles via its FYVE domain independently of ARL8? The authors need to test the localization of RUFY3.1-deltaCC2 in ARL8A/B-KO cells.  
+
+<|ref|>text<|/ref|><|det|>[[115, 368, 866, 422]]<|/det|>
+3. In Figure 5i, expression of RUFY3-GFP or RUFY4-GFP promoted retrograde transport of lysosomes in axons but also "reduced the number of moving lysosomes". Why? Please discuss the possible reason. Does expression of RUFY3-GFP or RUFY4-GFP also affect lysosome transport in "dendrites"?  
+
+<|ref|>text<|/ref|><|det|>[[115, 461, 881, 498]]<|/det|>
+4. Distinct Rab binding activities of the RUFY family members in the Discussion section is important (lines 397-404), but the information about the Rab33A binding activity of RUFY3.2 is missing.  
+
+<|ref|>text<|/ref|><|det|>[[115, 537, 181, 554]]<|/det|>
+5. Typos.  
+
+<|ref|>text<|/ref|><|det|>[[115, 564, 861, 600]]<|/det|>
+(line 40) unless; (line 357) RUFY3.3; (line 422) Once possibility; (line 683) Fig. 3c,4b should read Fig. 3b, 4b.  
+
+<|ref|>text<|/ref|><|det|>[[115, 611, 830, 630]]<|/det|>
+Figure 6c seems to be mis-labeled. The far right lane should be the GST-DLIC1 full-length protein.  
+
+<|ref|>text<|/ref|><|det|>[[115, 640, 836, 659]]<|/det|>
+Page numbers of several references are missing (e.g., lines 973,1026, 1064, 1082, 1096, and 1114).  
+
+<|ref|>text<|/ref|><|det|>[[115, 726, 393, 743]]<|/det|>
+Reviewer #5 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 782, 872, 874]]<|/det|>
+Keren- Kaplan et al. use an alternative BiolD2 assay called MitoID to identify proximal proteins of ARL8A and ARL8B. This resulted in the identification of RUFY3 and RUFY4 proteins that bind to the GTP- bound form of ARL8. The authors further demonstrate the interaction with dynein dynactin and link juxtanuclear redistribution of lysosomes under various conditions with the complex that they have described in this study.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 89, 872, 308]]<|/det|>
+For the proteomics part, the design of the experiment is not very clear in the current manuscript. While the authors briefly mention in the methods section that 3 biological replicates were used (with 3 'raw' in the data files), this is not clear from the results part, nor from fig. S1A or the processed tables. It would be better to make this clear to the readers in the figures and the text. Accordingly, the data processing to obtain Figures 1B and 1C and Supplementary dataset 1 should be better explained in the manuscript. BiolD and BiolD2 data have a tendency to generate a lot of background (especially upon massive overexpression as is the case here) so careful data analysis and processing is essential and should be transparent. For example, how exactly was the abundance ratio obtained and how was this transformed to fit the 0- 100 axis? Moreover, the proteomics data should be made available in a public repository so it becomes available to the community. It is not clear whether this is planned. Along this lines, it is also worthwhile to elaborate briefly on the benefits of MitolD over classical BiolD (or BiolD2) to better motivate the use of this interesting method to the readers.  
+
+<|ref|>text<|/ref|><|det|>[[114, 319, 880, 410]]<|/det|>
+Fig 4g. shows RT- qPCR analysis upon knockdown of the RUFY3. While I appreciate the other data supporting the link between RUFY3 and lysosomal localization, perturbance is important. The authors should clarify if this was a single siRNA or a pool of siRNAs. In the case of a single siRNA, it would be best to add another siRNA to eliminate the possibility of off- target effects. In the case of a pool, the pool should be deconvoluted to find (hopefully) at least 2 decent siRNAs that give the same phenotype.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 421, 245, 437]]<|/det|>
+## Minor comments  
+
+<|ref|>text<|/ref|><|det|>[[115, 450, 863, 504]]<|/det|>
+1. Line 110. The originally described MitolD procedure attaches BiolD and the mitochondrial targeting sequences C-terminally from the Ras family protein under investigation. Please provide some more explanation for switching the modules in this study.  
+
+<|ref|>text<|/ref|><|det|>[[112, 514, 864, 550]]<|/det|>
+2. Line 112. Provide a reference for the ARL8 mutants that lock the protein in the GDP and GTP-bound state.  
+
+<|ref|>text<|/ref|><|det|>[[112, 560, 864, 597]]<|/det|>
+3. Line 601: I assume that TECP is referring to the reducing chemical tris (2-carboxyethyl)phosphine so this should be TCEP.  
+
+<|ref|>text<|/ref|><|det|>[[112, 607, 800, 643]]<|/det|>
+4. Methods line 710-712. Transfection method and incubation time after transfection are not mentioned.  
+
+<|ref|>text<|/ref|><|det|>[[112, 654, 848, 690]]<|/det|>
+5. Methods line 718. "Following incubation, cells were washed" 1I think this should be "beads were washed"?  
+
+<|ref|>text<|/ref|><|det|>[[112, 701, 872, 738]]<|/det|>
+6. Fig 2b. On the bottom panel, many aspecific signals can be seen with very different intensities, please indicate the band sizes expected for each FLAG-tagged construct (for example in the figure legend).  
+
+<|ref|>text<|/ref|><|det|>[[113, 747, 867, 802]]<|/det|>
+7. Line 208, Fig 3c makes use of HeLa ARL8A-B KO cells, but these cells are not mentioned anywhere in the methods. Please describe their origin or how you established them, and provide evidence of successful KO if these cells have not been published elsewhere.  
+
+<|ref|>text<|/ref|><|det|>[[112, 812, 852, 848]]<|/det|>
+8. Fig 7e. Indicate which bars refer to rapalog-treated cells. Also check RUFY4 condition (NT and DHC siRNA)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[144, 125, 868, 177]]<|/det|>
+We thank the reviewers for their insightful and positive comments (italics). Changes to the text are indicated in red in the attached manuscript file. Below is a point- by- point response to the reviewers' comments:  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 194, 216, 210]]<|/det|>
+## Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[114, 226, 880, 378]]<|/det|>
+In this study, Keren- Kaplan et al identify 2 novel effectors of the small GTPase Arl8, namely RUIFY3 and RUIFY4. They use proximity labeling coupled with mitochondrial targeting (Mito ID) to identify these ARL8 interactors. Using imaging and biochemical approaches, they demonstrate that RUIFY3 and 4 localize to LAMP1 and ARL8 positive vesicles and drive perinuclear clustering of these organelles. They also show using in vitro experiments that RUIFY3 and 4 interact with dynein- dynactin complex. Using deletion constructs they identify the CC2 domain on RUIFY3 and 4 to be responsible for ARL8 binding and driving the perinuclear clustering. RUIFY3 also appears to localize to a subset of axonal LAMP1- positive organelles, in culture primary neurons. Lastly, RUIFY3/4 appears to be required for the juxtanuclear movement of lysosomes upon serum starvation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 394, 882, 462]]<|/det|>
+The study is interesting and novel in that it identifies new effectors of ARL8 GTPase and links this GTPase to retrograde transport for the first time. Through well- designed experiments, striking images and quantitative data, the authors convey most of the points mentioned above very clearly. There are some minor issues that should be addressed prior to publication:  
+
+<|ref|>text<|/ref|><|det|>[[144, 479, 850, 514]]<|/det|>
+We thank this reviewer for the positive assessment of our manuscript. We addressed the issues mentioned by this reviewer as described below.  
+
+<|ref|>text<|/ref|><|det|>[[115, 530, 576, 547]]<|/det|>
+1. Role of the different domains (CC2 and FYVE in particular):  
+
+<|ref|>text<|/ref|><|det|>[[115, 548, 858, 616]]<|/det|>
+a) In figure 3d, it appears that RUIFY3 lacking RUN or CC1 domains, while still localize to lysosomes/vesicles, are much less 'tightly' clustered to perinuclear region. This could suggest a role for these domains in supporting the retrograde movement or in interaction with retrograde transport machinery? It would be helpful to see a quantification of the strength of this clustering.  
+
+<|ref|>text<|/ref|><|det|>[[144, 632, 880, 855]]<|/det|>
+We have revised this figure to show transfected cells with more similar levels of expression, which allows for better comparison of the localization of RUIFY3- GFP deletion constructs (new Fig. 3f). Overall, we did not see significant changes in juxtanuclear clustering of RUIFY3 vesicles upon deletion of domains other than CC2 (more dispersed) (new Fig. 3h). Importantly, in the new version of the manuscript, we have additionally quantified the association of the RUIFY3- GFP deletion constructs with vesicles (new Fig. 3g). We found that deletion of the RUN domain, alone or in combination with the CC1 domain, did not significantly alter association of RUIFY3- GFP with vesicles (new Fig. 3g). Deletion of the CC1 domain alone decreased vesicle association, though to a much lesser extent than deletion of the CC2 or FYVE domains (Fig. 3g). These results clearly show that both the CC2 and FYVE domains promote membrane association of RUIFY3 vesicles, and that the CC2 domain additionally promotes juxtanuclear clustering of the vesicles. Other domains appear to be less important for these processes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 872, 857, 890]]<|/det|>
+b) There seems substantial recruitment of RUIFY3 to vesicles even in absence of CC2 (Fig 3d) although
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 88, 880, 124]]<|/det|>
+the perinuclear clustering is clearly reduced, while the version of protein lacking CC2 and FYVE domains is largely cytosolic. Is it more accurate to so both are involved in recruitment to vesicles?  
+
+<|ref|>text<|/ref|><|det|>[[144, 140, 870, 382]]<|/det|>
+The reviewer's observations are correct. The revised images in the new Fig. 3f better show the localization of different RUFY3- GFP deletion constructs with comparable levels of expression. The \(\Delta \mathrm{CC2}\) construct exhibits partially decreased association with vesicles (new Fig. 3f, g), and these are less clustered in the juxtanuclear area (new Fig. 3f, h). The \(\Delta \mathrm{FYVE}\) construct is also less associated with vesicles (new Fig. 3f, g) and the remaining vesicles are tightly clustered in the juxtanuclear area (new Fig. 3f, h). The CC2 or FYVE domains alone are completely cytosolic (new Fig. 3f, g). We have also added a new Supplementary Fig. 2a showing that ARL8 KO decreases the clustering of full- length RUFY3 and abrogates the localization of the \(\Delta \mathrm{FYVE}\) construct to the tight juxtanuclear cluster. From these experiments, we conclude that both the CC2 and FYVE domains contribute to association with vesicles, and that the CC2 domain additionally contributes to juxtanuclear clustering. These findings are now described in more detail in the Results section. We thank the reviewer for raising this issue, which helped us provide a more precise description of the roles of the CC2 and FYVE domains.  
+
+<|ref|>text<|/ref|><|det|>[[116, 398, 710, 416]]<|/det|>
+Also, can the authors explain why this looks different from the mito- experiments?  
+
+<|ref|>text<|/ref|><|det|>[[144, 432, 882, 536]]<|/det|>
+The localizations of RUFY3/4 in Figs. 1 (mito- ID with the T34N ARL8 construct) and 3 (nonmito- ID) are not so different. If there is a subtle difference, it could be due to the fact that in the experiment in Fig. 1 cells were fixed, whereas in that in Fig. 3 cells were imaged live. The difference in the membrane recruitment of the CC2 domain alone in Figs. 2d and 3f is likely due to the overexpression of the mitochondrially targeted ARL8 in Fig. 2d. This is now explained in the text.  
+
+<|ref|>text<|/ref|><|det|>[[115, 553, 360, 569]]<|/det|>
+2. RUFY3/4 function in neurons:  
+
+<|ref|>text<|/ref|><|det|>[[115, 570, 874, 620]]<|/det|>
+a) Kymographs in Fig 5e seem to suggest that RUFY4 is on both anterogradely and retrogradely moving LAMP1 vesicles while RUFY3 is largely on retrogradely moving ones. This looks a bit different from the quantification in 5h where both RUFY3/4 look to be more on retrogradely moving vesicles.  
+
+<|ref|>text<|/ref|><|det|>[[144, 637, 868, 722]]<|/det|>
+We thank the reviewer for pointing out this discrepancy. It was due to a poor choice of the representative images. We have now replaced the kymograph analyses and chosen images that are more representative of the quantification (new Fig. 6e, f). Both the images and the quantification show that RUFY3 and RUFY4 are more abundant in retrogradely- moving vesicles.  
+
+<|ref|>text<|/ref|><|det|>[[115, 739, 881, 808]]<|/det|>
+b) Likewise, while both seem to increase relative proportion of retrogradely moving LAMP1 vesicles, there is a strong reduction of total number of motile LAMP1 axonal vesicles. This data is very interesting overall but concluding that they increase retrograde transport seems a bit simplistic. Could RUFY3/4 affect Arl8 interaction/binding with SKIP and/or Kinesin?  
+
+<|ref|>text<|/ref|><|det|>[[144, 825, 880, 894]]<|/det|>
+We agree with the reviewer that RUFY3/4 could compete with SKIP and kinesins for binding to ARL8, and we now mention this caveat in the Discussion. However, the fact that RUFY3 KD disperses endolysosomes, that RUFY3/4 interact with dynein- dynactin, and that they redistribute peroxisomes to the cell center in a dynein- dynactin- dependent manner
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[142, 90, 868, 125]]<|/det|>
+support their role as dynein- dynactin adaptors. We think increased retrograde transport is the most sensible interpretation for the effect of RUFY3/4 on axonal LAMP1 vesicles.  
+
+<|ref|>text<|/ref|><|det|>[[115, 140, 880, 193]]<|/det|>
+Likewise, are there changes to pauses in LAMP1 vesicle motion, processivity, velocity? Further dissection of how RUFY3/4 over expression alters the motility will shed important information on the transport properties of these organelles.  
+
+<|ref|>text<|/ref|><|det|>[[144, 209, 872, 313]]<|/det|>
+New quantifications shown in Figs. 6h, i demonstrate that RUFY3/4 have little or no effect on the velocity and run length of both anterograde and retrograde LAMP1 vesicles in the axon. The main effects are a decrease in the relative number of anterograde LAMP1 tracks and the total number of LAMP1 tracks in the axon. While we appreciate the opportunity to provide these additional data, we think that further analyses of axonal transport exceed the scope of this study and our possibilities at this time.  
+
+<|ref|>text<|/ref|><|det|>[[115, 328, 867, 430]]<|/det|>
+c) As the authors point out in their discussion RUFY3/4 localizes to a subset of axonal LAMP1-positive organelles. Given the interest in these organelles, their maturation and transport in the neuronal cell biology field, it would help to further define this sub-population: are there Rab7 positive? Are the acidic (there is a strong correlation between retrogradely moving LAMP1 vesicles in axons and their acidic nature). These experiments will help determine more clearly, the nature of these RUFY3-positive endolysosomes.  
+
+<|ref|>text<|/ref|><|det|>[[144, 445, 866, 637]]<|/det|>
+The identity of LAMP1 vesicles in the axon is currently a matter of debate, and we do not aspire to resolve this important but complex problem in this study. Nevertheless, we are happy to provide additional characterization of these vesicles in the new Supplementary Fig. 4. We find that RUFY3 and RUFY4 exhibit \(100\%\) co- localization with ARL8 in axonal vesicles (new Supplementary Fig. 4d- f). We also find that \(\sim 85\%\) RUFY3 and \(\sim 40\%\) RUFY4 are associated with LysoTracker- positive vesicles (new Supplementary Fig. 4h), and that RUFY3/4- LysoTracker- positive vesicles move mostly in the retrograde direction (new Supplementary Fig. 4i, j). These data indicate that RUFY3 is largely associated, and RUFY4 partially associated, with retrogradely- moving acidic endolysosomes. In addition, RUFY4 seems to associate with another population of non- acidic vesicles, the identity of which remains to be determined.  
+
+<|ref|>text<|/ref|><|det|>[[145, 653, 872, 707]]<|/det|>
+We did not examine the co- localization with RAB7 because this is a very complex issue that deserves its own, dedicated study, and because RUFY3 and RUFY4 do not interact with RAB7 (new data in Supplementary fig. 1e, f).  
+
+<|ref|>text<|/ref|><|det|>[[115, 722, 870, 791]]<|/det|>
+d) While the authors clearly demonstrated a reduction of LAMP1 tracks, based on over expression, from their live imaging experiments (does this include stationary vesicles?), it would be better to examine the number of endogenous LAMP1 vesicles (per unit length of axon), to conclude that there are reduced number of lysosomes in axons.  
+
+<|ref|>text<|/ref|><|det|>[[145, 806, 880, 893]]<|/det|>
+In the new Supplementary fig. 4a, b, we show that expression of RUFY3 or RUFY4 decreases the number of endogenous LAMTOR4 (a bona fide endolysosomal marker) puncta per unit length of axon. We used this particular marker because the antibody gives stronger, more specific staining. This finding is consistent with the reduction of not only moving LAMP1 tracks but also the overall number of endolysosomes in the axon.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 108, 262, 123]]<|/det|>
+Other minor points:  
+
+<|ref|>text<|/ref|><|det|>[[115, 123, 866, 159]]<|/det|>
+In Fig 7 e, it will be good to include Raplog in the graph's X axis as it is a bit confusing without that- it appears as if the \(+ / -\) are for the siRNA.  
+
+<|ref|>text<|/ref|><|det|>[[144, 174, 515, 193]]<|/det|>
+We fixed this issue on the figure (now Fig. 8e).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 226, 217, 243]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[114, 259, 880, 428]]<|/det|>
+In this study, the authors reveal a new retrograde transport pathway for late endosomes/lysosomes mediated by a small GTPase Arl8b, which has thus far only been appreciated for its role in anterograde transport of the same organelles. Specifically, Keren- Kaplan et al identify RUY3 and 4 as the first minus- end- directed transport effectors for Arl8b. The authors show that RUY3 interacts with Arl8b using its CC2 domain, thereby allowing endosomes to be transported towards the perinuclear area in a dynein/dynactin dependent manner. The manuscript is well structured and easy to read. Most of the conclusions drawn are substantiated by the experiments presented, and the data appears reproducible and of high quality. It is important to point out however that the novelty of the findings is threatened by a recent preprint by Kumar and colleagues (DOI: 10.21203/rs.3.rs- 345822/v1). Irrespective of this, several important issues would need to be addressed prior to publication:  
+
+<|ref|>text<|/ref|><|det|>[[144, 444, 696, 462]]<|/det|>
+We thank this reviewer for his positive comments on our manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 480, 212, 495]]<|/det|>
+## Major points  
+
+<|ref|>text<|/ref|><|det|>[[115, 495, 878, 596]]<|/det|>
+1. My biggest concern is the implicit assumption by the authors that all retrograde transport mediated by RUY3/4 results downstream of Arl8b. This approach disregards the possibility that Rab7 may (also) be directing these actions. It is well established that a subset of late endosomes/lysosomes contains both Arl8 and Rab7, and several other Arl8 effectors, including SKIP and the HOPS complex are shared by these GTPases. The authors should address this point, for example by examining whether Rab7 (QL versus TN) can interact with RUY3/4.  
+
+<|ref|>text<|/ref|><|det|>[[144, 612, 880, 699]]<|/det|>
+In the new Supplementary fig. 1e, f, we show that neither Q67L nor T22N forms of mito- . RAB7A relocate RUY3 and RUY4 to mitochondria. This is in contrast to the mitochondrial relocation of a known RAB7A effector, RILP, by the Q67L construct (positive control). This demonstrates that RUY3/4 are not RAB7 effectors. In light of these results, we did not further pursue a possible relationship of RUY3/4 with RAB7A.  
+
+<|ref|>text<|/ref|><|det|>[[115, 715, 878, 767]]<|/det|>
+2. The authors mention that they find that binding of Arl8 to RUY involves the CC2 domain of RUY3. Although the truncation data indicates necessity, the authors should include evaluation of the CC2 domain (or combined with FYVE) to demonstrate sufficiency with respect to the recruitment to Arl8b.  
+
+<|ref|>text<|/ref|><|det|>[[144, 784, 878, 835]]<|/det|>
+We thank the reviewer for raising this important issue. We now provide new mito- targeting (new Fig. 2d, e) and pulldown (new Fig. 2h, i) data showing that the CC2 but not FYVE domain binds the active form of ARL8.  
+
+<|ref|>text<|/ref|><|det|>[[115, 852, 872, 887]]<|/det|>
+3. It would have been nice to see that engagement of RUY3/4 with peripheral Arl8b-positive endosomes drives their transport into the juxtanuclear region in live cells.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[144, 90, 876, 177]]<|/det|>
+While we acknowledge the importance of the suggested experiment, it would be difficult to perform in non- neuronal cells because RUFY3/4 causes strong clustering of endolysosomes in the juxtanuclear area. We think that our experiments in neurons provide evidence for an increase in the ratio of retrograde vs. anterograde LAMP1 vesicles (Fig. 6e, f), consistent with RUFY3/4 promoting retrograde transport.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 194, 214, 210]]<|/det|>
+## Minor points  
+
+<|ref|>text<|/ref|><|det|>[[115, 210, 880, 245]]<|/det|>
+4. In figure 2g, dCC2-FLAG and dCC2dFYVE-FLAG are not pulled down with GST-Arl8b-QL, however a band is visible for the condition where they used GST-Arl8b-TN. What could be the reason for this?  
+
+<|ref|>text<|/ref|><|det|>[[144, 261, 877, 330]]<|/det|>
+This is a faint non- specific band that is seen in overexposed blots. For better appreciation of the relative intensity of this non- specific band vs. specific bands, in the new Fig. 2g we show a shorter exposure of the blots. In any case, both the long and short exposures are shown in the source data file.  
+
+<|ref|>text<|/ref|><|det|>[[115, 346, 872, 381]]<|/det|>
+5. RUFY3 siRNAs need to be deconvoluted to show that multiple RUFY3 siRNA duplexes show similar affects.  
+
+<|ref|>text<|/ref|><|det|>[[144, 397, 835, 450]]<|/det|>
+In the new Supplementary Fig. 3, we show that 3 of the 4 single siRNAs in the original SMARTpool effectively knock down RUFY3, and that the 3 effective siRNAs cause peripheral dispersal of endolysosomes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 466, 738, 485]]<|/det|>
+6. Figure 4h: in the legend it is not mentioned what the yellow-colored arrows refer to.  
+
+<|ref|>text<|/ref|><|det|>[[142, 500, 883, 536]]<|/det|>
+We now indicate in the text that the yellow arrows point to LAMP1 structures at cell vertices (new Fig. 5d).  
+
+<|ref|>text<|/ref|><|det|>[[115, 552, 486, 570]]<|/det|>
+7. Figure 7e is missing the annotation +/- Rapalog.  
+
+<|ref|>text<|/ref|><|det|>[[144, 586, 595, 604]]<|/det|>
+This was fixed in the new version of Fig. 7e, now Fig. 8e.  
+
+<|ref|>text<|/ref|><|det|>[[115, 620, 639, 638]]<|/det|>
+8. Figure 6c Annotation of GST-DLIC1 and GST-DLIC-CT is swapped.  
+
+<|ref|>text<|/ref|><|det|>[[144, 655, 595, 673]]<|/det|>
+This was fixed in the new version of Fig. 6c, now Fig. 7c.  
+
+<|ref|>text<|/ref|><|det|>[[115, 689, 855, 742]]<|/det|>
+9. Some of the clustering pictures in Fig4 are not convincing enough (compared to the quantification). For example, \(\Delta\) RUN-GFP does not look as clustered as RUFY3-GFP, while the quantifications show similar average clustering.  
+
+<|ref|>text<|/ref|><|det|>[[144, 758, 870, 827]]<|/det|>
+We have replaced the image for \(\Delta\) RUN by another image that is more representative of the quantitative data (new Fig. 5a). Both the images and the quantification suggest that this construct is a bit less effective at clustering endolysosomes, although the differences do not rise to the level of statistical significance.  
+
+<|ref|>text<|/ref|><|det|>[[115, 863, 217, 879]]<|/det|>
+Reviewer #3
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 879, 258]]<|/det|>
+The manuscript by Keren- Kaplan et al. is a well written study that describes identification of novel ARL8 GTPase effectors, RUN- and FYVE- domain containing proteins, RUFY3 and RUFY4 that are important for retrograde lysosome movement in cells. The authors use a combination of cell biology and biochemical methods to show that RUFY3 and RUFY4, but not other members of the RUFY family, associate with ARL8- lysosomes. The authors also map regions in RUFY3 important for an interaction with ARL8. Furthermore, the authors provide evidence that the retrograde movement of ARL8- lysosomes is mediated by RUFY3 and RUFY4 association with the dynein complex, although direct evidence of binding to dynein is only shown for RUFY3. Together with previous studies from this group and others showing that ARL8 can associate with kinesin adaptors to move lysosomes in an anterograde direction, this work provides important insights into how ARL8 regulates movement of lysosomes in cells.  
+
+<|ref|>text<|/ref|><|det|>[[115, 272, 819, 324]]<|/det|>
+The experiments are described in sufficient detail to be replicated and statistical analysis appears appropriate. I think that this study will be of high interest to the field of cellular trafficking, and I recommend its publication in the journal of Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[144, 341, 735, 359]]<|/det|>
+We also thank this reviewer for the positive comments on our manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 375, 450, 392]]<|/det|>
+I have few minor suggestions described below:  
+
+<|ref|>text<|/ref|><|det|>[[115, 392, 879, 476]]<|/det|>
+1. In section titled: "ARL8B promotes recruitment of RUFY3 and RUFY4 to a juxtanuclear cluster of vesicles" authors conclude that ARL8 promotes the recruitment of RUFY3 and RUFY4 vesicles via the CC2 domain and that FYVE domain makes additional contributions to this recruitment. However, the contributions of RUFY4 domains are never tested. I would suggest clarifying this statement to focus only on domain contributions from RUFY3.  
+
+<|ref|>text<|/ref|><|det|>[[141, 493, 870, 528]]<|/det|>
+We thank the review for pointing out this misstatement. We have modified the text to refer only to domains of RUFY3.  
+
+<|ref|>text<|/ref|><|det|>[[115, 544, 867, 630]]<|/det|>
+2. The different RUFY3-GFP truncation constructs shown in inserts in figure 4e look very cytoplasmic, which is not fully consistent with images shown in figure 3d. Is there an increase in cytoplasmic localization of these constructs when co-expressed with LAMP1? If so, this would be unexpected and should be addressed in the text. It would also be helpful if the authors included full size images in the main figure or in supplementary figure.  
+
+<|ref|>text<|/ref|><|det|>[[144, 646, 858, 715]]<|/det|>
+The images in the insets of Fig. 4a look more cytosolic because 1) the constructs are overexpressed in order to maximize their effects on endolysosomes, and 2) the cells were fixed before staining, in contrast to those in Fig. 3d (now Fig. 3f), which were imaged live. We now clarify these differences in the text and figure legends.  
+
+<|ref|>text<|/ref|><|det|>[[115, 731, 627, 750]]<|/det|>
+3. There is a typo in the legend for figure 1, panel f is labeled with an e.  
+
+<|ref|>text<|/ref|><|det|>[[144, 766, 194, 782]]<|/det|>
+Fixed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 800, 686, 818]]<|/det|>
+4. Figure 2 panel a, please add description of what is being blotted in the input.  
+
+<|ref|>text<|/ref|><|det|>[[140, 834, 820, 852]]<|/det|>
+We now indicate in the figure that the input is immunoblotted for the FLAG epitope.  
+
+<|ref|>text<|/ref|><|det|>[[115, 868, 880, 886]]<|/det|>
+5. Figure 2 panel e has RUN-GFP included although there is no description of this construct anywhere in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 88, 870, 124]]<|/det|>
+the text or figure legend. Similarly, there is RUN- FLAG included in panel f and g, but again there is no description of this construct. Please add description of these constructs.  
+
+<|ref|>text<|/ref|><|det|>[[140, 140, 879, 177]]<|/det|>
+We now describe in the text the results with the RUN- FLAG construct in reference to Fig. 2f, 8.  
+
+<|ref|>text<|/ref|><|det|>[[115, 192, 857, 226]]<|/det|>
+6. In general, all labels for microscopy images are very small and hard to see. Could the authors please increase font size or rearrange labels to be above or below images, so it is easier to see them?  
+
+<|ref|>text<|/ref|><|det|>[[140, 243, 861, 278]]<|/det|>
+Whenever possible, we have increased the font size of the labels. In some cases, it was not possible for reasons of space.  
+
+<|ref|>text<|/ref|><|det|>[[115, 294, 549, 311]]<|/det|>
+7. Please add reference for the source of ARL8A- B- KO cells.  
+
+<|ref|>text<|/ref|><|det|>[[140, 329, 875, 363]]<|/det|>
+We added a reference for these cells (Keren- Kaplan and Bonifacino, 2021, Curr. Biol. 31, 540- 554 e545).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 397, 416, 415]]<|/det|>
+## Reviewer #4 (Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[114, 431, 874, 567]]<|/det|>
+In this manuscript, the authors identified RUFY3.1, a previously uncharacterized longer isoform of RUFY3, and RUFY4 as novel ARL8 effectors by using the MitoID method. They then found that both RUFY3.1 and RUFY4 promote retrograde transport of lysosomes through physical interaction with the dynein- dynactin motor complex and that their activity is required for accumulation of lysosomes in the juxtanuclear area of cells. Overall, the experiments are well conducted and there is good inclusion of positive and negative controls to reveal the selectivity and specificity of the observations. Thus, the study makes a significant contribution to the mechanistic understanding of ARL8- mediated lysosomal positioning that plays important roles in a wide range of cellular processes.  
+
+<|ref|>text<|/ref|><|det|>[[144, 583, 653, 600]]<|/det|>
+We thank this reviewer for the positive comments on our study.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 618, 255, 634]]<|/det|>
+## Specific comments:  
+
+<|ref|>text<|/ref|><|det|>[[115, 634, 882, 700]]<|/det|>
+1. In Figure S1c, the authors showed that RUFY3.2, an originally described RUFY3 isoform, is present in the cytosol in HeLa cells, strongly suggesting that RUFY3.2 is not an ARL8 effector. In the present manuscript, however, it is not clear whether RUFY3.2 binds to ARL8. The authors should test their interaction biochemically.  
+
+<|ref|>text<|/ref|><|det|>[[144, 717, 857, 786]]<|/det|>
+We have now added mitochondrial relocalization and pulldown analyses for RUFY3.1 vs RUFY3.2 (new Supplementary fig. 1c,d,g,h). The results show that the shorter RUFY3.2 isoform, lacking the FYVE domain and part of the CC2 domain, does not interact with ARL8.  
+
+<|ref|>text<|/ref|><|det|>[[115, 804, 874, 871]]<|/det|>
+2. Based on the results of Figure 3d, the authors concluded that in addition to the ARL8-interacting CC2 domain, the FYVE domain also contributes to the RYFY3.1 localization to cytoplasmic vesicles. Is RUFY3.1 localized to cytoplasmic vesicles via its FYVE domain independently of ARL8? The authors need to test the localization of RUFY3.1-deltaCC2 in ARL8A/B-KO cells.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[144, 90, 872, 210]]<|/det|>
+We thank the reviewer for suggesting this very important experiment. In the new Supplementary fig. 2a, we show that ARL8 KO reduces overall vesicular staining for RUFY3- GFP and the juxtanuclear clustering of RUFY3- GFP vesicles, does not change the already dispersed distribution of the \(\Delta \mathrm{CC2}\) construct, and abrogates the association of the \(\Delta \mathrm{FYVE}\) construct with vesicles. These findings confirm that the ARL8- binding CC2 domain is important for juxtanuclear clustering, while both the CC2 and the FYVE domain contribute to vesicle recruitment.  
+
+<|ref|>text<|/ref|><|det|>[[115, 226, 866, 278]]<|/det|>
+3. In Figure 5i, expression of RUFY3-GFP or RUFY4-GFP promoted retrograde transport of lysosomes in axons but also "reduced the number of moving lysosomes". Why? Please discuss the possible reason. Does expression of RUFY3-GFP or RUFY4-GFP also affect lysosome transport in "dendrites"?  
+
+<|ref|>text<|/ref|><|det|>[[144, 295, 877, 432]]<|/det|>
+In the new version of the manuscript, we explain more clearly that expression of RUFY3- GFP or RUFY4- GFP both reduces the total number of moving LAMP1 vesicles, and shifts the balance of the moving LAMP1 vesicles towards retrograde transport. This is consistent with RUFY3- GFP or RUFY4- GFP promoting retrograde over anterograde transport. As in non- neuronal cells, the reduction in LAMP1 tracks and LAMTOR4 puncta in the axon (new Fig. 6g and Supplementary Fig. 4a,b; see response to Reviewer #1, point d) likely results from accumulation of endolysosomes in the juxtanuclear area of the neurons. We now make this explanation clearer in the text.  
+
+<|ref|>text<|/ref|><|det|>[[144, 449, 870, 553]]<|/det|>
+RUFY3- FLAG and RUFY4- FLAG also colocalize with LAMP1- GFP and endogenous LAMTOR4 in dendrites (Fig. 6a- d). However, because dendrites have mixed microtubule polarity, it is harder to assess whether RUFY3- FLAG and RUFY4- FLAG alter the distribution or movement of lysosomes. For this reason, we limited our imaging of moving endolysosomes to axons, where microtubules are uniformly polarized and it is easier to assess anterograde vs. retrograde movement.  
+
+<|ref|>text<|/ref|><|det|>[[115, 569, 857, 604]]<|/det|>
+4. Distinct Rab binding activities of the RUFY family members in the Discussion section is important (lines 397-404), but the information about the Rab33A binding activity of RUFY3.2 is missing.  
+
+<|ref|>text<|/ref|><|det|>[[144, 621, 867, 672]]<|/det|>
+In the new version of the Discussion, we mention the fact that RUFY3.2 interacts with RAB33, and point out that the significance of this interaction for recruitment of RUFY3.2 to endolysosomes remains to be addressed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 689, 860, 724]]<|/det|>
+5. Typos. (line 40) unles; (line 357) RUFY3.3; (line 422) Once possibility; (line 683) Fig. 3c,4b should read Fig. 3b, 4b.  
+
+<|ref|>text<|/ref|><|det|>[[115, 740, 830, 758]]<|/det|>
+Figure 6c seems to be mis- labeled. The far- right lane should be the GST- DLIC1 full- length protein.  
+
+<|ref|>text<|/ref|><|det|>[[115, 774, 828, 792]]<|/det|>
+Page numbers of several references are missing (e.g., lines 973,1026, 1064, 1082, 1096, and 1114).  
+
+<|ref|>text<|/ref|><|det|>[[145, 809, 378, 826]]<|/det|>
+All of these typos were fixed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 862, 221, 878]]<|/det|>
+Reviewer #5:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 870, 174]]<|/det|>
+Keren- Kaplan et al. use an alternative BioID2 assay called MitoID to identify proximal proteins of ARL8A and ARL8B. This resulted in the identification of RUFY3 and RUFY4 proteins that bind to the GTP- bound form of ARL8. The authors further demonstrate the interaction with dynein dynactin and link juxtanuclear redistribution of lysozymes under various conditions with the complex that they have described in this study.  
+
+<|ref|>text<|/ref|><|det|>[[114, 174, 880, 309]]<|/det|>
+For the proteomics part, the design of the experiment is not very clear in the current manuscript. While the authors briefly mention in the methods section that 3 biological replicates were used (with 3 'raw' in the data files), this is not clear from the results part, nor from fig. S1A or the processed tables. It would be better to make this clear to the readers in the figures and the text. Accordingly, the data processing to obtain Figures 1B and 1C and Supplementary dataset 1 should be better explained in the manuscript. BioID and BioID2 data tend to generate a lot of background (especially upon massive overexpression as is the case here) so careful data analysis and processing is essential and should be transparent. For example, how exactly was the abundance ratio obtained and how was this transformed to fit the 0- 100 axis?  
+
+<|ref|>text<|/ref|><|det|>[[144, 323, 852, 377]]<|/det|>
+We now provide a more detailed description of the MitoID and mass spectrometry procedures in the legend to Fig. 1 and the Methods sections. This includes a statement in both the Fig. 1 legend and Methods that 3 biological replicates were used.  
+
+<|ref|>text<|/ref|><|det|>[[144, 392, 870, 445]]<|/det|>
+MitoID was chosen precisely because the targeting of the bait constructs to mitochondria provides for a more uniform background of non- specific hits, irrespective of the original localization of the baits. This is now mentioned in the first paragraph of the Results section.  
+
+<|ref|>text<|/ref|><|det|>[[144, 461, 863, 532]]<|/det|>
+The default value of 100 was set as the maximum fold change allowed. For example, if the calculated ratios are 50, 80, 120 and 150 for proteins A, B, C, and D, the ratios reported by PD software are 50, 80, 100, 100 for proteins A, B, C and D. The abundance ratio was not transformed in any way. It was plotted on the graph as presented in the Excel file.  
+
+<|ref|>text<|/ref|><|det|>[[115, 547, 867, 582]]<|/det|>
+Moreover, the proteomics data should be made available in a public repository, so it becomes available to the community. It is not clear whether this is planned.  
+
+<|ref|>text<|/ref|><|det|>[[148, 598, 872, 651]]<|/det|>
+Proteomics raw data and search results were deposited in the MassIVE repository and will be available upon publication: Title: Tal Keren- Kaplan and Juan Bonifacino LFQ data, link: ftp://massive.ucds.edu/MSV000087741/.  
+
+<|ref|>text<|/ref|><|det|>[[149, 667, 820, 702]]<|/det|>
+An Excel spreadsheet with all the proteins identified in the study is also included as Supplementary Data 1.  
+
+<|ref|>text<|/ref|><|det|>[[115, 718, 870, 753]]<|/det|>
+Along these lines, it is also worthwhile to elaborate briefly on the benefits of MitoID over classical BioID (or BioID2) to better motivate the use of this interesting method to the readers.  
+
+<|ref|>text<|/ref|><|det|>[[144, 769, 872, 857]]<|/det|>
+In the Results section (under section "Identification of RUFY3 and RUFY4 as ARL8 effectors"), we now mention that one advantage of MitoID over BioID is a more uniform identification of non- specific proteins due to the targeting of all constructs to mitochondria, irrespective of what compartment they normally associate with in the absence of the mitochondrial- targeting signal (e.g., endolysosomes vs. cytosol).  
+
+<|ref|>text<|/ref|><|det|>[[115, 872, 861, 907]]<|/det|>
+Fig 4g. shows RT- qPCR analysis upon knockdown of the RUFY3. While I appreciate the other data supporting the link between RUFY3 and lysosomal localization, perturbance is important. The authors
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 88, 866, 140]]<|/det|>
+should clarify if this was a single siRNA or a pool of siRNAs. In the case of a single siRNA, it would be best to add another siRNA to eliminate the possibility of off- target effects. In the case of a pool, the pool should be deconvoluted to find (hopefully) at least 2 decent siRNAs that give the same phenotype.  
+
+<|ref|>text<|/ref|><|det|>[[144, 157, 881, 244]]<|/det|>
+We thank the reviewer for this comment. We now indicate in the legend to the new Fig. 5 that RUY3 was knocked down using a SMARTpool. In addition, in the new Supplementary Fig. 3, we show RUY3 KD and LAMP1 redistribution data for single siRNAs from the SMARTpool. Three of the 4 siRNAs are effective in knocking down RUY3 and causing endolysosome dispersal.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 261, 243, 276]]<|/det|>
+## Minor comments  
+
+<|ref|>text<|/ref|><|det|>[[115, 277, 866, 328]]<|/det|>
+1. Line 110. The originally described MitoID procedure attaches BioID and the mitochondrial targeting sequences C-terminally from the Ras family protein under investigation. Please provide some more explanation for switching the modules in this study.  
+
+<|ref|>text<|/ref|><|det|>[[144, 345, 881, 430]]<|/det|>
+The RAB GTPases used in the original study are anchored to membranes via their C- terminus. In contrast, ARL GTPases, including the ARL8 used in our study, are anchored to membranes via their N-terminus, thus the different design of our constructs. We now briefly explain this in the first paragraph of the results and the corresponding section of the Methods (Recombinant DNAs).  
+
+<|ref|>text<|/ref|><|det|>[[115, 447, 871, 481]]<|/det|>
+2. Line 112. Provide a reference for the ARL8 mutants that lock the protein in the GDP and GTP-bound state.  
+
+<|ref|>text<|/ref|><|det|>[[144, 499, 481, 516]]<|/det|>
+We added a reference for these mutations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 533, 867, 567]]<|/det|>
+3. Line 601: I assume that TECP is referring to the reducing chemical tris (2-carboxyethyl)phosphine so this should be TCEP.  
+
+<|ref|>text<|/ref|><|det|>[[144, 585, 733, 620]]<|/det|>
+We corrected the abbreviation to TCEP and explained its meaning (Tris(2- carboxyethyl)phosphine).  
+
+<|ref|>text<|/ref|><|det|>[[115, 635, 869, 653]]<|/det|>
+4. Methods line 710-712. Transfection method and incubation time after transfection are not mentioned.  
+
+<|ref|>text<|/ref|><|det|>[[115, 670, 787, 687]]<|/det|>
+The transfection protocol was described in the "Cell culture and treatments" section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 717, 841, 750]]<|/det|>
+5. Methods line 718. "Following incubation, cells were washed" à I think this should be "beads were washed"?  
+
+<|ref|>text<|/ref|><|det|>[[144, 770, 229, 786]]<|/det|>
+Corrected.  
+
+<|ref|>text<|/ref|><|det|>[[115, 803, 866, 837]]<|/det|>
+6. Fig 2b. On the bottom panel, many aspecific signals can be seen with very different intensities, please indicate the band sizes expected for each FLAG-tagged construct (for example in the figure legend)  
+
+<|ref|>text<|/ref|><|det|>[[144, 855, 803, 889]]<|/det|>
+We added asterisks indicating the undegraded proteins and included the expected molecular masses in the figure legend.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 88, 874, 140]]<|/det|>
+7. Line 208, Fig 3c makes use of HeLa ARL8A-B KO cells, but these cells are not mentioned anywhere in the methods. Please describe their origin or how you established them and provide evidence of successful KO if these cells have not been published elsewhere.  
+
+<|ref|>text<|/ref|><|det|>[[142, 156, 844, 191]]<|/det|>
+The HeLa ARL8A- B KO cells were described previously (Keren- Kaplan and Bonifacino, 2021, Curr. Biol. 31, 540- 554 e545). We have added the reference to the text.  
+
+<|ref|>text<|/ref|><|det|>[[115, 208, 870, 243]]<|/det|>
+8. Fig 7e. Indicate which bars refer to rapalog-treated cells. Also check RUIFY4 condition (NT and DHC siRNA)  
+
+<|ref|>text<|/ref|><|det|>[[141, 259, 879, 294]]<|/det|>
+We modified Fig. 7e (now Fig. 8e) to indicate the rapalog- treated cells. We also corrected the labeling error for RUIFY4.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 300, 106]]<|/det|>
+REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[115, 146, 393, 163]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 230, 879, 360]]<|/det|>
+The study by Keren- Kaplan et al identifying 2 novel effectors of the small GTPase Arl8, namely RUFY3 and RUFY4 is interesting and novel in that it identifies these new effectors of ARL8 GTPase and links this GTPase to retrograde transport for the first time. As mentioned before in initial review the topic is important and the study itself is very well done with well- designed experiments, striking images and quantitative data. The authors have addressed the minor issues raised in initial review with additional experiments, quantitative data they have included and/or clarifications in text. I recommend publication of this interesting study.  
+
+<|ref|>text<|/ref|><|det|>[[115, 455, 393, 471]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 512, 361, 529]]<|/det|>
+The authors addressed my points  
+
+<|ref|>text<|/ref|><|det|>[[115, 596, 393, 613]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 654, 456, 671]]<|/det|>
+The authors have addressed all my comments.  
+
+<|ref|>text<|/ref|><|det|>[[115, 737, 393, 754]]<|/det|>
+Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 795, 855, 831]]<|/det|>
+In the revised manuscript, the authors properly addressed all of the concerns raised by this reviewer. Thus, I would like to recommend the revised manuscript for publication in its present form.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 91, 393, 107]]<|/det|>
+Reviewer #5 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 146, 875, 182]]<|/det|>
+Thank you for introducing clarifications in the manuscript. Please rephrase 'using four pooled siRNAs' to 'using a pool of four siRNAs'.  
+
+<|ref|>text<|/ref|><|det|>[[115, 221, 555, 238]]<|/det|>
+Manuscript has further improved and can be accepted now.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 309, 105]]<|/det|>
+Responses to Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[115, 118, 434, 135]]<|/det|>
+Authors' responses are indicated in red.  
+
+<|ref|>text<|/ref|><|det|>[[115, 149, 420, 166]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 179, 868, 289]]<|/det|>
+The study by Keren- Kaplan et al identifying 2 novel effectors of the small GTPase Arl8, namely RUY3 and RUY4 is interesting and novel in that it identifies these new effectors of ARL8 GTPase and links this GTPase to retrograde transport for the first time. As mentioned before in initial review the topic is important and the study itself is very well done with well- designed experiments, striking images and quantitative data. The authors have addressed the minor issues raised in initial review with additional experiments, quantitative data they have included and/or clarifications in text. I recommend publication of this interesting study.  
+
+<|ref|>text<|/ref|><|det|>[[115, 302, 420, 319]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 333, 355, 350]]<|/det|>
+The authors addressed my points  
+
+<|ref|>text<|/ref|><|det|>[[115, 363, 420, 380]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 393, 447, 410]]<|/det|>
+The authors have addressed all my comments.  
+
+<|ref|>text<|/ref|><|det|>[[115, 423, 420, 440]]<|/det|>
+Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 454, 839, 486]]<|/det|>
+In the revised manuscript, the authors properly addressed all of the concerns raised by this reviewer. Thus, I would like to recommend the revised manuscript for publication in its present form.  
+
+<|ref|>text<|/ref|><|det|>[[115, 499, 761, 517]]<|/det|>
+We are pleased to have been able to address all the comments by Reviewers #1- 4.  
+
+<|ref|>text<|/ref|><|det|>[[115, 530, 420, 547]]<|/det|>
+Reviewer #5 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 560, 864, 593]]<|/det|>
+Thank you for introducing clarifications in the manuscript. Please rephrase 'using four pooled siRNAs' to 'using a pool of four siRNAs'.  
+
+<|ref|>text<|/ref|><|det|>[[115, 606, 545, 623]]<|/det|>
+Manuscript has further improved and can be accepted now.  
+
+<|ref|>text<|/ref|><|det|>[[115, 636, 857, 670]]<|/det|>
+The text was changed as requested by this reviewer. We thank the review for recommending acceptance.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__722043b10d1bcf65fa536549eae28d4a71e0e9b474c02bc0420333a91997e235/images_list.json

@@ -0,0 +1,25 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_3b.jpg",
+    "caption": "Supplementary Fig.3b",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_7c.jpg",
+    "caption": "Supplementary Fig.7c",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        277,
+        850,
+        415
+      ]
+    ],
+    "page_idx": 2
+  }
+]

peer_reviews/supplementary_0_Peer Review File__722043b10d1bcf65fa536549eae28d4a71e0e9b474c02bc0420333a91997e235/supplementary_0_Peer Review File__722043b10d1bcf65fa536549eae28d4a71e0e9b474c02bc0420333a91997e235.mmd

@@ -0,0 +1,112 @@
+
+# nature portfolio  
+
+# Peer Review File  
+
+Environment- induced heritable variations are common in Arabidopsis thaliana  
+
+![](images/Supplementary_Figure_3b.jpg)
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to  
+
+the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## Response to reviewers' Comments:  
+
+## Reviewer #1:  
+
+"The authors reported transgenerational changes in several different \*Arabidopsis\* cultivars exposed to different stresses and found that most of the responses are genotype- dependent and correlate well with the transposon abundance in the genotypes. Overall, the work is convincing. I have some minor corrections/requests.  
+
+Response: Thank you for your positive comments and constructive suggestions. Following your suggestions, we have improved the method details and figures, and hope you will find the clarity of the manuscript further improved. Please refer to our point- to- point reply below, with comments in black, the reply in blue, and the revised texts highlighted.  
+
+Provide more detailed information (maybe I missed that) on the transposon occupancy in different cultivars, by type and genomic size. Show correlation for other transposons, for which the correlation was not as convincing."  
+
+Response: Thank you for the suggestion. For the information on transposon occupancy, please refer to Supplementary Table 1. We revised the table structure and referred to it in the main text (Line 139) to make it more accessible.  
+
+In this table, the number of transposons per superfamily is given, which was extracted from Quadrana. \*et al.\* eLife, 2016, 5:e15716. This number was originally estimated from Illumina read coverage (Lines 484- 488). To show the correlation for all transposon superfamilies, we revised the Supplementary Fig.3b (Test I) and Supplementary Fig. 7c (Test II). The corresponding correlation is also shown in Fig. 3b (Test I) and Supplementary Fig. 7b (Test II).
+
+<--- Page Split --->
+![](images/Supplementary_Figure_7c.jpg)
+
+<center>Supplementary Fig.3b </center>  
+
+![PLACEHOLDER_2_1]
+
+<center>Supplementary Fig.7c </center>  
+
+As the original paper did not provide size estimation, the transposon occupancy by genome size remained unknown. Recently, we reassembled two genotypes exhibiting the highest frequencies of induced variation, Abd- 0 and TRE- 1 (unpublished data), and found that the genome sizes are 135M for Abd- 0 and 138M for TRE- 1, much greater than reference genome Col- 0 (119 M, TAIR10), and even greater than a recent T- to- T version of Col- 0 (134 M, Hou et al., Mol. Plant, 2022, 15, 8). Although not perfectly estimated in this manuscript, we hope that our attempt to correlate transposon occupancy with induced heritable change still can serve as an incentive to bridge the fields of transposon and transgenerational effect studies, and to connect old ideas with new observations, offering valuable insights for future studies (Please refer to our reply to your points below).  
+
+"...The frequency of occurrence of fruit number was..." - delete "number"; is it actually just seed amount? "the frequency of occurrence of fruit" sounds weird - sounds like some plants had no fruits."
+
+<--- Page Split --->
+
+Response: Thanks for pointing out this issue. This is actually the number of siliques per plant. This phenotype is often adopted as a fitness proxy for A. thaliana because it is highly correlated with seed number, and counting the tiny and large quantity of Arabidopsis seeds is indeed challenging. To make this point clear, we have revised the sentence as follows: "The occurrence frequency estimated on the fitness proxy (=number of siliques per plant or fruit number) was the lowest among different phenotypes, marginally significantly lower than those of flowering time and plant height." (Lines 117- 120)  
+
+"the number of genes showing heritable expression changes can be predicted by the abundance of transposons in the genomes." - I would use the word "correlate"  
+
+Response: Thanks. We revised accordingly as: "...the number of genes showing heritable expression changes correlates with the abundance of transposons in the genomes." (Lines 221- 223)  
+
+"It was not very clear why you grown it for 7 generations and tested after 4th."  
+
+Response: Thanks. This is because there were two generations to remove the potential effect from seed storage and one to conduct the environmental treatments. Thus, there are three generations, and in 4th generation, we tested whether the treatments induced changes in descendants. To clarify the design, we revised the main text as follows: "To address the prevalence and predictability of environment- induced heritable changes, we subjected 14 natural accessions (genotypes) of Arabidopsis thaliana to the Control and ten environmental treatments to establish the ancestral generation. Before this generation, we planted the seeds of these accessions in the control environment for two generations to remove potential influence due to seed collection or storage. Then, we cultivated the offspring of the ancestral generation in the control environment to assess whether those environmental treatments induced phenotypic changes in offspring and
+
+<--- Page Split --->
+
+whether the induced changes were heritable over four offspring generations (Test I, Fig. 1, Supplementary Fig. 1a, Supplementary Table 1). "(Lines 50- 58)  
+
+"Please elaborate on pooling F1 – F4 seeds together for the Test II. Seeds from how many plants were pooled? By weight or by number?"  
+
+Response: Thank you for raising this point. Seeds from different plants or generations were not pooled, as we collected and reserved seeds for each plant and generation separately. We followed the single- seed descendant approach, that is to grow one individual descendent per plant per generation, to ensure there is no effect of selection. To clarify this point, we revised the methods as follows:  
+
+"Each treatment included six replicate plants per genotype, and the control included 12 plants per genotype, resulting in 1008 plants grown or lines established in the ancestral generation. Seeds were collected and reserved for each line, and in the offspring generation, we planted one descendant per line, following the single- seed descendant approach. In Test I, all 1008 lines were planted generation- by- generation in the control environment for four generations (F1- F4), and together 4032 plants were planted. For each offspring generation, seeds were collected and reserved for each line and generation separately. (Lines 343- 348)  
+
+Test II included a subset of ten genotypes (Abd- 0, Col- 0, Ang- 0, Dja- 1, Fei- 0, Ler- 0, Sapporo- 0, Tol- 0, TRE- 1 and Tri- 0), six different environments (the control, drought, low cadmium, high cadmium, low salinity and high salinity), and thus 360 lines (=10 genotypes \* 6 environments \* 6 replicates). For each line, we regenerated one plant from reserved seeds of the previous generation, and 1440 plants were grown to establish F1- F4. "(Lines 363- 365)  
+
+"A better Discussion about the role of transposons in the evolution is needed; there are reports demonstrating that transposons are activated by stress and genomic changes
+
+<--- Page Split --->
+
+prevail in transposons."  
+
+Response: Thank you for your constructive suggestions. Accordingly, we have revised the discussion section to incorporate evidence demonstrating that transposons become active under stress, leading to notable genomic changes. For the role of transposon, we highlighted their potential dual function as both selectors and inducers of genetic variation. Additionally, we have integrated suggestions from Reviewer #2. Please refer to the revised discussion section below:  
+
+...Transposons are mobile elements that constitute a significant proportion of eukaryotic genomes<sup>25</sup>. While these elements were once considered junk DNA, it became evident that they can contribute to the transcriptional binding region<sup>26</sup>, methylation loci<sup>21</sup>, transcriptional isoform sites<sup>27</sup>, and even coding sequences<sup>28</sup> to the host genomes. Barbara McClintock, who discovered transposons, proposed the idea of "genome shock," suggesting that stress can activate the movement of transposons, causing insertional mutations and structural variants<sup>29</sup>. This viewpoint has been increasingly supported by evidence that stress increases transposition rate<sup>30- 33</sup>, with the phenotypic effects of de novo transposon insertions including enhanced resistance of insects to pesticides<sup>34- 36</sup> and decreased sensitivity of plants to abiotic stress<sup>36- 39</sup>. The stress- induced transposition may be mediated by Hsp70 chaperone, a response factor to various biotic and abiotic stressors in plants and animals<sup>40,41</sup>, whose inducible expression could disturb the biogenesis of transposon- silencing piRNA<sup>42</sup>. These studies, along with our own, bridge and revive these previous ideas and explanations, suggesting that the environment is not only the selector but can also act as an inducer of genetic variation<sup>36,43</sup> (Lines 284- 299)  
+
+... Some notable examples of rapid evolution based on de novo transposon insertion include the industrial melanism mutation of the peppered moth<sup>23</sup> and the rapid increase in resistance to DDT in Drosophila melanogaster<sup>24</sup>. While many transposon insertions induced by environmental factors are deleterious, they enable the creation of alleles with significant, positively selected effects<sup>37,49,50</sup>. If the environment can act as both a
+
+<--- Page Split --->
+
+selector and an inducer of heritable variation, then the evolutionary speed constraints imposed by standing genetic variations will be liberated<sup>43,48</sup>. Therefore, exploring environmentally induced heritable variations and the molecular mechanisms related to transposon regulation is not only significant for fundamental research in genetics and evolution, but also crucial for assessing the threats of global changes and species responses.<sup>1</sup> (Lines 306- 316)
+
+<--- Page Split --->
+
+## Reviewer #2  
+
+"By such paper, the authors show that different types of environmental stress are capable to induce heritable phenotypic variation. Interestingly, it has also shown that such heritable changes are related to the abundance of transposons. Such data are interesting and significant together with other previous data, obtained on different organisms, that are changing the classical view on the environment as a major player only in selecting heritable changes. The very interesting point is that transposable elements are significant mediators in heritable changes induce by environmental stress. Such notion has been experimentally explored also in animal organisms. Among the several papers on such topic, I suggest to read some the following ones just to have a more complete view about the mechanisms involved on heritable changes produced by the environment through the activation of transposable elements:  
+
+Specchia V., L. Piacentini, P. Tritto, L. Fanti, R. D'Alessandro, G. Palumbo, S. Pimpinelli and MP. Bozzetti (2010). HSP90 prevents phenotypic variation by suppressing the mutagenic activity of transposons. Nature 463: 662- 665.  
+
+Piacentini L., Fanti L., Specchia V., Bozzetti MP., Berloco M., Palumbo G., Pimpinelli S. (2014) Transposons environmental changes and heritable induced phenotypic variability. Chromosoma. 123(4):345- 54.  
+
+Fanti L, Piacentini L, Cappucci U, Casale AM, Pimpinelli S. (2017) Canalization by Selection of de Novo Induced Mutations. Genetics. 206(4):1995- 2006.  
+
+Cappucci U, Noro F, Casale AM, Fanti L, Berloco M, Alagia AA, Grassi L, Le Pera L, Piacentini L, Pimpinelli S. (2019) The Hsp70 chaperone is a major player in stress- induced transposable element activation. Proc Natl Acad Sci U S A. 116(36):17943- 17950.  
+
+Pimpinelli S, Piacentini L. (2020) Environmental change and the evolution of genomes: Transposable elements as translators of phenotype plasticity into genotypic variability. Functional Ecology 34 (2), 426- 441.
+
+<--- Page Split --->
+
+In conclusion, the present work is well done and show very significant results. In my opinion, it deserves to be published."  
+
+Response: Thank you very much for your positive comments and for providing the references. These references have significantly expanded our knowledge of observations in various animals and plants, and the molecular mechanisms underlying stress- activating transposons. More importantly, these references have also summarized hallmark discoveries and bridges between those milestone ideas and concepts, offering thoughtful insights into this field. We have revised our discussion (please see below), and hope to put our findings into such a perspective.  
+
+...Transposons are mobile elements that constitute a significant proportion of eukaryotic genomes<sup>25</sup>. While these elements were once considered junk DNA, it became evident that they can contribute to the transcriptional binding region<sup>26</sup>, methylation loci<sup>21</sup>, transcriptional isoform sites<sup>27</sup>, and even coding sequences<sup>28</sup> to the host genomes. Barbara McClintock, who discovered transposons, proposed the idea of "genome shock," suggesting that stress can activate the movement of transposons, causing insertional mutations and structural variants<sup>29</sup>. This viewpoint has been increasingly supported by evidence that stress increases transposition rate<sup>30-33</sup>, with the phenotypic effects of de novo transposon insertions including enhanced resistance of insects to pesticides<sup>34-36</sup> and decreased sensitivity of plants to abiotic stress<sup>36-39</sup>. The stress- induced transposition may be mediated by Hsp70 chaperone, a response factor to various biotic and abiotic stressors in plants and animals<sup>40,41</sup>, whose inducible expression could disturb the biogenesis of transposon- silencing piRNA<sup>42</sup>. These studies, along with our own, bridge and revive these previous ideas and explanations, suggesting that the environment is not only the selector but can also act as an inducer of genetic variation<sup>36,43</sup> (Lines 284-299)
+
+<--- Page Split --->
+
+... Some notable examples of rapid evolution based on de novo transposon insertion include the industrial melanism mutation of the peppered moth \(^{23}\) and the rapid increase in resistance to DDT in *Drosophila melanogaster \(^{24}\) . While many transposon insertions induced by environmental factors are deleterious, they enable the creation of alleles with significant, positively selected effects \(^{37,49,50}\) . If the environment can act as both a selector and an inducer of heritable variation, then the evolutionary speed constraints imposed by standing genetic variations will be liberated \(^{43,48}\) . Therefore, exploring environmentally induced heritable variations and the molecular mechanisms related to transposon regulation is not only significant for fundamental research in genetics and evolution, but also crucial for assessing the threats of global changes and species responses. \(^{9}\) (Lines 306-316)
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__722043b10d1bcf65fa536549eae28d4a71e0e9b474c02bc0420333a91997e235/supplementary_0_Peer Review File__722043b10d1bcf65fa536549eae28d4a71e0e9b474c02bc0420333a91997e235_det.mmd

@@ -0,0 +1,143 @@
+<|ref|>title<|/ref|><|det|>[[61, 41, 507, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>title<|/ref|><|det|>[[68, 112, 348, 141]]<|/det|>
+# Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[68, 156, 928, 204]]<|/det|>
+Environment- induced heritable variations are common in Arabidopsis thaliana  
+
+<|ref|>image<|/ref|><|det|>[[56, 730, 240, 777]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[253, 732, 916, 784]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to  
+
+<|ref|>text<|/ref|><|det|>[[55, 784, 932, 923]]<|/det|>
+the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[149, 90, 501, 110]]<|/det|>
+## Response to reviewers' Comments:  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 124, 281, 143]]<|/det|>
+## Reviewer #1:  
+
+<|ref|>text<|/ref|><|det|>[[148, 159, 851, 266]]<|/det|>
+"The authors reported transgenerational changes in several different \*Arabidopsis\* cultivars exposed to different stresses and found that most of the responses are genotype- dependent and correlate well with the transposon abundance in the genotypes. Overall, the work is convincing. I have some minor corrections/requests.  
+
+<|ref|>text<|/ref|><|det|>[[147, 281, 851, 417]]<|/det|>
+Response: Thank you for your positive comments and constructive suggestions. Following your suggestions, we have improved the method details and figures, and hope you will find the clarity of the manuscript further improved. Please refer to our point- to- point reply below, with comments in black, the reply in blue, and the revised texts highlighted.  
+
+<|ref|>text<|/ref|><|det|>[[147, 466, 850, 544]]<|/det|>
+Provide more detailed information (maybe I missed that) on the transposon occupancy in different cultivars, by type and genomic size. Show correlation for other transposons, for which the correlation was not as convincing."  
+
+<|ref|>text<|/ref|><|det|>[[147, 560, 850, 636]]<|/det|>
+Response: Thank you for the suggestion. For the information on transposon occupancy, please refer to Supplementary Table 1. We revised the table structure and referred to it in the main text (Line 139) to make it more accessible.  
+
+<|ref|>text<|/ref|><|det|>[[147, 652, 851, 818]]<|/det|>
+In this table, the number of transposons per superfamily is given, which was extracted from Quadrana. \*et al.\* eLife, 2016, 5:e15716. This number was originally estimated from Illumina read coverage (Lines 484- 488). To show the correlation for all transposon superfamilies, we revised the Supplementary Fig.3b (Test I) and Supplementary Fig. 7c (Test II). The corresponding correlation is also shown in Fig. 3b (Test I) and Supplementary Fig. 7b (Test II).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[147, 88, 850, 234]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[405, 252, 590, 269]]<|/det|>
+<center>Supplementary Fig.3b </center>  
+
+<|ref|>image<|/ref|><|det|>[[147, 277, 850, 415]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[407, 425, 589, 444]]<|/det|>
+<center>Supplementary Fig.7c </center>  
+
+<|ref|>text<|/ref|><|det|>[[147, 460, 852, 770]]<|/det|>
+As the original paper did not provide size estimation, the transposon occupancy by genome size remained unknown. Recently, we reassembled two genotypes exhibiting the highest frequencies of induced variation, Abd- 0 and TRE- 1 (unpublished data), and found that the genome sizes are 135M for Abd- 0 and 138M for TRE- 1, much greater than reference genome Col- 0 (119 M, TAIR10), and even greater than a recent T- to- T version of Col- 0 (134 M, Hou et al., Mol. Plant, 2022, 15, 8). Although not perfectly estimated in this manuscript, we hope that our attempt to correlate transposon occupancy with induced heritable change still can serve as an incentive to bridge the fields of transposon and transgenerational effect studies, and to connect old ideas with new observations, offering valuable insights for future studies (Please refer to our reply to your points below).  
+
+<|ref|>text<|/ref|><|det|>[[147, 821, 850, 898]]<|/det|>
+"...The frequency of occurrence of fruit number was..." - delete "number"; is it actually just seed amount? "the frequency of occurrence of fruit" sounds weird - sounds like some plants had no fruits."
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 90, 852, 313]]<|/det|>
+Response: Thanks for pointing out this issue. This is actually the number of siliques per plant. This phenotype is often adopted as a fitness proxy for A. thaliana because it is highly correlated with seed number, and counting the tiny and large quantity of Arabidopsis seeds is indeed challenging. To make this point clear, we have revised the sentence as follows: "The occurrence frequency estimated on the fitness proxy (=number of siliques per plant or fruit number) was the lowest among different phenotypes, marginally significantly lower than those of flowering time and plant height." (Lines 117- 120)  
+
+<|ref|>text<|/ref|><|det|>[[148, 363, 850, 410]]<|/det|>
+"the number of genes showing heritable expression changes can be predicted by the abundance of transposons in the genomes." - I would use the word "correlate"  
+
+<|ref|>text<|/ref|><|det|>[[147, 425, 850, 505]]<|/det|>
+Response: Thanks. We revised accordingly as: "...the number of genes showing heritable expression changes correlates with the abundance of transposons in the genomes." (Lines 221- 223)  
+
+<|ref|>text<|/ref|><|det|>[[147, 554, 792, 574]]<|/det|>
+"It was not very clear why you grown it for 7 generations and tested after 4th."  
+
+<|ref|>text<|/ref|><|det|>[[147, 589, 852, 899]]<|/det|>
+Response: Thanks. This is because there were two generations to remove the potential effect from seed storage and one to conduct the environmental treatments. Thus, there are three generations, and in 4th generation, we tested whether the treatments induced changes in descendants. To clarify the design, we revised the main text as follows: "To address the prevalence and predictability of environment- induced heritable changes, we subjected 14 natural accessions (genotypes) of Arabidopsis thaliana to the Control and ten environmental treatments to establish the ancestral generation. Before this generation, we planted the seeds of these accessions in the control environment for two generations to remove potential influence due to seed collection or storage. Then, we cultivated the offspring of the ancestral generation in the control environment to assess whether those environmental treatments induced phenotypic changes in offspring and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 90, 848, 139]]<|/det|>
+whether the induced changes were heritable over four offspring generations (Test I, Fig. 1, Supplementary Fig. 1a, Supplementary Table 1). "(Lines 50- 58)  
+
+<|ref|>text<|/ref|><|det|>[[148, 189, 850, 237]]<|/det|>
+"Please elaborate on pooling F1 – F4 seeds together for the Test II. Seeds from how many plants were pooled? By weight or by number?"  
+
+<|ref|>text<|/ref|><|det|>[[147, 252, 852, 388]]<|/det|>
+Response: Thank you for raising this point. Seeds from different plants or generations were not pooled, as we collected and reserved seeds for each plant and generation separately. We followed the single- seed descendant approach, that is to grow one individual descendent per plant per generation, to ensure there is no effect of selection. To clarify this point, we revised the methods as follows:  
+
+<|ref|>text<|/ref|><|det|>[[147, 404, 852, 625]]<|/det|>
+"Each treatment included six replicate plants per genotype, and the control included 12 plants per genotype, resulting in 1008 plants grown or lines established in the ancestral generation. Seeds were collected and reserved for each line, and in the offspring generation, we planted one descendant per line, following the single- seed descendant approach. In Test I, all 1008 lines were planted generation- by- generation in the control environment for four generations (F1- F4), and together 4032 plants were planted. For each offspring generation, seeds were collected and reserved for each line and generation separately. (Lines 343- 348)  
+
+<|ref|>text<|/ref|><|det|>[[147, 641, 852, 805]]<|/det|>
+Test II included a subset of ten genotypes (Abd- 0, Col- 0, Ang- 0, Dja- 1, Fei- 0, Ler- 0, Sapporo- 0, Tol- 0, TRE- 1 and Tri- 0), six different environments (the control, drought, low cadmium, high cadmium, low salinity and high salinity), and thus 360 lines (=10 genotypes \* 6 environments \* 6 replicates). For each line, we regenerated one plant from reserved seeds of the previous generation, and 1440 plants were grown to establish F1- F4. "(Lines 363- 365)  
+
+<|ref|>text<|/ref|><|det|>[[148, 857, 850, 904]]<|/det|>
+"A better Discussion about the role of transposons in the evolution is needed; there are reports demonstrating that transposons are activated by stress and genomic changes
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[148, 91, 346, 108]]<|/det|>
+prevail in transposons."  
+
+<|ref|>text<|/ref|><|det|>[[147, 124, 852, 289]]<|/det|>
+Response: Thank you for your constructive suggestions. Accordingly, we have revised the discussion section to incorporate evidence demonstrating that transposons become active under stress, leading to notable genomic changes. For the role of transposon, we highlighted their potential dual function as both selectors and inducers of genetic variation. Additionally, we have integrated suggestions from Reviewer #2. Please refer to the revised discussion section below:  
+
+<|ref|>text<|/ref|><|det|>[[147, 303, 852, 760]]<|/det|>
+...Transposons are mobile elements that constitute a significant proportion of eukaryotic genomes<sup>25</sup>. While these elements were once considered junk DNA, it became evident that they can contribute to the transcriptional binding region<sup>26</sup>, methylation loci<sup>21</sup>, transcriptional isoform sites<sup>27</sup>, and even coding sequences<sup>28</sup> to the host genomes. Barbara McClintock, who discovered transposons, proposed the idea of "genome shock," suggesting that stress can activate the movement of transposons, causing insertional mutations and structural variants<sup>29</sup>. This viewpoint has been increasingly supported by evidence that stress increases transposition rate<sup>30- 33</sup>, with the phenotypic effects of de novo transposon insertions including enhanced resistance of insects to pesticides<sup>34- 36</sup> and decreased sensitivity of plants to abiotic stress<sup>36- 39</sup>. The stress- induced transposition may be mediated by Hsp70 chaperone, a response factor to various biotic and abiotic stressors in plants and animals<sup>40,41</sup>, whose inducible expression could disturb the biogenesis of transposon- silencing piRNA<sup>42</sup>. These studies, along with our own, bridge and revive these previous ideas and explanations, suggesting that the environment is not only the selector but can also act as an inducer of genetic variation<sup>36,43</sup> (Lines 284- 299)  
+
+<|ref|>text<|/ref|><|det|>[[147, 775, 850, 910]]<|/det|>
+... Some notable examples of rapid evolution based on de novo transposon insertion include the industrial melanism mutation of the peppered moth<sup>23</sup> and the rapid increase in resistance to DDT in Drosophila melanogaster<sup>24</sup>. While many transposon insertions induced by environmental factors are deleterious, they enable the creation of alleles with significant, positively selected effects<sup>37,49,50</sup>. If the environment can act as both a
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 88, 852, 255]]<|/det|>
+selector and an inducer of heritable variation, then the evolutionary speed constraints imposed by standing genetic variations will be liberated<sup>43,48</sup>. Therefore, exploring environmentally induced heritable variations and the molecular mechanisms related to transposon regulation is not only significant for fundamental research in genetics and evolution, but also crucial for assessing the threats of global changes and species responses.<sup>1</sup> (Lines 306- 316)
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[148, 90, 274, 109]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[147, 124, 853, 433]]<|/det|>
+"By such paper, the authors show that different types of environmental stress are capable to induce heritable phenotypic variation. Interestingly, it has also shown that such heritable changes are related to the abundance of transposons. Such data are interesting and significant together with other previous data, obtained on different organisms, that are changing the classical view on the environment as a major player only in selecting heritable changes. The very interesting point is that transposable elements are significant mediators in heritable changes induce by environmental stress. Such notion has been experimentally explored also in animal organisms. Among the several papers on such topic, I suggest to read some the following ones just to have a more complete view about the mechanisms involved on heritable changes produced by the environment through the activation of transposable elements:  
+
+<|ref|>text<|/ref|><|det|>[[147, 449, 850, 526]]<|/det|>
+Specchia V., L. Piacentini, P. Tritto, L. Fanti, R. D'Alessandro, G. Palumbo, S. Pimpinelli and MP. Bozzetti (2010). HSP90 prevents phenotypic variation by suppressing the mutagenic activity of transposons. Nature 463: 662- 665.  
+
+<|ref|>text<|/ref|><|det|>[[147, 542, 850, 618]]<|/det|>
+Piacentini L., Fanti L., Specchia V., Bozzetti MP., Berloco M., Palumbo G., Pimpinelli S. (2014) Transposons environmental changes and heritable induced phenotypic variability. Chromosoma. 123(4):345- 54.  
+
+<|ref|>text<|/ref|><|det|>[[147, 634, 850, 682]]<|/det|>
+Fanti L, Piacentini L, Cappucci U, Casale AM, Pimpinelli S. (2017) Canalization by Selection of de Novo Induced Mutations. Genetics. 206(4):1995- 2006.  
+
+<|ref|>text<|/ref|><|det|>[[147, 699, 850, 804]]<|/det|>
+Cappucci U, Noro F, Casale AM, Fanti L, Berloco M, Alagia AA, Grassi L, Le Pera L, Piacentini L, Pimpinelli S. (2019) The Hsp70 chaperone is a major player in stress- induced transposable element activation. Proc Natl Acad Sci U S A. 116(36):17943- 17950.  
+
+<|ref|>text<|/ref|><|det|>[[147, 821, 850, 897]]<|/det|>
+Pimpinelli S, Piacentini L. (2020) Environmental change and the evolution of genomes: Transposable elements as translators of phenotype plasticity into genotypic variability. Functional Ecology 34 (2), 426- 441.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 90, 850, 137]]<|/det|>
+In conclusion, the present work is well done and show very significant results. In my opinion, it deserves to be published."  
+
+<|ref|>text<|/ref|><|det|>[[147, 188, 852, 383]]<|/det|>
+Response: Thank you very much for your positive comments and for providing the references. These references have significantly expanded our knowledge of observations in various animals and plants, and the molecular mechanisms underlying stress- activating transposons. More importantly, these references have also summarized hallmark discoveries and bridges between those milestone ideas and concepts, offering thoughtful insights into this field. We have revised our discussion (please see below), and hope to put our findings into such a perspective.  
+
+<|ref|>text<|/ref|><|det|>[[147, 398, 852, 852]]<|/det|>
+...Transposons are mobile elements that constitute a significant proportion of eukaryotic genomes<sup>25</sup>. While these elements were once considered junk DNA, it became evident that they can contribute to the transcriptional binding region<sup>26</sup>, methylation loci<sup>21</sup>, transcriptional isoform sites<sup>27</sup>, and even coding sequences<sup>28</sup> to the host genomes. Barbara McClintock, who discovered transposons, proposed the idea of "genome shock," suggesting that stress can activate the movement of transposons, causing insertional mutations and structural variants<sup>29</sup>. This viewpoint has been increasingly supported by evidence that stress increases transposition rate<sup>30-33</sup>, with the phenotypic effects of de novo transposon insertions including enhanced resistance of insects to pesticides<sup>34-36</sup> and decreased sensitivity of plants to abiotic stress<sup>36-39</sup>. The stress- induced transposition may be mediated by Hsp70 chaperone, a response factor to various biotic and abiotic stressors in plants and animals<sup>40,41</sup>, whose inducible expression could disturb the biogenesis of transposon- silencing piRNA<sup>42</sup>. These studies, along with our own, bridge and revive these previous ideas and explanations, suggesting that the environment is not only the selector but can also act as an inducer of genetic variation<sup>36,43</sup> (Lines 284-299)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 88, 852, 400]]<|/det|>
+... Some notable examples of rapid evolution based on de novo transposon insertion include the industrial melanism mutation of the peppered moth \(^{23}\) and the rapid increase in resistance to DDT in *Drosophila melanogaster \(^{24}\) . While many transposon insertions induced by environmental factors are deleterious, they enable the creation of alleles with significant, positively selected effects \(^{37,49,50}\) . If the environment can act as both a selector and an inducer of heritable variation, then the evolutionary speed constraints imposed by standing genetic variations will be liberated \(^{43,48}\) . Therefore, exploring environmentally induced heritable variations and the molecular mechanisms related to transposon regulation is not only significant for fundamental research in genetics and evolution, but also crucial for assessing the threats of global changes and species responses. \(^{9}\) (Lines 306-316)
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__72255fb6ea7d72f2c8cbb94b80ae1203d10be66cb5998e4b40792b5d9c63447d/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__72414ca1b2a55915dfb98e7ef968c43a022581b3189cee78b898486569de958e/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__724ad6b2abe07ac657a748ae8fb931984135b3acbf597fb4f065a761d230f108/images_list.json

@@ -0,0 +1,109 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Reviewer Figure 1. The network model developed by Huang et al. has strong spiking coherence consistent with our definition of dense waves. Our sparse-wave model (purple line, left panel) does not induce strong spiking coherence, whereas our smaller-scale dense wave producing model (black line) has a significant peak in spiking coherence. The right panel shows the spiking coherence calculated from a simulation generated from the code published by Huang et al. corresponding to the model in their Figure 3Aiv. The large change in spiking probability in the Huang model is consistent with our definition of dense waves. Dotted lines denote the 95% confidence interval.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure S2. Distance dependence of pairwise spike coherence. (a) The maximum pairwise spike coherence calculated between neuron pools at various distances in the random (black) and topographic (red) large-scale networks in Figure 2. There was no change in spike coherence in either network at any distance. (b) Same as (a), but for the small-scale networks in Figure S1. There was a negative correlation with maximum spike coherence and distance in the small-scale topographic model (Pearson's \\(r = -0.72\\) ).",
+    "footnote": [],
+    "bbox": [
+      [
+        240,
+        496,
+        758,
+        651
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Figure S4. Delays are necessary for robust traveling waves in 1-D spiking network model. (a) Schematic of 1-D network model. 450,000 neurons were arranged on a ring with topographic connection probabilities and distance dependent delays. (b) 2-D FFT of the spatial (y-axis) and temporal (x-axis) frequencies of activity in the topographic network. The clear spectral line is consistent with waves traveling at 0.2 m/s. (c) No spectral line appears in a similar topographic 1D network without delays. (d) Spike rasters and LFP amplitude (pseudocolor) for the topographic network displays waves moving across space over time in the 1-D topographically connected network with delays. (e) Same as (d), but for the 1-D topographic network without delays. LFP fluctuations do not travel as waves but rather occur synchronously across regions of the network.",
+    "footnote": [],
+    "bbox": [
+      [
+        220,
+        128,
+        780,
+        510
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Figure S3. Topographic connections and distance-dependent delays combined are necessary to generate traveling waves. (a) Significant (white) and non-significant (black) wavelength values for each position in a linear slice through a large-scale 2D network simulation with topographic connections and no delays. (a') 2-D (space-time) FFT shows a concentration of spectral power corresponding to waves traveling at the velocity corresponding to propagation speeds (0.2 m/s). (b, b') Same as in (a), but for a network with topographic connectivity and no delays. Topographic connectivity is sufficient to generate significant spatially organized wavelengths. However, without delays, the spectral power does not concentrate along a joint spatial and temporal frequency band consistent with traveling waves. (c, c') Wavelengths and spatiotemporal FFT for a randomly connected network with delays. With random connectivity the network lacks strong spatial organization while delays are sufficient for the spatiotemporal flow of activity. (d, d'). Wavelengths and spatiotemporal FFT for a randomly connected network without delays. There is no spatial or temporal structure in this network suggestive of any wave activity.",
+    "footnote": [],
+    "bbox": [
+      [
+        225,
+        123,
+        770,
+        580
+      ]
+    ],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Figure S3. Topographic connections and distance-dependent delays combined are necessary to generate traveling waves. (a) Significant (white) and non-significant (black) wavelength values for each position in a linear slice through a large-scale 2D network simulation with topographic connections and no delays. (a') 2-D (space-time) FFT shows a concentration of spectral power corresponding to waves traveling at the velocity corresponding to propagation speeds (0.2 m/s). (b, b') Same as in (a), but for a network with topographic connectivity and no delays. Topographic connectivity is sufficient to generate significant spatially organized wavelengths. However, without delays, the spectral power does not concentrate along a joint spatial and temporal frequency band consistent with traveling waves. (c, c') Wavelengths and spatiotemporal FFT for a randomly connected network with delays. With random connectivity the network lacks strong spatial organization while delays are sufficient for the spatiotemporal flow of activity. (d, d'). Wavelengths and spatiotemporal FFT for a randomly connected network without delays. There is no spatial or temporal structure in this network suggestive of any wave activity.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Figure S4. Delays are necessary for robust traveling waves in 1-D spiking network model. (a) Schematic of 1-D network model. 450,000 neurons were arranged on a ring with topographic connection probabilities and distance dependent delays. (b) 2-D FFT of the spatial (y-axis) and temporal (x-axis) frequencies of activity in the topographic network. The clear spectral line is consistent with waves traveling at 0.2 m/s. (c) No spectral line appears in a similar topographic 1D network without delays. (d) Spike rasters and LFP amplitude (pseudocolor) for the topographic network displays waves moving across space over time in the 1-D topographically connected network with delays. (e) Same as (d), but for the 1-D topographic network without delays. LFP fluctuations do not travel as waves but rather occur synchronously across regions of the network.",
+    "footnote": [],
+    "bbox": [
+      [
+        220,
+        128,
+        780,
+        519
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Reviewer Figure 2. Estimated probability density functions of the combined excitatory and inhibitory conductances (ge+gi) for ten neurons in the dense (blue) and sparse (orange) networks.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 20
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2a.jpg",
+    "caption": "Figure S5. Randomly-connected spiking network model has weak spike-LFP phase coupling. Histogram showing the fraction of spikes that occurred during each phase of the LFP in the randomly connected network shown in Figure 2a (spike-phase index = 0.03).",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 23
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Reviewer Figure 3. Waves persist in a topographic spiking network model receiving feed-forward spiking input analogous to thalamocortical driving input. (a) A network simulation where the majority of the spiking activity was driven by feed-forward spiking inputs arriving analogous to a thalamic source. The mean firing rate of the network ( \\(\\sim 4.5 \\mathrm{~Hz}\\) ) during feed-forward input was significantly reduced when thalamic input was turned off, with a reduction similar in magnitude to that observed during experimental thalamic inactivation ( \\(\\sim 75\\%\\) reduction). (b) LFP amplitude (z-score) from a 1-D slice through the network during the period before and after removal of the driving inputs. Waves can be seen in both conditions as angled peaks in the LFP moving across the spatial extent of the network over time (examples indicated by arrows). (c) 2-D FFT during the period of feed-forward driving input reveals a band of spatiotemporal power consistent with the presence of traveling waves. (d) Same as in (c), but after feed-forward driving input is abolished. Waves are still present when the reduced network activity is entirely intrinsically driven.",
+    "footnote": [],
+    "bbox": [
+      [
+        184,
+        130,
+        800,
+        608
+      ]
+    ],
+    "page_idx": 24
+  }
+]

peer_reviews/supplementary_0_Peer Review File__72c9240b12f1e51345061f49bd71ff2dee4711feb40c1e8636f3ae3f31a9e972/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__72c9240b12f1e51345061f49bd71ff2dee4711feb40c1e8636f3ae3f31a9e972/supplementary_0_Peer Review File__72c9240b12f1e51345061f49bd71ff2dee4711feb40c1e8636f3ae3f31a9e972.mmd

@@ -0,0 +1,46 @@
+
+Editorial Note: This manuscript has been previously reviewed at another journal that is not operating a transparent peer review scheme. This document only contains reviewer comments and rebuttal letters for versions considered at Nature Communications.  
+
+Reviewers' Comments:  
+
+Reviewer #1:  
+
+Remarks to the Author:  
+
+Most of the previous comments and suggested corrections have been addressed. Unfortunately, the suggested additional experiments have not been performed.  
+
+Regarding "More insights on the dephasing processes could be obtained by performing temperature- dependent studies, especially below \(1 \text{K}\) , where dephasing processes should be significantly reduced." It is a pity that the authors rejected to include such important studies in the paper which definitely would have made the manuscript stronger. At the minimum, it would be good to include after the above sentence that studies below \(1 \text{K}\) are not within the scope of this contribution as those are rather time- consuming and elaborate. This is better than just postulating as this confuses the reader and raises the question why such studies have not been done then if those would have been so useful.  
+
+Regarding "The usefulness of such a transition for QIP applications could be enhanced by means of molecular engineering approaches. The tunable nature of molecular properties with atomic precision could provide access to quantum materials with technologically relevant spin and optical coherence lifetimes." These two sentences read too vague and cannot be backed- up currently as there are no molecular Eu systems known yet that have been implemented in applications. In addition, the optical coherence time (T2opt) determined for [Eu2] in this current manuscript is multiple orders of magnitude smaller than what has been observed in europium- doped system. Thus, the magnitude of the optical coherence time observed for [Eu2] is not special here, only the fact that an optical coherence time has been observed for the first time in a molecular system. The next logical step would be to design molecules that show optical coherence times comparable to or ideally even better than what has been demonstrated in europium- doped systems. Thus, I suggest rephrasing both sentences and lessening the "usefulness" reasoning as I think there is a long way to go until something like that could be really implemented for QIP applications.  
+
+After addressing my relatively minor comments, the manuscript may be considered for publication in Nature Communications.
+
+<--- Page Split --->
+
+## Answers to the comments  
+
+Note: The new additions are highlighted in yellow in the revised script. The revised script was edited by switching on the track change mode ON. The corrections could be seen by putting the script in the "All Markup" view in the review panel of MS Word.  
+
+Comment: Most of the previous comments and suggested corrections have been addressed.  
+
+Answer: We thank the referee for the constructive criticism and comments, which helped us to improve the quality and presentation of the script.  
+
+Comment: Unfortunately, the suggested additional experiments have not been performed. Regarding "More insights on the dephasing processes could be obtained by performing temperature- dependent studies, especially below 1 K, where dephasing processes should be significantly reduced." It is a pity that the authors rejected to include such important studies in the paper which definitely would have made the manuscript stronger. At the minimum, it would be good to include after the above sentence that studies below 1 K are not within the scope of this contribution as those are rather time- consuming and elaborate. This is better than just postulating as this confuses the reader and raises the question why such studies have not been done then if those would have been so useful.  
+
+Answer: We have added the following sentence, as suggested by the referee:  
+
+"However, experiments below 1 K require highly specialized equipment namely dilution refrigerators using \(\mathrm{He}_3 / \mathrm{He}_4\) mixtures with optical access. Although studying dephasing under these conditions is useful, it is out of the scope of the present study."  
+
+Comment: Regarding "The usefulness of such a transition for QIP applications could be enhanced by means of molecular engineering approaches. The tunable nature of molecular properties with atomic
+
+<--- Page Split --->
+
+precision could provide access to quantum materials with technologically relevant spin and optical coherence lifetimes." These two sentences read too vague and cannot be backed- up currently as there are no molecular Eu systems known yet that have been implemented in applications. In addition, the optical coherence time (T2opt) determined for [Eu2] in this current manuscript is multiple orders of magnitude smaller than what has been observed in europium- doped system. Thus, the magnitude of the optical coherence time observed for [Eu2] is not special here, only the fact that an optical coherence time has been observed for the first time in a molecular system. The next logical step would be to design molecules that show optical coherence times comparable to or ideally even better than what has been demonstrated in europium- doped systems. Thus, I suggest rephrasing both sentences and lessening the "usefulness" reasoning as I think there is a long way to go until something like that could be really implemented for QIP applications.  
+
+After addressing my relatively minor comments, the manuscript may be considered for publication in Nature Communications.  
+
+Answer: We have rephrased the sentence following the referee comment:  
+
+"To progress towards quantum technology applications, REI- based molecular systems featuring optical coherence lifetime larger than the \(\mathrm{T}_{2\mathrm{opt}} = 14.5 \pm 0.7\) ns reported for [Eu2] in this study should be obtained. The tunable nature of quantum properties of molecules with precision could provide access to REI- molecule- based quantum materials with technologically relevant spin and optical coherence lifetimes."
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__72c9240b12f1e51345061f49bd71ff2dee4711feb40c1e8636f3ae3f31a9e972/supplementary_0_Peer Review File__72c9240b12f1e51345061f49bd71ff2dee4711feb40c1e8636f3ae3f31a9e972_det.mmd

@@ -0,0 +1,63 @@
+<|ref|>text<|/ref|><|det|>[[119, 85, 867, 137]]<|/det|>
+Editorial Note: This manuscript has been previously reviewed at another journal that is not operating a transparent peer review scheme. This document only contains reviewer comments and rebuttal letters for versions considered at Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[119, 177, 293, 191]]<|/det|>
+Reviewers' Comments:  
+
+<|ref|>text<|/ref|><|det|>[[119, 207, 222, 220]]<|/det|>
+Reviewer #1:  
+
+<|ref|>text<|/ref|><|det|>[[119, 222, 300, 235]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[119, 237, 870, 266]]<|/det|>
+Most of the previous comments and suggested corrections have been addressed. Unfortunately, the suggested additional experiments have not been performed.  
+
+<|ref|>text<|/ref|><|det|>[[118, 267, 877, 386]]<|/det|>
+Regarding "More insights on the dephasing processes could be obtained by performing temperature- dependent studies, especially below \(1 \text{K}\) , where dephasing processes should be significantly reduced." It is a pity that the authors rejected to include such important studies in the paper which definitely would have made the manuscript stronger. At the minimum, it would be good to include after the above sentence that studies below \(1 \text{K}\) are not within the scope of this contribution as those are rather time- consuming and elaborate. This is better than just postulating as this confuses the reader and raises the question why such studies have not been done then if those would have been so useful.  
+
+<|ref|>text<|/ref|><|det|>[[117, 387, 877, 580]]<|/det|>
+Regarding "The usefulness of such a transition for QIP applications could be enhanced by means of molecular engineering approaches. The tunable nature of molecular properties with atomic precision could provide access to quantum materials with technologically relevant spin and optical coherence lifetimes." These two sentences read too vague and cannot be backed- up currently as there are no molecular Eu systems known yet that have been implemented in applications. In addition, the optical coherence time (T2opt) determined for [Eu2] in this current manuscript is multiple orders of magnitude smaller than what has been observed in europium- doped system. Thus, the magnitude of the optical coherence time observed for [Eu2] is not special here, only the fact that an optical coherence time has been observed for the first time in a molecular system. The next logical step would be to design molecules that show optical coherence times comparable to or ideally even better than what has been demonstrated in europium- doped systems. Thus, I suggest rephrasing both sentences and lessening the "usefulness" reasoning as I think there is a long way to go until something like that could be really implemented for QIP applications.  
+
+<|ref|>text<|/ref|><|det|>[[118, 581, 870, 610]]<|/det|>
+After addressing my relatively minor comments, the manuscript may be considered for publication in Nature Communications.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[70, 60, 290, 75]]<|/det|>
+## Answers to the comments  
+
+<|ref|>text<|/ref|><|det|>[[68, 102, 917, 191]]<|/det|>
+Note: The new additions are highlighted in yellow in the revised script. The revised script was edited by switching on the track change mode ON. The corrections could be seen by putting the script in the "All Markup" view in the review panel of MS Word.  
+
+<|ref|>text<|/ref|><|det|>[[68, 216, 830, 235]]<|/det|>
+Comment: Most of the previous comments and suggested corrections have been addressed.  
+
+<|ref|>text<|/ref|><|det|>[[68, 261, 925, 313]]<|/det|>
+Answer: We thank the referee for the constructive criticism and comments, which helped us to improve the quality and presentation of the script.  
+
+<|ref|>text<|/ref|><|det|>[[66, 339, 916, 602]]<|/det|>
+Comment: Unfortunately, the suggested additional experiments have not been performed. Regarding "More insights on the dephasing processes could be obtained by performing temperature- dependent studies, especially below 1 K, where dephasing processes should be significantly reduced." It is a pity that the authors rejected to include such important studies in the paper which definitely would have made the manuscript stronger. At the minimum, it would be good to include after the above sentence that studies below 1 K are not within the scope of this contribution as those are rather time- consuming and elaborate. This is better than just postulating as this confuses the reader and raises the question why such studies have not been done then if those would have been so useful.  
+
+<|ref|>text<|/ref|><|det|>[[68, 628, 707, 646]]<|/det|>
+Answer: We have added the following sentence, as suggested by the referee:  
+
+<|ref|>text<|/ref|><|det|>[[68, 671, 896, 760]]<|/det|>
+"However, experiments below 1 K require highly specialized equipment namely dilution refrigerators using \(\mathrm{He}_3 / \mathrm{He}_4\) mixtures with optical access. Although studying dephasing under these conditions is useful, it is out of the scope of the present study."  
+
+<|ref|>text<|/ref|><|det|>[[68, 821, 900, 874]]<|/det|>
+Comment: Regarding "The usefulness of such a transition for QIP applications could be enhanced by means of molecular engineering approaches. The tunable nature of molecular properties with atomic
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[65, 58, 930, 426]]<|/det|>
+precision could provide access to quantum materials with technologically relevant spin and optical coherence lifetimes." These two sentences read too vague and cannot be backed- up currently as there are no molecular Eu systems known yet that have been implemented in applications. In addition, the optical coherence time (T2opt) determined for [Eu2] in this current manuscript is multiple orders of magnitude smaller than what has been observed in europium- doped system. Thus, the magnitude of the optical coherence time observed for [Eu2] is not special here, only the fact that an optical coherence time has been observed for the first time in a molecular system. The next logical step would be to design molecules that show optical coherence times comparable to or ideally even better than what has been demonstrated in europium- doped systems. Thus, I suggest rephrasing both sentences and lessening the "usefulness" reasoning as I think there is a long way to go until something like that could be really implemented for QIP applications.  
+
+<|ref|>text<|/ref|><|det|>[[68, 440, 901, 494]]<|/det|>
+After addressing my relatively minor comments, the manuscript may be considered for publication in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[68, 520, 682, 538]]<|/det|>
+Answer: We have rephrased the sentence following the referee comment:  
+
+<|ref|>text<|/ref|><|det|>[[66, 563, 928, 722]]<|/det|>
+"To progress towards quantum technology applications, REI- based molecular systems featuring optical coherence lifetime larger than the \(\mathrm{T}_{2\mathrm{opt}} = 14.5 \pm 0.7\) ns reported for [Eu2] in this study should be obtained. The tunable nature of quantum properties of molecules with precision could provide access to REI- molecule- based quantum materials with technologically relevant spin and optical coherence lifetimes."
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__72f06a0c6f99ff975ae8e2e989c5343b6bc0a33378da8598ab6e8c6c95628b7c/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__72f06a0c6f99ff975ae8e2e989c5343b6bc0a33378da8598ab6e8c6c95628b7c/supplementary_0_Peer Review File__72f06a0c6f99ff975ae8e2e989c5343b6bc0a33378da8598ab6e8c6c95628b7c.mmd

@@ -0,0 +1,686 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+SurVIndel2: improving copy number variant calling from next- generation sequencing using hidden split reads  
+
+![PLACEHOLDER_0_0]
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors describe surVIndel2, a tool for CNV detection, leveraging a previously underappreciated signal - the so- called hidden split reads. This innovative approach appears successful in identifying CNVs, especially in a tandem repeat context, where typical split reads fail to identify the presence of a CNV. The observation that many CNVs are missed due to poor coverage is particularly intriguing, and I wonder if sequence- specific features beyond repetitive context contribute to this. The performance of this tool, compared to well- established tools, is impressive, and I wish to congratulate the authors on their work.  
+
+Sincerely, Wouter De Coster  
+
+Major:  
+
+I see the authors use SurVCluster to compare SVs from various callset, but the manuscript could be improved by more information on how exactly that is done, and how the precision and sensitivity are calculated based on that. Note that a commonly accepted method for comparing structural variant call sets is Truvari (https://github.com/ACEnglish/truvari). Can the authors confirm that they obtained similar results with their tool?  
+
+Minor:  
+
+It appears the authors use the 'HSR' abbreviation and only describe what it means later in the text.  
+
+The y- axis in Figure 1c and 1d is unlabeled.  
+
+There is no legend for the bar colors in Figure 5b, but presumably, the colors are the same as for 5a. However, this is not explicitly stated and could lead to confusion.  
+
+Mostly out of curiosity and not crucial for this paper, I wondered if it is possible to train the filter module on a combination of different species and what the performance of such a mixed model would be compared to the species- specific filter.  
+
+Reviewer #2 (Remarks to the Author):  
+
+This work presents a novel approach, SurVIndel2, to detect copy number variations (CNVs). It looks for sequences that can be aligned to alternative positions to the reference genome as split reads (referred to as hidden split read) and demonstrates enhanced sensitivity and precision from short- read sequencing data, particularly improving detection of duplication in repetitive regions of the genome.  
+
+Although long- read sequencing can significantly improve CNV calling in repetitive regions, this work acknowledges that short- read sequencing will remain the primary research data in the near future and therefore addressing the limitation of detecting CNVs in repetitive regions using short- read sequencing is of great relevance.  
+
+The manuscript provides an evaluation of SurVIndel2 with public datasets of multiple organisms and benchmarks its performance with existing CNV and SV callers. However, further analyses and validation on existing tools could strengthen the claims. Please see the list of revisions below for details.  
+
+The analysis appears methodologically sound overall, but a more detailed discussion on limitations and
+
+<--- Page Split --->
+
+potential biases would be beneficial. Any flaws do not seem to prohibit publication but may require revisions (see the list of revisions below) for clarity and completeness.  
+
+The manuscript clearly described the methodology and provides multiple benchmarking analyses of the method on cell line samples. I believe the work meets current standards in bioinformatics for CNV detection, given the list of revisions are addressed.  
+
+While SurVIndel2 is open- source software, for full reproducibility, I suggest also including scripts for running the analyses.  
+
+List of Revisions:  
+
+1. Figure 2 shows that over \(93\%\) of the tandem duplications of HG002 are in repeat regions. It would be good to see a breakdown of the types of repeat regions in the bar chart.  
+
+2. In section 2.3, the manuscript stated that filtering was implemented by leave-1-out training for SurVIndel2, but methods of the 'pre-trained random forest' filtering are not provided. Please include further description.  
+
+3. In addition to revision 2, section 2.3 didn't include descriptions of how filtering was performed on CNV calls of benchmarking tools. For example, Manta ranks the quality of SV calls using QUAL score and FILTER tags, and the choice of filtering strategy can impact the benchmarking results, particularly sensitivity. Figure 4 was generated with one set of calls per sample per caller. Possibilities of alternative filtering strategies were not explored. I recommend including multiple filtering options in the benchmarking, such as different QUAL cutoffs per sample per caller, and plot curves of precision-recall in figure 4(a), 4(b), and 5(a).  
+
+4. HG002 is a great independent sample, which was not specifically trained on by SurVIndel2, for benchmarking purposes. However, the discussion only includes sensitivity, not precision. I strongly recommend including HG002 in Figure 4(a) and 4(b) as well.  
+
+5. Very importantly, in section 2.3, Delly, Lumpy, and Manta were selected as benchmarking tools according to reference [4]. However, in this benchmarking study, GRIDSS is another top-performing SV caller. I strongly recommend including benchmarking with GRIDSS2 + Purple (Cameron, D.L., Baber, J., Shale, C. et al. GRIDSS2: comprehensive characterization of somatic structural variation using single breached variants and structural variant phasing. Genome Biol 22, 202 (2021). https://doi.org/10.1186/s13059-021-02423-x).  
+
+6. It's interesting to see that SurVIndel 1&2 are more sensitive than other callers with no HSR support. Section 2.7 discusses the reasons why these CNVs have a low level of evidence. It would be good to include more discussion on why SurVIndel2 can capture more such CNVs than other callers, as well as the true positive rate of unsupported CNVs called by SurVIndel2.  
+
+7. I recommend including a discussion on the limitations of the tool regarding the detection of somatic CNVs in cancer genomes. This discussion could acknowledge that, unlike the other tools benchmarked in the study, SurVIndel2 does not currently support the identification of somatic CNVs, which are critical for understanding cancer genomics. Highlight this as a significant caveat, and suggest future research directions or potential updates to the tool that could address this gap. Emphasize the importance of this functionality for comprehensive cancer genomics studies and how its inclusion could further elevate the tool's applicability and utility in the field. This addition will not only provide a more balanced view of SurVIndel2's capabilities but also align expectations for researchers interested in cancer genomics.  
+
+Reviewer #3 (Remarks to the Author):  
+
+This manuscript introduces a new tool SurVIndel2 for detecting the copy number variations (CNV) using short illumina sequencing reads with a novel technique that uses hidden split reads and machine learning techniques. The benchmarking results showed that SurVIndel2 performed better than other
+
+<--- Page Split --->
+
+short- read- based structural variant (SV) callers. However, concerns arise regarding the benchmarking evaluation methods and the availability of results produced in this study.  
+
+Major issues:  
+
+1. Throughout the manuscript, it seems the authors have considered all deletion (>=50bp) and tandem duplication type SVs as CNVs. There is confusion as all deletions (>=50bp) can not be called CNV. Is the tool designed to detect all DEL and DUP-type SVs? If that is the case, then it is not suggested to call it a CNV caller tool.  
+
+2. The authors have claimed that it is a new short-read-based CNV caller, however, in the last few steps it uses the machine learning technique that takes advantage of long-read-based CNV callers (steps g and h) for training purposes. The machine learning model of SurVIndel2 is used in these steps, right? So would it be fair to call it a short-read-based caller? It is expected to work a little better when long-reads or long-read-based callsets are used at the filtering stage, so it won't be a fair comparison with other "pure" short-read-based callers such as Manta, Lumpy, and Delly. Also, Lumpy only calls the DEL-type SVs, so it may not detect any INS (or DUP) type SVs.  
+
+3. The "Method" section is included inside the "Result" section. Also, that section only talks about the algorithms and techniques used in this tool. It would be better to include the benchmarking strategies such as what comparison tools and the truth set are used for evaluation. The results can be reproduced for verification if the VCF file for the truth sets, callsets, commands, or scripts are available in the "Methods" section. The authors could provide a link (e.g. GitHub repo) with all the result files.  
+
+4. The authors have used the HGSVC2 callsets as truth sets, so it is not clear if only DEL and DUP type SVs or the whole SV set is selected for the evaluations. Are these high-confidence call sets? Are they validated against GIAB truth sets? For each sample, the evaluation was performed using a training module of SurVIndel which was trained by using the features of other 33 long-read-based CNV callsets. So would it be also possible to compare the results of SurVIndel2 without using the last step to make a fair comparison with "pure" short-read-based callers? The evidence and claims can be better examined with the availability of callsets. Another suggestion is to use the GIAB SV truth set (v0.6) can also be used for the evaluation. The DEL-type SVs with length >=1kbp can be used as a benchmark set while evaluating all the callers. Also, why the precision is not computed for DEL types (last line of 1st paragraph of Section 2.3)?  
+
+5. All the other tools that are used in this comparison are not designed to call CNVs. Another most commonly used CNV caller tool is CNVnotar. So it is suggested to use that tool to compare the performance of SurVIndel2. Also, the 2nd best-performing tool, Manta, has been improved in the new DRAGEN pipeline released by Illumina. The authors are suggested to take a look at the pre-print (https://doi.org/10.1101/2024.01.02.573821) and the available callsets as some of the evaluation numbers presented in the pre-print for HG002 do not match the analysis done in this manuscript.  
+
+6. The authors used the terms like "tandem repetitive regions", and "repetitive regions" throughout the manuscript. It would be better if they are mentioned or a BED file is provided to identify those regions in reference. Are they telomere or centromere regions? It is difficult to understand the CNVs outside repetitive regions if they are not clearly defined or explained. Also, it is better to use the length information while using "shorter" terms for CNVs.
+
+<--- Page Split --->
+
+We thank the reviewers for the useful feedback. As requested, we have uploaded instructions and code to reproduce the figures of the paper in https://github.com/kensung-lab/survindel2_paper_experiments/We have also updated the data availability, since the data is now available in EBI. We will reply to their reviews point by point:  
+
+## Reviewer 1  
+
+The authors describe surVIndel2, a tool for CNV detection, leveraging a previously underappreciated signal - the so- called hidden split reads. This innovative approach appears successful in identifying CNVs, especially in a tandem repeat context, where typical split reads fail to identify the presence of a CNV. The observation that many CNVs are missed due to poor coverage is particularly intriguing, and I wonder if sequence- specific features beyond repetitive context contribute to this. The performance of this tool, compared to well- established tools, is impressive, and I wish to congratulate the authors on their work.  
+
+Sincerely, Wouter De Coster  
+
+We appreciate the kind words.  
+
+Regarding whether there are sequence- specific features that can explain low coverage, it would be a very interesting research direction. We have observed that many of these regions tend to be very repetitive, e.g., low complexity sequences or having long homopolymers. However, we have also observed pairs of very similar low complexity regions in totally different locations where one region was well sequenced while the other was very poorly sequenced. Therefore, there must be factors other than the sequence itself.  
+
+We would like to emphasize that we have not conducted any rigorous study on this at this point, and these are just simple observations made while developing the tool.  
+
+## Major:  
+
+I see the authors use SurVCluster to compare SVs from various callset, but the manuscript could be improved by more information on how exactly that is done, and how the precision and sensitivity are calculated based on that. Note that a commonly accepted method for comparing structural variant call sets is Truvari (https://github.com/ACEnglish/truvari). Can the authors confirm that they obtained similar results with their tool?  
+
+The main reason why we use SurVCluster for comparing SVs is that it employs a more sophisticated algorithm for comparing insertions and duplications. In long reads- based benchmarks such as HGSVC2, duplications are reported as point insertions. Short read based tools often report them as tandem duplications, denoted with the start and the end of the duplicated region.  
+
+One issue of such representation is that we do not know how many times the region was duplicated (the caller may report an estimate, but it is generally difficult to estimate accurately for duplications that are not extremely long, i.e., 1000s of bps).  
+
+Therefore, if a tool reports that a region B was duplicated, and the benchmark reports an insertion of a sequence BBBB in the same location, the comparison algorithm in SurVClusterer identifies that the two represent the same event. On the other hand, it appears to us (by reading the algorithm explained in the documentation) that truvari assumes every tandem duplication is duplicated exactly once.  
+
+One clear example of this is the benchmark SV chr6- 168229517- INS- 512 in NA24385. SurVIndel2 predicts a duplication of 128 bp, and the inserted sequence reported by the benchmark is identical to the duplicated region predicted by SurVIndel2, duplicated 4 times. SurVClusterer reports the two events as matching, while truvari does not.  
+
+In practice, many cases are less clear- cut; for example, many tandem duplications are caused by expansions of short motifs. When such expansions are not much smaller than the read length, it is nearly impossible to identify the exact length of the expansion using short reads (especially given that many of them are in the very noisy regions that we describe in Section 2.7), and it is often underestimated when compared to HiFi reads.
+
+<--- Page Split --->
+
+Secondly, SurVClusterer uses the TRF annotations to determine whether two events within the same repetitive region represent the same event but shifted.  
+
+In any case, we did try using truvari to compare Manta and SurVIndel2 (since they perform far better than the rest). Instructions and data are reported in https://github.com/kensung- lab/ survivedl2_paper_experiments/, under truvari.  
+
+For deletions, we obtained the following results:  
+
+HG00512  
+
+Manta: 0.36 recall, 0.72 precisionSurVIndel2: 0.56 recall, 0.80 precision  
+
+HG002  
+
+Manta: 0.36 recall, 0.71 precisionSurVIndel2: 0.56 recall, 0.75 precision  
+
+Since the matching criteria of the two comparison algorithms are different, the numbers are a bit different; however, the relative performance are the same.  
+
+For tandem duplications, because of the reasons explained above, we have disabled the comparison of the inserted sequences. If activated, the precision of the algorithms drop to around 0.6.  
+
+HG00512  
+
+Manta: 0.13 recall, 0.78 precisionSurVIndel2: 0.37 recall, 0.83 precisionHG002  
+
+Manta: 0.14 recall, 0.77 precisionSurVIndel2: 0.41 recall, 0.82 precision  
+
+However, this comparison is perhaps too relaxed, and we believe the one operated by SurVClusterer is closer to reality.  
+
+It must be noted that comparing SVs is very challenging, due to repetitive regions, imprecise breakpoints, imprecise inserted sequences, multiple valid representations for the same event, etc..  
+
+We are always looking for improvements and cases where our algorithm fails. Aside from benchmarking, being able to correctly identify matching SVs is especially important when clustering SVs in a population.  
+
+We also added a paragraph to the supplementary section "Comparing deletions and tandem duplications" stating how recall and precision are calculated.  
+
+Minor:  
+
+It appears the authors use the 'HSR' abbreviation and only describe what it means later in the text.  
+
+Indeed, we never introduced the HSR abbreviation formally. We added it to Section 2.1.  
+
+The y- axis in Figure 1c and 1d is unlabeled.  
+
+Since Figure 1 has no plot with x and y axes in it, we wonder if the reviewer is referring to a different figure. We have indeed found some panels without y- axis label:  
+
+- Fig. 2f and S6f- Fig. 4c and d
+
+<--- Page Split --->
+
+- Fig. S13c (now S15c) and S14c (now S16c) We added a label to all of them.  
+
+There is no legend for the bar colors in Figure 5b, but presumably, the colors are the same as for 5a. However, this is not explicitly stated and could lead to confusion.  
+
+We now explicitly state that the legend of panel (a) also applies to (b).  
+
+Mostly out of curiosity and not crucial for this paper, I wondered if it is possible to train the filter module on a combination of different species and what the performance of such a mixed model would be compared to the species- specific filter.  
+
+It is not easy to answer this question, as there are currently few (that we could find) datasets having both HiFi and Illumina reads. The reason we need HiFi is that we need to divide our training data into false and true positives. To do so, we require a benchmark that reports accurate inserted sequences. At the time of our experiments, we found that long read SV callers only produced accurate inserted sequences with HiFi reads.  
+
+Out of curiosity, we tried training a model on both humans and Arabidopsis data, and tried to predict an Arabidopsis sample (which was left out of training). The results are similar to using Arabidopsis data only, perhaps slightly worse: compared to using Arabidopsis only, for deletions, the mixed model has 0.02 lower sensitivity and 0.01 higher precision. For duplications, the mixed model has 0.01 higher sensitivity but 0.05 lower precision.  
+
+We also tried to classify a different organism (cattle), and there is no appreciable difference between the human- only model and the mixed model (very tiny increase in predicted TPs, but not enough to generate a change in sensitivity).  
+
+It is possible that a model that is larger or more sophisticated could utilize the mixed training model better.  
+
+## Reviewer 2  
+
+This work presents a novel approach, SurVlndel2, to detect copy number variations (CNVs). It looks for sequences that can be aligned to alternative positions to the reference genome as split reads (referred to as hidden split read) and demonstrates enhanced sensitivity and precision from short- read sequencing data, particularly improving detection of duplication in repetitive regions of the genome.  
+
+Although long- read sequencing can significantly improve CNV calling in repetitive regions, this work acknowledges that short- read sequencing will remain the primary research data in the near future and therefore addressing the limitation of detecting CNVs in repetitive regions using short- read sequencing is of great relevance.  
+
+The manuscript provides an evaluation of SurVlndel2 with public datasets of multiple organisms and benchmarks its performance with existing CNV and SV callers. However, further analyses and validation on existing tools could strengthen the claims. Please see the list of revisions below for details.  
+
+The analysis appears methodologically sound overall, but a more detailed discussion on limitations and potential biases would be beneficial. Any flaws do not seem to prohibit publication but may require revisions (see the list of revisions below) for clarity and completeness.  
+
+The manuscript clearly described the methodology and provides multiple benchmarking analyses of the method on cell line samples. I believe the work meets current standards in bioinformatics for CNV detection, given the list of revisions are addressed.  
+
+While SurVlndel2 is open- source software, for full reproducibility, I suggest also including scripts for running the analyses.  
+
+We would like to thank the reviewer. As suggested, we have published the scripts for generating the main figures in https://github.com/kensung- lab/survindel2_paper_experiments/.
+
+<--- Page Split --->
+
+Below, we will reply to the individual revisions.  
+
+## List of Revisions:  
+
+1. Figure 2 shows that over \(93\%\) of the tandem duplications of HG002 are in repeat regions. It would be good to see a breakdown of the types of repeat regions in the bar chart.  
+
+We use TRF to identify tandem repetitive regions (Supplementary). We have added a sentence in Section 2.1 that explicitly states this. As far as we know, TRF does not classify repetitive regions into subtypes, and I am currently not aware of such a classification (as opposed as general repeats, which can be classified into mobile elements, tandem repetitive regions, etc.).  
+
+2. In section 2.3, the manuscript stated that filtering was implemented by leave-1-out training for SurVIndel2, but methods of the 'pre-trained random forest' filtering are not provided. Please include further description.  
+
+We added a new Supplementary section called "Filtering module training".  
+
+3. In addition to revision 2, section 2.3 didn't include descriptions of how filtering was performed on CNV calls of benchmarking tools. For example, Manta ranks the quality of SV calls using QUAL score and FILTER tags, and the choice of filtering strategy can impact the benchmarking results, particularly sensitivity. Figure 4 was generated with one set of calls per sample per caller. Possibilities of alternative filtering strategies were not explored. I recommend including multiple filtering options in the benchmarking, such as different QUAL cutoffs per sample per caller, and plot curves of precision-recall in figure 4(a), 4(b), and 5(a).  
+
+We retained calls with FILTER \(= =\) PASS, since this is in our opinion the standard use case. We added a Supplementary section called "Selected callers" where we specify this. We have also generated a precision/recall curve for deletions in HG00512 and HG002 (Smoot, SurVIndel and SurVIndel2 do not report a QUAL value and therefore they are just single points).  
+
+![PLACEHOLDER_7_0]
+
+
+<--- Page Split --->
+![PLACEHOLDER_8_0]
+  
+
+We could not generate a curve for duplications because for Delly and Manta we use two different call sets to calculate recall and precision: precision is calculated on SVTYPE=DUP, while for recall we also include SVTYPE=INS (since Delly and Manta may decide to report events as either DUP or INS).  
+
+The code and data for generating the curves was uploaded to the aforementioned GitHub repository.  
+
+4. HG002 is a great independent sample, which was not specifically trained on by SurVlndel2, for benchmarking purposes. However, the discussion only includes sensitivity, not precision. I strongly recommend including HG002 in Figure 4(a) and 4(b) as well.  
+
+HG002 is part of the HGSVC2, as mentioned in Section 2.3: a catalogue (called HGSVC2) of insertions and deletions in 35 human genomes (34 from the 1000 genomes project, plus HG002). In HGSVC2, it is called NA24385. Therefore, it is in Figure 4a and 4b. The purpose of figures 4c and 4d is simply to show that existing callers struggle with HSR-supported calls. We have added two more benchmarks for HG002 (Fig. S7 and an additional paragraph in Section 2.3). One was the GIAB v0.6 benchmark, while the other was obtained by running Sniffles2 on HiFi reads. The relative performance of the tools are in line with what was observed on the HGSVC2 benchmark. Instructions and files to reproduce the benchmark are in the GitHub repository.  
+
+5. Very importantly, in section 2.3, Delly, Lumpy, and Manta were selected as benchmarking tools according to reference [4]. However, in this benchmarking study, GRIDSS is another top-performing SV caller. I strongly recommend including benchmarking with GRIDSS2 + Purple (Cameron, D.L., Baber, J., Shale, C. et al. GRIDSS2: comprehensive characterization of somatic
+
+<--- Page Split --->
+
+structural variation using single brekend variants and structural variant phasing. Genome Biol 22, 202 (2021). https://doi.org/10.1186/s13059- 021- 02423- x).  
+
+We have tried using GRIDSS, but it does not explicitly report deletions and duplications, only breakends. One criteria we used is that the callers should explicitly predict deletions and tandem duplications.  
+
+In the GitHub README, the authors mention that example/simple- event- annotation.R will annotate the calls with SIMPLE_TYPE and SVLEN, but the script is not made for general users as it has hardcoded paths inside. We modified the hardcoded path to point to our own GRIDSS results, but the script printed this:  
+
+Error in .subassign_columns(x, nsbs, value) : provided 68 variables to replace 67 variables Calls: [< - ... [<- -> mergeROWS -> mergeROWS -> .subassign_columns In addition: Warning message: info fields with no header: SVLEN Execution halted  
+
+In the benchmarking paper "Comprehensive evaluation and characterisation of short read general- purpose structural variant calling software" by Cameron DL et al. (the authors of GRIDSS), it is shown that GRIDSS performs similarly to Manta on HG002 (Fig.1, slightly more precise but less sensitive).  
+
+6. It's interesting to see that SurVlndel 1&2 are more sensitive than other callers with no HSR support. Section 2.7 discusses the reasons why these CNVs have a low level of evidence. It would be good to include more discussion on why SurVlndel2 can capture more such CNVs than other callers, as well as the true positive rate of unsupported CNVs called by SurVlndel2.  
+
+We investigated why SurVlndel2 can detect a portion of deletions marked as unsupported. There seem to be many factors contributing.  
+
+We found that \(38\%\) of these deletions are predicted using HSRs. As for why such deletions are classified as unsupported despite having HSRs supporting them, we identified three possible reasons:  
+
+(1) spoa built an alternative allele that is not \(100\%\) correct, and it escaped our filters; 
+(2) the HSRs are less than 5 (minimum requirement to be classified as HSR-supported), while SurVlndel2 is able to identify CNVs from as few as 3 reads;  
+
+(3) when mapping the short reads to the set of alternative alleles, BWA might have placed the relevant reads somewhere else (since we mapped the reads to all of the alternative alleles together, as running BWA for thousands of alternative allele separately would have been unfeasible with our computational resources).  
+
+Another \(38\%\) of them were predicted using the discordant pairs module, so no SR or HSR support was required. Finally, out of the remaining \(24\%\) that were predicted using split reads, nearly half of them had less than 5 split reads supporting them (minimum support required to be classified as SR-supported was again 5).  
+
+We have added a small paragraph explaining this to Section 2.3.  
+
+7. I recommend including a discussion on the limitations of the tool regarding the detection of somatic CNVs in cancer genomes. This discussion could acknowledge that, unlike the other tools benchmarked in the study, SurVlndel2 does not currently support the identification of somatic CNVs, which are critical for understanding cancer genomics. Highlight this as a significant caveat, and suggest future research directions or potential updates to the tool that could address this gap. Emphasize the importance of this functionality for comprehensive cancer genomics studies and how its inclusion could further elevate the tool's applicability and utility in the field. This addition will not only provide a more balanced view of SurVlndel2's capabilities but also align expectations for researchers interested in cancer genomics.  
+
+Indeed, SurVlndel2 does no explicitly support detecting somatic variants (excluding the naive approach of processing tumor and normal independently and operating a subtraction). I have added a paragraph to the Discussion explaining this.
+
+<--- Page Split --->
+
+## Reviewer 3  
+
+This manuscript introduces a new tool SurVlndel2 for detecting the copy number variations (CNV) using short illumina sequencing reads with a novel technique that uses hidden split reads and machine learning techniques. The benchmarking results showed that SurVlndel2 performed better than other short- read- based structural variant (SV) callers. However, concerns arise regarding the benchmarking evaluation methods and the availability of results produced in this study.  
+
+We would like to the thank the reviewer. Below, we reply to their comments individually:  
+
+## Major issues:  
+
+1. Throughout the manuscript, it seems the authors have considered all deletion (>= 50bp) and tandem duplication type SVs as CNVs. There is confusion as all deletions (>=50bp) can not be called CNV. Is the tool designed to detect all DEL and DUP-type SVs? If that is the case, then it is not suggested to call it a CNV caller tool.  
+
+By reading the literature, it appears to us that there is no universally accepted definition of CNV. Some definitions require a length of at least 1000 bp, other 50 bp: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9407502/ https://genomedicine.biomedcentral.com/articles/10.1186/s13073-021-00945-4 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9365719/  
+
+However, it is true that we do not detect all CNVs, because we only detect tandem duplications (as opposed to interspersed duplications). In the original SurVlndel, we created the term local CNV to refer to deletions and tandem duplications only. We have clarified in the introduction that in the paper we consider local CNVs, and we will refer to them simply as CNV for convenience.  
+
+2. The authors have claimed that it is a new short-read-based CNV caller, however, in the last few steps it uses the machine learning technique that takes advantage of long-read-based CNV callers (steps g and h) for training purposes. The machine learning model of SurVlndel2 is used in these steps, right? So would it be fair to call it a short-read-based caller? It is expected to work a little better when long-reads or long-read-based callsets are used at the filtering stage, so it won't be a fair comparison with other "pure" short-read-based callers such as Manta, Lumpy, and Delly. Also, Lumpy only calls the DEL-type SVs, so it may not detect any INS (or DUP) type SVs.  
+
+We distribute the data used to train the model (as the trained model is several GBs and difficult to distribute, and training the model requires modest computational resources). It is true that the data was obtained with the help of long reads. However, as mentioned, no long reads are required to run SurVlndel2, only short reads, and for this reason we believe it to be a short reads caller. It is true that if the user wants to train its model on different samples (for example, a different organism) he will need a high quality ground truth set for those samples, which are usually obtained by long reads. However, we have demonstrated in the paper that the default model provided works well in several non- human organisms. Lumpy calls deletions, duplications, inversions and generic "breakends".  
+
+Regarding the fairness of the comparison, we only want to demonstrate that the tool will be useful to the community. Delly, Lumpy and Manta are seminal works that have been developed nearly a decade ago. We have been able to develop our algorithm not only thanks to them, but also thanks to data that has been published since then, such as the many long reads datasets, novel catalogues such as HGSVC2 that are more complete than ever, and even high quality assemblies such as HG002. We have developed our insights and ultimately our algorithm using this data, which was not available at the time the other tools were developed.  
+
+3. The "Method" section is included inside the "Result" section. Also, that section only talks about the algorithms and techniques used in this tool. It would be better to include the benchmarking strategies such as what comparison tools and the truth set are used for evaluation. The results can be reproduced for verification if the VCF file for the truth sets, callsets, commands, or scripts are available in the "Methods" section. The authors could provide a link (e.g. GitHub repo) with all the result files.
+
+<--- Page Split --->
+
+4. The authors have used the HGSVC2 callsets as truth sets, so it is not clear if only DEL and DUP type SVs or the whole SV set is selected for the evaluations. Are these high-confidence call sets? Are they validated against GIAB truth sets? For each sample, the evaluation was performed using a training module of SurVlndel which was trained by using the features of other 33 long-read-based CNV callsets. So would it be also possible to compare the results of SurVlndel2 without using the last step to make a fair comparison with "pure" short-read-based callers? The evidence and claims can be better examined with the availability of callsets. Another suggestion is to use the GIAB SV truth set (v0.6) can also be used for the evaluation. The DEL-type SVs with length \(> = 1\) kbp can be used as a benchmark set while evaluating all the callers. Also, why the precision is not computed for DEL types (last line of 1st paragraph of Section 2.3)?  
+
+We will respond to both points together.  
+
+We understand this is confusing, since we failed to mention it explicitly. Tandem duplications in the HGSVC2 benchmark, but also in the Sniffles2 results, are reported as insertions. We have a method that we have previously published in the Supplementary of https://www.nature.com/articles/s41467- 023- 38870- 2 that identifies which insertions are due to tandem duplication. We tried to make it clearer by explicitly mentioning this in 2.3 and 2.4, and also clarifying how sensitivity and precision are calculated in the supplementary section "Comparing deletions and tandem duplications". We only use tandem duplications to calculate sensitivity, while we use all benchmark insertions to calculate precision.  
+
+We have published all the instructions, data and code to replicate our figures in https://github.com/kensung-lab/survindel2_paper_experiments/.  
+
+We believe HGSVC2 to be high- confidence, since it is produced by a well known consortium (the Human Genome SV Consortium) using HiFi reads. It is generally more complete than the GIAB benchmark (although, in our experience, it contains more false positives). They are hard to compare since GIAB is, as far as we know, hg19 only. In any case, we have added two more benchmarks for HG002: the GIAB one, and one generated by Sniffles2 (Fig. S7). Rather than \(\geq 1000\) bp, we used \(\geq 50\) bp, since all the tools involved are designed to call SV- type events.  
+
+Unfortunately the unfiltered callset of SurVlndel2 is very noisy, as it contains \(>100,000\) calls, and it is certainly not recommended to use it directly. For this reason, it is not meaningful to benchmark it.  
+
+Precision was computed for all samples, including HG002 (which is the sample NA24385 in HGSVC2). It is the y- axis in Fig. 4a and 4b. As mentioned, we have added two more benchmarks for HG002, and we computed sensitivity and precision.  
+
+5. All the other tools that are used in this comparison are not designed to call CNVs. Another most commonly used CNV caller tool is CNVnator. So it is suggested to use that tool to compare the performance of SurVlndel2. Also, the 2nd best-performing tool, Manta, has been improved in the new DRAGEN pipeline released by Illumina. The authors are suggested to take a look at the pre-print (https://doi.org/10.1101/2024.01.02.573821) and the available callsets as some of the evaluation numbers presented in the pre-print for HG002 do not match the analysis done in this manuscript.  
+
+In the original SurVlndel paper, we did include CNVnator, and it performed much worse than the callers included here: https://academic.oup.com/bioinformatics/article/37/11/1497/5466452. Also, the scope of CNVnator is different, as it is designed to detect very large CNVs, while SurVlndel2 and the other callers tested are designed to detect \(\geq 50\) bp events (or even smaller).  
+
+We did not compare DRAGEN to free and open- source because it is proprietary, closed and requires a license. However, we recognize its prominence in the field and since the paper the reviewer mentions does provide calls for 1KGP, including the samples in HGSVC2, we included a comparison (Section 2.3 and Fig. S8). DRAGEN is a very clear improvement over Manta, but its sensitivity and recall are still sensibly lower than SurVlndel2. As usual, the GitHub repository contains code and data for this comparison.
+
+<--- Page Split --->
+
+6. The authors used the terms like "tandem repetitive regions", and "repetitive regions" throughout the manuscript. It would be better if they are mentioned or a BED file is provided to identify those regions in reference. Are they telomere or centromere regions? It is difficult to understand the CNVs outside repetitive regions if they are not clearly defined or explained. Also, it is better to use the length information while using "shorter" terms for CNVs.  
+
+We now specify in Section 2.1 that we consider tandem repetitive regions identified by the Tandem Repeats Finder (TRF). In the GitHub we provide a file with the list of repetitive regions.  
+
+Sincerely,  
+
+Ramesh Rajaby and Wing- Kin Sung.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+Dear authors,  
+
+I am satisfied with the clarifications made regarding my previous request, and wish to congratulate you on your manuscript and tool. I have no further comments.  
+
+Sincerely, Wouter De Coster  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors have addressed my concerns and suggestions raised in my previous review. The revisions made to the manuscript and supplementary methods, including the clarified analyses and expanded discussions, have improved the clarity of the study. The additional instructions and code provided transparency and reproducibility of the published results. Therefore, I recommend this manuscript for publication.  
+
+Reviewer #2 (Remarks on code availability):  
+
+The repository for SurVIndel2 (https://github.com/kensung- lab/SurVIndel2) includes a clear README file with installation and running instructions. I successfully installed SurVIndel2 by following these instructions on Rocky Linux v8.9 (Green Obsidian).  
+
+The repository also provides clear instructions for reproducing the main figures. However, due to time constraints, I have not attempted to reproduce the results presented in the manuscript.  
+
+Reviewer #3 (Remarks to the Author):  
+
+I would like to thank the authors for putting a great effort and doing additional analysis as suggested by the reviewers. They have indeed provided more explanations for most of the comments and suggestions.  
+
+However, there are still a few additional concerns that may need to be addressed.  
+
+In the earlier version of the manuscript, it was a bit unclear about the validation tool/software that was used for all the analysis, and Reviewer #1 suggested using Truvari. I think Truvari has been considered a standard tool for SV validation. Also, some recent results have used Wittery (https://github.com/Illumina/witty.er) for this purpose. The authors have clarified that they have used a tool "SurVClusterer" which was developed by them and also clarified some reasons why it may be better than Truvari. They have also provided some additional results based on truvari. I have tried to replicate some of the results by running truvari. Here are some comments based on that result.  
+
+1. The file "1_data_preparation/by-sample/HG00512.DEL.vcf.gz" was not present on their Github repo (https://github.com/kensung-lab/survindel2_paper_experiments/). So it would be great if they upload at least DEL benchmark results for few samples.  
+
+2. For HG002 with hg19 reference calls, truvari failed with error "ERROR:root:No SAMPLE columns found in vcf", though it worked for MANta and DRAGEN vcfs.
+
+<--- Page Split --->
+
+I think it would be great if authors could show that the performance of their tool is better than Manta/DRAGEN (and others) by running Truvari on the DEL variant types.  
+
+Reviewer #3 (Remarks on code availability):  
+
+The codes are reproducible, but some files are still missing and also some info inside VCFs is missing.
+
+<--- Page Split --->
+
+We thank the reviewers for the kind feedback. We will reply to their reviews point by point:  
+
+## Reviewer 1  
+
+Dear authors,  
+
+I am satisfied with the clarifications made regarding my previous request, and wish to congratulate you on your manuscript and tool. I have no further comments.  
+
+Sincerely, Wouter De Coster  
+
+Thank you very much.  
+
+## Reviewer 2  
+
+The authors have addressed my concerns and suggestions raised in my previous review. The revisions made to the manuscript and supplementary methods, including the clarified analyses and expanded discussions, have improved the clarity of the study. The additional instructions and code provided transparency and reproducibility of the published results. Therefore, I recommend this manuscript for publication.  
+
+The repository for SurVlndel2 (https://github.com/kensung- lab/SurVlndel2) includes a clear README file with installation and running instructions. I successfully installed SurVlndel2 by following these instructions on Rocky Linux v8.9 (Green Obsidian).  
+
+The repository also provides clear instructions for reproducing the main figures. However, due to time constraints, I have not attempted to reproduce the results presented in the manuscript.  
+
+Thank you very much.  
+
+## Reviewer 3  
+
+I would like to thank the authors for putting a great effort and doing additional analysis as suggested by the reviewers. They have indeed provided more explanations for most of the comments and suggestions.  
+
+However, there are still a few additional concerns that may need to be addressed.  
+
+In the earlier version of the manuscript, it was a bit unclear about the validation tool/software that was used for all the analysis, and Reviewer#1 suggested using Truvari. I think Truvari has been considered a standard tool for SV validation. Also, some recent results have used Wittyer (https://github.com/Illumina/witty.er) for this purpose. The authors have clarified that they have used a tool "SurVClusterer" which was developed by them and also clarified some reasons why it may be better than Truvari. They have also provided some additional results based on truvari. I have tried to replicate some of the results by running truvari. Here are some come comments based on that result.  
+
+1. The file "1_data_preparation/by-sample/HG00512.DEL.vcf.gz" was not present on their Github repo (https://github.com/kensung-lab/survindel2_paper_experiments/). So it would be great if they upload at least DEL benchmark results for few samples.  
+
+Our approach in providing the steps for reproducing our analyses was to, as much as possible, start from the source public data. For this reason, the instructions in the README of 1_data_preparation should be executed before attempting to reproduce any other figure.  
+
+In this particular case, this means downloading the HGSVC2 dataset and generating the benchmark datasets for each individual sample:
+
+<--- Page Split --->
+
+wget https://ftp.1000genomes.ebi.ac.uk/vol1/ftp/data collections/HGSVC2/release/v2.0/integrated_callset/variants_freeze4_sv_insdel_alt.vcf.gz  
+
+(This downloads the HGSVC2 dataset)  
+
+(It is now necessary to divide benchmark insertions from tandem duplications, instructions are reported in https://github.com/kensung- lab/survindel2_paper_experiments/blob/main/1_data_preparation/README.txt)  
+
+mkdir by- sample  
+
+bcftools query - l variants_freeze4_sv_insdel_alt.vcf.gz | while read sample ; do bcftools view variants_freeze4_sv_insdel_alt.vcf.gz - i "SVTYPE \(=\) "DEL" - s \(\) sample - - min- ac=1 - Oz - o by- sample\) \ \(sample.DEL.vcf.gz\) ; done bcftools query - l variants_freeze4_sv_insdel_alt.vcf.gz | while read sample ; do bcftools view variants_freeze4_sv_insdel_alt.DUP.vcf.gz - i "SVTYPE \(=\) 'DEL' " - s \(\) sample - - min- ac=1 - Oz - o by- sample\) \ \(sample.DUP.vcf.gz\) ; done bcftools query - l variants_freeze4_sv_insdel_alt.vcf.gz | while read sample ; do bcftools view variants_freeze4_sv_insdel_alt.vcf.gz - i "SVTYPE \(=\) 'DEL' " - s \(\) sample - - min- ac=1 - Oz - o by- sample\) \ \(sample.INS.vcf.gz\) ; done  
+
+(This, barring unforeseen failures, generates benchmark datasets for each individual, including 1_data_preparation/by- sample/HG00512.DEL.vcf.gz.).  
+
+We find this approach more transparent, as it makes clear how we obtained the benchmark datasets, and that we did not filter/manipulate them unfairly.  
+
+It should be noted that there are also instances in which obtaining some of the results requires significant computation time/resources. In those cases, we also provided pre- computed results.  
+
+2. For HG002 with hg19 reference calls, truvari failed with error "ERROR:root:No SAMPLE columns found in vcf", though it worked for MAnta and DRAGEN vcfs.  
+
+I think it would be great if authors could show that the performance of their tool is better than Manta/DRAGEN (and others) by running Truvari on the DEL variant types.  
+
+We have included instructions to https://github.com/kensung- lab/survindel2_paper_experiments/blob/main/hg002- additional/README.txt on how to use truvari to benchmark the HG002 on hg19 calls, on both the GIAB and the Sniffles2 benchmark.  
+
+We did not encounter the error reported, but truvari failed on the GIAB benchmark because it was missing the contigs lengths (we show how to work around it in the README).  
+
+Here are the results:  
+
+Sniffles benchmark:  
+
+DEL:  
+
+Manta recall 0.39 Manta precision 0.80  
+
+SurVIndel2 recall 0.51 SurVIndel2 precision 0.76  
+
+DUP:  
+
+Manta recall: 0.15 Manta precision: 0.90  
+
+SurVIndel2 recall 0.40 SurVIndel2 precision 0.89
+
+<--- Page Split --->
+
+GIAB benchmark:  
+
+DEL:  
+
+Manta recall 0.52  Manta precision 0.79  
+
+SurVIndel2 recall 0.73  SurVIndel2 precision 0.90  
+
+DUP:  
+
+Manta recall 0.22  Manta precision 0.86  
+
+SurVIndel2 recall 0.50  SurVIndel2 precision 0.88  
+
+There are a couple of points worth mentioning. We have previously explained why we disable the comparison of inserted sequences in duplications when using truvari: truvari assumes that a tandem duplications duplicates the reference sequence exactly once, whereas SurVClusterer does not have this assumption.  
+
+This way, the precisions for duplications become very high. We briefly checked the Manta tandem duplications reported as true positives by truvari on the Sniffles benchmark. Here are the first two:  
+
+chr1 1584516 MantaDUP:TANDEM:61:2:6:0:0:0 A <DUP:TANDEM> 289 PASS  END=1651190;SVTYPE=DUP;SVLEN=66674;IMPRECISE;CIPOS=- 578,579;CIEND=- 437,437;Pct SeqSimilarity=0;PctSizeSimilarity=0.0587;PctRecOverlap=0.0297;SizeDiff=- 62761;StartDistance= 26;EndDistance=- 66648;TruScore=2;Matchld=32.1.0 GT:FT:GQ:PL:PR 0/1:PASS:289:339,0,761:53,26 chr1 1588374 MantaDUP:TANDEM:52:0:1:1:0:0 C <DUP:TANDEM> 23 PASS  END=1653752;SVTYPE=DUP;SVLEN=65378;IMPRECISE;CIPOS=- 297,297;CIEND=- 679,680;Pct SeqSimilarity=0;PctSizeSimilarity=0.0008;PctRecOverlap=0.0008;SizeDiff=- 65327;StartDistance= 61112;EndDistance=- 4266;GTMatch;TruScore=0;Matchld=33.0.0 GT:FT:GQ:PL:PR 0/1:PASS:23:73,0,999:133,25  
+
+For the first, Sniffles2 reported two insertions near the first breakpoint, one of size 681 and the other 3,913. The tandem duplication of Manta could not have generated such insertion, since it reports a duplication of 66,674 bps.  
+
+SurVClusterer takes this into account: it allows the duplicated region to repeat multiple times, if necessary, in order to match the inserted sequence; however, if the duplicated region is larger than the inserted sequence, it will not match the two.  
+
+We are very puzzled by the second one, since Sniffles2 does not report any insertion close to either breakpoints. We do not know why Truvari would consider this duplication as a true positive.  
+
+Second, in the Sniffles benchmark, it seems that truvari reports a lower precision for both Manta and SurVIndel2 (SurVClusterer reports nearly 0.9).  
+
+When manually checking the SurVIndel2 false positives according to truvari, however, many of them are clearly supported by long reads. We have uploaded the IGV screenshots for all false positives to the GitHub repository. We are not sure why truvari reports many of them as false positives.  
+
+Sincerely,  
+
+Ramesh Rajaby and Wing- Kin Sung.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #3 (Remarks to the Author):  
+
+I would like to thank the authors for the detailed response to my earlier suggestions. I understand that the authors have put a README file with all the commands that were used in the validation. In the earlier suggestion, I mentioned that some of the files were not available on Github repo. I totally agree with the authors that some executions may need significant time/resources. Therefore, it would be great if authors make only these three files available on their GitHub repo (or send the link to these files if they are already available)  
+
+HG00512  
+
+1. HG00512.DEL.vcf.gz 
+2. HG00512.survindel2.ml.DEL.alt.vcf.gz  
+
+HG002  
+
+1. HG002.survindel2.ml.DEL.alt.vcf.gz  
+
+It is also very interesting to know that truvari assumes tandem duplications are duplicated in the reference sequence exactly once. So it means that any benchmarking for tandem duplications using truvari were not accurate. The authors of Truvari have recently released a tandem- repeat benchmark (https://www.nature.com/articles/s41587-024-02225- z) that also used Truvari for validation, I am curious if they also used the same concept i.e. duplicated once in reference. It would be worth checking with Truvari developers or reporting these issues as this tool has become a standard tool for any validations.
+
+<--- Page Split --->
+
+We thank the reviewers for the kind feedback. We will reply to their reviews point by point:  
+
+## Reviewer 3  
+
+I would like to thank the authors for the detailed response to my earlier suggestions. I understand that the authors have put a README file with all the commands that were used in the validation. In the earlier suggestion, I mentioned that some of the files were not available on Github repo. I totally agree with the authors that some executions may need significant time/resources. Therefore, it would be great if authors make only these three files available on their GitHub repo (or send the link to these files if they are already available)  
+
+As suggested by the reviewer, we have uploaded the three requested files, in:  
+
+https://github.com/kensung- lab/survindel2 paper experiments/tree/main/1 data preparation/by- sample https://github.com/kensung- lab/survindel2 paper experiments/tree/main/truvari  
+
+It is also very interesting to know that truvari assumes tandem duplications are duplicated in the reference sequence exactly once. So it means that any benchmarking for tandem duplications using truvari were not accurate. The authors of Truvari have recently released a tandem- repeat benchmark (https://www.nature.com/articles/s41587- 024- 02225- z) that also used Truvari for validation, I am curious if they also used the same concept i.e. duplicated once in reference. It would be worth checking with Truvari developers or reporting these issues as this tool has become a standard tool for any validations.  
+
+Thank you for recommending the paper, which we were unaware of.  
+
+Upon reading it, it seems that Truvari now has a new comparison algorithm, called refine, which improves over the previous one in two ways:  
+
+1) A problem in comparing indels (whether small or large) in TR regions is that the indel can shift. For example, we can have two deletions, one deleting the first copy and one deleting the last copy of the same TR, and while they are disjoint, they should be recognized as producing the same haplotype. They improved the way in which this work. This problem was one of the reasons that pushed us to develop an in-house comparison algorithm.  
+
+2) An indel can not only shift, but also be "decomposed" (for lack of a better term) into multiple smaller indels. For example, if a region AA becomes AAAA, we can have a single insertion of AA, or two distinct insertions of A, and we are effectively representing the same indel. The authors have an algorithm that seems to be able to correctly compare indels under these conditions.  
+
+This is a difficult problem that does occur in practice (especially with noisy long reads).  
+
+However, none of these changes appear to improve the INS vs DUP comparison, which tends to be a problem mostly when comparing a long reads- derived benchmark dataset to a short read- derived called dataset.  
+
+Furthermore, it is not immediately clear to us how the refine command in Truvari directly translates to comparing sets of SVs. From its documentation (https://github.com/ACEnglish/truvari/wiki/ refine):  
+
+The regions spanned by subset.bed should be shorter and focused around the breakpoints of putative FNs/FPs. Haplotypes from these boundaries are fed into a realignment procedure which can take an extremely long time on e.g entire chromosomes.  
+
+However, for example, Manta calls several large deletions and duplications, often spanning almost a whole chromosome. Furthermore, they add:  
+
+Also, the genotypes within these regions must be phased.  
+
+However, this is not generally possible when using short reads.
+
+<--- Page Split --->
+
+We will try and get in touch with the authors of Truvari and see if we missed something and if there is indeed a way to extend the comparison algorithm of INS vs DUP.  
+
+Sincerely,  
+
+Ramesh Rajaby and Wing- Kin Sung.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #3 (Remarks to the Author):  
+
+I would like to thank the reviewers for providing the following VCF files for HG00512 and HG002 sample  
+
+HG00512 1. HG00512. DEL.vcf.gz 2. HG00512. survivdel2. ml.DEL.alt.vcf.gz  
+
+HG002 1. HG002. survivdel2. ml.DEL.alt.vcf.gz  
+
+The Truvari (v4.1) run on HG00512 (using the above two DEL vcf files) with the same parameters that the authors provided in the README file produced the following results. It would be great if the authors could compare this with their results and update them.  
+
+"precision": 0.8440798882173747, "recall": 0.610386228126738, "f1": 0.7084589093098503,  
+
+Regarding the comparison of DEL variants between SurVIndel2 (using the above DEL vcf) and DRAGEN based on Truvari (v4.1) produced the following results.  
+
+DRAGEN  
+
+"precision": 0.830099350472806, "recall": 0.6691725891079381, "f1": 0.7409993030606416,  
+
+SurVIndel2  
+
+"precision": 0.810636420815687, "recall": 0.6229756676916839, "f1": 0.7045235870445897,  
+
+So the Truvari- based analysis shows DRAGEN performs better than SurVIndel2 for DEL variants.  
+
+Although the authors have explained a few times the issues of Truvari based comparison and why the in- house evaluation tool is better than Truvari, I would still recommend providing all the results based on Truvari. Also, it is recommended to provide the valid reasons in the manuscript of using in- house tool against the standard evaluation tool.
+
+<--- Page Split --->
+
+We thank the reviewers for the feedback. We will reply to their reviews point by point:  
+
+## Reviewer 3  
+
+I would like to thank the reviewers for providing the following VCF files for HG00512 and HG002 sample  
+
+HG00512 1. HG00512. DEL.vcf.gz 2. HG00512. survindel2. ml.DEL.alt.vcf.gz  
+
+HG002 1. HG002. survindel2. ml.DEL.alt.vcf.gz  
+
+The Truvari (v4.1) run on HG00512 (using the above two DEL vcf files) with the same parameters that the authors provided in the README file produced the following results. It would be great if the authors could compare this with their results and update them.  
+
+"precision": 0.8440798882173747, "recall": 0.610386228126738, "f1": 0.7084589093098503,  
+
+Regarding the comparison of DEL variants between SurVIndel2 (using the above DEL vcf) and DRAGEN based on Truvari (v4.1) produced the following results.  
+
+DRAGEN  
+
+"precision": 0.830099350472806, "recall": 0.6691725891079381, "f1": 0.7409993036060416,  
+
+SurVIndel2  
+
+"precision": 0.810636420815687, "recall": 0.6229756676916839, "f1": 0.7045235870445897,  
+
+So the Truvari- based analysis shows DRAGEN performs better than SurVIndel2 for DEL variants.  
+
+Although the authors have explained a few times the issues of Truvari based comparison and why the in- house evaluation tool is better than Truvari, I would still recommend providing all the results based on Truvari. Also, it is recommended to provide the valid reasons in the manuscript of using in- house tool against the standard evaluation tool.  
+
+As recommended, we modified our supplementary section "Comparing deletions and tandem duplications" to include why we use the in- house tool rather than Truvari. Furthermore, we have added two panels to Fig. S8, that compare Manta, DRAGEN and SurVIndel2 using Truvari.  
+
+We have also modified the scripts in the truvari subfolder to make them work to Truvari 4+. We also noticed that when comparing Manta and DRAGEN, Truvari would give the following warning:  
+
+[WARNING] Unresolved SVs (e.g. ALT=<DEL>) are filtered when `--pctseq != 0`  
+
+And many calls would be ignored. Therefore, we now compare deletions with -- pctseq 0.  
+
+Regarding the comparison between SurVIndel2 and DRAGEN, we get very different numbers.
+
+<--- Page Split --->
+
+We have uploaded the DRAGEN deletions we have obtained for HG00512 to https://github.com/kensung-lab/survindel2 paper experiments/tree/main/dragen (along with the script to obtain them, in README.txt).  
+
+## DRAGEN  
+
+"precision": 0.8632424434580427, "recall": 0.44937403909510215, "f1": 0.5910618199468787,  
+
+## SurVIndel2  
+
+"precision": 0.8249924173491052, "recall": 0.5734680430485394, "f1": 0.6766108881942217,  
+
+(https://github.com/kensung-lab/survindel2 paper experiments/blob/main/truvari/README.txt contains the command we used)  
+
+DRAGEN reports 4732 deletions for HG00512, while the benchmark contains 9106 deletions. Therefore, it is unlikely to achieve a sensitivity higher than 4732/9106 = 0.52, regardless of the comparison method.  
+
+In case we are using different datasets or we have made a mistake in downloading the DRAGEN results, please let us know.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #3 (Remarks to the Author):  
+
+I would like to thank the authors for their satisfactory responses. There are no other concerns about the excellent work the authors did in this paper.  
+
+I have a last suggestion for authors to cross check the HG002 DEL comparison between DRAGEN and SurVIndel2 using truvari and report the correct outputs in the final version. The following results that was mentioned in the last review could belong to HG002 dataset (not HG00512).  
+
+DRAGEN  
+
+"precision": 0.830099350472806, "recall": 0.6691725891079381, "f1": 0.7409993036060416,  
+
+SurVIndel2  
+
+"precision": 0.810636420815687, "recall": 0.6229756676916839, "f1": 0.7045235870445897,
+
+<--- Page Split --->
+
+We thank the reviewers for the feedback. We will reply to their reviews point by point:  
+
+## Reviewer 3  
+
+I have a last suggestion for authors to cross check the HG002 DEL comparison between DRAGEN and SurVlndel2 using truvari and report the correct outputs in the final version. The following results that was mentioned in the last review could belong to HG002 dataset (not HG00512).  
+
+DRAGEN  
+
+"precision": 0.830099350472806, "recall": 0.6691725891079381, "f1": 0.7409993036060416, SurVlndel2  
+
+"precision": 0.810636420815687, "recall": 0.6229756676916839, "f1": 0.7045235870445897,  
+
+We could not replicate the numbers, using Truvari 4.1, even for SurVlndel2. We obtain  
+
+"precision": 0.7815208275090858, "recall": 0.5774434462604178, "f1": 0.6641588578474773,  
+
+While for the dataset provided by the reviewer for DRAGEN on HG002 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10802302/) we obtained the following numbers, again with Truvari 4.1  
+
+"precision": 0.800033892560583, "recall": 0.5116354583829419, "f1": 0.624129407381033,  
+
+It should be noted that we ran SurVlndel2 on a 50x HiSeq dataset, while the DRAGEN dataset was 35x NovaSeq 6000, so the comparison is not completely fair.  
+
+This brings us to the main reason why we think a comparison for HG002 would not be meaningful. A meaningful comparison requires, in our opinion, that the same input data is used. This was possible for the 34 samples in the HGSVC2 benchmark because the same set of reads, sequenced by the New York Genome Institute, were used to run SurVlndel2 and DRAGEN. However, for HG002, we selected a subset of Illumina reads (50x depth in total) from the GIAB project (which is 300x in total), and ran the different callers. We could not find a DRAGEN callset from the same set of reads.  
+
+To summarise, we would like not to include the comparison in the paper, because the comparison would not be not fair, as different coverage, sequencing platform, laboratories, etc. could influence the results. Furthermore, it would add little to the paper since  
+
+1) We have already compared SurVlndel2 to DRAGEN in 34 samples where a fair comparison is possible  
+2) DRAGEN is not a direct competitor of SurVlndel2, as it is a commercial product, while SurVlndel2 is open source and freely downloadable/usable  
+3) Even disregarding the above mentioned points, in the above mentioned test, SurVlndel2 still outperforms DRAGEN on HG002 (we would like to stress once again that it is not a completely fair comparison, and thus we do not think it should be reported in the paper)  
+
+Ramesh Rajaby and Wing- Kin Sung
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__72f06a0c6f99ff975ae8e2e989c5343b6bc0a33378da8598ab6e8c6c95628b7c/supplementary_0_Peer Review File__72f06a0c6f99ff975ae8e2e989c5343b6bc0a33378da8598ab6e8c6c95628b7c_det.mmd

@@ -0,0 +1,969 @@
+<|ref|>title<|/ref|><|det|>[[100, 40, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[108, 110, 373, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[108, 161, 810, 220]]<|/det|>
+SurVIndel2: improving copy number variant calling from next- generation sequencing using hidden split reads  
+
+<|ref|>image<|/ref|><|det|>[[93, 732, 262, 780]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[271, 732, 880, 784]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 88, 305, 104]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 118, 402, 134]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 148, 878, 255]]<|/det|>
+The authors describe surVIndel2, a tool for CNV detection, leveraging a previously underappreciated signal - the so- called hidden split reads. This innovative approach appears successful in identifying CNVs, especially in a tandem repeat context, where typical split reads fail to identify the presence of a CNV. The observation that many CNVs are missed due to poor coverage is particularly intriguing, and I wonder if sequence- specific features beyond repetitive context contribute to this. The performance of this tool, compared to well- established tools, is impressive, and I wish to congratulate the authors on their work.  
+
+<|ref|>text<|/ref|><|det|>[[115, 269, 248, 298]]<|/det|>
+Sincerely, Wouter De Coster  
+
+<|ref|>text<|/ref|><|det|>[[115, 313, 165, 327]]<|/det|>
+Major:  
+
+<|ref|>text<|/ref|><|det|>[[115, 328, 877, 402]]<|/det|>
+I see the authors use SurVCluster to compare SVs from various callset, but the manuscript could be improved by more information on how exactly that is done, and how the precision and sensitivity are calculated based on that. Note that a commonly accepted method for comparing structural variant call sets is Truvari (https://github.com/ACEnglish/truvari). Can the authors confirm that they obtained similar results with their tool?  
+
+<|ref|>text<|/ref|><|det|>[[115, 417, 163, 430]]<|/det|>
+Minor:  
+
+<|ref|>text<|/ref|><|det|>[[115, 431, 856, 447]]<|/det|>
+It appears the authors use the 'HSR' abbreviation and only describe what it means later in the text.  
+
+<|ref|>text<|/ref|><|det|>[[116, 461, 440, 477]]<|/det|>
+The y- axis in Figure 1c and 1d is unlabeled.  
+
+<|ref|>text<|/ref|><|det|>[[115, 491, 866, 522]]<|/det|>
+There is no legend for the bar colors in Figure 5b, but presumably, the colors are the same as for 5a. However, this is not explicitly stated and could lead to confusion.  
+
+<|ref|>text<|/ref|><|det|>[[115, 536, 874, 582]]<|/det|>
+Mostly out of curiosity and not crucial for this paper, I wondered if it is possible to train the filter module on a combination of different species and what the performance of such a mixed model would be compared to the species- specific filter.  
+
+<|ref|>text<|/ref|><|det|>[[116, 624, 402, 640]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 654, 880, 715]]<|/det|>
+This work presents a novel approach, SurVIndel2, to detect copy number variations (CNVs). It looks for sequences that can be aligned to alternative positions to the reference genome as split reads (referred to as hidden split read) and demonstrates enhanced sensitivity and precision from short- read sequencing data, particularly improving detection of duplication in repetitive regions of the genome.  
+
+<|ref|>text<|/ref|><|det|>[[115, 729, 880, 790]]<|/det|>
+Although long- read sequencing can significantly improve CNV calling in repetitive regions, this work acknowledges that short- read sequencing will remain the primary research data in the near future and therefore addressing the limitation of detecting CNVs in repetitive regions using short- read sequencing is of great relevance.  
+
+<|ref|>text<|/ref|><|det|>[[115, 804, 863, 864]]<|/det|>
+The manuscript provides an evaluation of SurVIndel2 with public datasets of multiple organisms and benchmarks its performance with existing CNV and SV callers. However, further analyses and validation on existing tools could strengthen the claims. Please see the list of revisions below for details.  
+
+<|ref|>text<|/ref|><|det|>[[113, 878, 880, 894]]<|/det|>
+The analysis appears methodologically sound overall, but a more detailed discussion on limitations and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 852, 120]]<|/det|>
+potential biases would be beneficial. Any flaws do not seem to prohibit publication but may require revisions (see the list of revisions below) for clarity and completeness.  
+
+<|ref|>text<|/ref|><|det|>[[115, 134, 874, 179]]<|/det|>
+The manuscript clearly described the methodology and provides multiple benchmarking analyses of the method on cell line samples. I believe the work meets current standards in bioinformatics for CNV detection, given the list of revisions are addressed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 193, 863, 223]]<|/det|>
+While SurVIndel2 is open- source software, for full reproducibility, I suggest also including scripts for running the analyses.  
+
+<|ref|>text<|/ref|><|det|>[[115, 238, 241, 253]]<|/det|>
+List of Revisions:  
+
+<|ref|>text<|/ref|><|det|>[[115, 268, 870, 298]]<|/det|>
+1. Figure 2 shows that over \(93\%\) of the tandem duplications of HG002 are in repeat regions. It would be good to see a breakdown of the types of repeat regions in the bar chart.  
+
+<|ref|>text<|/ref|><|det|>[[115, 299, 866, 343]]<|/det|>
+2. In section 2.3, the manuscript stated that filtering was implemented by leave-1-out training for SurVIndel2, but methods of the 'pre-trained random forest' filtering are not provided. Please include further description.  
+
+<|ref|>text<|/ref|><|det|>[[115, 344, 875, 449]]<|/det|>
+3. In addition to revision 2, section 2.3 didn't include descriptions of how filtering was performed on CNV calls of benchmarking tools. For example, Manta ranks the quality of SV calls using QUAL score and FILTER tags, and the choice of filtering strategy can impact the benchmarking results, particularly sensitivity. Figure 4 was generated with one set of calls per sample per caller. Possibilities of alternative filtering strategies were not explored. I recommend including multiple filtering options in the benchmarking, such as different QUAL cutoffs per sample per caller, and plot curves of precision-recall in figure 4(a), 4(b), and 5(a).  
+
+<|ref|>text<|/ref|><|det|>[[115, 449, 852, 493]]<|/det|>
+4. HG002 is a great independent sample, which was not specifically trained on by SurVIndel2, for benchmarking purposes. However, the discussion only includes sensitivity, not precision. I strongly recommend including HG002 in Figure 4(a) and 4(b) as well.  
+
+<|ref|>text<|/ref|><|det|>[[115, 493, 864, 582]]<|/det|>
+5. Very importantly, in section 2.3, Delly, Lumpy, and Manta were selected as benchmarking tools according to reference [4]. However, in this benchmarking study, GRIDSS is another top-performing SV caller. I strongly recommend including benchmarking with GRIDSS2 + Purple (Cameron, D.L., Baber, J., Shale, C. et al. GRIDSS2: comprehensive characterization of somatic structural variation using single breached variants and structural variant phasing. Genome Biol 22, 202 (2021). https://doi.org/10.1186/s13059-021-02423-x).  
+
+<|ref|>text<|/ref|><|det|>[[115, 583, 880, 642]]<|/det|>
+6. It's interesting to see that SurVIndel 1&2 are more sensitive than other callers with no HSR support. Section 2.7 discusses the reasons why these CNVs have a low level of evidence. It would be good to include more discussion on why SurVIndel2 can capture more such CNVs than other callers, as well as the true positive rate of unsupported CNVs called by SurVIndel2.  
+
+<|ref|>text<|/ref|><|det|>[[115, 643, 880, 774]]<|/det|>
+7. I recommend including a discussion on the limitations of the tool regarding the detection of somatic CNVs in cancer genomes. This discussion could acknowledge that, unlike the other tools benchmarked in the study, SurVIndel2 does not currently support the identification of somatic CNVs, which are critical for understanding cancer genomics. Highlight this as a significant caveat, and suggest future research directions or potential updates to the tool that could address this gap. Emphasize the importance of this functionality for comprehensive cancer genomics studies and how its inclusion could further elevate the tool's applicability and utility in the field. This addition will not only provide a more balanced view of SurVIndel2's capabilities but also align expectations for researchers interested in cancer genomics.  
+
+<|ref|>text<|/ref|><|det|>[[116, 819, 402, 834]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 849, 880, 894]]<|/det|>
+This manuscript introduces a new tool SurVIndel2 for detecting the copy number variations (CNV) using short illumina sequencing reads with a novel technique that uses hidden split reads and machine learning techniques. The benchmarking results showed that SurVIndel2 performed better than other
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 872, 120]]<|/det|>
+short- read- based structural variant (SV) callers. However, concerns arise regarding the benchmarking evaluation methods and the availability of results produced in this study.  
+
+<|ref|>text<|/ref|><|det|>[[115, 136, 214, 149]]<|/det|>
+Major issues:  
+
+<|ref|>text<|/ref|><|det|>[[115, 150, 868, 209]]<|/det|>
+1. Throughout the manuscript, it seems the authors have considered all deletion (>=50bp) and tandem duplication type SVs as CNVs. There is confusion as all deletions (>=50bp) can not be called CNV. Is the tool designed to detect all DEL and DUP-type SVs? If that is the case, then it is not suggested to call it a CNV caller tool.  
+
+<|ref|>text<|/ref|><|det|>[[115, 223, 881, 328]]<|/det|>
+2. The authors have claimed that it is a new short-read-based CNV caller, however, in the last few steps it uses the machine learning technique that takes advantage of long-read-based CNV callers (steps g and h) for training purposes. The machine learning model of SurVIndel2 is used in these steps, right? So would it be fair to call it a short-read-based caller? It is expected to work a little better when long-reads or long-read-based callsets are used at the filtering stage, so it won't be a fair comparison with other "pure" short-read-based callers such as Manta, Lumpy, and Delly. Also, Lumpy only calls the DEL-type SVs, so it may not detect any INS (or DUP) type SVs.  
+
+<|ref|>text<|/ref|><|det|>[[115, 342, 876, 431]]<|/det|>
+3. The "Method" section is included inside the "Result" section. Also, that section only talks about the algorithms and techniques used in this tool. It would be better to include the benchmarking strategies such as what comparison tools and the truth set are used for evaluation. The results can be reproduced for verification if the VCF file for the truth sets, callsets, commands, or scripts are available in the "Methods" section. The authors could provide a link (e.g. GitHub repo) with all the result files.  
+
+<|ref|>text<|/ref|><|det|>[[115, 446, 880, 596]]<|/det|>
+4. The authors have used the HGSVC2 callsets as truth sets, so it is not clear if only DEL and DUP type SVs or the whole SV set is selected for the evaluations. Are these high-confidence call sets? Are they validated against GIAB truth sets? For each sample, the evaluation was performed using a training module of SurVIndel which was trained by using the features of other 33 long-read-based CNV callsets. So would it be also possible to compare the results of SurVIndel2 without using the last step to make a fair comparison with "pure" short-read-based callers? The evidence and claims can be better examined with the availability of callsets. Another suggestion is to use the GIAB SV truth set (v0.6) can also be used for the evaluation. The DEL-type SVs with length >=1kbp can be used as a benchmark set while evaluating all the callers. Also, why the precision is not computed for DEL types (last line of 1st paragraph of Section 2.3)?  
+
+<|ref|>text<|/ref|><|det|>[[115, 610, 875, 700]]<|/det|>
+5. All the other tools that are used in this comparison are not designed to call CNVs. Another most commonly used CNV caller tool is CNVnotar. So it is suggested to use that tool to compare the performance of SurVIndel2. Also, the 2nd best-performing tool, Manta, has been improved in the new DRAGEN pipeline released by Illumina. The authors are suggested to take a look at the pre-print (https://doi.org/10.1101/2024.01.02.573821) and the available callsets as some of the evaluation numbers presented in the pre-print for HG002 do not match the analysis done in this manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 715, 863, 789]]<|/det|>
+6. The authors used the terms like "tandem repetitive regions", and "repetitive regions" throughout the manuscript. It would be better if they are mentioned or a BED file is provided to identify those regions in reference. Are they telomere or centromere regions? It is difficult to understand the CNVs outside repetitive regions if they are not clearly defined or explained. Also, it is better to use the length information while using "shorter" terms for CNVs.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 64, 900, 137]]<|/det|>
+We thank the reviewers for the useful feedback. As requested, we have uploaded instructions and code to reproduce the figures of the paper in https://github.com/kensung-lab/survindel2_paper_experiments/We have also updated the data availability, since the data is now available in EBI. We will reply to their reviews point by point:  
+
+<|ref|>sub_title<|/ref|><|det|>[[93, 165, 191, 180]]<|/det|>
+## Reviewer 1  
+
+<|ref|>text<|/ref|><|det|>[[92, 195, 904, 298]]<|/det|>
+The authors describe surVIndel2, a tool for CNV detection, leveraging a previously underappreciated signal - the so- called hidden split reads. This innovative approach appears successful in identifying CNVs, especially in a tandem repeat context, where typical split reads fail to identify the presence of a CNV. The observation that many CNVs are missed due to poor coverage is particularly intriguing, and I wonder if sequence- specific features beyond repetitive context contribute to this. The performance of this tool, compared to well- established tools, is impressive, and I wish to congratulate the authors on their work.  
+
+<|ref|>text<|/ref|><|det|>[[92, 311, 243, 339]]<|/det|>
+Sincerely, Wouter De Coster  
+
+<|ref|>text<|/ref|><|det|>[[93, 353, 344, 367]]<|/det|>
+We appreciate the kind words.  
+
+<|ref|>text<|/ref|><|det|>[[92, 367, 892, 452]]<|/det|>
+Regarding whether there are sequence- specific features that can explain low coverage, it would be a very interesting research direction. We have observed that many of these regions tend to be very repetitive, e.g., low complexity sequences or having long homopolymers. However, we have also observed pairs of very similar low complexity regions in totally different locations where one region was well sequenced while the other was very poorly sequenced. Therefore, there must be factors other than the sequence itself.  
+
+<|ref|>text<|/ref|><|det|>[[92, 452, 888, 482]]<|/det|>
+We would like to emphasize that we have not conducted any rigorous study on this at this point, and these are just simple observations made while developing the tool.  
+
+<|ref|>sub_title<|/ref|><|det|>[[92, 496, 147, 509]]<|/det|>
+## Major:  
+
+<|ref|>text<|/ref|><|det|>[[92, 509, 884, 580]]<|/det|>
+I see the authors use SurVCluster to compare SVs from various callset, but the manuscript could be improved by more information on how exactly that is done, and how the precision and sensitivity are calculated based on that. Note that a commonly accepted method for comparing structural variant call sets is Truvari (https://github.com/ACEnglish/truvari). Can the authors confirm that they obtained similar results with their tool?  
+
+<|ref|>text<|/ref|><|det|>[[92, 593, 875, 666]]<|/det|>
+The main reason why we use SurVCluster for comparing SVs is that it employs a more sophisticated algorithm for comparing insertions and duplications. In long reads- based benchmarks such as HGSVC2, duplications are reported as point insertions. Short read based tools often report them as tandem duplications, denoted with the start and the end of the duplicated region.  
+
+<|ref|>text<|/ref|><|det|>[[92, 666, 896, 708]]<|/det|>
+One issue of such representation is that we do not know how many times the region was duplicated (the caller may report an estimate, but it is generally difficult to estimate accurately for duplications that are not extremely long, i.e., 1000s of bps).  
+
+<|ref|>text<|/ref|><|det|>[[92, 708, 900, 780]]<|/det|>
+Therefore, if a tool reports that a region B was duplicated, and the benchmark reports an insertion of a sequence BBBB in the same location, the comparison algorithm in SurVClusterer identifies that the two represent the same event. On the other hand, it appears to us (by reading the algorithm explained in the documentation) that truvari assumes every tandem duplication is duplicated exactly once.  
+
+<|ref|>text<|/ref|><|det|>[[92, 793, 901, 850]]<|/det|>
+One clear example of this is the benchmark SV chr6- 168229517- INS- 512 in NA24385. SurVIndel2 predicts a duplication of 128 bp, and the inserted sequence reported by the benchmark is identical to the duplicated region predicted by SurVIndel2, duplicated 4 times. SurVClusterer reports the two events as matching, while truvari does not.  
+
+<|ref|>text<|/ref|><|det|>[[92, 851, 903, 923]]<|/det|>
+In practice, many cases are less clear- cut; for example, many tandem duplications are caused by expansions of short motifs. When such expansions are not much smaller than the read length, it is nearly impossible to identify the exact length of the expansion using short reads (especially given that many of them are in the very noisy regions that we describe in Section 2.7), and it is often underestimated when compared to HiFi reads.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 79, 872, 110]]<|/det|>
+Secondly, SurVClusterer uses the TRF annotations to determine whether two events within the same repetitive region represent the same event but shifted.  
+
+<|ref|>text<|/ref|><|det|>[[92, 121, 867, 167]]<|/det|>
+In any case, we did try using truvari to compare Manta and SurVIndel2 (since they perform far better than the rest). Instructions and data are reported in https://github.com/kensung- lab/ survivedl2_paper_experiments/, under truvari.  
+
+<|ref|>text<|/ref|><|det|>[[92, 179, 489, 195]]<|/det|>
+For deletions, we obtained the following results:  
+
+<|ref|>text<|/ref|><|det|>[[92, 207, 175, 222]]<|/det|>
+HG00512  
+
+<|ref|>text<|/ref|><|det|>[[92, 236, 407, 266]]<|/det|>
+Manta: 0.36 recall, 0.72 precisionSurVIndel2: 0.56 recall, 0.80 precision  
+
+<|ref|>text<|/ref|><|det|>[[92, 278, 154, 293]]<|/det|>
+HG002  
+
+<|ref|>text<|/ref|><|det|>[[92, 307, 407, 336]]<|/det|>
+Manta: 0.36 recall, 0.71 precisionSurVIndel2: 0.56 recall, 0.75 precision  
+
+<|ref|>text<|/ref|><|det|>[[92, 350, 890, 380]]<|/det|>
+Since the matching criteria of the two comparison algorithms are different, the numbers are a bit different; however, the relative performance are the same.  
+
+<|ref|>text<|/ref|><|det|>[[92, 392, 900, 437]]<|/det|>
+For tandem duplications, because of the reasons explained above, we have disabled the comparison of the inserted sequences. If activated, the precision of the algorithms drop to around 0.6.  
+
+<|ref|>text<|/ref|><|det|>[[92, 450, 175, 465]]<|/det|>
+HG00512  
+
+<|ref|>text<|/ref|><|det|>[[92, 479, 407, 521]]<|/det|>
+Manta: 0.13 recall, 0.78 precisionSurVIndel2: 0.37 recall, 0.83 precisionHG002  
+
+<|ref|>text<|/ref|><|det|>[[92, 535, 407, 564]]<|/det|>
+Manta: 0.14 recall, 0.77 precisionSurVIndel2: 0.41 recall, 0.82 precision  
+
+<|ref|>text<|/ref|><|det|>[[92, 577, 803, 607]]<|/det|>
+However, this comparison is perhaps too relaxed, and we believe the one operated by SurVClusterer is closer to reality.  
+
+<|ref|>text<|/ref|><|det|>[[92, 606, 866, 650]]<|/det|>
+It must be noted that comparing SVs is very challenging, due to repetitive regions, imprecise breakpoints, imprecise inserted sequences, multiple valid representations for the same event, etc..  
+
+<|ref|>text<|/ref|><|det|>[[92, 650, 835, 694]]<|/det|>
+We are always looking for improvements and cases where our algorithm fails. Aside from benchmarking, being able to correctly identify matching SVs is especially important when clustering SVs in a population.  
+
+<|ref|>text<|/ref|><|det|>[[92, 706, 860, 736]]<|/det|>
+We also added a paragraph to the supplementary section "Comparing deletions and tandem duplications" stating how recall and precision are calculated.  
+
+<|ref|>text<|/ref|><|det|>[[92, 749, 147, 763]]<|/det|>
+Minor:  
+
+<|ref|>text<|/ref|><|det|>[[92, 763, 866, 792]]<|/det|>
+It appears the authors use the 'HSR' abbreviation and only describe what it means later in the text.  
+
+<|ref|>text<|/ref|><|det|>[[92, 805, 813, 822]]<|/det|>
+Indeed, we never introduced the HSR abbreviation formally. We added it to Section 2.1.  
+
+<|ref|>text<|/ref|><|det|>[[92, 834, 452, 850]]<|/det|>
+The y- axis in Figure 1c and 1d is unlabeled.  
+
+<|ref|>text<|/ref|><|det|>[[92, 862, 844, 892]]<|/det|>
+Since Figure 1 has no plot with x and y axes in it, we wonder if the reviewer is referring to a different figure. We have indeed found some panels without y- axis label:  
+
+<|ref|>text<|/ref|><|det|>[[92, 892, 235, 920]]<|/det|>
+- Fig. 2f and S6f- Fig. 4c and d
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 65, 472, 95]]<|/det|>
+- Fig. S13c (now S15c) and S14c (now S16c) We added a label to all of them.  
+
+<|ref|>text<|/ref|><|det|>[[92, 107, 890, 138]]<|/det|>
+There is no legend for the bar colors in Figure 5b, but presumably, the colors are the same as for 5a. However, this is not explicitly stated and could lead to confusion.  
+
+<|ref|>text<|/ref|><|det|>[[92, 150, 664, 167]]<|/det|>
+We now explicitly state that the legend of panel (a) also applies to (b).  
+
+<|ref|>text<|/ref|><|det|>[[92, 178, 885, 223]]<|/det|>
+Mostly out of curiosity and not crucial for this paper, I wondered if it is possible to train the filter module on a combination of different species and what the performance of such a mixed model would be compared to the species- specific filter.  
+
+<|ref|>text<|/ref|><|det|>[[92, 236, 898, 309]]<|/det|>
+It is not easy to answer this question, as there are currently few (that we could find) datasets having both HiFi and Illumina reads. The reason we need HiFi is that we need to divide our training data into false and true positives. To do so, we require a benchmark that reports accurate inserted sequences. At the time of our experiments, we found that long read SV callers only produced accurate inserted sequences with HiFi reads.  
+
+<|ref|>text<|/ref|><|det|>[[92, 321, 897, 395]]<|/det|>
+Out of curiosity, we tried training a model on both humans and Arabidopsis data, and tried to predict an Arabidopsis sample (which was left out of training). The results are similar to using Arabidopsis data only, perhaps slightly worse: compared to using Arabidopsis only, for deletions, the mixed model has 0.02 lower sensitivity and 0.01 higher precision. For duplications, the mixed model has 0.01 higher sensitivity but 0.05 lower precision.  
+
+<|ref|>text<|/ref|><|det|>[[92, 394, 898, 438]]<|/det|>
+We also tried to classify a different organism (cattle), and there is no appreciable difference between the human- only model and the mixed model (very tiny increase in predicted TPs, but not enough to generate a change in sensitivity).  
+
+<|ref|>text<|/ref|><|det|>[[92, 436, 860, 465]]<|/det|>
+It is possible that a model that is larger or more sophisticated could utilize the mixed training model better.  
+
+<|ref|>sub_title<|/ref|><|det|>[[93, 479, 193, 494]]<|/det|>
+## Reviewer 2  
+
+<|ref|>text<|/ref|><|det|>[[92, 508, 899, 582]]<|/det|>
+This work presents a novel approach, SurVlndel2, to detect copy number variations (CNVs). It looks for sequences that can be aligned to alternative positions to the reference genome as split reads (referred to as hidden split read) and demonstrates enhanced sensitivity and precision from short- read sequencing data, particularly improving detection of duplication in repetitive regions of the genome.  
+
+<|ref|>text<|/ref|><|det|>[[92, 594, 892, 653]]<|/det|>
+Although long- read sequencing can significantly improve CNV calling in repetitive regions, this work acknowledges that short- read sequencing will remain the primary research data in the near future and therefore addressing the limitation of detecting CNVs in repetitive regions using short- read sequencing is of great relevance.  
+
+<|ref|>text<|/ref|><|det|>[[92, 665, 901, 723]]<|/det|>
+The manuscript provides an evaluation of SurVlndel2 with public datasets of multiple organisms and benchmarks its performance with existing CNV and SV callers. However, further analyses and validation on existing tools could strengthen the claims. Please see the list of revisions below for details.  
+
+<|ref|>text<|/ref|><|det|>[[92, 736, 900, 780]]<|/det|>
+The analysis appears methodologically sound overall, but a more detailed discussion on limitations and potential biases would be beneficial. Any flaws do not seem to prohibit publication but may require revisions (see the list of revisions below) for clarity and completeness.  
+
+<|ref|>text<|/ref|><|det|>[[92, 793, 897, 837]]<|/det|>
+The manuscript clearly described the methodology and provides multiple benchmarking analyses of the method on cell line samples. I believe the work meets current standards in bioinformatics for CNV detection, given the list of revisions are addressed.  
+
+<|ref|>text<|/ref|><|det|>[[92, 850, 889, 880]]<|/det|>
+While SurVlndel2 is open- source software, for full reproducibility, I suggest also including scripts for running the analyses.  
+
+<|ref|>text<|/ref|><|det|>[[92, 893, 888, 923]]<|/det|>
+We would like to thank the reviewer. As suggested, we have published the scripts for generating the main figures in https://github.com/kensung- lab/survindel2_paper_experiments/.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[93, 66, 473, 80]]<|/det|>
+Below, we will reply to the individual revisions.  
+
+<|ref|>sub_title<|/ref|><|det|>[[93, 110, 234, 125]]<|/det|>
+## List of Revisions:  
+
+<|ref|>text<|/ref|><|det|>[[93, 140, 870, 170]]<|/det|>
+1. Figure 2 shows that over \(93\%\) of the tandem duplications of HG002 are in repeat regions. It would be good to see a breakdown of the types of repeat regions in the bar chart.  
+
+<|ref|>text<|/ref|><|det|>[[92, 182, 890, 241]]<|/det|>
+We use TRF to identify tandem repetitive regions (Supplementary). We have added a sentence in Section 2.1 that explicitly states this. As far as we know, TRF does not classify repetitive regions into subtypes, and I am currently not aware of such a classification (as opposed as general repeats, which can be classified into mobile elements, tandem repetitive regions, etc.).  
+
+<|ref|>text<|/ref|><|det|>[[92, 254, 890, 297]]<|/det|>
+2. In section 2.3, the manuscript stated that filtering was implemented by leave-1-out training for SurVIndel2, but methods of the 'pre-trained random forest' filtering are not provided. Please include further description.  
+
+<|ref|>text<|/ref|><|det|>[[92, 310, 707, 325]]<|/det|>
+We added a new Supplementary section called "Filtering module training".  
+
+<|ref|>text<|/ref|><|det|>[[92, 339, 901, 440]]<|/det|>
+3. In addition to revision 2, section 2.3 didn't include descriptions of how filtering was performed on CNV calls of benchmarking tools. For example, Manta ranks the quality of SV calls using QUAL score and FILTER tags, and the choice of filtering strategy can impact the benchmarking results, particularly sensitivity. Figure 4 was generated with one set of calls per sample per caller. Possibilities of alternative filtering strategies were not explored. I recommend including multiple filtering options in the benchmarking, such as different QUAL cutoffs per sample per caller, and plot curves of precision-recall in figure 4(a), 4(b), and 5(a).  
+
+<|ref|>text<|/ref|><|det|>[[92, 453, 884, 511]]<|/det|>
+We retained calls with FILTER \(= =\) PASS, since this is in our opinion the standard use case. We added a Supplementary section called "Selected callers" where we specify this. We have also generated a precision/recall curve for deletions in HG00512 and HG002 (Smoot, SurVIndel and SurVIndel2 do not report a QUAL value and therefore they are just single points).  
+
+<|ref|>image<|/ref|><|det|>[[150, 570, 828, 875]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[149, 193, 828, 500]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[92, 537, 898, 597]]<|/det|>
+We could not generate a curve for duplications because for Delly and Manta we use two different call sets to calculate recall and precision: precision is calculated on SVTYPE=DUP, while for recall we also include SVTYPE=INS (since Delly and Manta may decide to report events as either DUP or INS).  
+
+<|ref|>text<|/ref|><|det|>[[92, 609, 835, 640]]<|/det|>
+The code and data for generating the curves was uploaded to the aforementioned GitHub repository.  
+
+<|ref|>text<|/ref|><|det|>[[92, 652, 896, 696]]<|/det|>
+4. HG002 is a great independent sample, which was not specifically trained on by SurVlndel2, for benchmarking purposes. However, the discussion only includes sensitivity, not precision. I strongly recommend including HG002 in Figure 4(a) and 4(b) as well.  
+
+<|ref|>text<|/ref|><|det|>[[92, 709, 896, 835]]<|/det|>
+HG002 is part of the HGSVC2, as mentioned in Section 2.3: a catalogue (called HGSVC2) of insertions and deletions in 35 human genomes (34 from the 1000 genomes project, plus HG002). In HGSVC2, it is called NA24385. Therefore, it is in Figure 4a and 4b. The purpose of figures 4c and 4d is simply to show that existing callers struggle with HSR-supported calls. We have added two more benchmarks for HG002 (Fig. S7 and an additional paragraph in Section 2.3). One was the GIAB v0.6 benchmark, while the other was obtained by running Sniffles2 on HiFi reads. The relative performance of the tools are in line with what was observed on the HGSVC2 benchmark. Instructions and files to reproduce the benchmark are in the GitHub repository.  
+
+<|ref|>text<|/ref|><|det|>[[92, 855, 895, 911]]<|/det|>
+5. Very importantly, in section 2.3, Delly, Lumpy, and Manta were selected as benchmarking tools according to reference [4]. However, in this benchmarking study, GRIDSS is another top-performing SV caller. I strongly recommend including benchmarking with GRIDSS2 + Purple (Cameron, D.L., Baber, J., Shale, C. et al. GRIDSS2: comprehensive characterization of somatic
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 66, 900, 95]]<|/det|>
+structural variation using single brekend variants and structural variant phasing. Genome Biol 22, 202 (2021). https://doi.org/10.1186/s13059- 021- 02423- x).  
+
+<|ref|>text<|/ref|><|det|>[[92, 108, 896, 150]]<|/det|>
+We have tried using GRIDSS, but it does not explicitly report deletions and duplications, only breakends. One criteria we used is that the callers should explicitly predict deletions and tandem duplications.  
+
+<|ref|>text<|/ref|><|det|>[[92, 150, 897, 209]]<|/det|>
+In the GitHub README, the authors mention that example/simple- event- annotation.R will annotate the calls with SIMPLE_TYPE and SVLEN, but the script is not made for general users as it has hardcoded paths inside. We modified the hardcoded path to point to our own GRIDSS results, but the script printed this:  
+
+<|ref|>text<|/ref|><|det|>[[92, 223, 688, 309]]<|/det|>
+Error in .subassign_columns(x, nsbs, value) : provided 68 variables to replace 67 variables Calls: [< - ... [<- -> mergeROWS -> mergeROWS -> .subassign_columns In addition: Warning message: info fields with no header: SVLEN Execution halted  
+
+<|ref|>text<|/ref|><|det|>[[92, 321, 904, 380]]<|/det|>
+In the benchmarking paper "Comprehensive evaluation and characterisation of short read general- purpose structural variant calling software" by Cameron DL et al. (the authors of GRIDSS), it is shown that GRIDSS performs similarly to Manta on HG002 (Fig.1, slightly more precise but less sensitive).  
+
+<|ref|>text<|/ref|><|det|>[[92, 393, 896, 451]]<|/det|>
+6. It's interesting to see that SurVlndel 1&2 are more sensitive than other callers with no HSR support. Section 2.7 discusses the reasons why these CNVs have a low level of evidence. It would be good to include more discussion on why SurVlndel2 can capture more such CNVs than other callers, as well as the true positive rate of unsupported CNVs called by SurVlndel2.  
+
+<|ref|>text<|/ref|><|det|>[[92, 464, 895, 494]]<|/det|>
+We investigated why SurVlndel2 can detect a portion of deletions marked as unsupported. There seem to be many factors contributing.  
+
+<|ref|>text<|/ref|><|det|>[[92, 494, 880, 536]]<|/det|>
+We found that \(38\%\) of these deletions are predicted using HSRs. As for why such deletions are classified as unsupported despite having HSRs supporting them, we identified three possible reasons:  
+
+<|ref|>text<|/ref|><|det|>[[92, 536, 870, 580]]<|/det|>
+(1) spoa built an alternative allele that is not \(100\%\) correct, and it escaped our filters; 
+(2) the HSRs are less than 5 (minimum requirement to be classified as HSR-supported), while SurVlndel2 is able to identify CNVs from as few as 3 reads;  
+
+<|ref|>text<|/ref|><|det|>[[92, 580, 870, 636]]<|/det|>
+(3) when mapping the short reads to the set of alternative alleles, BWA might have placed the relevant reads somewhere else (since we mapped the reads to all of the alternative alleles together, as running BWA for thousands of alternative allele separately would have been unfeasible with our computational resources).  
+
+<|ref|>text<|/ref|><|det|>[[92, 636, 888, 694]]<|/det|>
+Another \(38\%\) of them were predicted using the discordant pairs module, so no SR or HSR support was required. Finally, out of the remaining \(24\%\) that were predicted using split reads, nearly half of them had less than 5 split reads supporting them (minimum support required to be classified as SR-supported was again 5).  
+
+<|ref|>text<|/ref|><|det|>[[92, 694, 625, 709]]<|/det|>
+We have added a small paragraph explaining this to Section 2.3.  
+
+<|ref|>text<|/ref|><|det|>[[92, 722, 899, 850]]<|/det|>
+7. I recommend including a discussion on the limitations of the tool regarding the detection of somatic CNVs in cancer genomes. This discussion could acknowledge that, unlike the other tools benchmarked in the study, SurVlndel2 does not currently support the identification of somatic CNVs, which are critical for understanding cancer genomics. Highlight this as a significant caveat, and suggest future research directions or potential updates to the tool that could address this gap. Emphasize the importance of this functionality for comprehensive cancer genomics studies and how its inclusion could further elevate the tool's applicability and utility in the field. This addition will not only provide a more balanced view of SurVlndel2's capabilities but also align expectations for researchers interested in cancer genomics.  
+
+<|ref|>text<|/ref|><|det|>[[92, 863, 865, 907]]<|/det|>
+Indeed, SurVlndel2 does no explicitly support detecting somatic variants (excluding the naive approach of processing tumor and normal independently and operating a subtraction). I have added a paragraph to the Discussion explaining this.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[93, 66, 193, 81]]<|/det|>
+## Reviewer 3  
+
+<|ref|>text<|/ref|><|det|>[[93, 95, 902, 170]]<|/det|>
+This manuscript introduces a new tool SurVlndel2 for detecting the copy number variations (CNV) using short illumina sequencing reads with a novel technique that uses hidden split reads and machine learning techniques. The benchmarking results showed that SurVlndel2 performed better than other short- read- based structural variant (SV) callers. However, concerns arise regarding the benchmarking evaluation methods and the availability of results produced in this study.  
+
+<|ref|>text<|/ref|><|det|>[[93, 181, 812, 198]]<|/det|>
+We would like to the thank the reviewer. Below, we reply to their comments individually:  
+
+<|ref|>sub_title<|/ref|><|det|>[[93, 212, 204, 226]]<|/det|>
+## Major issues:  
+
+<|ref|>text<|/ref|><|det|>[[93, 226, 899, 283]]<|/det|>
+1. Throughout the manuscript, it seems the authors have considered all deletion (>= 50bp) and tandem duplication type SVs as CNVs. There is confusion as all deletions (>=50bp) can not be called CNV. Is the tool designed to detect all DEL and DUP-type SVs? If that is the case, then it is not suggested to call it a CNV caller tool.  
+
+<|ref|>text<|/ref|><|det|>[[92, 296, 881, 370]]<|/det|>
+By reading the literature, it appears to us that there is no universally accepted definition of CNV. Some definitions require a length of at least 1000 bp, other 50 bp: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9407502/ https://genomedicine.biomedcentral.com/articles/10.1186/s13073-021-00945-4 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9365719/  
+
+<|ref|>text<|/ref|><|det|>[[92, 381, 890, 442]]<|/det|>
+However, it is true that we do not detect all CNVs, because we only detect tandem duplications (as opposed to interspersed duplications). In the original SurVlndel, we created the term local CNV to refer to deletions and tandem duplications only. We have clarified in the introduction that in the paper we consider local CNVs, and we will refer to them simply as CNV for convenience.  
+
+<|ref|>text<|/ref|><|det|>[[92, 455, 901, 557]]<|/det|>
+2. The authors have claimed that it is a new short-read-based CNV caller, however, in the last few steps it uses the machine learning technique that takes advantage of long-read-based CNV callers (steps g and h) for training purposes. The machine learning model of SurVlndel2 is used in these steps, right? So would it be fair to call it a short-read-based caller? It is expected to work a little better when long-reads or long-read-based callsets are used at the filtering stage, so it won't be a fair comparison with other "pure" short-read-based callers such as Manta, Lumpy, and Delly. Also, Lumpy only calls the DEL-type SVs, so it may not detect any INS (or DUP) type SVs.  
+
+<|ref|>text<|/ref|><|det|>[[92, 569, 903, 690]]<|/det|>
+We distribute the data used to train the model (as the trained model is several GBs and difficult to distribute, and training the model requires modest computational resources). It is true that the data was obtained with the help of long reads. However, as mentioned, no long reads are required to run SurVlndel2, only short reads, and for this reason we believe it to be a short reads caller. It is true that if the user wants to train its model on different samples (for example, a different organism) he will need a high quality ground truth set for those samples, which are usually obtained by long reads. However, we have demonstrated in the paper that the default model provided works well in several non- human organisms. Lumpy calls deletions, duplications, inversions and generic "breakends".  
+
+<|ref|>text<|/ref|><|det|>[[92, 714, 904, 812]]<|/det|>
+Regarding the fairness of the comparison, we only want to demonstrate that the tool will be useful to the community. Delly, Lumpy and Manta are seminal works that have been developed nearly a decade ago. We have been able to develop our algorithm not only thanks to them, but also thanks to data that has been published since then, such as the many long reads datasets, novel catalogues such as HGSVC2 that are more complete than ever, and even high quality assemblies such as HG002. We have developed our insights and ultimately our algorithm using this data, which was not available at the time the other tools were developed.  
+
+<|ref|>text<|/ref|><|det|>[[92, 825, 902, 911]]<|/det|>
+3. The "Method" section is included inside the "Result" section. Also, that section only talks about the algorithms and techniques used in this tool. It would be better to include the benchmarking strategies such as what comparison tools and the truth set are used for evaluation. The results can be reproduced for verification if the VCF file for the truth sets, callsets, commands, or scripts are available in the "Methods" section. The authors could provide a link (e.g. GitHub repo) with all the result files.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 65, 904, 210]]<|/det|>
+4. The authors have used the HGSVC2 callsets as truth sets, so it is not clear if only DEL and DUP type SVs or the whole SV set is selected for the evaluations. Are these high-confidence call sets? Are they validated against GIAB truth sets? For each sample, the evaluation was performed using a training module of SurVlndel which was trained by using the features of other 33 long-read-based CNV callsets. So would it be also possible to compare the results of SurVlndel2 without using the last step to make a fair comparison with "pure" short-read-based callers? The evidence and claims can be better examined with the availability of callsets. Another suggestion is to use the GIAB SV truth set (v0.6) can also be used for the evaluation. The DEL-type SVs with length \(> = 1\) kbp can be used as a benchmark set while evaluating all the callers. Also, why the precision is not computed for DEL types (last line of 1st paragraph of Section 2.3)?  
+
+<|ref|>text<|/ref|><|det|>[[94, 223, 422, 238]]<|/det|>
+We will respond to both points together.  
+
+<|ref|>text<|/ref|><|det|>[[92, 250, 883, 366]]<|/det|>
+We understand this is confusing, since we failed to mention it explicitly. Tandem duplications in the HGSVC2 benchmark, but also in the Sniffles2 results, are reported as insertions. We have a method that we have previously published in the Supplementary of https://www.nature.com/articles/s41467- 023- 38870- 2 that identifies which insertions are due to tandem duplication. We tried to make it clearer by explicitly mentioning this in 2.3 and 2.4, and also clarifying how sensitivity and precision are calculated in the supplementary section "Comparing deletions and tandem duplications". We only use tandem duplications to calculate sensitivity, while we use all benchmark insertions to calculate precision.  
+
+<|ref|>text<|/ref|><|det|>[[92, 378, 810, 408]]<|/det|>
+We have published all the instructions, data and code to replicate our figures in https://github.com/kensung-lab/survindel2_paper_experiments/.  
+
+<|ref|>text<|/ref|><|det|>[[92, 420, 896, 520]]<|/det|>
+We believe HGSVC2 to be high- confidence, since it is produced by a well known consortium (the Human Genome SV Consortium) using HiFi reads. It is generally more complete than the GIAB benchmark (although, in our experience, it contains more false positives). They are hard to compare since GIAB is, as far as we know, hg19 only. In any case, we have added two more benchmarks for HG002: the GIAB one, and one generated by Sniffles2 (Fig. S7). Rather than \(\geq 1000\) bp, we used \(\geq 50\) bp, since all the tools involved are designed to call SV- type events.  
+
+<|ref|>text<|/ref|><|det|>[[92, 533, 901, 578]]<|/det|>
+Unfortunately the unfiltered callset of SurVlndel2 is very noisy, as it contains \(>100,000\) calls, and it is certainly not recommended to use it directly. For this reason, it is not meaningful to benchmark it.  
+
+<|ref|>text<|/ref|><|det|>[[92, 591, 896, 636]]<|/det|>
+Precision was computed for all samples, including HG002 (which is the sample NA24385 in HGSVC2). It is the y- axis in Fig. 4a and 4b. As mentioned, we have added two more benchmarks for HG002, and we computed sensitivity and precision.  
+
+<|ref|>text<|/ref|><|det|>[[92, 649, 900, 751]]<|/det|>
+5. All the other tools that are used in this comparison are not designed to call CNVs. Another most commonly used CNV caller tool is CNVnator. So it is suggested to use that tool to compare the performance of SurVlndel2. Also, the 2nd best-performing tool, Manta, has been improved in the new DRAGEN pipeline released by Illumina. The authors are suggested to take a look at the pre-print (https://doi.org/10.1101/2024.01.02.573821) and the available callsets as some of the evaluation numbers presented in the pre-print for HG002 do not match the analysis done in this manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[92, 763, 887, 822]]<|/det|>
+In the original SurVlndel paper, we did include CNVnator, and it performed much worse than the callers included here: https://academic.oup.com/bioinformatics/article/37/11/1497/5466452. Also, the scope of CNVnator is different, as it is designed to detect very large CNVs, while SurVlndel2 and the other callers tested are designed to detect \(\geq 50\) bp events (or even smaller).  
+
+<|ref|>text<|/ref|><|det|>[[92, 835, 894, 921]]<|/det|>
+We did not compare DRAGEN to free and open- source because it is proprietary, closed and requires a license. However, we recognize its prominence in the field and since the paper the reviewer mentions does provide calls for 1KGP, including the samples in HGSVC2, we included a comparison (Section 2.3 and Fig. S8). DRAGEN is a very clear improvement over Manta, but its sensitivity and recall are still sensibly lower than SurVlndel2. As usual, the GitHub repository contains code and data for this comparison.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 78, 902, 152]]<|/det|>
+6. The authors used the terms like "tandem repetitive regions", and "repetitive regions" throughout the manuscript. It would be better if they are mentioned or a BED file is provided to identify those regions in reference. Are they telomere or centromere regions? It is difficult to understand the CNVs outside repetitive regions if they are not clearly defined or explained. Also, it is better to use the length information while using "shorter" terms for CNVs.  
+
+<|ref|>text<|/ref|><|det|>[[92, 164, 872, 195]]<|/det|>
+We now specify in Section 2.1 that we consider tandem repetitive regions identified by the Tandem Repeats Finder (TRF). In the GitHub we provide a file with the list of repetitive regions.  
+
+<|ref|>text<|/ref|><|det|>[[93, 207, 173, 222]]<|/det|>
+Sincerely,  
+
+<|ref|>text<|/ref|><|det|>[[93, 234, 395, 251]]<|/det|>
+Ramesh Rajaby and Wing- Kin Sung.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 89, 305, 105]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 119, 404, 135]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 149, 217, 163]]<|/det|>
+Dear authors,  
+
+<|ref|>text<|/ref|><|det|>[[115, 179, 857, 223]]<|/det|>
+I am satisfied with the clarifications made regarding my previous request, and wish to congratulate you on your manuscript and tool. I have no further comments.  
+
+<|ref|>text<|/ref|><|det|>[[115, 238, 250, 268]]<|/det|>
+Sincerely, Wouter De Coster  
+
+<|ref|>text<|/ref|><|det|>[[116, 297, 404, 312]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 327, 876, 401]]<|/det|>
+The authors have addressed my concerns and suggestions raised in my previous review. The revisions made to the manuscript and supplementary methods, including the clarified analyses and expanded discussions, have improved the clarity of the study. The additional instructions and code provided transparency and reproducibility of the published results. Therefore, I recommend this manuscript for publication.  
+
+<|ref|>text<|/ref|><|det|>[[115, 416, 447, 431]]<|/det|>
+Reviewer #2 (Remarks on code availability):  
+
+<|ref|>text<|/ref|><|det|>[[115, 446, 872, 491]]<|/det|>
+The repository for SurVIndel2 (https://github.com/kensung- lab/SurVIndel2) includes a clear README file with installation and running instructions. I successfully installed SurVIndel2 by following these instructions on Rocky Linux v8.9 (Green Obsidian).  
+
+<|ref|>text<|/ref|><|det|>[[115, 506, 872, 536]]<|/det|>
+The repository also provides clear instructions for reproducing the main figures. However, due to time constraints, I have not attempted to reproduce the results presented in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[116, 565, 404, 580]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 595, 870, 640]]<|/det|>
+I would like to thank the authors for putting a great effort and doing additional analysis as suggested by the reviewers. They have indeed provided more explanations for most of the comments and suggestions.  
+
+<|ref|>text<|/ref|><|det|>[[115, 655, 722, 670]]<|/det|>
+However, there are still a few additional concerns that may need to be addressed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 685, 877, 790]]<|/det|>
+In the earlier version of the manuscript, it was a bit unclear about the validation tool/software that was used for all the analysis, and Reviewer #1 suggested using Truvari. I think Truvari has been considered a standard tool for SV validation. Also, some recent results have used Wittery (https://github.com/Illumina/witty.er) for this purpose. The authors have clarified that they have used a tool "SurVClusterer" which was developed by them and also clarified some reasons why it may be better than Truvari. They have also provided some additional results based on truvari. I have tried to replicate some of the results by running truvari. Here are some comments based on that result.  
+
+<|ref|>text<|/ref|><|det|>[[115, 804, 877, 849]]<|/det|>
+1. The file "1_data_preparation/by-sample/HG00512.DEL.vcf.gz" was not present on their Github repo (https://github.com/kensung-lab/survindel2_paper_experiments/). So it would be great if they upload at least DEL benchmark results for few samples.  
+
+<|ref|>text<|/ref|><|det|>[[112, 863, 847, 893]]<|/det|>
+2. For HG002 with hg19 reference calls, truvari failed with error "ERROR:root:No SAMPLE columns found in vcf", though it worked for MANta and DRAGEN vcfs.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 103, 820, 135]]<|/det|>
+I think it would be great if authors could show that the performance of their tool is better than Manta/DRAGEN (and others) by running Truvari on the DEL variant types.  
+
+<|ref|>text<|/ref|><|det|>[[115, 148, 448, 164]]<|/det|>
+Reviewer #3 (Remarks on code availability):  
+
+<|ref|>text<|/ref|><|det|>[[113, 178, 870, 195]]<|/det|>
+The codes are reproducible, but some files are still missing and also some info inside VCFs is missing.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 65, 835, 81]]<|/det|>
+We thank the reviewers for the kind feedback. We will reply to their reviews point by point:  
+
+<|ref|>sub_title<|/ref|><|det|>[[93, 109, 191, 124]]<|/det|>
+## Reviewer 1  
+
+<|ref|>text<|/ref|><|det|>[[93, 139, 205, 154]]<|/det|>
+Dear authors,  
+
+<|ref|>text<|/ref|><|det|>[[92, 167, 798, 210]]<|/det|>
+I am satisfied with the clarifications made regarding my previous request, and wish to congratulate you on your manuscript and tool. I have no further comments.  
+
+<|ref|>text<|/ref|><|det|>[[93, 225, 245, 253]]<|/det|>
+Sincerely, Wouter De Coster  
+
+<|ref|>text<|/ref|><|det|>[[94, 266, 275, 282]]<|/det|>
+Thank you very much.  
+
+<|ref|>sub_title<|/ref|><|det|>[[93, 296, 192, 311]]<|/det|>
+## Reviewer 2  
+
+<|ref|>text<|/ref|><|det|>[[93, 325, 900, 399]]<|/det|>
+The authors have addressed my concerns and suggestions raised in my previous review. The revisions made to the manuscript and supplementary methods, including the clarified analyses and expanded discussions, have improved the clarity of the study. The additional instructions and code provided transparency and reproducibility of the published results. Therefore, I recommend this manuscript for publication.  
+
+<|ref|>text<|/ref|><|det|>[[93, 411, 856, 455]]<|/det|>
+The repository for SurVlndel2 (https://github.com/kensung- lab/SurVlndel2) includes a clear README file with installation and running instructions. I successfully installed SurVlndel2 by following these instructions on Rocky Linux v8.9 (Green Obsidian).  
+
+<|ref|>text<|/ref|><|det|>[[92, 468, 890, 498]]<|/det|>
+The repository also provides clear instructions for reproducing the main figures. However, due to time constraints, I have not attempted to reproduce the results presented in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[94, 511, 275, 526]]<|/det|>
+Thank you very much.  
+
+<|ref|>sub_title<|/ref|><|det|>[[93, 541, 192, 555]]<|/det|>
+## Reviewer 3  
+
+<|ref|>text<|/ref|><|det|>[[93, 570, 840, 614]]<|/det|>
+I would like to thank the authors for putting a great effort and doing additional analysis as suggested by the reviewers. They have indeed provided more explanations for most of the comments and suggestions.  
+
+<|ref|>text<|/ref|><|det|>[[93, 627, 765, 643]]<|/det|>
+However, there are still a few additional concerns that may need to be addressed.  
+
+<|ref|>text<|/ref|><|det|>[[92, 655, 903, 771]]<|/det|>
+In the earlier version of the manuscript, it was a bit unclear about the validation tool/software that was used for all the analysis, and Reviewer#1 suggested using Truvari. I think Truvari has been considered a standard tool for SV validation. Also, some recent results have used Wittyer (https://github.com/Illumina/witty.er) for this purpose. The authors have clarified that they have used a tool "SurVClusterer" which was developed by them and also clarified some reasons why it may be better than Truvari. They have also provided some additional results based on truvari. I have tried to replicate some of the results by running truvari. Here are some come comments based on that result.  
+
+<|ref|>text<|/ref|><|det|>[[93, 784, 895, 828]]<|/det|>
+1. The file "1_data_preparation/by-sample/HG00512.DEL.vcf.gz" was not present on their Github repo (https://github.com/kensung-lab/survindel2_paper_experiments/). So it would be great if they upload at least DEL benchmark results for few samples.  
+
+<|ref|>text<|/ref|><|det|>[[93, 841, 877, 885]]<|/det|>
+Our approach in providing the steps for reproducing our analyses was to, as much as possible, start from the source public data. For this reason, the instructions in the README of 1_data_preparation should be executed before attempting to reproduce any other figure.  
+
+<|ref|>text<|/ref|><|det|>[[92, 898, 821, 927]]<|/det|>
+In this particular case, this means downloading the HGSVC2 dataset and generating the benchmark datasets for each individual sample:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 65, 830, 95]]<|/det|>
+wget https://ftp.1000genomes.ebi.ac.uk/vol1/ftp/data collections/HGSVC2/release/v2.0/integrated_callset/variants_freeze4_sv_insdel_alt.vcf.gz  
+
+<|ref|>text<|/ref|><|det|>[[94, 107, 412, 123]]<|/det|>
+(This downloads the HGSVC2 dataset)  
+
+<|ref|>text<|/ref|><|det|>[[92, 136, 870, 180]]<|/det|>
+(It is now necessary to divide benchmark insertions from tandem duplications, instructions are reported in https://github.com/kensung- lab/survindel2_paper_experiments/blob/main/1_data_preparation/README.txt)  
+
+<|ref|>text<|/ref|><|det|>[[92, 193, 234, 207]]<|/det|>
+mkdir by- sample  
+
+<|ref|>text<|/ref|><|det|>[[92, 207, 890, 339]]<|/det|>
+bcftools query - l variants_freeze4_sv_insdel_alt.vcf.gz | while read sample ; do bcftools view variants_freeze4_sv_insdel_alt.vcf.gz - i "SVTYPE \(=\) "DEL" - s \(\) sample - - min- ac=1 - Oz - o by- sample\) \ \(sample.DEL.vcf.gz\) ; done bcftools query - l variants_freeze4_sv_insdel_alt.vcf.gz | while read sample ; do bcftools view variants_freeze4_sv_insdel_alt.DUP.vcf.gz - i "SVTYPE \(=\) 'DEL' " - s \(\) sample - - min- ac=1 - Oz - o by- sample\) \ \(sample.DUP.vcf.gz\) ; done bcftools query - l variants_freeze4_sv_insdel_alt.vcf.gz | while read sample ; do bcftools view variants_freeze4_sv_insdel_alt.vcf.gz - i "SVTYPE \(=\) 'DEL' " - s \(\) sample - - min- ac=1 - Oz - o by- sample\) \ \(sample.INS.vcf.gz\) ; done  
+
+<|ref|>text<|/ref|><|det|>[[92, 350, 870, 380]]<|/det|>
+(This, barring unforeseen failures, generates benchmark datasets for each individual, including 1_data_preparation/by- sample/HG00512.DEL.vcf.gz.).  
+
+<|ref|>text<|/ref|><|det|>[[92, 392, 848, 422]]<|/det|>
+We find this approach more transparent, as it makes clear how we obtained the benchmark datasets, and that we did not filter/manipulate them unfairly.  
+
+<|ref|>text<|/ref|><|det|>[[92, 435, 884, 465]]<|/det|>
+It should be noted that there are also instances in which obtaining some of the results requires significant computation time/resources. In those cases, we also provided pre- computed results.  
+
+<|ref|>text<|/ref|><|det|>[[92, 478, 835, 508]]<|/det|>
+2. For HG002 with hg19 reference calls, truvari failed with error "ERROR:root:No SAMPLE columns found in vcf", though it worked for MAnta and DRAGEN vcfs.  
+
+<|ref|>text<|/ref|><|det|>[[92, 521, 864, 551]]<|/det|>
+I think it would be great if authors could show that the performance of their tool is better than Manta/DRAGEN (and others) by running Truvari on the DEL variant types.  
+
+<|ref|>text<|/ref|><|det|>[[92, 564, 898, 618]]<|/det|>
+We have included instructions to https://github.com/kensung- lab/survindel2_paper_experiments/blob/main/hg002- additional/README.txt on how to use truvari to benchmark the HG002 on hg19 calls, on both the GIAB and the Sniffles2 benchmark.  
+
+<|ref|>text<|/ref|><|det|>[[92, 608, 900, 637]]<|/det|>
+We did not encounter the error reported, but truvari failed on the GIAB benchmark because it was missing the contigs lengths (we show how to work around it in the README).  
+
+<|ref|>text<|/ref|><|det|>[[93, 650, 260, 665]]<|/det|>
+Here are the results:  
+
+<|ref|>text<|/ref|><|det|>[[93, 678, 258, 693]]<|/det|>
+Sniffles benchmark:  
+
+<|ref|>text<|/ref|><|det|>[[93, 707, 135, 720]]<|/det|>
+DEL:  
+
+<|ref|>text<|/ref|><|det|>[[93, 735, 270, 763]]<|/det|>
+Manta recall 0.39 Manta precision 0.80  
+
+<|ref|>text<|/ref|><|det|>[[93, 777, 306, 806]]<|/det|>
+SurVIndel2 recall 0.51 SurVIndel2 precision 0.76  
+
+<|ref|>text<|/ref|><|det|>[[93, 820, 138, 834]]<|/det|>
+DUP:  
+
+<|ref|>text<|/ref|><|det|>[[93, 848, 275, 877]]<|/det|>
+Manta recall: 0.15 Manta precision: 0.90  
+
+<|ref|>text<|/ref|><|det|>[[93, 891, 305, 920]]<|/det|>
+SurVIndel2 recall 0.40 SurVIndel2 precision 0.89
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[93, 65, 241, 80]]<|/det|>
+GIAB benchmark:  
+
+<|ref|>text<|/ref|><|det|>[[93, 93, 134, 108]]<|/det|>
+DEL:  
+
+<|ref|>text<|/ref|><|det|>[[93, 122, 270, 152]]<|/det|>
+Manta recall 0.52  Manta precision 0.79  
+
+<|ref|>text<|/ref|><|det|>[[93, 165, 306, 195]]<|/det|>
+SurVIndel2 recall 0.73  SurVIndel2 precision 0.90  
+
+<|ref|>text<|/ref|><|det|>[[93, 208, 139, 223]]<|/det|>
+DUP:  
+
+<|ref|>text<|/ref|><|det|>[[93, 236, 270, 266]]<|/det|>
+Manta recall 0.22  Manta precision 0.86  
+
+<|ref|>text<|/ref|><|det|>[[93, 278, 306, 308]]<|/det|>
+SurVIndel2 recall 0.50  SurVIndel2 precision 0.88  
+
+<|ref|>text<|/ref|><|det|>[[93, 321, 899, 378]]<|/det|>
+There are a couple of points worth mentioning. We have previously explained why we disable the comparison of inserted sequences in duplications when using truvari: truvari assumes that a tandem duplications duplicates the reference sequence exactly once, whereas SurVClusterer does not have this assumption.  
+
+<|ref|>text<|/ref|><|det|>[[93, 378, 900, 409]]<|/det|>
+This way, the precisions for duplications become very high. We briefly checked the Manta tandem duplications reported as true positives by truvari on the Sniffles benchmark. Here are the first two:  
+
+<|ref|>text<|/ref|><|det|>[[92, 420, 900, 565]]<|/det|>
+chr1 1584516 MantaDUP:TANDEM:61:2:6:0:0:0 A <DUP:TANDEM> 289 PASS  END=1651190;SVTYPE=DUP;SVLEN=66674;IMPRECISE;CIPOS=- 578,579;CIEND=- 437,437;Pct SeqSimilarity=0;PctSizeSimilarity=0.0587;PctRecOverlap=0.0297;SizeDiff=- 62761;StartDistance= 26;EndDistance=- 66648;TruScore=2;Matchld=32.1.0 GT:FT:GQ:PL:PR 0/1:PASS:289:339,0,761:53,26 chr1 1588374 MantaDUP:TANDEM:52:0:1:1:0:0 C <DUP:TANDEM> 23 PASS  END=1653752;SVTYPE=DUP;SVLEN=65378;IMPRECISE;CIPOS=- 297,297;CIEND=- 679,680;Pct SeqSimilarity=0;PctSizeSimilarity=0.0008;PctRecOverlap=0.0008;SizeDiff=- 65327;StartDistance= 61112;EndDistance=- 4266;GTMatch;TruScore=0;Matchld=33.0.0 GT:FT:GQ:PL:PR 0/1:PASS:23:73,0,999:133,25  
+
+<|ref|>text<|/ref|><|det|>[[92, 577, 881, 618]]<|/det|>
+For the first, Sniffles2 reported two insertions near the first breakpoint, one of size 681 and the other 3,913. The tandem duplication of Manta could not have generated such insertion, since it reports a duplication of 66,674 bps.  
+
+<|ref|>text<|/ref|><|det|>[[93, 618, 880, 663]]<|/det|>
+SurVClusterer takes this into account: it allows the duplicated region to repeat multiple times, if necessary, in order to match the inserted sequence; however, if the duplicated region is larger than the inserted sequence, it will not match the two.  
+
+<|ref|>text<|/ref|><|det|>[[93, 676, 900, 707]]<|/det|>
+We are very puzzled by the second one, since Sniffles2 does not report any insertion close to either breakpoints. We do not know why Truvari would consider this duplication as a true positive.  
+
+<|ref|>text<|/ref|><|det|>[[92, 720, 889, 750]]<|/det|>
+Second, in the Sniffles benchmark, it seems that truvari reports a lower precision for both Manta and SurVIndel2 (SurVClusterer reports nearly 0.9).  
+
+<|ref|>text<|/ref|><|det|>[[93, 750, 874, 808]]<|/det|>
+When manually checking the SurVIndel2 false positives according to truvari, however, many of them are clearly supported by long reads. We have uploaded the IGV screenshots for all false positives to the GitHub repository. We are not sure why truvari reports many of them as false positives.  
+
+<|ref|>text<|/ref|><|det|>[[93, 821, 173, 836]]<|/det|>
+Sincerely,  
+
+<|ref|>text<|/ref|><|det|>[[93, 849, 395, 865]]<|/det|>
+Ramesh Rajaby and Wing- Kin Sung.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 89, 305, 104]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 119, 402, 134]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 148, 878, 239]]<|/det|>
+I would like to thank the authors for the detailed response to my earlier suggestions. I understand that the authors have put a README file with all the commands that were used in the validation. In the earlier suggestion, I mentioned that some of the files were not available on Github repo. I totally agree with the authors that some executions may need significant time/resources. Therefore, it would be great if authors make only these three files available on their GitHub repo (or send the link to these files if they are already available)  
+
+<|ref|>text<|/ref|><|det|>[[115, 253, 170, 266]]<|/det|>
+HG00512  
+
+<|ref|>text<|/ref|><|det|>[[115, 270, 420, 312]]<|/det|>
+1. HG00512.DEL.vcf.gz 
+2. HG00512.survindel2.ml.DEL.alt.vcf.gz  
+
+<|ref|>text<|/ref|><|det|>[[115, 327, 169, 340]]<|/det|>
+HG002  
+
+<|ref|>text<|/ref|><|det|>[[115, 356, 400, 370]]<|/det|>
+1. HG002.survindel2.ml.DEL.alt.vcf.gz  
+
+<|ref|>text<|/ref|><|det|>[[115, 400, 872, 506]]<|/det|>
+It is also very interesting to know that truvari assumes tandem duplications are duplicated in the reference sequence exactly once. So it means that any benchmarking for tandem duplications using truvari were not accurate. The authors of Truvari have recently released a tandem- repeat benchmark (https://www.nature.com/articles/s41587-024-02225- z) that also used Truvari for validation, I am curious if they also used the same concept i.e. duplicated once in reference. It would be worth checking with Truvari developers or reporting these issues as this tool has become a standard tool for any validations.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 64, 833, 80]]<|/det|>
+We thank the reviewers for the kind feedback. We will reply to their reviews point by point:  
+
+<|ref|>sub_title<|/ref|><|det|>[[93, 109, 192, 124]]<|/det|>
+## Reviewer 3  
+
+<|ref|>text<|/ref|><|det|>[[92, 138, 896, 226]]<|/det|>
+I would like to thank the authors for the detailed response to my earlier suggestions. I understand that the authors have put a README file with all the commands that were used in the validation. In the earlier suggestion, I mentioned that some of the files were not available on Github repo. I totally agree with the authors that some executions may need significant time/resources. Therefore, it would be great if authors make only these three files available on their GitHub repo (or send the link to these files if they are already available)  
+
+<|ref|>text<|/ref|><|det|>[[92, 238, 732, 254]]<|/det|>
+As suggested by the reviewer, we have uploaded the three requested files, in:  
+
+<|ref|>text<|/ref|><|det|>[[92, 266, 900, 310]]<|/det|>
+https://github.com/kensung- lab/survindel2 paper experiments/tree/main/1 data preparation/by- sample https://github.com/kensung- lab/survindel2 paper experiments/tree/main/truvari  
+
+<|ref|>text<|/ref|><|det|>[[92, 323, 884, 425]]<|/det|>
+It is also very interesting to know that truvari assumes tandem duplications are duplicated in the reference sequence exactly once. So it means that any benchmarking for tandem duplications using truvari were not accurate. The authors of Truvari have recently released a tandem- repeat benchmark (https://www.nature.com/articles/s41587- 024- 02225- z) that also used Truvari for validation, I am curious if they also used the same concept i.e. duplicated once in reference. It would be worth checking with Truvari developers or reporting these issues as this tool has become a standard tool for any validations.  
+
+<|ref|>text<|/ref|><|det|>[[92, 438, 653, 453]]<|/det|>
+Thank you for recommending the paper, which we were unaware of.  
+
+<|ref|>text<|/ref|><|det|>[[92, 453, 884, 481]]<|/det|>
+Upon reading it, it seems that Truvari now has a new comparison algorithm, called refine, which improves over the previous one in two ways:  
+
+<|ref|>text<|/ref|><|det|>[[92, 481, 896, 556]]<|/det|>
+1) A problem in comparing indels (whether small or large) in TR regions is that the indel can shift. For example, we can have two deletions, one deleting the first copy and one deleting the last copy of the same TR, and while they are disjoint, they should be recognized as producing the same haplotype. They improved the way in which this work. This problem was one of the reasons that pushed us to develop an in-house comparison algorithm.  
+
+<|ref|>text<|/ref|><|det|>[[92, 567, 888, 638]]<|/det|>
+2) An indel can not only shift, but also be "decomposed" (for lack of a better term) into multiple smaller indels. For example, if a region AA becomes AAAA, we can have a single insertion of AA, or two distinct insertions of A, and we are effectively representing the same indel. The authors have an algorithm that seems to be able to correctly compare indels under these conditions.  
+
+<|ref|>text<|/ref|><|det|>[[110, 638, 835, 654]]<|/det|>
+This is a difficult problem that does occur in practice (especially with noisy long reads).  
+
+<|ref|>text<|/ref|><|det|>[[92, 666, 891, 710]]<|/det|>
+However, none of these changes appear to improve the INS vs DUP comparison, which tends to be a problem mostly when comparing a long reads- derived benchmark dataset to a short read- derived called dataset.  
+
+<|ref|>text<|/ref|><|det|>[[92, 710, 903, 753]]<|/det|>
+Furthermore, it is not immediately clear to us how the refine command in Truvari directly translates to comparing sets of SVs. From its documentation (https://github.com/ACEnglish/truvari/wiki/ refine):  
+
+<|ref|>text<|/ref|><|det|>[[92, 767, 888, 812]]<|/det|>
+The regions spanned by subset.bed should be shorter and focused around the breakpoints of putative FNs/FPs. Haplotypes from these boundaries are fed into a realignment procedure which can take an extremely long time on e.g entire chromosomes.  
+
+<|ref|>text<|/ref|><|det|>[[92, 825, 903, 855]]<|/det|>
+However, for example, Manta calls several large deletions and duplications, often spanning almost a whole chromosome. Furthermore, they add:  
+
+<|ref|>text<|/ref|><|det|>[[92, 870, 568, 885]]<|/det|>
+Also, the genotypes within these regions must be phased.  
+
+<|ref|>text<|/ref|><|det|>[[92, 899, 611, 914]]<|/det|>
+However, this is not generally possible when using short reads.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 65, 860, 95]]<|/det|>
+We will try and get in touch with the authors of Truvari and see if we missed something and if there is indeed a way to extend the comparison algorithm of INS vs DUP.  
+
+<|ref|>text<|/ref|><|det|>[[92, 108, 173, 123]]<|/det|>
+Sincerely,  
+
+<|ref|>text<|/ref|><|det|>[[92, 136, 395, 152]]<|/det|>
+Ramesh Rajaby and Wing- Kin Sung.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 89, 306, 105]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 119, 404, 135]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 148, 833, 179]]<|/det|>
+I would like to thank the reviewers for providing the following VCF files for HG00512 and HG002 sample  
+
+<|ref|>text<|/ref|><|det|>[[115, 194, 420, 260]]<|/det|>
+HG00512 1. HG00512. DEL.vcf.gz 2. HG00512. survivdel2. ml.DEL.alt.vcf.gz  
+
+<|ref|>text<|/ref|><|det|>[[115, 269, 401, 313]]<|/det|>
+HG002 1. HG002. survivdel2. ml.DEL.alt.vcf.gz  
+
+<|ref|>text<|/ref|><|det|>[[115, 327, 880, 372]]<|/det|>
+The Truvari (v4.1) run on HG00512 (using the above two DEL vcf files) with the same parameters that the authors provided in the README file produced the following results. It would be great if the authors could compare this with their results and update them.  
+
+<|ref|>text<|/ref|><|det|>[[115, 386, 379, 430]]<|/det|>
+"precision": 0.8440798882173747, "recall": 0.610386228126738, "f1": 0.7084589093098503,  
+
+<|ref|>text<|/ref|><|det|>[[115, 445, 815, 475]]<|/det|>
+Regarding the comparison of DEL variants between SurVIndel2 (using the above DEL vcf) and DRAGEN based on Truvari (v4.1) produced the following results.  
+
+<|ref|>text<|/ref|><|det|>[[115, 490, 180, 504]]<|/det|>
+DRAGEN  
+
+<|ref|>text<|/ref|><|det|>[[115, 520, 368, 564]]<|/det|>
+"precision": 0.830099350472806, "recall": 0.6691725891079381, "f1": 0.7409993030606416,  
+
+<|ref|>text<|/ref|><|det|>[[115, 580, 201, 593]]<|/det|>
+SurVIndel2  
+
+<|ref|>text<|/ref|><|det|>[[115, 608, 370, 652]]<|/det|>
+"precision": 0.810636420815687, "recall": 0.6229756676916839, "f1": 0.7045235870445897,  
+
+<|ref|>text<|/ref|><|det|>[[115, 667, 830, 683]]<|/det|>
+So the Truvari- based analysis shows DRAGEN performs better than SurVIndel2 for DEL variants.  
+
+<|ref|>text<|/ref|><|det|>[[115, 698, 875, 758]]<|/det|>
+Although the authors have explained a few times the issues of Truvari based comparison and why the in- house evaluation tool is better than Truvari, I would still recommend providing all the results based on Truvari. Also, it is recommended to provide the valid reasons in the manuscript of using in- house tool against the standard evaluation tool.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 64, 794, 80]]<|/det|>
+We thank the reviewers for the feedback. We will reply to their reviews point by point:  
+
+<|ref|>sub_title<|/ref|><|det|>[[93, 109, 193, 124]]<|/det|>
+## Reviewer 3  
+
+<|ref|>text<|/ref|><|det|>[[92, 138, 885, 170]]<|/det|>
+I would like to thank the reviewers for providing the following VCF files for HG00512 and HG002 sample  
+
+<|ref|>text<|/ref|><|det|>[[93, 181, 430, 241]]<|/det|>
+HG00512 1. HG00512. DEL.vcf.gz 2. HG00512. survindel2. ml.DEL.alt.vcf.gz  
+
+<|ref|>text<|/ref|><|det|>[[93, 253, 430, 300]]<|/det|>
+HG002 1. HG002. survindel2. ml.DEL.alt.vcf.gz  
+
+<|ref|>text<|/ref|><|det|>[[93, 309, 884, 353]]<|/det|>
+The Truvari (v4.1) run on HG00512 (using the above two DEL vcf files) with the same parameters that the authors provided in the README file produced the following results. It would be great if the authors could compare this with their results and update them.  
+
+<|ref|>text<|/ref|><|det|>[[93, 366, 381, 409]]<|/det|>
+"precision": 0.8440798882173747, "recall": 0.610386228126738, "f1": 0.7084589093098503,  
+
+<|ref|>text<|/ref|><|det|>[[93, 422, 870, 452]]<|/det|>
+Regarding the comparison of DEL variants between SurVIndel2 (using the above DEL vcf) and DRAGEN based on Truvari (v4.1) produced the following results.  
+
+<|ref|>text<|/ref|><|det|>[[93, 465, 172, 479]]<|/det|>
+DRAGEN  
+
+<|ref|>text<|/ref|><|det|>[[93, 494, 370, 536]]<|/det|>
+"precision": 0.830099350472806, "recall": 0.6691725891079381, "f1": 0.7409993036060416,  
+
+<|ref|>text<|/ref|><|det|>[[93, 550, 185, 565]]<|/det|>
+SurVIndel2  
+
+<|ref|>text<|/ref|><|det|>[[93, 580, 370, 622]]<|/det|>
+"precision": 0.810636420815687, "recall": 0.6229756676916839, "f1": 0.7045235870445897,  
+
+<|ref|>text<|/ref|><|det|>[[92, 636, 888, 652]]<|/det|>
+So the Truvari- based analysis shows DRAGEN performs better than SurVIndel2 for DEL variants.  
+
+<|ref|>text<|/ref|><|det|>[[93, 664, 898, 724]]<|/det|>
+Although the authors have explained a few times the issues of Truvari based comparison and why the in- house evaluation tool is better than Truvari, I would still recommend providing all the results based on Truvari. Also, it is recommended to provide the valid reasons in the manuscript of using in- house tool against the standard evaluation tool.  
+
+<|ref|>text<|/ref|><|det|>[[93, 736, 884, 780]]<|/det|>
+As recommended, we modified our supplementary section "Comparing deletions and tandem duplications" to include why we use the in- house tool rather than Truvari. Furthermore, we have added two panels to Fig. S8, that compare Manta, DRAGEN and SurVIndel2 using Truvari.  
+
+<|ref|>text<|/ref|><|det|>[[92, 793, 896, 823]]<|/det|>
+We have also modified the scripts in the truvari subfolder to make them work to Truvari 4+. We also noticed that when comparing Manta and DRAGEN, Truvari would give the following warning:  
+
+<|ref|>text<|/ref|><|det|>[[92, 835, 740, 852]]<|/det|>
+[WARNING] Unresolved SVs (e.g. ALT=<DEL>) are filtered when `--pctseq != 0`  
+
+<|ref|>text<|/ref|><|det|>[[92, 864, 830, 880]]<|/det|>
+And many calls would be ignored. Therefore, we now compare deletions with -- pctseq 0.  
+
+<|ref|>text<|/ref|><|det|>[[92, 892, 861, 908]]<|/det|>
+Regarding the comparison between SurVIndel2 and DRAGEN, we get very different numbers.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 65, 890, 110]]<|/det|>
+We have uploaded the DRAGEN deletions we have obtained for HG00512 to https://github.com/kensung-lab/survindel2 paper experiments/tree/main/dragen (along with the script to obtain them, in README.txt).  
+
+<|ref|>sub_title<|/ref|><|det|>[[92, 122, 172, 138]]<|/det|>
+## DRAGEN  
+
+<|ref|>text<|/ref|><|det|>[[112, 151, 400, 195]]<|/det|>
+"precision": 0.8632424434580427, "recall": 0.44937403909510215, "f1": 0.5910618199468787,  
+
+<|ref|>sub_title<|/ref|><|det|>[[92, 208, 185, 223]]<|/det|>
+## SurVIndel2  
+
+<|ref|>text<|/ref|><|det|>[[112, 236, 400, 280]]<|/det|>
+"precision": 0.8249924173491052, "recall": 0.5734680430485394, "f1": 0.6766108881942217,  
+
+<|ref|>text<|/ref|><|det|>[[92, 293, 875, 323]]<|/det|>
+(https://github.com/kensung-lab/survindel2 paper experiments/blob/main/truvari/README.txt contains the command we used)  
+
+<|ref|>text<|/ref|><|det|>[[92, 335, 870, 380]]<|/det|>
+DRAGEN reports 4732 deletions for HG00512, while the benchmark contains 9106 deletions. Therefore, it is unlikely to achieve a sensitivity higher than 4732/9106 = 0.52, regardless of the comparison method.  
+
+<|ref|>text<|/ref|><|det|>[[92, 392, 888, 423]]<|/det|>
+In case we are using different datasets or we have made a mistake in downloading the DRAGEN results, please let us know.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 88, 320, 105]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 119, 404, 134]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 148, 864, 179]]<|/det|>
+I would like to thank the authors for their satisfactory responses. There are no other concerns about the excellent work the authors did in this paper.  
+
+<|ref|>text<|/ref|><|det|>[[115, 193, 870, 238]]<|/det|>
+I have a last suggestion for authors to cross check the HG002 DEL comparison between DRAGEN and SurVIndel2 using truvari and report the correct outputs in the final version. The following results that was mentioned in the last review could belong to HG002 dataset (not HG00512).  
+
+<|ref|>text<|/ref|><|det|>[[115, 253, 180, 266]]<|/det|>
+DRAGEN  
+
+<|ref|>text<|/ref|><|det|>[[115, 272, 369, 326]]<|/det|>
+"precision": 0.830099350472806, "recall": 0.6691725891079381, "f1": 0.7409993036060416,  
+
+<|ref|>text<|/ref|><|det|>[[115, 342, 200, 355]]<|/det|>
+SurVIndel2  
+
+<|ref|>text<|/ref|><|det|>[[115, 370, 369, 414]]<|/det|>
+"precision": 0.810636420815687, "recall": 0.6229756676916839, "f1": 0.7045235870445897,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 64, 794, 80]]<|/det|>
+We thank the reviewers for the feedback. We will reply to their reviews point by point:  
+
+<|ref|>sub_title<|/ref|><|det|>[[93, 109, 193, 124]]<|/det|>
+## Reviewer 3  
+
+<|ref|>text<|/ref|><|det|>[[92, 138, 860, 199]]<|/det|>
+I have a last suggestion for authors to cross check the HG002 DEL comparison between DRAGEN and SurVlndel2 using truvari and report the correct outputs in the final version. The following results that was mentioned in the last review could belong to HG002 dataset (not HG00512).  
+
+<|ref|>text<|/ref|><|det|>[[93, 199, 175, 211]]<|/det|>
+DRAGEN  
+
+<|ref|>text<|/ref|><|det|>[[93, 225, 370, 280]]<|/det|>
+"precision": 0.830099350472806, "recall": 0.6691725891079381, "f1": 0.7409993036060416, SurVlndel2  
+
+<|ref|>text<|/ref|><|det|>[[93, 295, 370, 339]]<|/det|>
+"precision": 0.810636420815687, "recall": 0.6229756676916839, "f1": 0.7045235870445897,  
+
+<|ref|>text<|/ref|><|det|>[[92, 352, 799, 368]]<|/det|>
+We could not replicate the numbers, using Truvari 4.1, even for SurVlndel2. We obtain  
+
+<|ref|>text<|/ref|><|det|>[[111, 380, 401, 424]]<|/det|>
+"precision": 0.7815208275090858, "recall": 0.5774434462604178, "f1": 0.6641588578474773,  
+
+<|ref|>text<|/ref|><|det|>[[92, 437, 875, 481]]<|/det|>
+While for the dataset provided by the reviewer for DRAGEN on HG002 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10802302/) we obtained the following numbers, again with Truvari 4.1  
+
+<|ref|>text<|/ref|><|det|>[[111, 494, 390, 538]]<|/det|>
+"precision": 0.800033892560583, "recall": 0.5116354583829419, "f1": 0.624129407381033,  
+
+<|ref|>text<|/ref|><|det|>[[92, 551, 870, 581]]<|/det|>
+It should be noted that we ran SurVlndel2 on a 50x HiSeq dataset, while the DRAGEN dataset was 35x NovaSeq 6000, so the comparison is not completely fair.  
+
+<|ref|>text<|/ref|><|det|>[[92, 593, 902, 695]]<|/det|>
+This brings us to the main reason why we think a comparison for HG002 would not be meaningful. A meaningful comparison requires, in our opinion, that the same input data is used. This was possible for the 34 samples in the HGSVC2 benchmark because the same set of reads, sequenced by the New York Genome Institute, were used to run SurVlndel2 and DRAGEN. However, for HG002, we selected a subset of Illumina reads (50x depth in total) from the GIAB project (which is 300x in total), and ran the different callers. We could not find a DRAGEN callset from the same set of reads.  
+
+<|ref|>text<|/ref|><|det|>[[92, 708, 900, 752]]<|/det|>
+To summarise, we would like not to include the comparison in the paper, because the comparison would not be not fair, as different coverage, sequencing platform, laboratories, etc. could influence the results. Furthermore, it would add little to the paper since  
+
+<|ref|>text<|/ref|><|det|>[[92, 765, 900, 865]]<|/det|>
+1) We have already compared SurVlndel2 to DRAGEN in 34 samples where a fair comparison is possible  
+2) DRAGEN is not a direct competitor of SurVlndel2, as it is a commercial product, while SurVlndel2 is open source and freely downloadable/usable  
+3) Even disregarding the above mentioned points, in the above mentioned test, SurVlndel2 still outperforms DRAGEN on HG002 (we would like to stress once again that it is not a completely fair comparison, and thus we do not think it should be reported in the paper)  
+
+<|ref|>text<|/ref|><|det|>[[93, 878, 390, 893]]<|/det|>
+Ramesh Rajaby and Wing- Kin Sung
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__72f5c47559c66648327d1e06e829ba3510af74f9ae1229b38b159158d672cb65/images_list.json

@@ -0,0 +1,55 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Caption: Black line represents the mirror Hall conductivity from all bands. Red line represents the contribution from the four gapless bands with \\(n = 1\\) . Blue line represents the contribution from the second-lowest energy gapped bands with \\(n = 2\\) .",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Fig. R2:(left) The bandstructure of Bi2Se3 along [001] direction with a twin boundary. The color of the dots represents the mean position of the states along the [001] direction.A gate voltage of \\(\\mathrm{V = 0.1eV}\\) is applied to the bottom surface. The remaining parameters are the same as in the main text. (right) The mirror Hall conductance \\(\\sigma_{xy}^{M2}\\) as a function of \\(\\mu\\) . The red dashed line serves as a visual guide to indicate the value of -1. The blue shadowed region indicates the gap regions for the bulk states.",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        100,
+        872,
+        360
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Fig. R1: The blue and red solid lines represent the line integral of the Berry connection over the Fermi surface \\(\\mu\\) for these two Hamiltonians. The dashed lines with triangles represent the corresponding Hall conductivities. \\(\\nu = 1, \\lambda = 1, b = 0.2\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        170,
+        460,
+        545,
+        686
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Fig. R1: (left) The mirror Hall conductivity \\(\\sigma_{xy}^{Mz}\\) as a function of \\(\\mu\\) for various gate voltages V applied to the layer slightly above the midpoint of the film. (right) The mirror Hall conductivity \\(\\sigma_{xy}^{Mz}\\) as a function of \\(\\mu\\) for gate voltage of V applied to the bottom surface. The remaining parameters are the same as in the main text. The gray dashed line serves as a visual guide to indicate the value of -1.",
+    "footnote": [],
+    "bbox": [
+      [
+        118,
+        81,
+        872,
+        456
+      ]
+    ],
+    "page_idx": 23
+  }
+]

peer_reviews/supplementary_0_Peer Review File__72f5c47559c66648327d1e06e829ba3510af74f9ae1229b38b159158d672cb65/supplementary_0_Peer Review File__72f5c47559c66648327d1e06e829ba3510af74f9ae1229b38b159158d672cb65.mmd

@@ -0,0 +1,426 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Half Quantum Mirror Hall Effect  
+
+![](images/Figure_unknown_0.jpg)
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+In this paper the authors present an account of generating a mirror Hall effect through surface states of a three dimensional strong topological insulator (TI). The underlying source is the half- quantized Hall conductivity supported by an isolated two dimensional Dirac point. At its core, utilizing the surface Dirac cones of an 3D TI to generate half- quantized Hall effect is not new, and many papers, both theoretical and experimental, have been written on this topic over the past decade (most of these papers are missing from the reference list). Since some form of magnetization is involved in these studies, the role of the mirror symmetry explored here is perhaps the only new aspect.  
+
+Beyond the generalities, I have the following specific questions:  
+
+1. The authors appear to be using a lattice version of the Hamiltonian introduced in Eq (1) of Nat Phys 5, 438 (2009) [at least for the concrete calculations reported in the 'Methods' section]. For this case, I do not see how the z-mirror operator globally commutes with the Hamiltonian, as mentioned in section 2 ("Results"). Therefore, it is unclear to me how one may obtain the direct-sum/block-diagonal form of the 3D Hamiltonian.  
+
+2. Usually surface states do not exist at all points in the surface Brillouin zone, as discussed, for example, by König et al. in Physical Society of Japan 77, 031007 (2008) [arxiv 0801.0901]. I do not see such constraint in Eq (2).  
+
+3. I am afraid I could not understand the reason behind the difference between results obtained by employing Eq (2) vs. Eq (3), and, consequently, the contrasting data presented in Fig 2(b).  
+
+4. In their calculations that lead to Eq (2), the supplemental materials seems to suggest they are using plane waves as a solution. In the absence of translation invariance along z-direction, why is this a good choice for eigenstates (it seems \(\backslash \mathrm{xi}\) is allowed to be complex; if so, it should be stated at the beginning)? Moreover, how does their method and its results compare with that by Creutz & Horvath in Phys. Rev. D 50, 2297 (1994) that was used for deriving surface states by König et al.? The form of the wavefunctions appears to be quite different at face value.  
+
+5. A few additional points  
+
+a) Subfigure labels are missing in Fig 1.
+
+<--- Page Split --->
+
+b) In Fig 2(a) the red circles are hardly visible, especially on a printout.  
+
+c) The subsection "Material candidates" could be placed at the end of section 2 in order to smoothen the narrative.  
+
+d) Much of the content of the section "Topological field theory for quantum mirror Hall effect on a lattice" in 'Methods' can be suppressed or moved to the Supplementary Materials, since the methodology is fairly well known [see for example, Phys. Rev. B 99, 235144 (2019) and references therein]  
+
+e) The notation for the 4x4 matrices used to in subsection "Mirror plane with time-reversal symmetry breaking" is not consistent — \tauau and \sigmaigma are interchanged below Eq (1).  
+
+f) What is meant by "magnetization" below Eq (1)?  
+
+g) Below Eq (2) the authors write "This is distinct from the conventional effective model for the surface states which is only ...". What do they mean?  
+
+h) The notations for position and momentum space coordinates should be made uniform — while bold r_|perp is used for the former, only bold k is used for the latter.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The manuscript by Fu et al presents a theory proposal of half quantum mirror Hall effect. The concept is quite novel and can be of interest. However, before I can recommend publication, I have the following concerns that the authors should address  
+
+1. Unclear presentation and lack of schematic illustrations: The presentation is in general not clear. There is significant disconnection between the general introduction, the mathematical derivation and the figures. Below I give some examples:  
+
+a. The authors emphasized the H_x, which seems to be a Hamiltonian with a given mirror eigenvalue in their math derivations. However, in their figures, it is not clear how to understand H_x at all. People typically think about a topological insulator with one Dirac cone at the top surface and another Dirac cone at the bottom surface. But I am guessing that the bands described by H_x does not come from a single surface. Instead, the bands described by H_x has contribution from both surface Dirac cones. Is that correct? Can the authors make some figure to label the bands of H_x and H_x?  
+
+b. Similarly, the schematic in Fig. 3a is confusing and lacks critical information. E.g., where is the mirror plane respect to the sample? I am guessing that the mirror plane is the horizontal x-y plane at the center of the sample, is that correct?
+
+<--- Page Split --->
+
+c. The authors constantly make analogy to the spin Hall effect. In the spin Hall effect, there is a very nice schematic, where as a current flows along \(x\) , the electrons with opposite spins bent toward the \(+y\) and \(-y\) directions, respectively. In this way, there is no net electrical Hall voltage but there is a spin voltage along the \(y\) direction. Can the authors make a similar plot to help people understand the mirror Hall effect they propose?  
+
+d. It is further confusing why the mirror Hall effect (or the spin Hall effect) can be directly measured by the two-probe conductance. The authors made some description in the text that is very difficult for readers to understand: "The charge current contains two parts \(\mathrm{Jc,y = Jssc,y + JmHc,y}\) : the first term Jss \(\mathrm{c,y}\) comes from the conducting surface states and the second term \(\mathrm{JmHc,y}\propto \tan 20\mathrm{m}\) arises from the spatial accumulation of the mirror polarization density and is an effect due to the existence of the mirror Hall effect. This mirror Hall mediated charge transport can be understood as follows: the electric field first induces a mirror charge accumulation on the boundary via the mirror Hall effect and then is converted into the charge current along the electric field via the inverse mirror Hall effect". Some graphics will help a lot  
+
+2. What is the role of nontrivial topology here? The authors constantly make analogy to the spin Hall effect. However, spin Hall effect is widely studied in heavy metals such as Ta, W. However, in the present work, the authors consider a topological insulator and seems to be claiming some quantized response. It is unclear what kind of transport is quantized? E.g., even in quantum spin Hall, the edge conductance is famously non-quantized due to inelastic scattering. No discussion of that kind is provided. What if I have a metal that has M_z mirror symmetry? Do I have mirror Hall effect (just not quantized)?  
+
+In general, I think the paper may contain some interesting concept that deserves to be published in a top journal, however, the current manuscript is written in a way that is very hard for people to judge. The authors also did not show great care and desire to make the presentation readable and understandable. I cannot recommend the publication in the present form but is happy to consider a significantly improved version.  
+
+Reviewer #3 (Remarks to the Author):  
+
+In this paper, the authors propose a novel topological phenomenon dubbed the half- quantum mirror Hall effect. They consider a strong topological insulator film with mirror reflection symmetry. Using mirror symmetry, they split a pair of surface Dirac modes on the top and bottom surfaces into a single one and derive a half- quantized Hall conductance on each mirror subsector. Furthermore, they also reveal that the half- quantized mirror Hall conductance results in a net current due to the accumulation of a mirror charge and the inverse mirror Hall effect. The idea of this paper looks novel and exciting. However, the following points should be clarified before the decision on the recommendation.
+
+<--- Page Split --->
+
+1. Whereas the authors assume an exact mirror-reflection symmetry, actual materials do not support it due to the imperfection of the crystal structures. The author should clarify how the imperfection affects the obtained result.  
+
+2. It needs to be clarified that the half-quantized value of the Hall conductance can be concluded through the two-terminal measurement. I understand that a non-zero mirror Hall conductance can be observed, but I do not know how to determine its value.  
+
+3. The superscript and subscript m should be modified as Mz to avoid confusion with the band index.
+
+<--- Page Split --->
+
+## Reply to the referees' reports  
+
+Reviewer #1 (Remarks to the Author):  
+
+Remark A1: In this paper the authors present an account of generating a mirror Hall effect through surface states of a three dimensional strong topological insulator (TI). The underlying source is the half- quantized Hall conductivity supported by an isolated two dimensional Dirac point. At its core, utilizing the surface Dirac cones of an 3D TI to generate half- quantized Hall effect is not new, and many papers, both theoretical and experimental, have been written on this topic over the past decade (most of these papers are missing from the reference list). Since some form of magnetization is involved in these studies, the role of the mirror symmetry explored here is perhaps the only new aspect.  
+
+Reply A1: We appreciate the referee's comments and would like to clarify the novelty of our study. While it is true that the concept of using the surface Dirac cones in a three- dimensional topological insulator to produce a half- quantized Hall effect has been explored in the literatures, it is important to note that previous works on this topic have primarily focused on the gapped surface states. In contrast, the half- quantized Hall conductivity in our study is attributed to the two- dimensional gapless Dirac cone, which is a new and distinct contribution. We have discussed this in detail in Ref. [26] (PRB 107, 125153 (2023)), where we highlight that the result that is derived from the massive Dirac fermions in the continuous model does not hold for a single gapped surface Dirac cone on a lattice model as the first Brillouin zone is finite according to the TKNN theorem. Therefore, the half- quantized Hall conductance of the gapless Dirac cone is a new and significant contribution to the field. Furthermore, we would like to emphasize that in our study, the strong topological insulator film does not break the time- reversal symmetry. Instead, the additional mirror symmetry plays a decisive role in the formation of this new effect. We believe that this represents an important advancement in our understanding of the role of symmetry in the generation of novel electronic phenomena.  
+
+Beyond the generalities, I have the following specific questions:  
+
+Remark A2: 1. The authors appear to be using a lattice version of the Hamiltonian introduced in Eq (1) of Nat Phys 5, 438 (2009) [at least for the concrete calculations reported in the 'Methods' section]. For this case, I do not see how the z- mirror operator globally commutes with the Hamiltonian, as mentioned in section 2 ("Results"). Therefore, it is unclear to me how one may obtain the direct- sum/block- diagonal form of the 3D Hamiltonian.  
+
+Reply A2: Yes, the model we used is a lattice version of Eq. (1) in Nat. Phys. (Ref. 39). The model has the mirror symmetry if the system is symmetric with open boundary condition along the z- axis. We have carefully reviewed the manuscript and made the necessary modifications to present a clearer explanation. We have added a paragraph where we explicitly include the z- mirror symmetry operator and explain how it commutes with the lattice Hamiltonian. This addition will address the referee's question and provide a more detailed understanding of the relationship between the mirror symmetry operator and the Hamiltonian. Thank you for bringing this issue to our attention.  
+
+Remark A3: 2. Usually surface states do not exist at all points in the surface Brillouin zone, as
+
+<--- Page Split --->
+
+discussed, for example, by König et al. in Physical Society of Japan 77, 031007 (2008) [arxiv 0801.0901]. I don not see such constraint in Eq (2).  
+
+Reply A3: The reviewer is right, the surface states only exist in a small portion of the first Brillouin zone, and its energy dispersion is located within between the bulk gap of three- dimensional topological insulator. Beyond the regime, the states evolve into the bulk one. The two parts consist of a complete band in the first Brillouin zone. In Eq. (8) [Originally Eq. (2)], for \(\Theta (x) = 0\) , the mass term \(\Delta (k) = 0\) , which represents the surface states while for \(\Theta (x) = 1\) , \(\Delta (k) = m_0(k)\) , which represent the bulk states. The dispersion of Eq. (2) forms a complete Dirac cone, which is defined in the first Brillouin zone. The surface electrons are only part of the Dirac cone near the Dirac point. As shown in "Methods" section "Derivation of Gapless Dirac Cone Model with Parity Symmetry Breaking", Eq. (8) is derived based on the tight- binding model (not \(k^*\) p model) which is applicable across the entire 2D Brillouin zone and not restricted to the vicinity of the Gamma point. The basis function in Eqs. (23) and (24) which are mirror eigenfunctions correspond to surface states for small values of \(k\) and naturally evolve into bulks state as \(k\) approaches the corners of the Brillouin zone as shown in Fig. S2. This is the key difference between our model (Eq. (8)) and the conventional effective model for the surface states in "Journal of Physical Society of Japan 77, 031007 (2008)".  
+
+Remark A4: 3. I am afraid I could not understand the reason behind the difference between results obtained by employing Eq (2) vs. Eq (3), and, consequently, the contrasting data presented in Fig 2(b).  
+
+Reply A4: Eq. (8) [Originally Eq. (2)] only describes the band of the gapless Dirac cone, which contributes one half Hall conductance. Eq.(9) [Originally Eq. (3)] is a formula for the mirror Hall conductivity for all the bands of the thin film as shown in Fig.2b[Originally Fig. 2a], which contains the bands from \(n = 1\) to \(Lz\) in Eq. (26) [Originally Eq. (14)]. Eq. (8) [Originally Eq. (2)] is only the bands of \(n = 1\) , i.e., the gapless Dirac cones labelled by the green line in Fig. 2b[Originally Fig. 2a]. The black line with squares in Fig. 2c[Originally Fig. 2b] is for the total mirror Hall conductivity for all the bands, and the green dashed line is just for the band for the gapless Dirac cones. Apart from the four gapless bands with \(n = 1\) , all other bands are gapped and topologically trivial (when chemical potential lies within their gaps, the mirror Hall conductivity is zero), however they also contribute when chemical potential shifts into the bulk states as shown by the figure below. The contributions from \(n = 3,4,\ldots ,Lz\) bands, not displayed in the figure below, should also be summed to produce the final results indicated by the black line with squares. That's why the two results are different.
+
+<--- Page Split --->
+![](images/Figure_unknown_1.jpg)
+
+<center>Caption: Black line represents the mirror Hall conductivity from all bands. Red line represents the contribution from the four gapless bands with \(n = 1\) . Blue line represents the contribution from the second-lowest energy gapped bands with \(n = 2\) . </center>  
+
+Remark A5: 4. In their calculations that lead to Eq (2), the supplemental materials seems to suggest they are using plane waves as a solution. In the absence of translation invariance along z- direction, why is this a good choice for eigenstates (it seems \(\vert x\vert\) is allowed to be complex; if so, it should be stated at the beginning)? Moreover, how does their method and its results compare with that by Creutz & Horvath in Phys. Rev. D 50, 2297 (1994) that was used for deriving surface states by König et al.? The form of the wavefunctions appears to be quite different at face value.  
+
+Reply A5: We take the periodic boundary condition along the \(x\) and \(y\) direction, and open boundary condition in the \(z\) direction. The plane wave is used only for the \(x - y\) plane. The effective mass in the 1- D Hamiltonian is a function of \(k\) . There is no translational symmetry, and H_{1d} in Eq. (21) (Originally Eq.(9)) is a one- dimensional lattice open chain. The eigen wave function is a function of the lattice site \(L_{z}\) , not a plane wave as shown in Eq. (23) and Eq. (24). We clarify this point in the revised version.  
+
+We read Creutz & Horvath's paper very carefully. We are surprised they have intended to solve this problem long before the birth of topological insulator. From Eq. 15 in PRD paper, it stated clearly that they tried to find the surface states of 1+ D- dimensional film, (D=1 and 3). We noted that in Creutz and Horvath's paper: 1) They obtained the dispersion of the boundary states of 1+1 and 1+ 3 dimension film, that is the one- dimensional chiral edge state and three- dimensional gapless Dirac cone which are eigenvectors of chiral operator \(\gamma_{5}\) ; 2) they noted the high energy part which violates the chiral symmetry. 3) they presented the numerical dispersion of the lattice model. However, 4) They didn't establish a set of equations to determine the variable \(\backslash \mathrm{lambda}\) and the energy eigenvalues, and failed to present the general solutions of the wave functions. Most important, due to the double degeneracy of the surface states, a pair of gapless Dirac fermions are usually regarded to cancel the chiral anomaly with each other as the two authors pointed out in the introduction.  
+
+In the present work, we focus on the two- dimensional gapless Dirac fermions reduced from a three- dimensional system. We established a complete set of equations to determine the
+
+<--- Page Split --->
+
+energy dispersion and \lambda lambda as a function of a finite thickness Lz. By means of the additional mirror symmetry, we classify the energy eigenstates into two separated classes of the gapless Dirac cones with even and odd parity [Eq. (23) and Eq. (24)]. The complete dispersions of the two gapless Dirac cone are presented, each gapless Dirac cone with parity anomaly.  
+
+We summarize the key differences between two works:  
+
+1) Our model possesses time reversal (TR) symmetry thus belongs to class All which is characterized by a \(Z_{2}\) topological invariant and helical edge states. In contrast, the model employed in PRD lacks TR symmetry and belongs to class A, which is characterized by a Chern number and chiral edge states. Consequently, the quantum anomalies addressed in the two works are fundamentally different: PRD deals with chiral anomaly, whereas our work concentrates on parity anomaly.  
+
+2) The analytical study in the Appendix of PRD only focuses on the solution of the chiral edge states which is restricted in a small portion of the Brillion zone. In contrast, our work analytically provides the all the eigenfunctions across the entire Brillion zone for the film geometry by utilizing the mirror symmetry.  
+
+3) PRD points out that the double degeneracy of surface states on two surfaces cancels the chiral anomaly. However, in our work, we find the inclusion of additional mirror symmetry reintroduces the parity anomaly which can have physical consequences.  
+
+We want to emphasize that the study of mirror symmetry not only facilitates an analytic solution but also introduces new physics in this three-dimensional topological insulator film.  
+
+## Remark A6:  
+
+5. A few additional points a) Subfigure labels are missing in Fig 1. Reply A6: we replotted and labelled of the subfigures in Fig. 1  
+
+b) In Fig 2(a) the red circles are hardly visible, especially on a printout.  
+
+Reply A6: we replotted Fig. 2.  
+
+c) The subsection "Material candidates" could be placed at the end of section 2 in order to smoothen the narrative.  
+
+Reply A6: we accept the suggestion to place the subsection in section 3 "Discussion and conclusion".  
+
+d) Much of the content of the section "Topological field theory for quantum mirror Hall effect on a lattice" in "Methods' can be suppressed or moved to the Supplementary Materials, since the methodology is fairly well known [see for example, Phys. Rev. B 99, 235144 (2019) and references therein]  
+
+Reply A: We accept the suggestion to suppress this section in the revised manuscript. We leave this section in the "Method" to emphasize how the topological field theory is modified when time reversal symmetry and an additional mirror symmetry are present.  
+
+e) The notation for the 4x4 matrices used to in subsection "Mirror plane with time-reversal symmetry breaking" is not consistent — \tauau and \sigmaigma are interchanged below Eq (1).
+
+<--- Page Split --->
+
+Reply A: We have updated the notation for the \(4 \times 4\) matrices throughout the manuscript for consistency.  
+
+f) What is meant by "magnetization" below Eq (1)?  
+
+Reply A: In the Hamiltonian \(H_{x}\) , a symmetry broken term appears near the mirror plane, which can be cancelled with the term in \(H_{- x}\) . The word "magnetization" may not be appropriate, it contains two symmetry broken fields.  
+
+g) Below Eq (2) the authors write "This is distinct from the conventional effective model for the surface states which is only ...". What do they mean?  
+
+Reply A: The conventional effective model for the surface states is only lower energy part of the model in Eq. (8) (Eq. (2) in the original version). It does not contain the symmetry broken part for higher energy. We have added a new section titled "Gapless Dirac Cones with Parity Symmetry Breaking" which includes an expanded discussion and a schematic diagram to clarify the distinctions between the conventional effective model for surface states and our theory.  
+
+h) The notations for position and momentum space coordinates should be made uniform — while bold \(r\) \perp perp is used for the former, only bold \(k\) is used for the latter.  
+
+Reply A: We used the unified notation as the reviewer suggested.  
+
+Reviewer #2 (Remarks to the Author):  
+
+Remark B1: The manuscript by Fu et al presents a theory proposal of half quantum mirror Hall effect. The concept is quite novel and can be of interest. However, before I can recommend publication, I have the following concerns that the authors should address  
+
+Reply B1: We would like to thank the referee for the positive comments.  
+
+Remark B2: 1. Unclear presentation and lack of schematic illustrations: The presentation is in general not clear. There is significant disconnection between the general introduction, the mathematical derivation and the figures. Below I give some examples:  
+
+a. The authors emphasized the \(H_{-x}\) , which seems to be a Hamiltonian with a given mirror eigenvalue in their math derivations. However, in their figures, it is not clear how to understand \(H_{-x}\) at all. People typically think about a topological insulator with one Dirac cone at the top surface and another Dirac cone at the bottom surface. But I am guessing that the bands described by \(H_{-x}\) does not come from a single surface. Instead, the bands described by \(H_{-x}\) has contribution from both surface Dirac cones. Is that correct? Can the authors make some figure to label the bands of \(H_{-x}\) and \(H_{-x}\) ?  
+
+Reply B2: We replotted Fig. 1 to illustrate how the Hamiltonian with the mirror symmetry decomposes into two separate sub- Hamiltonians, in which the states possess even and odd parity, respectively. To be clearer, we add the schematic of the states with even and odd parity and the corresponding wave function distribution. Then we further illustrate how the sub- Hamiltonian with a thickness \(Lz\) can be equivalently folded into a Hamiltonian with a thickness \(Lz / 2\) by utilizing the projected symmetric field operators, which is equivalent to a semimagnetic topological insulator film with a magnetic field only in one side. Hopefully the new figure is clearer.
+
+<--- Page Split --->
+
+Remark B3: b. Similarly, the schematic in Fig. 3a is confusing and lacks critical information. E.g., where is the mirror plane respect to the sample? I am guessing that the mirror plane is the horizontal \(x - y\) plane at the center of the sample, is that correct?  
+
+c. The authors constantly make analogy to the spin Hall effect. In the spin Hall effect, there is a very nice schematic, where as a current flows along \(x\) , the electrons with opposite spins bent toward the \(+y\) and \(-y\) directions, respectively. In this way, there is no net electrical Hall voltage but there is a spin voltage along the \(y\) direction. Can the authors make a similar plot to help people understand the mirror Hall effect they propose?  
+
+Reply B3: Here we response b and c together. We accept the reviewer's suggestion to replot Figure 3. We insert Fig. 3a to illustrate the mirror Hall effect induced by the electric current. The mirror plane is normal to the \(z\) axis and located in the middle of the film. An electric current can induce the charge carriers with even parity deflects to one side and the charge carriers with odd parity deflects to the opposite side. There is not net Hall current, but there is a "mirror parity" voltage, which is very similar to the spin Hall effect.  
+
+Remark B4: d. It is further confusing why the mirror Hall effect (or the spin Hall effect) can be directly measured by the two- probe conductance. The authors made some description in the text that is very difficult for readers to understand: "The charge current contains two parts \(Jc, y = Jss c, y + JmH c, y\) : the first term \(Jss c, y\) comes from the conducting surface states and the second term \(JmH c, y \propto \tan 2\Phi m\) arises from the spatial accumulation of the mirror polarization density and is an effect due to the existence of the mirror Hall effect. This mirror Hall mediated charge transport can be understood as follows: the electric field first induces a mirror charge accumulation on the boundary via the mirror Hall effect and then is converted into the charge current along the electric field via the inverse mirror Hall effect". Some graphics will help a lot Reply B: First if the anomalous Hall effect exists, the two- probe conductance includes a correction part attributed by the Hall conductance as well as the longitudinal conductance, \(\sigma_{two} = \sigma_{xx} + \frac{|\sigma_{xy}|^2}{\sigma_{yy}}\) if \(j_y = 0\) and the longitudinal conductance is nonzero by means of the relation between the electric current and electric field. In quantum anomalous Hall effect and quantum spin Hall effect, they are insulating phases, which longitudinal conductance is zero. However, it is a two- probe conductance is \(e^2 /h\) for quantum anomalous Hall effect, and \(2e^2 /h\) for quantum spin Hall effect. We add more detailed discussion in the main context.  
+
+Remark B4 2. What is the role of nontrivial topology here? The authors constantly make analogy to the spin Hall effect. However, spin Hall effect is widely studied in heavy metals such as Ta, W. However, in the present work, the authors consider a topological insulator and seems to be claiming some quantized response. It is unclear what kind of transport is quantized? E.g., even in quantum spin Hall, the edge conductance is famously non- quantized due to inelastic scattering. No discussion of that kind is provided. What if I have a metal that has \(M_{-}z\) mirror symmetry? Do I have mirror Hall effect (just not quantized)?  
+
+Reply B4: The half quantum mirror Hall effect is similar to the spin Hall effect as both of them are metallic phases, but its mirror Hall conductance is quantized, only one half of the quantum spin Hall conductance. The role of nontrivial topology reveals from the half quantized Hall conductance of a single gapless Dirac cone on a lattice. This is opposite to the graphene which hosts a pair of the gapless Dirac cone (if we ignore the spin degree of freedom). Existence of a single gapless Dirac cone on a lattice has to break the time reversal symmetry as required
+
+<--- Page Split --->
+
+by the fermion doubling theorem. As shown in Fig. 2(c), when chemical potential is situated within the bulk band gap \((|\mu |< m_0)\) , intersecting solely with the gapless Dirac cone, the mirror Hall conductivity is quantized at \(\sigma_{xy}^{M_z} = e^2 /h\) . When chemical potential shifts into the bulk states, the mirror Hall effect persists, though it derives from the quantized value due to the contributions from partially filled trivial gapped band. For our spin 1/2 system, the horizontal mirror operator \(\hat{M}_z\) anticommutes with the vertical mirror operator (or parity operator in 2D) \(\hat{M}_x\) , i.e., \(\{\hat{M}_z, \hat{M}_x\} = 0\) . Consequently, each mirror eigensector of \(\hat{M}_z\) violates the parity symmetry \(\hat{M}_x\) explicitly, leading to nonzero Hall effect for each sector, which in turn results in a nonzero mirror Hall effect. The relation \(\{\hat{M}_z, \hat{M}_x\} = 0\) serves as a sufficient but not necessary condition for the existence of a nonzero mirror Hall effect in a conventional metal that has Mz symmetry. As for a mirror symmetric metal, it is possible to have a mirror Hall effect based on the picture of Fig. 1 for the topologically trivial case of the lattice model we use. We are grateful that the reviewer attracted us to this point.  
+
+Remark B5 In general, I think the paper may contain some interesting concept that deserves to be published in a top journal, however, the current manuscript is written in a way that is very hard for people to judge. The authors also did not show great care and desire to make the presentation readable and understandable. I cannot recommend the publication in the present form but is happy to consider a significantly improved version.  
+
+Reply B5: Thank you for taking the time to review our manuscript and for providing valuable feedback. We appreciate your effort in assessing our work and are grateful for your constructive comments. We have carefully considered your suggestions and have made significant improvements to address the issues you raised. Specifically, we have focused on enhancing the clarity and readability of our paper. We have taken your feedback regarding the presentation of our figures into account and have replotted them to ensure better visual representation of our results. Furthermore, we have expanded the discussions in our manuscript to provide a more comprehensive analysis of our findings. We believe these additions have significantly enhanced the overall quality of our work.  
+
+Reviewer #3 (Remarks to the Author):  
+
+Remark C1: In this paper, the authors propose a novel topological phenomenon dubbed the half- quantum mirror Hall effect. They consider a strong topological insulator film with mirror reflection symmetry. Using mirror symmetry, they split a pair of surface Dirac modes on the top and bottom surfaces into a single one and derive a half- quantized Hall conductance on each mirror subsector. Furthermore, they also reveal that the half- quantized mirror Hall conductance results in a net current due to the accumulation of a mirror charge and the inverse mirror Hall effect. The idea of this paper looks novel and exciting. However, the following points should be clarified before the decision on the recommendation.  
+
+Reply C1: We would like to thank the referee for the positive comments and constructive suggestions.  
+
+Remark C2: 1. Whereas the authors assume an exact mirror- reflection symmetry, actual materials do not support it due to the imperfection of the crystal structures. The author should clarify how the imperfection affects the obtained result.
+
+<--- Page Split --->
+
+Reply C2: We appreciate the referee's insightful question regarding the imperfection of the crystal structure. Indeed, in real materials, exact mirror- reflection symmetry may not be achievable due to imperfections in the lattice structure. However, we believe that if the imperfection is not significantly strong, we can treat it as a perturbation that breaks the symmetry. This perturbation introduces inter- mirror scattering, and the scattering length described in Eqs. (11) and (12) (originally Eqs. (4) and (5)) reflects the strength of the imperfection. Although the exact mirror- reflection symmetry assumption may not hold in the presence of strong imperfections, we argue that for weak imperfections, the effect can still be observed. To address this concern, we have added a dedicated discussion in the revised manuscript (section "Discussion and conclusion") to further elaborate on the impact of lattice structure imperfections and the robustness of our findings. We believe that this clarification will provide a more comprehensive understanding of the phenomenon and its relevance to real materials.  
+
+Remark C3: 2. It needs to be clarified that the half- quantized value of the Hall conductance can be concluded through the two- terminal measurement. I understand that a non- zero mirror Hall conductance can be observed, but I do not know how to determine its value.  
+
+Reply C3: We appreciate the referee's question regarding the determination of the value of the mirror Hall conductance. If the exact mirror- reflection symmetry nearly holds (i.e., \(l_{MZ} \to \infty\) ), to deduce the value of the mirror Hall conductance, we propose a two- step measurement approach. In the first step, we suggest performing a conventional six- probe measurement to measure the longitudinal conductance of the system. This measurement provides valuable information about the overall conductivity of the material. In the second step, we propose a two- probe measurement to determine the mirror Hall conductance. By comparing the two- probe conductance with the longitudinal conductance obtained from the first step, we can calculate the difference. This difference corresponds to the mirror Hall conductance, allowing us to determine its value. We acknowledge that this two- step measurement approach is necessary to accurately determine the mirror Hall conductance. If the mirror- reflection symmetry is weakly broken (with \(l_{MZ}\) being finite), we propose utilizing a multi- terminal measurement to extract the mirror Hall conductance of the system. We have included a detailed explanation of a measurement procedure in the revised manuscript to provide further clarity on how the value of the mirror Hall conductance can be determined experimentally.  
+
+Remark C4: 3. The superscript and subscript m should be modified as Mz to avoid confusion with the band index.  
+
+Reply C4: To address this concern, we have carefully reviewed the manuscript and made the necessary modifications to clarify the notation. We have replaced the superscript and subscript "m" with "M_z" to differentiate it from the band index notation. We apologize for any confusion that may have arisen due to the previous notation, and we appreciate the referee's feedback in improving the clarity of our manuscript. Thank you for bringing this issue to our attention, and we hope that these revisions will address your concerns.
+
+<--- Page Split --->
+
+Reviewers' comments:  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors have responded to all points raised in my report. Their response has clarified several key aspects of the paper; in particular, the interplay between mirror symmetry and surface states. While they have identified an interesting role of mirror symmetry in the topological response from surface states, I remain unconvinced by the extent of progress reported in this paper, given the large body of literature, including recent works by (at least) a subset of the authors, that have already explored ideas that overlap with those explored here. In my view, the key new result here is the generalization of the notion of the mirror- Chern number (generally used as a bulk invariant) to half- quantized Hall conductivity from surface states — individually, the notions of mirror- Chern number and half- quantized Hall response from isolated 2D Weyl/Dirac cones are not new. Further, the fact that the conclusions rest on the analyses of a 4- band model is a concern, given the maturity of the field. A minimal- model based calculation, while important and the essential first step (the emphasis on obtaining analytical results is appreciated), does not unambiguously establish the authors' claim that this work represents "an important advancement in our understanding of the role of symmetry in the generation of novel electronic phenomena". I would urge the authors to numerically demonstrate their key results in more realistic models [eg. the Hamiltonian in Eq. (52) of Phys. Rev. B 82, 045122 (2010)] or ab initio data. While the former type of models are important for comparing with experimental data in topological insulators, the latter is the best demonstration of any principle borne out of minimal models.  
+
+Some specific comments following the authors' response:  
+
+1. I do not yet see why the authors emphasize the lattice-based calculations for surface states. On mathematical grounds, the fact that generally surface states exist over a finite sub-region of the surface Brillouin zone can be demonstrated within a k.p model [see for example, section V of Phys. Rev. B 82, 045122 (2010)]. On physical grounds, the half-quantized response should be accessible to a k.p model because the anomalous Hall response for a 2D metal boils down to a line-integral over the Fermi surface. Beyond concerns over quantitative agreements, the only other physically interesting reason for favoring lattice-based models appears to be tied to the discussion around the new Fig. 2 (d). This aspect has remained unclear to me, even after multiple re-reading of the text.  
+
+2. In their response the authors agree with the finite extent of the surface states, but this is not reflected by the Hamiltonian in Eq (8). If the states do not exist beyond a particular region of the surface Brillouin zone, why does the Hamiltonian exist? Moreover, terms proportional to \(\S \backslash \mathrm{tilde} \backslash \mathrm{sigma} \_ z \S\) [according to the convention of Eq (8)] appears due to the so-called hexagonal warping, and is an important feature of surface states of topological insulators, as discussed in Phys. Rev. Lett. 103, 266801 (2009). In light of the above two points, I would conclude Eq (8) is incorrect, which casts doubt over any analysis following Eq (8).
+
+<--- Page Split --->
+
+3. Regarding "Reply A4": What I understood from this response is that there are bulk states (states that are not exponentially localized on the two surfaces of the slab) that contribute to Hall conductivity if the overall filling is sufficiently large. While I agree with this statement, I find this distinction can be better utilized in making a physically important statement regarding the thickness dependence of the mirror-Hall response. Otherwise, this is a secondary point. Moreover, the language used to describe additional states contributing to the Hall response needs to be sharpened (also see my concern in point 2 above while interpreting the numerical results in this context).  
+
+4. While I appreciate the authors' contrasting their results with that obtained by Creutz & Horvath (CH), I would like to emphasize that the CH method itself is more general than its initial implementation. This is clearly demonstrated in the paper by König et al. noted in my previous report. Therefore, I would suggest taking both papers under consideration while contrasting between the two methods above the section "Quantum mirror Hall conductance".  
+
+5. Overall, I still find the narrative is unnecessarily convoluted. Even after the substantial revision of the text in the present version, the key ideas developed here are obfuscated by a lot of details that appear to be secondary. Is it impossible to demonstrate the main message with an 4-band k.p model?  
+
+If so, then this is an important aspect of this work, and should be demonstrated more directly.  
+
+If not, then the lattice-based calculations add quantitative details, and only the final outcome is relevant to the main text. Moreover, in this case, it is unclear whether a more realistic 8-band k.p model is less useful than using a lattice-version of a 4-band model.  
+
+6. While I have not noted this in my previous report, I find the word "discovery" in the abstract to be too strong for a theoretical proposal, whose applicability to actual materials is yet to be demonstrated.
+
+<--- Page Split --->
+
+Regarding C2: This is an important point, and underscores my suggestion above on working with ab initio data to unambiguously demonstrate the existence of the mirror Hall effect in actual materials. The authors have responded to this point qualitatively. My concern here is that if the mirror symmetry is broken (weakly or strongly), then, in principle, the two Dirac cones can hybridize and mutually gap out. This will revert the system to the very regime the authors want to avoid — topological response from gapped surface states (see "Reply A1").  
+
+Regarding C3: This is also an important issue, since measuring lattice- symmetry protected topological response is not obvious. I am unaware of any existing work where a lattice- symmetry protected transport response has been measured. The authors seem to recognize this challenge, and have made an effort to device a measurement protocol, but this may require/invite further scrutiny. Therefore, it would be helpful to cite other similar proposals (i.e. measurement of lattice- symmetry protected response) in this context.  
+
+Reviewer #2 (Remarks to the Author):  
+
+I have read the authors' responses. I think that they have thoroughly address all the comments. I recommend the work for publication.
+
+<--- Page Split --->
+
+Reply to the reviewer #1's Remarks to the Authors.  
+
+Remark 1: The authors have responded to all points raised in my report. Their response has clarified several key aspects of the paper; in particular, the interplay between mirror symmetry and surface states. While they have identified an interesting role of mirror symmetry in the topological response from surface states, I remain unconvinced by the extent of progress reported in this paper, given the large body of literature, including recent works by (at least) a subset of the authors, that have already explored ideas that overlap with those explored here. In my view, the key new result here is the generalization of the notion of the mirror- Chern number (generally used as a bulk invariant) to half- quantized Hall conductivity from surface states — individually, the notions of mirror- Chern number and half- quantized Hall response from isolated 2D Weyl/Dirac cones are not new. Further, the fact that the conclusions rest on the analyses of a 4- band model is a concern, given the maturity of the field. A minimal- model based calculation, while important and the essential first step (the emphasis on obtaining analytical results is appreciated), does not unambiguously establish the authors' claim that this work represents "an important advancement in our understanding of the role of symmetry in the generation of novel electronic phenomena". I would urge the authors to numerically demonstrate their key results in more realistic models [eg. the Hamiltonian in Eq. (52) of Phys. Rev. B 82, 045122 (2010)] or ab initio data. While the former type of models are important for comparing with experimental data in topological insulators, the latter is the best demonstration of any principle borne out of minimal models.  
+
+Reply 1: Our findings represent more than a mere generalization of previous results. So far it is the first metallic system to exhibit the quantum anomaly under a time- reversal symmetry. Firstly, in an insulating system, the mirror Chern number must be integer. However, in a metallic system, in which the surface states are gapless, it is the first example to exhibit half- quantization associated with a pair of gapless Dirac cones. To our best knowledge, there are currently no other known systems that exhibit half- quantization while maintaining time- reversal symmetry. Secondly, a lattice itself provides a gauge- invariant regularization, such as in the case of a strong topological insulator film or graphene. Our findings demonstrate that when an additional intrinsic symmetry is present to separate massless surface Dirac bands (including the surface states near Gamma point and other states within the bulk gap) across the entire energy spectrum, quantum anomalies can emerge and induce measurable physical effects. Therefore, our findings not only establish quantized topological invariants for metals but also revive the occurrence of quantum anomalies in systems that are usually expected to be free from them.
+
+<--- Page Split --->
+
+To convince the referee, we accept the referee's suggestion to do numerical calculation based on realistic tight- binding models with the same crystalline symmetries as in \(\mathrm{Bi}_2\mathrm{Se}_3\) and \(\mathrm{Bi}_2\mathrm{Te}_3\) , although the lattice construction is really complicated. Each unit cell contains six sites and each site has four orbitals. Fortunately, the band spectra with the warping effect that the reviewer was concerning are well reproduced. The main findings, such as the quantized mirror Hall conductance, are reproduced as we expected. It is important to emphasize that our main results are derived from a symmetry analysis and are not reliant on specific model details. The use of a simplified model allows for a more concise presentation of the findings.  
+
+## Remark 2:  
+
+1. I do not yet see why the authors emphasize the lattice-based calculations for surface states. On mathematical grounds, the fact that generally surface states exist over a finite sub-region of the surface Brillouin zone can be demonstrated within a k.p model [see for example, section V of Phys. Rev. B 82, 045122 (2010)]. On physical grounds, the half-quantized response should be accessible to a k.p model because the anomalous Hall response for a 2D metal boils down to a line-integral over the Fermi surface. Beyond concerns over quantitative agreements, the only other physically interesting reason for favoring lattice-based models appears to be tied to the discussion around the new Fig. 2 (d). This aspect has remained unclear to me, even after multiple re-reading of the text.  
+
+Reply 2: The reviewer said, "the half-quantized response should be accessible to a k.p model because the anomalous Hall response for a 2D metal boils down to a line-integral over the Fermi surface". This statement is not always true: it is only valid for a k.p model that can be placed on a lattice without introducing a doubling problem. A famous example is the linear Dirac cone cannot be realized on a lattice if no additional symmetry breaking term is introduced. If the symmetry is not broken, the Hall conductivity is zero. To realize a gapless Dirac cone we have to breaking the symmetry as a whole, which is actually a central issue in the lattice gauge theory.  
+
+In a k.p model there exists a singularity at the point \(k \to +\infty\) , we cannot simply use the Stokes' theorem to transfer the area integral to line integral around the Fermi surface in the TKNN formula because the contribution around the singularity is ignored. If we insist on using the formula with line integral by ignoring the singularity, and the \(\pi\) Berry phase in the k.p model in which \(\pm \pi\) are equivalent, it can utmost give rise one half of the quantum Hall conductance, but its sign has to be determined by the filled states far away from the Fermi surface. For a comprehensive model example supporting this statement, please consult the Appendix.  
+
+There is no similar issue on a lattice model, which can provides an ambiguous answer to the issue. In another word, the lattice model provides a natural
+
+<--- Page Split --->
+
+regularization of low- energy effective model on a finite Brillouin zone. That is why we rely on calculations based on the lattice model to study surface states. This approach helps to address the ambiguity inherent in the issue and provides clearer and more reliable results. We have already addressed half quantized Hall conductivity of single gapless Dirac cone on a lattice in our previous works.  
+
+There is an example to illustrate the shortcoming of the k.p model. The massive Dirac model in the k.p approximation gives a one- half quantized Hall conductance, which is impossible to be realized on a lattice according to the TKNN theorem that a fully- filled gapped band always gives an integer quantum Hall conductance. So, we should be very careful when using the k.p model to explore the physics related to topology.  
+
+Remark 3: 2. In their response the authors agree with the finite extent of the surface states, but this is not reflected by the Hamiltonian in Eq (8). If the states do not exist beyond a particular region of the surface Brillouin zone, why does the Hamiltonian exist? Moreover, terms proportional to \(S\) tilde \(\backslash \mathrm{sigma\_z}\$ [according to the convention of Eq (8)] appears due to the so- called hexagonal warping, and is an important feature of surface states of topological insulators, as discussed in Phys. Rev. Lett. 103, 266801 (2009). In light of the above two points, I would conclude Eq (8) is incorrect, which casts doubt over any analysis following Eq (8).  
+
+Reply 3: The Hamiltonian in Eq. 8 contains either the surface states \((\Delta (k) = 0)\) or the bulk states \((\Delta (k) \neq 0)\) , which forms a whole band in the first Brillouin zone. The \(\sigma_z\) term in Eq. 8 breaks the parity symmetry and time reversal symmetry, but the combination of the pair of gapless Dirac cones respect the mirror symmetry and time reversal symmetry. This is opposite to the hexagonal warping term introduced by Liang Fu, in which the term is cubic in momentum and does not breaking time reversal symmetry. The warping effect appears in the systems with the threefold rotation symmetry. Although it changes the shape of the dispersion spectra, it does not change the band inversion or topology of the band. The appearance of the warping term does not affect the quantization of the mirror Hall conductivity in the mirror- symmetric \(\mathrm{Bi}_2\mathrm{Te}_3\) with twin boundary. Use the reviewer's word, this is a secondary effect. We use the model with threefold rotation in newly added section "Material candidates". The warping effect is successfully reproduced, and the main results remain unchanged.  
+
+Eq. 8 is an analytical result from a three- dimensional lattice model for a strong topological insulator, which has fourfold (NOT threefold) rotation symmetry. That's why there is no warping effect as in Fu's paper.  
+
+Remark 4: 3. Regarding "Reply A4": What I understood from this response is that there are bulk states (states that are not exponentially localized on the two
+
+<--- Page Split --->
+
+surfaces of the slab) that contribute to Hall conductivity if the overall filling is sufficiently large. While I agree with this statement, I find this distinction can be better utilized in making a physically important statement regarding the thickness dependence of the mirror- Hall response. Otherwise, this is a secondary point. Moreover, the language used to describe additional states contributing to the Hall response needs to be sharpened (also see my concern in point 2 above while interpreting the numerical results in this context).  
+
+Reply 4: We appreciate that the reviewer accepted our explanation. In this work we focus on the quantized mirror Hall conductivity within the bulk gap, which is also the gap of the states of \(\mathrm{n} = 2\) . It would be interesting question for the mirror Hall conductivity out of the gap. When the chemical potential sweeps more bands, the mirror Hall conductance is no longer quantized. The thickness of a sample will change the density of states around the Fermi level, and change the mirror Hall conductivity. However the bulk gap remains unchanged. We agree that the thickness dependence should be an interesting issue, deserving further investigation in the future. We accept the suggestion by adding a few sentences related to Fig. 2c.  
+
+Remark 5: 4. While I appreciate the authors' contrasting their results with that obtained by Creutz & Horvath (CH), I would like to emphasize that the CH method itself is more general than its initial implementation. This is clearly demonstrated in the paper by König et al. noted in my previous report. Therefore, I would suggest taking both papers under consideration while contrasting between the two methods above the section "Quantum mirror Hall conductance". Reply 5: Thank the referee' for the suggestion. As we pointed out in the last reply, the CH method is incomplete to have the final solution. For example, the method cannot give the finite size effect in quantum spin Hall effect. The approach proposed by Creutz and Horvath (CH) has been employed to address the edge states in the quantum spin Hall system as discussed in Konig et al.'s paper. In that paper, Konig et al. utilize the similar model as us (shown from Eqs 12, 13, and 14). However, they also only provide an analytical solution for the edge states, as demonstrated in Eqs. 21 and 22 as CH's work. This can also be observed in the lower panel of Fig. 3. The bulk states (represented by blue lines) are only included to reproduce the appearance of exact diagonalization results that the number of bands at \(\mathrm{k} = 0\) and \(\mathrm{k} = \mathrm{pi}\) are not equal. Therefore, both the work of Creutz and Horvath (CH) and Konig et al.'s paper only provide solutions for the surface states, which are limited to a specific region of the surface Brillouin zone. As shown in Fig. S1 and Fig. S2 in the supplementary material of our manuscript, our analytical solutions, which are founded on the tight- binding model, exhibit remarkable consistency throughout the entirety of the Brillouin zone, also encompassing the bulk states. However, the mirror Hall effect originates from high- energy bulk states with mirror Berry curvature, rather than the surface states, as depicted in Fig. 5e. This mirror Hall effect can
+
+<--- Page Split --->
+
+be captured by our solution in Eq. 8, which includes contributions from both surface states and bulk states. This has already gone beyond the CH and Konig et al's papers. The solution of the surface states is not the key point in the present work.  
+
+Remark 6: 5. Overall, I still find the narrative is unnecessarily convoluted. Even after the substantial revision of the text in the present version, the key ideas developed here are obfuscated by a lot of details that appear to be secondary. Is it impossible to demonstrate the main message with an 4- band k.p model?  
+
+If so, then this is an important aspect of this work, and should be demonstrated more directly.  
+
+If not, then the lattice- based calculations add quantitative details, and only the final outcome is relevant to the main text. Moreover, in this case, it is unclear whether a more realistic 8- band k.p model is less useful than using a lattice- version of a 4- band model.  
+
+## Reply 6:  
+
+We would like to emphasize that our principal findings are broadly applicable, rooted in the topological characteristics and symmetries of the system, rather than dependent on the particulars of any single model. This universality has been thoroughly addressed in the section titled "The Mirror Symmetry and Single Gapless Dirac Cones." The subsequent discussion, which employs a detailed model, serves merely to provide a concrete example that validates our theoretical framework.  
+
+While the effect is straightforward from topology band structure and symmetry, it becomes a key issue how to measure it in experiment. Part of the manuscript is devoted to exploring transport signature of the effect. We think this is necessary to predict a new physical effect.  
+
+It is possible to demonstrate the main message with the k.p model if the model can be realized on a lattice without introducing the doubling problem. The lattice calculation here aims to avoid the shortcoming of the k.p theory, as we mentioned in Reply 1. The 4- and 8- band k.p models may make the band dispersions closer to the DFT calculation. The most important point here is whether the model can capture the band inversion for nontrivial topology, and whether the system has the mirror symmetry. As clarified in Reply 2, our preference for utilizing the tight- binding model is to yield an unambiguous result for the problem at hand.  
+
+In the newly added section "Material candidates", we use the tight- binding approximation based on the four- band k.p model. The numerical results fully support our point of view.
+
+<--- Page Split --->
+
+Remark 7: 6. While I have not noted this in my previous report, I find the word "discovery" in the abstract to be too strong for a theoretical proposal, whose applicability to actual materials is yet to be demonstrated.  
+
+Reply 7: We have replaced the term "discovery" with "finding" in the text.  
+
+Remark 8: I was asked to comment on the response to the 3rd referee's comments. I will limit myself to comments \(C2\) & \(C3\) , since the other two are rather straightforward which the authors have addressed.  
+
+Regarding \(C2\) : This is an important point, and underscores my suggestion above on working with ab initio data to unambiguously demonstrate the existence of the mirror Hall effect in actual materials. The authors have responded to this point qualitatively. My concern here is that if the mirror symmetry is broken (weakly or strongly), then, in principle, the two Dirac cones can hybridize and mutually gap out. This will revert the system to the very regime the authors want to avoid — topological response from gapped surface states (see "Reply A1").  
+
+Reply 8: The opening of the gap in the Dirac cones can occur through two mechanisms: i) magnetic doping on the surface, which breaks the time- reversal symmetry, or ii) the coupling between two surface Dirac cones induced by finite size effects. Actually, these two mechanisms do not necessarily break mirror symmetry. Magnetic doping on both surfaces in an equal manner preserves mirror symmetry, and reducing the thickness of a mirror- symmetric film where two Dirac cones hybridize and form a gap, also does not break mirror symmetry. If mirror symmetry is maintained, the mirror eigenvalue remains a valid quantum number, and the mirror Hall conductivity can still be defined. However, in this scenario, the Hall conductivity for each mirror sector becomes an integer (the mirror Hall conductivity is also an integer) when the chemical potential is located within the gap of the Dirac cones, as discussed in the "Discussion and Conclusion" section of the previous revised version.  
+
+In epitaxially grown films, the charge transfer from the substrate will induce a potential difference between the two surfaces of the topological insulator which breaks the mirror symmetry explicitly. The substrate- induced potential can be modelled as \(H_{V} = \sum_{k,z} \psi_{k,z}^{+} V(z) \psi_{k,z}\) where \(V(z)\) is nonzero at several bottom layers and nearly uniform in x- y plane. As depicted in the left panel of Figure R2, the presence of a potential breaks the degeneracy between the top and bottom surface states, resulting in the explicit breaking of mirror symmetry. However, in the right panel of Figure R2, it can be observed that when the chemical potential is situated within the bulk band gap (blue region), the mirror Hall conductivity remains quantized at - 1. Indeed, the mirror Hall effect mainly originates from high- energy states primarily located within the bulk of the material. The potential on the surface does not significantly affect these states.
+
+<--- Page Split --->
+![](images/Figure_unknown_2.jpg)
+
+<center>Fig. R2:(left) The bandstructure of Bi2Se3 along [001] direction with a twin boundary. The color of the dots represents the mean position of the states along the [001] direction.A gate voltage of \(\mathrm{V = 0.1eV}\) is applied to the bottom surface. The remaining parameters are the same as in the main text. (right) The mirror Hall conductance \(\sigma_{xy}^{M2}\) as a function of \(\mu\) . The red dashed line serves as a visual guide to indicate the value of -1. The blue shadowed region indicates the gap regions for the bulk states. </center>  
+
+Remark 9: Regarding C3: This is also an important issue, since measuring lattice- symmetry protected topological response is not obvious. I am unaware of any existing work where a lattice- symmetry protected transport response has been measured. The authors seem to recognize this challenge, and have made an effort to device a measurement protocol, but this may require/invite further scrutiny. Therefore, it would be helpful to cite other similar proposals (i.e. measurement of lattice- symmetry protected response) in this context.  
+
+Reply 9: This is also new to the authors. We know this new effect is similar to the spin Hall effect in conductors or semiconductors, but the mirror Hall conductance is only one half of the quantum spin Hall effect in an insulating phase. Maybe the techniques to measure spin Hall effect are helpful for this new effect (Ref. 48- 52). Furthermore, we have included several recent experimental measurements of lattice- symmetry protected transport responses in the section "transport signature".  
+
+## Appendix:  
+
+To offer a more succinct clarification in response to Reply 2, let's consider two models as examples:  
+
+\[\mathrm{Model}1:H_{1} = \nu \big(k_{x}\sigma_{y} - k_{y}\sigma_{x}\big) + \lambda \big(3k_{x}^{2}k_{y} - k_{y}^{3}\big)\sigma_{z}\]
+
+<--- Page Split --->
+
+Model 2: \(H_{2} = H_{1} + b\left(k_{x}^{2} + k_{y}^{2}\right)^{2}\sigma_{z}\)  
+
+\(H_{1}\) represents the Hamiltonian for topological insulator surface states with a warping effect as in Liang Fu's paper, while \(H_{2}\) is \(H_{1}\) with an additional time- reversal symmetry- breaking term, \(b\left(k_{x}^{2} + k_{y}^{2}\right)^{2}\sigma_{z}\) . Up to the third order in k, \(H_{2}\) is identical to \(H_{1}\) . When chemical potential \(\mu\) is near the Dirac point, the line integral of Berry connection over the Fermi surface \(\mu\) for these two Hamiltonians is \(-\pi\) as \(\mu\) approaches zero, as shown by the red and blue lines in Fig. R1. However, when we evaluate the Hall conductivity for these two models, we find that for model 2, the Hall conductivity is same as Berry phase (only multiplied by a constant factor). However, for Model 1, due to the time- reversal symmetry, the Hall conductivity is zero (blue triangle with dashed line), which conflicts with its Berry phase value on Fermi surface. The reason for the discrepancy in the results is that the relation \(\int d^{2}k\Omega (k)\Theta (\mu -\epsilon (k)) = \oint_{FS}dl\cdot A\) is based on Stokes' theorem, which requires that the integral region be free of singularities. However, in Model 1, the point at \(k\rightarrow \infty\) can be considered a singularity due to the spin texture pointing in different directions as the amplitude angle of \(k\) varies from 0 to \(2\pi\) . This singularity violates the conditions required for the application of Stokes' theorem, resulting in different outcomes for these two models.  
+
+![](images/Figure_unknown_3.jpg)
+
+<center>Fig. R1: The blue and red solid lines represent the line integral of the Berry connection over the Fermi surface \(\mu\) for these two Hamiltonians. The dashed lines with triangles represent the corresponding Hall conductivities. \(\nu = 1, \lambda = 1, b = 0.2\) . </center>  
+
+To further analyze the distinction between the models, we can place both k.p models on a lattice using the substitutions \(k_{x} \rightarrow \sin k_{x}\) and \(k_{x}^{2} \rightarrow 2(1 - \cos k_{x})\) :
+
+<--- Page Split --->
+
+\[\mathrm{Model~1:}H_{1}^{l a t t i c e} = v\sin k_{x}\sigma_{y} - v\sin k_{y}\sigma_{x} + \lambda \sin k_{y}(4 - 6\cos k_{x} +\] \[2\cos k_{y})\sigma_{z\] \[\mathrm{Model~2:~}H_{2}^{l a t t i c e} = v\sin k_{x}\sigma_{y} - v\sin k_{y}\sigma_{x} + [\lambda \sin k_{y}(4 - 6\cos k_{x} +\] \[2\cos k_{y}) + b(4 - 2\cos k_{x} - 2\cos k_{y})^{2}]\sigma_{z\]  
+
+It is easily to check for \(H_{1}^{l a t t i c e}\) lattice, except at the \(\Gamma = (0,0)\) point, there are additional massless Dirac fermions at other time- reversal invariant momenta (TRIMs). However, for \(H_{2}^{l a t t i c e}\) lattice, \(\Gamma = (0,0)\) point is the only point hosting a massless Dirac fermion.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+I appreciate that the authors have demonstrated the robustness of their results for realistic models and they have clarified the distinction between the present work vs. their earlier work [Phys. Rev. B 107, 125153 (2023)]. I was also happy to learn that the authors agree that their main findings can be demonstrated via suitable k.p models.  
+
+Their response to "Remark 8", however, has confused me. In particular, a comparison of the mirror- Hall conductivity in Fig. 2 vs. Fig. S3 (Fig R2 of the response) seems to suggest that the mirror symmetry is not necessary for the quantized response (somehow, even when mirror symmetry is broken "explicitly", the authors are defining a mirror- Hall conductivity in Fig S31).  
+
+Further, the authors attribute the quantized response in Fig. S3 to "high energy states primarily located in the bulk". This appears to be in conflict with a central assertion noted in the abstract "... mirror symmetry assigns a unique mirror parity to each Dirac cone, resulting in a half- quantized Hall conductance of \(+ / - e^{\wedge}2 / 2h\) for each cone.". Therefore, in an experiment, if one detects a quantized mirror- Hall conductivity, how would one know its source?  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors have satisfactorily address all my questions. I support publication.
+
+<--- Page Split --->
+
+## Reply to the report of Reviewer #1  
+
+Comment: Their response to "Remark 8", however, has confused me. In particular, a comparison of the mirror- Hall conductivity in Fig. 2 vs. Fig. S3 (Fig R2 of the response) seems to suggest that the mirror symmetry is not necessary for the quantized response(somehow, even when mirror symmetry is broken "explicitly", the authors are defining a mirror- Hall conductivity in Fig S3!).  
+
+Reply: The Reviewer #1 raised an interesting and new question, which is generated in our previous reply. The mirror symmetry is of course necessary for the exact quantized response. This has already been shown in "transport signature" of Section 2 that an increase of symmetry- breaking causes a reduction in \(l_{M_z}\) and a corresponding reduction of half- quantum mirror Hall effect related transport phenomena. Here, \(l_{M_z}\) is the inter mirror scattering length which can be used to characterize the extent of mirror symmetry breaking.  
+
+Generally, when the symmetry is broken explicitly, the mirror Hall conductance will deviate from the quantized value. In Fig. S3, we consider a special symmetry breaking term which acts only on the bottom layer. It removes the degeneracy of the surface states on the top and bottom layer, which are almost localized at the two surface layers. However, as illustrated in the right panel of Fig. R1 (also shown in Fig. S3), we observe that mirror symmetry breaking on the surface layers has a minimal impact on the quantization of mirror Hall conductivity (approximated as \(- e^2 /h\) ) when the chemical potential intersects with the surface states. This minimal effect is due to the sum of Berry curvatures across an equal energy surface of the surface states being zero, which does not contribute to the Hall conductivity. This lack of contribution underlies the emergence of the mirror Hall plateau as the chemical potential shifts within the bulk gap.  
+
+For comparison, we have also analysed the effects of applying this voltage to a layer just above the mirror plane of the film, as depicted in the left panel of Fig. R1. Notably, under these conditions, there is a significant change in the quantization of mirror Hall conductivity. This change occurs because this mirror symmetry breaking voltage impacts the high- energy bulk states which primarily contribute to the half- quantized mirror Hall conductivity.  
+
+These results reflect unique properties of the half- quantum mirror Hall effect in topological insulator films with mirror symmetry, meriting further investigation in the future.
+
+<--- Page Split --->
+![PLACEHOLDER_27_0]
+
+<center>Fig. R1: (left) The mirror Hall conductivity \(\sigma_{xy}^{Mz}\) as a function of \(\mu\) for various gate voltages V applied to the layer slightly above the midpoint of the film. (right) The mirror Hall conductivity \(\sigma_{xy}^{Mz}\) as a function of \(\mu\) for gate voltage of V applied to the bottom surface. The remaining parameters are the same as in the main text. The gray dashed line serves as a visual guide to indicate the value of -1. </center>  
+
+Comment: .Further, the authors attribute the quantized response in Fig. S3 to "high energy states primarily located in the bulk". This appears to be in conflict with a central assertion noted in the abstract "... mirror symmetry assigns a unique mirror parity to each Dirac cone, resulting in a half-quantized Hall conductance of +/- e^2/2h for each cone.". Therefore, in an experiment, if one detects a quantized mirror-Hall conductivity, how would one know its source?  
+
+Reply: In the abstract, the term 'Dirac cones associated with surface electrons' refers to the entire gapless band spanning the first Brillouin zone, which includes both the low- energy surface states and the high- energy bulk states, also as illustrated in Fig. 2d of the main text. This distinction is important because the analysis of the Hall effect requires consideration of all occupied states within a band. We have revised this description in the abstract for better clarity, 'Dirac cones in the first Brillouin zone associated with surface electrons'.  
+
+In experiment, it is hard to figure out the source if one just detects a single data of quantized mirror Hall conductivity. However, if a plateau of mirror Hall conductivity is measured as the chemical potential varies in a finite range within the bulk gap, one can conclude that the surface electrons has no contribution to the value of the mirror Hall conductivity, but the plateau (no change with the chemical potential) is attributable to the presence of gapless surface states. The
+
+<--- Page Split --->
+
+nonzero value does arise from contributions by high- energy bulk states. Therefore, there is no conflict here.
+
+<--- Page Split --->
+
+REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors have responded to the two points raised in my previous report. I find the results of investigations in to the first point to be intriguing, and it supports the claim made by the authors that the quantized response arises from the bulk states. This also indicates that the quantized response that is obtained in this paper requires a curious mixture of both bulk and surface states - - the surface states provide the basis for the Dirac cones, while the bulk states are necessary for the quantized response. The response to the second point appears to highlight this complex source of the phenomenon. Because of this feature, the thickness of the slab may play a non- trivial role.  
+
+While unusual for what one would expect for a topological response, this work represents an instance where surface and bulk states play a joint role. This property and the eventual outcome of a quantized mirror- Hall conductivity could be expected to be of immediate interest to both experimentalists and theorists working on topological materials.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__72f5c47559c66648327d1e06e829ba3510af74f9ae1229b38b159158d672cb65/supplementary_0_Peer Review File__72f5c47559c66648327d1e06e829ba3510af74f9ae1229b38b159158d672cb65_det.mmd

@@ -0,0 +1,573 @@
+<|ref|>title<|/ref|><|det|>[[60, 40, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[68, 110, 362, 140]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[70, 154, 513, 180]]<|/det|>
+Half Quantum Mirror Hall Effect  
+
+<|ref|>image<|/ref|><|det|>[[57, 732, 240, 783]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[250, 732, 912, 785]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 145, 392, 161]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 202, 880, 330]]<|/det|>
+In this paper the authors present an account of generating a mirror Hall effect through surface states of a three dimensional strong topological insulator (TI). The underlying source is the half- quantized Hall conductivity supported by an isolated two dimensional Dirac point. At its core, utilizing the surface Dirac cones of an 3D TI to generate half- quantized Hall effect is not new, and many papers, both theoretical and experimental, have been written on this topic over the past decade (most of these papers are missing from the reference list). Since some form of magnetization is involved in these studies, the role of the mirror symmetry explored here is perhaps the only new aspect.  
+
+<|ref|>text<|/ref|><|det|>[[115, 341, 579, 358]]<|/det|>
+Beyond the generalities, I have the following specific questions:  
+
+<|ref|>text<|/ref|><|det|>[[115, 397, 879, 488]]<|/det|>
+1. The authors appear to be using a lattice version of the Hamiltonian introduced in Eq (1) of Nat Phys 5, 438 (2009) [at least for the concrete calculations reported in the 'Methods' section]. For this case, I do not see how the z-mirror operator globally commutes with the Hamiltonian, as mentioned in section 2 ("Results"). Therefore, it is unclear to me how one may obtain the direct-sum/block-diagonal form of the 3D Hamiltonian.  
+
+<|ref|>text<|/ref|><|det|>[[115, 527, 876, 582]]<|/det|>
+2. Usually surface states do not exist at all points in the surface Brillouin zone, as discussed, for example, by König et al. in Physical Society of Japan 77, 031007 (2008) [arxiv 0801.0901]. I do not see such constraint in Eq (2).  
+
+<|ref|>text<|/ref|><|det|>[[115, 620, 840, 658]]<|/det|>
+3. I am afraid I could not understand the reason behind the difference between results obtained by employing Eq (2) vs. Eq (3), and, consequently, the contrasting data presented in Fig 2(b).  
+
+<|ref|>text<|/ref|><|det|>[[115, 696, 875, 807]]<|/det|>
+4. In their calculations that lead to Eq (2), the supplemental materials seems to suggest they are using plane waves as a solution. In the absence of translation invariance along z-direction, why is this a good choice for eigenstates (it seems \(\backslash \mathrm{xi}\) is allowed to be complex; if so, it should be stated at the beginning)? Moreover, how does their method and its results compare with that by Creutz & Horvath in Phys. Rev. D 50, 2297 (1994) that was used for deriving surface states by König et al.? The form of the wavefunctions appears to be quite different at face value.  
+
+<|ref|>text<|/ref|><|det|>[[115, 846, 303, 863]]<|/det|>
+5. A few additional points  
+
+<|ref|>text<|/ref|><|det|>[[115, 874, 397, 891]]<|/det|>
+a) Subfigure labels are missing in Fig 1.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 625, 107]]<|/det|>
+b) In Fig 2(a) the red circles are hardly visible, especially on a printout.  
+
+<|ref|>text<|/ref|><|det|>[[113, 118, 884, 154]]<|/det|>
+c) The subsection "Material candidates" could be placed at the end of section 2 in order to smoothen the narrative.  
+
+<|ref|>text<|/ref|><|det|>[[115, 165, 835, 237]]<|/det|>
+d) Much of the content of the section "Topological field theory for quantum mirror Hall effect on a lattice" in 'Methods' can be suppressed or moved to the Supplementary Materials, since the methodology is fairly well known [see for example, Phys. Rev. B 99, 235144 (2019) and references therein]  
+
+<|ref|>text<|/ref|><|det|>[[115, 248, 860, 284]]<|/det|>
+e) The notation for the 4x4 matrices used to in subsection "Mirror plane with time-reversal symmetry breaking" is not consistent — \tauau and \sigmaigma are interchanged below Eq (1).  
+
+<|ref|>text<|/ref|><|det|>[[115, 295, 488, 312]]<|/det|>
+f) What is meant by "magnetization" below Eq (1)?  
+
+<|ref|>text<|/ref|><|det|>[[115, 323, 866, 359]]<|/det|>
+g) Below Eq (2) the authors write "This is distinct from the conventional effective model for the surface states which is only ...". What do they mean?  
+
+<|ref|>text<|/ref|><|det|>[[115, 370, 867, 405]]<|/det|>
+h) The notations for position and momentum space coordinates should be made uniform — while bold r_|perp is used for the former, only bold k is used for the latter.  
+
+<|ref|>text<|/ref|><|det|>[[115, 473, 392, 490]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 529, 874, 584]]<|/det|>
+The manuscript by Fu et al presents a theory proposal of half quantum mirror Hall effect. The concept is quite novel and can be of interest. However, before I can recommend publication, I have the following concerns that the authors should address  
+
+<|ref|>text<|/ref|><|det|>[[115, 623, 879, 678]]<|/det|>
+1. Unclear presentation and lack of schematic illustrations: The presentation is in general not clear. There is significant disconnection between the general introduction, the mathematical derivation and the figures. Below I give some examples:  
+
+<|ref|>text<|/ref|><|det|>[[115, 688, 869, 798]]<|/det|>
+a. The authors emphasized the H_x, which seems to be a Hamiltonian with a given mirror eigenvalue in their math derivations. However, in their figures, it is not clear how to understand H_x at all. People typically think about a topological insulator with one Dirac cone at the top surface and another Dirac cone at the bottom surface. But I am guessing that the bands described by H_x does not come from a single surface. Instead, the bands described by H_x has contribution from both surface Dirac cones. Is that correct? Can the authors make some figure to label the bands of H_x and H_x?  
+
+<|ref|>text<|/ref|><|det|>[[115, 808, 877, 862]]<|/det|>
+b. Similarly, the schematic in Fig. 3a is confusing and lacks critical information. E.g., where is the mirror plane respect to the sample? I am guessing that the mirror plane is the horizontal x-y plane at the center of the sample, is that correct?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 881, 180]]<|/det|>
+c. The authors constantly make analogy to the spin Hall effect. In the spin Hall effect, there is a very nice schematic, where as a current flows along \(x\) , the electrons with opposite spins bent toward the \(+y\) and \(-y\) directions, respectively. In this way, there is no net electrical Hall voltage but there is a spin voltage along the \(y\) direction. Can the authors make a similar plot to help people understand the mirror Hall effect they propose?  
+
+<|ref|>text<|/ref|><|det|>[[114, 191, 880, 355]]<|/det|>
+d. It is further confusing why the mirror Hall effect (or the spin Hall effect) can be directly measured by the two-probe conductance. The authors made some description in the text that is very difficult for readers to understand: "The charge current contains two parts \(\mathrm{Jc,y = Jssc,y + JmHc,y}\) : the first term Jss \(\mathrm{c,y}\) comes from the conducting surface states and the second term \(\mathrm{JmHc,y}\propto \tan 20\mathrm{m}\) arises from the spatial accumulation of the mirror polarization density and is an effect due to the existence of the mirror Hall effect. This mirror Hall mediated charge transport can be understood as follows: the electric field first induces a mirror charge accumulation on the boundary via the mirror Hall effect and then is converted into the charge current along the electric field via the inverse mirror Hall effect". Some graphics will help a lot  
+
+<|ref|>text<|/ref|><|det|>[[114, 366, 879, 476]]<|/det|>
+2. What is the role of nontrivial topology here? The authors constantly make analogy to the spin Hall effect. However, spin Hall effect is widely studied in heavy metals such as Ta, W. However, in the present work, the authors consider a topological insulator and seems to be claiming some quantized response. It is unclear what kind of transport is quantized? E.g., even in quantum spin Hall, the edge conductance is famously non-quantized due to inelastic scattering. No discussion of that kind is provided. What if I have a metal that has M_z mirror symmetry? Do I have mirror Hall effect (just not quantized)?  
+
+<|ref|>text<|/ref|><|det|>[[114, 486, 880, 576]]<|/det|>
+In general, I think the paper may contain some interesting concept that deserves to be published in a top journal, however, the current manuscript is written in a way that is very hard for people to judge. The authors also did not show great care and desire to make the presentation readable and understandable. I cannot recommend the publication in the present form but is happy to consider a significantly improved version.  
+
+<|ref|>text<|/ref|><|det|>[[115, 673, 392, 689]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 729, 866, 856]]<|/det|>
+In this paper, the authors propose a novel topological phenomenon dubbed the half- quantum mirror Hall effect. They consider a strong topological insulator film with mirror reflection symmetry. Using mirror symmetry, they split a pair of surface Dirac modes on the top and bottom surfaces into a single one and derive a half- quantized Hall conductance on each mirror subsector. Furthermore, they also reveal that the half- quantized mirror Hall conductance results in a net current due to the accumulation of a mirror charge and the inverse mirror Hall effect. The idea of this paper looks novel and exciting. However, the following points should be clarified before the decision on the recommendation.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 90, 872, 143]]<|/det|>
+1. Whereas the authors assume an exact mirror-reflection symmetry, actual materials do not support it due to the imperfection of the crystal structures. The author should clarify how the imperfection affects the obtained result.  
+
+<|ref|>text<|/ref|><|det|>[[114, 155, 861, 209]]<|/det|>
+2. It needs to be clarified that the half-quantized value of the Hall conductance can be concluded through the two-terminal measurement. I understand that a non-zero mirror Hall conductance can be observed, but I do not know how to determine its value.  
+
+<|ref|>text<|/ref|><|det|>[[112, 220, 850, 238]]<|/det|>
+3. The superscript and subscript m should be modified as Mz to avoid confusion with the band index.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[120, 89, 647, 122]]<|/det|>
+## Reply to the referees' reports  
+
+<|ref|>text<|/ref|><|det|>[[120, 142, 430, 159]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 177, 881, 317]]<|/det|>
+Remark A1: In this paper the authors present an account of generating a mirror Hall effect through surface states of a three dimensional strong topological insulator (TI). The underlying source is the half- quantized Hall conductivity supported by an isolated two dimensional Dirac point. At its core, utilizing the surface Dirac cones of an 3D TI to generate half- quantized Hall effect is not new, and many papers, both theoretical and experimental, have been written on this topic over the past decade (most of these papers are missing from the reference list). Since some form of magnetization is involved in these studies, the role of the mirror symmetry explored here is perhaps the only new aspect.  
+
+<|ref|>text<|/ref|><|det|>[[118, 317, 880, 595]]<|/det|>
+Reply A1: We appreciate the referee's comments and would like to clarify the novelty of our study. While it is true that the concept of using the surface Dirac cones in a three- dimensional topological insulator to produce a half- quantized Hall effect has been explored in the literatures, it is important to note that previous works on this topic have primarily focused on the gapped surface states. In contrast, the half- quantized Hall conductivity in our study is attributed to the two- dimensional gapless Dirac cone, which is a new and distinct contribution. We have discussed this in detail in Ref. [26] (PRB 107, 125153 (2023)), where we highlight that the result that is derived from the massive Dirac fermions in the continuous model does not hold for a single gapped surface Dirac cone on a lattice model as the first Brillouin zone is finite according to the TKNN theorem. Therefore, the half- quantized Hall conductance of the gapless Dirac cone is a new and significant contribution to the field. Furthermore, we would like to emphasize that in our study, the strong topological insulator film does not break the time- reversal symmetry. Instead, the additional mirror symmetry plays a decisive role in the formation of this new effect. We believe that this represents an important advancement in our understanding of the role of symmetry in the generation of novel electronic phenomena.  
+
+<|ref|>text<|/ref|><|det|>[[120, 612, 634, 630]]<|/det|>
+Beyond the generalities, I have the following specific questions:  
+
+<|ref|>text<|/ref|><|det|>[[118, 647, 880, 735]]<|/det|>
+Remark A2: 1. The authors appear to be using a lattice version of the Hamiltonian introduced in Eq (1) of Nat Phys 5, 438 (2009) [at least for the concrete calculations reported in the 'Methods' section]. For this case, I do not see how the z- mirror operator globally commutes with the Hamiltonian, as mentioned in section 2 ("Results"). Therefore, it is unclear to me how one may obtain the direct- sum/block- diagonal form of the 3D Hamiltonian.  
+
+<|ref|>text<|/ref|><|det|>[[118, 735, 880, 874]]<|/det|>
+Reply A2: Yes, the model we used is a lattice version of Eq. (1) in Nat. Phys. (Ref. 39). The model has the mirror symmetry if the system is symmetric with open boundary condition along the z- axis. We have carefully reviewed the manuscript and made the necessary modifications to present a clearer explanation. We have added a paragraph where we explicitly include the z- mirror symmetry operator and explain how it commutes with the lattice Hamiltonian. This addition will address the referee's question and provide a more detailed understanding of the relationship between the mirror symmetry operator and the Hamiltonian. Thank you for bringing this issue to our attention.  
+
+<|ref|>text<|/ref|><|det|>[[116, 891, 878, 909]]<|/det|>
+Remark A3: 2. Usually surface states do not exist at all points in the surface Brillouin zone, as
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 878, 120]]<|/det|>
+discussed, for example, by König et al. in Physical Society of Japan 77, 031007 (2008) [arxiv 0801.0901]. I don not see such constraint in Eq (2).  
+
+<|ref|>text<|/ref|><|det|>[[117, 121, 880, 380]]<|/det|>
+Reply A3: The reviewer is right, the surface states only exist in a small portion of the first Brillouin zone, and its energy dispersion is located within between the bulk gap of three- dimensional topological insulator. Beyond the regime, the states evolve into the bulk one. The two parts consist of a complete band in the first Brillouin zone. In Eq. (8) [Originally Eq. (2)], for \(\Theta (x) = 0\) , the mass term \(\Delta (k) = 0\) , which represents the surface states while for \(\Theta (x) = 1\) , \(\Delta (k) = m_0(k)\) , which represent the bulk states. The dispersion of Eq. (2) forms a complete Dirac cone, which is defined in the first Brillouin zone. The surface electrons are only part of the Dirac cone near the Dirac point. As shown in "Methods" section "Derivation of Gapless Dirac Cone Model with Parity Symmetry Breaking", Eq. (8) is derived based on the tight- binding model (not \(k^*\) p model) which is applicable across the entire 2D Brillouin zone and not restricted to the vicinity of the Gamma point. The basis function in Eqs. (23) and (24) which are mirror eigenfunctions correspond to surface states for small values of \(k\) and naturally evolve into bulks state as \(k\) approaches the corners of the Brillouin zone as shown in Fig. S2. This is the key difference between our model (Eq. (8)) and the conventional effective model for the surface states in "Journal of Physical Society of Japan 77, 031007 (2008)".  
+
+<|ref|>text<|/ref|><|det|>[[117, 397, 880, 450]]<|/det|>
+Remark A4: 3. I am afraid I could not understand the reason behind the difference between results obtained by employing Eq (2) vs. Eq (3), and, consequently, the contrasting data presented in Fig 2(b).  
+
+<|ref|>text<|/ref|><|det|>[[117, 451, 880, 678]]<|/det|>
+Reply A4: Eq. (8) [Originally Eq. (2)] only describes the band of the gapless Dirac cone, which contributes one half Hall conductance. Eq.(9) [Originally Eq. (3)] is a formula for the mirror Hall conductivity for all the bands of the thin film as shown in Fig.2b[Originally Fig. 2a], which contains the bands from \(n = 1\) to \(Lz\) in Eq. (26) [Originally Eq. (14)]. Eq. (8) [Originally Eq. (2)] is only the bands of \(n = 1\) , i.e., the gapless Dirac cones labelled by the green line in Fig. 2b[Originally Fig. 2a]. The black line with squares in Fig. 2c[Originally Fig. 2b] is for the total mirror Hall conductivity for all the bands, and the green dashed line is just for the band for the gapless Dirac cones. Apart from the four gapless bands with \(n = 1\) , all other bands are gapped and topologically trivial (when chemical potential lies within their gaps, the mirror Hall conductivity is zero), however they also contribute when chemical potential shifts into the bulk states as shown by the figure below. The contributions from \(n = 3,4,\ldots ,Lz\) bands, not displayed in the figure below, should also be summed to produce the final results indicated by the black line with squares. That's why the two results are different.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[145, 90, 533, 316]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[117, 326, 880, 378]]<|/det|>
+<center>Caption: Black line represents the mirror Hall conductivity from all bands. Red line represents the contribution from the four gapless bands with \(n = 1\) . Blue line represents the contribution from the second-lowest energy gapped bands with \(n = 2\) . </center>  
+
+<|ref|>text<|/ref|><|det|>[[117, 395, 880, 516]]<|/det|>
+Remark A5: 4. In their calculations that lead to Eq (2), the supplemental materials seems to suggest they are using plane waves as a solution. In the absence of translation invariance along z- direction, why is this a good choice for eigenstates (it seems \(\vert x\vert\) is allowed to be complex; if so, it should be stated at the beginning)? Moreover, how does their method and its results compare with that by Creutz & Horvath in Phys. Rev. D 50, 2297 (1994) that was used for deriving surface states by König et al.? The form of the wavefunctions appears to be quite different at face value.  
+
+<|ref|>text<|/ref|><|det|>[[117, 517, 880, 621]]<|/det|>
+Reply A5: We take the periodic boundary condition along the \(x\) and \(y\) direction, and open boundary condition in the \(z\) direction. The plane wave is used only for the \(x - y\) plane. The effective mass in the 1- D Hamiltonian is a function of \(k\) . There is no translational symmetry, and H_{1d} in Eq. (21) (Originally Eq.(9)) is a one- dimensional lattice open chain. The eigen wave function is a function of the lattice site \(L_{z}\) , not a plane wave as shown in Eq. (23) and Eq. (24). We clarify this point in the revised version.  
+
+<|ref|>text<|/ref|><|det|>[[117, 639, 880, 848]]<|/det|>
+We read Creutz & Horvath's paper very carefully. We are surprised they have intended to solve this problem long before the birth of topological insulator. From Eq. 15 in PRD paper, it stated clearly that they tried to find the surface states of 1+ D- dimensional film, (D=1 and 3). We noted that in Creutz and Horvath's paper: 1) They obtained the dispersion of the boundary states of 1+1 and 1+ 3 dimension film, that is the one- dimensional chiral edge state and three- dimensional gapless Dirac cone which are eigenvectors of chiral operator \(\gamma_{5}\) ; 2) they noted the high energy part which violates the chiral symmetry. 3) they presented the numerical dispersion of the lattice model. However, 4) They didn't establish a set of equations to determine the variable \(\backslash \mathrm{lambda}\) and the energy eigenvalues, and failed to present the general solutions of the wave functions. Most important, due to the double degeneracy of the surface states, a pair of gapless Dirac fermions are usually regarded to cancel the chiral anomaly with each other as the two authors pointed out in the introduction.  
+
+<|ref|>text<|/ref|><|det|>[[117, 866, 880, 900]]<|/det|>
+In the present work, we focus on the two- dimensional gapless Dirac fermions reduced from a three- dimensional system. We established a complete set of equations to determine the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 880, 172]]<|/det|>
+energy dispersion and \lambda lambda as a function of a finite thickness Lz. By means of the additional mirror symmetry, we classify the energy eigenstates into two separated classes of the gapless Dirac cones with even and odd parity [Eq. (23) and Eq. (24)]. The complete dispersions of the two gapless Dirac cone are presented, each gapless Dirac cone with parity anomaly.  
+
+<|ref|>text<|/ref|><|det|>[[118, 190, 576, 206]]<|/det|>
+We summarize the key differences between two works:  
+
+<|ref|>text<|/ref|><|det|>[[118, 207, 880, 310]]<|/det|>
+1) Our model possesses time reversal (TR) symmetry thus belongs to class All which is characterized by a \(Z_{2}\) topological invariant and helical edge states. In contrast, the model employed in PRD lacks TR symmetry and belongs to class A, which is characterized by a Chern number and chiral edge states. Consequently, the quantum anomalies addressed in the two works are fundamentally different: PRD deals with chiral anomaly, whereas our work concentrates on parity anomaly.  
+
+<|ref|>text<|/ref|><|det|>[[118, 311, 880, 380]]<|/det|>
+2) The analytical study in the Appendix of PRD only focuses on the solution of the chiral edge states which is restricted in a small portion of the Brillion zone. In contrast, our work analytically provides the all the eigenfunctions across the entire Brillion zone for the film geometry by utilizing the mirror symmetry.  
+
+<|ref|>text<|/ref|><|det|>[[118, 381, 880, 433]]<|/det|>
+3) PRD points out that the double degeneracy of surface states on two surfaces cancels the chiral anomaly. However, in our work, we find the inclusion of additional mirror symmetry reintroduces the parity anomaly which can have physical consequences.  
+
+<|ref|>text<|/ref|><|det|>[[118, 450, 880, 485]]<|/det|>
+We want to emphasize that the study of mirror symmetry not only facilitates an analytic solution but also introduces new physics in this three-dimensional topological insulator film.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 503, 218, 518]]<|/det|>
+## Remark A6:  
+
+<|ref|>text<|/ref|><|det|>[[118, 521, 639, 572]]<|/det|>
+5. A few additional points a) Subfigure labels are missing in Fig 1. Reply A6: we replotted and labelled of the subfigures in Fig. 1  
+
+<|ref|>text<|/ref|><|det|>[[118, 589, 688, 605]]<|/det|>
+b) In Fig 2(a) the red circles are hardly visible, especially on a printout.  
+
+<|ref|>text<|/ref|><|det|>[[118, 607, 363, 623]]<|/det|>
+Reply A6: we replotted Fig. 2.  
+
+<|ref|>text<|/ref|><|det|>[[118, 641, 880, 675]]<|/det|>
+c) The subsection "Material candidates" could be placed at the end of section 2 in order to smoothen the narrative.  
+
+<|ref|>text<|/ref|><|det|>[[118, 676, 880, 711]]<|/det|>
+Reply A6: we accept the suggestion to place the subsection in section 3 "Discussion and conclusion".  
+
+<|ref|>text<|/ref|><|det|>[[118, 729, 880, 799]]<|/det|>
+d) Much of the content of the section "Topological field theory for quantum mirror Hall effect on a lattice" in "Methods' can be suppressed or moved to the Supplementary Materials, since the methodology is fairly well known [see for example, Phys. Rev. B 99, 235144 (2019) and references therein]  
+
+<|ref|>text<|/ref|><|det|>[[118, 801, 880, 852]]<|/det|>
+Reply A: We accept the suggestion to suppress this section in the revised manuscript. We leave this section in the "Method" to emphasize how the topological field theory is modified when time reversal symmetry and an additional mirror symmetry are present.  
+
+<|ref|>text<|/ref|><|det|>[[118, 870, 880, 904]]<|/det|>
+e) The notation for the 4x4 matrices used to in subsection "Mirror plane with time-reversal symmetry breaking" is not consistent — \tauau and \sigmaigma are interchanged below Eq (1).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 85, 878, 119]]<|/det|>
+Reply A: We have updated the notation for the \(4 \times 4\) matrices throughout the manuscript for consistency.  
+
+<|ref|>text<|/ref|><|det|>[[118, 137, 537, 154]]<|/det|>
+f) What is meant by "magnetization" below Eq (1)?  
+
+<|ref|>text<|/ref|><|det|>[[118, 155, 879, 209]]<|/det|>
+Reply A: In the Hamiltonian \(H_{x}\) , a symmetry broken term appears near the mirror plane, which can be cancelled with the term in \(H_{- x}\) . The word "magnetization" may not be appropriate, it contains two symmetry broken fields.  
+
+<|ref|>text<|/ref|><|det|>[[118, 226, 879, 261]]<|/det|>
+g) Below Eq (2) the authors write "This is distinct from the conventional effective model for the surface states which is only ...". What do they mean?  
+
+<|ref|>text<|/ref|><|det|>[[118, 262, 880, 366]]<|/det|>
+Reply A: The conventional effective model for the surface states is only lower energy part of the model in Eq. (8) (Eq. (2) in the original version). It does not contain the symmetry broken part for higher energy. We have added a new section titled "Gapless Dirac Cones with Parity Symmetry Breaking" which includes an expanded discussion and a schematic diagram to clarify the distinctions between the conventional effective model for surface states and our theory.  
+
+<|ref|>text<|/ref|><|det|>[[118, 382, 879, 418]]<|/det|>
+h) The notations for position and momentum space coordinates should be made uniform — while bold \(r\) \perp perp is used for the former, only bold \(k\) is used for the latter.  
+
+<|ref|>text<|/ref|><|det|>[[118, 418, 657, 435]]<|/det|>
+Reply A: We used the unified notation as the reviewer suggested.  
+
+<|ref|>text<|/ref|><|det|>[[118, 452, 430, 469]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 487, 880, 540]]<|/det|>
+Remark B1: The manuscript by Fu et al presents a theory proposal of half quantum mirror Hall effect. The concept is quite novel and can be of interest. However, before I can recommend publication, I have the following concerns that the authors should address  
+
+<|ref|>text<|/ref|><|det|>[[118, 540, 713, 557]]<|/det|>
+Reply B1: We would like to thank the referee for the positive comments.  
+
+<|ref|>text<|/ref|><|det|>[[118, 574, 880, 625]]<|/det|>
+Remark B2: 1. Unclear presentation and lack of schematic illustrations: The presentation is in general not clear. There is significant disconnection between the general introduction, the mathematical derivation and the figures. Below I give some examples:  
+
+<|ref|>text<|/ref|><|det|>[[118, 626, 880, 749]]<|/det|>
+a. The authors emphasized the \(H_{-x}\) , which seems to be a Hamiltonian with a given mirror eigenvalue in their math derivations. However, in their figures, it is not clear how to understand \(H_{-x}\) at all. People typically think about a topological insulator with one Dirac cone at the top surface and another Dirac cone at the bottom surface. But I am guessing that the bands described by \(H_{-x}\) does not come from a single surface. Instead, the bands described by \(H_{-x}\) has contribution from both surface Dirac cones. Is that correct? Can the authors make some figure to label the bands of \(H_{-x}\) and \(H_{-x}\) ?  
+
+<|ref|>text<|/ref|><|det|>[[118, 750, 880, 887]]<|/det|>
+Reply B2: We replotted Fig. 1 to illustrate how the Hamiltonian with the mirror symmetry decomposes into two separate sub- Hamiltonians, in which the states possess even and odd parity, respectively. To be clearer, we add the schematic of the states with even and odd parity and the corresponding wave function distribution. Then we further illustrate how the sub- Hamiltonian with a thickness \(Lz\) can be equivalently folded into a Hamiltonian with a thickness \(Lz / 2\) by utilizing the projected symmetric field operators, which is equivalent to a semimagnetic topological insulator film with a magnetic field only in one side. Hopefully the new figure is clearer.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 878, 138]]<|/det|>
+Remark B3: b. Similarly, the schematic in Fig. 3a is confusing and lacks critical information. E.g., where is the mirror plane respect to the sample? I am guessing that the mirror plane is the horizontal \(x - y\) plane at the center of the sample, is that correct?  
+
+<|ref|>text<|/ref|><|det|>[[118, 154, 877, 242]]<|/det|>
+c. The authors constantly make analogy to the spin Hall effect. In the spin Hall effect, there is a very nice schematic, where as a current flows along \(x\) , the electrons with opposite spins bent toward the \(+y\) and \(-y\) directions, respectively. In this way, there is no net electrical Hall voltage but there is a spin voltage along the \(y\) direction. Can the authors make a similar plot to help people understand the mirror Hall effect they propose?  
+
+<|ref|>text<|/ref|><|det|>[[118, 241, 880, 346]]<|/det|>
+Reply B3: Here we response b and c together. We accept the reviewer's suggestion to replot Figure 3. We insert Fig. 3a to illustrate the mirror Hall effect induced by the electric current. The mirror plane is normal to the \(z\) axis and located in the middle of the film. An electric current can induce the charge carriers with even parity deflects to one side and the charge carriers with odd parity deflects to the opposite side. There is not net Hall current, but there is a "mirror parity" voltage, which is very similar to the spin Hall effect.  
+
+<|ref|>text<|/ref|><|det|>[[117, 362, 881, 660]]<|/det|>
+Remark B4: d. It is further confusing why the mirror Hall effect (or the spin Hall effect) can be directly measured by the two- probe conductance. The authors made some description in the text that is very difficult for readers to understand: "The charge current contains two parts \(Jc, y = Jss c, y + JmH c, y\) : the first term \(Jss c, y\) comes from the conducting surface states and the second term \(JmH c, y \propto \tan 2\Phi m\) arises from the spatial accumulation of the mirror polarization density and is an effect due to the existence of the mirror Hall effect. This mirror Hall mediated charge transport can be understood as follows: the electric field first induces a mirror charge accumulation on the boundary via the mirror Hall effect and then is converted into the charge current along the electric field via the inverse mirror Hall effect". Some graphics will help a lot Reply B: First if the anomalous Hall effect exists, the two- probe conductance includes a correction part attributed by the Hall conductance as well as the longitudinal conductance, \(\sigma_{two} = \sigma_{xx} + \frac{|\sigma_{xy}|^2}{\sigma_{yy}}\) if \(j_y = 0\) and the longitudinal conductance is nonzero by means of the relation between the electric current and electric field. In quantum anomalous Hall effect and quantum spin Hall effect, they are insulating phases, which longitudinal conductance is zero. However, it is a two- probe conductance is \(e^2 /h\) for quantum anomalous Hall effect, and \(2e^2 /h\) for quantum spin Hall effect. We add more detailed discussion in the main context.  
+
+<|ref|>text<|/ref|><|det|>[[118, 676, 880, 799]]<|/det|>
+Remark B4 2. What is the role of nontrivial topology here? The authors constantly make analogy to the spin Hall effect. However, spin Hall effect is widely studied in heavy metals such as Ta, W. However, in the present work, the authors consider a topological insulator and seems to be claiming some quantized response. It is unclear what kind of transport is quantized? E.g., even in quantum spin Hall, the edge conductance is famously non- quantized due to inelastic scattering. No discussion of that kind is provided. What if I have a metal that has \(M_{-}z\) mirror symmetry? Do I have mirror Hall effect (just not quantized)?  
+
+<|ref|>text<|/ref|><|det|>[[118, 799, 880, 902]]<|/det|>
+Reply B4: The half quantum mirror Hall effect is similar to the spin Hall effect as both of them are metallic phases, but its mirror Hall conductance is quantized, only one half of the quantum spin Hall conductance. The role of nontrivial topology reveals from the half quantized Hall conductance of a single gapless Dirac cone on a lattice. This is opposite to the graphene which hosts a pair of the gapless Dirac cone (if we ignore the spin degree of freedom). Existence of a single gapless Dirac cone on a lattice has to break the time reversal symmetry as required
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 880, 323]]<|/det|>
+by the fermion doubling theorem. As shown in Fig. 2(c), when chemical potential is situated within the bulk band gap \((|\mu |< m_0)\) , intersecting solely with the gapless Dirac cone, the mirror Hall conductivity is quantized at \(\sigma_{xy}^{M_z} = e^2 /h\) . When chemical potential shifts into the bulk states, the mirror Hall effect persists, though it derives from the quantized value due to the contributions from partially filled trivial gapped band. For our spin 1/2 system, the horizontal mirror operator \(\hat{M}_z\) anticommutes with the vertical mirror operator (or parity operator in 2D) \(\hat{M}_x\) , i.e., \(\{\hat{M}_z, \hat{M}_x\} = 0\) . Consequently, each mirror eigensector of \(\hat{M}_z\) violates the parity symmetry \(\hat{M}_x\) explicitly, leading to nonzero Hall effect for each sector, which in turn results in a nonzero mirror Hall effect. The relation \(\{\hat{M}_z, \hat{M}_x\} = 0\) serves as a sufficient but not necessary condition for the existence of a nonzero mirror Hall effect in a conventional metal that has Mz symmetry. As for a mirror symmetric metal, it is possible to have a mirror Hall effect based on the picture of Fig. 1 for the topologically trivial case of the lattice model we use. We are grateful that the reviewer attracted us to this point.  
+
+<|ref|>text<|/ref|><|det|>[[117, 339, 880, 427]]<|/det|>
+Remark B5 In general, I think the paper may contain some interesting concept that deserves to be published in a top journal, however, the current manuscript is written in a way that is very hard for people to judge. The authors also did not show great care and desire to make the presentation readable and understandable. I cannot recommend the publication in the present form but is happy to consider a significantly improved version.  
+
+<|ref|>text<|/ref|><|det|>[[117, 427, 880, 583]]<|/det|>
+Reply B5: Thank you for taking the time to review our manuscript and for providing valuable feedback. We appreciate your effort in assessing our work and are grateful for your constructive comments. We have carefully considered your suggestions and have made significant improvements to address the issues you raised. Specifically, we have focused on enhancing the clarity and readability of our paper. We have taken your feedback regarding the presentation of our figures into account and have replotted them to ensure better visual representation of our results. Furthermore, we have expanded the discussions in our manuscript to provide a more comprehensive analysis of our findings. We believe these additions have significantly enhanced the overall quality of our work.  
+
+<|ref|>text<|/ref|><|det|>[[119, 617, 429, 633]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[117, 652, 880, 792]]<|/det|>
+Remark C1: In this paper, the authors propose a novel topological phenomenon dubbed the half- quantum mirror Hall effect. They consider a strong topological insulator film with mirror reflection symmetry. Using mirror symmetry, they split a pair of surface Dirac modes on the top and bottom surfaces into a single one and derive a half- quantized Hall conductance on each mirror subsector. Furthermore, they also reveal that the half- quantized mirror Hall conductance results in a net current due to the accumulation of a mirror charge and the inverse mirror Hall effect. The idea of this paper looks novel and exciting. However, the following points should be clarified before the decision on the recommendation.  
+
+<|ref|>text<|/ref|><|det|>[[117, 792, 879, 826]]<|/det|>
+Reply C1: We would like to thank the referee for the positive comments and constructive suggestions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 844, 880, 896]]<|/det|>
+Remark C2: 1. Whereas the authors assume an exact mirror- reflection symmetry, actual materials do not support it due to the imperfection of the crystal structures. The author should clarify how the imperfection affects the obtained result.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 880, 310]]<|/det|>
+Reply C2: We appreciate the referee's insightful question regarding the imperfection of the crystal structure. Indeed, in real materials, exact mirror- reflection symmetry may not be achievable due to imperfections in the lattice structure. However, we believe that if the imperfection is not significantly strong, we can treat it as a perturbation that breaks the symmetry. This perturbation introduces inter- mirror scattering, and the scattering length described in Eqs. (11) and (12) (originally Eqs. (4) and (5)) reflects the strength of the imperfection. Although the exact mirror- reflection symmetry assumption may not hold in the presence of strong imperfections, we argue that for weak imperfections, the effect can still be observed. To address this concern, we have added a dedicated discussion in the revised manuscript (section "Discussion and conclusion") to further elaborate on the impact of lattice structure imperfections and the robustness of our findings. We believe that this clarification will provide a more comprehensive understanding of the phenomenon and its relevance to real materials.  
+
+<|ref|>text<|/ref|><|det|>[[118, 328, 880, 380]]<|/det|>
+Remark C3: 2. It needs to be clarified that the half- quantized value of the Hall conductance can be concluded through the two- terminal measurement. I understand that a non- zero mirror Hall conductance can be observed, but I do not know how to determine its value.  
+
+<|ref|>text<|/ref|><|det|>[[118, 381, 880, 658]]<|/det|>
+Reply C3: We appreciate the referee's question regarding the determination of the value of the mirror Hall conductance. If the exact mirror- reflection symmetry nearly holds (i.e., \(l_{MZ} \to \infty\) ), to deduce the value of the mirror Hall conductance, we propose a two- step measurement approach. In the first step, we suggest performing a conventional six- probe measurement to measure the longitudinal conductance of the system. This measurement provides valuable information about the overall conductivity of the material. In the second step, we propose a two- probe measurement to determine the mirror Hall conductance. By comparing the two- probe conductance with the longitudinal conductance obtained from the first step, we can calculate the difference. This difference corresponds to the mirror Hall conductance, allowing us to determine its value. We acknowledge that this two- step measurement approach is necessary to accurately determine the mirror Hall conductance. If the mirror- reflection symmetry is weakly broken (with \(l_{MZ}\) being finite), we propose utilizing a multi- terminal measurement to extract the mirror Hall conductance of the system. We have included a detailed explanation of a measurement procedure in the revised manuscript to provide further clarity on how the value of the mirror Hall conductance can be determined experimentally.  
+
+<|ref|>text<|/ref|><|det|>[[118, 676, 876, 710]]<|/det|>
+Remark C4: 3. The superscript and subscript m should be modified as Mz to avoid confusion with the band index.  
+
+<|ref|>text<|/ref|><|det|>[[118, 711, 879, 814]]<|/det|>
+Reply C4: To address this concern, we have carefully reviewed the manuscript and made the necessary modifications to clarify the notation. We have replaced the superscript and subscript "m" with "M_z" to differentiate it from the band index notation. We apologize for any confusion that may have arisen due to the previous notation, and we appreciate the referee's feedback in improving the clarity of our manuscript. Thank you for bringing this issue to our attention, and we hope that these revisions will address your concerns.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 91, 280, 106]]<|/det|>
+Reviewers' comments:  
+
+<|ref|>text<|/ref|><|det|>[[115, 147, 392, 163]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[113, 202, 874, 514]]<|/det|>
+The authors have responded to all points raised in my report. Their response has clarified several key aspects of the paper; in particular, the interplay between mirror symmetry and surface states. While they have identified an interesting role of mirror symmetry in the topological response from surface states, I remain unconvinced by the extent of progress reported in this paper, given the large body of literature, including recent works by (at least) a subset of the authors, that have already explored ideas that overlap with those explored here. In my view, the key new result here is the generalization of the notion of the mirror- Chern number (generally used as a bulk invariant) to half- quantized Hall conductivity from surface states — individually, the notions of mirror- Chern number and half- quantized Hall response from isolated 2D Weyl/Dirac cones are not new. Further, the fact that the conclusions rest on the analyses of a 4- band model is a concern, given the maturity of the field. A minimal- model based calculation, while important and the essential first step (the emphasis on obtaining analytical results is appreciated), does not unambiguously establish the authors' claim that this work represents "an important advancement in our understanding of the role of symmetry in the generation of novel electronic phenomena". I would urge the authors to numerically demonstrate their key results in more realistic models [eg. the Hamiltonian in Eq. (52) of Phys. Rev. B 82, 045122 (2010)] or ab initio data. While the former type of models are important for comparing with experimental data in topological insulators, the latter is the best demonstration of any principle borne out of minimal models.  
+
+<|ref|>text<|/ref|><|det|>[[115, 553, 536, 569]]<|/det|>
+Some specific comments following the authors' response:  
+
+<|ref|>text<|/ref|><|det|>[[114, 608, 880, 756]]<|/det|>
+1. I do not yet see why the authors emphasize the lattice-based calculations for surface states. On mathematical grounds, the fact that generally surface states exist over a finite sub-region of the surface Brillouin zone can be demonstrated within a k.p model [see for example, section V of Phys. Rev. B 82, 045122 (2010)]. On physical grounds, the half-quantized response should be accessible to a k.p model because the anomalous Hall response for a 2D metal boils down to a line-integral over the Fermi surface. Beyond concerns over quantitative agreements, the only other physically interesting reason for favoring lattice-based models appears to be tied to the discussion around the new Fig. 2 (d). This aspect has remained unclear to me, even after multiple re-reading of the text.  
+
+<|ref|>text<|/ref|><|det|>[[114, 766, 881, 894]]<|/det|>
+2. In their response the authors agree with the finite extent of the surface states, but this is not reflected by the Hamiltonian in Eq (8). If the states do not exist beyond a particular region of the surface Brillouin zone, why does the Hamiltonian exist? Moreover, terms proportional to \(\S \backslash \mathrm{tilde} \backslash \mathrm{sigma} \_ z \S\) [according to the convention of Eq (8)] appears due to the so-called hexagonal warping, and is an important feature of surface states of topological insulators, as discussed in Phys. Rev. Lett. 103, 266801 (2009). In light of the above two points, I would conclude Eq (8) is incorrect, which casts doubt over any analysis following Eq (8).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 117, 876, 247]]<|/det|>
+3. Regarding "Reply A4": What I understood from this response is that there are bulk states (states that are not exponentially localized on the two surfaces of the slab) that contribute to Hall conductivity if the overall filling is sufficiently large. While I agree with this statement, I find this distinction can be better utilized in making a physically important statement regarding the thickness dependence of the mirror-Hall response. Otherwise, this is a secondary point. Moreover, the language used to describe additional states contributing to the Hall response needs to be sharpened (also see my concern in point 2 above while interpreting the numerical results in this context).  
+
+<|ref|>text<|/ref|><|det|>[[114, 285, 882, 377]]<|/det|>
+4. While I appreciate the authors' contrasting their results with that obtained by Creutz & Horvath (CH), I would like to emphasize that the CH method itself is more general than its initial implementation. This is clearly demonstrated in the paper by König et al. noted in my previous report. Therefore, I would suggest taking both papers under consideration while contrasting between the two methods above the section "Quantum mirror Hall conductance".  
+
+<|ref|>text<|/ref|><|det|>[[114, 414, 882, 470]]<|/det|>
+5. Overall, I still find the narrative is unnecessarily convoluted. Even after the substantial revision of the text in the present version, the key ideas developed here are obfuscated by a lot of details that appear to be secondary. Is it impossible to demonstrate the main message with an 4-band k.p model?  
+
+<|ref|>text<|/ref|><|det|>[[113, 481, 825, 500]]<|/det|>
+If so, then this is an important aspect of this work, and should be demonstrated more directly.  
+
+<|ref|>text<|/ref|><|det|>[[114, 510, 876, 566]]<|/det|>
+If not, then the lattice-based calculations add quantitative details, and only the final outcome is relevant to the main text. Moreover, in this case, it is unclear whether a more realistic 8-band k.p model is less useful than using a lattice-version of a 4-band model.  
+
+<|ref|>text<|/ref|><|det|>[[113, 604, 876, 641]]<|/det|>
+6. While I have not noted this in my previous report, I find the word "discovery" in the abstract to be too strong for a theoretical proposal, whose applicability to actual materials is yet to be demonstrated.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 879, 201]]<|/det|>
+Regarding C2: This is an important point, and underscores my suggestion above on working with ab initio data to unambiguously demonstrate the existence of the mirror Hall effect in actual materials. The authors have responded to this point qualitatively. My concern here is that if the mirror symmetry is broken (weakly or strongly), then, in principle, the two Dirac cones can hybridize and mutually gap out. This will revert the system to the very regime the authors want to avoid — topological response from gapped surface states (see "Reply A1").  
+
+<|ref|>text<|/ref|><|det|>[[114, 238, 883, 347]]<|/det|>
+Regarding C3: This is also an important issue, since measuring lattice- symmetry protected topological response is not obvious. I am unaware of any existing work where a lattice- symmetry protected transport response has been measured. The authors seem to recognize this challenge, and have made an effort to device a measurement protocol, but this may require/invite further scrutiny. Therefore, it would be helpful to cite other similar proposals (i.e. measurement of lattice- symmetry protected response) in this context.  
+
+<|ref|>text<|/ref|><|det|>[[115, 416, 392, 432]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 471, 820, 507]]<|/det|>
+I have read the authors' responses. I think that they have thoroughly address all the comments. I recommend the work for publication.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 103, 600, 122]]<|/det|>
+Reply to the reviewer #1's Remarks to the Authors.  
+
+<|ref|>text<|/ref|><|det|>[[117, 140, 881, 563]]<|/det|>
+Remark 1: The authors have responded to all points raised in my report. Their response has clarified several key aspects of the paper; in particular, the interplay between mirror symmetry and surface states. While they have identified an interesting role of mirror symmetry in the topological response from surface states, I remain unconvinced by the extent of progress reported in this paper, given the large body of literature, including recent works by (at least) a subset of the authors, that have already explored ideas that overlap with those explored here. In my view, the key new result here is the generalization of the notion of the mirror- Chern number (generally used as a bulk invariant) to half- quantized Hall conductivity from surface states — individually, the notions of mirror- Chern number and half- quantized Hall response from isolated 2D Weyl/Dirac cones are not new. Further, the fact that the conclusions rest on the analyses of a 4- band model is a concern, given the maturity of the field. A minimal- model based calculation, while important and the essential first step (the emphasis on obtaining analytical results is appreciated), does not unambiguously establish the authors' claim that this work represents "an important advancement in our understanding of the role of symmetry in the generation of novel electronic phenomena". I would urge the authors to numerically demonstrate their key results in more realistic models [eg. the Hamiltonian in Eq. (52) of Phys. Rev. B 82, 045122 (2010)] or ab initio data. While the former type of models are important for comparing with experimental data in topological insulators, the latter is the best demonstration of any principle borne out of minimal models.  
+
+<|ref|>text<|/ref|><|det|>[[117, 581, 881, 887]]<|/det|>
+Reply 1: Our findings represent more than a mere generalization of previous results. So far it is the first metallic system to exhibit the quantum anomaly under a time- reversal symmetry. Firstly, in an insulating system, the mirror Chern number must be integer. However, in a metallic system, in which the surface states are gapless, it is the first example to exhibit half- quantization associated with a pair of gapless Dirac cones. To our best knowledge, there are currently no other known systems that exhibit half- quantization while maintaining time- reversal symmetry. Secondly, a lattice itself provides a gauge- invariant regularization, such as in the case of a strong topological insulator film or graphene. Our findings demonstrate that when an additional intrinsic symmetry is present to separate massless surface Dirac bands (including the surface states near Gamma point and other states within the bulk gap) across the entire energy spectrum, quantum anomalies can emerge and induce measurable physical effects. Therefore, our findings not only establish quantized topological invariants for metals but also revive the occurrence of quantum anomalies in systems that are usually expected to be free from them.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 880, 276]]<|/det|>
+To convince the referee, we accept the referee's suggestion to do numerical calculation based on realistic tight- binding models with the same crystalline symmetries as in \(\mathrm{Bi}_2\mathrm{Se}_3\) and \(\mathrm{Bi}_2\mathrm{Te}_3\) , although the lattice construction is really complicated. Each unit cell contains six sites and each site has four orbitals. Fortunately, the band spectra with the warping effect that the reviewer was concerning are well reproduced. The main findings, such as the quantized mirror Hall conductance, are reproduced as we expected. It is important to emphasize that our main results are derived from a symmetry analysis and are not reliant on specific model details. The use of a simplified model allows for a more concise presentation of the findings.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 295, 225, 313]]<|/det|>
+## Remark 2:  
+
+<|ref|>text<|/ref|><|det|>[[117, 315, 881, 504]]<|/det|>
+1. I do not yet see why the authors emphasize the lattice-based calculations for surface states. On mathematical grounds, the fact that generally surface states exist over a finite sub-region of the surface Brillouin zone can be demonstrated within a k.p model [see for example, section V of Phys. Rev. B 82, 045122 (2010)]. On physical grounds, the half-quantized response should be accessible to a k.p model because the anomalous Hall response for a 2D metal boils down to a line-integral over the Fermi surface. Beyond concerns over quantitative agreements, the only other physically interesting reason for favoring lattice-based models appears to be tied to the discussion around the new Fig. 2 (d). This aspect has remained unclear to me, even after multiple re-reading of the text.  
+
+<|ref|>text<|/ref|><|det|>[[117, 505, 880, 677]]<|/det|>
+Reply 2: The reviewer said, "the half-quantized response should be accessible to a k.p model because the anomalous Hall response for a 2D metal boils down to a line-integral over the Fermi surface". This statement is not always true: it is only valid for a k.p model that can be placed on a lattice without introducing a doubling problem. A famous example is the linear Dirac cone cannot be realized on a lattice if no additional symmetry breaking term is introduced. If the symmetry is not broken, the Hall conductivity is zero. To realize a gapless Dirac cone we have to breaking the symmetry as a whole, which is actually a central issue in the lattice gauge theory.  
+
+<|ref|>text<|/ref|><|det|>[[117, 678, 880, 850]]<|/det|>
+In a k.p model there exists a singularity at the point \(k \to +\infty\) , we cannot simply use the Stokes' theorem to transfer the area integral to line integral around the Fermi surface in the TKNN formula because the contribution around the singularity is ignored. If we insist on using the formula with line integral by ignoring the singularity, and the \(\pi\) Berry phase in the k.p model in which \(\pm \pi\) are equivalent, it can utmost give rise one half of the quantum Hall conductance, but its sign has to be determined by the filled states far away from the Fermi surface. For a comprehensive model example supporting this statement, please consult the Appendix.  
+
+<|ref|>text<|/ref|><|det|>[[118, 869, 878, 907]]<|/det|>
+There is no similar issue on a lattice model, which can provides an ambiguous answer to the issue. In another word, the lattice model provides a natural
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 879, 182]]<|/det|>
+regularization of low- energy effective model on a finite Brillouin zone. That is why we rely on calculations based on the lattice model to study surface states. This approach helps to address the ambiguity inherent in the issue and provides clearer and more reliable results. We have already addressed half quantized Hall conductivity of single gapless Dirac cone on a lattice in our previous works.  
+
+<|ref|>text<|/ref|><|det|>[[118, 202, 879, 317]]<|/det|>
+There is an example to illustrate the shortcoming of the k.p model. The massive Dirac model in the k.p approximation gives a one- half quantized Hall conductance, which is impossible to be realized on a lattice according to the TKNN theorem that a fully- filled gapped band always gives an integer quantum Hall conductance. So, we should be very careful when using the k.p model to explore the physics related to topology.  
+
+<|ref|>text<|/ref|><|det|>[[117, 335, 880, 508]]<|/det|>
+Remark 3: 2. In their response the authors agree with the finite extent of the surface states, but this is not reflected by the Hamiltonian in Eq (8). If the states do not exist beyond a particular region of the surface Brillouin zone, why does the Hamiltonian exist? Moreover, terms proportional to \(S\) tilde \(\backslash \mathrm{sigma\_z}\$ [according to the convention of Eq (8)] appears due to the so- called hexagonal warping, and is an important feature of surface states of topological insulators, as discussed in Phys. Rev. Lett. 103, 266801 (2009). In light of the above two points, I would conclude Eq (8) is incorrect, which casts doubt over any analysis following Eq (8).  
+
+<|ref|>text<|/ref|><|det|>[[117, 508, 880, 777]]<|/det|>
+Reply 3: The Hamiltonian in Eq. 8 contains either the surface states \((\Delta (k) = 0)\) or the bulk states \((\Delta (k) \neq 0)\) , which forms a whole band in the first Brillouin zone. The \(\sigma_z\) term in Eq. 8 breaks the parity symmetry and time reversal symmetry, but the combination of the pair of gapless Dirac cones respect the mirror symmetry and time reversal symmetry. This is opposite to the hexagonal warping term introduced by Liang Fu, in which the term is cubic in momentum and does not breaking time reversal symmetry. The warping effect appears in the systems with the threefold rotation symmetry. Although it changes the shape of the dispersion spectra, it does not change the band inversion or topology of the band. The appearance of the warping term does not affect the quantization of the mirror Hall conductivity in the mirror- symmetric \(\mathrm{Bi}_2\mathrm{Te}_3\) with twin boundary. Use the reviewer's word, this is a secondary effect. We use the model with threefold rotation in newly added section "Material candidates". The warping effect is successfully reproduced, and the main results remain unchanged.  
+
+<|ref|>text<|/ref|><|det|>[[118, 796, 878, 853]]<|/det|>
+Eq. 8 is an analytical result from a three- dimensional lattice model for a strong topological insulator, which has fourfold (NOT threefold) rotation symmetry. That's why there is no warping effect as in Fu's paper.  
+
+<|ref|>text<|/ref|><|det|>[[118, 872, 878, 910]]<|/det|>
+Remark 4: 3. Regarding "Reply A4": What I understood from this response is that there are bulk states (states that are not exponentially localized on the two
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 880, 219]]<|/det|>
+surfaces of the slab) that contribute to Hall conductivity if the overall filling is sufficiently large. While I agree with this statement, I find this distinction can be better utilized in making a physically important statement regarding the thickness dependence of the mirror- Hall response. Otherwise, this is a secondary point. Moreover, the language used to describe additional states contributing to the Hall response needs to be sharpened (also see my concern in point 2 above while interpreting the numerical results in this context).  
+
+<|ref|>text<|/ref|><|det|>[[117, 219, 882, 410]]<|/det|>
+Reply 4: We appreciate that the reviewer accepted our explanation. In this work we focus on the quantized mirror Hall conductivity within the bulk gap, which is also the gap of the states of \(\mathrm{n} = 2\) . It would be interesting question for the mirror Hall conductivity out of the gap. When the chemical potential sweeps more bands, the mirror Hall conductance is no longer quantized. The thickness of a sample will change the density of states around the Fermi level, and change the mirror Hall conductivity. However the bulk gap remains unchanged. We agree that the thickness dependence should be an interesting issue, deserving further investigation in the future. We accept the suggestion by adding a few sentences related to Fig. 2c.  
+
+<|ref|>text<|/ref|><|det|>[[117, 428, 880, 907]]<|/det|>
+Remark 5: 4. While I appreciate the authors' contrasting their results with that obtained by Creutz & Horvath (CH), I would like to emphasize that the CH method itself is more general than its initial implementation. This is clearly demonstrated in the paper by König et al. noted in my previous report. Therefore, I would suggest taking both papers under consideration while contrasting between the two methods above the section "Quantum mirror Hall conductance". Reply 5: Thank the referee' for the suggestion. As we pointed out in the last reply, the CH method is incomplete to have the final solution. For example, the method cannot give the finite size effect in quantum spin Hall effect. The approach proposed by Creutz and Horvath (CH) has been employed to address the edge states in the quantum spin Hall system as discussed in Konig et al.'s paper. In that paper, Konig et al. utilize the similar model as us (shown from Eqs 12, 13, and 14). However, they also only provide an analytical solution for the edge states, as demonstrated in Eqs. 21 and 22 as CH's work. This can also be observed in the lower panel of Fig. 3. The bulk states (represented by blue lines) are only included to reproduce the appearance of exact diagonalization results that the number of bands at \(\mathrm{k} = 0\) and \(\mathrm{k} = \mathrm{pi}\) are not equal. Therefore, both the work of Creutz and Horvath (CH) and Konig et al.'s paper only provide solutions for the surface states, which are limited to a specific region of the surface Brillouin zone. As shown in Fig. S1 and Fig. S2 in the supplementary material of our manuscript, our analytical solutions, which are founded on the tight- binding model, exhibit remarkable consistency throughout the entirety of the Brillouin zone, also encompassing the bulk states. However, the mirror Hall effect originates from high- energy bulk states with mirror Berry curvature, rather than the surface states, as depicted in Fig. 5e. This mirror Hall effect can
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 864, 160]]<|/det|>
+be captured by our solution in Eq. 8, which includes contributions from both surface states and bulk states. This has already gone beyond the CH and Konig et al's papers. The solution of the surface states is not the key point in the present work.  
+
+<|ref|>text<|/ref|><|det|>[[118, 180, 879, 258]]<|/det|>
+Remark 6: 5. Overall, I still find the narrative is unnecessarily convoluted. Even after the substantial revision of the text in the present version, the key ideas developed here are obfuscated by a lot of details that appear to be secondary. Is it impossible to demonstrate the main message with an 4- band k.p model?  
+
+<|ref|>text<|/ref|><|det|>[[118, 277, 879, 315]]<|/det|>
+If so, then this is an important aspect of this work, and should be demonstrated more directly.  
+
+<|ref|>text<|/ref|><|det|>[[118, 334, 879, 410]]<|/det|>
+If not, then the lattice- based calculations add quantitative details, and only the final outcome is relevant to the main text. Moreover, in this case, it is unclear whether a more realistic 8- band k.p model is less useful than using a lattice- version of a 4- band model.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 429, 202, 448]]<|/det|>
+## Reply 6:  
+
+<|ref|>text<|/ref|><|det|>[[118, 449, 879, 581]]<|/det|>
+We would like to emphasize that our principal findings are broadly applicable, rooted in the topological characteristics and symmetries of the system, rather than dependent on the particulars of any single model. This universality has been thoroughly addressed in the section titled "The Mirror Symmetry and Single Gapless Dirac Cones." The subsequent discussion, which employs a detailed model, serves merely to provide a concrete example that validates our theoretical framework.  
+
+<|ref|>text<|/ref|><|det|>[[118, 582, 879, 658]]<|/det|>
+While the effect is straightforward from topology band structure and symmetry, it becomes a key issue how to measure it in experiment. Part of the manuscript is devoted to exploring transport signature of the effect. We think this is necessary to predict a new physical effect.  
+
+<|ref|>text<|/ref|><|det|>[[118, 677, 879, 829]]<|/det|>
+It is possible to demonstrate the main message with the k.p model if the model can be realized on a lattice without introducing the doubling problem. The lattice calculation here aims to avoid the shortcoming of the k.p theory, as we mentioned in Reply 1. The 4- and 8- band k.p models may make the band dispersions closer to the DFT calculation. The most important point here is whether the model can capture the band inversion for nontrivial topology, and whether the system has the mirror symmetry. As clarified in Reply 2, our preference for utilizing the tight- binding model is to yield an unambiguous result for the problem at hand.  
+
+<|ref|>text<|/ref|><|det|>[[118, 849, 879, 906]]<|/det|>
+In the newly added section "Material candidates", we use the tight- binding approximation based on the four- band k.p model. The numerical results fully support our point of view.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 102, 880, 161]]<|/det|>
+Remark 7: 6. While I have not noted this in my previous report, I find the word "discovery" in the abstract to be too strong for a theoretical proposal, whose applicability to actual materials is yet to be demonstrated.  
+
+<|ref|>text<|/ref|><|det|>[[118, 161, 825, 181]]<|/det|>
+Reply 7: We have replaced the term "discovery" with "finding" in the text.  
+
+<|ref|>text<|/ref|><|det|>[[118, 199, 881, 257]]<|/det|>
+Remark 8: I was asked to comment on the response to the 3rd referee's comments. I will limit myself to comments \(C2\) & \(C3\) , since the other two are rather straightforward which the authors have addressed.  
+
+<|ref|>text<|/ref|><|det|>[[117, 275, 880, 409]]<|/det|>
+Regarding \(C2\) : This is an important point, and underscores my suggestion above on working with ab initio data to unambiguously demonstrate the existence of the mirror Hall effect in actual materials. The authors have responded to this point qualitatively. My concern here is that if the mirror symmetry is broken (weakly or strongly), then, in principle, the two Dirac cones can hybridize and mutually gap out. This will revert the system to the very regime the authors want to avoid — topological response from gapped surface states (see "Reply A1").  
+
+<|ref|>text<|/ref|><|det|>[[117, 409, 880, 658]]<|/det|>
+Reply 8: The opening of the gap in the Dirac cones can occur through two mechanisms: i) magnetic doping on the surface, which breaks the time- reversal symmetry, or ii) the coupling between two surface Dirac cones induced by finite size effects. Actually, these two mechanisms do not necessarily break mirror symmetry. Magnetic doping on both surfaces in an equal manner preserves mirror symmetry, and reducing the thickness of a mirror- symmetric film where two Dirac cones hybridize and form a gap, also does not break mirror symmetry. If mirror symmetry is maintained, the mirror eigenvalue remains a valid quantum number, and the mirror Hall conductivity can still be defined. However, in this scenario, the Hall conductivity for each mirror sector becomes an integer (the mirror Hall conductivity is also an integer) when the chemical potential is located within the gap of the Dirac cones, as discussed in the "Discussion and Conclusion" section of the previous revised version.  
+
+<|ref|>text<|/ref|><|det|>[[117, 676, 880, 910]]<|/det|>
+In epitaxially grown films, the charge transfer from the substrate will induce a potential difference between the two surfaces of the topological insulator which breaks the mirror symmetry explicitly. The substrate- induced potential can be modelled as \(H_{V} = \sum_{k,z} \psi_{k,z}^{+} V(z) \psi_{k,z}\) where \(V(z)\) is nonzero at several bottom layers and nearly uniform in x- y plane. As depicted in the left panel of Figure R2, the presence of a potential breaks the degeneracy between the top and bottom surface states, resulting in the explicit breaking of mirror symmetry. However, in the right panel of Figure R2, it can be observed that when the chemical potential is situated within the bulk band gap (blue region), the mirror Hall conductivity remains quantized at - 1. Indeed, the mirror Hall effect mainly originates from high- energy states primarily located within the bulk of the material. The potential on the surface does not significantly affect these states.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[120, 100, 872, 360]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 365, 880, 506]]<|/det|>
+<center>Fig. R2:(left) The bandstructure of Bi2Se3 along [001] direction with a twin boundary. The color of the dots represents the mean position of the states along the [001] direction.A gate voltage of \(\mathrm{V = 0.1eV}\) is applied to the bottom surface. The remaining parameters are the same as in the main text. (right) The mirror Hall conductance \(\sigma_{xy}^{M2}\) as a function of \(\mu\) . The red dashed line serves as a visual guide to indicate the value of -1. The blue shadowed region indicates the gap regions for the bulk states. </center>  
+
+<|ref|>text<|/ref|><|det|>[[116, 523, 881, 658]]<|/det|>
+Remark 9: Regarding C3: This is also an important issue, since measuring lattice- symmetry protected topological response is not obvious. I am unaware of any existing work where a lattice- symmetry protected transport response has been measured. The authors seem to recognize this challenge, and have made an effort to device a measurement protocol, but this may require/invite further scrutiny. Therefore, it would be helpful to cite other similar proposals (i.e. measurement of lattice- symmetry protected response) in this context.  
+
+<|ref|>text<|/ref|><|det|>[[116, 659, 881, 792]]<|/det|>
+Reply 9: This is also new to the authors. We know this new effect is similar to the spin Hall effect in conductors or semiconductors, but the mirror Hall conductance is only one half of the quantum spin Hall effect in an insulating phase. Maybe the techniques to measure spin Hall effect are helpful for this new effect (Ref. 48- 52). Furthermore, we have included several recent experimental measurements of lattice- symmetry protected transport responses in the section "transport signature".  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 831, 226, 849]]<|/det|>
+## Appendix:  
+
+<|ref|>text<|/ref|><|det|>[[116, 850, 880, 888]]<|/det|>
+To offer a more succinct clarification in response to Reply 2, let's consider two models as examples:  
+
+<|ref|>equation<|/ref|><|det|>[[116, 887, 622, 912]]<|/det|>
+\[\mathrm{Model}1:H_{1} = \nu \big(k_{x}\sigma_{y} - k_{y}\sigma_{x}\big) + \lambda \big(3k_{x}^{2}k_{y} - k_{y}^{3}\big)\sigma_{z}\]
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 85, 472, 111]]<|/det|>
+Model 2: \(H_{2} = H_{1} + b\left(k_{x}^{2} + k_{y}^{2}\right)^{2}\sigma_{z}\)  
+
+<|ref|>text<|/ref|><|det|>[[115, 110, 880, 457]]<|/det|>
+\(H_{1}\) represents the Hamiltonian for topological insulator surface states with a warping effect as in Liang Fu's paper, while \(H_{2}\) is \(H_{1}\) with an additional time- reversal symmetry- breaking term, \(b\left(k_{x}^{2} + k_{y}^{2}\right)^{2}\sigma_{z}\) . Up to the third order in k, \(H_{2}\) is identical to \(H_{1}\) . When chemical potential \(\mu\) is near the Dirac point, the line integral of Berry connection over the Fermi surface \(\mu\) for these two Hamiltonians is \(-\pi\) as \(\mu\) approaches zero, as shown by the red and blue lines in Fig. R1. However, when we evaluate the Hall conductivity for these two models, we find that for model 2, the Hall conductivity is same as Berry phase (only multiplied by a constant factor). However, for Model 1, due to the time- reversal symmetry, the Hall conductivity is zero (blue triangle with dashed line), which conflicts with its Berry phase value on Fermi surface. The reason for the discrepancy in the results is that the relation \(\int d^{2}k\Omega (k)\Theta (\mu -\epsilon (k)) = \oint_{FS}dl\cdot A\) is based on Stokes' theorem, which requires that the integral region be free of singularities. However, in Model 1, the point at \(k\rightarrow \infty\) can be considered a singularity due to the spin texture pointing in different directions as the amplitude angle of \(k\) varies from 0 to \(2\pi\) . This singularity violates the conditions required for the application of Stokes' theorem, resulting in different outcomes for these two models.  
+
+<|ref|>image<|/ref|><|det|>[[170, 460, 545, 686]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[155, 694, 856, 815]]<|/det|>
+<center>Fig. R1: The blue and red solid lines represent the line integral of the Berry connection over the Fermi surface \(\mu\) for these two Hamiltonians. The dashed lines with triangles represent the corresponding Hall conductivities. \(\nu = 1, \lambda = 1, b = 0.2\) . </center>  
+
+<|ref|>text<|/ref|><|det|>[[117, 852, 880, 913]]<|/det|>
+To further analyze the distinction between the models, we can place both k.p models on a lattice using the substitutions \(k_{x} \rightarrow \sin k_{x}\) and \(k_{x}^{2} \rightarrow 2(1 - \cos k_{x})\) :
+
+<--- Page Split --->
+<|ref|>equation<|/ref|><|det|>[[115, 81, 825, 180]]<|/det|>
+\[\mathrm{Model~1:}H_{1}^{l a t t i c e} = v\sin k_{x}\sigma_{y} - v\sin k_{y}\sigma_{x} + \lambda \sin k_{y}(4 - 6\cos k_{x} +\] \[2\cos k_{y})\sigma_{z\] \[\mathrm{Model~2:~}H_{2}^{l a t t i c e} = v\sin k_{x}\sigma_{y} - v\sin k_{y}\sigma_{x} + [\lambda \sin k_{y}(4 - 6\cos k_{x} +\] \[2\cos k_{y}) + b(4 - 2\cos k_{x} - 2\cos k_{y})^{2}]\sigma_{z\]  
+
+<|ref|>text<|/ref|><|det|>[[116, 197, 880, 279]]<|/det|>
+It is easily to check for \(H_{1}^{l a t t i c e}\) lattice, except at the \(\Gamma = (0,0)\) point, there are additional massless Dirac fermions at other time- reversal invariant momenta (TRIMs). However, for \(H_{2}^{l a t t i c e}\) lattice, \(\Gamma = (0,0)\) point is the only point hosting a massless Dirac fermion.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 145, 392, 161]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 202, 870, 275]]<|/det|>
+I appreciate that the authors have demonstrated the robustness of their results for realistic models and they have clarified the distinction between the present work vs. their earlier work [Phys. Rev. B 107, 125153 (2023)]. I was also happy to learn that the authors agree that their main findings can be demonstrated via suitable k.p models.  
+
+<|ref|>text<|/ref|><|det|>[[115, 315, 882, 388]]<|/det|>
+Their response to "Remark 8", however, has confused me. In particular, a comparison of the mirror- Hall conductivity in Fig. 2 vs. Fig. S3 (Fig R2 of the response) seems to suggest that the mirror symmetry is not necessary for the quantized response (somehow, even when mirror symmetry is broken "explicitly", the authors are defining a mirror- Hall conductivity in Fig S31).  
+
+<|ref|>text<|/ref|><|det|>[[115, 398, 867, 490]]<|/det|>
+Further, the authors attribute the quantized response in Fig. S3 to "high energy states primarily located in the bulk". This appears to be in conflict with a central assertion noted in the abstract "... mirror symmetry assigns a unique mirror parity to each Dirac cone, resulting in a half- quantized Hall conductance of \(+ / - e^{\wedge}2 / 2h\) for each cone.". Therefore, in an experiment, if one detects a quantized mirror- Hall conductivity, how would one know its source?  
+
+<|ref|>text<|/ref|><|det|>[[115, 558, 393, 574]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 614, 685, 631]]<|/det|>
+The authors have satisfactorily address all my questions. I support publication.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 102, 409, 119]]<|/det|>
+## Reply to the report of Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[118, 137, 880, 224]]<|/det|>
+Comment: Their response to "Remark 8", however, has confused me. In particular, a comparison of the mirror- Hall conductivity in Fig. 2 vs. Fig. S3 (Fig R2 of the response) seems to suggest that the mirror symmetry is not necessary for the quantized response(somehow, even when mirror symmetry is broken "explicitly", the authors are defining a mirror- Hall conductivity in Fig S3!).  
+
+<|ref|>text<|/ref|><|det|>[[118, 241, 880, 345]]<|/det|>
+Reply: The Reviewer #1 raised an interesting and new question, which is generated in our previous reply. The mirror symmetry is of course necessary for the exact quantized response. This has already been shown in "transport signature" of Section 2 that an increase of symmetry- breaking causes a reduction in \(l_{M_z}\) and a corresponding reduction of half- quantum mirror Hall effect related transport phenomena. Here, \(l_{M_z}\) is the inter mirror scattering length which can be used to characterize the extent of mirror symmetry breaking.  
+
+<|ref|>text<|/ref|><|det|>[[118, 360, 880, 543]]<|/det|>
+Generally, when the symmetry is broken explicitly, the mirror Hall conductance will deviate from the quantized value. In Fig. S3, we consider a special symmetry breaking term which acts only on the bottom layer. It removes the degeneracy of the surface states on the top and bottom layer, which are almost localized at the two surface layers. However, as illustrated in the right panel of Fig. R1 (also shown in Fig. S3), we observe that mirror symmetry breaking on the surface layers has a minimal impact on the quantization of mirror Hall conductivity (approximated as \(- e^2 /h\) ) when the chemical potential intersects with the surface states. This minimal effect is due to the sum of Berry curvatures across an equal energy surface of the surface states being zero, which does not contribute to the Hall conductivity. This lack of contribution underlies the emergence of the mirror Hall plateau as the chemical potential shifts within the bulk gap.  
+
+<|ref|>text<|/ref|><|det|>[[118, 558, 879, 642]]<|/det|>
+For comparison, we have also analysed the effects of applying this voltage to a layer just above the mirror plane of the film, as depicted in the left panel of Fig. R1. Notably, under these conditions, there is a significant change in the quantization of mirror Hall conductivity. This change occurs because this mirror symmetry breaking voltage impacts the high- energy bulk states which primarily contribute to the half- quantized mirror Hall conductivity.  
+
+<|ref|>text<|/ref|><|det|>[[118, 657, 860, 691]]<|/det|>
+These results reflect unique properties of the half- quantum mirror Hall effect in topological insulator films with mirror symmetry, meriting further investigation in the future.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[118, 81, 872, 456]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[117, 460, 881, 554]]<|/det|>
+<center>Fig. R1: (left) The mirror Hall conductivity \(\sigma_{xy}^{Mz}\) as a function of \(\mu\) for various gate voltages V applied to the layer slightly above the midpoint of the film. (right) The mirror Hall conductivity \(\sigma_{xy}^{Mz}\) as a function of \(\mu\) for gate voltage of V applied to the bottom surface. The remaining parameters are the same as in the main text. The gray dashed line serves as a visual guide to indicate the value of -1. </center>  
+
+<|ref|>text<|/ref|><|det|>[[117, 570, 880, 673]]<|/det|>
+Comment: .Further, the authors attribute the quantized response in Fig. S3 to "high energy states primarily located in the bulk". This appears to be in conflict with a central assertion noted in the abstract "... mirror symmetry assigns a unique mirror parity to each Dirac cone, resulting in a half-quantized Hall conductance of +/- e^2/2h for each cone.". Therefore, in an experiment, if one detects a quantized mirror-Hall conductivity, how would one know its source?  
+
+<|ref|>text<|/ref|><|det|>[[117, 691, 880, 791]]<|/det|>
+Reply: In the abstract, the term 'Dirac cones associated with surface electrons' refers to the entire gapless band spanning the first Brillouin zone, which includes both the low- energy surface states and the high- energy bulk states, also as illustrated in Fig. 2d of the main text. This distinction is important because the analysis of the Hall effect requires consideration of all occupied states within a band. We have revised this description in the abstract for better clarity, 'Dirac cones in the first Brillouin zone associated with surface electrons'.  
+
+<|ref|>text<|/ref|><|det|>[[117, 807, 880, 890]]<|/det|>
+In experiment, it is hard to figure out the source if one just detects a single data of quantized mirror Hall conductivity. However, if a plateau of mirror Hall conductivity is measured as the chemical potential varies in a finite range within the bulk gap, one can conclude that the surface electrons has no contribution to the value of the mirror Hall conductivity, but the plateau (no change with the chemical potential) is attributable to the presence of gapless surface states. The
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 880, 118]]<|/det|>
+nonzero value does arise from contributions by high- energy bulk states. Therefore, there is no conflict here.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 91, 300, 105]]<|/det|>
+REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[115, 147, 392, 163]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 202, 884, 330]]<|/det|>
+The authors have responded to the two points raised in my previous report. I find the results of investigations in to the first point to be intriguing, and it supports the claim made by the authors that the quantized response arises from the bulk states. This also indicates that the quantized response that is obtained in this paper requires a curious mixture of both bulk and surface states - - the surface states provide the basis for the Dirac cones, while the bulk states are necessary for the quantized response. The response to the second point appears to highlight this complex source of the phenomenon. Because of this feature, the thickness of the slab may play a non- trivial role.  
+
+<|ref|>text<|/ref|><|det|>[[115, 369, 863, 441]]<|/det|>
+While unusual for what one would expect for a topological response, this work represents an instance where surface and bulk states play a joint role. This property and the eventual outcome of a quantized mirror- Hall conductivity could be expected to be of immediate interest to both experimentalists and theorists working on topological materials.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__730848c60c5285f995e47458af790980361a7c821b4a57d0174037b393207bf4/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__730848c60c5285f995e47458af790980361a7c821b4a57d0174037b393207bf4/supplementary_0_Peer Review File__730848c60c5285f995e47458af790980361a7c821b4a57d0174037b393207bf4.mmd

@@ -0,0 +1,381 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Pricing indirect emissions accelerates low- carbon transition of US light vehicle sector  
+
+![PLACEHOLDER_0_0]
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+Reviewer comments, first round review -  
+
+Reviewer #1 (Remarks to the Author):  
+
+The overall premise of the paper is interesting, though the level of novelty is perhaps not the highest.  
+
+The overall results could be more clearly described - particularly the relationship between pricing, materials and emissions towards changing the vehicle mix.  
+
+The NEMS model is, as far as I understand, a CGE model for energy consuming sectors of the US economy, but it is not clear if supply- demand effects on price that are considered for the oil market (for example) are similarly considered for producers of non- energy materials (such as lithium and lithium batteries). In the present study, how is the effect of price considered on the material sectors - particularly when many of the "other" materials are not domestically produced in the US?  
+
+There is a particular practical issue of whether offshore emissions could be effectively included in the price or not.  
+
+Likewise, in the LCA data, the "other" materials would be expected to be important - otherwise the only(?) material difference between the vehicle types with functional relevance is copper? The remainder of the materials would appear to be structural? In this case, should the vehicle chassis be unified and a clearer delineation of the impacts that arise solely due to the vehicle power train given?  
+
+An inventory of the "other" materials would be useful.  
+
+Is there any consideration of the need for additional electricity generation to power a high BEV load in transport? (i.e. the electricity demand in the sector would increase, requiring additional capacity installation - not just a change from high CO2 to low CO2 mix)  
+
+It would be useful to include a diagram showing the flows of information and steps in the model to enhance understanding of the modelling approach.  
+
+Line 36 - is this \(26\%\) additional to the 1.5Gt emissions, or included? Line 48 - aren't these models referred to as Integrated Assessment Models, not specifically Integrated Energy Models?  
+
+Reviewer #2 (Remarks to the Author):  
+
+This paper uses a quantitative model to simulate the transition from internal combustion engine vehicles to electric vehicles. The paper uses a life cycle analysis model of vehicle inputs and an energy model to look at how carbon pricing affects vehicle adoption. The paper shows that if renewables could provide more than \(75\%\) of electricity, then a carbon price on the extraction/processing of oil and the combustion of gasoline for conventional vehicles (and a similar carbon price for all other emissions like vehicle construction and the remaining polluting electricity sources) would lead to greater EV adoption than a carbon price just on gasoline combustion. The authors conclude that: "the simultaneous reduction of both direct and indirect emissions indicates a win- win situation for climate change mitigation, meaning that climate policy with very high shares of BEVs represents a no- regrets strategy." And furthermore, that the US "should target deployment of BEVs and largely disregard competing technologies."  
+
+## Main comments  
+
+1. Low prices of renewables are not sufficient for large scale adoption. Electricity reliability requires that supply equal demand at all moments. When the sun is not shining and the wind is not blowing, how does a grid with \(75\%\) renewables operate? Without explicit modelling of the very large costs of storage, additional transmission, and investments to address issues of reliability, the model likely understates the costs of, or even overstates the feasibility of, large-scale renewable
+
+<--- Page Split --->
+
+deployment. A model that looks at the grid on an annual basis will not capture these issues.  
+
+2. Are fugitive methane emissions from extracting natural gas in the model?  
+
+3. The authors' conclusions ignore uncertainty. I suggest toning down the language like "no-regrets" and "disregard competing technologies".  
+
+4. In fact, showing how sensitive the paper's results are to the assumptions on renewables penetration would be helpful. If renewables do not get cheaper and coal plants do not retire, what does the model find? I would think that the indirect emissions of extracting coal and natural gas for producing electricity (especially if fugitive emissions are high) could be similar to the range of the indirect emissions from extracting and processing crude oil.  
+
+## Minor comments  
+
+5. Page 2, lines 34-35: gasoline production emissions presumably are in g CO2/km, not kWh.  
+6. Page 2, line 59 EIA stands for Energy Information Administration, not Agency.  
+7. Nature Communications does not use footnotes.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The submitted manuscript has investigated the life cycle environmental and economic implications of pricing, through the carbon tax, GHG emissions from passenger vehicles throughout their life cycles as an enviro- economical policy measure. In this pursuit, the researchers claim that pricing the GHG emissions associated with a vehicle's life cycle (though they have not provided a figure showing the system boundaries drawn for the life cycle assessment model) could be a more effective policy in reducing the environmental impacts of U.S. transportation than pricing only the GHG emissions from a vehicle's tailpipe. They further claim that such a policy measure could accelerate the phase- out of conventional vehicles while leading to higher penetration of battery electric vehicles and hence increased GHG savings.  
+
+Environmental policies have not been as efficient and effective as they are supposed to be for mitigating the negative impacts of anthropogenic activities (e.g. transportation). Pricing vehicle's life cycle emissions through carbon tax is one of such policies that is likely to have far- reaching implications in terms of the sustainability profile of U.S. passenger vehicles. Therefore, I think that gaining insights into these implications is of interest to the field of industrial ecology and others in the community and the wider field, especially given the significance of effective enviro- economical policies for mitigating the climate crisis studied in this field.  
+
+Two models - i) Yale- NEMS and ii) LCA model previously developed and published- were combined to carry out the quantitative analysis. Even though the supplementary information, including the input data, have been provided and includes data on many variables used in the models mentioned above, previously published works have been referred for some information, e.g. cost estimates for engines, electric motors, transmissions, fuel cells, and hydrogen storage tanks (and their specifications). The researchers are recommended to provide such information in a table, at least, in the SI. Furthermore, even though the source code for the LCA model has been provided, the mathematical notations of the model formulations employed in the study, as well as of the incorporation of the assumptions into these models, are missing. I would recommend that the researchers consider addressing these in the SI, at least if the word limit does not allow them to be addressed in the main manuscript. This would make it easier for the reader to have a better understanding of the analytical work done and for other researchers to be able to reproduce the results and build upon the models. Also, the researchers are recommended to provide a figure, depicting the system boundary for the LCA model developed.  
+
+I have submitted my comments as annotations and attached the annotated manuscript.
+
+<--- Page Split --->
+
+## Response to the reviewers  
+
+Dear anonymous reviewers,  
+
+We would like to thank you very much for your thoughtful reviews of our previously submitted manuscript "Pricing indirect emissions accelerates low- carbon transition of US light vehicle sector". We have worked hard to successfully address all the concerns that have been expressed and we believe that the paper has improved as a result. We look forward to hearing from you.  
+
+Kindest regards,  
+
+Paul Wolfram (on behalf of all authors)  
+
+## Reviewer 1  
+
+The overall premise of the paper is interesting, though the level of novelty is perhaps not the highest.  
+
+Thank you for this thought. We appreciate that you find the overall premise interesting. In the comment on the novelty, we believe that the reviewer may be referring to the fact that previous static life cycle assessments have highlighted the importance of the electricity mix for both charging vehicles and producing the vehicles in the first place. We agree with this assessment of the literature. However, the present work differs in being the first large- scale dynamic assessment of electric vehicle roll- out scenarios considering both the life cycle coefficients and the dynamic effects of the entire energy system at the same time. This combination of approaches is novel and leads to a new result to the literature: that the negative impacts that occur with electric vehicle adoption can be largely avoided. Thus, we see our manuscript to be of great interest to those engaged in the lively debate about EV policy. We would like to point out that the article pre- print on ResearchSquare attracted more than 300 viewers, with 100 alone within the first five days. This perhaps can be seen as at least one indication that the broader academic community will find the work novel.
+
+<--- Page Split --->
+
+The overall results could be more clearly described - particularly the relationship between pricing, materials and emissions towards changing the vehicle mix.  
+
+Thank you for this suggestion - it is one that we have taken seriously. One new aspect of this work above the previous literature is that the pricing of all emissions, including those stemming from material production, influence the optimal vehicle fleet. Thus, we feel it is especially important to document this part of our approach clearly and we appreciate your nudge to do so. We added three mathematical equations in the methods section providing additional detail on how embodied emissions from material production and energy chains affect vehicle prices. We further added two sections in the supplementary information which serve as a more detailed description of the interconnection between prices, materials and emissions (Supplementary Sections 2 and 10). We believe that this additional information complements the existing descriptions in the main manuscript well and we thank you again for this comment.  
+
+The NEMS model is, as far as I understand, a CGE model for energy consuming sectors of the US economy, but it is not clear if supply- demand effects on price that are considered for the oil market (for example) are similarly considered for producers of non- energy materials (such as lithium and lithium batteries). In the present study, how is the effect of price considered on the material sectors - particularly when many of the 'other' materials are not domestically produced in the US? There is a particular practical issue of whether offshore emissions could be effectively included in the price or not.  
+
+The reviewer's understanding of the model is correct. NEMS is indeed a CGE model focusing on the US economy. (In addition, the model includes world energy prices, world energy supply and demand, as well as US energy imports and exports. It is also correct that the market for these materials is not represented explicitly in NEMS.) In our model, the GHG emissions of these materials are part of the life- cycle assessment (LCA). Our single- region LCA model indeed assumes that vehicle production takes place in the US. We updated the text on lines 317- 320 accordingly: "For simplicity purposes, the LCA model assumes that vehicle production takes place in the US [...] (see Supplementary Section 10 for a discussion of the error invoked from these assumptions)." Accordingly, we added a new section (Supplementary Section 10) and there we added: "Further, our single- region LCA model assumes that vehicle production takes place in the US. In reality,
+
+<--- Page Split --->
+
+many of the vehicles bought in the US are made elsewhere in the world. On the other hand, quite a few are exported. \(^{1}\) We present evidence that the error invoked from disregarding these trade relations is small however. To that end we analyze the differences between the 'full pricing' and the 'well- to- wheel' pricing scenario. The 'full pricing' scenario fully prices embodied emissions of vehicle and battery production while the 'well- to- wheel pricing' scenario excludes pricing of vehicle and battery production emissions altogether. Hence, the cost of embodied vehicle and battery production emissions are zero under 'well- to- wheel' pricing. Yet, the differences in sales are marginal: merely a few longer- range EVs are partially replaced by shorter- range EVs (Figure S6h). Furthermore, these sales differences do not notably affect overall emissions outcomes (Figure 2a). As documented in Supplementary Table 2, the difference in total fleet- wide life cycle emissions between the two scenarios amounts to 71 Mt \(\mathrm{CO_2}\) cumulatively over the period 2010- 2050. Hence, the results of this study are fairly unsusceptible to assumptions regarding the carbon intensity of vehicle and battery production which in turn partly depend on the location of production (inside versus outside of the US).  
+
+In addition, the mentioned simplifications do not prevent us from capturing potential changes to material prices in our model, at least in a simplified way. Therefore, inspired by the reviewer's thoughtful feedback, we added six new sensitivity cases, of which four assume constant prices of EV batteries from 2021 on. This constant price can be interpreted either as insufficient investments into battery technology or as growing raw material prices. IEA's newly published World Energy Outlook \(^{2}\) states that "a doubling of lithium or nickel prices would induce a 6% increase in battery costs. If both lithium and nickel prices were to double at the same time, this would offset all the anticipated unit cost reductions associated with a doubling of battery production capacity." We note the results of this investigation in a new section called "Uncertainty analysis" as well as in Supplementary Section 6.  
+
+Likewise, in the LCA data, the 'other' materials would be expected to be important - otherwise the only(?) material difference between the vehicle types with functional relevance is copper? The remainder of the materials would appear to be structural? In this case, should the vehicle chassis be unified and a clearer delineation of the impacts that arise solely due to the
+
+<--- Page Split --->
+
+vehicle power train given? An inventory of the ‘other’ materials would be useful.  
+
+We apologize for not being fully clear in our previous draft. Accordingly, we added a caveat to lines 317- 320: “For simplicity purposes […] the model includes the most climate- relevant vehicle materials and disregards other minor materials (see Supplementary Section 10 for a discussion of the error invoked from these assumptions).”  
+
+In addition, we added to Supplementary Section 10: “our detailed process- based model allows for an explicit differentiation in composition and mass of the glider and power train. We consider the seven most common materials used for vehicle production: cast iron, stainless steel, automotive steel, wrought aluminum, cast aluminum, copper, and plastics. Combined these contribute more than 92% of the weight of the vehicle. The ‘other’ materials category is mostly comprised of glass and rubber, and for reasons of simplicity, we estimate the emissions of these other minor materials at 2 kg \(\mathrm{CO_2 / kg}\) material in the base year. This is well within the range of emission factors of rubber and glass which make up the vast majority of the ‘other’ category (by weight). For example, according to the ecoinvent 3.5 database using IPCC’s 2013 GWP- 100 indicator, the production of synthetic rubber emits 2.75 kg \(\mathrm{CO_2 / kg}\) material, while natural rubber and uncoated flat glass emit 2.02 and 0.99 kg \(\mathrm{CO_2 / kg}\) .  
+
+Other potentially important materials specific to EV batteries may be cobalt, nickel and lithium, which are not considered in our model. Comparing our inventory and GHG emission results to that of a recently published and very detailed study,3 these materials account for 15% of the emissions associated with battery production. This translates into an omission of about 6% for the production of the vehicle as a whole, and of merely 1% for the entire EV life cycle including charging electricity in the model’s base year. We acknowledge that the relative error could be higher in future years assuming a decarbonization of the electricity mix but not the metal production. However, simultaneous improvements in battery technology are conceivable as well. We therefore anticipate that including the embodied \(\mathrm{CO_2}\) emissions from cobalt, nickel and lithium would not notably change scenario outcomes.” As mentioned above, this assertion is also confirmed if one analyzes the small differences in vehicle sales and resulting fleet life- cycle emissions between the ‘full pricing’ scenario and the ‘well- to- wheel pricing’ scenario (Figure 2a, Figure S6h, Supplementary Table 2).
+
+<--- Page Split --->
+
+Is there any consideration of the need for additional electricity generation to power a high BEV load in transport? (i.e. the electricity demand in the sector would increase, requiring additional capacity installation - not just a change from high CO2 to low CO2 mix)  
+
+We completely agree that this is important. Fortunately, the model indeed considers the additional electricity generation to power a high BEV load. In fact, this was one of the reasons why we chose the model. The effects of the additional electricity load should be clear from the additional information provided in the SI, such as Supplementary Figure 12, showing the increase in overall electricity demand, and Supplementary Figure 14, illustrating new additions to electricity generation capacity as well as retirements of old power plants.  
+
+It would be useful to include a diagram showing the flows of information and steps in the model to enhance understanding of the modelling approach.  
+
+We thank the reviewer for this suggestion. We now provide a graph showing an overview of the modelling framework in Supplementary Section 10.  
+
+Line 36 - is this \(26\%\) additional to the 1.5Gt emissions, or included?  
+
+The \(26\%\) are included in the 1.5 Gt. We double- checked the sentence to make sure that our language is clear.  
+
+Line 48 - aren't these models referred to as Integrated Assessment Models, not specifically Integrated Energy Models?  
+
+The definition of the term 'Integrated Assessment Model' is not consistent throughout the literature. Given that most of these models do not contain an assessment of the damage of climate change, we prefer to use the term 'Integrated Energy Model' for which some of the authors provided a detailed definition in a recent peer- reviewed paper.4 We also provide this definition in a new section in the SI (Supplementary Section 9).  
+
+## Reviewer 2  
+
+This paper uses a quantitative model to simulate the transition from internal combustion engine vehicles to electric vehicles. The paper uses a life cycle analysis model of vehicle inputs and an energy model to look at how
+
+<--- Page Split --->
+
+carbon pricing affects vehicle adoption. The paper shows that if renewables could provide more than \(75\%\) of electricity, then a carbon price on the extraction/processing of oil and the combustion of gasoline for conventional vehicles (and a similar carbon price for all other emissions like vehicle construction and the remaining polluting electricity sources) would lead to greater EV adoption than a carbon price just on gasoline combustion. The authors conclude that: "the simultaneous reduction of both direct and indirect emissions indicates a win- win situation for climate change mitigation, meaning that climate policy with very high shares of BEVs represents a no- regrets strategy." And furthermore, that the US "should target deployment of BEVs and largely disregard competing technologies."  
+
+Main comments:  
+
+1. Low prices of renewables are not sufficient for large scale adoption. Electricity reliability requires that supply equal demand at all moments. When the sun is not shining and the wind is not blowing, how does a grid with \(75\%\) renewables operate? Without explicit modelling of the very large costs of storage, additional transmission, and investments to address issues of reliability, the model likely understates the costs of, or even overstates the feasibility of, large-scale renewable deployment. A model that looks at the grid on an annual basis will not capture these issues.  
+
+We agree with you that reliability must be included in the analysis. Fortunately, NEMS includes a dispatch model that determines electricity supply, demand and prices at sub- annual level (three seasons by three times of day). It is important to note that NEMS explicitly models the cost of additional investments needed to allow for intermittent renewable electricity generation capacity. This is a core part of the capacity additions modeling. Thus, NEMS has the characteristics you identify as crucial for exploring our research question. We apologize if we were unclear and implied that NEMS was run only at an annual basis and did not model reliability. We have rectified this by adding the following sentence (lines 276- 278): "A dispatch model determines electricity supply, demand and prices at sub- annual level (three seasons by three times of day)."  
+
+As for the feasibility of large- scale renewable development, we find that our results are backed by several previous studies indicating that such a high share of renewables
+
+<--- Page Split --->
+
+is possible. These studies even include work published in Nature Communications. \(^{5}\) \(^{6}\) \(^{7}\) Of course achieving a high share of renewables is not the only way to achieve decarbonization of the electricity system. For example, see the careful discussion in the IPCC's recent Special Report on Global Warming of 1.5 C, which is widely accepted by the scientific community. In the United States, the Biden administration aims to decarbonize electricity generation with a Clean Electricity Standard as the central policy, which will likely encourage development of renewables, but also opens the door to other technologies. These questions of the cost and feasibility of high renewables are very interesting, but they are not the research question at hand in our paper. Rather, we view the high renewables scenario as a starting point for our analysis.  
+
+## 2. Are fugitive methane emissions from extracting natural gas in the model?  
+
+We apologize for our missing description of methane emissions. We agree with you that methane emissions from fossil fuel extraction and transportation can be significant and should ideally be taken into consideration. Unfortunately, NEMS only reports emissions of \(\mathrm{CO_2}\) and a range of air pollutants. Thus, we have now made this very clear by updating the model description accordingly (lines 248- 252): "Yale- NEMS provides a full account of \(\mathrm{CO_2}\) emissions across all industries and a range of air pollutants from vehicles and power plants. \(\mathrm{CO_2}\) accounted for \(97\%\) of total GHG emissions in the US electricity and transport sectors in 2019. \(^{8}\) Other GHGs such as methane emissions from fossil- fuel and hydroelectric power plants are not included." We also now include a more detailed description of natural gas \(\mathrm{CO_2}\) emissions factors used in this work in our new Supplementary Section 6.  
+
+## 3. The authors' conclusions ignore uncertainty. I suggest toning down the language like "no-regrets" and "disregard competing technologies".  
+
+We agree with you; our previous language was too strong. In response to your comment, we deleted the clause "disregarding competing technologies". In addition, we double- checked our language to make it more clear that BEVs are a "no regrets" strategy only if electricity decarbonizes, as has been assumed in our main scenarios. Examples include lines 20- 22: "Given continued decarbonization of electricity supply, results show
+
+<--- Page Split --->
+
+that a large- scale adoption of electric vehicles is able to reduce \(\mathrm{CO_2}\) emissions through more channels than previously expected"; lines 204- 208: "In fact, the simultaneous reduction of both direct and indirect emissions indicates a win- win situation for climate change mitigation, meaning that climate policy with very high shares of BEVs represents a no- regrets strategy (but only if electricity continues to decarbonize as has been assumed in our main scenarios)"; and lines 404- 407: "In terms of emissions, a carbon tax on supply chain emissions is not able to yield the desired results if the electricity grid does not face substantial decarbonization. In this case, pricing supply chain emissions leads to higher emissions compared to pricing direct emissions only (see Supplementary Section 6.1 for more details and results)". Our new sensitivity cases which have been requested by the reviewer now make it clear that large adverse effects can occur if electricity slowly or barely decarbonizes.  
+
+4. In fact, showing how sensitive the paper's results are to the assumptions on renewables penetration would be helpful. If renewables do not get cheaper and coal plants do not retire, what does the model find? I would think that the indirect emissions of extracting coal and natural gas for producing electricity (especially if fugitive emissions are high) could be similar to the range of the indirect emissions from extracting and processing crude oil.  
+
+We thank the reviewer for this thoughtful comment. As indicated above we include six new scenarios which explore the uncertainty in the costs of electric vehicles and renewable power plants (see section "Uncertainty analysis" and Supplementary Section 6). From the results of these scenarios it becomes apparent that the reviewer is correct in their assertion that the emissions from fossil- fuel based electricity could exceed those of gasoline production if the electricity sector fails to decarbonize substantially. In addition, we now provide a full list and brief description of our 28 scenarios in Supplementary Section 4. We believe that these 28 scenarios more than adequately explore the main uncertainties present in this work. The results of the extreme (i.e., most optimistic and most pessimistic) scenarios are described in detail throughout the main manuscript and the SI.  
+
+Minor comments:  
+
+5. Page 2, lines 34-35: gasoline production emissions presumably are in g CO2/km, not kWh.
+
+<--- Page Split --->
+
+We chose to provide gasoline production emissions in \(\mathrm{gCO_2 / kWh}\) instead of \(\mathrm{gCO_2 / km}\) in order to avoid additional assumptions on vehicle fuel consumption (e.g., \(\mathrm{kWh / 100km}\) ). We deeply considered providing \(\mathrm{gCO_2 / km}\) numbers along with the \(\mathrm{gCO_2 / kWh}\) numbers but we ask the reviewer for understanding that we refrained from this idea in the end. With the range of different vehicle segments and technologies analyzed in this work, we worry it would make the sentence too convoluted and thus difficult to communicate the key point to readers. However, we decided to provide values in \(\mathrm{gCO_2 / gallon}\) gasoline equivalents in Supplementary Section 1. We hope that this is an adequate solution and if this is a crucial point to you, we are open to reassessing.  
+
+## 6. Page 2, line 59 EIA stands for Energy Information Administration, not Agency.  
+
+Thank you for spotting this. We corrected accordingly.  
+
+## 7. Nature Communications does not use footnotes.  
+
+Thank you again. We made sure to avoid footnotes throughout the entire manuscript.  
+
+## Reviewer 3  
+
+The submitted manuscript has investigated the life cycle environmental and economic implications of pricing, through the carbon tax, GHG emissions from passenger vehicles throughout their life cycles as an enviro- economical policy measure. In this pursuit, the researchers claim that pricing the GHG emissions associated with a vehicle's life cycle (though they have not provided a figure showing the system boundaries drawn for the life cycle assessment model) could be a more effective policy in reducing the environmental impacts of U.S. transportation than pricing only the GHG emissions from a vehicle's tailpipe. They further claim that such a policy measure could accelerate the phase- out of conventional vehicles while leading to higher penetration of battery electric vehicles and hence increased GHG savings. Environmental policies have not been as efficient and effective as they are supposed to be for mitigating the negative impacts of anthropogenic activities (e.g. transportation). Pricing vehicle's life cycle emissions through carbon tax is one of such policies that is likely to have far- reaching implications in terms of the sustainability profile
+
+<--- Page Split --->
+
+of U.S. passenger vehicles. Therefore, I think that gaining insights into these implications is of interest to the field of industrial ecology and others in the community and the wider field, especially given the significance of effective enviro- economical policies for mitigating the climate crisis studied in this field.  
+
+We deeply thank the reader for this assessment and very much share the view that this research is of interest to a wide range of research communities.  
+
+Two models - i) Yale- NEMS and ii) LCA model previously developed and published- were combined to carry out the quantitative analysis. Even though the supplementary information, including the input data, have been provided and includes data on many variables used in the models mentioned above, previously published works have been referred for some information, e.g. cost estimates for engines, electric motors, transmissions, fuel cells, and hydrogen storage tanks (and their specifications). The researchers are recommended to provide such information in a table, at least, in the SI.  
+
+We apologize for not having included this information previously. We now provide detailed cost figures for all modeled powertrain components in Supplementary Table 9 in the provided Excel spreadsheet. Similarly, detailed specifications of individual vehicles and vehicle components, including battery capacity and weight, battery depth of discharge, weight of motors, energy consumption, motorization, purchase price, fuel tank size, and total vehicle weight, have been included in Supplementary Table 11. Scenario- specific fleet- average fuel efficiencies, purchase prices, and degree of lightweighting are endogenous modelling results of NEMS, and have been documented in Supplementary Tables 5, 14 and 12.  
+
+Furthermore, even though the source code for the LCA model has been provided, the mathematical notations of the model formulations employed in the study, as well as of the incorporation of the assumptions into these models, are missing. I would recommend that the researchers consider addressing these in the SI, at least if the word limit does not allow them to be addressed in the main manuscript. This would make it easier for the reader to have a better understanding of the analytical work done and for other researchers to be able to reproduce the results and build upon the models. Also, the researchers are recommended to provide a figure, depicting the system boundary for the LCA model developed.
+
+<--- Page Split --->
+
+Thank you for this important suggestion. We now provide the three central equations of our approach in the methods section. For additional equations describing the LCA model and NEMS, interested readers can consult Wolfram et al. \(^{9}\) and the NEMS model documentation. We made sure that these references are shown more prominently in the methods section. In addition, as noted above, we now provide a figure of the modeling approach and the systems boundary in Supplementary Section 2.  
+
+I have submitted my comments as annotations and attached the annotated manuscript.  
+
+Line 30: The reference number 3 investigates the life cycle energy and environmental assessment of natural gas as transportation fuel in Pakistan and is not quite relevant for the statement it was given to support it. Given that the spatial and technological scope of the study is the United States, the reference number 3 can be replaced with a more relevant reference, e.g. having the same spatial scope such as Onat, N., Kucukvar, M., and Tatari O. (2014). Towards Life Cycle Sustainability Assessment of Alternative Passenger Vehicles. Sustainability 6 (12). Here, please, consider citing that work, instead, or adding that reference, as well, to acknowledge quite a relevant work.  
+
+We added this important reference.  
+
+Lines 31- 32: This reads like an incomplete sentence and I could not understand. Please, consider revising this sentence.  
+
+We revised this sentence accordingly in order to make it more understandable: "These emissions occur off- site, or indirectly, and include generation of electricity to charge electric vehicles, in this work \(\sim 66 - 86 \mathrm{g} \mathrm{CO}_{2}\) per electric- vehicle km driven in 2020, as well as the production of vehicles, here \(\sim 16 - 38 \mathrm{g} \mathrm{CO}_{2}\) per vehicle- km driven in 2020 (Supplementary Section 1 and Supplementary Table 1)."  
+
+Line 34: Could the researchers explain why they preferred using such a unit when referring to the emissions from gasoline production? Did they use the conversion factor of \(1 \mathrm{kWh} = 0.03 \mathrm{GGE}\) ?  
+
+We use kWh because it is a unit listed in the International System of Units (SI). For better readability we would wish to stick to kWh in the main text. In Supplementary
+
+<--- Page Split --->
+
+Section 1 however, we decided to also provide values in GGE using the conversion factor that the reviewer kindly provided.  
+
+Line 39: Since they are the main subject of the study, the researchers are recommended to indicate more clearly (maybe, in paranthesis) what these indirect emissions include to get rid of the confusion caused by the ambiguity of the term.  
+
+Agreed. We changed the sentence accordingly. It now reads as follows (now lines 38- 40): "The introduction of the Low Carbon Fuel Standard (LCFS) in California, which regulates all fuel and electricity production and combustion emissions, shows that transport policy in practice can at least partly address indirect vehicle emissions."  
+
+Lines 44- 45: The effect on production decision of changing costs due to regulatory standards such as CAFE has been investigated before. The researchers are recommended to refer to the study titled CAFE's impact on the market share of electric vehicles by Sen et al. (2017) to acknowledge a relevant work.  
+
+We included this interesting reference.  
+
+Line 75: What is meant by this term? Please, elaborate.  
+
+We changed the sentence to make it more clear. It now reads (now lines 77- 79): "Here we address these knowledge gaps by applying a novel conceptual framework by Creutzig et al., which focuses on energy- demand side (rather than energy- supply side) solutions to climate change mitigation."  
+
+Lines 79- 80: These sectors are already parts of an EV's life cycle under the IO modeling setting (which can be represented as unit processes under the process modeling settin). Such responses are reflected upon in the life cycle sustainability assessment studies of different vehicle classes, if I have understood this sentence, correctly. That's why the researchers are recommended to clarify what are meant by direct and indirect emissions and by 'vehicle sectors', and provide a figure, depicting the system boundary adopted for this study.  
+
+The reviewer is absolutely correct in stating that the sectors material production, vehicle manufacturing and electric charging are reflected and adequately linked to each other in the LCA model. On the other hand, NEMS has the advantage of better captur
+
+<--- Page Split --->
+
+ing the non- linear dynamics of the entire energy system including changes in electricity supply. In order to make use of the advantages of both models, we soft- link our LCA model to NEMS. We hope that this becomes clear from our description, especially now that we included a new figure depicting the system boundary of this work (Supplementary Section 2).  
+
+Line 87: So, Scenario 1 assumes that the emissions from the tailpipe are accounted for, priced, and optimized for, whereas Scenario 2 assumes that the emissions from the entire vehicle supply chain are accounted for, priced, and optimized for. Is that correct?  
+
+Yes, the reviewer is correct. We double- checked our scenario descriptions throughout the main manuscript and the SI and we hope that they are adequate and clear.  
+
+Line 95: Does "direct tailpipe emissions" mean that tailpipe emissions are the direct emissions? I believe there is a need to provide a clarification as to what are meant by direct and indirect emissions. Are direct emissions those from tailpipe or from vehicle sectors? Similarly, are indirect emissions those from non- tailpipe emissions or from material extraction sector? The researchers are recommended to clearly define these emissions to avoid any confusion.  
+
+The reviewer interpreted correctly what is meant by 'direct tailpipe emissions'. In order to avoid any ambiguities we state on line 35 that we use the term 'direct emissions' and 'tailpipe emissions' as synonyms.  
+
+Line 104: Is this referring to the "well to tank" emissions? Please elaborate.  
+
+Energy- chain emissions describe emissions invoked by the production and use of energy carriers (well- to- wheel). We added to line 104 (now line 107) that the term 'energy- chain' emissions is synonymous with 'well- to- wheel' emissions in order to avoid confusion.  
+
+Lines 104- 105: The researchers assumed that hydrogen production becomes carbon- netural by 2050 through the hydrogen production from biomethane, with CCS. Why have the researchers not considered the green hydrogen production, at all? How do they think that such a consideration would affect the conclusions of the study?  
+
+In Supplementary Section 1 we provide an example on how carbon neutral produc
+
+<--- Page Split --->
+
+tion of hydrogen could be achieved. We added the reviewer's example of a green hydrogen pathway to the text. We also explain how this pathway would influence modelling outcomes. The section now reads as follows: "Net- zero could be achieved in various different ways, for example by producing hydrogen exclusively from wind or solar power or by using a hydrogen production mix consisting primarily of hydrogen from biomethane with carbon capture and storage, representing a carbon sink with about \(- 36 \mathrm{g} \mathrm{CO}_{2} \mathrm{e} / \mathrm{kWh}\) , and a small remainder, around \(7 - 8\%\) , of hydrogen from SMR, emitting around \(450 \mathrm{g}\) . Since carbon- neutral hydrogen production is only considered in one of our side cases it has been modeled in less detail: For the carbon- neutral hydrogen side case NEMS only receives the hydrogen production emissions factor from the LCA model, which can be assumed equal under both production pathways. Hence, for the purpose of this paper, whether hydrogen is produced from renewables or from a combination of biomethane CCS and natural gas, has no effect on the emissions outcomes because both production pathways would ensure net- zero emissions."  
+
+Lines 113- 114: Here the researchers are recommended to cite a study on the material footprint of electric vehicles through MRiO, conducted by Sen et al. (2019). This is one of the very few studies in the literature that shows the material intensity of EVs, which is very relevant to cite to acknowledge the previous work in this regard.  
+
+Thank you for this suggestion. We added the reference.  
+
+## Line 116: Are these ones different from the a, b, and c previously mentioned?  
+
+They are in fact the same. In order to avoid confusion we added a reference to Supplementary Figure 6 and changed the sentence. It now reads (now lines 120- 123): "As mentioned earlier, we explore a range of side cases (Supplementary Section 3) which show some variation in their potential for emission reductions (also see dotted lines in Figure 2a- j) but the overall trend is robust among these cases."  
+
+## Line 152: The researchers are recommended to provide provide the percentages, as well, to help the reader relate better.  
+
+We made sure that percentages are provided as well. Please see the previous paragraph on lines 151 to 158. In addition, relative changes are also indicated in Figure 4. We would also like to note that all underlying data for Figure 4 is included in Supplementary Table 7 in the accompanying Excel spreadsheet.
+
+<--- Page Split --->
+
+Line 172: Given how significantly battery manufacturing influences the sustainability impacts of EVs, why have the researchers not considered discussing the implications of such a policy measure in terms of the future of EV batteries and how increasing demand on BEVs would likely influence battery technologies and the demand for them, as well as their prices?  
+
+We agree with the reviewer that EV batteries have a significant influence on sustainability issues. Our focal point is however not on emissions from batteries in particular but rather on the overall influence of indirect emissions on optimal vehicle fleets. We ask for the reviewer's understanding that, due to the limited space in the main manuscript, we had to focus on a high level discussion of the topic. However, in response to your comment, we added some additional discussion items on vehicle and battery materials to Supplementary Section 10.  
+
+Lines 187- 188: This is a common finding of the studies on vehicle LCA that do not consider pricing, at all. So, how does the pricing affect this? Through a higher market penetration of BEVs thanks to taxing the full LC emissions that influence the consumer's purchase behavior?  
+
+The reviewer is correct again. A useful aspect of our study is the consideration of prices and the ability to take consumer decisions into account. As the reviewer points out correctly, this ability of the model enables us to study the effects of holistic pricing of all embodied emissions of different vehicle options. The model finds that a pricing of all emission sources leads to a different optimal solution than pricing of tailpipe emissions alone.  
+
+## Line 189: Does this refer to the extraction of crude oil?  
+
+We thank the reviewer for spotting this. In fact, we wanted to refer to the entire production process of crude oil, including both exploration and extraction. In order to avoid any ambiguities we rephrased the sentence. It now reads as follows (now lines 197- 198): "However, higher electricity emissions are more than offset by lower gasoline supply- chain emissions stemming from the production of crude oil (Figure 2k)."  
+
+Line 256: The researchers are recommended to provide these assumptions/values, along with their references, in a table. In fact, given the large scope of the study, it might even be better to provide all your assumptions in a table.  
+
+We agree with you completely and fully support the reproducibility of scientific re
+
+<--- Page Split --->
+
+sults. We therefore provide all cost assumptions of various vehicle components in Supplementary Table 9 in the provided Excel spreadsheet. In the same spreadsheet we provide 21 additional data tabs representing an exhaustive modelling database.  
+
+Line 269: The researchers are recommended to consider normalizing Fig. 8b or present a separate figure, with normalized price information per BTU, for example.  
+
+We provide an additional panel in Supplementary Figure 8 depicting normalized gasoline and electricity prices per BTU as requested by the reviewer. For conversion we use conversion factors provided by the U.S. EIA. \(^{10}\)  
+
+Lines 273- 275: How was this formulated and incorporated into the model? The researchers are recommended to consider providing the mathematical notations of all their formulations.  
+
+In this instance, apart from updating the costs of rooftop solar PVs, no changes have been made to the NEMS model by the authors. Due to space constraints unfortunately we cannot reproduce the detailed equations and descriptions contained in the NEMS modelling documentation. However, we added a reference to the specific section of the NEMS documentation (section "Distributed Generation and Combined Heat and Power (CHP) Submodule" within the "Commercial Demand" section of the NEMS documentation). Interested readers will be able to find all the required information there. Please also refer to the new figure in Supplementary Section 2 where we depict a simplified representation of the commercial and residential buildings sectors.  
+
+## Lines 276-279: Are these the researchers' assumptions?  
+
+This paragraph describes future electricity demand growth, the development of electricity emissions as well as the resulting electricity carbon intensity, all of which are NEMS modelling results. These developments are a direct result of our cost assumptions on renewable electricity generators. We double- checked lines 274 to 277 to make sure that these assumptions are clearly explained there.  
+
+Line 303: What do C\$, LCC\$, WTW\$. mean, and what do the values provided represent, exactly?
+
+<--- Page Split --->
+
+We apologize for this mishap. These acronyms used in the supplementary Excel spreadsheet are only meant for internal use. We renamed them accordingly in order to be consistent with the scenario names in the manuscript. We made sure that the first tab of the Excel spreadsheet file describes the content of each following tab. In addition, we double- checked that each tab has a descriptive header and that all units are provided.  
+
+## Lines 309–310: Overall, I think there is a lack of mathematical representation of the model(s) formulations and of the incorporation of assumptions into these formulas. Please consider providing the mathematical notations.  
+
+As mentioned above, we now provide the three central equations of our approach in the methods section. For additional equations describing the LCA model as well as NEMS, interested readers can consult Wolfram et al. \(^{11}\) as well as the NEMS model documentation. We made sure that these references are shown more prominently in the methods section.  
+
+## Line 327: Why have the battery capacities been assumed constant after 2025? And, have the researchers assumed any improvements in the GHG efficiency of battery production over the years?  
+
+In order to address these important questions, we added two new sections (Supplementary Sections 6.2 and 11). We address the first part of the question in Supplementary Section 6.2: “A standard assumption of Yale–NEMS is that of constant EV battery capacities (as well as constant battery weights and densities and hence EV ranges) after 2025 (Supplementary Table 11). In most of our scenario runs we adopted this assumption because, on the one hand, batteries could become more energetically dense in the future, hence requiring smaller capacities. On the other hand, larger capacities may be needed if BEVs continue to increase in driving range. Both factors could cancel each other out, hence the constant capacity assumption. However, in this section we present two sensitivity cases that explore the effects of increasing battery densities, leading to smaller, lighter, less material-intensive and cheaper EV batteries while providing the same driving range. We assume that – averaged over all technologies – battery densities continue to increase by about 1.5% per year after 2025. This rate is somewhat higher than the assumed average 2010–2025 increase of about 0.9% per year. This development helps especially with cost reductions of longer–range BEVs. This is true under both direct–emissions–only pricing
+
+<--- Page Split --->
+
+and marginally more so under full- emissions pricing because the emissions penalty of the smaller batteries is also lowered. Hence, a shift in sales from 100- mile to 200- mile range BEVs can be observed. In addition, this trend is accompanied by an overall increase in BEV sales shares (Figure S10). "  
+
+The second part of the question is addressed in Supplementary Section 11: "The GHG intensity of batteries in particular (and vehicle production in general) improves in several ways in our model. The main scenarios assume a falling carbon intensity of electricity, which reduces both the emissions during the material production stage as well as emissions invoked during the battery assembly stage (Supplementary Table 18). The assumed energy mix of the material production and battery assembly stages is comprised of heat from fossil fuels as well as electricity (see Supplementary Tables 19, 20). In addition, some scenarios assume improved recycling of materials and reuse of components such as batteries (see Supplementary Tables 21, 22), further reducing GHG emissions." We also hope that the figure provided in Supplementary Section 2 further helps to communicate these relationships.
+
+<--- Page Split --->
+
+Reviewer comments, second round review -  
+
+Reviewer #1 (Remarks to the Author):  
+
+The paper has been much improved.  
+
+I would prefer that the figure describing the modelling was in the main text rather than the supplementary material, as it would better give context to the results. The paper as a whole tends to lean on the supplementary material to a significant extent, which detracts from understanding - perhaps the authors could add some additional commentary to the supplementary material so that it could be read more as a report, so that those who are interested in understanding in detail could skip the paper and go to the supplement.  
+
+Reviewer #2 (Remarks to the Author):  
+
+Thank you for addressing my previous concerns. I have no further comments.  
+
+Reviewer #3 (Remarks to the Author):  
+
+I thank the authors for addressing and clarifying all my comments and concerns regarding the submitted manuscript. Nature, as a leading multidisciplinary science journal, is a source of scientific knowledge that is highly trusted both by academia and society, in general. Every single sentence that is written in any manuscript that is to be potentially published in Nature must be paid due attention, and all the assumptions considered in any analysis must be justified e.g. by means of providing relevant publication(s), accordingly. So, as my last comment, I suggest that the authors make an extra effort to highlight, at least, any significant assumptions throughout the manuscript that will enhance the understanding of the readers, especially the primary target audience.  
+
+Overall, given the comment above, the manuscript is suggested for publication after that very minor revision.
+
+<--- Page Split --->
+
+## Response to the reviewers  
+
+Dear anonymous reviewers,  
+
+We would like to thank you again for reviewing our revised manuscript entitled "Pricing indirect emissions accelerates low- carbon transition of US light vehicle sector". We incorporated all of your remaining suggestions and look forward to hearing from you.  
+
+Kindest regards,  
+
+Paul Wolfram (on behalf of all authors)  
+
+## Reviewer 1  
+
+The paper has been much improved. I would prefer that the figure describing the modelling was in the main text rather than the supplementary material, as it would better give context to the results. The paper as a whole tends to lean on the supplementary material to a significant extent, which detracts from understanding - perhaps the authors could add some additional commentary to the supplementary material so that it could be read more as a report, so that those who are interested in understanding in detail could skip the paper and go to the supplement.  
+
+Thank you for this comment. In response to your suggestion, we copied a simplified, more compact version of the figure describing the modelling framework into the main manuscript. The full version of the figure would require many more words in the main text, muddling the flow of the manuscript and requiring cuts elsewhere, but we believe that the compact version allows us to convey the main ideas in a concise way. We also added a new section to the beginning of the supplementary material briefly summarizing the results of the paper. In addition, we strictly adhered to the style guidelines of Nature Communications when preparing the supplementary material.
+
+<--- Page Split --->
+
+## Reviewer 2  
+
+Thank you for addressing my previous concerns. I have no further comments. Thank you again!  
+
+## Reviewer 3  
+
+I thank the authors for addressing and clarifying all my comments and concerns regarding the submitted manuscript. Nature, as a leading multidisciplinary science journal, is a source of scientific knowledge that is highly trusted both by academia and society, in general. Every single sentence that is written in any manuscript that is to be potentially published in Nature must be paid due attention, and all the assumptions considered in any analysis must be justified e.g. by means of providing relevant publication(s), accordingly. So, as my last comment, I suggest that the authors make an extra effort to highlight, at least, any significant assumptions throughout the manuscript that will enhance the understanding of the readers, especially the primary target audience. Overall, given the comment above, the manuscript is suggested for publication after that very minor revision.  
+
+Thank you for this suggestion. We made several final changes that further improved the readability and the understanding of our manuscript. First, we restructured the introduction. Second, we included a simplified version of Figure S3 in the main text. Finally, we carefully double- checked each and every sentence and made sure that all assumptions made in our work are clearly documented. We strongly believe that the manuscript is now ready for publication in Nature Communications.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__730848c60c5285f995e47458af790980361a7c821b4a57d0174037b393207bf4/supplementary_0_Peer Review File__730848c60c5285f995e47458af790980361a7c821b4a57d0174037b393207bf4_det.mmd

@@ -0,0 +1,520 @@
+<|ref|>title<|/ref|><|det|>[[61, 40, 508, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[67, 111, 362, 140]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[70, 155, 806, 211]]<|/det|>
+Pricing indirect emissions accelerates low- carbon transition of US light vehicle sector  
+
+<|ref|>image<|/ref|><|det|>[[57, 732, 240, 780]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[250, 732, 912, 785]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[120, 85, 421, 99]]<|/det|>
+Reviewer comments, first round review -  
+
+<|ref|>text<|/ref|><|det|>[[120, 136, 415, 150]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 164, 833, 193]]<|/det|>
+The overall premise of the paper is interesting, though the level of novelty is perhaps not the highest.  
+
+<|ref|>text<|/ref|><|det|>[[118, 193, 870, 222]]<|/det|>
+The overall results could be more clearly described - particularly the relationship between pricing, materials and emissions towards changing the vehicle mix.  
+
+<|ref|>text<|/ref|><|det|>[[118, 222, 877, 306]]<|/det|>
+The NEMS model is, as far as I understand, a CGE model for energy consuming sectors of the US economy, but it is not clear if supply- demand effects on price that are considered for the oil market (for example) are similarly considered for producers of non- energy materials (such as lithium and lithium batteries). In the present study, how is the effect of price considered on the material sectors - particularly when many of the "other" materials are not domestically produced in the US?  
+
+<|ref|>text<|/ref|><|det|>[[118, 319, 866, 348]]<|/det|>
+There is a particular practical issue of whether offshore emissions could be effectively included in the price or not.  
+
+<|ref|>text<|/ref|><|det|>[[118, 361, 875, 431]]<|/det|>
+Likewise, in the LCA data, the "other" materials would be expected to be important - otherwise the only(?) material difference between the vehicle types with functional relevance is copper? The remainder of the materials would appear to be structural? In this case, should the vehicle chassis be unified and a clearer delineation of the impacts that arise solely due to the vehicle power train given?  
+
+<|ref|>text<|/ref|><|det|>[[118, 431, 530, 445]]<|/det|>
+An inventory of the "other" materials would be useful.  
+
+<|ref|>text<|/ref|><|det|>[[118, 459, 847, 502]]<|/det|>
+Is there any consideration of the need for additional electricity generation to power a high BEV load in transport? (i.e. the electricity demand in the sector would increase, requiring additional capacity installation - not just a change from high CO2 to low CO2 mix)  
+
+<|ref|>text<|/ref|><|det|>[[118, 515, 872, 544]]<|/det|>
+It would be useful to include a diagram showing the flows of information and steps in the model to enhance understanding of the modelling approach.  
+
+<|ref|>text<|/ref|><|det|>[[118, 557, 822, 600]]<|/det|>
+Line 36 - is this \(26\%\) additional to the 1.5Gt emissions, or included? Line 48 - aren't these models referred to as Integrated Assessment Models, not specifically Integrated Energy Models?  
+
+<|ref|>text<|/ref|><|det|>[[119, 627, 415, 641]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 655, 875, 811]]<|/det|>
+This paper uses a quantitative model to simulate the transition from internal combustion engine vehicles to electric vehicles. The paper uses a life cycle analysis model of vehicle inputs and an energy model to look at how carbon pricing affects vehicle adoption. The paper shows that if renewables could provide more than \(75\%\) of electricity, then a carbon price on the extraction/processing of oil and the combustion of gasoline for conventional vehicles (and a similar carbon price for all other emissions like vehicle construction and the remaining polluting electricity sources) would lead to greater EV adoption than a carbon price just on gasoline combustion. The authors conclude that: "the simultaneous reduction of both direct and indirect emissions indicates a win- win situation for climate change mitigation, meaning that climate policy with very high shares of BEVs represents a no- regrets strategy." And furthermore, that the US "should target deployment of BEVs and largely disregard competing technologies."  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 825, 241, 838]]<|/det|>
+## Main comments  
+
+<|ref|>text<|/ref|><|det|>[[118, 839, 877, 908]]<|/det|>
+1. Low prices of renewables are not sufficient for large scale adoption. Electricity reliability requires that supply equal demand at all moments. When the sun is not shining and the wind is not blowing, how does a grid with \(75\%\) renewables operate? Without explicit modelling of the very large costs of storage, additional transmission, and investments to address issues of reliability, the model likely understates the costs of, or even overstates the feasibility of, large-scale renewable
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 822, 97]]<|/det|>
+deployment. A model that looks at the grid on an annual basis will not capture these issues.  
+
+<|ref|>text<|/ref|><|det|>[[118, 98, 700, 111]]<|/det|>
+2. Are fugitive methane emissions from extracting natural gas in the model?  
+
+<|ref|>text<|/ref|><|det|>[[118, 112, 825, 139]]<|/det|>
+3. The authors' conclusions ignore uncertainty. I suggest toning down the language like "no-regrets" and "disregard competing technologies".  
+
+<|ref|>text<|/ref|><|det|>[[118, 140, 870, 210]]<|/det|>
+4. In fact, showing how sensitive the paper's results are to the assumptions on renewables penetration would be helpful. If renewables do not get cheaper and coal plants do not retire, what does the model find? I would think that the indirect emissions of extracting coal and natural gas for producing electricity (especially if fugitive emissions are high) could be similar to the range of the indirect emissions from extracting and processing crude oil.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 238, 247, 251]]<|/det|>
+## Minor comments  
+
+<|ref|>text<|/ref|><|det|>[[118, 265, 833, 308]]<|/det|>
+5. Page 2, lines 34-35: gasoline production emissions presumably are in g CO2/km, not kWh.  
+6. Page 2, line 59 EIA stands for Energy Information Administration, not Agency.  
+7. Nature Communications does not use footnotes.  
+
+<|ref|>text<|/ref|><|det|>[[120, 350, 415, 363]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 377, 872, 504]]<|/det|>
+The submitted manuscript has investigated the life cycle environmental and economic implications of pricing, through the carbon tax, GHG emissions from passenger vehicles throughout their life cycles as an enviro- economical policy measure. In this pursuit, the researchers claim that pricing the GHG emissions associated with a vehicle's life cycle (though they have not provided a figure showing the system boundaries drawn for the life cycle assessment model) could be a more effective policy in reducing the environmental impacts of U.S. transportation than pricing only the GHG emissions from a vehicle's tailpipe. They further claim that such a policy measure could accelerate the phase- out of conventional vehicles while leading to higher penetration of battery electric vehicles and hence increased GHG savings.  
+
+<|ref|>text<|/ref|><|det|>[[118, 517, 875, 616]]<|/det|>
+Environmental policies have not been as efficient and effective as they are supposed to be for mitigating the negative impacts of anthropogenic activities (e.g. transportation). Pricing vehicle's life cycle emissions through carbon tax is one of such policies that is likely to have far- reaching implications in terms of the sustainability profile of U.S. passenger vehicles. Therefore, I think that gaining insights into these implications is of interest to the field of industrial ecology and others in the community and the wider field, especially given the significance of effective enviro- economical policies for mitigating the climate crisis studied in this field.  
+
+<|ref|>text<|/ref|><|det|>[[118, 629, 872, 826]]<|/det|>
+Two models - i) Yale- NEMS and ii) LCA model previously developed and published- were combined to carry out the quantitative analysis. Even though the supplementary information, including the input data, have been provided and includes data on many variables used in the models mentioned above, previously published works have been referred for some information, e.g. cost estimates for engines, electric motors, transmissions, fuel cells, and hydrogen storage tanks (and their specifications). The researchers are recommended to provide such information in a table, at least, in the SI. Furthermore, even though the source code for the LCA model has been provided, the mathematical notations of the model formulations employed in the study, as well as of the incorporation of the assumptions into these models, are missing. I would recommend that the researchers consider addressing these in the SI, at least if the word limit does not allow them to be addressed in the main manuscript. This would make it easier for the reader to have a better understanding of the analytical work done and for other researchers to be able to reproduce the results and build upon the models. Also, the researchers are recommended to provide a figure, depicting the system boundary for the LCA model developed.  
+
+<|ref|>text<|/ref|><|det|>[[118, 840, 790, 854]]<|/det|>
+I have submitted my comments as annotations and attached the annotated manuscript.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[117, 140, 430, 160]]<|/det|>
+## Response to the reviewers  
+
+<|ref|>text<|/ref|><|det|>[[117, 188, 353, 205]]<|/det|>
+Dear anonymous reviewers,  
+
+<|ref|>text<|/ref|><|det|>[[115, 228, 883, 343]]<|/det|>
+We would like to thank you very much for your thoughtful reviews of our previously submitted manuscript "Pricing indirect emissions accelerates low- carbon transition of US light vehicle sector". We have worked hard to successfully address all the concerns that have been expressed and we believe that the paper has improved as a result. We look forward to hearing from you.  
+
+<|ref|>text<|/ref|><|det|>[[117, 365, 259, 383]]<|/det|>
+Kindest regards,  
+
+<|ref|>text<|/ref|><|det|>[[115, 406, 460, 424]]<|/det|>
+Paul Wolfram (on behalf of all authors)  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 487, 249, 506]]<|/det|>
+## Reviewer 1  
+
+<|ref|>text<|/ref|><|det|>[[115, 521, 882, 565]]<|/det|>
+The overall premise of the paper is interesting, though the level of novelty is perhaps not the highest.  
+
+<|ref|>text<|/ref|><|det|>[[115, 570, 884, 902]]<|/det|>
+Thank you for this thought. We appreciate that you find the overall premise interesting. In the comment on the novelty, we believe that the reviewer may be referring to the fact that previous static life cycle assessments have highlighted the importance of the electricity mix for both charging vehicles and producing the vehicles in the first place. We agree with this assessment of the literature. However, the present work differs in being the first large- scale dynamic assessment of electric vehicle roll- out scenarios considering both the life cycle coefficients and the dynamic effects of the entire energy system at the same time. This combination of approaches is novel and leads to a new result to the literature: that the negative impacts that occur with electric vehicle adoption can be largely avoided. Thus, we see our manuscript to be of great interest to those engaged in the lively debate about EV policy. We would like to point out that the article pre- print on ResearchSquare attracted more than 300 viewers, with 100 alone within the first five days. This perhaps can be seen as at least one indication that the broader academic community will find the work novel.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 100, 881, 165]]<|/det|>
+The overall results could be more clearly described - particularly the relationship between pricing, materials and emissions towards changing the vehicle mix.  
+
+<|ref|>text<|/ref|><|det|>[[115, 173, 883, 431]]<|/det|>
+Thank you for this suggestion - it is one that we have taken seriously. One new aspect of this work above the previous literature is that the pricing of all emissions, including those stemming from material production, influence the optimal vehicle fleet. Thus, we feel it is especially important to document this part of our approach clearly and we appreciate your nudge to do so. We added three mathematical equations in the methods section providing additional detail on how embodied emissions from material production and energy chains affect vehicle prices. We further added two sections in the supplementary information which serve as a more detailed description of the interconnection between prices, materials and emissions (Supplementary Sections 2 and 10). We believe that this additional information complements the existing descriptions in the main manuscript well and we thank you again for this comment.  
+
+<|ref|>text<|/ref|><|det|>[[115, 449, 883, 637]]<|/det|>
+The NEMS model is, as far as I understand, a CGE model for energy consuming sectors of the US economy, but it is not clear if supply- demand effects on price that are considered for the oil market (for example) are similarly considered for producers of non- energy materials (such as lithium and lithium batteries). In the present study, how is the effect of price considered on the material sectors - particularly when many of the 'other' materials are not domestically produced in the US? There is a particular practical issue of whether offshore emissions could be effectively included in the price or not.  
+
+<|ref|>text<|/ref|><|det|>[[115, 643, 883, 902]]<|/det|>
+The reviewer's understanding of the model is correct. NEMS is indeed a CGE model focusing on the US economy. (In addition, the model includes world energy prices, world energy supply and demand, as well as US energy imports and exports. It is also correct that the market for these materials is not represented explicitly in NEMS.) In our model, the GHG emissions of these materials are part of the life- cycle assessment (LCA). Our single- region LCA model indeed assumes that vehicle production takes place in the US. We updated the text on lines 317- 320 accordingly: "For simplicity purposes, the LCA model assumes that vehicle production takes place in the US [...] (see Supplementary Section 10 for a discussion of the error invoked from these assumptions)." Accordingly, we added a new section (Supplementary Section 10) and there we added: "Further, our single- region LCA model assumes that vehicle production takes place in the US. In reality,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 98, 884, 455]]<|/det|>
+many of the vehicles bought in the US are made elsewhere in the world. On the other hand, quite a few are exported. \(^{1}\) We present evidence that the error invoked from disregarding these trade relations is small however. To that end we analyze the differences between the 'full pricing' and the 'well- to- wheel' pricing scenario. The 'full pricing' scenario fully prices embodied emissions of vehicle and battery production while the 'well- to- wheel pricing' scenario excludes pricing of vehicle and battery production emissions altogether. Hence, the cost of embodied vehicle and battery production emissions are zero under 'well- to- wheel' pricing. Yet, the differences in sales are marginal: merely a few longer- range EVs are partially replaced by shorter- range EVs (Figure S6h). Furthermore, these sales differences do not notably affect overall emissions outcomes (Figure 2a). As documented in Supplementary Table 2, the difference in total fleet- wide life cycle emissions between the two scenarios amounts to 71 Mt \(\mathrm{CO_2}\) cumulatively over the period 2010- 2050. Hence, the results of this study are fairly unsusceptible to assumptions regarding the carbon intensity of vehicle and battery production which in turn partly depend on the location of production (inside versus outside of the US).  
+
+<|ref|>text<|/ref|><|det|>[[115, 461, 884, 720]]<|/det|>
+In addition, the mentioned simplifications do not prevent us from capturing potential changes to material prices in our model, at least in a simplified way. Therefore, inspired by the reviewer's thoughtful feedback, we added six new sensitivity cases, of which four assume constant prices of EV batteries from 2021 on. This constant price can be interpreted either as insufficient investments into battery technology or as growing raw material prices. IEA's newly published World Energy Outlook \(^{2}\) states that "a doubling of lithium or nickel prices would induce a 6% increase in battery costs. If both lithium and nickel prices were to double at the same time, this would offset all the anticipated unit cost reductions associated with a doubling of battery production capacity." We note the results of this investigation in a new section called "Uncertainty analysis" as well as in Supplementary Section 6.  
+
+<|ref|>text<|/ref|><|det|>[[115, 739, 884, 855]]<|/det|>
+Likewise, in the LCA data, the 'other' materials would be expected to be important - otherwise the only(?) material difference between the vehicle types with functional relevance is copper? The remainder of the materials would appear to be structural? In this case, should the vehicle chassis be unified and a clearer delineation of the impacts that arise solely due to the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 100, 881, 142]]<|/det|>
+vehicle power train given? An inventory of the ‘other’ materials would be useful.  
+
+<|ref|>text<|/ref|><|det|>[[115, 150, 882, 240]]<|/det|>
+We apologize for not being fully clear in our previous draft. Accordingly, we added a caveat to lines 317- 320: “For simplicity purposes […] the model includes the most climate- relevant vehicle materials and disregards other minor materials (see Supplementary Section 10 for a discussion of the error invoked from these assumptions).”  
+
+<|ref|>text<|/ref|><|det|>[[115, 248, 883, 531]]<|/det|>
+In addition, we added to Supplementary Section 10: “our detailed process- based model allows for an explicit differentiation in composition and mass of the glider and power train. We consider the seven most common materials used for vehicle production: cast iron, stainless steel, automotive steel, wrought aluminum, cast aluminum, copper, and plastics. Combined these contribute more than 92% of the weight of the vehicle. The ‘other’ materials category is mostly comprised of glass and rubber, and for reasons of simplicity, we estimate the emissions of these other minor materials at 2 kg \(\mathrm{CO_2 / kg}\) material in the base year. This is well within the range of emission factors of rubber and glass which make up the vast majority of the ‘other’ category (by weight). For example, according to the ecoinvent 3.5 database using IPCC’s 2013 GWP- 100 indicator, the production of synthetic rubber emits 2.75 kg \(\mathrm{CO_2 / kg}\) material, while natural rubber and uncoated flat glass emit 2.02 and 0.99 kg \(\mathrm{CO_2 / kg}\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 538, 883, 868]]<|/det|>
+Other potentially important materials specific to EV batteries may be cobalt, nickel and lithium, which are not considered in our model. Comparing our inventory and GHG emission results to that of a recently published and very detailed study,3 these materials account for 15% of the emissions associated with battery production. This translates into an omission of about 6% for the production of the vehicle as a whole, and of merely 1% for the entire EV life cycle including charging electricity in the model’s base year. We acknowledge that the relative error could be higher in future years assuming a decarbonization of the electricity mix but not the metal production. However, simultaneous improvements in battery technology are conceivable as well. We therefore anticipate that including the embodied \(\mathrm{CO_2}\) emissions from cobalt, nickel and lithium would not notably change scenario outcomes.” As mentioned above, this assertion is also confirmed if one analyzes the small differences in vehicle sales and resulting fleet life- cycle emissions between the ‘full pricing’ scenario and the ‘well- to- wheel pricing’ scenario (Figure 2a, Figure S6h, Supplementary Table 2).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 100, 883, 191]]<|/det|>
+Is there any consideration of the need for additional electricity generation to power a high BEV load in transport? (i.e. the electricity demand in the sector would increase, requiring additional capacity installation - not just a change from high CO2 to low CO2 mix)  
+
+<|ref|>text<|/ref|><|det|>[[115, 197, 884, 360]]<|/det|>
+We completely agree that this is important. Fortunately, the model indeed considers the additional electricity generation to power a high BEV load. In fact, this was one of the reasons why we chose the model. The effects of the additional electricity load should be clear from the additional information provided in the SI, such as Supplementary Figure 12, showing the increase in overall electricity demand, and Supplementary Figure 14, illustrating new additions to electricity generation capacity as well as retirements of old power plants.  
+
+<|ref|>text<|/ref|><|det|>[[115, 377, 882, 421]]<|/det|>
+It would be useful to include a diagram showing the flows of information and steps in the model to enhance understanding of the modelling approach.  
+
+<|ref|>text<|/ref|><|det|>[[115, 427, 882, 470]]<|/det|>
+We thank the reviewer for this suggestion. We now provide a graph showing an overview of the modelling framework in Supplementary Section 10.  
+
+<|ref|>text<|/ref|><|det|>[[115, 488, 794, 507]]<|/det|>
+Line 36 - is this \(26\%\) additional to the 1.5Gt emissions, or included?  
+
+<|ref|>text<|/ref|><|det|>[[115, 513, 882, 556]]<|/det|>
+The \(26\%\) are included in the 1.5 Gt. We double- checked the sentence to make sure that our language is clear.  
+
+<|ref|>text<|/ref|><|det|>[[115, 574, 881, 618]]<|/det|>
+Line 48 - aren't these models referred to as Integrated Assessment Models, not specifically Integrated Energy Models?  
+
+<|ref|>text<|/ref|><|det|>[[115, 623, 883, 739]]<|/det|>
+The definition of the term 'Integrated Assessment Model' is not consistent throughout the literature. Given that most of these models do not contain an assessment of the damage of climate change, we prefer to use the term 'Integrated Energy Model' for which some of the authors provided a detailed definition in a recent peer- reviewed paper.4 We also provide this definition in a new section in the SI (Supplementary Section 9).  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 772, 250, 791]]<|/det|>
+## Reviewer 2  
+
+<|ref|>text<|/ref|><|det|>[[115, 808, 882, 875]]<|/det|>
+This paper uses a quantitative model to simulate the transition from internal combustion engine vehicles to electric vehicles. The paper uses a life cycle analysis model of vehicle inputs and an energy model to look at how
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 99, 884, 358]]<|/det|>
+carbon pricing affects vehicle adoption. The paper shows that if renewables could provide more than \(75\%\) of electricity, then a carbon price on the extraction/processing of oil and the combustion of gasoline for conventional vehicles (and a similar carbon price for all other emissions like vehicle construction and the remaining polluting electricity sources) would lead to greater EV adoption than a carbon price just on gasoline combustion. The authors conclude that: "the simultaneous reduction of both direct and indirect emissions indicates a win- win situation for climate change mitigation, meaning that climate policy with very high shares of BEVs represents a no- regrets strategy." And furthermore, that the US "should target deployment of BEVs and largely disregard competing technologies."  
+
+<|ref|>text<|/ref|><|det|>[[116, 380, 281, 397]]<|/det|>
+Main comments:  
+
+<|ref|>text<|/ref|><|det|>[[115, 419, 884, 606]]<|/det|>
+1. Low prices of renewables are not sufficient for large scale adoption. Electricity reliability requires that supply equal demand at all moments. When the sun is not shining and the wind is not blowing, how does a grid with \(75\%\) renewables operate? Without explicit modelling of the very large costs of storage, additional transmission, and investments to address issues of reliability, the model likely understates the costs of, or even overstates the feasibility of, large-scale renewable deployment. A model that looks at the grid on an annual basis will not capture these issues.  
+
+<|ref|>text<|/ref|><|det|>[[115, 615, 884, 850]]<|/det|>
+We agree with you that reliability must be included in the analysis. Fortunately, NEMS includes a dispatch model that determines electricity supply, demand and prices at sub- annual level (three seasons by three times of day). It is important to note that NEMS explicitly models the cost of additional investments needed to allow for intermittent renewable electricity generation capacity. This is a core part of the capacity additions modeling. Thus, NEMS has the characteristics you identify as crucial for exploring our research question. We apologize if we were unclear and implied that NEMS was run only at an annual basis and did not model reliability. We have rectified this by adding the following sentence (lines 276- 278): "A dispatch model determines electricity supply, demand and prices at sub- annual level (three seasons by three times of day)."  
+
+<|ref|>text<|/ref|><|det|>[[116, 859, 882, 901]]<|/det|>
+As for the feasibility of large- scale renewable development, we find that our results are backed by several previous studies indicating that such a high share of renewables
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 98, 883, 335]]<|/det|>
+is possible. These studies even include work published in Nature Communications. \(^{5}\) \(^{6}\) \(^{7}\) Of course achieving a high share of renewables is not the only way to achieve decarbonization of the electricity system. For example, see the careful discussion in the IPCC's recent Special Report on Global Warming of 1.5 C, which is widely accepted by the scientific community. In the United States, the Biden administration aims to decarbonize electricity generation with a Clean Electricity Standard as the central policy, which will likely encourage development of renewables, but also opens the door to other technologies. These questions of the cost and feasibility of high renewables are very interesting, but they are not the research question at hand in our paper. Rather, we view the high renewables scenario as a starting point for our analysis.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 354, 877, 374]]<|/det|>
+## 2. Are fugitive methane emissions from extracting natural gas in the model?  
+
+<|ref|>text<|/ref|><|det|>[[115, 380, 883, 639]]<|/det|>
+We apologize for our missing description of methane emissions. We agree with you that methane emissions from fossil fuel extraction and transportation can be significant and should ideally be taken into consideration. Unfortunately, NEMS only reports emissions of \(\mathrm{CO_2}\) and a range of air pollutants. Thus, we have now made this very clear by updating the model description accordingly (lines 248- 252): "Yale- NEMS provides a full account of \(\mathrm{CO_2}\) emissions across all industries and a range of air pollutants from vehicles and power plants. \(\mathrm{CO_2}\) accounted for \(97\%\) of total GHG emissions in the US electricity and transport sectors in 2019. \(^{8}\) Other GHGs such as methane emissions from fossil- fuel and hydroelectric power plants are not included." We also now include a more detailed description of natural gas \(\mathrm{CO_2}\) emissions factors used in this work in our new Supplementary Section 6.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 659, 881, 702]]<|/det|>
+## 3. The authors' conclusions ignore uncertainty. I suggest toning down the language like "no-regrets" and "disregard competing technologies".  
+
+<|ref|>text<|/ref|><|det|>[[115, 708, 883, 823]]<|/det|>
+We agree with you; our previous language was too strong. In response to your comment, we deleted the clause "disregarding competing technologies". In addition, we double- checked our language to make it more clear that BEVs are a "no regrets" strategy only if electricity decarbonizes, as has been assumed in our main scenarios. Examples include lines 20- 22: "Given continued decarbonization of electricity supply, results show
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 98, 884, 383]]<|/det|>
+that a large- scale adoption of electric vehicles is able to reduce \(\mathrm{CO_2}\) emissions through more channels than previously expected"; lines 204- 208: "In fact, the simultaneous reduction of both direct and indirect emissions indicates a win- win situation for climate change mitigation, meaning that climate policy with very high shares of BEVs represents a no- regrets strategy (but only if electricity continues to decarbonize as has been assumed in our main scenarios)"; and lines 404- 407: "In terms of emissions, a carbon tax on supply chain emissions is not able to yield the desired results if the electricity grid does not face substantial decarbonization. In this case, pricing supply chain emissions leads to higher emissions compared to pricing direct emissions only (see Supplementary Section 6.1 for more details and results)". Our new sensitivity cases which have been requested by the reviewer now make it clear that large adverse effects can occur if electricity slowly or barely decarbonizes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 400, 884, 540]]<|/det|>
+4. In fact, showing how sensitive the paper's results are to the assumptions on renewables penetration would be helpful. If renewables do not get cheaper and coal plants do not retire, what does the model find? I would think that the indirect emissions of extracting coal and natural gas for producing electricity (especially if fugitive emissions are high) could be similar to the range of the indirect emissions from extracting and processing crude oil.  
+
+<|ref|>text<|/ref|><|det|>[[115, 545, 884, 803]]<|/det|>
+We thank the reviewer for this thoughtful comment. As indicated above we include six new scenarios which explore the uncertainty in the costs of electric vehicles and renewable power plants (see section "Uncertainty analysis" and Supplementary Section 6). From the results of these scenarios it becomes apparent that the reviewer is correct in their assertion that the emissions from fossil- fuel based electricity could exceed those of gasoline production if the electricity sector fails to decarbonize substantially. In addition, we now provide a full list and brief description of our 28 scenarios in Supplementary Section 4. We believe that these 28 scenarios more than adequately explore the main uncertainties present in this work. The results of the extreme (i.e., most optimistic and most pessimistic) scenarios are described in detail throughout the main manuscript and the SI.  
+
+<|ref|>text<|/ref|><|det|>[[117, 823, 292, 840]]<|/det|>
+Minor comments:  
+
+<|ref|>text<|/ref|><|det|>[[115, 859, 884, 902]]<|/det|>
+5. Page 2, lines 34-35: gasoline production emissions presumably are in g CO2/km, not kWh.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 100, 885, 310]]<|/det|>
+We chose to provide gasoline production emissions in \(\mathrm{gCO_2 / kWh}\) instead of \(\mathrm{gCO_2 / km}\) in order to avoid additional assumptions on vehicle fuel consumption (e.g., \(\mathrm{kWh / 100km}\) ). We deeply considered providing \(\mathrm{gCO_2 / km}\) numbers along with the \(\mathrm{gCO_2 / kWh}\) numbers but we ask the reviewer for understanding that we refrained from this idea in the end. With the range of different vehicle segments and technologies analyzed in this work, we worry it would make the sentence too convoluted and thus difficult to communicate the key point to readers. However, we decided to provide values in \(\mathrm{gCO_2 / gallon}\) gasoline equivalents in Supplementary Section 1. We hope that this is an adequate solution and if this is a crucial point to you, we are open to reassessing.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 330, 881, 373]]<|/det|>
+## 6. Page 2, line 59 EIA stands for Energy Information Administration, not Agency.  
+
+<|ref|>text<|/ref|><|det|>[[155, 379, 632, 398]]<|/det|>
+Thank you for spotting this. We corrected accordingly.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 417, 623, 435]]<|/det|>
+## 7. Nature Communications does not use footnotes.  
+
+<|ref|>text<|/ref|><|det|>[[152, 443, 879, 461]]<|/det|>
+Thank you again. We made sure to avoid footnotes throughout the entire manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 497, 250, 517]]<|/det|>
+## Reviewer 3  
+
+<|ref|>text<|/ref|><|det|>[[115, 545, 884, 903]]<|/det|>
+The submitted manuscript has investigated the life cycle environmental and economic implications of pricing, through the carbon tax, GHG emissions from passenger vehicles throughout their life cycles as an enviro- economical policy measure. In this pursuit, the researchers claim that pricing the GHG emissions associated with a vehicle's life cycle (though they have not provided a figure showing the system boundaries drawn for the life cycle assessment model) could be a more effective policy in reducing the environmental impacts of U.S. transportation than pricing only the GHG emissions from a vehicle's tailpipe. They further claim that such a policy measure could accelerate the phase- out of conventional vehicles while leading to higher penetration of battery electric vehicles and hence increased GHG savings. Environmental policies have not been as efficient and effective as they are supposed to be for mitigating the negative impacts of anthropogenic activities (e.g. transportation). Pricing vehicle's life cycle emissions through carbon tax is one of such policies that is likely to have far- reaching implications in terms of the sustainability profile
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 99, 883, 191]]<|/det|>
+of U.S. passenger vehicles. Therefore, I think that gaining insights into these implications is of interest to the field of industrial ecology and others in the community and the wider field, especially given the significance of effective enviro- economical policies for mitigating the climate crisis studied in this field.  
+
+<|ref|>text<|/ref|><|det|>[[115, 199, 882, 242]]<|/det|>
+We deeply thank the reader for this assessment and very much share the view that this research is of interest to a wide range of research communities.  
+
+<|ref|>text<|/ref|><|det|>[[115, 250, 884, 439]]<|/det|>
+Two models - i) Yale- NEMS and ii) LCA model previously developed and published- were combined to carry out the quantitative analysis. Even though the supplementary information, including the input data, have been provided and includes data on many variables used in the models mentioned above, previously published works have been referred for some information, e.g. cost estimates for engines, electric motors, transmissions, fuel cells, and hydrogen storage tanks (and their specifications). The researchers are recommended to provide such information in a table, at least, in the SI.  
+
+<|ref|>text<|/ref|><|det|>[[115, 446, 884, 658]]<|/det|>
+We apologize for not having included this information previously. We now provide detailed cost figures for all modeled powertrain components in Supplementary Table 9 in the provided Excel spreadsheet. Similarly, detailed specifications of individual vehicles and vehicle components, including battery capacity and weight, battery depth of discharge, weight of motors, energy consumption, motorization, purchase price, fuel tank size, and total vehicle weight, have been included in Supplementary Table 11. Scenario- specific fleet- average fuel efficiencies, purchase prices, and degree of lightweighting are endogenous modelling results of NEMS, and have been documented in Supplementary Tables 5, 14 and 12.  
+
+<|ref|>text<|/ref|><|det|>[[115, 667, 884, 903]]<|/det|>
+Furthermore, even though the source code for the LCA model has been provided, the mathematical notations of the model formulations employed in the study, as well as of the incorporation of the assumptions into these models, are missing. I would recommend that the researchers consider addressing these in the SI, at least if the word limit does not allow them to be addressed in the main manuscript. This would make it easier for the reader to have a better understanding of the analytical work done and for other researchers to be able to reproduce the results and build upon the models. Also, the researchers are recommended to provide a figure, depicting the system boundary for the LCA model developed.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 100, 883, 239]]<|/det|>
+Thank you for this important suggestion. We now provide the three central equations of our approach in the methods section. For additional equations describing the LCA model and NEMS, interested readers can consult Wolfram et al. \(^{9}\) and the NEMS model documentation. We made sure that these references are shown more prominently in the methods section. In addition, as noted above, we now provide a figure of the modeling approach and the systems boundary in Supplementary Section 2.  
+
+<|ref|>text<|/ref|><|det|>[[115, 258, 882, 301]]<|/det|>
+I have submitted my comments as annotations and attached the annotated manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 319, 883, 531]]<|/det|>
+Line 30: The reference number 3 investigates the life cycle energy and environmental assessment of natural gas as transportation fuel in Pakistan and is not quite relevant for the statement it was given to support it. Given that the spatial and technological scope of the study is the United States, the reference number 3 can be replaced with a more relevant reference, e.g. having the same spatial scope such as Onat, N., Kucukvar, M., and Tatari O. (2014). Towards Life Cycle Sustainability Assessment of Alternative Passenger Vehicles. Sustainability 6 (12). Here, please, consider citing that work, instead, or adding that reference, as well, to acknowledge quite a relevant work.  
+
+<|ref|>text<|/ref|><|det|>[[155, 538, 461, 555]]<|/det|>
+We added this important reference.  
+
+<|ref|>text<|/ref|><|det|>[[115, 574, 881, 617]]<|/det|>
+Lines 31- 32: This reads like an incomplete sentence and I could not understand. Please, consider revising this sentence.  
+
+<|ref|>text<|/ref|><|det|>[[115, 624, 882, 738]]<|/det|>
+We revised this sentence accordingly in order to make it more understandable: "These emissions occur off- site, or indirectly, and include generation of electricity to charge electric vehicles, in this work \(\sim 66 - 86 \mathrm{g} \mathrm{CO}_{2}\) per electric- vehicle km driven in 2020, as well as the production of vehicles, here \(\sim 16 - 38 \mathrm{g} \mathrm{CO}_{2}\) per vehicle- km driven in 2020 (Supplementary Section 1 and Supplementary Table 1)."  
+
+<|ref|>text<|/ref|><|det|>[[115, 757, 881, 824]]<|/det|>
+Line 34: Could the researchers explain why they preferred using such a unit when referring to the emissions from gasoline production? Did they use the conversion factor of \(1 \mathrm{kWh} = 0.03 \mathrm{GGE}\) ?  
+
+<|ref|>text<|/ref|><|det|>[[115, 831, 881, 873]]<|/det|>
+We use kWh because it is a unit listed in the International System of Units (SI). For better readability we would wish to stick to kWh in the main text. In Supplementary
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 100, 881, 141]]<|/det|>
+Section 1 however, we decided to also provide values in GGE using the conversion factor that the reviewer kindly provided.  
+
+<|ref|>text<|/ref|><|det|>[[115, 161, 882, 252]]<|/det|>
+Line 39: Since they are the main subject of the study, the researchers are recommended to indicate more clearly (maybe, in paranthesis) what these indirect emissions include to get rid of the confusion caused by the ambiguity of the term.  
+
+<|ref|>text<|/ref|><|det|>[[115, 260, 882, 350]]<|/det|>
+Agreed. We changed the sentence accordingly. It now reads as follows (now lines 38- 40): "The introduction of the Low Carbon Fuel Standard (LCFS) in California, which regulates all fuel and electricity production and combustion emissions, shows that transport policy in practice can at least partly address indirect vehicle emissions."  
+
+<|ref|>text<|/ref|><|det|>[[115, 369, 882, 460]]<|/det|>
+Lines 44- 45: The effect on production decision of changing costs due to regulatory standards such as CAFE has been investigated before. The researchers are recommended to refer to the study titled CAFE's impact on the market share of electric vehicles by Sen et al. (2017) to acknowledge a relevant work.  
+
+<|ref|>text<|/ref|><|det|>[[155, 468, 486, 485]]<|/det|>
+We included this interesting reference.  
+
+<|ref|>text<|/ref|><|det|>[[115, 506, 679, 524]]<|/det|>
+Line 75: What is meant by this term? Please, elaborate.  
+
+<|ref|>text<|/ref|><|det|>[[115, 532, 882, 620]]<|/det|>
+We changed the sentence to make it more clear. It now reads (now lines 77- 79): "Here we address these knowledge gaps by applying a novel conceptual framework by Creutzig et al., which focuses on energy- demand side (rather than energy- supply side) solutions to climate change mitigation."  
+
+<|ref|>text<|/ref|><|det|>[[115, 641, 883, 828]]<|/det|>
+Lines 79- 80: These sectors are already parts of an EV's life cycle under the IO modeling setting (which can be represented as unit processes under the process modeling settin). Such responses are reflected upon in the life cycle sustainability assessment studies of different vehicle classes, if I have understood this sentence, correctly. That's why the researchers are recommended to clarify what are meant by direct and indirect emissions and by 'vehicle sectors', and provide a figure, depicting the system boundary adopted for this study.  
+
+<|ref|>text<|/ref|><|det|>[[115, 836, 881, 901]]<|/det|>
+The reviewer is absolutely correct in stating that the sectors material production, vehicle manufacturing and electric charging are reflected and adequately linked to each other in the LCA model. On the other hand, NEMS has the advantage of better captur
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 100, 882, 215]]<|/det|>
+ing the non- linear dynamics of the entire energy system including changes in electricity supply. In order to make use of the advantages of both models, we soft- link our LCA model to NEMS. We hope that this becomes clear from our description, especially now that we included a new figure depicting the system boundary of this work (Supplementary Section 2).  
+
+<|ref|>text<|/ref|><|det|>[[115, 234, 882, 325]]<|/det|>
+Line 87: So, Scenario 1 assumes that the emissions from the tailpipe are accounted for, priced, and optimized for, whereas Scenario 2 assumes that the emissions from the entire vehicle supply chain are accounted for, priced, and optimized for. Is that correct?  
+
+<|ref|>text<|/ref|><|det|>[[115, 332, 881, 375]]<|/det|>
+Yes, the reviewer is correct. We double- checked our scenario descriptions throughout the main manuscript and the SI and we hope that they are adequate and clear.  
+
+<|ref|>text<|/ref|><|det|>[[115, 394, 883, 533]]<|/det|>
+Line 95: Does "direct tailpipe emissions" mean that tailpipe emissions are the direct emissions? I believe there is a need to provide a clarification as to what are meant by direct and indirect emissions. Are direct emissions those from tailpipe or from vehicle sectors? Similarly, are indirect emissions those from non- tailpipe emissions or from material extraction sector? The researchers are recommended to clearly define these emissions to avoid any confusion.  
+
+<|ref|>text<|/ref|><|det|>[[115, 540, 881, 608]]<|/det|>
+The reviewer interpreted correctly what is meant by 'direct tailpipe emissions'. In order to avoid any ambiguities we state on line 35 that we use the term 'direct emissions' and 'tailpipe emissions' as synonyms.  
+
+<|ref|>text<|/ref|><|det|>[[115, 626, 876, 646]]<|/det|>
+Line 104: Is this referring to the "well to tank" emissions? Please elaborate.  
+
+<|ref|>text<|/ref|><|det|>[[115, 652, 882, 742]]<|/det|>
+Energy- chain emissions describe emissions invoked by the production and use of energy carriers (well- to- wheel). We added to line 104 (now line 107) that the term 'energy- chain' emissions is synonymous with 'well- to- wheel' emissions in order to avoid confusion.  
+
+<|ref|>text<|/ref|><|det|>[[115, 761, 882, 877]]<|/det|>
+Lines 104- 105: The researchers assumed that hydrogen production becomes carbon- netural by 2050 through the hydrogen production from biomethane, with CCS. Why have the researchers not considered the green hydrogen production, at all? How do they think that such a consideration would affect the conclusions of the study?  
+
+<|ref|>text<|/ref|><|det|>[[150, 884, 880, 903]]<|/det|>
+In Supplementary Section 1 we provide an example on how carbon neutral produc
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 98, 884, 431]]<|/det|>
+tion of hydrogen could be achieved. We added the reviewer's example of a green hydrogen pathway to the text. We also explain how this pathway would influence modelling outcomes. The section now reads as follows: "Net- zero could be achieved in various different ways, for example by producing hydrogen exclusively from wind or solar power or by using a hydrogen production mix consisting primarily of hydrogen from biomethane with carbon capture and storage, representing a carbon sink with about \(- 36 \mathrm{g} \mathrm{CO}_{2} \mathrm{e} / \mathrm{kWh}\) , and a small remainder, around \(7 - 8\%\) , of hydrogen from SMR, emitting around \(450 \mathrm{g}\) . Since carbon- neutral hydrogen production is only considered in one of our side cases it has been modeled in less detail: For the carbon- neutral hydrogen side case NEMS only receives the hydrogen production emissions factor from the LCA model, which can be assumed equal under both production pathways. Hence, for the purpose of this paper, whether hydrogen is produced from renewables or from a combination of biomethane CCS and natural gas, has no effect on the emissions outcomes because both production pathways would ensure net- zero emissions."  
+
+<|ref|>text<|/ref|><|det|>[[115, 448, 884, 565]]<|/det|>
+Lines 113- 114: Here the researchers are recommended to cite a study on the material footprint of electric vehicles through MRiO, conducted by Sen et al. (2019). This is one of the very few studies in the literature that shows the material intensity of EVs, which is very relevant to cite to acknowledge the previous work in this regard.  
+
+<|ref|>text<|/ref|><|det|>[[154, 569, 636, 588]]<|/det|>
+Thank you for this suggestion. We added the reference.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 606, 881, 625]]<|/det|>
+## Line 116: Are these ones different from the a, b, and c previously mentioned?  
+
+<|ref|>text<|/ref|><|det|>[[115, 630, 884, 746]]<|/det|>
+They are in fact the same. In order to avoid confusion we added a reference to Supplementary Figure 6 and changed the sentence. It now reads (now lines 120- 123): "As mentioned earlier, we explore a range of side cases (Supplementary Section 3) which show some variation in their potential for emission reductions (also see dotted lines in Figure 2a- j) but the overall trend is robust among these cases."  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 763, 880, 806]]<|/det|>
+## Line 152: The researchers are recommended to provide provide the percentages, as well, to help the reader relate better.  
+
+<|ref|>text<|/ref|><|det|>[[115, 811, 883, 903]]<|/det|>
+We made sure that percentages are provided as well. Please see the previous paragraph on lines 151 to 158. In addition, relative changes are also indicated in Figure 4. We would also like to note that all underlying data for Figure 4 is included in Supplementary Table 7 in the accompanying Excel spreadsheet.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 100, 883, 215]]<|/det|>
+Line 172: Given how significantly battery manufacturing influences the sustainability impacts of EVs, why have the researchers not considered discussing the implications of such a policy measure in terms of the future of EV batteries and how increasing demand on BEVs would likely influence battery technologies and the demand for them, as well as their prices?  
+
+<|ref|>text<|/ref|><|det|>[[115, 220, 883, 384]]<|/det|>
+We agree with the reviewer that EV batteries have a significant influence on sustainability issues. Our focal point is however not on emissions from batteries in particular but rather on the overall influence of indirect emissions on optimal vehicle fleets. We ask for the reviewer's understanding that, due to the limited space in the main manuscript, we had to focus on a high level discussion of the topic. However, in response to your comment, we added some additional discussion items on vehicle and battery materials to Supplementary Section 10.  
+
+<|ref|>text<|/ref|><|det|>[[115, 401, 882, 492]]<|/det|>
+Lines 187- 188: This is a common finding of the studies on vehicle LCA that do not consider pricing, at all. So, how does the pricing affect this? Through a higher market penetration of BEVs thanks to taxing the full LC emissions that influence the consumer's purchase behavior?  
+
+<|ref|>text<|/ref|><|det|>[[115, 497, 884, 635]]<|/det|>
+The reviewer is correct again. A useful aspect of our study is the consideration of prices and the ability to take consumer decisions into account. As the reviewer points out correctly, this ability of the model enables us to study the effects of holistic pricing of all embodied emissions of different vehicle options. The model finds that a pricing of all emission sources leads to a different optimal solution than pricing of tailpipe emissions alone.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 654, 670, 672]]<|/det|>
+## Line 189: Does this refer to the extraction of crude oil?  
+
+<|ref|>text<|/ref|><|det|>[[115, 679, 884, 792]]<|/det|>
+We thank the reviewer for spotting this. In fact, we wanted to refer to the entire production process of crude oil, including both exploration and extraction. In order to avoid any ambiguities we rephrased the sentence. It now reads as follows (now lines 197- 198): "However, higher electricity emissions are more than offset by lower gasoline supply- chain emissions stemming from the production of crude oil (Figure 2k)."  
+
+<|ref|>text<|/ref|><|det|>[[115, 811, 909, 878]]<|/det|>
+Line 256: The researchers are recommended to provide these assumptions/values, along with their references, in a table. In fact, given the large scope of the study, it might even be better to provide all your assumptions in a table.  
+
+<|ref|>text<|/ref|><|det|>[[150, 884, 881, 902]]<|/det|>
+We agree with you completely and fully support the reproducibility of scientific re
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 100, 882, 165]]<|/det|>
+sults. We therefore provide all cost assumptions of various vehicle components in Supplementary Table 9 in the provided Excel spreadsheet. In the same spreadsheet we provide 21 additional data tabs representing an exhaustive modelling database.  
+
+<|ref|>text<|/ref|><|det|>[[115, 185, 882, 252]]<|/det|>
+Line 269: The researchers are recommended to consider normalizing Fig. 8b or present a separate figure, with normalized price information per BTU, for example.  
+
+<|ref|>text<|/ref|><|det|>[[115, 260, 882, 326]]<|/det|>
+We provide an additional panel in Supplementary Figure 8 depicting normalized gasoline and electricity prices per BTU as requested by the reviewer. For conversion we use conversion factors provided by the U.S. EIA. \(^{10}\)  
+
+<|ref|>text<|/ref|><|det|>[[115, 345, 882, 412]]<|/det|>
+Lines 273- 275: How was this formulated and incorporated into the model? The researchers are recommended to consider providing the mathematical notations of all their formulations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 418, 883, 629]]<|/det|>
+In this instance, apart from updating the costs of rooftop solar PVs, no changes have been made to the NEMS model by the authors. Due to space constraints unfortunately we cannot reproduce the detailed equations and descriptions contained in the NEMS modelling documentation. However, we added a reference to the specific section of the NEMS documentation (section "Distributed Generation and Combined Heat and Power (CHP) Submodule" within the "Commercial Demand" section of the NEMS documentation). Interested readers will be able to find all the required information there. Please also refer to the new figure in Supplementary Section 2 where we depict a simplified representation of the commercial and residential buildings sectors.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 650, 670, 668]]<|/det|>
+## Lines 276-279: Are these the researchers' assumptions?  
+
+<|ref|>text<|/ref|><|det|>[[115, 676, 883, 789]]<|/det|>
+This paragraph describes future electricity demand growth, the development of electricity emissions as well as the resulting electricity carbon intensity, all of which are NEMS modelling results. These developments are a direct result of our cost assumptions on renewable electricity generators. We double- checked lines 274 to 277 to make sure that these assumptions are clearly explained there.  
+
+<|ref|>text<|/ref|><|det|>[[115, 809, 882, 850]]<|/det|>
+Line 303: What do C\$, LCC\$, WTW\$. mean, and what do the values provided represent, exactly?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 100, 883, 215]]<|/det|>
+We apologize for this mishap. These acronyms used in the supplementary Excel spreadsheet are only meant for internal use. We renamed them accordingly in order to be consistent with the scenario names in the manuscript. We made sure that the first tab of the Excel spreadsheet file describes the content of each following tab. In addition, we double- checked that each tab has a descriptive header and that all units are provided.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 234, 881, 301]]<|/det|>
+## Lines 309–310: Overall, I think there is a lack of mathematical representation of the model(s) formulations and of the incorporation of assumptions into these formulas. Please consider providing the mathematical notations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 308, 883, 422]]<|/det|>
+As mentioned above, we now provide the three central equations of our approach in the methods section. For additional equations describing the LCA model as well as NEMS, interested readers can consult Wolfram et al. \(^{11}\) as well as the NEMS model documentation. We made sure that these references are shown more prominently in the methods section.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 442, 881, 509]]<|/det|>
+## Line 327: Why have the battery capacities been assumed constant after 2025? And, have the researchers assumed any improvements in the GHG efficiency of battery production over the years?  
+
+<|ref|>text<|/ref|><|det|>[[115, 515, 883, 871]]<|/det|>
+In order to address these important questions, we added two new sections (Supplementary Sections 6.2 and 11). We address the first part of the question in Supplementary Section 6.2: “A standard assumption of Yale–NEMS is that of constant EV battery capacities (as well as constant battery weights and densities and hence EV ranges) after 2025 (Supplementary Table 11). In most of our scenario runs we adopted this assumption because, on the one hand, batteries could become more energetically dense in the future, hence requiring smaller capacities. On the other hand, larger capacities may be needed if BEVs continue to increase in driving range. Both factors could cancel each other out, hence the constant capacity assumption. However, in this section we present two sensitivity cases that explore the effects of increasing battery densities, leading to smaller, lighter, less material-intensive and cheaper EV batteries while providing the same driving range. We assume that – averaged over all technologies – battery densities continue to increase by about 1.5% per year after 2025. This rate is somewhat higher than the assumed average 2010–2025 increase of about 0.9% per year. This development helps especially with cost reductions of longer–range BEVs. This is true under both direct–emissions–only pricing
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 100, 883, 191]]<|/det|>
+and marginally more so under full- emissions pricing because the emissions penalty of the smaller batteries is also lowered. Hence, a shift in sales from 100- mile to 200- mile range BEVs can be observed. In addition, this trend is accompanied by an overall increase in BEV sales shares (Figure S10). "  
+
+<|ref|>text<|/ref|><|det|>[[115, 196, 884, 455]]<|/det|>
+The second part of the question is addressed in Supplementary Section 11: "The GHG intensity of batteries in particular (and vehicle production in general) improves in several ways in our model. The main scenarios assume a falling carbon intensity of electricity, which reduces both the emissions during the material production stage as well as emissions invoked during the battery assembly stage (Supplementary Table 18). The assumed energy mix of the material production and battery assembly stages is comprised of heat from fossil fuels as well as electricity (see Supplementary Tables 19, 20). In addition, some scenarios assume improved recycling of materials and reuse of components such as batteries (see Supplementary Tables 21, 22), further reducing GHG emissions." We also hope that the figure provided in Supplementary Section 2 further helps to communicate these relationships.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[120, 85, 444, 98]]<|/det|>
+Reviewer comments, second round review -  
+
+<|ref|>text<|/ref|><|det|>[[120, 136, 415, 150]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[120, 165, 400, 179]]<|/det|>
+The paper has been much improved.  
+
+<|ref|>text<|/ref|><|det|>[[118, 179, 875, 264]]<|/det|>
+I would prefer that the figure describing the modelling was in the main text rather than the supplementary material, as it would better give context to the results. The paper as a whole tends to lean on the supplementary material to a significant extent, which detracts from understanding - perhaps the authors could add some additional commentary to the supplementary material so that it could be read more as a report, so that those who are interested in understanding in detail could skip the paper and go to the supplement.  
+
+<|ref|>text<|/ref|><|det|>[[120, 304, 415, 319]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[120, 333, 714, 348]]<|/det|>
+Thank you for addressing my previous concerns. I have no further comments.  
+
+<|ref|>text<|/ref|><|det|>[[120, 375, 415, 390]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 403, 870, 529]]<|/det|>
+I thank the authors for addressing and clarifying all my comments and concerns regarding the submitted manuscript. Nature, as a leading multidisciplinary science journal, is a source of scientific knowledge that is highly trusted both by academia and society, in general. Every single sentence that is written in any manuscript that is to be potentially published in Nature must be paid due attention, and all the assumptions considered in any analysis must be justified e.g. by means of providing relevant publication(s), accordingly. So, as my last comment, I suggest that the authors make an extra effort to highlight, at least, any significant assumptions throughout the manuscript that will enhance the understanding of the readers, especially the primary target audience.  
+
+<|ref|>text<|/ref|><|det|>[[118, 542, 839, 571]]<|/det|>
+Overall, given the comment above, the manuscript is suggested for publication after that very minor revision.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[117, 156, 430, 176]]<|/det|>
+## Response to the reviewers  
+
+<|ref|>text<|/ref|><|det|>[[117, 206, 353, 222]]<|/det|>
+Dear anonymous reviewers,  
+
+<|ref|>text<|/ref|><|det|>[[117, 248, 882, 315]]<|/det|>
+We would like to thank you again for reviewing our revised manuscript entitled "Pricing indirect emissions accelerates low- carbon transition of US light vehicle sector". We incorporated all of your remaining suggestions and look forward to hearing from you.  
+
+<|ref|>text<|/ref|><|det|>[[117, 341, 259, 358]]<|/det|>
+Kindest regards,  
+
+<|ref|>text<|/ref|><|det|>[[116, 384, 460, 402]]<|/det|>
+Paul Wolfram (on behalf of all authors)  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 480, 249, 499]]<|/det|>
+## Reviewer 1  
+
+<|ref|>text<|/ref|><|det|>[[115, 520, 884, 708]]<|/det|>
+The paper has been much improved. I would prefer that the figure describing the modelling was in the main text rather than the supplementary material, as it would better give context to the results. The paper as a whole tends to lean on the supplementary material to a significant extent, which detracts from understanding - perhaps the authors could add some additional commentary to the supplementary material so that it could be read more as a report, so that those who are interested in understanding in detail could skip the paper and go to the supplement.  
+
+<|ref|>text<|/ref|><|det|>[[115, 715, 884, 902]]<|/det|>
+Thank you for this comment. In response to your suggestion, we copied a simplified, more compact version of the figure describing the modelling framework into the main manuscript. The full version of the figure would require many more words in the main text, muddling the flow of the manuscript and requiring cuts elsewhere, but we believe that the compact version allows us to convey the main ideas in a concise way. We also added a new section to the beginning of the supplementary material briefly summarizing the results of the paper. In addition, we strictly adhered to the style guidelines of Nature Communications when preparing the supplementary material.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 99, 250, 118]]<|/det|>
+## Reviewer 2  
+
+<|ref|>text<|/ref|><|det|>[[116, 145, 880, 189]]<|/det|>
+Thank you for addressing my previous concerns. I have no further comments. Thank you again!  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 218, 250, 238]]<|/det|>
+## Reviewer 3  
+
+<|ref|>text<|/ref|><|det|>[[114, 264, 883, 549]]<|/det|>
+I thank the authors for addressing and clarifying all my comments and concerns regarding the submitted manuscript. Nature, as a leading multidisciplinary science journal, is a source of scientific knowledge that is highly trusted both by academia and society, in general. Every single sentence that is written in any manuscript that is to be potentially published in Nature must be paid due attention, and all the assumptions considered in any analysis must be justified e.g. by means of providing relevant publication(s), accordingly. So, as my last comment, I suggest that the authors make an extra effort to highlight, at least, any significant assumptions throughout the manuscript that will enhance the understanding of the readers, especially the primary target audience. Overall, given the comment above, the manuscript is suggested for publication after that very minor revision.  
+
+<|ref|>text<|/ref|><|det|>[[115, 553, 883, 693]]<|/det|>
+Thank you for this suggestion. We made several final changes that further improved the readability and the understanding of our manuscript. First, we restructured the introduction. Second, we included a simplified version of Figure S3 in the main text. Finally, we carefully double- checked each and every sentence and made sure that all assumptions made in our work are clearly documented. We strongly believe that the manuscript is now ready for publication in Nature Communications.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__7360ebdf58d8c1045eda153ebd75eac6739f0dd879ad16ad9a400079e48134a8/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__7360ebdf58d8c1045eda153ebd75eac6739f0dd879ad16ad9a400079e48134a8/supplementary_0_Peer Review File__7360ebdf58d8c1045eda153ebd75eac6739f0dd879ad16ad9a400079e48134a8.mmd

@@ -0,0 +1,67 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Injectable hydrogel electrodes as conduction highways to restore native pacing  
+
+![PLACEHOLDER_0_0]
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+Editorial Note: This manuscript has been previously reviewed at another journal that is not operating a transparent peer review scheme. This document only contains reviewer comments and rebuttal letters for versions considered at Nature Communications.  
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The novelty of the approach with the initial experiments are interesting. The authors responded to earlier comments recognizing the need for additional experiments and development as well as longer- term follow- up.  
+
+Reviewer #2 (Remarks to the Author):  
+
+Given the journal transfer, I do not think additional data is required. However, text changes are needed per previous comments to more accurately reflect the data presented in the manuscript:  
+
+The added limitation at the end of the manuscript is good, but all mention of testing the hydrogel in scarred myocardium or labelling ablated tissue as "scar" needs to be removed. The tissue was acutely evaluated after ablation. It should be referred to what it is (ablated tissue), not as scar since no scar has had time to develop.  
+
+Similarly, the manuscript should be revised to say potentially deliverable via dual lumen catheter and not described as a given since the authors have still not provided data showing this is feasible without clogging the tip of the catheter, which is not trivial. This reviewer appreciates this is in development, but it is not yet achieved and should not be stated as fact in the manuscript. Given that S31 just shows a prototype without delivering the material and therefore does not address the original concern, I would suggest this be removed from this manuscript, so it does not interfere with publishing a separate manuscript on the catheter once it is fully developed.  
+
+Other limitations have been adequately addressed with the revision.
+
+<--- Page Split --->
+
+Reviewer #3 (Remarks to the Author):  
+
+The authors have nicely addressed the comments. This work is suitable for publication in Nature Communications.
+
+<--- Page Split --->
+
+## RESPONSE TO REVIEWERS:  
+
+RESPONSE TO REVIEWERS:The authors would like to thank the reviewers for their comments and enthusiasm for the innovative nature of this approach. We appreciate that the editorial team agreed that long- term validation was not necessary for this manuscript and have included detailed responses to individual reviewer comments below as well as a more detailed discussion of the evidence that supports the translational potential of this technology in the revised manuscript.  
+
+## REVIEWER #1  
+
+The novelty of the approach with the initial experiments are interesting. The authors responded to earlier comments recognizing the need for additional experiments and development as well as longer- term follow- up.  
+
+## REVIEWER #2  
+
+Given the journal transfer, I do not think additional data is required. However, text changes are needed per previous comments to more accurately reflect the data presented in the manuscript:  
+
+Comment 1: The added limitation at the end of the manuscript is good, but all mention of testing the hydrogel in scarred myocardium or labelling ablated tissue as "scar" needs to be removed. The tissue was acutely evaluated after ablation. It should be referred to what it is (ablated tissue), not as scar since no scar has had time to develop.  
+
+Response: We have removed the remaining reference to "scar" from the SI figures and text and referred to it as ablation lesion per standard terminology for this model.  
+
+Comment 2: Similarly, the manuscript should be revised to say potentially deliverable via dual lumen catheter and not described as a given since the authors have still not provided data showing this is feasible without clogging the tip of the catheter, which is not trivial. This reviewer appreciates this is in development, but it is not yet achieved and should not be stated as fact in the manuscript. Given that S31 just shows a prototype without delivering the material and therefore does not address the original concern, I would suggest this be removed from this manuscript, so it does not interfere with publishing a separate manuscript on the catheter once it is fully developed.  
+
+Response: We have revised the text accordingly to indicate that this material design allows for the potential of endovascular delivery via a dual lumen catheter. We retained the supplemental figure and modified the details to ensure that it does not limit future publication.  
+
+"Clinical experience together with the current data supports the potential of transvenous catheter- based delivery of the hydrogel electrode to the AIV through the subclavian vein or internal jugular."  
+
+## REVIEWER #3  
+
+The authors have nicely addressed the comments. This work is suitable for publication in Nature Communications.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__7360ebdf58d8c1045eda153ebd75eac6739f0dd879ad16ad9a400079e48134a8/supplementary_0_Peer Review File__7360ebdf58d8c1045eda153ebd75eac6739f0dd879ad16ad9a400079e48134a8_det.mmd

@@ -0,0 +1,90 @@
+<|ref|>title<|/ref|><|det|>[[61, 40, 508, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[69, 110, 362, 140]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[70, 155, 860, 215]]<|/det|>
+Injectable hydrogel electrodes as conduction highways to restore native pacing  
+
+<|ref|>image<|/ref|><|det|>[[57, 732, 240, 782]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[250, 732, 912, 785]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 845, 163]]<|/det|>
+Editorial Note: This manuscript has been previously reviewed at another journal that is not operating a transparent peer review scheme. This document only contains reviewer comments and rebuttal letters for versions considered at Nature Communications.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 204, 288, 220]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[115, 260, 393, 277]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 317, 865, 371]]<|/det|>
+The novelty of the approach with the initial experiments are interesting. The authors responded to earlier comments recognizing the need for additional experiments and development as well as longer- term follow- up.  
+
+<|ref|>text<|/ref|><|det|>[[115, 468, 393, 484]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 523, 872, 560]]<|/det|>
+Given the journal transfer, I do not think additional data is required. However, text changes are needed per previous comments to more accurately reflect the data presented in the manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[115, 599, 875, 671]]<|/det|>
+The added limitation at the end of the manuscript is good, but all mention of testing the hydrogel in scarred myocardium or labelling ablated tissue as "scar" needs to be removed. The tissue was acutely evaluated after ablation. It should be referred to what it is (ablated tissue), not as scar since no scar has had time to develop.  
+
+<|ref|>text<|/ref|><|det|>[[115, 710, 880, 837]]<|/det|>
+Similarly, the manuscript should be revised to say potentially deliverable via dual lumen catheter and not described as a given since the authors have still not provided data showing this is feasible without clogging the tip of the catheter, which is not trivial. This reviewer appreciates this is in development, but it is not yet achieved and should not be stated as fact in the manuscript. Given that S31 just shows a prototype without delivering the material and therefore does not address the original concern, I would suggest this be removed from this manuscript, so it does not interfere with publishing a separate manuscript on the catheter once it is fully developed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 876, 617, 893]]<|/det|>
+Other limitations have been adequately addressed with the revision.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 175, 394, 191]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 232, 820, 267]]<|/det|>
+The authors have nicely addressed the comments. This work is suitable for publication in Nature Communications.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 384, 107]]<|/det|>
+## RESPONSE TO REVIEWERS:  
+
+<|ref|>text<|/ref|><|det|>[[115, 108, 882, 196]]<|/det|>
+RESPONSE TO REVIEWERS:The authors would like to thank the reviewers for their comments and enthusiasm for the innovative nature of this approach. We appreciate that the editorial team agreed that long- term validation was not necessary for this manuscript and have included detailed responses to individual reviewer comments below as well as a more detailed discussion of the evidence that supports the translational potential of this technology in the revised manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 212, 250, 230]]<|/det|>
+## REVIEWER #1  
+
+<|ref|>text<|/ref|><|det|>[[115, 247, 882, 300]]<|/det|>
+The novelty of the approach with the initial experiments are interesting. The authors responded to earlier comments recognizing the need for additional experiments and development as well as longer- term follow- up.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 315, 251, 333]]<|/det|>
+## REVIEWER #2  
+
+<|ref|>text<|/ref|><|det|>[[115, 350, 882, 386]]<|/det|>
+Given the journal transfer, I do not think additional data is required. However, text changes are needed per previous comments to more accurately reflect the data presented in the manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[115, 402, 882, 473]]<|/det|>
+Comment 1: The added limitation at the end of the manuscript is good, but all mention of testing the hydrogel in scarred myocardium or labelling ablated tissue as "scar" needs to be removed. The tissue was acutely evaluated after ablation. It should be referred to what it is (ablated tissue), not as scar since no scar has had time to develop.  
+
+<|ref|>text<|/ref|><|det|>[[115, 490, 882, 525]]<|/det|>
+Response: We have removed the remaining reference to "scar" from the SI figures and text and referred to it as ablation lesion per standard terminology for this model.  
+
+<|ref|>text<|/ref|><|det|>[[115, 542, 882, 666]]<|/det|>
+Comment 2: Similarly, the manuscript should be revised to say potentially deliverable via dual lumen catheter and not described as a given since the authors have still not provided data showing this is feasible without clogging the tip of the catheter, which is not trivial. This reviewer appreciates this is in development, but it is not yet achieved and should not be stated as fact in the manuscript. Given that S31 just shows a prototype without delivering the material and therefore does not address the original concern, I would suggest this be removed from this manuscript, so it does not interfere with publishing a separate manuscript on the catheter once it is fully developed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 682, 882, 735]]<|/det|>
+Response: We have revised the text accordingly to indicate that this material design allows for the potential of endovascular delivery via a dual lumen catheter. We retained the supplemental figure and modified the details to ensure that it does not limit future publication.  
+
+<|ref|>text<|/ref|><|det|>[[115, 752, 882, 805]]<|/det|>
+"Clinical experience together with the current data supports the potential of transvenous catheter- based delivery of the hydrogel electrode to the AIV through the subclavian vein or internal jugular."  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 821, 250, 839]]<|/det|>
+## REVIEWER #3  
+
+<|ref|>text<|/ref|><|det|>[[115, 856, 882, 891]]<|/det|>
+The authors have nicely addressed the comments. This work is suitable for publication in Nature Communications.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__738b10253eb29318b5141db0c5e308c59681bf9d30847d5513c5cf52add3c9e9/images_list.json

@@ -0,0 +1,77 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Figure.S2 (SI): Analysis of spectral line shape and plasmon-exciton coupling.",
+    "footnote": [],
+    "bbox": [
+      [
+        265,
+        52,
+        732,
+        310
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Figure.S3 (SI): Mixing coefficients in the strong plasmon-exciton coupling.",
+    "footnote": [],
+    "bbox": [
+      [
+        264,
+        498,
+        732,
+        640
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Figure.S13 (SI): Light absorption in PC-WS \\(_2\\) system at the pump frequency (3.1 eV)",
+    "footnote": [],
+    "bbox": [
+      [
+        246,
+        346,
+        744,
+        542
+      ]
+    ],
+    "page_idx": 8
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Figure.S2 (SI): Analysis of spectral line shape and plasmon-exciton coupling.",
+    "footnote": [],
+    "bbox": [
+      [
+        266,
+        46,
+        731,
+        508
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure.1 (Main Text): Sample structure and analysis of spectral line shape and plasmon-exciton coupling.",
+    "footnote": [],
+    "bbox": [
+      [
+        175,
+        197,
+        825,
+        458
+      ]
+    ],
+    "page_idx": 20
+  }
+]

peer_reviews/supplementary_0_Peer Review File__738b10253eb29318b5141db0c5e308c59681bf9d30847d5513c5cf52add3c9e9/supplementary_0_Peer Review File__738b10253eb29318b5141db0c5e308c59681bf9d30847d5513c5cf52add3c9e9.mmd

@@ -0,0 +1,434 @@
+
+## REVIEWER COMMENTS  
+
+## Reviewer #1 (Remarks to the Author):  
+
+The authors report an interesting study on doping two- dimensional semiconductors using plasmonic hot electrons. Specifically, the proposed plasmonic crystal (PC)- tungsten disulfide (WS2) structure is investigated extensively with transient absorption (TA) spectroscopy. While this study provides a unique way to engineer the optoelectronic properties of 2D semiconductors, there are some issues that need to be resolved as listed below:  
+
+1. The main conclusion of this work is that the bandgap of WS2 can be significantly modified by hot electrons injected from the PC. However, there is no strong evidence that hot electrons dominate the observed transient features in TA spectroscopy:  
+
+1.1. According to the discussions, hot electrons are believed to be injected into WS2 by tunneling through the Al2O3 spacer layer. While the spacer layer helps to prevent injected hot electrons from tunneling back into the PC, it also decreases the hot electron injection efficiency and this effect depends on the spacer layer thickness. It is not clear how the authors determine the thickness of the Al2O3 to use for this work and if this thickness is favorable for DET.  
+
+1.2. Similarly, the authors need to explain why the tunneling barrier height can be set to be 1 eV in Equation (2).  
+
+1.3. Since losses are not included in Equation (2), the actual number of injected hot electrons could be much lower than calculated. The authors should provide a better evaluation of the effect of losses either theoretically or experimentally.  
+
+2. Since the quality of the silver cap layer can affect the generated hot electrons and local electric field distribution, did the authors check the optical properties of the deposited silver layer? If so, they should show results of the optical measurements. Otherwise, those measurements need to be made for the next revision.  
+
+3. To compare the broad maxima in Figure 2d and 2g, it would be more helpful to show both figures with the same y-scale.  
+
+4. The authors also mentioned the enhanced nonlinear optical responses of the PC-WS2 structure. How is it related to the rest of this work? The authors should elaborate more on the relevance of the enhanced nonlinear response.  
+
+I would recommend a major revision for the authors to address the questions above.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+The authors study nonlinear optical properties of plasmonic crystals covered with WS2 monolayers using time- resolved transient absorption measurements. The time resolution is of the order of 100 fs, insufficient to resolve coherent exciton- plasmon coupling phenomena in the time domain. The samples show some signatures of exciton- plasmon coupling. Their linear optical properties have been studied in earlier work.  
+
+The authors do claim that the samples are in the strong coupling regime but, in my opinion, this is actually not convincingly supported by the data shown in the paper or in the SI. It seems to me that all the spectra that are shown are consistent with an exciton- plasmon system in the intermediate coupling regime (exciton lifetime < rabi period < plasmon lifetime). Support for the strong coupling claim in the paper should be given by a quantitative analysis of the line shape of the spectra. This criticism, however, is only partially relevant for the main results of the present paper probing incoherent optical nonlinearities occurring on time scales that are longer than the lifetimes of coherent excitations in the coupled system. The main experimental observations are potentially of interest even when the strong- coupling claim is withdrawn.
+
+<--- Page Split --->
+
+In Fig. 2, the authors show angle- resolved studies of the nonlinear optical response of their system. Several interesting features are observed. First of all, some spectra display a transient below- bandgap response, similar to what has been shown by Chernikov et al. for bare WS2 layers in Nature Phot. 9, 466 (2015). Second, the authors see a strongly angle- dependent differential transmission lineshape around the WS2 A exciton at 2.1 eV.  
+
+Let us start with the second feature - which is not discussed at all in the paper. What is the origin of the lineshapes in Fig. 2d and g? What explains the angle- dependent change in lineshape seen in Fig. S12? If the excitons and plasmons in the present system are indeed coupled, it should be possible to analyze these lineshapes in terms of a phenomenological coupled Lorentz oscillator model as routinely used for describing such coupled systems. I would like to strongly encourage the authors to perform such an analysis. It would provide interesting new insight into the nonlinear optical properties of TMDC excitons coupled to plasmons. Also, it could certainly help the authors to better understand some of their results. Based on such analysis, a more quantitative and less speculative analysis of the dynamical behavior should be possible.  
+
+Now, let us turn to the below- bandgap absorption peak. The spectral signatures are indeed very similar to what has been seen by Chernikov in his Nature Photonics paper. I therefore did not quite understand why the authors argue that their results "sharply contrast" with those earlier observations. I would guess that the nature of the below- bandgap peak is very much the same as demonstrated earlier: bandgap renormalization due to a nonequilibrium carrier concentration in WS2 film. The fact that this is seen here at lower pump powers seems not particularly surprising to me: the presence of the plasmonic crystal enhances the absorption (even at the off- resonant pump energy). Since I agree with the authors that the data show this below- bandgap peak at lower fluence than observed before, I believe that the findings are interesting. I would therefore like to ask the authors to thoroughly discuss their observations and to comparatively discuss different physical mechanisms that could account for it. In the present manuscript, they focus very much on a "hot electron from plasmonic crystals" explanation which - in my opinion - is not necessarily supported by the data. I think that it is realistic to estimate plasmonic absorption enhancements at the pump energy and to carefully compare photoinduced carrier densities with and without plasmonic field enhancement. Can plasmonic field enhancement account already for the increase in bandgap renormalization? This question should be discussed in much more depth than in the present manuscript and - if possible - supported by simulations of the linear optical properties of the sample.  
+
+More technical issues:  
+
+1. As mentioned already, I do not believe that the claim of "fast and repeated hot electron population" (aka strong coupling) on page 8 is justified.  
+
+2. A color code in Fig. 1b is missing.  
+
+3. A color code in Fig. 3a is missing.  
+
+4. Why are the data in Fig. 3c, d so noisy. The time-dependent spectra in Fig. 3a look reasonably clean. Has Fourier filtering and smoothing been applied to them.  
+
+5. The physics behind Eq. 2 remains unclear since the model is not explained in the text. If the authors want to maintain their "hot electron" explanation, the model needs more thorough explanation.  
+
+In summary, the reported show some new aspects of the transient optical nonlinearities of exciton-plasmon-coupled systems and hence may be of interest to the research community. In order to make the data publishable a much more in-depth analysis and discussion is needed. Claims of "strong coupling" should either be proven or withdrawn.  
+
+Christoph Lienau
+
+<--- Page Split --->
+
+## Reviewer #3 (Remarks to the Author):  
+
+The authors reported optical control of bandgap of a monolayer WS2 integrated with a plasmonic structure. The sample studied contains a monolayer WS2 on a self- assembled plasmonic crystal. By coupling the plasmonic resonance with A- excitons of WS2, the authors demonstrate control of the WS2 bandgap by optically exciting plasmonic hot electrons that transfer to WS2. Ultrafast control of bandgap renormalization can have important applications in ultrafast photonics such as ultrafast optical switch. As such, the results are significant. However, there are a few issues that I wish the authors could clarify.  
+
+1. The main conclusion of the study is a 650-meV bandgap renormalization at RT. The authors need to provide strong evidence and more detailed analysis to support this claim. It appears that the authors assumed an exciton binding energy of 200 meV. However, this is the value for a WS2 monolayer on an insulating substrate. When coupled with a plasmonic crystal, this is likely to change, since such a large exciton binding energy is due to the lack of screening in the 2D form. Without knowing the precise value of the exciton binding energy in their structure, the claimed renormalization may not be accurate.  
+
+2. It is also a bit concerning to use the peaks in TA spectra as the parameter to determine the bandgap. The TA peak aligns with the resonance only if the phase-space filling is the dominant mechanisms of TA. In 2D materials, this may not be the case.  
+
+3. The authors compared with Ref. 9 in terms of the achieved normalization and the used injection level. However, the hot-electron inject method used here results in net charges in WS2, which is quite different from optical doping. The authors should discuss about the potential implications. In addition, a direct comparison of the pump fluence used may not be fair, since the absorptions are different.
+
+<--- Page Split --->
+
+## Reviewer #1  
+
+We are glad that Reviewer #1 thinks that "The authors report an interesting study on doping twodimensional semiconductors using plasmonic hot electrons.". In addition, the reviewer also suggests that:  
+
+1. We should provide stronger evidence that hot electrons dominate the observed transient features by  
+
+1.1 providing more details on the determination of the thickness of the \(\mathrm{Al}_2\mathrm{O}_3\) spacer.  
+
+Author response: We very much thank the reviewer and agree with her/him that the spacer thickness is critical in forming a favourable tunneling barrier for hot electron transfer. We therefore have added the following sentence in the main text (line 246ff):  
+
+"...The thickness of the metal and dielectric layers are characterised using an ellipsometer measuring identical evaporations on flat silicon substrates."  
+
+In addition, we have also added a schematic in Fig.S14 in the Supplementary Information (SI) to show the surface morphology of \(\mathrm{Al}_2\mathrm{O}_3\) layer. Relevant discussion can be found in the response to the below suggestion.  
+
+1.2 We should explain why the tunneling barrier height can be set to be \(1\mathrm{eV}\) in Eq.(2) in the main text and the relation between the barrier height and the \(\mathrm{Al}_2\mathrm{O}_3\) spacer thickness.  
+
+Author response: We appreciate that Reviewer#1 gives us this opportunity to better explain the height of tunneling barrier. As the reviewer suggested, the barrier height closely relates to the spacer thickness. In our sample, the spacer thickness varies with different locations at a level of \(2.5\pm 2\mathrm{nm}\) According to the previous work [Nanoscale, 11, 4811, (2019)], the \(\mathrm{Al}_2\mathrm{O}_3\) spacer with a thickness of \(\sim 2.1\mathrm{nm}\) can form a tunneling barrier of \(\sim 0.8\mathrm{eV}\) at the metal- \(\mathrm{WS}_2\) interface. Therefore here we take \(\Delta \phi_{\mathrm{TB}} = 1\mathrm{eV}\) , which is also a value that is commonly used in other studies, e.g. [Adv. Opt. Mater., 5, 1600594, (2017)] and [ACS Photonics, 4, 2759, (2017)]. To clearly elucidate the value setting, we have added a paragraph (line 419ff) and Fig.S14 in the SI and the following sentence (line 195ff) in the main text:  
+
+"...In addition, the tunneling barrier \(\Delta \phi_{\mathrm{TB}}\) is set to be \(1\mathrm{eV}\) , which is a typical for the ultrathin \(\mathrm{Al}_2\mathrm{O}_3\) layers used in our system[31, 36], and this setting can help address other dissipations that are not considered in the whole excitation process...."  
+
+with references:  
+
+[31] Xiang Tian Kong, Zhiming Wang, and Alexander O. Govorov. "Plasmonic Nanostars with Hot Spots for Efficient Generation of Hot Electrons under Solar Illumination", Adv. Opt. Mater., 5, 1600594, 2017.  
+
+[36] Shan Zheng, Haichang Lu, Huan Liu, Dameng Liu, and John Robertson. "Insertion of an ultrathin \(\mathrm{Al}_2\mathrm{O}_3\) interfacial layer for Schottky barrier height reduction in \(\mathrm{WS}_2\) field- effect transistors", Nanoscale, 11, 4811, 2019.  
+
+1.3 Reviewer #1 suggests that we should provide a better evaluation of the losses for calculation of hot electron density using Eq.(2).
+
+<--- Page Split --->
+
+Author response: We are very much grateful that the reviewer has raised this suggestion and agree with the reviewer that it is necessary to include losses in the calculation of hot electron density, as the density is the critical factor in identifying the role of hot electrons in the observed bandgap renormalisation. We have therefore revised Eq.(2) to a new form:  
+
+\[N_{\mathrm{e}} = \frac{F_{\mathrm{pump}}\cdot\eta_{\mathrm{A}}\cdot\eta_{\mathrm{D}}\cdot\eta_{\mathrm{pl}}}{2c\epsilon_{0}}\cdot \mathbf{F}\cdot \frac{1}{\pi^{2}}\frac{e^{2}E_{\mathrm{F}}^{2}}{\hbar}\frac{\hbar\omega - \Delta\phi_{\mathrm{TB}}}{(\hbar\omega)^{4}} \quad (2)\]  
+
+In short, three major losses have been included: (i) \(\eta_{\mathrm{A}} = 55\%\) is the ratio of pump energy that is absorbed by the system; (ii) \(\eta_{\mathrm{D}} = 2 / 3\) presents the down- converted energy ratio to excite polaritons; and \(\eta_{\mathrm{pl}} = 50\%\) characterises the ratio of plamonic component in polaritons. The detailed discussions of these major losses and the deduction of the revised Eq.(2) have been added into both the SI and the main text. Specifically, a new sub- section "Including losses" has been added in the Section 6 of the SI, which includes Fig.S13 and relevant discussions. In addition, the following sentence has been added into the main text (line 173ff):  
+
+"To prove that hot electron doping can induce the observed bandgap renormalisation, we need to figure out how the hot electrons are generated in our system as well as quantify the net carrier density in the lattice. As explained before, due to the off- resonance frequency of the pump, plasmons in the PC- WS \(_2\) system can only be effectively excited by coupling to excitons. Specifically the pump energy is absorbed by the semiconductor and down- converted to excite the plasmon- exciton polaritons, which, as half- plasmon half- exciton hybrid states, naturally excite their plasmonic component and result in the generation of plasmonic hot electrons. These charges then overcome the tunneling barrier \((\Delta \phi_{\mathrm{TB}})\) formed at the Ag- Al \(_2\) O \(_3\) - WS \(_2\) interface to dope the WS \(_2\) lattice. During this process, the hot electron doping is subject to several major losses, including (i) the limited pump absorption by the WS \(_2\) MLs, (ii) the losses in energy down- conversion and (iii) the losses due to the hybrid nature of polaritons."  
+
+and in line 188ff in the main text:  
+
+"...Here we take \(\eta_{\mathrm{A}} \approx 55\%\) as the absorption coefficient, \(\eta_{\mathrm{D}} \approx 66\%\) as the energy down- conversion ratio and \(\eta_{\mathrm{pl}} \approx 50\%\) as the excitation ratio of the plasmon component in polaritons. As a result, \(F_{\mathrm{pump}} \cdot \eta_{\mathrm{A}} \cdot \eta_{\mathrm{D}} \cdot \eta_{\mathrm{pl}}\) corresponds to the process that the pump energy is absorbed, down- converted and coupled to the plasmonic components in polaritons with major losses included..."  
+
+It turns out that with losses included, the hot electron density in the WS \(_2\) monolayer can still typically achieve \(\sim 10^{13} \mathrm{cm}^{- 2}\) , even approaching \(10^{14} \mathrm{cm}^{- 2}\) at hot spots. These numbers have reached what is required \((3 \times 10^{13} - 1.1 \times 10^{14} \mathrm{cm}^{- 2})\) to induce a \(\sim 550 \mathrm{meV}\) bandgap renormalisation by electrical doping, meaning that the hot electron densities are sufficient to result in the observed bandgap restructuring. For more details, please see the revised Fig.4 and relevant paragraphs in the main text (lines \(200 - 219 \mathrm{ff}\) ).  
+
+2. Reviewer #1 suggests that we demonstrate the optical properties of the bare plasmonic crystal (PC) in the manuscript, since the quality of the silver PC can affect the generated hot electrons and local electric field distribution.  
+
+Author response: We agree with the reviewer. Due to the frame limit, we have included the angle
+
+<--- Page Split --->
+
+resolved transmission spectra of the bare PC sample in Fig.S1(a) of the SI for comparison with the spectra of the PC sample integrated with a \(\mathrm{WS}_2\) monolayer [PC- \(\mathrm{WS}_2\) , Fig.S1(b)].  
+
+3. Reviewer #1 suggests that we we show Fig.2d and Fig.2g with the same y-scale, which will provide a better comparison for the bandgap renormalisation effect.  
+
+Author response: We thank the reviewer for her/his careful observation. We therefore have made relevant revision to Fig.2d and 2g in the main text.  
+
+4. Reviewer #1 suggests that we elaborate more on the relevance of the enhanced nonlinear response in our system (Fig.3a and 3b in the main text, and Fig.S15 and S16 in the SI).  
+
+Author response: We very much thank Reviewer #1 for this valueable suggestion, since this has inspired us to explore a new research direction. The nonlinear response of our system under high- power pump mainly includes: (i) the spectral shift (Fig.S15) and (ii) the delayed occurence (Fig.S16) of the polariton maxima.  
+
+In short, by comparing the relative magnitudes of the polariton split maxima, we find that the spectral shift highly relates to the excitonic resonance shift that is induced by carrier density enhancement. Likewisely, the delayed occurence of maxima can also relate to the enhancement of carrier density in the lattice. These results correspond to the main conclusion of this manuscript, i.e. the strong coupling between plasmons and excitons facilitates the generation of hot electrons, resulting in the elevation of carrier density in the lattice. We have added these discussions into Section 8 of the SI.  
+
+## Reviewer #2  
+
+We are very grateful that Reviewer #2 positively commented on our results, e.g. "The main experimental observations are potentially of interest", "I believe that the findings are interesting." and "the reported show some new aspects of the transient optical nonlinearities of exciton plasmon- coupled systems and hence may be of interest to the research community". In addition, s/he has provided many valueable suggestions:  
+
+1. Reviewer #2 strongly encourages that we perform lineshape analysis on spectral features of our system using a phenomelogical coupled Lorentz oscillator model, since this enables more quantitative and less speculative analysis on the coupling state of the system, providing a solid ground for other discussions in the manuscript.  
+
+Author response: We fully agree with Reviewer #2 and very much appreciate this suggestion, as it allows us to gain a much better understanding of the coupling behaviours in our system. Following the her/his instructions, we have used a coupled Lorentz model to fit the steady-state transmission spectra of both the bare PC and PC- \(\mathrm{WS}_2\) systems at different angles, which can be seen from Fig.S2 in the SI, where only the spectra at \(\theta = 22^{\circ}\) are shown.  
+
+It turns out that the rate of energy exchange between plasmons and excitons \((2g)\) is slower than the plasmon dephasing \((\kappa)\) but faster than the exciton decay rate \((\gamma)\) , i.e. \(\kappa > 2g > \gamma\) , but the coupling strength in our system can still achieve \(2g > (\kappa + \gamma) / 2\) , which is typically treated as in the strong coupling regime, according to an extensively used criterion in coupling systems, e.g. in the works [Khitrova et al., Nat. Phys., 2, 81 (2006)] and [Lienau et al., ACS Nano, 8, 1056 (2014)].
+
+<--- Page Split --->
+![](images/Figure_unknown_0.jpg)
+
+<center>Figure.S2 (SI): Analysis of spectral line shape and plasmon-exciton coupling. </center>  
+
+We want to point out that the criterion of strong coupling highly depends on specific scenarios. For example, in traditional studies of cavity quantum electrodynamics, a vacuum Rabi splitting that is smaller than the cavity loss may result in unwanted decoherence among excitations. Therefore it requires a strict coupling criterion. But for some other situations, like in our system, the strong coupling effect is only used to provide a channel for energy transfer between plasmons and excitons. In this case, the relatively loose coupling criterion \([2g > (\kappa + \gamma) / 2]\) may apply, because the plasmon- exciton hybrid state can still remain effective in the presence of moderate cavity losses.  
+
+![](images/Figure_unknown_1.jpg)
+
+<center>Figure.S3 (SI): Mixing coefficients in the strong plasmon-exciton coupling. </center>  
+
+The effectiveness of the hybrid state can be determined through the relative degree of mixing between the cavity- plasmon and the exciton in plasmon- exciton polaritons. In short, following the method from [Lienau et al., ACS Nano, 8, 1056 (2014)], we have included the damping factors (e.g. \(i\kappa\) and \(i\gamma\) ), using complex frequency of plasmons, excitons and polaritons to calculate the mixing coefficients for both upper and lower polaritons. As shown in the Fig.S3(b) of SI, it turns out that the coefficients calculated from experimentally acquired parameters slightly drift from the damping- free curves, but still reach \(\sim 50\%\) at the tuned state. This means that the plasmonic component and excitonic component account for half of the polariton energy respectively, indicating that moderate losses in plasmons do not significantly change the strong coupling nature of the PC- WS \(_2\) system in terms of energy exchange. Theoretical simulations also support this conclusion [Fig.S3(a)].
+
+<--- Page Split --->
+
+Therefore, here we kindly ask the reviewer that if s/he can consider the situation and allow us to treat the system as in the strong coupling regime. To better demonstrate the coupling behaviours in our system, we have rewritten the whole Section 1 in the SI, including newly added figures Fig.S1, S2 and S3. We have also added the following sentence in the main text (line 66ff):  
+
+"The splitting at the tuned state \((\theta = 22^{\circ})\) can be characterised by a vacuum Rabi splitting of \(\hbar \cdot \Omega_{R} =\) \(\hbar \cdot 2g\approx 140meV\) exceeding the widely used coherent strong coupling criterion \(2g > (\kappa +\gamma) / 2\) [21,22], where \(\kappa\) is the dissipation of plasmon modes and \(\gamma\) is the exciton decay rate."  
+
+with references:  
+
+"[21] Khitrova, G., Gibbs, H. M., Kira, M., Koch, S. W. and Scherer, A. Vacuum Rabi splitting in semiconductors., Nat. Phys., 2, 81, 2006."  
+
+"[22] Wang, W., Vasa, P., Pomraenke, R., Vogelgesang, R., De Sio, A., Sommer, E., Maiuri, M., Manzoni, C., Cerullo, G. and Lienau, C., Interplay between strong coupling and radiative damping of excitons and surface plasmon polaritons in hybrid nanostructures, ACS Nano, 8, 1056, 2014."  
+
+In addition, we would like to point out that here all lineshape analyses were based on the steady- state transmission spectra, because the steady- state spectra provide stable, collective and power (and time) independent spectral features, which can simplify the discussion and is typically used for the analysis of plasmon- exciton coupling. In contrast, the spectral positions and linewidth of features in transient spectra \(\Delta \mathrm{T} / \mathrm{T}\) are highly sensitive to the pump power and delay time (see Fig.2 and 3 in the main text and Fig.S4, S6, S8, S15, S16 and S17 and relevant discussions in the SI), which brings enoumours difficulties and uncertainties in analysing the coupling.  
+
+2. Reviewer #2 suggests that we discuss the "strongly angle-dependent differential transmission lineshape around the \(W S_{2}\) exciton \(A\) ". Specifically, it is necessary to figure out "the origin of the lineshapes in Fig.2d and \(2g\) " as well as to explain "the angle-dependent change in lineshape seen in Fig.S12". The reviewer suggests that we use a Lorentz oscillator model to analyse the lineshapes.  
+
+Author response: We agree with the reviewer that it is important to discuss the lineshape change in transient differential spectra at distinct incident angles, as this will provide us a deeper understanding of the coupled system's transient dynamics.  
+
+First of all, we want to point out that the coupled Lorentz oscillator model may not be applicable for analysing the lineshapes of differential spectra. Specifically, Lorentz models can describe relatively simple transition processes, e.g. excitations from ground to excited state or decay from excited to ground state, which, therefore, are typically used to analyse the simple resonant features in steady- state spectra. However, the differential spectra used here is \((\mathrm{T} - \mathrm{T}_{0}) / \mathrm{T}_{0}\) , which is the normalised subtractions between two spectra at different delay times, comprising of very complicated physical processes in addition to energy transition, e.g. the time- dependent many- body interactions between excitations. As a result, resonance features may broaden and shift at different delay times, leading to negative magnitudes and peak shifts as compared to the steady- state spectra. In this case, using Lorentz oscillator models to analyse the lineshapes of the differential spectra does not provide effective information.  
+
+Instead, we have phenomenologically analysed the lineshape change of differential spectra at different
+
+<--- Page Split --->
+
+angles and delay times, mainly drawing the following conclusions:  
+
+(i) The splitting in differential spectra correspond to the splitting in the steady-state spectra, but with some frequency shifts and smaller splitting magnitudes.  
+
+(ii) The frequency shifts and smaller splitting magnitudes should be the result of lineshape broadening and shift induced by many-body Auger recombination.  
+
+These discussions have been added to Section 2 in the SI, particularly with panel (d), (e) and (f) of Fig.S4 (Fig.S12 in the last version of SI)  
+
+3. Reviewer #2 suggests that we "thoroughly discuss their observations and to comparatively discuss different physical mechanisms that could account for it." Specifically, the reviewer would like us to "estimate plasmonic absorption enhancements at the pump energy and to carefully compare photoinduced carrier densities with and without plasmonic field enhancement." This should be "discussed in much more depth than in the present manuscript" and "if possible – supported by simulations of the linear optical properties of the sample."  
+
+![](images/Figure_unknown_2.jpg)
+
+<center>Figure.S13 (SI): Light absorption in PC-WS \(_2\) system at the pump frequency (3.1 eV) </center>  
+
+Author response: First we would like to express our sincere appreciation to Reviewer #2, since this suggestion has extended our understanding of carrier population in the system and greatly helped us reshape the later discussions on hot electron doping.  
+
+Specifically, we have modelled the intensity distribution of the PC- WS \(_2\) system at the pump frequency [Fig. S13 (a)]. It turns out that the intensity enhancement at this frequency is no larger than \(\sim 7\) times at the position of the WS \(_2\) monolayer, which, according to our calculation, results in an average (over the whole area) optical absorption of \(\sim 55\%\) in the monolayer. It means that the absorption in WS \(_2\) is highly enhanced as compared to the bare monolayer ( \(\sim 10\%\) ) that is not integrated with the plasmonic crystal [Fig. S13(b)]. As a result, if the pump pulse that has a fluence of \(12 \mu \mathrm{J} / \mathrm{cm}^2\) can be absorbed and fully converted, the carrier population in the WS \(_2\) ML can reach at a density of \(\sim 1.2 \times 10^{13} \mathrm{~cm}^{- 2}\) .  
+
+This overestimated value, however, is still one order of magnitude lower than the one ( \(\sim 10^{14} \mathrm{~cm}^{- 2}\) ) that is required to develop a Mott- transition at a cryogenic temperature \(70 \mathrm{~K}\) [Heinz et al., Nat. Photon., 9, 466, 2015], let alone the level at room- temperature. We also want to point out that even if all the absorbed energy in the PC- WS \(_2\) system (i.e. \(\sim 80\%\) absorption of pump, see the absorption spec
+
+<--- Page Split --->
+
+trum of the PC- WS \(_2\) at \(3.1\mathrm{eV}\) in Fig.S13) can be converted to excite excitons in the WS \(_2\) lattice, the generated carrier density is still lower than what is required to induce a large bandgap renormalisation as in our experiment.  
+
+In addition, the density increase induced by absorption enhancement should appear at transient spectra of all incident angles, but in our case, only the tuned state ( \(\theta = 22^{\circ}\) ) show a large bandgap renormalisation. Therefore, we do not think plasmonic absorption enhancement is the main factor that can induce the observed bandgap renormalisation in our experiments.  
+
+To present readers with this argument, we have added a new paragraph in the main text (line 129ff):  
+
+"In our system, the carrier density may be increased by plasmonic absorption enhancement (PAE), which, however, can not provide enough carrier population according to our calculation. Specifically, the excitation of plasmons can enhance the absorption of the pump by the system, which naturally results in an elevation of carrier numbers in the lattice. In the PC- WS \(_2\) system, the pump intensity can be amplified at the position of the WS \(_2\) ML (Fig.S13 in SI), which, according to our calculation, gives a \(\sim 5\) times average increase of absorption in the semiconductor. As a result, the carrier density can achieve up to \(\sim 1.2 \times 10^{13}\mathrm{cm}^{- 2}\) if the absorbed pump energy is fully converted. However, even this overestimated value is still one order of magnitude lower than the density level ( \(\sim 10^{14}\mathrm{cm}^{- 2}\) )[9] required to cause a Mott- transition at \(70\mathrm{K}\) , let alone the level at room- temperature. Furthermore, PAE should also enhance carrier generation in the detuned systems. However, in our experiments, only the tuned system shows a large bandgap renormalisation (Fig.2b and 2d). Hence there must be other mechanisms that can enhance carrier population in addition to PAE."  
+
+with references:  
+
+"[9] Alexey Chernikov, Claudia Ruppert, Heather M. Hill, Albert F. Rigosi, and Tony F. Heinz., Population inversion and giant bandgap renormalization in atomically thin WS2 layers. Nat. Photon., 9, 466, 2015."  
+
+Furthermore, have added a subsection in Section 6 in the SI including Fig.S13.  
+
+In addition, Reviewer #2 has raised some technical issues:  
+
+(1) Reviewer #2 suggests that we justify ""fast and repeated hot electron population" (aka strong coupling)" in our system.  
+
+Author response: Please see above the answers for point 1.  
+
+(2) & (3) Reviewer #2 suggests that we add colour codes in Fig.1b and Fig.3a.  
+
+Author response: We are grateful that the reviewer has carefully read our manuscript and we have therefore added colour codes in these figures.  
+
+(4) Reviewer #2 suggests that we clarify why data in Fig.3c and 3d look noisy while the intensity plot in Fig.3a looks clean, as they are plotted from the same set of data.  
+
+Author response: We have checked these data and plots, finding that no filtering or smoothing have been applied to them. The inconsistency in noise level might be from different axis range taken. Please note that
+
+<--- Page Split --->
+
+panel c and d only show spectra with an energy range from 1.5 to \(1.9\mathrm{eV}\) , while panel a shows the intensity plot from 1.6 to \(2.5\mathrm{eV}\) . The denser distribution of data point may reduce the illustration of noise.  
+
+(5) Reviewer #2 suggests that we better explain the model (Eq.2) and elucidate our arguement of "coherent hot electron doping" with more thorough analysis.  
+
+Author response: We fully agree with the reviewer and therefore have redeveloped our model, having included several major loss factors that may suppress the hot electron doping. By doing so, we have been able to obtain a more accurate estimation of the hot electron density in the \(\mathrm{WS}_2\) lattice. As a result, we find that with major losses included, the hot electron density can still reach the level that is required to develop a large bandgap renormalisation as observed in our experiments. Specifically, we have added a new paragraph in the main text (line 173ff):  
+
+"To prove that hot electron doping can induce the observed bandgap renormalisation, we need to understand how the hot electrons are generated in our system as well as quantify the net carrier density in lattice. As explained before, due to the off- resonance frequency of pump, plasmons in the PC- \(\mathrm{WS}_2\) system can only be effectively excited by coupling to excitons. Specifically the pump energy is absorbed by the semiconductor and down- converted to excite the plasmon- exciton polaritons, which, as half- plasmon half- exciton hybrid states, naturally excite their plasmonic component and result in the generation of plasmonic hot electrons. These charges then overcome the tunneling barrier \((\Delta \phi_{\mathrm{TB}})\) formed at the Ag- \(\mathrm{Al}_2\mathrm{O}_3\) - \(\mathrm{WS}_2\) interface to dope the \(\mathrm{WS}_2\) lattice. During this process, the hot electron doping is subject to several major losses, including (i) the limited pump absorption by the \(\mathrm{WS}_2\) MLs, (ii) the losses in energy down- conversion and (iii) the losses due to the hybrid nature of polaritons."  
+
+Following this process, as described in the response to Reviewer #1, we have developed a new model to numerically estimate the density \((N_{\mathrm{e}})\) .  
+
+\[N_{\mathrm{e}} = \frac{F_{\mathrm{pump}}\cdot\eta_{\mathrm{A}}\cdot\eta_{\mathrm{D}}\cdot\eta_{\mathrm{pl}}}{2c\epsilon_{0}}\cdot \mathcal{F}\cdot \frac{1}{\pi^{2}}\frac{e^{2}E_{\mathrm{F}}^{2}}{\hbar}\frac{\hbar\omega - \Delta\phi_{\mathrm{TB}}}{(\hbar\omega)^{4}} \quad (2)\]  
+
+Relevant explanations are given at line 188ff in the main text:  
+
+"...Here we take \(\eta_{\mathrm{A}}\approx 55\%\) as the absorption coefficient, \(\eta_{\mathrm{D}}\approx 66\%\) as the energy down- conversion ratio and \(\eta_{\mathrm{pl}}\approx 50\%\) as the excitation ratio of the plasmon component in polaritons. As a result, \(F_{\mathrm{pump}}\cdot \eta_{\mathrm{A}}\cdot\) \(\eta_{\mathrm{D}}\cdot \eta_{\mathrm{pl}}\) corresponds to the process that the pump energy is absorbed, down- converted and coupled to the plasmonic components in polaritons with major losses included. As optical modes, the excited polaritons gain a spatial distribution (Fig.4b) at the tuned frequency, spreading over the Ag cap surface with hot spots at the interstices between caps. This can be mathematically expressed as \(\mathcal{F} = |\mathbf{E} / \mathbf{E}_0|^2\dots\) "  
+
+Here we want to thank Reviewer #2 that the efficiency parameters \(\eta_{\mathrm{A}}\) and \(\eta_{\mathrm{pl}}\) are all inspired by her/his suggestions, which are the absorption coefficient and the mixing coefficient respectively. Please see above the suggestions for more details. As a result, we have been able to calculate the carrier density, which is stated in the main text (line 200ff):  
+
+"Using Eq.2, we are able to plot the spatially distributed hot electron density in the \(\mathrm{WS}_2\) monolayer (Fig.4c). The density naturally acquires identical distributions as do the plasmonic excitations, exhibiting inhomogeneous distribution over the area. It has values typically higher than \(1 \times 10^{13} \mathrm{~cm}^{- 2}\) in most of the
+
+<--- Page Split --->
+
+areas, peaking at the interstices between caps with maxima larger than \(2 \times 10^{14} \mathrm{cm}^{- 2} \ldots\) "  
+
+In addition, a more detailed discussions of these major losses and the deduction of the revised Eq.(2) have been added into the SI. Specifically, a new sub- section "Including losses" has been added in the Section 6 of the SI, which includes Fig.S13 and relevant discussions.  
+
+## Reviewer #3  
+
+We are very happy that Reviewer #3 thinks highly of our manuscript, e.g. "Ultrafast control of bandgap renormalization can have important applications in ultrafast photonics such as ultrafast optical switch. As such, the results are significant." In addition, the reviewer has raised a few issues:  
+
+1. Reviewer #3 suggests that we figure out the binding energy of a \(\mathrm{WS}_2\) monolayer deposited on the plasmonic crystal, as this could be largely different from the value for monolayers deposited on a insulating substrate, which will be critical in determining the magnitude of bandgap renormalisation.  
+
+Author response: We fully agree with the reviewer and very much appreciate this suggestion. After carefully reviewing literature, we have found that the binding energy of a \(\mathrm{WS}_2\) monolayer can be reduced to a half of its original value when deposited on a metal substrate. We have therefore revised our conclusion that the bandgap renormalisation can achieve up to \(\sim 550 \mathrm{meV}\) (about \(100 \mathrm{meV}\) smaller than the previous estimation).  
+
+Specifically, we have corrected the value in the following sentence in the main text (line 118ff):  
+
+"...It means that in our experiments, the renormalised bandgap starts at \(E_{g} \approx 1.60 \mathrm{eV}\) , lying \(\sim 400 \mathrm{meV}\) below LP and \(\sim 550 \mathrm{meV}\) below the bandgap of \(\mathrm{WS}_2\) MLs (given that the binding energy of exciton A is decreased to \(\sim 100 \mathrm{meV}\) when deposited on metal substrates[28], i.e. about a half of the initial value[19])."  
+
+with references:  
+
+"[19] Paul D. Cunningham, Aubrey T. Hanbicki, Kathleen M. McCreary, and Berend T. Jonker. Photoinduced Bandgap Renormalization and Exciton Binding Energy Reduction in WS2., ACS Nano, 11, 12601, 2017."  
+
+"[28] Park Soohyung, Mutz Niklas, Schultz Thorsten, Blumstengel Sylke, Han Ali, Aljarb Areej, Li Lain Jong, List-Kratochvil Emil J.W., Amsalem Patrick, and Koch Norbert. Direct determination of monolayer MoS2 and WSe2 exciton binding energies on insulating and metallic substrates. 2D Mater, 5, 025003, 2018."  
+
+With the correction, the main conclusion in the manuscript remains unchanged, because \(\sim 550 \mathrm{meV}\) renormalisation is still a large one. But here we want to deeply thank Reviewer #3. Without her/his suggestion, we would have made a serious mistake on the magnitude of bandgap renormalisation.  
+
+2. Reviewer #3 suggests that we clarify the method to determine the spectral position of bandgap, since using transient absorption peaks to tell the bandgap position may not apply to 2D semiconductors.  
+
+Author response: We agree with the reviewer and have rechecked our measurements for finding the bandgap. Specifically, we have followed the method used in a previous work [Heinz et al., Nat.
+
+<--- Page Split --->
+
+Photon., 9, 466, 2015], using the onset of a spectral feature (i.e. the low energy end) but not the peak to determine the band edge position. Using this method, we can always find the point with the lowest energy, avoiding misjudgement of a bandgap position induced by the not fully filled phase- space. In particular, we have added the following sentence in the main text (line 116ff):  
+
+"The onset of the new bandgap can be extracted from the low- energy end of the broad maximum[9] (red dashed vertical line in Fig.4c)"  
+
+with a reference:  
+
+"[9] Alexey Chernikov, Claudia Ruppert, Heather M. Hill, Albert F. Rigosi, and Tony F. Heinz. Population inversion and giant bandgap renormalization in atomically thin WS2 layers. Nat. Photon., 9, 466, 2015."  
+
+3. Reviewer #3 suggests that we should discuss the potential implications of hot electron doping, since it is different from the case in Ref[9], where the optical doping (dessociated electron-hole pairs induced by a strong optical pump) is the dominant mechanism.  
+
+Author response: We fully agree with the reviewer. We therefore have added new discussions and a reference to discuss the implications of hot electron doping. Specifically, we have added the following sentence in the main text (line 212ff):  
+
+"...However, the injected carriers in our system are hot electrons, which are different from the dessociated electron-hole pairs induced by pure optical pumping[9,26], but are more similar to free charge carriers by electrical injection[8,39], where \(a \sim 550 \mathrm{meV}\) bandgap renormalisation in \(\mathrm{WS}_2\) MLs can occur at the electron density of \(3 \times 10^{13} - 1.1 \times 10^{14} \mathrm{cm}^{-2}\) [8]. These results share high similarity with our observation, suggesting that the hot electron doping in our system is able to achieve the threshold to induce a bandgap redshift up to \(\sim 500 \mathrm{meV}\) with carriers draining from conduction band \(K\) to \(\Sigma\) valley[38] that renders the semiconductor indirect"  
+
+with references:  
+
+"[8] Alexey Chernikov, Arend M. Van Der Zande, Heather M. Hill, Albert F. Rigosi, Ajanth Velauthapillai, James Hone, and Tony F. Heinz. Electrical Tuning of Exciton Binding Energies in Monolayer WS2. Phys. Rev. Lett., 115, 126802, 2015."  
+
+"[9] Alexey Chernikov, Claudia Ruppert, Heather M. Hill, Albert F. Rigosi, and Tony F. Heinz. Population inversion and giant bandgap renormalization in atomically thin WS2 layers. Nat. Photon., 9, 466, 2015."  
+
+"[26] D. Erben, A. Steinhoff, C. Gies, G. Schonhoff, T. O. Wehling, and F. Jahnke. Excitation- induced transition to indirect band gaps in atomically thin transition- metal dichalcogenide semiconductors. Phys. Rev. B, 98, 035434, 2018."  
+
+"[38] F. Lohof, A. Steinhoff, M. Florian, M. Lorke, D. Erben, F. Jahnke, and C. Gies. Prospects and Limitations of Transition Metal Dichalcogenide Laser Gain Materials. Nano Lett. 19, 210, 2019."  
+
+"[39] Qiu Zhizhan, Trushin Maxim, Fang Hanyan, Verzhbitskiy Ivan, Gao Shiyuan, Laksono Evan,
+
+<--- Page Split --->
+
+Yang Ming, Lyu Pin, Li Jing, Su Jie, Telychko Mykola, Watanabe Kenji, Taniguchi Takashi, Wu Jishan, Neto A H Castro, Yang Li, Eda Goki, Adam Shaffique and Lu Jiong. Giant gate-tunable bandgap renormalization and excitonic effects in a 2D semiconductor. Sci. Adv., 5, eaaw2347, 2019."  
+
+In addition, Reviewer #3 suggests that we should compare the carrier population in our system with that in the ref[9] in a fairer ground but not only based on a fluence comparison.  
+
+Author response: We agree with the reviewer and have therefore directly compare the carrier density. Specifically, we have added a new paragraph in the main text (line 129ff):  
+
+"In our system, the carrier density may be increased by plasmonic absorption enhancement (PAE), which, however, can not provide enough carrier population according to our calculation. Specifically, the excitation of plasmons can enhance the absorption of pump by the system, which naturally results in an elevation of carrier numbers in the lattice. In the PC- WS₂ system, the pump intensity can be amplified at the position of the WS₂ ML (Fig.S13 in SI), which, according to our calculation, gives a \(\sim 5\) times average increase of absorption in the semiconductor. As a result, the carrier density can achieve up to \(\sim 1.2 \times 10^{13} \mathrm{~cm}^{- 2}\) if the absorbed pump energy is fully converted. However, even this overestimated value is still one order of magnitude lower than the density level \(\sim 10^{14} \mathrm{~cm}^{- 2}\) [9] required to cause a Mott- transition at 70 K, let alone the level at room- temperature. Furthermore, PAE should also enhance carrier generation in the detuned systems. However, in our experiments, only the tuned system shows a large bandgap renormalisation (Fig.2b and 2d). Hence there must be other mechanisms that can enhance carrier population in addition to PAE."  
+
+with references:  
+
+"[9] Alexey Chernikov, Claudia Ruppert, Heather M. Hill, Albert F. Rigosi, and Tony F. Heinz., Population inversion and giant bandgap renormalization in atomically thin WS2 layers. Nat. Photon., 9, 466, 2015."
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+## Reviewer #1 (Remarks to the Author):  
+
+All my concerns were addressed properly. In principle, the manuscript can be published now.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+The authors have responded in great detail to the questions and comments of the Reviewers. The have also added much helpful additional analysis.  
+
+I first want to specifically address the point of "strong coupling".  
+
+1. The authors show, in Fig. S1a angle-resolved transmission spectra of the bare PC sample. This shows two transmission peaks in the wavelength range of the A and B excitons. Out of these two modes, only one of the modes is considered in the coupling scenario that is shown in Fig. S2c. I do - honestly - believe that it is an oversimplification to neglect the existence of these two modes in the coupling scenario. A simple 2x2 model does not work. The same holds for the WS2 sample. Both Xs should be considered in a relativistic coupling scenario, which should be based, at least, on the analysis of a 4 x 4 matrix.  
+
+2. If I understand it correctly, the "coupled Lorentz oscillator" model that is used by the authors is, again, oversimplified. I see in in Fig. S2b that the spectra are fit two to Lorentz resonances with absorptive lineshape. In the experimental spectra, however, distinct signatures of Fano resonances are seen (for example in the spectra in fig. S1 at \(40^{\circ}\) around the XA exciton). Such Fano resonances arise if the emission of a broader resonance interferes with the emission from a narrower resonance. Such Fano-type lineshapes are neglected in the analysis (if I understand it correctly), even though they are evidently seen in experiment. I think that it is necessary to really calculate the linear transmission spectra for the discussed 4x4 model if the authors want to gain a detailed understanding of the optical properties of their sample. For these reasons, I am skeptical about the "blue bullets" shown in Fig. 2c (and the normal mode splitting deduced from them). I have the suspicion that the actual normal mode splitting may turn out to be smaller when analyzing a more appropriate 4x4 model.  
+
+3. In their analysis, the authors conclude (page 4 of the SI, 2nd par) that \(\backslash \mathrm{kappa} > 2\mathrm{g} > \backslash \mathrm{gamma}\) . This implies that the sample is \\*not\\* in the strong coupling regime but in an intermediate coupling regime. I would therefore very much suggest to omit the claim that the sample is in the strong coupling regime. This holds in particular since the 2x2 model that is used for extracting the parameters \(\backslash \mathrm{kappa}\) , \(\backslash \mathrm{g}\) and \(\backslash \mathrm{gamma}\) is oversimplified. I therefore firmly suggest to remove the claims that the sample is in the strong coupling regime. Otherwise, a complete lineshape analysis, including all relevant resonances in the sample, together with a demonstrating that strong coupling is truly reached, is necessary. I also point out that I believe that the conclusions drawn from the pump-probe measurements (which cannot resolve the Rabi oscillations anyway) will not change if the authors admit that the sample is "only" in the intermediate coupling regime.  
+
+4. The authors probe a normal mode coupling of many excitons to several plasmon resonances. This "classical harmonic oscillator coupling" is different from a Vacuum Rabi splitting in the optical spectra of one single quantum emitter coupled to a single cavity mode. The term "Vacuum Rabi splitting" should therefore be avoided.  
+
+In summary, I think that the manuscript would greatly benefit from a much more thorough analysis of the linear optical transmission spectra in Fig. S1. Such an analysis must include all relevant spectral resonances (in the present case it seems that this requires at least a 4x4 model). I still believe that the results may be of interest for the audience of Nature Communications.
+
+<--- Page Split --->
+
+## Reviewer #3 (Remarks to the Author):  
+
+The authors have adequately addressed my comments and improved their manuscript. I recommend acceptance of the revised manuscript.
+
+<--- Page Split --->
+
+## Reviewer #2  
+
+We are glad that Reviewer #2 noted our efforts in improving the manuscript and we are happy that s/he "still believe the results may be of interest for the readers of Nature Communications". In responding to her/his new suggestions, we figure that the problems in our system have now been much better addressed and the role of plasmon- exciton coupling has been further clarified. Thank you!  
+
+In general, Reviewer #2 suggests that we should perform more detailed analyses on the transmission spectra and the coupling state of our PC- WS \(_2\) system. Specifically,  
+
+1. Reviewer #2 first suggests that we should analyse the transmission spectra using a coupled model that includes at least 4 oscillators, because there are 4 sets of resonances co-existing and mutually interacting in our system, which are PC-01 mode, PC-02 mode, exciton A and exciton B, respectively.  
+
+2. In addition to point 1, Reviewer #2 suggests that we should study the optical spectra using a more sophisticated model than the current one, which may better explain the special lineshapes, e.g. Fano-like lineshapes at oblique incident angles, so that the resonance dispersions can be more accurately reproduced.  
+
+Author response: Since point 1 and 2 are very relevant, here we respond them together. First of all, we are grateful for these suggestions, which enable us to gain a more comprehensive understanding of our own system. We fully agree with the reviewer on these points and have made relevant revisions.  
+
+(1) We have used the transmission coefficients that build on 4 coupled Lorentz oscillators to reproduce the spectral lineshapes, which is:  
+
+\[T(\omega) = |t(\omega)|^{2} = \left|a + \sum_{j = 1}b_{j}\frac{\pi\frac{\gamma_{1}}{2}}{(\omega - \omega_{j}) - i\frac{\gamma_{1}}{2}}\right|^{2} \quad (S1)\]  
+
+As compared to our previous model, i.e. the linear superposition of the amplitudes of component Lorentz oscillators, Eq.S1 allows us to calculate the transmission spectra that result from the interplay between all 4 oscillators in our system. As the result, we are able to well reproduce the spectral lineshapes from all incident angles, including the Fano- like lineshape at \(40^{\circ}\) . Fig.S2(b) exhibits examples of these fittings. Then the spectral positions of the fitted resonances were extracted and plotted as a function of angles, shown as the blue dots in Fig.S2(c).  
+
+(2) We then used a \(4 \times 4\) matrix to fit the dispersion of these blue dots, which is:  
+
+\[\begin{array}{r}{\left(\begin{array}{c c c c}{\tilde{E}_{\kappa_{1}}(\theta)}&{g_{1\mathrm{A}}}&{0}&{g_{1\mathrm{B}}}\\ {g_{1\mathrm{A}}}&{\tilde{E}_{\gamma_{\mathrm{A}}}}&{g_{2\mathrm{A}}}&{0}\\ {0}&{g_{2\mathrm{A}}}&{\tilde{E}_{\kappa_{2}}(\theta)}&{g_{2\mathrm{B}}}\\ {g_{1\mathrm{B}}}&{0}&{g_{2\mathrm{B}}}&{\tilde{E}_{\gamma_{\mathrm{B}}}}\end{array}\right)\left(\begin{array}{c}{\alpha_{\kappa_{1}}(\theta)}\\ {\alpha_{\gamma_{\mathrm{A}}}(\theta)}\\ {\alpha_{\kappa_{2}}(\theta)}\\ {\alpha_{\gamma_{\mathrm{B}}}(\theta)}\end{array}\right)=\tilde{E}_{p}(\theta)\left(\begin{array}{c}{\alpha_{\kappa_{1}}(\theta)}\\ {\alpha_{\gamma_{\mathrm{A}}}(\theta)}\\ {\alpha_{\kappa_{2}}(\theta)}\\ {\alpha_{\gamma_{\mathrm{B}}}(\theta)}\end{array}\right)}\end{array} \quad (S2)\]  
+
+Using Eq.S2, we have been able to plot the curves [orange curves in Fig.S2(c)] that follow the dispersions of the fitted resonances. We note that the fitted dispersion curves agree well with spectral positions near the intersection point between PC- 01 mode and exciton A, exhibiting a split spectral feature. In contrast, the curves slightly drift from the fitted resonances near the intersection point between PC- 02 mode and exciton B. We think it is due to the lossy nature of the coupled oscillators, e.g. both PC- 02 mode and exciton B
+
+<--- Page Split --->
+![](images/Figure_unknown_3.jpg)
+
+<center>Figure.S2 (SI): Analysis of spectral line shape and plasmon-exciton coupling. </center>  
+
+acquire broad linewidths, which may temper the fitting accuracy using Eq.S1.  
+
+(3) From the fitting, we've learned that the strength of coupling between PC-01 and exciton A is \(g_{1\mathrm{A}}\approx 87meV\) and the strength of coupling between PC-02 and exciton B is \(g_{2\mathrm{B}}\approx 30meV\) . What's worth noting is that there is also a coupling between PC-01 mode and exciton B. This can be seen from the blue shift of the resonances near exciton B at low incident angles \((\theta = 0 - 6^{\circ})\) , yielding a strength of \(g_{1\mathrm{B}}\approx 70\mathrm{meV}\) . Here we want to point out that we wouldn't have been able to find this 01-B coupling without using the \(4\times 4\) coupled oscillator model. Therefore we very much appreciate this suggestion from Reviewer #2.  
+
+Detailed revisions corresponding to these two points can be found in Section 1 of supplementary information (SI).  
+
+3. Reviewer #2 suggests that we should define our system as in "intermediate" coupling state, if the coupling strength only meets \(2g > (\kappa + \gamma) / 2\) but does not satisfy \(2g > \kappa > \gamma\) .  
+
+Author response: We fully agree with this point, as an accurate identification of coupling state will provide clear application conditions for the coherent doping of plasmonic hot electrons, which is critical for this work. As mentioned above, the coupling strength \(g_{1\mathrm{A}}\approx 87meV\) , which satisfies \(2g_{1\mathrm{A}} > (\kappa_{1} + \gamma_{\mathrm{A}}) / 2\) , yet
+
+<--- Page Split --->
+
+is slower than the plasmonic dephasing. Therefore, the coupling between PC- 01 mode and exciton A should be defined as intermediate coupling. We have added this new definition and removed the previous “strong coupling” definition throughout the manuscript including the main text and SI.  
+
+4. Reviewer #2 suggests that we should avoid the term “Vacuum Rabi splitting”, as this term is specially used to describe the coupling between a single emitter and a single cavity photon, which is different from our case, in which multiple plasmons couple with multiple excitons.  
+
+Author response: We fully agree with Reviewer #2 on this point, and therefore have replaced “Vacuum Rabi splitting” as “spectral splitting”.  
+
+In addition to the specific points raised by the reviewers we have also revised some other parts in the manuscript, e.g. abstract, some paragraphs and figure captions.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+## Reviewer #2 (Remarks to the Author):  
+
+Dear authors, I quickly looked at your response. I am very happy to read that you have now performed such a thorough analysis of your data. I believe that this new analysis gives a physically much more accurate description of your data.  
+
+I also acknowledge the removal of claims on "strong coupling" and VRS. This certainly improves the quality of the paper.  
+
+If I may, I would like to suggest to add Fig. S2 to the main manuscript. It is crucial for understanding the time- resolved results and will certainly be of interest for many readers.  
+
+I gladly recommend your manuscript for publication in Nature Comm. Christoph Lienau
+
+<--- Page Split --->
+
+## Reviewer #2  
+
+We are very much glad that Reviewer #2 recommended our manuscript to be published in Nature Communications. Thank you!  
+
+In addition, Reviewer#2 suggests that we add Supplementary Figure 2 into the main text, which will help readers better understand our system. We fully agree on this point and have integrated the main contents, i.e. panel (b) and (c), of Supplementary Figure 2 into Figure 1 in the main text. Please see below.  
+
+![](images/Figure_1.jpg)
+
+<center>Figure.1 (Main Text): Sample structure and analysis of spectral line shape and plasmon-exciton coupling. </center>  
+
+In addition, we have adapted the main text accordingly to explain the newly added panel (c) and (d) of Figure 1. Specifically, we have added the following text (line 69ff):  
+
+"As a result, within the angle range \((0 - 40^{\circ})\) , PC- 01, PC- 02, \(X_{A}\) and \(X_{B}\) mutually interact. To analyse the couplings between them, we have used a coupled Lorentz model that build on 4 sets of Lorentz oscillators (Supplementary Eq. S1) to fit the transmission spectra of the PC- WS₂ sample. Fig. 1c shows examples of these fittings at different incident angles. The spectral positions of the fitted resonances were then extracted and plotted as a function of angles (blue dots in Fig. 1d). The complicated dispersive behaviours of these resonances were then fitted using a \((4 \times 4)\) matrix of coupled oscillators (Supplementary Eq. S2) to give the critical coupling parameters."  
+
+In addition to the specific points raised by the reviewers we have also revised some other parts in the manuscript, e.g. abstract, some paragraphs and figure captions.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__738b10253eb29318b5141db0c5e308c59681bf9d30847d5513c5cf52add3c9e9/supplementary_0_Peer Review File__738b10253eb29318b5141db0c5e308c59681bf9d30847d5513c5cf52add3c9e9_det.mmd

@@ -0,0 +1,602 @@
+<|ref|>sub_title<|/ref|><|det|>[[119, 84, 356, 101]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 138, 448, 153]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 166, 877, 238]]<|/det|>
+The authors report an interesting study on doping two- dimensional semiconductors using plasmonic hot electrons. Specifically, the proposed plasmonic crystal (PC)- tungsten disulfide (WS2) structure is investigated extensively with transient absorption (TA) spectroscopy. While this study provides a unique way to engineer the optoelectronic properties of 2D semiconductors, there are some issues that need to be resolved as listed below:  
+
+<|ref|>text<|/ref|><|det|>[[118, 238, 860, 280]]<|/det|>
+1. The main conclusion of this work is that the bandgap of WS2 can be significantly modified by hot electrons injected from the PC. However, there is no strong evidence that hot electrons dominate the observed transient features in TA spectroscopy:  
+
+<|ref|>text<|/ref|><|det|>[[118, 280, 864, 351]]<|/det|>
+1.1. According to the discussions, hot electrons are believed to be injected into WS2 by tunneling through the Al2O3 spacer layer. While the spacer layer helps to prevent injected hot electrons from tunneling back into the PC, it also decreases the hot electron injection efficiency and this effect depends on the spacer layer thickness. It is not clear how the authors determine the thickness of the Al2O3 to use for this work and if this thickness is favorable for DET.  
+
+<|ref|>text<|/ref|><|det|>[[118, 350, 864, 378]]<|/det|>
+1.2. Similarly, the authors need to explain why the tunneling barrier height can be set to be 1 eV in Equation (2).  
+
+<|ref|>text<|/ref|><|det|>[[118, 378, 866, 420]]<|/det|>
+1.3. Since losses are not included in Equation (2), the actual number of injected hot electrons could be much lower than calculated. The authors should provide a better evaluation of the effect of losses either theoretically or experimentally.  
+
+<|ref|>text<|/ref|><|det|>[[118, 419, 875, 476]]<|/det|>
+2. Since the quality of the silver cap layer can affect the generated hot electrons and local electric field distribution, did the authors check the optical properties of the deposited silver layer? If so, they should show results of the optical measurements. Otherwise, those measurements need to be made for the next revision.  
+
+<|ref|>text<|/ref|><|det|>[[118, 476, 833, 504]]<|/det|>
+3. To compare the broad maxima in Figure 2d and 2g, it would be more helpful to show both figures with the same y-scale.  
+
+<|ref|>text<|/ref|><|det|>[[118, 504, 868, 546]]<|/det|>
+4. The authors also mentioned the enhanced nonlinear optical responses of the PC-WS2 structure. How is it related to the rest of this work? The authors should elaborate more on the relevance of the enhanced nonlinear response.  
+
+<|ref|>text<|/ref|><|det|>[[118, 546, 768, 561]]<|/det|>
+I would recommend a major revision for the authors to address the questions above.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 681, 448, 696]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 708, 874, 780]]<|/det|>
+The authors study nonlinear optical properties of plasmonic crystals covered with WS2 monolayers using time- resolved transient absorption measurements. The time resolution is of the order of 100 fs, insufficient to resolve coherent exciton- plasmon coupling phenomena in the time domain. The samples show some signatures of exciton- plasmon coupling. Their linear optical properties have been studied in earlier work.  
+
+<|ref|>text<|/ref|><|det|>[[118, 780, 872, 906]]<|/det|>
+The authors do claim that the samples are in the strong coupling regime but, in my opinion, this is actually not convincingly supported by the data shown in the paper or in the SI. It seems to me that all the spectra that are shown are consistent with an exciton- plasmon system in the intermediate coupling regime (exciton lifetime < rabi period < plasmon lifetime). Support for the strong coupling claim in the paper should be given by a quantitative analysis of the line shape of the spectra. This criticism, however, is only partially relevant for the main results of the present paper probing incoherent optical nonlinearities occurring on time scales that are longer than the lifetimes of coherent excitations in the coupled system. The main experimental observations are potentially of interest even when the strong- coupling claim is withdrawn.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 875, 155]]<|/det|>
+In Fig. 2, the authors show angle- resolved studies of the nonlinear optical response of their system. Several interesting features are observed. First of all, some spectra display a transient below- bandgap response, similar to what has been shown by Chernikov et al. for bare WS2 layers in Nature Phot. 9, 466 (2015). Second, the authors see a strongly angle- dependent differential transmission lineshape around the WS2 A exciton at 2.1 eV.  
+
+<|ref|>text<|/ref|><|det|>[[117, 155, 880, 281]]<|/det|>
+Let us start with the second feature - which is not discussed at all in the paper. What is the origin of the lineshapes in Fig. 2d and g? What explains the angle- dependent change in lineshape seen in Fig. S12? If the excitons and plasmons in the present system are indeed coupled, it should be possible to analyze these lineshapes in terms of a phenomenological coupled Lorentz oscillator model as routinely used for describing such coupled systems. I would like to strongly encourage the authors to perform such an analysis. It would provide interesting new insight into the nonlinear optical properties of TMDC excitons coupled to plasmons. Also, it could certainly help the authors to better understand some of their results. Based on such analysis, a more quantitative and less speculative analysis of the dynamical behavior should be possible.  
+
+<|ref|>text<|/ref|><|det|>[[117, 281, 880, 536]]<|/det|>
+Now, let us turn to the below- bandgap absorption peak. The spectral signatures are indeed very similar to what has been seen by Chernikov in his Nature Photonics paper. I therefore did not quite understand why the authors argue that their results "sharply contrast" with those earlier observations. I would guess that the nature of the below- bandgap peak is very much the same as demonstrated earlier: bandgap renormalization due to a nonequilibrium carrier concentration in WS2 film. The fact that this is seen here at lower pump powers seems not particularly surprising to me: the presence of the plasmonic crystal enhances the absorption (even at the off- resonant pump energy). Since I agree with the authors that the data show this below- bandgap peak at lower fluence than observed before, I believe that the findings are interesting. I would therefore like to ask the authors to thoroughly discuss their observations and to comparatively discuss different physical mechanisms that could account for it. In the present manuscript, they focus very much on a "hot electron from plasmonic crystals" explanation which - in my opinion - is not necessarily supported by the data. I think that it is realistic to estimate plasmonic absorption enhancements at the pump energy and to carefully compare photoinduced carrier densities with and without plasmonic field enhancement. Can plasmonic field enhancement account already for the increase in bandgap renormalization? This question should be discussed in much more depth than in the present manuscript and - if possible - supported by simulations of the linear optical properties of the sample.  
+
+<|ref|>text<|/ref|><|det|>[[119, 536, 288, 548]]<|/det|>
+More technical issues:  
+
+<|ref|>text<|/ref|><|det|>[[117, 549, 810, 576]]<|/det|>
+1. As mentioned already, I do not believe that the claim of "fast and repeated hot electron population" (aka strong coupling) on page 8 is justified.  
+
+<|ref|>text<|/ref|><|det|>[[119, 576, 393, 590]]<|/det|>
+2. A color code in Fig. 1b is missing.  
+
+<|ref|>text<|/ref|><|det|>[[119, 590, 395, 603]]<|/det|>
+3. A color code in Fig. 3a is missing.  
+
+<|ref|>text<|/ref|><|det|>[[117, 603, 863, 631]]<|/det|>
+4. Why are the data in Fig. 3c, d so noisy. The time-dependent spectra in Fig. 3a look reasonably clean. Has Fourier filtering and smoothing been applied to them.  
+
+<|ref|>text<|/ref|><|det|>[[117, 631, 850, 673]]<|/det|>
+5. The physics behind Eq. 2 remains unclear since the model is not explained in the text. If the authors want to maintain their "hot electron" explanation, the model needs more thorough explanation.  
+
+<|ref|>text<|/ref|><|det|>[[119, 674, 848, 728]]<|/det|>
+In summary, the reported show some new aspects of the transient optical nonlinearities of exciton-plasmon-coupled systems and hence may be of interest to the research community. In order to make the data publishable a much more in-depth analysis and discussion is needed. Claims of "strong coupling" should either be proven or withdrawn.  
+
+<|ref|>text<|/ref|><|det|>[[119, 743, 248, 756]]<|/det|>
+Christoph Lienau
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 97, 448, 112]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 125, 868, 224]]<|/det|>
+The authors reported optical control of bandgap of a monolayer WS2 integrated with a plasmonic structure. The sample studied contains a monolayer WS2 on a self- assembled plasmonic crystal. By coupling the plasmonic resonance with A- excitons of WS2, the authors demonstrate control of the WS2 bandgap by optically exciting plasmonic hot electrons that transfer to WS2. Ultrafast control of bandgap renormalization can have important applications in ultrafast photonics such as ultrafast optical switch. As such, the results are significant. However, there are a few issues that I wish the authors could clarify.  
+
+<|ref|>text<|/ref|><|det|>[[118, 237, 867, 336]]<|/det|>
+1. The main conclusion of the study is a 650-meV bandgap renormalization at RT. The authors need to provide strong evidence and more detailed analysis to support this claim. It appears that the authors assumed an exciton binding energy of 200 meV. However, this is the value for a WS2 monolayer on an insulating substrate. When coupled with a plasmonic crystal, this is likely to change, since such a large exciton binding energy is due to the lack of screening in the 2D form. Without knowing the precise value of the exciton binding energy in their structure, the claimed renormalization may not be accurate.  
+
+<|ref|>text<|/ref|><|det|>[[118, 350, 846, 392]]<|/det|>
+2. It is also a bit concerning to use the peaks in TA spectra as the parameter to determine the bandgap. The TA peak aligns with the resonance only if the phase-space filling is the dominant mechanisms of TA. In 2D materials, this may not be the case.  
+
+<|ref|>text<|/ref|><|det|>[[118, 405, 874, 476]]<|/det|>
+3. The authors compared with Ref. 9 in terms of the achieved normalization and the used injection level. However, the hot-electron inject method used here results in net charges in WS2, which is quite different from optical doping. The authors should discuss about the potential implications. In addition, a direct comparison of the pump fluence used may not be fair, since the absorptions are different.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 108, 210, 125]]<|/det|>
+## Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[115, 140, 881, 177]]<|/det|>
+We are glad that Reviewer #1 thinks that "The authors report an interesting study on doping twodimensional semiconductors using plasmonic hot electrons.". In addition, the reviewer also suggests that:  
+
+<|ref|>text<|/ref|><|det|>[[135, 190, 870, 208]]<|/det|>
+1. We should provide stronger evidence that hot electrons dominate the observed transient features by  
+
+<|ref|>text<|/ref|><|det|>[[159, 221, 768, 239]]<|/det|>
+1.1 providing more details on the determination of the thickness of the \(\mathrm{Al}_2\mathrm{O}_3\) spacer.  
+
+<|ref|>text<|/ref|><|det|>[[159, 253, 882, 306]]<|/det|>
+Author response: We very much thank the reviewer and agree with her/him that the spacer thickness is critical in forming a favourable tunneling barrier for hot electron transfer. We therefore have added the following sentence in the main text (line 246ff):  
+
+<|ref|>text<|/ref|><|det|>[[159, 320, 881, 356]]<|/det|>
+"...The thickness of the metal and dielectric layers are characterised using an ellipsometer measuring identical evaporations on flat silicon substrates."  
+
+<|ref|>text<|/ref|><|det|>[[159, 370, 882, 423]]<|/det|>
+In addition, we have also added a schematic in Fig.S14 in the Supplementary Information (SI) to show the surface morphology of \(\mathrm{Al}_2\mathrm{O}_3\) layer. Relevant discussion can be found in the response to the below suggestion.  
+
+<|ref|>text<|/ref|><|det|>[[159, 436, 882, 472]]<|/det|>
+1.2 We should explain why the tunneling barrier height can be set to be \(1\mathrm{eV}\) in Eq.(2) in the main text and the relation between the barrier height and the \(\mathrm{Al}_2\mathrm{O}_3\) spacer thickness.  
+
+<|ref|>text<|/ref|><|det|>[[159, 485, 882, 646]]<|/det|>
+Author response: We appreciate that Reviewer#1 gives us this opportunity to better explain the height of tunneling barrier. As the reviewer suggested, the barrier height closely relates to the spacer thickness. In our sample, the spacer thickness varies with different locations at a level of \(2.5\pm 2\mathrm{nm}\) According to the previous work [Nanoscale, 11, 4811, (2019)], the \(\mathrm{Al}_2\mathrm{O}_3\) spacer with a thickness of \(\sim 2.1\mathrm{nm}\) can form a tunneling barrier of \(\sim 0.8\mathrm{eV}\) at the metal- \(\mathrm{WS}_2\) interface. Therefore here we take \(\Delta \phi_{\mathrm{TB}} = 1\mathrm{eV}\) , which is also a value that is commonly used in other studies, e.g. [Adv. Opt. Mater., 5, 1600594, (2017)] and [ACS Photonics, 4, 2759, (2017)]. To clearly elucidate the value setting, we have added a paragraph (line 419ff) and Fig.S14 in the SI and the following sentence (line 195ff) in the main text:  
+
+<|ref|>text<|/ref|><|det|>[[159, 660, 882, 713]]<|/det|>
+"...In addition, the tunneling barrier \(\Delta \phi_{\mathrm{TB}}\) is set to be \(1\mathrm{eV}\) , which is a typical for the ultrathin \(\mathrm{Al}_2\mathrm{O}_3\) layers used in our system[31, 36], and this setting can help address other dissipations that are not considered in the whole excitation process...."  
+
+<|ref|>text<|/ref|><|det|>[[160, 727, 275, 743]]<|/det|>
+with references:  
+
+<|ref|>text<|/ref|><|det|>[[160, 757, 882, 810]]<|/det|>
+[31] Xiang Tian Kong, Zhiming Wang, and Alexander O. Govorov. "Plasmonic Nanostars with Hot Spots for Efficient Generation of Hot Electrons under Solar Illumination", Adv. Opt. Mater., 5, 1600594, 2017.  
+
+<|ref|>text<|/ref|><|det|>[[160, 825, 882, 878]]<|/det|>
+[36] Shan Zheng, Haichang Lu, Huan Liu, Dameng Liu, and John Robertson. "Insertion of an ultrathin \(\mathrm{Al}_2\mathrm{O}_3\) interfacial layer for Schottky barrier height reduction in \(\mathrm{WS}_2\) field- effect transistors", Nanoscale, 11, 4811, 2019.  
+
+<|ref|>text<|/ref|><|det|>[[160, 892, 884, 927]]<|/det|>
+1.3 Reviewer #1 suggests that we should provide a better evaluation of the losses for calculation of hot electron density using Eq.(2).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[160, 45, 883, 116]]<|/det|>
+Author response: We are very much grateful that the reviewer has raised this suggestion and agree with the reviewer that it is necessary to include losses in the calculation of hot electron density, as the density is the critical factor in identifying the role of hot electrons in the observed bandgap renormalisation. We have therefore revised Eq.(2) to a new form:  
+
+<|ref|>equation<|/ref|><|det|>[[329, 140, 880, 180]]<|/det|>
+\[N_{\mathrm{e}} = \frac{F_{\mathrm{pump}}\cdot\eta_{\mathrm{A}}\cdot\eta_{\mathrm{D}}\cdot\eta_{\mathrm{pl}}}{2c\epsilon_{0}}\cdot \mathbf{F}\cdot \frac{1}{\pi^{2}}\frac{e^{2}E_{\mathrm{F}}^{2}}{\hbar}\frac{\hbar\omega - \Delta\phi_{\mathrm{TB}}}{(\hbar\omega)^{4}} \quad (2)\]  
+
+<|ref|>text<|/ref|><|det|>[[159, 198, 883, 324]]<|/det|>
+In short, three major losses have been included: (i) \(\eta_{\mathrm{A}} = 55\%\) is the ratio of pump energy that is absorbed by the system; (ii) \(\eta_{\mathrm{D}} = 2 / 3\) presents the down- converted energy ratio to excite polaritons; and \(\eta_{\mathrm{pl}} = 50\%\) characterises the ratio of plamonic component in polaritons. The detailed discussions of these major losses and the deduction of the revised Eq.(2) have been added into both the SI and the main text. Specifically, a new sub- section "Including losses" has been added in the Section 6 of the SI, which includes Fig.S13 and relevant discussions. In addition, the following sentence has been added into the main text (line 173ff):  
+
+<|ref|>text<|/ref|><|det|>[[158, 336, 883, 516]]<|/det|>
+"To prove that hot electron doping can induce the observed bandgap renormalisation, we need to figure out how the hot electrons are generated in our system as well as quantify the net carrier density in the lattice. As explained before, due to the off- resonance frequency of the pump, plasmons in the PC- WS \(_2\) system can only be effectively excited by coupling to excitons. Specifically the pump energy is absorbed by the semiconductor and down- converted to excite the plasmon- exciton polaritons, which, as half- plasmon half- exciton hybrid states, naturally excite their plasmonic component and result in the generation of plasmonic hot electrons. These charges then overcome the tunneling barrier \((\Delta \phi_{\mathrm{TB}})\) formed at the Ag- Al \(_2\) O \(_3\) - WS \(_2\) interface to dope the WS \(_2\) lattice. During this process, the hot electron doping is subject to several major losses, including (i) the limited pump absorption by the WS \(_2\) MLs, (ii) the losses in energy down- conversion and (iii) the losses due to the hybrid nature of polaritons."  
+
+<|ref|>text<|/ref|><|det|>[[160, 529, 400, 546]]<|/det|>
+and in line 188ff in the main text:  
+
+<|ref|>text<|/ref|><|det|>[[159, 560, 883, 632]]<|/det|>
+"...Here we take \(\eta_{\mathrm{A}} \approx 55\%\) as the absorption coefficient, \(\eta_{\mathrm{D}} \approx 66\%\) as the energy down- conversion ratio and \(\eta_{\mathrm{pl}} \approx 50\%\) as the excitation ratio of the plasmon component in polaritons. As a result, \(F_{\mathrm{pump}} \cdot \eta_{\mathrm{A}} \cdot \eta_{\mathrm{D}} \cdot \eta_{\mathrm{pl}}\) corresponds to the process that the pump energy is absorbed, down- converted and coupled to the plasmonic components in polaritons with major losses included..."  
+
+<|ref|>text<|/ref|><|det|>[[158, 644, 883, 752]]<|/det|>
+It turns out that with losses included, the hot electron density in the WS \(_2\) monolayer can still typically achieve \(\sim 10^{13} \mathrm{cm}^{- 2}\) , even approaching \(10^{14} \mathrm{cm}^{- 2}\) at hot spots. These numbers have reached what is required \((3 \times 10^{13} - 1.1 \times 10^{14} \mathrm{cm}^{- 2})\) to induce a \(\sim 550 \mathrm{meV}\) bandgap renormalisation by electrical doping, meaning that the hot electron densities are sufficient to result in the observed bandgap restructuring. For more details, please see the revised Fig.4 and relevant paragraphs in the main text (lines \(200 - 219 \mathrm{ff}\) ).  
+
+<|ref|>text<|/ref|><|det|>[[137, 765, 882, 819]]<|/det|>
+2. Reviewer #1 suggests that we demonstrate the optical properties of the bare plasmonic crystal (PC) in the manuscript, since the quality of the silver PC can affect the generated hot electrons and local electric field distribution.  
+
+<|ref|>text<|/ref|><|det|>[[156, 833, 880, 851]]<|/det|>
+Author response: We agree with the reviewer. Due to the frame limit, we have included the angle
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[157, 45, 883, 81]]<|/det|>
+resolved transmission spectra of the bare PC sample in Fig.S1(a) of the SI for comparison with the spectra of the PC sample integrated with a \(\mathrm{WS}_2\) monolayer [PC- \(\mathrm{WS}_2\) , Fig.S1(b)].  
+
+<|ref|>text<|/ref|><|det|>[[137, 95, 883, 130]]<|/det|>
+3. Reviewer #1 suggests that we we show Fig.2d and Fig.2g with the same y-scale, which will provide a better comparison for the bandgap renormalisation effect.  
+
+<|ref|>text<|/ref|><|det|>[[156, 144, 883, 179]]<|/det|>
+Author response: We thank the reviewer for her/his careful observation. We therefore have made relevant revision to Fig.2d and 2g in the main text.  
+
+<|ref|>text<|/ref|><|det|>[[137, 193, 883, 228]]<|/det|>
+4. Reviewer #1 suggests that we elaborate more on the relevance of the enhanced nonlinear response in our system (Fig.3a and 3b in the main text, and Fig.S15 and S16 in the SI).  
+
+<|ref|>text<|/ref|><|det|>[[158, 242, 883, 312]]<|/det|>
+Author response: We very much thank Reviewer #1 for this valueable suggestion, since this has inspired us to explore a new research direction. The nonlinear response of our system under high- power pump mainly includes: (i) the spectral shift (Fig.S15) and (ii) the delayed occurence (Fig.S16) of the polariton maxima.  
+
+<|ref|>text<|/ref|><|det|>[[158, 327, 883, 433]]<|/det|>
+In short, by comparing the relative magnitudes of the polariton split maxima, we find that the spectral shift highly relates to the excitonic resonance shift that is induced by carrier density enhancement. Likewisely, the delayed occurence of maxima can also relate to the enhancement of carrier density in the lattice. These results correspond to the main conclusion of this manuscript, i.e. the strong coupling between plasmons and excitons facilitates the generation of hot electrons, resulting in the elevation of carrier density in the lattice. We have added these discussions into Section 8 of the SI.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 448, 211, 465]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[115, 480, 883, 550]]<|/det|>
+We are very grateful that Reviewer #2 positively commented on our results, e.g. "The main experimental observations are potentially of interest", "I believe that the findings are interesting." and "the reported show some new aspects of the transient optical nonlinearities of exciton plasmon- coupled systems and hence may be of interest to the research community". In addition, s/he has provided many valueable suggestions:  
+
+<|ref|>text<|/ref|><|det|>[[137, 564, 883, 635]]<|/det|>
+1. Reviewer #2 strongly encourages that we perform lineshape analysis on spectral features of our system using a phenomelogical coupled Lorentz oscillator model, since this enables more quantitative and less speculative analysis on the coupling state of the system, providing a solid ground for other discussions in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[158, 649, 883, 737]]<|/det|>
+Author response: We fully agree with Reviewer #2 and very much appreciate this suggestion, as it allows us to gain a much better understanding of the coupling behaviours in our system. Following the her/his instructions, we have used a coupled Lorentz model to fit the steady-state transmission spectra of both the bare PC and PC- \(\mathrm{WS}_2\) systems at different angles, which can be seen from Fig.S2 in the SI, where only the spectra at \(\theta = 22^{\circ}\) are shown.  
+
+<|ref|>text<|/ref|><|det|>[[158, 752, 883, 840]]<|/det|>
+It turns out that the rate of energy exchange between plasmons and excitons \((2g)\) is slower than the plasmon dephasing \((\kappa)\) but faster than the exciton decay rate \((\gamma)\) , i.e. \(\kappa > 2g > \gamma\) , but the coupling strength in our system can still achieve \(2g > (\kappa + \gamma) / 2\) , which is typically treated as in the strong coupling regime, according to an extensively used criterion in coupling systems, e.g. in the works [Khitrova et al., Nat. Phys., 2, 81 (2006)] and [Lienau et al., ACS Nano, 8, 1056 (2014)].
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[265, 52, 732, 310]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[217, 317, 775, 335]]<|/det|>
+<center>Figure.S2 (SI): Analysis of spectral line shape and plasmon-exciton coupling. </center>  
+
+<|ref|>text<|/ref|><|det|>[[159, 358, 883, 484]]<|/det|>
+We want to point out that the criterion of strong coupling highly depends on specific scenarios. For example, in traditional studies of cavity quantum electrodynamics, a vacuum Rabi splitting that is smaller than the cavity loss may result in unwanted decoherence among excitations. Therefore it requires a strict coupling criterion. But for some other situations, like in our system, the strong coupling effect is only used to provide a channel for energy transfer between plasmons and excitons. In this case, the relatively loose coupling criterion \([2g > (\kappa + \gamma) / 2]\) may apply, because the plasmon- exciton hybrid state can still remain effective in the presence of moderate cavity losses.  
+
+<|ref|>image<|/ref|><|det|>[[264, 498, 732, 640]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[223, 648, 765, 666]]<|/det|>
+<center>Figure.S3 (SI): Mixing coefficients in the strong plasmon-exciton coupling. </center>  
+
+<|ref|>text<|/ref|><|det|>[[159, 681, 883, 858]]<|/det|>
+The effectiveness of the hybrid state can be determined through the relative degree of mixing between the cavity- plasmon and the exciton in plasmon- exciton polaritons. In short, following the method from [Lienau et al., ACS Nano, 8, 1056 (2014)], we have included the damping factors (e.g. \(i\kappa\) and \(i\gamma\) ), using complex frequency of plasmons, excitons and polaritons to calculate the mixing coefficients for both upper and lower polaritons. As shown in the Fig.S3(b) of SI, it turns out that the coefficients calculated from experimentally acquired parameters slightly drift from the damping- free curves, but still reach \(\sim 50\%\) at the tuned state. This means that the plasmonic component and excitonic component account for half of the polariton energy respectively, indicating that moderate losses in plasmons do not significantly change the strong coupling nature of the PC- WS \(_2\) system in terms of energy exchange. Theoretical simulations also support this conclusion [Fig.S3(a)].
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[160, 45, 883, 116]]<|/det|>
+Therefore, here we kindly ask the reviewer that if s/he can consider the situation and allow us to treat the system as in the strong coupling regime. To better demonstrate the coupling behaviours in our system, we have rewritten the whole Section 1 in the SI, including newly added figures Fig.S1, S2 and S3. We have also added the following sentence in the main text (line 66ff):  
+
+<|ref|>text<|/ref|><|det|>[[160, 130, 883, 184]]<|/det|>
+"The splitting at the tuned state \((\theta = 22^{\circ})\) can be characterised by a vacuum Rabi splitting of \(\hbar \cdot \Omega_{R} =\) \(\hbar \cdot 2g\approx 140meV\) exceeding the widely used coherent strong coupling criterion \(2g > (\kappa +\gamma) / 2\) [21,22], where \(\kappa\) is the dissipation of plasmon modes and \(\gamma\) is the exciton decay rate."  
+
+<|ref|>text<|/ref|><|det|>[[160, 199, 275, 214]]<|/det|>
+with references:  
+
+<|ref|>text<|/ref|><|det|>[[160, 230, 883, 264]]<|/det|>
+"[21] Khitrova, G., Gibbs, H. M., Kira, M., Koch, S. W. and Scherer, A. Vacuum Rabi splitting in semiconductors., Nat. Phys., 2, 81, 2006."  
+
+<|ref|>text<|/ref|><|det|>[[160, 279, 883, 331]]<|/det|>
+"[22] Wang, W., Vasa, P., Pomraenke, R., Vogelgesang, R., De Sio, A., Sommer, E., Maiuri, M., Manzoni, C., Cerullo, G. and Lienau, C., Interplay between strong coupling and radiative damping of excitons and surface plasmon polaritons in hybrid nanostructures, ACS Nano, 8, 1056, 2014."  
+
+<|ref|>text<|/ref|><|det|>[[159, 345, 883, 470]]<|/det|>
+In addition, we would like to point out that here all lineshape analyses were based on the steady- state transmission spectra, because the steady- state spectra provide stable, collective and power (and time) independent spectral features, which can simplify the discussion and is typically used for the analysis of plasmon- exciton coupling. In contrast, the spectral positions and linewidth of features in transient spectra \(\Delta \mathrm{T} / \mathrm{T}\) are highly sensitive to the pump power and delay time (see Fig.2 and 3 in the main text and Fig.S4, S6, S8, S15, S16 and S17 and relevant discussions in the SI), which brings enoumours difficulties and uncertainties in analysing the coupling.  
+
+<|ref|>text<|/ref|><|det|>[[137, 484, 882, 555]]<|/det|>
+2. Reviewer #2 suggests that we discuss the "strongly angle-dependent differential transmission lineshape around the \(W S_{2}\) exciton \(A\) ". Specifically, it is necessary to figure out "the origin of the lineshapes in Fig.2d and \(2g\) " as well as to explain "the angle-dependent change in lineshape seen in Fig.S12". The reviewer suggests that we use a Lorentz oscillator model to analyse the lineshapes.  
+
+<|ref|>text<|/ref|><|det|>[[159, 569, 882, 621]]<|/det|>
+Author response: We agree with the reviewer that it is important to discuss the lineshape change in transient differential spectra at distinct incident angles, as this will provide us a deeper understanding of the coupled system's transient dynamics.  
+
+<|ref|>text<|/ref|><|det|>[[159, 636, 882, 831]]<|/det|>
+First of all, we want to point out that the coupled Lorentz oscillator model may not be applicable for analysing the lineshapes of differential spectra. Specifically, Lorentz models can describe relatively simple transition processes, e.g. excitations from ground to excited state or decay from excited to ground state, which, therefore, are typically used to analyse the simple resonant features in steady- state spectra. However, the differential spectra used here is \((\mathrm{T} - \mathrm{T}_{0}) / \mathrm{T}_{0}\) , which is the normalised subtractions between two spectra at different delay times, comprising of very complicated physical processes in addition to energy transition, e.g. the time- dependent many- body interactions between excitations. As a result, resonance features may broaden and shift at different delay times, leading to negative magnitudes and peak shifts as compared to the steady- state spectra. In this case, using Lorentz oscillator models to analyse the lineshapes of the differential spectra does not provide effective information.  
+
+<|ref|>text<|/ref|><|det|>[[159, 845, 882, 863]]<|/det|>
+Instead, we have phenomenologically analysed the lineshape change of differential spectra at different
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[160, 46, 637, 63]]<|/det|>
+angles and delay times, mainly drawing the following conclusions:  
+
+<|ref|>text<|/ref|><|det|>[[157, 77, 883, 112]]<|/det|>
+(i) The splitting in differential spectra correspond to the splitting in the steady-state spectra, but with some frequency shifts and smaller splitting magnitudes.  
+
+<|ref|>text<|/ref|><|det|>[[157, 125, 883, 161]]<|/det|>
+(ii) The frequency shifts and smaller splitting magnitudes should be the result of lineshape broadening and shift induced by many-body Auger recombination.  
+
+<|ref|>text<|/ref|><|det|>[[159, 175, 883, 211]]<|/det|>
+These discussions have been added to Section 2 in the SI, particularly with panel (d), (e) and (f) of Fig.S4 (Fig.S12 in the last version of SI)  
+
+<|ref|>text<|/ref|><|det|>[[137, 224, 883, 333]]<|/det|>
+3. Reviewer #2 suggests that we "thoroughly discuss their observations and to comparatively discuss different physical mechanisms that could account for it." Specifically, the reviewer would like us to "estimate plasmonic absorption enhancements at the pump energy and to carefully compare photoinduced carrier densities with and without plasmonic field enhancement." This should be "discussed in much more depth than in the present manuscript" and "if possible – supported by simulations of the linear optical properties of the sample."  
+
+<|ref|>image<|/ref|><|det|>[[246, 346, 744, 542]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[190, 549, 801, 567]]<|/det|>
+<center>Figure.S13 (SI): Light absorption in PC-WS \(_2\) system at the pump frequency (3.1 eV) </center>  
+
+<|ref|>text<|/ref|><|det|>[[159, 581, 883, 635]]<|/det|>
+Author response: First we would like to express our sincere appreciation to Reviewer #2, since this suggestion has extended our understanding of carrier population in the system and greatly helped us reshape the later discussions on hot electron doping.  
+
+<|ref|>text<|/ref|><|det|>[[159, 649, 883, 774]]<|/det|>
+Specifically, we have modelled the intensity distribution of the PC- WS \(_2\) system at the pump frequency [Fig. S13 (a)]. It turns out that the intensity enhancement at this frequency is no larger than \(\sim 7\) times at the position of the WS \(_2\) monolayer, which, according to our calculation, results in an average (over the whole area) optical absorption of \(\sim 55\%\) in the monolayer. It means that the absorption in WS \(_2\) is highly enhanced as compared to the bare monolayer ( \(\sim 10\%\) ) that is not integrated with the plasmonic crystal [Fig. S13(b)]. As a result, if the pump pulse that has a fluence of \(12 \mu \mathrm{J} / \mathrm{cm}^2\) can be absorbed and fully converted, the carrier population in the WS \(_2\) ML can reach at a density of \(\sim 1.2 \times 10^{13} \mathrm{~cm}^{- 2}\) .  
+
+<|ref|>text<|/ref|><|det|>[[159, 787, 883, 859]]<|/det|>
+This overestimated value, however, is still one order of magnitude lower than the one ( \(\sim 10^{14} \mathrm{~cm}^{- 2}\) ) that is required to develop a Mott- transition at a cryogenic temperature \(70 \mathrm{~K}\) [Heinz et al., Nat. Photon., 9, 466, 2015], let alone the level at room- temperature. We also want to point out that even if all the absorbed energy in the PC- WS \(_2\) system (i.e. \(\sim 80\%\) absorption of pump, see the absorption spec
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[160, 45, 882, 98]]<|/det|>
+trum of the PC- WS \(_2\) at \(3.1\mathrm{eV}\) in Fig.S13) can be converted to excite excitons in the WS \(_2\) lattice, the generated carrier density is still lower than what is required to induce a large bandgap renormalisation as in our experiment.  
+
+<|ref|>text<|/ref|><|det|>[[160, 112, 882, 183]]<|/det|>
+In addition, the density increase induced by absorption enhancement should appear at transient spectra of all incident angles, but in our case, only the tuned state ( \(\theta = 22^{\circ}\) ) show a large bandgap renormalisation. Therefore, we do not think plasmonic absorption enhancement is the main factor that can induce the observed bandgap renormalisation in our experiments.  
+
+<|ref|>text<|/ref|><|det|>[[160, 197, 880, 215]]<|/det|>
+To present readers with this argument, we have added a new paragraph in the main text (line 129ff):  
+
+<|ref|>text<|/ref|><|det|>[[159, 229, 882, 443]]<|/det|>
+"In our system, the carrier density may be increased by plasmonic absorption enhancement (PAE), which, however, can not provide enough carrier population according to our calculation. Specifically, the excitation of plasmons can enhance the absorption of the pump by the system, which naturally results in an elevation of carrier numbers in the lattice. In the PC- WS \(_2\) system, the pump intensity can be amplified at the position of the WS \(_2\) ML (Fig.S13 in SI), which, according to our calculation, gives a \(\sim 5\) times average increase of absorption in the semiconductor. As a result, the carrier density can achieve up to \(\sim 1.2 \times 10^{13}\mathrm{cm}^{- 2}\) if the absorbed pump energy is fully converted. However, even this overestimated value is still one order of magnitude lower than the density level ( \(\sim 10^{14}\mathrm{cm}^{- 2}\) )[9] required to cause a Mott- transition at \(70\mathrm{K}\) , let alone the level at room- temperature. Furthermore, PAE should also enhance carrier generation in the detuned systems. However, in our experiments, only the tuned system shows a large bandgap renormalisation (Fig.2b and 2d). Hence there must be other mechanisms that can enhance carrier population in addition to PAE."  
+
+<|ref|>text<|/ref|><|det|>[[160, 458, 275, 474]]<|/det|>
+with references:  
+
+<|ref|>text<|/ref|><|det|>[[160, 488, 881, 540]]<|/det|>
+"[9] Alexey Chernikov, Claudia Ruppert, Heather M. Hill, Albert F. Rigosi, and Tony F. Heinz., Population inversion and giant bandgap renormalization in atomically thin WS2 layers. Nat. Photon., 9, 466, 2015."  
+
+<|ref|>text<|/ref|><|det|>[[158, 555, 722, 572]]<|/det|>
+Furthermore, have added a subsection in Section 6 in the SI including Fig.S13.  
+
+<|ref|>text<|/ref|><|det|>[[115, 587, 531, 603]]<|/det|>
+In addition, Reviewer #2 has raised some technical issues:  
+
+<|ref|>text<|/ref|><|det|>[[115, 618, 881, 654]]<|/det|>
+(1) Reviewer #2 suggests that we justify ""fast and repeated hot electron population" (aka strong coupling)" in our system.  
+
+<|ref|>text<|/ref|><|det|>[[115, 668, 535, 685]]<|/det|>
+Author response: Please see above the answers for point 1.  
+
+<|ref|>text<|/ref|><|det|>[[115, 699, 675, 717]]<|/det|>
+(2) & (3) Reviewer #2 suggests that we add colour codes in Fig.1b and Fig.3a.  
+
+<|ref|>text<|/ref|><|det|>[[115, 730, 881, 765]]<|/det|>
+Author response: We are grateful that the reviewer has carefully read our manuscript and we have therefore added colour codes in these figures.  
+
+<|ref|>text<|/ref|><|det|>[[115, 780, 881, 815]]<|/det|>
+(4) Reviewer #2 suggests that we clarify why data in Fig.3c and 3d look noisy while the intensity plot in Fig.3a looks clean, as they are plotted from the same set of data.  
+
+<|ref|>text<|/ref|><|det|>[[115, 829, 881, 864]]<|/det|>
+Author response: We have checked these data and plots, finding that no filtering or smoothing have been applied to them. The inconsistency in noise level might be from different axis range taken. Please note that
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 45, 881, 81]]<|/det|>
+panel c and d only show spectra with an energy range from 1.5 to \(1.9\mathrm{eV}\) , while panel a shows the intensity plot from 1.6 to \(2.5\mathrm{eV}\) . The denser distribution of data point may reduce the illustration of noise.  
+
+<|ref|>text<|/ref|><|det|>[[115, 94, 883, 130]]<|/det|>
+(5) Reviewer #2 suggests that we better explain the model (Eq.2) and elucidate our arguement of "coherent hot electron doping" with more thorough analysis.  
+
+<|ref|>text<|/ref|><|det|>[[115, 143, 882, 252]]<|/det|>
+Author response: We fully agree with the reviewer and therefore have redeveloped our model, having included several major loss factors that may suppress the hot electron doping. By doing so, we have been able to obtain a more accurate estimation of the hot electron density in the \(\mathrm{WS}_2\) lattice. As a result, we find that with major losses included, the hot electron density can still reach the level that is required to develop a large bandgap renormalisation as observed in our experiments. Specifically, we have added a new paragraph in the main text (line 173ff):  
+
+<|ref|>text<|/ref|><|det|>[[114, 264, 882, 444]]<|/det|>
+"To prove that hot electron doping can induce the observed bandgap renormalisation, we need to understand how the hot electrons are generated in our system as well as quantify the net carrier density in lattice. As explained before, due to the off- resonance frequency of pump, plasmons in the PC- \(\mathrm{WS}_2\) system can only be effectively excited by coupling to excitons. Specifically the pump energy is absorbed by the semiconductor and down- converted to excite the plasmon- exciton polaritons, which, as half- plasmon half- exciton hybrid states, naturally excite their plasmonic component and result in the generation of plasmonic hot electrons. These charges then overcome the tunneling barrier \((\Delta \phi_{\mathrm{TB}})\) formed at the Ag- \(\mathrm{Al}_2\mathrm{O}_3\) - \(\mathrm{WS}_2\) interface to dope the \(\mathrm{WS}_2\) lattice. During this process, the hot electron doping is subject to several major losses, including (i) the limited pump absorption by the \(\mathrm{WS}_2\) MLs, (ii) the losses in energy down- conversion and (iii) the losses due to the hybrid nature of polaritons."  
+
+<|ref|>text<|/ref|><|det|>[[115, 456, 882, 492]]<|/det|>
+Following this process, as described in the response to Reviewer #1, we have developed a new model to numerically estimate the density \((N_{\mathrm{e}})\) .  
+
+<|ref|>equation<|/ref|><|det|>[[304, 514, 880, 555]]<|/det|>
+\[N_{\mathrm{e}} = \frac{F_{\mathrm{pump}}\cdot\eta_{\mathrm{A}}\cdot\eta_{\mathrm{D}}\cdot\eta_{\mathrm{pl}}}{2c\epsilon_{0}}\cdot \mathcal{F}\cdot \frac{1}{\pi^{2}}\frac{e^{2}E_{\mathrm{F}}^{2}}{\hbar}\frac{\hbar\omega - \Delta\phi_{\mathrm{TB}}}{(\hbar\omega)^{4}} \quad (2)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 560, 556, 578]]<|/det|>
+Relevant explanations are given at line 188ff in the main text:  
+
+<|ref|>text<|/ref|><|det|>[[115, 591, 882, 700]]<|/det|>
+"...Here we take \(\eta_{\mathrm{A}}\approx 55\%\) as the absorption coefficient, \(\eta_{\mathrm{D}}\approx 66\%\) as the energy down- conversion ratio and \(\eta_{\mathrm{pl}}\approx 50\%\) as the excitation ratio of the plasmon component in polaritons. As a result, \(F_{\mathrm{pump}}\cdot \eta_{\mathrm{A}}\cdot\) \(\eta_{\mathrm{D}}\cdot \eta_{\mathrm{pl}}\) corresponds to the process that the pump energy is absorbed, down- converted and coupled to the plasmonic components in polaritons with major losses included. As optical modes, the excited polaritons gain a spatial distribution (Fig.4b) at the tuned frequency, spreading over the Ag cap surface with hot spots at the interstices between caps. This can be mathematically expressed as \(\mathcal{F} = |\mathbf{E} / \mathbf{E}_0|^2\dots\) "  
+
+<|ref|>text<|/ref|><|det|>[[115, 712, 882, 784]]<|/det|>
+Here we want to thank Reviewer #2 that the efficiency parameters \(\eta_{\mathrm{A}}\) and \(\eta_{\mathrm{pl}}\) are all inspired by her/his suggestions, which are the absorption coefficient and the mixing coefficient respectively. Please see above the suggestions for more details. As a result, we have been able to calculate the carrier density, which is stated in the main text (line 200ff):  
+
+<|ref|>text<|/ref|><|det|>[[115, 796, 882, 850]]<|/det|>
+"Using Eq.2, we are able to plot the spatially distributed hot electron density in the \(\mathrm{WS}_2\) monolayer (Fig.4c). The density naturally acquires identical distributions as do the plasmonic excitations, exhibiting inhomogeneous distribution over the area. It has values typically higher than \(1 \times 10^{13} \mathrm{~cm}^{- 2}\) in most of the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 45, 760, 64]]<|/det|>
+areas, peaking at the interstices between caps with maxima larger than \(2 \times 10^{14} \mathrm{cm}^{- 2} \ldots\) "  
+
+<|ref|>text<|/ref|><|det|>[[115, 77, 883, 131]]<|/det|>
+In addition, a more detailed discussions of these major losses and the deduction of the revised Eq.(2) have been added into the SI. Specifically, a new sub- section "Including losses" has been added in the Section 6 of the SI, which includes Fig.S13 and relevant discussions.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 145, 211, 161]]<|/det|>
+## Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[115, 176, 883, 230]]<|/det|>
+We are very happy that Reviewer #3 thinks highly of our manuscript, e.g. "Ultrafast control of bandgap renormalization can have important applications in ultrafast photonics such as ultrafast optical switch. As such, the results are significant." In addition, the reviewer has raised a few issues:  
+
+<|ref|>text<|/ref|><|det|>[[137, 243, 883, 297]]<|/det|>
+1. Reviewer #3 suggests that we figure out the binding energy of a \(\mathrm{WS}_2\) monolayer deposited on the plasmonic crystal, as this could be largely different from the value for monolayers deposited on a insulating substrate, which will be critical in determining the magnitude of bandgap renormalisation.  
+
+<|ref|>text<|/ref|><|det|>[[158, 310, 883, 399]]<|/det|>
+Author response: We fully agree with the reviewer and very much appreciate this suggestion. After carefully reviewing literature, we have found that the binding energy of a \(\mathrm{WS}_2\) monolayer can be reduced to a half of its original value when deposited on a metal substrate. We have therefore revised our conclusion that the bandgap renormalisation can achieve up to \(\sim 550 \mathrm{meV}\) (about \(100 \mathrm{meV}\) smaller than the previous estimation).  
+
+<|ref|>text<|/ref|><|det|>[[157, 412, 842, 431]]<|/det|>
+Specifically, we have corrected the value in the following sentence in the main text (line 118ff):  
+
+<|ref|>text<|/ref|><|det|>[[159, 444, 884, 515]]<|/det|>
+"...It means that in our experiments, the renormalised bandgap starts at \(E_{g} \approx 1.60 \mathrm{eV}\) , lying \(\sim 400 \mathrm{meV}\) below LP and \(\sim 550 \mathrm{meV}\) below the bandgap of \(\mathrm{WS}_2\) MLs (given that the binding energy of exciton A is decreased to \(\sim 100 \mathrm{meV}\) when deposited on metal substrates[28], i.e. about a half of the initial value[19])."  
+
+<|ref|>text<|/ref|><|det|>[[159, 530, 275, 546]]<|/det|>
+with references:  
+
+<|ref|>text<|/ref|><|det|>[[160, 560, 882, 613]]<|/det|>
+"[19] Paul D. Cunningham, Aubrey T. Hanbicki, Kathleen M. McCreary, and Berend T. Jonker. Photoinduced Bandgap Renormalization and Exciton Binding Energy Reduction in WS2., ACS Nano, 11, 12601, 2017."  
+
+<|ref|>text<|/ref|><|det|>[[159, 627, 883, 698]]<|/det|>
+"[28] Park Soohyung, Mutz Niklas, Schultz Thorsten, Blumstengel Sylke, Han Ali, Aljarb Areej, Li Lain Jong, List-Kratochvil Emil J.W., Amsalem Patrick, and Koch Norbert. Direct determination of monolayer MoS2 and WSe2 exciton binding energies on insulating and metallic substrates. 2D Mater, 5, 025003, 2018."  
+
+<|ref|>text<|/ref|><|det|>[[159, 712, 882, 765]]<|/det|>
+With the correction, the main conclusion in the manuscript remains unchanged, because \(\sim 550 \mathrm{meV}\) renormalisation is still a large one. But here we want to deeply thank Reviewer #3. Without her/his suggestion, we would have made a serious mistake on the magnitude of bandgap renormalisation.  
+
+<|ref|>text<|/ref|><|det|>[[137, 779, 883, 815]]<|/det|>
+2. Reviewer #3 suggests that we clarify the method to determine the spectral position of bandgap, since using transient absorption peaks to tell the bandgap position may not apply to 2D semiconductors.  
+
+<|ref|>text<|/ref|><|det|>[[157, 828, 883, 864]]<|/det|>
+Author response: We agree with the reviewer and have rechecked our measurements for finding the bandgap. Specifically, we have followed the method used in a previous work [Heinz et al., Nat.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[160, 45, 883, 116]]<|/det|>
+Photon., 9, 466, 2015], using the onset of a spectral feature (i.e. the low energy end) but not the peak to determine the band edge position. Using this method, we can always find the point with the lowest energy, avoiding misjudgement of a bandgap position induced by the not fully filled phase- space. In particular, we have added the following sentence in the main text (line 116ff):  
+
+<|ref|>text<|/ref|><|det|>[[160, 130, 883, 165]]<|/det|>
+"The onset of the new bandgap can be extracted from the low- energy end of the broad maximum[9] (red dashed vertical line in Fig.4c)"  
+
+<|ref|>text<|/ref|><|det|>[[160, 181, 281, 197]]<|/det|>
+with a reference:  
+
+<|ref|>text<|/ref|><|det|>[[160, 211, 882, 264]]<|/det|>
+"[9] Alexey Chernikov, Claudia Ruppert, Heather M. Hill, Albert F. Rigosi, and Tony F. Heinz. Population inversion and giant bandgap renormalization in atomically thin WS2 layers. Nat. Photon., 9, 466, 2015."  
+
+<|ref|>text<|/ref|><|det|>[[137, 279, 882, 332]]<|/det|>
+3. Reviewer #3 suggests that we should discuss the potential implications of hot electron doping, since it is different from the case in Ref[9], where the optical doping (dessociated electron-hole pairs induced by a strong optical pump) is the dominant mechanism.  
+
+<|ref|>text<|/ref|><|det|>[[160, 345, 882, 398]]<|/det|>
+Author response: We fully agree with the reviewer. We therefore have added new discussions and a reference to discuss the implications of hot electron doping. Specifically, we have added the following sentence in the main text (line 212ff):  
+
+<|ref|>text<|/ref|><|det|>[[160, 411, 883, 536]]<|/det|>
+"...However, the injected carriers in our system are hot electrons, which are different from the dessociated electron-hole pairs induced by pure optical pumping[9,26], but are more similar to free charge carriers by electrical injection[8,39], where \(a \sim 550 \mathrm{meV}\) bandgap renormalisation in \(\mathrm{WS}_2\) MLs can occur at the electron density of \(3 \times 10^{13} - 1.1 \times 10^{14} \mathrm{cm}^{-2}\) [8]. These results share high similarity with our observation, suggesting that the hot electron doping in our system is able to achieve the threshold to induce a bandgap redshift up to \(\sim 500 \mathrm{meV}\) with carriers draining from conduction band \(K\) to \(\Sigma\) valley[38] that renders the semiconductor indirect"  
+
+<|ref|>text<|/ref|><|det|>[[160, 552, 275, 567]]<|/det|>
+with references:  
+
+<|ref|>text<|/ref|><|det|>[[160, 582, 882, 635]]<|/det|>
+"[8] Alexey Chernikov, Arend M. Van Der Zande, Heather M. Hill, Albert F. Rigosi, Ajanth Velauthapillai, James Hone, and Tony F. Heinz. Electrical Tuning of Exciton Binding Energies in Monolayer WS2. Phys. Rev. Lett., 115, 126802, 2015."  
+
+<|ref|>text<|/ref|><|det|>[[160, 649, 882, 702]]<|/det|>
+"[9] Alexey Chernikov, Claudia Ruppert, Heather M. Hill, Albert F. Rigosi, and Tony F. Heinz. Population inversion and giant bandgap renormalization in atomically thin WS2 layers. Nat. Photon., 9, 466, 2015."  
+
+<|ref|>text<|/ref|><|det|>[[160, 716, 882, 769]]<|/det|>
+"[26] D. Erben, A. Steinhoff, C. Gies, G. Schonhoff, T. O. Wehling, and F. Jahnke. Excitation- induced transition to indirect band gaps in atomically thin transition- metal dichalcogenide semiconductors. Phys. Rev. B, 98, 035434, 2018."  
+
+<|ref|>text<|/ref|><|det|>[[160, 783, 882, 818]]<|/det|>
+"[38] F. Lohof, A. Steinhoff, M. Florian, M. Lorke, D. Erben, F. Jahnke, and C. Gies. Prospects and Limitations of Transition Metal Dichalcogenide Laser Gain Materials. Nano Lett. 19, 210, 2019."  
+
+<|ref|>text<|/ref|><|det|>[[160, 832, 880, 850]]<|/det|>
+"[39] Qiu Zhizhan, Trushin Maxim, Fang Hanyan, Verzhbitskiy Ivan, Gao Shiyuan, Laksono Evan,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[160, 45, 882, 99]]<|/det|>
+Yang Ming, Lyu Pin, Li Jing, Su Jie, Telychko Mykola, Watanabe Kenji, Taniguchi Takashi, Wu Jishan, Neto A H Castro, Yang Li, Eda Goki, Adam Shaffique and Lu Jiong. Giant gate-tunable bandgap renormalization and excitonic effects in a 2D semiconductor. Sci. Adv., 5, eaaw2347, 2019."  
+
+<|ref|>text<|/ref|><|det|>[[160, 112, 882, 148]]<|/det|>
+In addition, Reviewer #3 suggests that we should compare the carrier population in our system with that in the ref[9] in a fairer ground but not only based on a fluence comparison.  
+
+<|ref|>text<|/ref|><|det|>[[160, 161, 882, 197]]<|/det|>
+Author response: We agree with the reviewer and have therefore directly compare the carrier density. Specifically, we have added a new paragraph in the main text (line 129ff):  
+
+<|ref|>text<|/ref|><|det|>[[159, 210, 883, 424]]<|/det|>
+"In our system, the carrier density may be increased by plasmonic absorption enhancement (PAE), which, however, can not provide enough carrier population according to our calculation. Specifically, the excitation of plasmons can enhance the absorption of pump by the system, which naturally results in an elevation of carrier numbers in the lattice. In the PC- WS₂ system, the pump intensity can be amplified at the position of the WS₂ ML (Fig.S13 in SI), which, according to our calculation, gives a \(\sim 5\) times average increase of absorption in the semiconductor. As a result, the carrier density can achieve up to \(\sim 1.2 \times 10^{13} \mathrm{~cm}^{- 2}\) if the absorbed pump energy is fully converted. However, even this overestimated value is still one order of magnitude lower than the density level \(\sim 10^{14} \mathrm{~cm}^{- 2}\) [9] required to cause a Mott- transition at 70 K, let alone the level at room- temperature. Furthermore, PAE should also enhance carrier generation in the detuned systems. However, in our experiments, only the tuned system shows a large bandgap renormalisation (Fig.2b and 2d). Hence there must be other mechanisms that can enhance carrier population in addition to PAE."  
+
+<|ref|>text<|/ref|><|det|>[[160, 439, 275, 455]]<|/det|>
+with references:  
+
+<|ref|>text<|/ref|><|det|>[[160, 469, 882, 522]]<|/det|>
+"[9] Alexey Chernikov, Claudia Ruppert, Heather M. Hill, Albert F. Rigosi, and Tony F. Heinz., Population inversion and giant bandgap renormalization in atomically thin WS2 layers. Nat. Photon., 9, 466, 2015."
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 84, 357, 101]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 138, 448, 152]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 166, 830, 181]]<|/det|>
+All my concerns were addressed properly. In principle, the manuscript can be published now.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 255, 448, 270]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 283, 866, 311]]<|/det|>
+The authors have responded in great detail to the questions and comments of the Reviewers. The have also added much helpful additional analysis.  
+
+<|ref|>text<|/ref|><|det|>[[119, 311, 610, 325]]<|/det|>
+I first want to specifically address the point of "strong coupling".  
+
+<|ref|>text<|/ref|><|det|>[[118, 325, 876, 424]]<|/det|>
+1. The authors show, in Fig. S1a angle-resolved transmission spectra of the bare PC sample. This shows two transmission peaks in the wavelength range of the A and B excitons. Out of these two modes, only one of the modes is considered in the coupling scenario that is shown in Fig. S2c. I do - honestly - believe that it is an oversimplification to neglect the existence of these two modes in the coupling scenario. A simple 2x2 model does not work. The same holds for the WS2 sample. Both Xs should be considered in a relativistic coupling scenario, which should be based, at least, on the analysis of a 4 x 4 matrix.  
+
+<|ref|>text<|/ref|><|det|>[[118, 424, 878, 592]]<|/det|>
+2. If I understand it correctly, the "coupled Lorentz oscillator" model that is used by the authors is, again, oversimplified. I see in in Fig. S2b that the spectra are fit two to Lorentz resonances with absorptive lineshape. In the experimental spectra, however, distinct signatures of Fano resonances are seen (for example in the spectra in fig. S1 at \(40^{\circ}\) around the XA exciton). Such Fano resonances arise if the emission of a broader resonance interferes with the emission from a narrower resonance. Such Fano-type lineshapes are neglected in the analysis (if I understand it correctly), even though they are evidently seen in experiment. I think that it is necessary to really calculate the linear transmission spectra for the discussed 4x4 model if the authors want to gain a detailed understanding of the optical properties of their sample. For these reasons, I am skeptical about the "blue bullets" shown in Fig. 2c (and the normal mode splitting deduced from them). I have the suspicion that the actual normal mode splitting may turn out to be smaller when analyzing a more appropriate 4x4 model.  
+
+<|ref|>text<|/ref|><|det|>[[118, 592, 876, 730]]<|/det|>
+3. In their analysis, the authors conclude (page 4 of the SI, 2nd par) that \(\backslash \mathrm{kappa} > 2\mathrm{g} > \backslash \mathrm{gamma}\) . This implies that the sample is \\*not\\* in the strong coupling regime but in an intermediate coupling regime. I would therefore very much suggest to omit the claim that the sample is in the strong coupling regime. This holds in particular since the 2x2 model that is used for extracting the parameters \(\backslash \mathrm{kappa}\) , \(\backslash \mathrm{g}\) and \(\backslash \mathrm{gamma}\) is oversimplified. I therefore firmly suggest to remove the claims that the sample is in the strong coupling regime. Otherwise, a complete lineshape analysis, including all relevant resonances in the sample, together with a demonstrating that strong coupling is truly reached, is necessary. I also point out that I believe that the conclusions drawn from the pump-probe measurements (which cannot resolve the Rabi oscillations anyway) will not change if the authors admit that the sample is "only" in the intermediate coupling regime.  
+
+<|ref|>text<|/ref|><|det|>[[118, 730, 870, 788]]<|/det|>
+4. The authors probe a normal mode coupling of many excitons to several plasmon resonances. This "classical harmonic oscillator coupling" is different from a Vacuum Rabi splitting in the optical spectra of one single quantum emitter coupled to a single cavity mode. The term "Vacuum Rabi splitting" should therefore be avoided.  
+
+<|ref|>text<|/ref|><|det|>[[118, 789, 875, 844]]<|/det|>
+In summary, I think that the manuscript would greatly benefit from a much more thorough analysis of the linear optical transmission spectra in Fig. S1. Such an analysis must include all relevant spectral resonances (in the present case it seems that this requires at least a 4x4 model). I still believe that the results may be of interest for the audience of Nature Communications.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 84, 448, 99]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 112, 797, 141]]<|/det|>
+The authors have adequately addressed my comments and improved their manuscript. I recommend acceptance of the revised manuscript.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 54, 212, 70]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[115, 77, 882, 148]]<|/det|>
+We are glad that Reviewer #2 noted our efforts in improving the manuscript and we are happy that s/he "still believe the results may be of interest for the readers of Nature Communications". In responding to her/his new suggestions, we figure that the problems in our system have now been much better addressed and the role of plasmon- exciton coupling has been further clarified. Thank you!  
+
+<|ref|>text<|/ref|><|det|>[[115, 161, 882, 198]]<|/det|>
+In general, Reviewer #2 suggests that we should perform more detailed analyses on the transmission spectra and the coupling state of our PC- WS \(_2\) system. Specifically,  
+
+<|ref|>text<|/ref|><|det|>[[115, 211, 882, 265]]<|/det|>
+1. Reviewer #2 first suggests that we should analyse the transmission spectra using a coupled model that includes at least 4 oscillators, because there are 4 sets of resonances co-existing and mutually interacting in our system, which are PC-01 mode, PC-02 mode, exciton A and exciton B, respectively.  
+
+<|ref|>text<|/ref|><|det|>[[115, 279, 882, 331]]<|/det|>
+2. In addition to point 1, Reviewer #2 suggests that we should study the optical spectra using a more sophisticated model than the current one, which may better explain the special lineshapes, e.g. Fano-like lineshapes at oblique incident angles, so that the resonance dispersions can be more accurately reproduced.  
+
+<|ref|>text<|/ref|><|det|>[[115, 345, 882, 398]]<|/det|>
+Author response: Since point 1 and 2 are very relevant, here we respond them together. First of all, we are grateful for these suggestions, which enable us to gain a more comprehensive understanding of our own system. We fully agree with the reviewer on these points and have made relevant revisions.  
+
+<|ref|>text<|/ref|><|det|>[[115, 411, 882, 447]]<|/det|>
+(1) We have used the transmission coefficients that build on 4 coupled Lorentz oscillators to reproduce the spectral lineshapes, which is:  
+
+<|ref|>equation<|/ref|><|det|>[[328, 459, 880, 515]]<|/det|>
+\[T(\omega) = |t(\omega)|^{2} = \left|a + \sum_{j = 1}b_{j}\frac{\pi\frac{\gamma_{1}}{2}}{(\omega - \omega_{j}) - i\frac{\gamma_{1}}{2}}\right|^{2} \quad (S1)\]  
+
+<|ref|>text<|/ref|><|det|>[[114, 525, 883, 633]]<|/det|>
+As compared to our previous model, i.e. the linear superposition of the amplitudes of component Lorentz oscillators, Eq.S1 allows us to calculate the transmission spectra that result from the interplay between all 4 oscillators in our system. As the result, we are able to well reproduce the spectral lineshapes from all incident angles, including the Fano- like lineshape at \(40^{\circ}\) . Fig.S2(b) exhibits examples of these fittings. Then the spectral positions of the fitted resonances were extracted and plotted as a function of angles, shown as the blue dots in Fig.S2(c).  
+
+<|ref|>text<|/ref|><|det|>[[115, 646, 691, 664]]<|/det|>
+(2) We then used a \(4 \times 4\) matrix to fit the dispersion of these blue dots, which is:  
+
+<|ref|>equation<|/ref|><|det|>[[260, 675, 880, 754]]<|/det|>
+\[\begin{array}{r}{\left(\begin{array}{c c c c}{\tilde{E}_{\kappa_{1}}(\theta)}&{g_{1\mathrm{A}}}&{0}&{g_{1\mathrm{B}}}\\ {g_{1\mathrm{A}}}&{\tilde{E}_{\gamma_{\mathrm{A}}}}&{g_{2\mathrm{A}}}&{0}\\ {0}&{g_{2\mathrm{A}}}&{\tilde{E}_{\kappa_{2}}(\theta)}&{g_{2\mathrm{B}}}\\ {g_{1\mathrm{B}}}&{0}&{g_{2\mathrm{B}}}&{\tilde{E}_{\gamma_{\mathrm{B}}}}\end{array}\right)\left(\begin{array}{c}{\alpha_{\kappa_{1}}(\theta)}\\ {\alpha_{\gamma_{\mathrm{A}}}(\theta)}\\ {\alpha_{\kappa_{2}}(\theta)}\\ {\alpha_{\gamma_{\mathrm{B}}}(\theta)}\end{array}\right)=\tilde{E}_{p}(\theta)\left(\begin{array}{c}{\alpha_{\kappa_{1}}(\theta)}\\ {\alpha_{\gamma_{\mathrm{A}}}(\theta)}\\ {\alpha_{\kappa_{2}}(\theta)}\\ {\alpha_{\gamma_{\mathrm{B}}}(\theta)}\end{array}\right)}\end{array} \quad (S2)\]  
+
+<|ref|>text<|/ref|><|det|>[[114, 763, 883, 852]]<|/det|>
+Using Eq.S2, we have been able to plot the curves [orange curves in Fig.S2(c)] that follow the dispersions of the fitted resonances. We note that the fitted dispersion curves agree well with spectral positions near the intersection point between PC- 01 mode and exciton A, exhibiting a split spectral feature. In contrast, the curves slightly drift from the fitted resonances near the intersection point between PC- 02 mode and exciton B. We think it is due to the lossy nature of the coupled oscillators, e.g. both PC- 02 mode and exciton B
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[266, 46, 731, 508]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[215, 516, 773, 535]]<|/det|>
+<center>Figure.S2 (SI): Analysis of spectral line shape and plasmon-exciton coupling. </center>  
+
+<|ref|>text<|/ref|><|det|>[[115, 558, 668, 576]]<|/det|>
+acquire broad linewidths, which may temper the fitting accuracy using Eq.S1.  
+
+<|ref|>text<|/ref|><|det|>[[115, 589, 883, 696]]<|/det|>
+(3) From the fitting, we've learned that the strength of coupling between PC-01 and exciton A is \(g_{1\mathrm{A}}\approx 87meV\) and the strength of coupling between PC-02 and exciton B is \(g_{2\mathrm{B}}\approx 30meV\) . What's worth noting is that there is also a coupling between PC-01 mode and exciton B. This can be seen from the blue shift of the resonances near exciton B at low incident angles \((\theta = 0 - 6^{\circ})\) , yielding a strength of \(g_{1\mathrm{B}}\approx 70\mathrm{meV}\) . Here we want to point out that we wouldn't have been able to find this 01-B coupling without using the \(4\times 4\) coupled oscillator model. Therefore we very much appreciate this suggestion from Reviewer #2.  
+
+<|ref|>text<|/ref|><|det|>[[115, 710, 880, 745]]<|/det|>
+Detailed revisions corresponding to these two points can be found in Section 1 of supplementary information (SI).  
+
+<|ref|>text<|/ref|><|det|>[[115, 759, 881, 796]]<|/det|>
+3. Reviewer #2 suggests that we should define our system as in "intermediate" coupling state, if the coupling strength only meets \(2g > (\kappa + \gamma) / 2\) but does not satisfy \(2g > \kappa > \gamma\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 809, 882, 862]]<|/det|>
+Author response: We fully agree with this point, as an accurate identification of coupling state will provide clear application conditions for the coherent doping of plasmonic hot electrons, which is critical for this work. As mentioned above, the coupling strength \(g_{1\mathrm{A}}\approx 87meV\) , which satisfies \(2g_{1\mathrm{A}} > (\kappa_{1} + \gamma_{\mathrm{A}}) / 2\) , yet
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 45, 882, 99]]<|/det|>
+is slower than the plasmonic dephasing. Therefore, the coupling between PC- 01 mode and exciton A should be defined as intermediate coupling. We have added this new definition and removed the previous “strong coupling” definition throughout the manuscript including the main text and SI.  
+
+<|ref|>text<|/ref|><|det|>[[115, 112, 882, 166]]<|/det|>
+4. Reviewer #2 suggests that we should avoid the term “Vacuum Rabi splitting”, as this term is specially used to describe the coupling between a single emitter and a single cavity photon, which is different from our case, in which multiple plasmons couple with multiple excitons.  
+
+<|ref|>text<|/ref|><|det|>[[115, 179, 882, 216]]<|/det|>
+Author response: We fully agree with Reviewer #2 on this point, and therefore have replaced “Vacuum Rabi splitting” as “spectral splitting”.  
+
+<|ref|>text<|/ref|><|det|>[[115, 228, 882, 265]]<|/det|>
+In addition to the specific points raised by the reviewers we have also revised some other parts in the manuscript, e.g. abstract, some paragraphs and figure captions.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 85, 377, 101]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 138, 448, 153]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 166, 878, 209]]<|/det|>
+Dear authors, I quickly looked at your response. I am very happy to read that you have now performed such a thorough analysis of your data. I believe that this new analysis gives a physically much more accurate description of your data.  
+
+<|ref|>text<|/ref|><|det|>[[118, 222, 855, 251]]<|/det|>
+I also acknowledge the removal of claims on "strong coupling" and VRS. This certainly improves the quality of the paper.  
+
+<|ref|>text<|/ref|><|det|>[[118, 264, 808, 292]]<|/det|>
+If I may, I would like to suggest to add Fig. S2 to the main manuscript. It is crucial for understanding the time- resolved results and will certainly be of interest for many readers.  
+
+<|ref|>text<|/ref|><|det|>[[118, 306, 650, 334]]<|/det|>
+I gladly recommend your manuscript for publication in Nature Comm. Christoph Lienau
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 52, 211, 68]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[115, 87, 881, 123]]<|/det|>
+We are very much glad that Reviewer #2 recommended our manuscript to be published in Nature Communications. Thank you!  
+
+<|ref|>text<|/ref|><|det|>[[115, 127, 882, 181]]<|/det|>
+In addition, Reviewer#2 suggests that we add Supplementary Figure 2 into the main text, which will help readers better understand our system. We fully agree on this point and have integrated the main contents, i.e. panel (b) and (c), of Supplementary Figure 2 into Figure 1 in the main text. Please see below.  
+
+<|ref|>image<|/ref|><|det|>[[175, 197, 825, 458]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 465, 875, 483]]<|/det|>
+<center>Figure.1 (Main Text): Sample structure and analysis of spectral line shape and plasmon-exciton coupling. </center>  
+
+<|ref|>text<|/ref|><|det|>[[115, 497, 884, 533]]<|/det|>
+In addition, we have adapted the main text accordingly to explain the newly added panel (c) and (d) of Figure 1. Specifically, we have added the following text (line 69ff):  
+
+<|ref|>text<|/ref|><|det|>[[115, 546, 883, 672]]<|/det|>
+"As a result, within the angle range \((0 - 40^{\circ})\) , PC- 01, PC- 02, \(X_{A}\) and \(X_{B}\) mutually interact. To analyse the couplings between them, we have used a coupled Lorentz model that build on 4 sets of Lorentz oscillators (Supplementary Eq. S1) to fit the transmission spectra of the PC- WS₂ sample. Fig. 1c shows examples of these fittings at different incident angles. The spectral positions of the fitted resonances were then extracted and plotted as a function of angles (blue dots in Fig. 1d). The complicated dispersive behaviours of these resonances were then fitted using a \((4 \times 4)\) matrix of coupled oscillators (Supplementary Eq. S2) to give the critical coupling parameters."  
+
+<|ref|>text<|/ref|><|det|>[[115, 686, 883, 722]]<|/det|>
+In addition to the specific points raised by the reviewers we have also revised some other parts in the manuscript, e.g. abstract, some paragraphs and figure captions.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__739c81ea9cb34f52413dd3041875cdd436c559c1e1bf01921729db46d9491416/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__743a77233d14fadc6182cccbb642cde9d35e1f029140d8f485ff42cc15bd1141/images_list.json

@@ -0,0 +1,70 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1: Installed capacity and power generation of coal power and VRE during the planning period in the BAU scenarios",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2: Differences of the percentage trajectories between ATB 2020 and ATB 2021 under the moderate case[14, 1]",
+    "footnote": [],
+    "bbox": [
+      [
+        145,
+        250,
+        844,
+        781
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3: Comparison between the electricity supply costs using ATB 2020 and ATB 2021",
+    "footnote": [],
+    "bbox": [
+      [
+        337,
+        94,
+        647,
+        283
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4: Annual and Total Carbon Emission in the key years",
+    "footnote": [],
+    "bbox": [
+      [
+        157,
+        544,
+        864,
+        759
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5: Annual carbon reduction trajectory under each scenario",
+    "footnote": [],
+    "bbox": [
+      [
+        313,
+        230,
+        685,
+        415
+      ]
+    ],
+    "page_idx": 11
+  }
+]

peer_reviews/supplementary_0_Peer Review File__748dba856ef92eeef811d187e996cf1aa44407a69761618224ab8f3d40930e14/images_list.json

@@ -0,0 +1,453 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Fig. R1 Schematic representation of the method used to identify a flash drought event. a Schematic representation of the whole phase of a flash drought event. SM decreases from above the 40th percentile to below the 20th percentile with an average decline rate of no less than the 5th percentile for each pentad, and SM below the 20th percentile should last for no less than 3 pentads. The blue solid line represents the 5-day mean SM percentile for a grid point. The orange and green dashed lines represent the wet (the 40th percentile at a particular time of the year during the period 2000–2020) and the dry (the 20th percentiles at a particular time of the year during the period 2000–2020) conditions of SM, respectively. The purple shaded area represents the onset development of flash droughts. b Schematic representation of the onset phase of a flash drought event.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Fig. R2 Comparison of percentages of flash droughts at different lead times relative to all flash droughts and their temporal evolutions. a The global proportion of flash droughts at different lead times relative to all flash droughts. The red dashed lines at the top of each bar represent the range of uncertainty in three different datasets. b The proportion of flash droughts at different lead times relative to all flash droughts over 21 regions. c, d, and e Temporal evolutions of the percentages of flash droughts with the lead times of 1-, 2-3, and 4-5 pentads relative to all flash droughts. The linear annual trends in the proportion of flash droughts at different lead times are estimated based on the Sen's slope estimator, and statistical significances in trends are determined based on the MK test for the whole study period (2000–2020).",
+    "footnote": [],
+    "bbox": [
+      [
+        135,
+        103,
+        850,
+        533
+      ]
+    ],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Fig. R3 Comparison of SM from four datasets. a–d Spatial patterns of annual total SM for GLEAM, Noah, CLSM, and VIC datasets. e–h Same as a–d, but for the 40th percentile SM. i–l Same as a–d, but for the 20th percentile SM.",
+    "footnote": [],
+    "bbox": [
+      [
+        152,
+        180,
+        844,
+        502
+      ]
+    ],
+    "page_idx": 14
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Fig. R4 Comparison of frequencies of occurrence of flash droughts identified by GLEAM, Noah, and CLSM. a–c Spatial pattern of frequencies of occurrence of flash droughts identified",
+    "footnote": [],
+    "bbox": [
+      [
+        157,
+        603,
+        857,
+        850
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "Fig. R5 Comparison of frequencies of occurrence of traditional droughts identified by percentile-based SM, SSI, and SMI for GLEAM, Noah, and CLSM. a–c Spatial pattern of frequencies of occurrence of traditional droughts identified by percentile-based SM for GLEAM, Noah, and CLSM. d–f Spatial pattern of frequencies of occurrence of traditional droughts identified by SSI for GLEAM, Noah, and CLSM. g–i Spatial pattern of frequencies of occurrence of traditional droughts identified by SMI for GLEAM, Noah, and CLSM.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        457,
+        850,
+        767
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_5.jpg",
+    "caption": "Fig. R6 Comparison of frequencies of occurrence of flash droughts identified by the percentile-based SM, SSI, and SMI for GLEAM, Noah, and CLSM. a–c Spatial pattern of frequencies of occurrence of flash droughts identified by the percentile-based SM for GLEAM, Noah, and CLSM. d–f Spatial pattern of frequencies of occurrence of flash droughts identified by SSI for GLEAM, Noah, and CLSM. g–i Spatial pattern of frequencies of occurrence of flash droughts identified by SMI for GLEAM, Noah, and CLSM.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        476,
+        835,
+        782
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_6.jpg",
+    "caption": "Fig. R7 Temporal evolution of the percentages of flash droughts with the lead times of 1-, 2-3, and 4-5 pentads relative to all flash droughts identified by SMI (a-c) and SSI (d-f).",
+    "footnote": [],
+    "bbox": [
+      [
+        140,
+        389,
+        805,
+        697
+      ]
+    ],
+    "page_idx": 18
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 Interannual variations of flash drought-affected land area and its related variables over the NEU-MEU from 2000 to 2020. a, e Anomalies of flash drought-affected land area and relative SM change over the NEU-MEU from three different datasets. The relative SM change is defined as the change of SM percentiles between adjacent pentads (the SM percentile at the current pentad minus the SM percentile at the next pentad), where positive values represent a decline in SM at the current pentad. b, c, and d Anomalies of annual mean T, annual total P, and annual mean VPD over the NEU-MEU.",
+    "footnote": [],
+    "bbox": [
+      [
+        135,
+        98,
+        863,
+        670
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4 Relationship between SM and VPD. a Mean PMF of concurrent extreme SM (below the 30th percentile) and VPD (above the 90th percentile) for the period 2000–2020 based on three datasets. b Mean probability of percentile bins of SM and VPD across all land grid points. c Spatial relationship between the frequency of occurrence of flash droughts and the PMF of SM and VPD extremes (below the 30th percentile SM and above the 90th percentile VPD). d Spatial relationship between the frequency of occurrence of flash droughts and the PMF of SM and VPD extremes (below the 20th percentile SM and above the 90th percentile VPD). Each point in c and d is the mean of the results from three datasets.",
+    "footnote": [],
+    "bbox": [
+      [
+        133,
+        92,
+        860,
+        420
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 Interannual variations of flash drought-affected land area and its related variables over the NEU-MEU from 2000 to 2020. a, e Anomalies of flash drought-affected land area and relative SM change over the NEU-MEU from three different datasets. The relative SM change is defined as the change of SM percentiles between adjacent pentads (the SM percentile at the current pentad minus the SM percentile at the next pentad), where positive values represent a decline in SM at the current pentad. b, c, and d Anomalies of annual mean T, annual total P, and annual mean VPD over the NEU-MEU.",
+    "footnote": [],
+    "bbox": [
+      [
+        135,
+        99,
+        882,
+        690
+      ]
+    ],
+    "page_idx": 24
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4 Relationship between SM and VPD. a Mean PMF of concurrent extreme SM (below the 30th percentile) and VPD (above the 90th percentile) for the period 2000–2020 based on three datasets. b Mean probability of percentile bins of SM and VPD across all land grid points. c Spatial relationship between the frequency of occurrence of flash droughts and the PMF of SM and VPD extremes (below the 30th percentile SM and above the 90th percentile VPD). d Spatial relationship between the frequency of occurrence of flash droughts and the PMF of SM and VPD extremes (below the 20th percentile SM and above the 90th percentile VPD). Each point in c and d is the mean of the results from three datasets.",
+    "footnote": [],
+    "bbox": [
+      [
+        135,
+        92,
+        899,
+        440
+      ]
+    ],
+    "page_idx": 30
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_7.jpg",
+    "caption": "Fig. R8 Variation of SM percentiles for flash droughts of all grid points. a Variation of SM percentiles for flash droughts of all grid points, identified by intensification rate. b Variation of SM percentiles for flash droughts of all grid points, identified by duration. The color lines represent the mean of all flash drought detected from three models.",
+    "footnote": [],
+    "bbox": [
+      [
+        122,
+        128,
+        884,
+        328
+      ]
+    ],
+    "page_idx": 31
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_8.jpg",
+    "caption": "Fig. R9 Variation in SM percentiles for each grid point of flash droughts at 1-, 2-, 3-, 4-, and 5-pentad lead times without considering duration that can diminish crop productivity and yield (SM should not only drop below the 20th percentile but also last for at least three pentads).",
+    "footnote": [],
+    "bbox": [
+      [
+        207,
+        263,
+        800,
+        653
+      ]
+    ],
+    "page_idx": 33
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_9.jpg",
+    "caption": "Fig. R10 Comparison of percentages of flash droughts at different lead times relative to all flash droughts and their temporal evolutions. a The global proportion of flash droughts at different lead times relative to all flash droughts. The red dashed lines at the top of each bar represent the range of uncertainty in three different datasets. b The proportion of flash droughts at different lead times relative to all flash droughts over 21 regions. c, d, and e Temporal evolutions of percentages of flash droughts with the lead times of 1-, 2-3, and 4-5 pentads relative to all flash droughts.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        293,
+        848,
+        725
+      ]
+    ],
+    "page_idx": 34
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_5.jpg",
+    "caption": "Supplementary Fig. 5 Temporal evolution of the percentage of flash droughts developing at 1 pentad relative to all flash droughts across different regions of the world. The red line represents the mean of results obtained from three datasets, and the grey shadows represent the ranges of results derived from three datasets.",
+    "footnote": [],
+    "bbox": [
+      [
+        140,
+        92,
+        785,
+        592
+      ]
+    ],
+    "page_idx": 40
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_6.jpg",
+    "caption": "Supplementary Fig. 6 Temporal evolution of the percentage of flash droughts developing at 2-3 pentads relative to all flash droughts across different regions of the world. The red line represents the mean of results obtained from three datasets, and the grey shadows represent the ranges of results derived from three datasets.",
+    "footnote": [],
+    "bbox": [
+      [
+        161,
+        155,
+        808,
+        658
+      ]
+    ],
+    "page_idx": 41
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_7.jpg",
+    "caption": "Supplementary Fig. 7 Temporal evolution of the percentage of flash droughts developing at 4-5 pentads relative to all flash droughts across different regions of the world. The red line represents the mean of results obtained from three datasets, and the grey shadows represent the ranges of results derived from three datasets.",
+    "footnote": [],
+    "bbox": [
+      [
+        161,
+        152,
+        808,
+        658
+      ]
+    ],
+    "page_idx": 42
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 Interannual variations of flash drought-affected land area and its related variables over the NEU-MEU from 2000 to 2020. a, e Anomalies of flash drought-affected land area and relative SM change over the NEU-MEU from three different datasets. The relative SM change is defined as the change of SM percentiles between adjacent pentads (the SM percentile at the current pentad minus the SM percentile at the next pentad), where positive values represent a decline in SM at the current pentad. b, c, and d Anomalies of annual mean T, annual total P, and annual mean VPD over the NEU-MEU.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 43
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_10.jpg",
+    "caption": "Fig. R11 Relationship between SM and VPD. a Mean PMF of concurrent extreme SM (below the 30th percentile) and VPD (above the 90th percentile) for the period 2000–2020 based on three datasets. b Mean probability of percentile bins of SM and VPD across all land grid points. c Spatial relationship between the frequency of occurrence of flash droughts and the PMF of SM and VPD extremes (below the 30th percentile SM and above the 90th percentile VPD). d Spatial relationship between the frequency of occurrence of flash droughts and the PMF of SM and VPD extremes (below the 20th percentile SM and above the 90th percentile VPD). Each point in c and d is the mean of the results from three datasets.",
+    "footnote": [],
+    "bbox": [
+      [
+        137,
+        160,
+        863,
+        485
+      ]
+    ],
+    "page_idx": 44
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_11.jpg",
+    "caption": "Fig. R1 Variation in SM percentiles for each grid point of flash droughts at 1-, 2-, 3-, 4-, and 5-pentad onset times without considering the duration that can diminish crop productivity and yield (SM should not only drop below the 20th percentile but also last for at least three pentads). The red boxes highlight the events with SM decreasing to below the 40th percentile occur, but then rapidly recover up to the 40th percentile within only one or two pentads under abnormally dry conditions.",
+    "footnote": [],
+    "bbox": [
+      [
+        125,
+        90,
+        833,
+        555
+      ]
+    ],
+    "page_idx": 46
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_12.jpg",
+    "caption": "Fig. R2 Spatial pattern of flash drought duration based on different models. a–d Mean duration of occurrence of flash droughts based on Noah, CLSM, GLEAM, and the ensemble of results from three models. e Probability density of the mean duration of flash droughts based on Noah, CLSM, GLEAM, and the ensemble of resultsfrom three models.",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        186,
+        812,
+        600
+      ]
+    ],
+    "page_idx": 48
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_13.jpg",
+    "caption": "Fig. R3 Meteorological conditions during before (two pentads prior to the occurrence of flash drought), onset and recovery phases of flash drought, as well as the difference between different phases of flash drought in light of flash droughts captured based on SM from Noah, CLSM, and GLEAM models. a–c T, VPD, and P for the before phase of flash drought. d–f Same as a–c but for the onset phase of flash drought. g–i Same as a-c but for the recovery phase of flash drought. j–l Differences of T, VPD, and P between before and onset phases of flash drought. m–o Differences of T, VPD, and P between onset and recovery phases of flash drought. These results are the ensemble of the results from three datasets.",
+    "footnote": [],
+    "bbox": [
+      [
+        139,
+        98,
+        835,
+        597
+      ]
+    ],
+    "page_idx": 56
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_14.jpg",
+    "caption": "Fig. R4 Meteorological conditions during before (two pentads prior to the occurrence of flash drought), onset and recovery phases of flash drought, as well as the difference between different phases of flash drought in light of flash droughts captured based on SM from Noah model. a–c T, VPD, and P for the before phase of flash drought. d–f Same as a–c but for the onset phase of flash drought. g–i Same as a–c but for the recovery phase of flash drought. j–l Differences of T, VPD, and P between before and onset phases of flash drought. m–o Differences of T, VPD, and P between onset and recovery phases of flash drought.",
+    "footnote": [],
+    "bbox": [
+      [
+        144,
+        94,
+        835,
+        590
+      ]
+    ],
+    "page_idx": 57
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_15.jpg",
+    "caption": "Fig. R5 Meteorological conditions during before (two pentads prior to the occurrence of flash drought), onset and recovery phases of flash drought, as well as the difference between different phases of flash drought in light of flash droughts captured based on SM from CLSM model. a–c T, VPD, and P for the before phase of flash drought. d–f Same as a–c but for the onset phase of flash drought. g–i Same as a-c but for the recovery phase of flash drought. j–l Differences of T, VPD, and P between before and onset phases of flash drought. m–o Differences of T, VPD, and P between onset and recovery phases of flash drought.",
+    "footnote": [],
+    "bbox": [
+      [
+        144,
+        93,
+        815,
+        577
+      ]
+    ],
+    "page_idx": 59
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_16.jpg",
+    "caption": "Fig. R6 Meteorological conditions during before (two pentads prior to the occurrence of flash drought), onset and recovery phases of flash drought, as well as the difference between different phases of flash drought in light of flash droughts captured based on SM from GLEAM model. a–c T, VPD, and P for the before phase of flash drought. d–f Same as a–c but for the onset phase of flash drought. g–i Same as a–c but for the recovery phase of flash drought. j–l Differences of T, VPD, and P between before and onset phases of flash drought. m–o Differences of T, VPD, and P between onset and recovery phases of flash drought.",
+    "footnote": [],
+    "bbox": [
+      [
+        133,
+        92,
+        844,
+        600
+      ]
+    ],
+    "page_idx": 60
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_17.jpg",
+    "caption": "Fig. R7 Comparison of meteorological conditions between different phases of flash droughts in light of flash droughts captured based on SM from Noah, CLSM, and GLEAM models. a–c Differences of T, VPD, and P between before (two pentads prior to flash drought) and onset phases of flash droughts in light of flash droughts captured based on SM from Noah model. d–f Differences of T, VPD, and P between onset and recovery phases of flash droughts in light of flash droughts captured based on SM from Noah model. g–l Same as a–e but for the CLSM model. m–r Same as a–e but for the GLEAM model.",
+    "footnote": [],
+    "bbox": [
+      [
+        231,
+        288,
+        758,
+        738
+      ]
+    ],
+    "page_idx": 61
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_18.jpg",
+    "caption": "Fig. R8 Comparison of percentages of flash droughts at different onset times relative to all flash droughts over NEU and MED based on GLEAM, Noah, and CLSM models. a–c Temporal evolutions of percentages of flash droughts with the onset times of 1-, 2-3, and 4-5 pentads relative to all flash droughts over NEU. d–f Temporal evolutions of percentages of flash droughts with the onset times of 1-, 2-3, and 4-5 pentads relative to all flash droughts over MED. The linear annual trends in the proportion of flash droughts at different onset times are estimated based on the Sen's slope estimator, and statistical significances in trends are determined based on the MK test for the entire study period (2000–2020).",
+    "footnote": [],
+    "bbox": [
+      [
+        137,
+        95,
+        835,
+        413
+      ]
+    ],
+    "page_idx": 62
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_19.jpg",
+    "caption": "Fig. R9 Temporal evolutions of annual mean VPD, annual mean T, and annual total P over NEU and MED. a–c annual mean VPD, annual mean T, and annual total P over NEU. d–f Same as a–b but for MED. The linear annual trends in the annual mean VPD, annual mean T, and annual total P are estimated based on the Sen's slope estimator, and statistical",
+    "footnote": [],
+    "bbox": [
+      [
+        135,
+        560,
+        832,
+        820
+      ]
+    ],
+    "page_idx": 64
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_20.jpg",
+    "caption": "Fig. R10 Comparison of percentages of 1-pentad onset flash droughts relative to all flash droughts for different regions based on GLEAM, Noah, and CLSM models. The linear annual trends in the proportion of 1-pentad onset flash droughts are estimated based on the Sen's slope estimator, and statistical significances in trends are determined based on the MK test for the entire study period (2000–2020).",
+    "footnote": [],
+    "bbox": [
+      [
+        131,
+        141,
+        840,
+        664
+      ]
+    ],
+    "page_idx": 65
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_21.jpg",
+    "caption": "Fig. R11 Temporal evolutions of annual mean VPD, annual mean T, and annual total P over AUS, AMZ, SSA, ALA, SAF, and SEA. The linear annual trends in the annual mean VPD, annual mean T, and annual total P are estimated based on the Sen's slope estimator, and statistical significances in trends are determined based on the MK test for the entire study period (2000–2020).",
+    "footnote": [],
+    "bbox": [
+      [
+        135,
+        90,
+        840,
+        633
+      ]
+    ],
+    "page_idx": 65
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Fig. 5 Schematic representation of the method used to identify a flash drought event. a Schematic representation of the whole phase of a flash drought event. SM decreases from above the 40th percentile to below the 20th percentile with an average decline rate of no less than the 5th percentile for each pentad, and SM below the 20th percentile should last for no less than 3 pentads. The blue solid line represents the 5-day mean SM percentile for a grid point. The orange and green dashed lines represent the wet (the 40th percentile at a particular time of the year during the period 2000–2020) and the dry (the 20th percentiles at a particular time of the year during the period 2000–2020) conditions of SM, respectively. The purple shaded area represents the onset development of flash droughts. b Schematic representation of the onset phase of a flash drought event.",
+    "footnote": [],
+    "bbox": [
+      [
+        130,
+        95,
+        848,
+        432
+      ]
+    ],
+    "page_idx": 66
+  }
+]